summaryrefslogtreecommitdiffstats
path: root/xlators/features/locks
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/features/locks')
-rw-r--r--xlators/features/locks/src/Makefile.am14
-rw-r--r--xlators/features/locks/src/clear.c689
-rw-r--r--xlators/features/locks/src/clear.h71
-rw-r--r--xlators/features/locks/src/common.c1966
-rw-r--r--xlators/features/locks/src/common.h258
-rw-r--r--xlators/features/locks/src/entrylk.c1458
-rw-r--r--xlators/features/locks/src/inodelk.c1548
-rw-r--r--xlators/features/locks/src/locks-mem-types.h23
-rw-r--r--xlators/features/locks/src/locks.h328
-rw-r--r--xlators/features/locks/src/pl-messages.h29
-rw-r--r--xlators/features/locks/src/posix.c6322
-rw-r--r--xlators/features/locks/src/reservelk.c587
-rw-r--r--xlators/features/locks/tests/unit-test.c96
13 files changed, 8538 insertions, 4851 deletions
diff --git a/xlators/features/locks/src/Makefile.am b/xlators/features/locks/src/Makefile.am
index 0f79731b415..0b174c19d2d 100644
--- a/xlators/features/locks/src/Makefile.am
+++ b/xlators/features/locks/src/Makefile.am
@@ -1,23 +1,29 @@
+if WITH_SERVER
xlator_LTLIBRARIES = locks.la
+endif
xlatordir = $(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator/features
-locks_la_LDFLAGS = -module -avoid-version
+locks_la_LDFLAGS = -module $(GF_XLATOR_DEFAULT_LDFLAGS)
locks_la_SOURCES = common.c posix.c entrylk.c inodelk.c reservelk.c \
- clear.c
+ clear.c
+
locks_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la
-noinst_HEADERS = locks.h common.h locks-mem-types.h clear.h
+noinst_HEADERS = locks.h common.h locks-mem-types.h clear.h pl-messages.h
-AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src
+AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \
+ -I$(top_srcdir)/rpc/xdr/src -I$(top_builddir)/rpc/xdr/src
AM_CFLAGS = -Wall -fno-strict-aliasing $(GF_CFLAGS)
CLEANFILES =
+if WITH_SERVER
uninstall-local:
rm -f $(DESTDIR)$(xlatordir)/posix-locks.so
install-data-hook:
ln -sf locks.so $(DESTDIR)$(xlatordir)/posix-locks.so
+endif
diff --git a/xlators/features/locks/src/clear.c b/xlators/features/locks/src/clear.c
index 75593b8988c..ab1eac68a53 100644
--- a/xlators/features/locks/src/clear.c
+++ b/xlators/features/locks/src/clear.c
@@ -12,412 +12,449 @@
#include <limits.h>
#include <pthread.h>
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
#include "locks.h"
#include "common.h"
-#include "statedump.h"
+#include <glusterfs/statedump.h>
#include "clear.h"
+const char *clrlk_type_names[CLRLK_TYPE_MAX] = {
+ [CLRLK_INODE] = "inode",
+ [CLRLK_ENTRY] = "entry",
+ [CLRLK_POSIX] = "posix",
+};
+
int
-clrlk_get_kind (char *kind)
+clrlk_get_kind(char *kind)
{
- char *clrlk_kinds[CLRLK_KIND_MAX] = {"dummy", "blocked", "granted",
- "all"};
- int ret_kind = CLRLK_KIND_MAX;
- int i = 0;
-
- for (i = CLRLK_BLOCKED; i < CLRLK_KIND_MAX; i++) {
- if (!strcmp (clrlk_kinds[i], kind)) {
- ret_kind = i;
- break;
- }
+ char *clrlk_kinds[CLRLK_KIND_MAX] = {"dummy", "blocked", "granted", "all"};
+ int ret_kind = CLRLK_KIND_MAX;
+ int i = 0;
+
+ for (i = CLRLK_BLOCKED; i < CLRLK_KIND_MAX; i++) {
+ if (!strcmp(clrlk_kinds[i], kind)) {
+ ret_kind = i;
+ break;
}
+ }
- return ret_kind;
+ return ret_kind;
}
int
-clrlk_get_type (char *type)
+clrlk_get_type(char *type)
{
- char *clrlk_types[CLRLK_TYPE_MAX] = {"inode", "entry", "posix"};
- int ret_type = CLRLK_TYPE_MAX;
- int i = 0;
-
- for (i = CLRLK_INODE; i < CLRLK_TYPE_MAX; i++) {
- if (!strcmp (clrlk_types[i], type)) {
- ret_type = i;
- break;
- }
+ char *clrlk_types[CLRLK_TYPE_MAX] = {"inode", "entry", "posix"};
+ int ret_type = CLRLK_TYPE_MAX;
+ int i = 0;
+
+ for (i = CLRLK_INODE; i < CLRLK_TYPE_MAX; i++) {
+ if (!strcmp(clrlk_types[i], type)) {
+ ret_type = i;
+ break;
}
+ }
- return ret_type;
+ return ret_type;
}
int
-clrlk_get_lock_range (char *range_str, struct gf_flock *ulock,
- gf_boolean_t *chk_range)
+clrlk_get_lock_range(char *range_str, struct gf_flock *ulock,
+ gf_boolean_t *chk_range)
{
- int ret = -1;
-
- if (!chk_range)
- goto out;
+ int ret = -1;
- if (!range_str) {
- ret = 0;
- *chk_range = _gf_false;
- goto out;
- }
-
- if (sscanf (range_str, "%hd,%"PRId64"-""%"PRId64, &ulock->l_whence,
- &ulock->l_start, &ulock->l_len) != 3) {
- goto out;
- }
+ if (!chk_range)
+ goto out;
+ if (!range_str) {
ret = 0;
- *chk_range = _gf_true;
+ *chk_range = _gf_false;
+ goto out;
+ }
+
+ if (sscanf(range_str,
+ "%hd,%" PRId64 "-"
+ "%" PRId64,
+ &ulock->l_whence, &ulock->l_start, &ulock->l_len) != 3) {
+ goto out;
+ }
+
+ ret = 0;
+ *chk_range = _gf_true;
out:
- return ret;
+ return ret;
}
int
-clrlk_parse_args (const char* cmd, clrlk_args *args)
+clrlk_parse_args(const char *cmd, clrlk_args *args)
{
- char *opts = NULL;
- char *cur = NULL;
- char *tok = NULL;
- char *sptr = NULL;
- char *free_ptr = NULL;
- char kw[KW_MAX] = {[KW_TYPE] = 't',
- [KW_KIND] = 'k',
- };
- int ret = -1;
- int i = 0;
-
- GF_ASSERT (cmd);
- free_ptr = opts = GF_CALLOC (1, strlen (cmd), gf_common_mt_char);
- if (!opts)
- goto out;
-
- if (sscanf (cmd, GF_XATTR_CLRLK_CMD".%s", opts) < 1) {
- ret = -1;
- goto out;
+ char *opts = NULL;
+ char *cur = NULL;
+ char *tok = NULL;
+ char *sptr = NULL;
+ char *free_ptr = NULL;
+ char kw[KW_MAX] = {
+ [KW_TYPE] = 't',
+ [KW_KIND] = 'k',
+ };
+ int ret = -1;
+ int i = 0;
+
+ GF_ASSERT(cmd);
+ free_ptr = opts = GF_CALLOC(1, strlen(cmd), gf_common_mt_char);
+ if (!opts)
+ goto out;
+
+ if (sscanf(cmd, GF_XATTR_CLRLK_CMD ".%s", opts) < 1) {
+ ret = -1;
+ goto out;
+ }
+
+ /*clr_lk_prefix.ttype.kkind.args, args - type specific*/
+ cur = opts;
+ for (i = 0; i < KW_MAX && (tok = strtok_r(cur, ".", &sptr));
+ cur = NULL, i++) {
+ if (tok[0] != kw[i]) {
+ ret = -1;
+ goto out;
}
-
- /*clr_lk_prefix.ttype.kkind.args, args - type specific*/
- cur = opts;
- for (i = 0; i < KW_MAX && (tok = strtok_r (cur, ".", &sptr));
- cur = NULL, i++) {
- if (tok[0] != kw[i]) {
- ret = -1;
- goto out;
- }
- if (i == KW_TYPE)
- args->type = clrlk_get_type (tok+1);
- if (i == KW_KIND)
- args->kind = clrlk_get_kind (tok+1);
- }
-
- if ((args->type == CLRLK_TYPE_MAX) || (args->kind == CLRLK_KIND_MAX))
- goto out;
-
- /*optional args, neither range nor basename can 'legally' contain
- * "/" in them*/
- tok = strtok_r (NULL, "/", &sptr);
- if (tok)
- args->opts = gf_strdup (tok);
-
- ret = 0;
+ if (i == KW_TYPE)
+ args->type = clrlk_get_type(tok + 1);
+ if (i == KW_KIND)
+ args->kind = clrlk_get_kind(tok + 1);
+ }
+
+ if ((args->type == CLRLK_TYPE_MAX) || (args->kind == CLRLK_KIND_MAX))
+ goto out;
+
+ /*optional args, neither range nor basename can 'legally' contain
+ * "/" in them*/
+ tok = strtok_r(NULL, "/", &sptr);
+ if (tok)
+ args->opts = gf_strdup(tok);
+
+ ret = 0;
out:
- GF_FREE (free_ptr);
- return ret;
+ GF_FREE(free_ptr);
+ return ret;
}
int
-clrlk_clear_posixlk (xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
- int *blkd, int *granted, int *op_errno)
+clrlk_clear_posixlk(xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
+ int *blkd, int *granted, int *op_errno)
{
- posix_lock_t *plock = NULL;
- posix_lock_t *tmp = NULL;
- struct gf_flock ulock = {0, };
- int ret = -1;
- int bcount = 0;
- int gcount = 0;
- gf_boolean_t chk_range = _gf_false;
-
- if (clrlk_get_lock_range (args->opts, &ulock, &chk_range)) {
- *op_errno = EINVAL;
- goto out;
- }
-
- pthread_mutex_lock (&pl_inode->mutex);
+ posix_lock_t *plock = NULL;
+ posix_lock_t *tmp = NULL;
+ struct gf_flock ulock = {
+ 0,
+ };
+ int ret = -1;
+ int bcount = 0;
+ int gcount = 0;
+ gf_boolean_t chk_range = _gf_false;
+
+ if (clrlk_get_lock_range(args->opts, &ulock, &chk_range)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(plock, tmp, &pl_inode->ext_list, list)
{
- list_for_each_entry_safe (plock, tmp, &pl_inode->ext_list,
- list) {
- if ((plock->blocked &&
- !(args->kind & CLRLK_BLOCKED)) ||
- (!plock->blocked &&
- !(args->kind & CLRLK_GRANTED)))
- continue;
-
- if (chk_range &&
- (plock->user_flock.l_whence != ulock.l_whence
- || plock->user_flock.l_start != ulock.l_start
- || plock->user_flock.l_len != ulock.l_len))
- continue;
-
- list_del_init (&plock->list);
- if (plock->blocked) {
- bcount++;
- pl_trace_out (this, plock->frame, NULL, NULL,
- F_SETLKW, &plock->user_flock,
- -1, EAGAIN, NULL);
-
- STACK_UNWIND_STRICT (lk, plock->frame, -1, EAGAIN,
- &plock->user_flock, NULL);
-
- } else {
- gcount++;
- }
- GF_FREE (plock);
- }
+ if ((plock->blocked && !(args->kind & CLRLK_BLOCKED)) ||
+ (!plock->blocked && !(args->kind & CLRLK_GRANTED)))
+ continue;
+
+ if (chk_range && (plock->user_flock.l_whence != ulock.l_whence ||
+ plock->user_flock.l_start != ulock.l_start ||
+ plock->user_flock.l_len != ulock.l_len))
+ continue;
+
+ list_del_init(&plock->list);
+ if (plock->blocked) {
+ bcount++;
+ pl_trace_out(this, plock->frame, NULL, NULL, F_SETLKW,
+ &plock->user_flock, -1, EINTR, NULL);
+
+ STACK_UNWIND_STRICT(lk, plock->frame, -1, EINTR,
+ &plock->user_flock, NULL);
+
+ } else {
+ gcount++;
+ }
+ __destroy_lock(plock);
}
- pthread_mutex_unlock (&pl_inode->mutex);
- grant_blocked_locks (this, pl_inode);
- ret = 0;
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ grant_blocked_locks(this, pl_inode);
+ ret = 0;
out:
- *blkd = bcount;
- *granted = gcount;
- return ret;
+ *blkd = bcount;
+ *granted = gcount;
+ return ret;
}
/* Returns 0 on success and -1 on failure */
int
-clrlk_clear_inodelk (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
- clrlk_args *args, int *blkd, int *granted, int *op_errno)
+clrlk_clear_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+ clrlk_args *args, int *blkd, int *granted, int *op_errno)
{
- pl_inode_lock_t *ilock = NULL;
- pl_inode_lock_t *tmp = NULL;
- struct gf_flock ulock = {0, };
- int ret = -1;
- int bcount = 0;
- int gcount = 0;
- gf_boolean_t chk_range = _gf_false;
- struct list_head released;
-
- INIT_LIST_HEAD (&released);
- if (clrlk_get_lock_range (args->opts, &ulock, &chk_range)) {
- *op_errno = EINVAL;
- goto out;
- }
-
- if (args->kind & CLRLK_BLOCKED)
- goto blkd;
-
- if (args->kind & CLRLK_GRANTED)
- goto granted;
+ posix_locks_private_t *priv;
+ pl_inode_lock_t *ilock = NULL;
+ pl_inode_lock_t *tmp = NULL;
+ struct gf_flock ulock = {
+ 0,
+ };
+ int ret = -1;
+ int bcount = 0;
+ int gcount = 0;
+ gf_boolean_t chk_range = _gf_false;
+ struct list_head *pcontend = NULL;
+ struct list_head released;
+ struct list_head contend;
+ struct timespec now = {};
+
+ INIT_LIST_HEAD(&released);
+
+ priv = this->private;
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ if (clrlk_get_lock_range(args->opts, &ulock, &chk_range)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ if (args->kind & CLRLK_BLOCKED)
+ goto blkd;
+
+ if (args->kind & CLRLK_GRANTED)
+ goto granted;
blkd:
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(ilock, tmp, &dom->blocked_inodelks,
+ blocked_locks)
{
- list_for_each_entry_safe (ilock, tmp, &dom->blocked_inodelks,
- blocked_locks) {
- if (chk_range &&
- (ilock->user_flock.l_whence != ulock.l_whence
- || ilock->user_flock.l_start != ulock.l_start
- || ilock->user_flock.l_len != ulock.l_len))
- continue;
-
- bcount++;
- list_del_init (&ilock->blocked_locks);
- list_add (&ilock->blocked_locks, &released);
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
-
- list_for_each_entry_safe (ilock, tmp, &released, blocked_locks) {
- list_del_init (&ilock->blocked_locks);
- pl_trace_out (this, ilock->frame, NULL, NULL, F_SETLKW,
- &ilock->user_flock, -1, EAGAIN,
- ilock->volume);
- STACK_UNWIND_STRICT (inodelk, ilock->frame, -1,
- EAGAIN, NULL);
- //No need to take lock as the locks are only in one list
- __pl_inodelk_unref (ilock);
+ if (chk_range && (ilock->user_flock.l_whence != ulock.l_whence ||
+ ilock->user_flock.l_start != ulock.l_start ||
+ ilock->user_flock.l_len != ulock.l_len))
+ continue;
+
+ bcount++;
+ list_del_init(&ilock->client_list);
+ list_del_init(&ilock->blocked_locks);
+ list_add(&ilock->blocked_locks, &released);
}
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- if (!(args->kind & CLRLK_GRANTED)) {
- ret = 0;
- goto out;
+ if (!list_empty(&released)) {
+ list_for_each_entry_safe(ilock, tmp, &released, blocked_locks)
+ {
+ list_del_init(&ilock->blocked_locks);
+ pl_trace_out(this, ilock->frame, NULL, NULL, F_SETLKW,
+ &ilock->user_flock, -1, EAGAIN, ilock->volume);
+ STACK_UNWIND_STRICT(inodelk, ilock->frame, -1, EAGAIN, NULL);
+ // No need to take lock as the locks are only in one list
+ __pl_inodelk_unref(ilock);
}
+ }
+
+ if (!(args->kind & CLRLK_GRANTED)) {
+ ret = 0;
+ goto out;
+ }
granted:
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(ilock, tmp, &dom->inodelk_list, list)
{
- list_for_each_entry_safe (ilock, tmp, &dom->inodelk_list,
- list) {
- if (chk_range &&
- (ilock->user_flock.l_whence != ulock.l_whence
- || ilock->user_flock.l_start != ulock.l_start
- || ilock->user_flock.l_len != ulock.l_len))
- continue;
-
- gcount++;
- list_del_init (&ilock->list);
- list_add (&ilock->list, &released);
- }
+ if (chk_range && (ilock->user_flock.l_whence != ulock.l_whence ||
+ ilock->user_flock.l_start != ulock.l_start ||
+ ilock->user_flock.l_len != ulock.l_len))
+ continue;
+
+ gcount++;
+ list_del_init(&ilock->client_list);
+ list_del_init(&ilock->list);
+ list_add(&ilock->list, &released);
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- list_for_each_entry_safe (ilock, tmp, &released, list) {
- list_del_init (&ilock->list);
- //No need to take lock as the locks are only in one list
- __pl_inodelk_unref (ilock);
- }
+ list_for_each_entry_safe(ilock, tmp, &released, list)
+ {
+ list_del_init(&ilock->list);
+ // No need to take lock as the locks are only in one list
+ __pl_inodelk_unref(ilock);
+ }
- ret = 0;
+ ret = 0;
out:
- grant_blocked_inode_locks (this, pl_inode, dom);
- *blkd = bcount;
- *granted = gcount;
- return ret;
+ grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
+ if (pcontend != NULL) {
+ inodelk_contention_notify(this, pcontend);
+ }
+ *blkd = bcount;
+ *granted = gcount;
+ return ret;
}
/* Returns 0 on success and -1 on failure */
int
-clrlk_clear_entrylk (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
- clrlk_args *args, int *blkd, int *granted, int *op_errno)
+clrlk_clear_entrylk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+ clrlk_args *args, int *blkd, int *granted, int *op_errno)
{
- pl_entry_lock_t *elock = NULL;
- pl_entry_lock_t *tmp = NULL;
- int bcount = 0;
- int gcount = 0;
- int ret = -1;
- struct list_head removed;
- struct list_head released;
-
- INIT_LIST_HEAD (&released);
- if (args->kind & CLRLK_BLOCKED)
- goto blkd;
-
- if (args->kind & CLRLK_GRANTED)
- goto granted;
+ posix_locks_private_t *priv;
+ pl_entry_lock_t *elock = NULL;
+ pl_entry_lock_t *tmp = NULL;
+ int bcount = 0;
+ int gcount = 0;
+ int ret = -1;
+ struct list_head *pcontend = NULL;
+ struct list_head removed;
+ struct list_head released;
+ struct list_head contend;
+ struct timespec now;
+
+ INIT_LIST_HEAD(&released);
+
+ priv = this->private;
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ if (args->kind & CLRLK_BLOCKED)
+ goto blkd;
+
+ if (args->kind & CLRLK_GRANTED)
+ goto granted;
blkd:
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(elock, tmp, &dom->blocked_entrylks,
+ blocked_locks)
{
- list_for_each_entry_safe (elock, tmp, &dom->blocked_entrylks,
- blocked_locks) {
- if (args->opts) {
- if (!elock->basename ||
- strcmp (elock->basename, args->opts))
- continue;
- }
-
- bcount++;
-
- list_del_init (&elock->blocked_locks);
- list_add_tail (&elock->blocked_locks, &released);
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ if (args->opts) {
+ if (!elock->basename || strcmp(elock->basename, args->opts))
+ continue;
+ }
- list_for_each_entry_safe (elock, tmp, &released, blocked_locks) {
- list_del_init (&elock->blocked_locks);
- entrylk_trace_out (this, elock->frame, elock->volume, NULL, NULL,
- elock->basename, ENTRYLK_LOCK, elock->type,
- -1, EAGAIN);
- STACK_UNWIND_STRICT (entrylk, elock->frame, -1, EAGAIN, NULL);
+ bcount++;
- __pl_entrylk_unref (elock);
+ list_del_init(&elock->client_list);
+ list_del_init(&elock->blocked_locks);
+ list_add_tail(&elock->blocked_locks, &released);
}
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ if (!list_empty(&released)) {
+ list_for_each_entry_safe(elock, tmp, &released, blocked_locks)
+ {
+ list_del_init(&elock->blocked_locks);
+ entrylk_trace_out(this, elock->frame, elock->volume, NULL, NULL,
+ elock->basename, ENTRYLK_LOCK, elock->type, -1,
+ EAGAIN);
+ STACK_UNWIND_STRICT(entrylk, elock->frame, -1, EAGAIN, NULL);
- if (!(args->kind & CLRLK_GRANTED)) {
- ret = 0;
- goto out;
+ __pl_entrylk_unref(elock);
}
+ }
+
+ if (!(args->kind & CLRLK_GRANTED)) {
+ ret = 0;
+ goto out;
+ }
granted:
- INIT_LIST_HEAD (&removed);
- pthread_mutex_lock (&pl_inode->mutex);
+ INIT_LIST_HEAD(&removed);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(elock, tmp, &dom->entrylk_list, domain_list)
{
- list_for_each_entry_safe (elock, tmp, &dom->entrylk_list,
- domain_list) {
- if (args->opts) {
- if (!elock->basename ||
- strcmp (elock->basename, args->opts))
- continue;
- }
-
- gcount++;
- list_del_init (&elock->domain_list);
- list_add_tail (&elock->domain_list, &removed);
-
- __pl_entrylk_unref (elock);
- }
+ if (args->opts) {
+ if (!elock->basename || strcmp(elock->basename, args->opts))
+ continue;
+ }
+
+ gcount++;
+ list_del_init(&elock->client_list);
+ list_del_init(&elock->domain_list);
+ list_add_tail(&elock->domain_list, &removed);
+
+ __pl_entrylk_unref(elock);
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- grant_blocked_entry_locks (this, pl_inode, dom);
+ grant_blocked_entry_locks(this, pl_inode, dom, &now, pcontend);
+ if (pcontend != NULL) {
+ entrylk_contention_notify(this, pcontend);
+ }
- ret = 0;
+ ret = 0;
out:
- *blkd = bcount;
- *granted = gcount;
- return ret;
+ *blkd = bcount;
+ *granted = gcount;
+ return ret;
}
int
-clrlk_clear_lks_in_all_domains (xlator_t *this, pl_inode_t *pl_inode,
- clrlk_args *args, int *blkd, int *granted,
- int *op_errno)
+clrlk_clear_lks_in_all_domains(xlator_t *this, pl_inode_t *pl_inode,
+ clrlk_args *args, int *blkd, int *granted,
+ int *op_errno)
{
- pl_dom_list_t *dom = NULL;
- int ret = -1;
- int tmp_bcount = 0;
- int tmp_gcount = 0;
-
- if (list_empty (&pl_inode->dom_list)) {
- ret = 0;
- goto out;
- }
+ pl_dom_list_t *dom = NULL;
+ int ret = -1;
+ int tmp_bcount = 0;
+ int tmp_gcount = 0;
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
- tmp_bcount = tmp_gcount = 0;
-
- switch (args->type)
- {
- case CLRLK_INODE:
- ret = clrlk_clear_inodelk (this, pl_inode, dom, args,
- &tmp_bcount, &tmp_gcount,
- op_errno);
- if (ret)
- goto out;
- break;
- case CLRLK_ENTRY:
- ret = clrlk_clear_entrylk (this, pl_inode, dom, args,
- &tmp_bcount, &tmp_gcount,
- op_errno);
- if (ret)
- goto out;
- break;
- }
-
- *blkd += tmp_bcount;
- *granted += tmp_gcount;
+ if (list_empty(&pl_inode->dom_list)) {
+ ret = 0;
+ goto out;
+ }
+
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ tmp_bcount = tmp_gcount = 0;
+
+ switch (args->type) {
+ case CLRLK_INODE:
+ ret = clrlk_clear_inodelk(this, pl_inode, dom, args,
+ &tmp_bcount, &tmp_gcount, op_errno);
+ if (ret)
+ goto out;
+ break;
+ case CLRLK_ENTRY:
+ ret = clrlk_clear_entrylk(this, pl_inode, dom, args,
+ &tmp_bcount, &tmp_gcount, op_errno);
+ if (ret)
+ goto out;
+ break;
}
- ret = 0;
+ *blkd += tmp_bcount;
+ *granted += tmp_gcount;
+ }
+
+ ret = 0;
out:
- return ret;
+ return ret;
}
diff --git a/xlators/features/locks/src/clear.h b/xlators/features/locks/src/clear.h
index 511f3f74ae5..bc118cb1b81 100644
--- a/xlators/features/locks/src/clear.h
+++ b/xlators/features/locks/src/clear.h
@@ -10,67 +10,64 @@
#ifndef __CLEAR_H__
#define __CLEAR_H__
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "compat-errno.h"
-#include "stack.h"
-#include "call-stub.h"
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/stack.h>
+#include <glusterfs/call-stub.h>
#include "locks.h"
typedef enum {
- CLRLK_INODE,
- CLRLK_ENTRY,
- CLRLK_POSIX,
- CLRLK_TYPE_MAX
+ CLRLK_INODE,
+ CLRLK_ENTRY,
+ CLRLK_POSIX,
+ CLRLK_TYPE_MAX
} clrlk_type;
+extern const char *clrlk_type_names[];
+
typedef enum {
- CLRLK_BLOCKED = 1,
- CLRLK_GRANTED,
- CLRLK_ALL,
- CLRLK_KIND_MAX
+ CLRLK_BLOCKED = 1,
+ CLRLK_GRANTED,
+ CLRLK_ALL,
+ CLRLK_KIND_MAX
} clrlk_kind;
typedef enum {
- KW_TYPE,
- KW_KIND,
- /*add new keywords here*/
- KW_MAX
+ KW_TYPE,
+ KW_KIND,
+ /*add new keywords here*/
+ KW_MAX
} clrlk_opts;
struct _clrlk_args;
typedef struct _clrlk_args clrlk_args;
struct _clrlk_args {
- int type;
- int kind;
- char *opts;
+ int type;
+ int kind;
+ char *opts;
};
int
-clrlk_get__kind (char *kind);
+clrlk_get__kind(char *kind);
int
-clrlk_get_type (char *type);
+clrlk_get_type(char *type);
int
-clrlk_get_lock_range (char *range_str, struct gf_flock *ulock,
- gf_boolean_t *chk_range);
+clrlk_get_lock_range(char *range_str, struct gf_flock *ulock,
+ gf_boolean_t *chk_range);
int
-clrlk_parse_args (const char* cmd, clrlk_args *args);
+clrlk_parse_args(const char *cmd, clrlk_args *args);
int
-clrlk_clear_posixlk (xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
- int *blkd, int *granted, int *op_errno);
+clrlk_clear_posixlk(xlator_t *this, pl_inode_t *pl_inode, clrlk_args *args,
+ int *blkd, int *granted, int *op_errno);
int
-clrlk_clear_inodelk (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
- clrlk_args *args, int *blkd, int *granted, int *op_errno);
+clrlk_clear_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+ clrlk_args *args, int *blkd, int *granted, int *op_errno);
int
-clrlk_clear_entrylk (xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
- clrlk_args *args, int *blkd, int *granted, int *op_errno);
+clrlk_clear_entrylk(xlator_t *this, pl_inode_t *pl_inode, pl_dom_list_t *dom,
+ clrlk_args *args, int *blkd, int *granted, int *op_errno);
int
-clrlk_clear_lks_in_all_domains (xlator_t *this, pl_inode_t *pl_inode,
- clrlk_args *args, int *blkd, int *granted,
- int *op_errno);
+clrlk_clear_lks_in_all_domains(xlator_t *this, pl_inode_t *pl_inode,
+ clrlk_args *args, int *blkd, int *granted,
+ int *op_errno);
#endif /* __CLEAR_H__ */
diff --git a/xlators/features/locks/src/common.c b/xlators/features/locks/src/common.c
index f6c71c1cf86..a2c6be93e03 100644
--- a/xlators/features/locks/src/common.c
+++ b/xlators/features/locks/src/common.c
@@ -1,5 +1,5 @@
/*
- Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+ Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
This file is licensed to you under your choice of the GNU Lesser
@@ -12,737 +12,775 @@
#include <limits.h>
#include <pthread.h>
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/syncop.h>
#include "locks.h"
#include "common.h"
-
static int
-__is_lock_grantable (pl_inode_t *pl_inode, posix_lock_t *lock);
+__is_lock_grantable(pl_inode_t *pl_inode, posix_lock_t *lock);
static void
-__insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock);
+__insert_and_merge(pl_inode_t *pl_inode, posix_lock_t *lock);
static int
-pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
- posix_lock_t *old_lock);
+pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode,
+ posix_lock_t *old_lock);
static pl_dom_list_t *
-__allocate_domain (const char *volume)
+__allocate_domain(const char *volume)
{
- pl_dom_list_t *dom = NULL;
+ pl_dom_list_t *dom = NULL;
- dom = GF_CALLOC (1, sizeof (*dom),
- gf_locks_mt_pl_dom_list_t);
- if (!dom)
- goto out;
+ dom = GF_CALLOC(1, sizeof(*dom), gf_locks_mt_pl_dom_list_t);
+ if (!dom)
+ goto out;
- dom->domain = gf_strdup(volume);
- if (!dom->domain)
- goto out;
+ dom->domain = gf_strdup(volume);
+ if (!dom->domain)
+ goto out;
- gf_log ("posix-locks", GF_LOG_TRACE,
- "New domain allocated: %s", dom->domain);
+ gf_log("posix-locks", GF_LOG_TRACE, "New domain allocated: %s",
+ dom->domain);
- INIT_LIST_HEAD (&dom->inode_list);
- INIT_LIST_HEAD (&dom->entrylk_list);
- INIT_LIST_HEAD (&dom->blocked_entrylks);
- INIT_LIST_HEAD (&dom->inodelk_list);
- INIT_LIST_HEAD (&dom->blocked_inodelks);
+ INIT_LIST_HEAD(&dom->inode_list);
+ INIT_LIST_HEAD(&dom->entrylk_list);
+ INIT_LIST_HEAD(&dom->blocked_entrylks);
+ INIT_LIST_HEAD(&dom->inodelk_list);
+ INIT_LIST_HEAD(&dom->blocked_inodelks);
out:
- if (dom && (NULL == dom->domain)) {
- GF_FREE (dom);
- dom = NULL;
- }
+ if (dom && (NULL == dom->domain)) {
+ GF_FREE(dom);
+ dom = NULL;
+ }
- return dom;
+ return dom;
}
/* Returns domain for the lock. If domain is not present,
* allocates a domain and returns it
*/
pl_dom_list_t *
-get_domain (pl_inode_t *pl_inode, const char *volume)
+get_domain(pl_inode_t *pl_inode, const char *volume)
{
- pl_dom_list_t *dom = NULL;
+ pl_dom_list_t *dom = NULL;
- GF_VALIDATE_OR_GOTO ("posix-locks", pl_inode, out);
- GF_VALIDATE_OR_GOTO ("posix-locks", volume, out);
+ GF_VALIDATE_OR_GOTO("posix-locks", pl_inode, out);
+ GF_VALIDATE_OR_GOTO("posix-locks", volume, out);
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
{
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
- if (strcmp (dom->domain, volume) == 0)
- goto unlock;
- }
-
- dom = __allocate_domain (volume);
- if (dom)
- list_add (&dom->inode_list, &pl_inode->dom_list);
+ if (strcmp(dom->domain, volume) == 0)
+ goto unlock;
}
+
+ dom = __allocate_domain(volume);
+ if (dom)
+ list_add(&dom->inode_list, &pl_inode->dom_list);
+ }
unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
- if (dom) {
- gf_log ("posix-locks", GF_LOG_TRACE, "Domain %s found", volume);
- } else {
- gf_log ("posix-locks", GF_LOG_TRACE, "Domain %s not found", volume);
- }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ if (dom) {
+ gf_log("posix-locks", GF_LOG_TRACE, "Domain %s found", volume);
+ } else {
+ gf_log("posix-locks", GF_LOG_TRACE, "Domain %s not found", volume);
+ }
out:
- return dom;
+ return dom;
}
unsigned long
-fd_to_fdnum (fd_t *fd)
+fd_to_fdnum(fd_t *fd)
{
- return ((unsigned long) fd);
+ return ((unsigned long)fd);
}
fd_t *
-fd_from_fdnum (posix_lock_t *lock)
+fd_from_fdnum(posix_lock_t *lock)
{
- return ((fd_t *) lock->fd_num);
+ return ((fd_t *)lock->fd_num);
}
int
-__pl_inode_is_empty (pl_inode_t *pl_inode)
+__pl_inode_is_empty(pl_inode_t *pl_inode)
{
- pl_dom_list_t *dom = NULL;
- int is_empty = 1;
-
- if (!list_empty (&pl_inode->ext_list))
- is_empty = 0;
-
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
- if (!list_empty (&dom->entrylk_list))
- is_empty = 0;
-
- if (!list_empty (&dom->inodelk_list))
- is_empty = 0;
- }
-
- return is_empty;
+ return (list_empty(&pl_inode->ext_list));
}
void
-pl_print_locker (char *str, int size, xlator_t *this, call_frame_t *frame)
+pl_print_locker(char *str, int size, xlator_t *this, call_frame_t *frame)
{
- snprintf (str, size, "Pid=%llu, lk-owner=%s, Client=%p, Frame=%llu",
- (unsigned long long) frame->root->pid,
- lkowner_utoa (&frame->root->lk_owner),
- frame->root->client,
- (unsigned long long) frame->root->unique);
+ snprintf(str, size, "Pid=%llu, lk-owner=%s, Client=%p, Frame=%llu",
+ (unsigned long long)frame->root->pid,
+ lkowner_utoa(&frame->root->lk_owner), frame->root->client,
+ (unsigned long long)frame->root->unique);
}
-
void
-pl_print_lockee (char *str, int size, fd_t *fd, loc_t *loc)
+pl_print_lockee(char *str, int size, fd_t *fd, loc_t *loc)
{
- inode_t *inode = NULL;
- char *ipath = NULL;
- int ret = 0;
+ inode_t *inode = NULL;
+ char *ipath = NULL;
+ int ret = 0;
- if (fd)
- inode = fd->inode;
- if (loc)
- inode = loc->inode;
+ if (fd)
+ inode = fd->inode;
+ if (loc)
+ inode = loc->inode;
- if (!inode) {
- snprintf (str, size, "<nul>");
- return;
- }
+ if (!inode) {
+ snprintf(str, size, "<nul>");
+ return;
+ }
- if (loc && loc->path) {
- ipath = gf_strdup (loc->path);
- } else {
- ret = inode_path (inode, NULL, &ipath);
- if (ret <= 0)
- ipath = NULL;
- }
+ if (loc && loc->path) {
+ ipath = gf_strdup(loc->path);
+ } else {
+ ret = inode_path(inode, NULL, &ipath);
+ if (ret <= 0)
+ ipath = NULL;
+ }
- snprintf (str, size, "gfid=%s, fd=%p, path=%s",
- uuid_utoa (inode->gfid), fd,
- ipath ? ipath : "<nul>");
+ snprintf(str, size, "gfid=%s, fd=%p, path=%s", uuid_utoa(inode->gfid), fd,
+ ipath ? ipath : "<nul>");
- GF_FREE (ipath);
+ GF_FREE(ipath);
}
-
void
-pl_print_lock (char *str, int size, int cmd,
- struct gf_flock *flock, gf_lkowner_t *owner)
+pl_print_lock(char *str, int size, int cmd, struct gf_flock *flock,
+ gf_lkowner_t *owner)
{
- char *cmd_str = NULL;
- char *type_str = NULL;
+ char *cmd_str = NULL;
+ char *type_str = NULL;
- switch (cmd) {
+ switch (cmd) {
#if F_GETLK != F_GETLK64
case F_GETLK64:
#endif
case F_GETLK:
- cmd_str = "GETLK";
- break;
+ cmd_str = "GETLK";
+ break;
#if F_SETLK != F_SETLK64
case F_SETLK64:
#endif
case F_SETLK:
- cmd_str = "SETLK";
- break;
+ cmd_str = "SETLK";
+ break;
#if F_SETLKW != F_SETLKW64
case F_SETLKW64:
#endif
case F_SETLKW:
- cmd_str = "SETLKW";
- break;
+ cmd_str = "SETLKW";
+ break;
default:
- cmd_str = "UNKNOWN";
- break;
- }
+ cmd_str = "UNKNOWN";
+ break;
+ }
- switch (flock->l_type) {
+ switch (flock->l_type) {
case F_RDLCK:
- type_str = "READ";
- break;
+ type_str = "READ";
+ break;
case F_WRLCK:
- type_str = "WRITE";
- break;
+ type_str = "WRITE";
+ break;
case F_UNLCK:
- type_str = "UNLOCK";
- break;
+ type_str = "UNLOCK";
+ break;
default:
- type_str = "UNKNOWN";
- break;
- }
-
- snprintf (str, size, "lock=FCNTL, cmd=%s, type=%s, "
- "start=%llu, len=%llu, pid=%llu, lk-owner=%s",
- cmd_str, type_str, (unsigned long long) flock->l_start,
- (unsigned long long) flock->l_len,
- (unsigned long long) flock->l_pid,
- lkowner_utoa (owner));
+ type_str = "UNKNOWN";
+ break;
+ }
+
+ snprintf(str, size,
+ "lock=FCNTL, cmd=%s, type=%s, "
+ "start=%llu, len=%llu, pid=%llu, lk-owner=%s",
+ cmd_str, type_str, (unsigned long long)flock->l_start,
+ (unsigned long long)flock->l_len, (unsigned long long)flock->l_pid,
+ lkowner_utoa(owner));
}
-
void
-pl_trace_in (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, const char *domain)
+pl_trace_in(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+ struct gf_flock *flock, const char *domain)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_lock[256];
-
- priv = this->private;
+ posix_locks_private_t *priv = this->private;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_lock[256];
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- if (domain)
- pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
- else
- pl_print_lock (pl_lock, 256, cmd, flock, &frame->root->lk_owner);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ if (domain)
+ pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+ else
+ pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
- gf_log (this->name, GF_LOG_INFO,
- "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}",
- pl_locker, pl_lockee, pl_lock);
+ gf_log(this->name, GF_LOG_INFO,
+ "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+ pl_lockee, pl_lock);
}
-
void
-pl_print_verdict (char *str, int size, int op_ret, int op_errno)
+pl_print_verdict(char *str, int size, int op_ret, int op_errno)
{
- char *verdict = NULL;
-
- if (op_ret == 0) {
- verdict = "GRANTED";
- } else {
- switch (op_errno) {
- case EAGAIN:
- verdict = "TRYAGAIN";
- break;
- default:
- verdict = strerror (op_errno);
- }
+ char *verdict = NULL;
+
+ if (op_ret == 0) {
+ verdict = "GRANTED";
+ } else {
+ switch (op_errno) {
+ case EAGAIN:
+ verdict = "TRYAGAIN";
+ break;
+ default:
+ verdict = strerror(op_errno);
}
+ }
- snprintf (str, size, "%s", verdict);
+ snprintf(str, size, "%s", verdict);
}
-
void
-pl_trace_out (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, int op_ret, int op_errno, const char *domain)
+pl_trace_out(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+ struct gf_flock *flock, int op_ret, int op_errno,
+ const char *domain)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_lock[256];
- char verdict[32];
+ posix_locks_private_t *priv = NULL;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_lock[256];
+ char verdict[32];
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- if (domain)
- pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
- else
- pl_print_lock (pl_lock, 256, cmd, flock, &frame->root->lk_owner);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ if (domain)
+ pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+ else
+ pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
- pl_print_verdict (verdict, 32, op_ret, op_errno);
+ pl_print_verdict(verdict, 32, op_ret, op_errno);
- gf_log (this->name, GF_LOG_INFO,
- "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}",
- verdict, pl_locker, pl_lockee, pl_lock);
+ gf_log(this->name, GF_LOG_INFO,
+ "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}", verdict, pl_locker,
+ pl_lockee, pl_lock);
}
-
void
-pl_trace_block (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, const char *domain)
+pl_trace_block(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
+ int cmd, struct gf_flock *flock, const char *domain)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_lock[256];
-
- priv = this->private;
+ posix_locks_private_t *priv = this->private;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_lock[256];
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- if (domain)
- pl_print_inodelk (pl_lock, 256, cmd, flock, domain);
- else
- pl_print_lock (pl_lock, 256, cmd, flock, &frame->root->lk_owner);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ if (domain)
+ pl_print_inodelk(pl_lock, 256, cmd, flock, domain);
+ else
+ pl_print_lock(pl_lock, 256, cmd, flock, &frame->root->lk_owner);
- gf_log (this->name, GF_LOG_INFO,
- "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}",
- pl_locker, pl_lockee, pl_lock);
+ gf_log(this->name, GF_LOG_INFO,
+ "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+ pl_lockee, pl_lock);
}
-
void
-pl_trace_flush (xlator_t *this, call_frame_t *frame, fd_t *fd)
+pl_trace_flush(xlator_t *this, call_frame_t *frame, fd_t *fd)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- pl_inode_t *pl_inode = NULL;
+ posix_locks_private_t *priv = NULL;
+ char pl_locker[256];
+ char pl_lockee[256];
+ pl_inode_t *pl_inode = NULL;
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_inode = pl_inode_get (this, fd->inode);
+ pl_inode = pl_inode_get(this, fd->inode, NULL);
- if (pl_inode && __pl_inode_is_empty (pl_inode))
- return;
+ if (pl_inode && __pl_inode_is_empty(pl_inode))
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, NULL);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, NULL);
- gf_log (this->name, GF_LOG_INFO,
- "[FLUSH] Locker = {%s} Lockee = {%s}",
- pl_locker, pl_lockee);
+ gf_log(this->name, GF_LOG_INFO, "[FLUSH] Locker = {%s} Lockee = {%s}",
+ pl_locker, pl_lockee);
}
void
-pl_trace_release (xlator_t *this, fd_t *fd)
+pl_trace_release(xlator_t *this, fd_t *fd)
{
- posix_locks_private_t *priv = NULL;
- char pl_lockee[256];
+ posix_locks_private_t *priv = NULL;
+ char pl_lockee[256];
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_lockee (pl_lockee, 256, fd, NULL);
+ pl_print_lockee(pl_lockee, 256, fd, NULL);
- gf_log (this->name, GF_LOG_INFO,
- "[RELEASE] Lockee = {%s}", pl_lockee);
+ gf_log(this->name, GF_LOG_INFO, "[RELEASE] Lockee = {%s}", pl_lockee);
}
-
void
-pl_update_refkeeper (xlator_t *this, inode_t *inode)
+pl_update_refkeeper(xlator_t *this, inode_t *inode)
{
- pl_inode_t *pl_inode = NULL;
- int is_empty = 0;
- int need_unref = 0;
- int need_ref = 0;
+ pl_inode_t *pl_inode = NULL;
+ int is_empty = 0;
+ int need_unref = 0;
+ int need_ref = 0;
- pl_inode = pl_inode_get (this, inode);
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (!pl_inode)
+ return;
- pthread_mutex_lock (&pl_inode->mutex);
- {
- is_empty = __pl_inode_is_empty (pl_inode);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ is_empty = __pl_inode_is_empty(pl_inode);
- if (is_empty && pl_inode->refkeeper) {
- need_unref = 1;
- pl_inode->refkeeper = NULL;
- }
+ if (is_empty && pl_inode->refkeeper) {
+ need_unref = 1;
+ pl_inode->refkeeper = NULL;
+ }
- if (!is_empty && !pl_inode->refkeeper) {
- need_ref = 1;
- pl_inode->refkeeper = inode;
- }
+ if (!is_empty && !pl_inode->refkeeper) {
+ need_ref = 1;
+ pl_inode->refkeeper = inode;
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- if (need_unref)
- inode_unref (inode);
+ if (need_unref)
+ inode_unref(inode);
- if (need_ref)
- inode_ref (inode);
+ if (need_ref)
+ inode_ref(inode);
}
-
-pl_inode_t *
-pl_inode_get (xlator_t *this, inode_t *inode)
+/* Get lock enforcement info from disk */
+int
+pl_fetch_mlock_info_from_disk(xlator_t *this, pl_inode_t *pl_inode,
+ pl_local_t *local)
{
- uint64_t tmp_pl_inode = 0;
- pl_inode_t *pl_inode = NULL;
- int ret = 0;
-
- LOCK (&inode->lock);
- {
- ret = __inode_ctx_get (inode, this, &tmp_pl_inode);
- if (ret == 0) {
- pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
- goto unlock;
- }
- pl_inode = GF_CALLOC (1, sizeof (*pl_inode),
- gf_locks_mt_pl_inode_t);
- if (!pl_inode) {
- goto unlock;
- }
+ dict_t *xdata_rsp = NULL;
+ int ret = 0;
+ int op_ret = 0;
+
+ if (!local) {
+ return -1;
+ }
+
+ if (local->fd) {
+ op_ret = syncop_fgetxattr(this, local->fd, &xdata_rsp,
+ GF_ENFORCE_MANDATORY_LOCK, NULL, NULL);
+ } else {
+ op_ret = syncop_getxattr(this, &local->loc[0], &xdata_rsp,
+ GF_ENFORCE_MANDATORY_LOCK, NULL, NULL);
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (op_ret >= 0) {
+ pl_inode->mlock_enforced = _gf_true;
+ pl_inode->check_mlock_info = _gf_false;
+ } else {
+ gf_msg(this->name, GF_LOG_WARNING, -op_ret, 0,
+ "getxattr failed with %d", op_ret);
+ pl_inode->mlock_enforced = _gf_false;
+
+ if (-op_ret == ENODATA) {
+ pl_inode->check_mlock_info = _gf_false;
+ } else {
+ pl_inode->check_mlock_info = _gf_true;
+ }
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- gf_log (this->name, GF_LOG_TRACE,
- "Allocating new pl inode");
+ return ret;
+}
- pthread_mutex_init (&pl_inode->mutex, NULL);
+pl_inode_t *
+pl_inode_get(xlator_t *this, inode_t *inode, pl_local_t *local)
+{
+ uint64_t tmp_pl_inode = 0;
+ pl_inode_t *pl_inode = NULL;
+ int ret = 0;
+
+ LOCK(&inode->lock);
+ {
+ ret = __inode_ctx_get(inode, this, &tmp_pl_inode);
+ if (ret == 0) {
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+ goto unlock;
+ }
- INIT_LIST_HEAD (&pl_inode->dom_list);
- INIT_LIST_HEAD (&pl_inode->ext_list);
- INIT_LIST_HEAD (&pl_inode->rw_list);
- INIT_LIST_HEAD (&pl_inode->reservelk_list);
- INIT_LIST_HEAD (&pl_inode->blocked_reservelks);
- INIT_LIST_HEAD (&pl_inode->blocked_calls);
+ pl_inode = GF_CALLOC(1, sizeof(*pl_inode), gf_locks_mt_pl_inode_t);
+ if (!pl_inode) {
+ goto unlock;
+ }
- __inode_ctx_put (inode, this, (uint64_t)(long)(pl_inode));
+ gf_log(this->name, GF_LOG_TRACE, "Allocating new pl inode");
+
+ pthread_mutex_init(&pl_inode->mutex, NULL);
+ pthread_cond_init(&pl_inode->check_fop_wind_count, 0);
+
+ INIT_LIST_HEAD(&pl_inode->dom_list);
+ INIT_LIST_HEAD(&pl_inode->ext_list);
+ INIT_LIST_HEAD(&pl_inode->rw_list);
+ INIT_LIST_HEAD(&pl_inode->reservelk_list);
+ INIT_LIST_HEAD(&pl_inode->blocked_reservelks);
+ INIT_LIST_HEAD(&pl_inode->blocked_calls);
+ INIT_LIST_HEAD(&pl_inode->metalk_list);
+ INIT_LIST_HEAD(&pl_inode->queued_locks);
+ INIT_LIST_HEAD(&pl_inode->waiting);
+ gf_uuid_copy(pl_inode->gfid, inode->gfid);
+
+ pl_inode->check_mlock_info = _gf_true;
+ pl_inode->mlock_enforced = _gf_false;
+
+ /* -2 means never looked up. -1 means something went wrong and link
+ * tracking is disabled. */
+ pl_inode->links = -2;
+
+ ret = __inode_ctx_put(inode, this, (uint64_t)(long)(pl_inode));
+ if (ret) {
+ pthread_mutex_destroy(&pl_inode->mutex);
+ GF_FREE(pl_inode);
+ pl_inode = NULL;
+ goto unlock;
}
+ }
unlock:
- UNLOCK (&inode->lock);
+ UNLOCK(&inode->lock);
- return pl_inode;
-}
+ if ((pl_inode != NULL) && pl_is_mandatory_locking_enabled(pl_inode) &&
+ pl_inode->check_mlock_info && local) {
+ /* Note: The lock enforcement information per file can be stored in the
+ attribute flag of stat(x) in posix. With that there won't be a need
+ for doing getxattr post a reboot
+ */
+ pl_fetch_mlock_info_from_disk(this, pl_inode, local);
+ }
+ return pl_inode;
+}
/* Create a new posix_lock_t */
posix_lock_t *
-new_posix_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
- gf_lkowner_t *owner, fd_t *fd)
+new_posix_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+ gf_lkowner_t *owner, fd_t *fd, uint32_t lk_flags, int blocking,
+ int32_t *op_errno)
{
- posix_lock_t *lock = NULL;
+ posix_lock_t *lock = NULL;
- GF_VALIDATE_OR_GOTO ("posix-locks", flock, out);
- GF_VALIDATE_OR_GOTO ("posix-locks", client, out);
- GF_VALIDATE_OR_GOTO ("posix-locks", fd, out);
+ GF_VALIDATE_OR_GOTO("posix-locks", flock, out);
+ GF_VALIDATE_OR_GOTO("posix-locks", client, out);
+ GF_VALIDATE_OR_GOTO("posix-locks", fd, out);
- lock = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!lock) {
- goto out;
- }
+ if (!pl_is_lk_owner_valid(owner, client)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ lock = GF_CALLOC(1, sizeof(posix_lock_t), gf_locks_mt_posix_lock_t);
+ if (!lock) {
+ *op_errno = ENOMEM;
+ goto out;
+ }
- lock->fl_start = flock->l_start;
- lock->fl_type = flock->l_type;
+ lock->fl_start = flock->l_start;
+ lock->fl_type = flock->l_type;
- if (flock->l_len == 0)
- lock->fl_end = LLONG_MAX;
- else
- lock->fl_end = flock->l_start + flock->l_len - 1;
+ if (flock->l_len == 0)
+ lock->fl_end = LLONG_MAX;
+ else
+ lock->fl_end = flock->l_start + flock->l_len - 1;
- lock->client = client;
- lock->fd_num = fd_to_fdnum (fd);
- lock->fd = fd;
- lock->client_pid = client_pid;
- lock->owner = *owner;
+ lock->client = client;
- INIT_LIST_HEAD (&lock->list);
+ lock->client_uid = gf_strdup(client->client_uid);
+ if (lock->client_uid == NULL) {
+ GF_FREE(lock);
+ lock = NULL;
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ lock->fd_num = fd_to_fdnum(fd);
+ lock->fd = fd;
+ lock->client_pid = client_pid;
+ lock->owner = *owner;
+ lock->lk_flags = lk_flags;
+
+ lock->blocking = blocking;
+ memcpy(&lock->user_flock, flock, sizeof(lock->user_flock));
+
+ INIT_LIST_HEAD(&lock->list);
out:
- return lock;
+ return lock;
}
-
/* Delete a lock from the inode's lock list */
void
-__delete_lock (pl_inode_t *pl_inode, posix_lock_t *lock)
+__delete_lock(posix_lock_t *lock)
{
- list_del_init (&lock->list);
+ list_del_init(&lock->list);
}
-
/* Destroy a posix_lock */
void
-__destroy_lock (posix_lock_t *lock)
+__destroy_lock(posix_lock_t *lock)
{
- GF_FREE (lock);
+ GF_FREE(lock->client_uid);
+ GF_FREE(lock);
}
+static posix_lock_t *
+__copy_lock(posix_lock_t *src)
+{
+ posix_lock_t *dst;
+
+ dst = GF_MALLOC(sizeof(posix_lock_t), gf_locks_mt_posix_lock_t);
+ if (dst != NULL) {
+ memcpy(dst, src, sizeof(posix_lock_t));
+ dst->client_uid = gf_strdup(src->client_uid);
+ if (dst->client_uid == NULL) {
+ GF_FREE(dst);
+ dst = NULL;
+ }
+
+ if (dst != NULL)
+ INIT_LIST_HEAD(&dst->list);
+ }
+
+ return dst;
+}
/* Convert a posix_lock to a struct gf_flock */
void
-posix_lock_to_flock (posix_lock_t *lock, struct gf_flock *flock)
+posix_lock_to_flock(posix_lock_t *lock, struct gf_flock *flock)
{
- flock->l_pid = lock->client_pid;
- flock->l_type = lock->fl_type;
- flock->l_start = lock->fl_start;
- flock->l_owner = lock->owner;
-
- if (lock->fl_end == LLONG_MAX)
- flock->l_len = 0;
- else
- flock->l_len = lock->fl_end - lock->fl_start + 1;
+ flock->l_pid = lock->user_flock.l_pid;
+ flock->l_type = lock->fl_type;
+ flock->l_start = lock->fl_start;
+ flock->l_owner = lock->owner;
+
+ if (lock->fl_end == LLONG_MAX)
+ flock->l_len = 0;
+ else
+ flock->l_len = lock->fl_end - lock->fl_start + 1;
}
/* Insert the lock into the inode's lock list */
static void
-__insert_lock (pl_inode_t *pl_inode, posix_lock_t *lock)
+__insert_lock(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- if (lock->blocked)
- gettimeofday (&lock->blkd_time, NULL);
- else
- gettimeofday (&lock->granted_time, NULL);
+ if (lock->blocked)
+ lock->blkd_time = gf_time();
+ else
+ lock->granted_time = gf_time();
- list_add_tail (&lock->list, &pl_inode->ext_list);
-
- return;
+ list_add_tail(&lock->list, &pl_inode->ext_list);
}
-
/* Return true if the locks overlap, false otherwise */
int
-locks_overlap (posix_lock_t *l1, posix_lock_t *l2)
+locks_overlap(posix_lock_t *l1, posix_lock_t *l2)
{
- /*
- Note:
- FUSE always gives us absolute offsets, so no need to worry
- about SEEK_CUR or SEEK_END
- */
+ /*
+ Note:
+ FUSE always gives us absolute offsets, so no need to worry
+ about SEEK_CUR or SEEK_END
+ */
- return ((l1->fl_end >= l2->fl_start) &&
- (l2->fl_end >= l1->fl_start));
+ return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
}
-
/* Return true if the locks have the same owner */
int
-same_owner (posix_lock_t *l1, posix_lock_t *l2)
+same_owner(posix_lock_t *l1, posix_lock_t *l2)
{
-
- return (is_same_lkowner (&l1->owner, &l2->owner) &&
- (l1->client == l2->client));
-
+ return (is_same_lkowner(&l1->owner, &l2->owner) &&
+ (l1->client == l2->client));
}
-
/* Delete all F_UNLCK locks */
void
-__delete_unlck_locks (pl_inode_t *pl_inode)
+__delete_unlck_locks(pl_inode_t *pl_inode)
{
- posix_lock_t *l = NULL;
- posix_lock_t *tmp = NULL;
-
- list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
- if (l->fl_type == F_UNLCK) {
- __delete_lock (pl_inode, l);
- __destroy_lock (l);
- }
+ posix_lock_t *l = NULL;
+ posix_lock_t *tmp = NULL;
+
+ list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+ {
+ if (l->fl_type == F_UNLCK) {
+ __delete_lock(l);
+ __destroy_lock(l);
}
+ }
}
-
/* Add two locks */
static posix_lock_t *
-add_locks (posix_lock_t *l1, posix_lock_t *l2)
+add_locks(posix_lock_t *l1, posix_lock_t *l2, posix_lock_t *dst)
{
- posix_lock_t *sum = NULL;
+ posix_lock_t *sum = NULL;
- sum = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!sum)
- return NULL;
+ sum = __copy_lock(dst);
+ if (!sum)
+ return NULL;
- sum->fl_start = min (l1->fl_start, l2->fl_start);
- sum->fl_end = max (l1->fl_end, l2->fl_end);
+ sum->fl_start = min(l1->fl_start, l2->fl_start);
+ sum->fl_end = max(l1->fl_end, l2->fl_end);
- return sum;
+ posix_lock_to_flock(sum, &sum->user_flock);
+
+ return sum;
}
/* Subtract two locks */
struct _values {
- posix_lock_t *locks[3];
+ posix_lock_t *locks[3];
};
/* {big} must always be contained inside {small} */
static struct _values
-subtract_locks (posix_lock_t *big, posix_lock_t *small)
+subtract_locks(posix_lock_t *big, posix_lock_t *small)
{
+ struct _values v = {.locks = {0, 0, 0}};
- struct _values v = { .locks = {0, 0, 0} };
-
- if ((big->fl_start == small->fl_start) &&
- (big->fl_end == small->fl_end)) {
- /* both edges coincide with big */
- v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[0])
- goto out;
- memcpy (v.locks[0], big, sizeof (posix_lock_t));
- v.locks[0]->fl_type = small->fl_type;
- goto done;
+ if ((big->fl_start == small->fl_start) && (big->fl_end == small->fl_end)) {
+ /* both edges coincide with big */
+ v.locks[0] = __copy_lock(big);
+ if (!v.locks[0]) {
+ goto out;
}
- if ((small->fl_start > big->fl_start) &&
- (small->fl_end < big->fl_end)) {
- /* both edges lie inside big */
- v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[0])
- goto out;
-
- v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[1])
- goto out;
-
- v.locks[2] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[1])
- goto out;
-
- memcpy (v.locks[0], big, sizeof (posix_lock_t));
- v.locks[0]->fl_end = small->fl_start - 1;
-
- memcpy (v.locks[1], small, sizeof (posix_lock_t));
-
- memcpy (v.locks[2], big, sizeof (posix_lock_t));
- v.locks[2]->fl_start = small->fl_end + 1;
- goto done;
-
+ v.locks[0]->fl_type = small->fl_type;
+ v.locks[0]->user_flock.l_type = small->fl_type;
+ goto done;
+ }
+
+ if ((small->fl_start > big->fl_start) && (small->fl_end < big->fl_end)) {
+ /* both edges lie inside big */
+ v.locks[0] = __copy_lock(big);
+ v.locks[1] = __copy_lock(small);
+ v.locks[2] = __copy_lock(big);
+ if ((v.locks[0] == NULL) || (v.locks[1] == NULL) ||
+ (v.locks[2] == NULL)) {
+ goto out;
}
- /* one edge coincides with big */
- if (small->fl_start == big->fl_start) {
- v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[0])
- goto out;
-
- v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[1])
- goto out;
-
- memcpy (v.locks[0], big, sizeof (posix_lock_t));
- v.locks[0]->fl_start = small->fl_end + 1;
-
- memcpy (v.locks[1], small, sizeof (posix_lock_t));
- goto done;
+ v.locks[0]->fl_end = small->fl_start - 1;
+ v.locks[2]->fl_start = small->fl_end + 1;
+ posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+ posix_lock_to_flock(v.locks[2], &v.locks[2]->user_flock);
+ goto done;
+ }
+
+ /* one edge coincides with big */
+ if (small->fl_start == big->fl_start) {
+ v.locks[0] = __copy_lock(big);
+ v.locks[1] = __copy_lock(small);
+ if ((v.locks[0] == NULL) || (v.locks[1] == NULL)) {
+ goto out;
}
- if (small->fl_end == big->fl_end) {
- v.locks[0] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[0])
- goto out;
-
- v.locks[1] = GF_CALLOC (1, sizeof (posix_lock_t),
- gf_locks_mt_posix_lock_t);
- if (!v.locks[1])
- goto out;
-
- memcpy (v.locks[0], big, sizeof (posix_lock_t));
- v.locks[0]->fl_end = small->fl_start - 1;
+ v.locks[0]->fl_start = small->fl_end + 1;
+ posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+ goto done;
+ }
- memcpy (v.locks[1], small, sizeof (posix_lock_t));
- goto done;
+ if (small->fl_end == big->fl_end) {
+ v.locks[0] = __copy_lock(big);
+ v.locks[1] = __copy_lock(small);
+ if ((v.locks[0] == NULL) || (v.locks[1] == NULL)) {
+ goto out;
}
- GF_ASSERT (0);
- gf_log ("posix-locks", GF_LOG_ERROR, "Unexpected case in subtract_locks");
+ v.locks[0]->fl_end = small->fl_start - 1;
+ posix_lock_to_flock(v.locks[0], &v.locks[0]->user_flock);
+ goto done;
+ }
+
+ GF_ASSERT(0);
+ gf_log("posix-locks", GF_LOG_ERROR, "Unexpected case in subtract_locks");
out:
- if (v.locks[0]) {
- GF_FREE (v.locks[0]);
- v.locks[0] = NULL;
- }
- if (v.locks[1]) {
- GF_FREE (v.locks[1]);
- v.locks[1] = NULL;
- }
- if (v.locks[2]) {
- GF_FREE (v.locks[2]);
- v.locks[2] = NULL;
- }
+ if (v.locks[0]) {
+ __destroy_lock(v.locks[0]);
+ v.locks[0] = NULL;
+ }
+ if (v.locks[1]) {
+ __destroy_lock(v.locks[1]);
+ v.locks[1] = NULL;
+ }
+ if (v.locks[2]) {
+ __destroy_lock(v.locks[2]);
+ v.locks[2] = NULL;
+ }
done:
- return v;
+ return v;
}
static posix_lock_t *
-first_conflicting_overlap (pl_inode_t *pl_inode, posix_lock_t *lock)
+first_conflicting_overlap(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *l = NULL;
- posix_lock_t *conf = NULL;
+ posix_lock_t *l = NULL;
+ posix_lock_t *conf = NULL;
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry(l, &pl_inode->ext_list, list)
{
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->blocked)
- continue;
-
- if (locks_overlap (l, lock)) {
- if (same_owner (l, lock))
- continue;
-
- if ((l->fl_type == F_WRLCK) ||
- (lock->fl_type == F_WRLCK)) {
- conf = l;
- goto unlock;
- }
- }
+ if (l->blocked)
+ continue;
+
+ if (locks_overlap(l, lock)) {
+ if (same_owner(l, lock))
+ continue;
+
+ if ((l->fl_type == F_WRLCK) || (lock->fl_type == F_WRLCK)) {
+ conf = l;
+ goto unlock;
}
+ }
}
+ }
unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
+ pthread_mutex_unlock(&pl_inode->mutex);
- return conf;
+ return conf;
}
/*
@@ -751,351 +789,803 @@ unlock:
If {begin} is NULL, then start from the beginning of the list
*/
static posix_lock_t *
-first_overlap (pl_inode_t *pl_inode, posix_lock_t *lock)
+first_overlap(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *l = NULL;
+ posix_lock_t *l = NULL;
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->blocked)
- continue;
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (l->blocked)
+ continue;
- if (locks_overlap (l, lock))
- return l;
- }
+ if (locks_overlap(l, lock))
+ return l;
+ }
- return NULL;
+ return NULL;
}
-
-
/* Return true if lock is grantable */
static int
-__is_lock_grantable (pl_inode_t *pl_inode, posix_lock_t *lock)
+__is_lock_grantable(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *l = NULL;
- int ret = 1;
-
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (!l->blocked && locks_overlap (lock, l)) {
- if (((l->fl_type == F_WRLCK)
- || (lock->fl_type == F_WRLCK))
- && (lock->fl_type != F_UNLCK)
- && !same_owner (l, lock)) {
- ret = 0;
- break;
- }
- }
+ posix_lock_t *l = NULL;
+ int ret = 1;
+
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (!l->blocked && locks_overlap(lock, l)) {
+ if (((l->fl_type == F_WRLCK) || (lock->fl_type == F_WRLCK)) &&
+ (lock->fl_type != F_UNLCK) && !same_owner(l, lock)) {
+ ret = 0;
+ break;
+ }
}
- return ret;
+ }
+ return ret;
}
-
-extern void do_blocked_rw (pl_inode_t *);
-
+extern void
+do_blocked_rw(pl_inode_t *);
static void
-__insert_and_merge (pl_inode_t *pl_inode, posix_lock_t *lock)
+__insert_and_merge(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *conf = NULL;
- posix_lock_t *t = NULL;
- posix_lock_t *sum = NULL;
- int i = 0;
- struct _values v = { .locks = {0, 0, 0} };
-
- list_for_each_entry_safe (conf, t, &pl_inode->ext_list, list) {
- if (conf->blocked)
- continue;
- if (!locks_overlap (conf, lock))
- continue;
+ posix_lock_t *conf = NULL;
+ posix_lock_t *t = NULL;
+ posix_lock_t *sum = NULL;
+ int i = 0;
+ struct _values v = {.locks = {0, 0, 0}};
+
+ list_for_each_entry_safe(conf, t, &pl_inode->ext_list, list)
+ {
+ if (conf->blocked)
+ continue;
+ if (!locks_overlap(conf, lock))
+ continue;
+
+ if (same_owner(conf, lock)) {
+ if (conf->fl_type == lock->fl_type &&
+ conf->lk_flags == lock->lk_flags) {
+ sum = add_locks(lock, conf, lock);
+
+ __delete_lock(conf);
+ __destroy_lock(conf);
+
+ __destroy_lock(lock);
+ INIT_LIST_HEAD(&sum->list);
+ posix_lock_to_flock(sum, &sum->user_flock);
+ __insert_and_merge(pl_inode, sum);
- if (same_owner (conf, lock)) {
- if (conf->fl_type == lock->fl_type) {
- sum = add_locks (lock, conf);
+ return;
+ } else {
+ sum = add_locks(lock, conf, conf);
- sum->fl_type = lock->fl_type;
- sum->client = lock->client;
- sum->fd_num = lock->fd_num;
- sum->client_pid = lock->client_pid;
- sum->owner = lock->owner;
+ v = subtract_locks(sum, lock);
- __delete_lock (pl_inode, conf);
- __destroy_lock (conf);
+ __delete_lock(conf);
+ __destroy_lock(conf);
- __destroy_lock (lock);
- INIT_LIST_HEAD (&sum->list);
- posix_lock_to_flock (sum, &sum->user_flock);
- __insert_and_merge (pl_inode, sum);
+ __delete_lock(lock);
+ __destroy_lock(lock);
- return;
- } else {
- sum = add_locks (lock, conf);
+ __destroy_lock(sum);
- sum->fl_type = conf->fl_type;
- sum->client = conf->client;
- sum->fd_num = conf->fd_num;
- sum->client_pid = conf->client_pid;
- sum->owner = conf->owner;
+ for (i = 0; i < 3; i++) {
+ if (!v.locks[i])
+ continue;
- v = subtract_locks (sum, lock);
+ __insert_and_merge(pl_inode, v.locks[i]);
+ }
- __delete_lock (pl_inode, conf);
- __destroy_lock (conf);
+ __delete_unlck_locks(pl_inode);
+ return;
+ }
+ }
- __delete_lock (pl_inode, lock);
- __destroy_lock (lock);
+ if (lock->fl_type == F_UNLCK) {
+ continue;
+ }
- __destroy_lock (sum);
+ if ((conf->fl_type == F_RDLCK) && (lock->fl_type == F_RDLCK)) {
+ __insert_lock(pl_inode, lock);
+ return;
+ }
+ }
+
+ /* no conflicts, so just insert */
+ if (lock->fl_type != F_UNLCK) {
+ __insert_lock(pl_inode, lock);
+ } else {
+ __destroy_lock(lock);
+ }
+}
- for (i = 0; i < 3; i++) {
- if (!v.locks[i])
- continue;
+void
+__grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted)
+{
+ struct list_head tmp_list;
+ posix_lock_t *l = NULL;
+ posix_lock_t *tmp = NULL;
+ posix_lock_t *conf = NULL;
+
+ INIT_LIST_HEAD(&tmp_list);
+
+ list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+ {
+ if (l->blocked) {
+ conf = first_overlap(pl_inode, l);
+ if (conf)
+ continue;
+
+ l->blocked = 0;
+ list_move_tail(&l->list, &tmp_list);
+ }
+ }
- INIT_LIST_HEAD (&v.locks[i]->list);
- posix_lock_to_flock (v.locks[i],
- &v.locks[i]->user_flock);
- __insert_and_merge (pl_inode,
- v.locks[i]);
- }
+ list_for_each_entry_safe(l, tmp, &tmp_list, list)
+ {
+ list_del_init(&l->list);
- __delete_unlck_locks (pl_inode);
- return;
- }
- }
+ if (__is_lock_grantable(pl_inode, l)) {
+ conf = GF_CALLOC(1, sizeof(*conf), gf_locks_mt_posix_lock_t);
- if (lock->fl_type == F_UNLCK) {
- continue;
- }
+ if (!conf) {
+ l->blocked = 1;
+ __insert_lock(pl_inode, l);
+ continue;
+ }
- if ((conf->fl_type == F_RDLCK) && (lock->fl_type == F_RDLCK)) {
- __insert_lock (pl_inode, lock);
- return;
- }
- }
+ conf->frame = l->frame;
+ l->frame = NULL;
- /* no conflicts, so just insert */
- if (lock->fl_type != F_UNLCK) {
- __insert_lock (pl_inode, lock);
+ posix_lock_to_flock(l, &conf->user_flock);
+
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64
+ " => Granted",
+ l->fl_type == F_UNLCK ? "Unlock" : "Lock", l->client_pid,
+ lkowner_utoa(&l->owner), l->user_flock.l_start,
+ l->user_flock.l_len);
+
+ __insert_and_merge(pl_inode, l);
+
+ list_add(&conf->list, granted);
} else {
- __destroy_lock (lock);
+ l->blocked = 1;
+ __insert_lock(pl_inode, l);
}
+ }
}
-
void
-__grant_blocked_locks (xlator_t *this, pl_inode_t *pl_inode, struct list_head *granted)
+grant_blocked_locks(xlator_t *this, pl_inode_t *pl_inode)
{
- struct list_head tmp_list;
- posix_lock_t *l = NULL;
- posix_lock_t *tmp = NULL;
- posix_lock_t *conf = NULL;
+ struct list_head granted_list;
+ posix_lock_t *tmp = NULL;
+ posix_lock_t *lock = NULL;
+ pl_local_t *local = NULL;
+ INIT_LIST_HEAD(&granted_list);
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __grant_blocked_locks(this, pl_inode, &granted_list);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ list_for_each_entry_safe(lock, tmp, &granted_list, list)
+ {
+ list_del_init(&lock->list);
+
+ pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+ 0, 0, NULL);
+ local = lock->frame->local;
+ PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0,
+ &lock->user_flock, NULL);
+ __destroy_lock(lock);
+ }
+
+ return;
+}
- INIT_LIST_HEAD (&tmp_list);
+static int
+pl_send_prelock_unlock(xlator_t *this, pl_inode_t *pl_inode,
+ posix_lock_t *old_lock)
+{
+ struct gf_flock flock = {
+ 0,
+ };
+ posix_lock_t *unlock_lock = NULL;
+ int32_t op_errno = 0;
- list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
- if (l->blocked) {
- conf = first_overlap (pl_inode, l);
- if (conf)
- continue;
+ struct list_head granted_list;
+ posix_lock_t *tmp = NULL;
+ posix_lock_t *lock = NULL;
+ pl_local_t *local = NULL;
- l->blocked = 0;
- list_move_tail (&l->list, &tmp_list);
- }
- }
+ int ret = -1;
- list_for_each_entry_safe (l, tmp, &tmp_list, list) {
- list_del_init (&l->list);
+ INIT_LIST_HEAD(&granted_list);
- if (__is_lock_grantable (pl_inode, l)) {
- conf = GF_CALLOC (1, sizeof (*conf),
- gf_locks_mt_posix_lock_t);
+ flock.l_type = F_UNLCK;
+ flock.l_whence = old_lock->user_flock.l_whence;
+ flock.l_start = old_lock->user_flock.l_start;
+ flock.l_len = old_lock->user_flock.l_len;
+ flock.l_pid = old_lock->user_flock.l_pid;
- if (!conf) {
- l->blocked = 1;
- __insert_lock (pl_inode, l);
- continue;
- }
+ unlock_lock = new_posix_lock(&flock, old_lock->client, old_lock->client_pid,
+ &old_lock->owner, old_lock->fd,
+ old_lock->lk_flags, 0, &op_errno);
+ GF_VALIDATE_OR_GOTO(this->name, unlock_lock, out);
+ ret = 0;
- conf->frame = l->frame;
- l->frame = NULL;
+ __insert_and_merge(pl_inode, unlock_lock);
- posix_lock_to_flock (l, &conf->user_flock);
+ __grant_blocked_locks(this, pl_inode, &granted_list);
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => Granted",
- l->fl_type == F_UNLCK ? "Unlock" : "Lock",
- l->client_pid, lkowner_utoa (&l->owner),
- l->user_flock.l_start,
- l->user_flock.l_len);
+ list_for_each_entry_safe(lock, tmp, &granted_list, list)
+ {
+ list_del_init(&lock->list);
- __insert_and_merge (pl_inode, l);
+ pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+ 0, 0, NULL);
+ local = lock->frame->local;
+ PL_STACK_UNWIND_AND_FREE(local, lk, lock->frame, 0, 0,
+ &lock->user_flock, NULL);
+ __destroy_lock(lock);
+ }
- list_add (&conf->list, granted);
- } else {
- l->blocked = 1;
- __insert_lock (pl_inode, l);
- }
- }
+out:
+ return ret;
}
+int
+pl_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ int can_block)
+{
+ int ret = 0;
-void
-grant_blocked_locks (xlator_t *this, pl_inode_t *pl_inode)
+ errno = 0;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ /* Send unlock before the actual lock to
+ prevent lock upgrade / downgrade
+ problems only if:
+ - it is a blocking call
+ - it has other conflicting locks
+ */
+
+ if (can_block && !(__is_lock_grantable(pl_inode, lock))) {
+ ret = pl_send_prelock_unlock(this, pl_inode, lock);
+ if (ret)
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Could not send pre-lock "
+ "unlock");
+ }
+
+ if (__is_lock_grantable(pl_inode, lock)) {
+ if (pl_metalock_is_active(pl_inode)) {
+ __pl_queue_lock(pl_inode, lock);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ ret = -2;
+ goto out;
+ }
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => OK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->user_flock.l_start, lock->user_flock.l_len);
+ __insert_and_merge(pl_inode, lock);
+ } else if (can_block) {
+ if (pl_metalock_is_active(pl_inode)) {
+ __pl_queue_lock(pl_inode, lock);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ ret = -2;
+ goto out;
+ }
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64
+ " => Blocked",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->user_flock.l_start, lock->user_flock.l_len);
+
+ pl_trace_block(this, lock->frame, NULL, NULL, F_SETLKW,
+ &lock->user_flock, NULL);
+
+ lock->blocked = 1;
+ __insert_lock(pl_inode, lock);
+ ret = -1;
+ } else {
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => NOK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->user_flock.l_start, lock->user_flock.l_len);
+ errno = EAGAIN;
+ ret = -1;
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ grant_blocked_locks(this, pl_inode);
+
+ do_blocked_rw(pl_inode);
+
+out:
+ return ret;
+}
+
+posix_lock_t *
+pl_getlk(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- struct list_head granted_list;
- posix_lock_t *tmp = NULL;
- posix_lock_t *lock = NULL;
+ posix_lock_t *conf = first_conflicting_overlap(pl_inode, lock);
+ if (conf == NULL) {
+ lock->fl_type = F_UNLCK;
+ return lock;
+ }
+
+ return conf;
+}
- INIT_LIST_HEAD (&granted_list);
+gf_boolean_t
+pl_does_monkey_want_stuck_lock()
+{
+ long int monkey_unlock_rand = 0;
+ long int monkey_unlock_rand_rem = 0;
+
+ /* coverity[DC.WEAK_CRYPTO] */
+ monkey_unlock_rand = random();
+ monkey_unlock_rand_rem = monkey_unlock_rand % 100;
+ if (monkey_unlock_rand_rem == 0)
+ return _gf_true;
+ return _gf_false;
+}
- pthread_mutex_lock (&pl_inode->mutex);
+int
+pl_lock_preempt(pl_inode_t *pl_inode, posix_lock_t *reqlock)
+{
+ posix_lock_t *lock = NULL;
+ posix_lock_t *i = NULL;
+ pl_rw_req_t *rw = NULL;
+ pl_rw_req_t *itr = NULL;
+ struct list_head unwind_blist = {
+ 0,
+ };
+ struct list_head unwind_rw_list = {
+ 0,
+ };
+ int ret = 0;
+
+ INIT_LIST_HEAD(&unwind_blist);
+ INIT_LIST_HEAD(&unwind_rw_list);
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ /*
+ - go through the lock list
+ - remove all locks from different owners
+ - same owner locks will be added or substracted based on
+ the new request
+ - add the new lock
+ */
+ list_for_each_entry_safe(lock, i, &pl_inode->ext_list, list)
{
- __grant_blocked_locks (this, pl_inode, &granted_list);
+ if (lock->blocked) {
+ list_del_init(&lock->list);
+ list_add(&lock->list, &unwind_blist);
+ continue;
+ }
+
+ if (locks_overlap(lock, reqlock)) {
+ if (same_owner(lock, reqlock))
+ continue;
+
+ /* remove conflicting locks */
+ list_del_init(&lock->list);
+ __delete_lock(lock);
+ __destroy_lock(lock);
+ }
}
- pthread_mutex_unlock (&pl_inode->mutex);
- list_for_each_entry_safe (lock, tmp, &granted_list, list) {
- list_del_init (&lock->list);
+ __insert_and_merge(pl_inode, reqlock);
- pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
- &lock->user_flock, 0, 0, NULL);
+ list_for_each_entry_safe(rw, itr, &pl_inode->rw_list, list)
+ {
+ list_del_init(&rw->list);
+ list_add(&rw->list, &unwind_rw_list);
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ /* unwind blocked locks */
+ list_for_each_entry_safe(lock, i, &unwind_blist, list)
+ {
+ PL_STACK_UNWIND_AND_FREE(((pl_local_t *)lock->frame->local), lk,
+ lock->frame, -1, EBUSY, &lock->user_flock,
+ NULL);
+ __destroy_lock(lock);
+ }
+
+ /* unwind blocked IOs */
+ list_for_each_entry_safe(rw, itr, &unwind_rw_list, list)
+ {
+ pl_clean_local(rw->stub->frame->local);
+ call_unwind_error(rw->stub, -1, EBUSY);
+ }
+
+ return ret;
+}
- STACK_UNWIND_STRICT (lk, lock->frame, 0, 0,
- &lock->user_flock, NULL);
+/* Return true in case we need to ensure mandatory-locking
+ * semantics under different modes.
+ */
+gf_boolean_t
+pl_is_mandatory_locking_enabled(pl_inode_t *pl_inode)
+{
+ posix_locks_private_t *priv = THIS->private;
- GF_FREE (lock);
- }
+ if (priv->mandatory_mode == MLK_FILE_BASED && pl_inode->mandatory)
+ return _gf_true;
+ else if (priv->mandatory_mode == MLK_FORCED ||
+ priv->mandatory_mode == MLK_OPTIMAL)
+ return _gf_true;
- return;
+ return _gf_false;
}
-static int
-pl_send_prelock_unlock (xlator_t *this, pl_inode_t *pl_inode,
- posix_lock_t *old_lock)
+void
+pl_clean_local(pl_local_t *local)
{
- struct gf_flock flock = {0,};
- posix_lock_t *unlock_lock = NULL;
+ if (!local)
+ return;
- struct list_head granted_list;
- posix_lock_t *tmp = NULL;
- posix_lock_t *lock = NULL;
+ if (local->inodelk_dom_count_req)
+ data_unref(local->inodelk_dom_count_req);
+ loc_wipe(&local->loc[0]);
+ loc_wipe(&local->loc[1]);
+ if (local->fd)
+ fd_unref(local->fd);
+ if (local->inode)
+ inode_unref(local->inode);
+ mem_put(local);
+}
- int ret = -1;
+/*
+TODO: detach local initialization from PL_LOCAL_GET_REQUESTS and add it here
+*/
+int
+pl_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd)
+{
+ pl_local_t *local = NULL;
+
+ if (!loc && !fd) {
+ return -1;
+ }
+
+ if (!frame->local) {
+ local = mem_get0(this->local_pool);
+ if (!local) {
+ gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
+ "mem allocation failed");
+ return -1;
+ }
- INIT_LIST_HEAD (&granted_list);
+ local->inode = (loc ? inode_ref(loc->inode) : inode_ref(fd->inode));
- flock.l_type = F_UNLCK;
- flock.l_whence = old_lock->user_flock.l_whence;
- flock.l_start = old_lock->user_flock.l_start;
- flock.l_len = old_lock->user_flock.l_len;
+ frame->local = local;
+ }
+ return 0;
+}
- unlock_lock = new_posix_lock (&flock, old_lock->client,
- old_lock->client_pid, &old_lock->owner,
- old_lock->fd);
- GF_VALIDATE_OR_GOTO (this->name, unlock_lock, out);
- ret = 0;
+gf_boolean_t
+pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client)
+{
+ if (client && (client->opversion < GD_OP_VERSION_7_0)) {
+ return _gf_true;
+ }
+
+ if (is_lk_owner_null(owner)) {
+ return _gf_false;
+ }
+ return _gf_true;
+}
- __insert_and_merge (pl_inode, unlock_lock);
+static int32_t
+pl_inode_from_loc(loc_t *loc, inode_t **pinode)
+{
+ inode_t *inode = NULL;
+ int32_t error = 0;
+
+ if (loc->inode != NULL) {
+ inode = inode_ref(loc->inode);
+ goto done;
+ }
+
+ if (loc->parent == NULL) {
+ error = EINVAL;
+ goto done;
+ }
+
+ if (!gf_uuid_is_null(loc->gfid)) {
+ inode = inode_find(loc->parent->table, loc->gfid);
+ if (inode != NULL) {
+ goto done;
+ }
+ }
- __grant_blocked_locks (this, pl_inode, &granted_list);
+ if (loc->name == NULL) {
+ error = EINVAL;
+ goto done;
+ }
- list_for_each_entry_safe (lock, tmp, &granted_list, list) {
- list_del_init (&lock->list);
+ inode = inode_grep(loc->parent->table, loc->parent, loc->name);
+ if (inode == NULL) {
+ /* We haven't found any inode. This means that the file doesn't exist
+ * or that even if it exists, we don't have any knowledge about it, so
+ * we don't have locks on it either, which is fine for our purposes. */
+ goto done;
+ }
- pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
- &lock->user_flock, 0, 0, NULL);
+done:
+ *pinode = inode;
- STACK_UNWIND_STRICT (lk, lock->frame, 0, 0,
- &lock->user_flock, NULL);
+ return error;
+}
+
+static gf_boolean_t
+pl_inode_has_owners(xlator_t *xl, client_t *client, pl_inode_t *pl_inode,
+ struct timespec *now, struct list_head *contend)
+{
+ pl_dom_list_t *dom;
+ pl_inode_lock_t *lock;
+ gf_boolean_t has_owners = _gf_false;
- GF_FREE (lock);
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ list_for_each_entry(lock, &dom->inodelk_list, list)
+ {
+ /* If the lock belongs to the same client, we assume it's related
+ * to the same operation, so we allow the removal to continue. */
+ if (lock->client == client) {
+ continue;
+ }
+ /* If the lock belongs to an internal process, we don't block the
+ * removal. */
+ if (lock->client_pid < 0) {
+ continue;
+ }
+ if (contend == NULL) {
+ return _gf_true;
+ }
+ has_owners = _gf_true;
+ inodelk_contention_notify_check(xl, lock, now, contend);
}
+ }
-out:
- return ret;
+ return has_owners;
}
-int
-pl_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
- int can_block)
+int32_t
+pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc,
+ pl_inode_t **ppl_inode, struct list_head *contend)
{
- int ret = 0;
+ struct timespec now;
+ inode_t *inode;
+ pl_inode_t *pl_inode;
+ int32_t error;
+
+ pl_inode = NULL;
+
+ error = pl_inode_from_loc(loc, &inode);
+ if ((error != 0) || (inode == NULL)) {
+ goto done;
+ }
+
+ pl_inode = pl_inode_get(xl, inode, NULL);
+ if (pl_inode == NULL) {
+ inode_unref(inode);
+ error = ENOMEM;
+ goto done;
+ }
+
+ /* pl_inode_from_loc() already increments ref count for inode, so
+ * we only assign here our reference. */
+ pl_inode->inode = inode;
+
+ timespec_now(&now);
+
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ if (pl_inode->removed) {
+ error = ESTALE;
+ goto unlock;
+ }
+
+ if (pl_inode_has_owners(xl, frame->root->client, pl_inode, &now, contend)) {
+ error = -1;
+ /* We skip the unlock here because the caller must create a stub when
+ * we return -1 and do a call to pl_inode_remove_complete(), which
+ * assumes the lock is still acquired and will release it once
+ * everything else is prepared. */
+ goto done;
+ }
+
+ pl_inode->is_locked = _gf_true;
+ pl_inode->remove_running++;
- errno = 0;
+unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
- pthread_mutex_lock (&pl_inode->mutex);
- {
- /* Send unlock before the actual lock to
- prevent lock upgrade / downgrade
- problems only if:
- - it is a blocking call
- - it has other conflicting locks
- */
-
- if (can_block &&
- !(__is_lock_grantable (pl_inode, lock))) {
- ret = pl_send_prelock_unlock (this, pl_inode,
- lock);
- if (ret)
- gf_log (this->name, GF_LOG_DEBUG,
- "Could not send pre-lock "
- "unlock");
- }
+done:
+ *ppl_inode = pl_inode;
- if (__is_lock_grantable (pl_inode, lock)) {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => OK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
- __insert_and_merge (pl_inode, lock);
- } else if (can_block) {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => Blocked",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
- lock->blocked = 1;
- __insert_lock (pl_inode, lock);
- ret = -1;
- } else {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => NOK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
- errno = EAGAIN;
- ret = -1;
- }
+ return error;
+}
+
+int32_t
+pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub,
+ struct list_head *contend)
+{
+ pl_inode_lock_t *lock;
+ int32_t error = -1;
+
+ if (stub != NULL) {
+ list_add_tail(&stub->list, &pl_inode->waiting);
+ pl_inode->is_locked = _gf_true;
+ } else {
+ error = ENOMEM;
+
+ while (!list_empty(contend)) {
+ lock = list_first_entry(contend, pl_inode_lock_t, list);
+ list_del_init(&lock->list);
+ __pl_inodelk_unref(lock);
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
- grant_blocked_locks (this, pl_inode);
+ pthread_mutex_unlock(&pl_inode->mutex);
- do_blocked_rw (pl_inode);
+ if (error < 0) {
+ inodelk_contention_notify(xl, contend);
+ }
- return ret;
+ inode_unref(pl_inode->inode);
+
+ return error;
}
+void
+pl_inode_remove_wake(struct list_head *list)
+{
+ call_stub_t *stub;
-posix_lock_t *
-pl_getlk (pl_inode_t *pl_inode, posix_lock_t *lock)
+ while (!list_empty(list)) {
+ stub = list_first_entry(list, call_stub_t, list);
+ list_del_init(&stub->list);
+
+ call_resume(stub);
+ }
+}
+
+void
+pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error)
{
- posix_lock_t *conf = NULL;
+ struct list_head contend, granted;
+ struct timespec now;
+ pl_dom_list_t *dom;
+
+ if (pl_inode == NULL) {
+ return;
+ }
- conf = first_conflicting_overlap (pl_inode, lock);
+ INIT_LIST_HEAD(&contend);
+ INIT_LIST_HEAD(&granted);
+ timespec_now(&now);
- if (conf == NULL) {
- lock->fl_type = F_UNLCK;
- return lock;
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ if (error == 0) {
+ if (pl_inode->links >= 0) {
+ pl_inode->links--;
+ }
+ if (pl_inode->links == 0) {
+ pl_inode->removed = _gf_true;
+ }
+ }
+
+ pl_inode->remove_running--;
+
+ if ((pl_inode->remove_running == 0) && list_empty(&pl_inode->waiting)) {
+ pl_inode->is_locked = _gf_false;
+
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ __grant_blocked_inode_locks(xl, pl_inode, &granted, dom, &now,
+ &contend);
}
+ }
+
+ pthread_mutex_unlock(&pl_inode->mutex);
- return conf;
+ unwind_granted_inodes(xl, pl_inode, &granted);
+
+ inodelk_contention_notify(xl, &contend);
+
+ inode_unref(pl_inode->inode);
+}
+
+void
+pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode,
+ struct list_head *list)
+{
+ call_stub_t *stub, *tmp;
+
+ if (!pl_inode->is_locked) {
+ return;
+ }
+
+ list_for_each_entry_safe(stub, tmp, &pl_inode->waiting, list)
+ {
+ if (!pl_inode_has_owners(xl, stub->frame->root->client, pl_inode, NULL,
+ NULL)) {
+ list_move_tail(&stub->list, list);
+ }
+ }
}
+/* This function determines if an inodelk attempt can be done now or it needs
+ * to wait.
+ *
+ * Possible return values:
+ * < 0: An error occurred. Currently only -ESTALE can be returned if the
+ * inode has been deleted previously by unlink/rmdir/rename
+ * = 0: The lock can be attempted.
+ * > 0: The lock needs to wait because a conflicting remove operation is
+ * ongoing.
+ */
+int32_t
+pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock)
+{
+ pl_dom_list_t *dom;
+ pl_inode_lock_t *ilock;
+
+ /* If the inode has been deleted, we won't allow any lock. */
+ if (pl_inode->removed) {
+ return -ESTALE;
+ }
+
+ /* We only synchronize with locks made for regular operations coming from
+ * the user. Locks done for internal purposes are hard to control and could
+ * lead to long delays or deadlocks quite easily. */
+ if (lock->client_pid < 0) {
+ return 0;
+ }
+ if (!pl_inode->is_locked) {
+ return 0;
+ }
+ if (pl_inode->remove_running > 0) {
+ return 1;
+ }
+
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ list_for_each_entry(ilock, &dom->inodelk_list, list)
+ {
+ /* If a lock from the same client is already granted, we allow this
+ * one to continue. This is necessary to prevent deadlocks when
+ * multiple locks are taken for the same operation.
+ *
+ * On the other side it's unlikely that the same client sends
+ * completely unrelated locks for the same inode.
+ */
+ if (ilock->client == lock->client) {
+ return 0;
+ }
+ }
+ }
+
+ return 1;
+}
diff --git a/xlators/features/locks/src/common.h b/xlators/features/locks/src/common.h
index 5ec630ee857..281223bf3b8 100644
--- a/xlators/features/locks/src/common.h
+++ b/xlators/features/locks/src/common.h
@@ -1,5 +1,5 @@
/*
- Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+ Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
This file is licensed to you under your choice of the GNU Lesser
@@ -10,149 +10,253 @@
#ifndef __COMMON_H__
#define __COMMON_H__
-#include "lkowner.h"
/*dump locks format strings */
-#define RANGE_FMT "type=%s, whence=%hd, start=%llu, len=%llu"
-#define ENTRY_FMT "type=%s on basename=%s"
-#define DUMP_GEN_FMT "pid = %llu, owner=%s, client=%p"
-#define GRNTD_AT "granted at %s"
-#define BLKD_AT "blocked at %s"
-#define CONN_ID "connection-id=%s"
-#define DUMP_BLKD_FMT DUMP_GEN_FMT", "CONN_ID", "BLKD_AT
-#define DUMP_GRNTD_FMT DUMP_GEN_FMT", "CONN_ID", "GRNTD_AT
-#define DUMP_BLKD_GRNTD_FMT DUMP_GEN_FMT", "CONN_ID", "BLKD_AT", "GRNTD_AT
-
-#define ENTRY_BLKD_FMT ENTRY_FMT", "DUMP_BLKD_FMT
-#define ENTRY_GRNTD_FMT ENTRY_FMT", "DUMP_GRNTD_FMT
-#define ENTRY_BLKD_GRNTD_FMT ENTRY_FMT", "DUMP_BLKD_GRNTD_FMT
-
-#define RANGE_BLKD_FMT RANGE_FMT", "DUMP_BLKD_FMT
-#define RANGE_GRNTD_FMT RANGE_FMT", "DUMP_GRNTD_FMT
-#define RANGE_BLKD_GRNTD_FMT RANGE_FMT", "DUMP_BLKD_GRNTD_FMT
+#define RANGE_FMT "type=%s, whence=%hd, start=%llu, len=%llu"
+#define ENTRY_FMT "type=%s on basename=%s"
+#define DUMP_GEN_FMT "pid = %llu, owner=%s, client=%p"
+#define GRNTD_AT "granted at %s"
+#define BLKD_AT "blocked at %s"
+#define CONN_ID "connection-id=%s"
+#define DUMP_BLKD_FMT DUMP_GEN_FMT ", " CONN_ID ", " BLKD_AT
+#define DUMP_GRNTD_FMT DUMP_GEN_FMT ", " CONN_ID ", " GRNTD_AT
+#define DUMP_BLKD_GRNTD_FMT DUMP_GEN_FMT ", " CONN_ID ", " BLKD_AT ", " GRNTD_AT
+
+#define ENTRY_BLKD_FMT ENTRY_FMT ", " DUMP_BLKD_FMT
+#define ENTRY_GRNTD_FMT ENTRY_FMT ", " DUMP_GRNTD_FMT
+#define ENTRY_BLKD_GRNTD_FMT ENTRY_FMT ", " DUMP_BLKD_GRNTD_FMT
+
+#define RANGE_BLKD_FMT RANGE_FMT ", " DUMP_BLKD_FMT
+#define RANGE_GRNTD_FMT RANGE_FMT ", " DUMP_GRNTD_FMT
+#define RANGE_BLKD_GRNTD_FMT RANGE_FMT ", " DUMP_BLKD_GRNTD_FMT
#define SET_FLOCK_PID(flock, lock) ((flock)->l_pid = lock->client_pid)
+#define PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params...) \
+ do { \
+ frame->local = NULL; \
+ STACK_UNWIND_STRICT(fop, frame, op_ret, params); \
+ if (__local) { \
+ if (__local->inodelk_dom_count_req) \
+ data_unref(__local->inodelk_dom_count_req); \
+ loc_wipe(&__local->loc[0]); \
+ loc_wipe(&__local->loc[1]); \
+ if (__local->fd) \
+ fd_unref(__local->fd); \
+ if (__local->inode) \
+ inode_unref(__local->inode); \
+ if (__local->xdata) { \
+ dict_unref(__local->xdata); \
+ __local->xdata = NULL; \
+ } \
+ mem_put(__local); \
+ } \
+ } while (0)
posix_lock_t *
-new_posix_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
- gf_lkowner_t *owner, fd_t *fd);
+new_posix_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+ gf_lkowner_t *owner, fd_t *fd, uint32_t lk_flags, int blocking,
+ int32_t *op_errno);
pl_inode_t *
-pl_inode_get (xlator_t *this, inode_t *inode);
+pl_inode_get(xlator_t *this, inode_t *inode, pl_local_t *local);
posix_lock_t *
-pl_getlk (pl_inode_t *inode, posix_lock_t *lock);
+pl_getlk(pl_inode_t *inode, posix_lock_t *lock);
int
-pl_setlk (xlator_t *this, pl_inode_t *inode, posix_lock_t *lock,
- int can_block);
+pl_setlk(xlator_t *this, pl_inode_t *inode, posix_lock_t *lock, int can_block);
+
+int
+pl_lock_preempt(pl_inode_t *pl_inode, posix_lock_t *reqlock);
void
-grant_blocked_locks (xlator_t *this, pl_inode_t *inode);
+grant_blocked_locks(xlator_t *this, pl_inode_t *inode);
void
-posix_lock_to_flock (posix_lock_t *lock, struct gf_flock *flock);
+posix_lock_to_flock(posix_lock_t *lock, struct gf_flock *flock);
int
-locks_overlap (posix_lock_t *l1, posix_lock_t *l2);
+locks_overlap(posix_lock_t *l1, posix_lock_t *l2);
int
-same_owner (posix_lock_t *l1, posix_lock_t *l2);
+same_owner(posix_lock_t *l1, posix_lock_t *l2);
-void __delete_lock (pl_inode_t *, posix_lock_t *);
+void
+__delete_lock(posix_lock_t *);
-void __destroy_lock (posix_lock_t *);
+void
+__destroy_lock(posix_lock_t *);
pl_dom_list_t *
-get_domain (pl_inode_t *pl_inode, const char *volume);
+get_domain(pl_inode_t *pl_inode, const char *volume);
void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
- pl_dom_list_t *dom);
+grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+ pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend);
void
-__delete_inode_lock (pl_inode_lock_t *lock);
+inodelk_contention_notify(xlator_t *this, struct list_head *contend);
void
-__pl_inodelk_unref (pl_inode_lock_t *lock);
+__delete_inode_lock(pl_inode_lock_t *lock);
void
-grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
- pl_dom_list_t *dom);
+__pl_inodelk_unref(pl_inode_lock_t *lock);
-void pl_update_refkeeper (xlator_t *this, inode_t *inode);
+void
+__grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted, pl_dom_list_t *dom,
+ struct timespec *now, struct list_head *contend);
+
+void
+unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted);
+
+void
+grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+ pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend);
+
+void
+entrylk_contention_notify(xlator_t *this, struct list_head *contend);
+
+void
+pl_update_refkeeper(xlator_t *this, inode_t *inode);
int32_t
-__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode, char *domname);
+__get_inodelk_count(xlator_t *this, pl_inode_t *pl_inode, char *domname);
int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode, char *domname);
+get_inodelk_count(xlator_t *this, inode_t *inode, char *domname);
int32_t
-__get_entrylk_count (xlator_t *this, pl_inode_t *pl_inode);
+__get_entrylk_count(xlator_t *this, pl_inode_t *pl_inode);
int32_t
-get_entrylk_count (xlator_t *this, inode_t *inode);
+get_entrylk_count(xlator_t *this, inode_t *inode);
-void pl_trace_in (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, const char *domain);
+void
+pl_trace_in(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+ struct gf_flock *flock, const char *domain);
-void pl_trace_out (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, int op_ret, int op_errno, const char *domain);
+void
+pl_trace_out(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc, int cmd,
+ struct gf_flock *flock, int op_ret, int op_errno,
+ const char *domain);
-void pl_trace_block (xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
- int cmd, struct gf_flock *flock, const char *domain);
+void
+pl_trace_block(xlator_t *this, call_frame_t *frame, fd_t *fd, loc_t *loc,
+ int cmd, struct gf_flock *flock, const char *domain);
-void pl_trace_flush (xlator_t *this, call_frame_t *frame, fd_t *fd);
+void
+pl_trace_flush(xlator_t *this, call_frame_t *frame, fd_t *fd);
-void entrylk_trace_in (xlator_t *this, call_frame_t *frame, const char *volume,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type);
+void
+entrylk_trace_in(xlator_t *this, call_frame_t *frame, const char *volume,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type);
-void entrylk_trace_out (xlator_t *this, call_frame_t *frame, const char *volume,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type,
- int op_ret, int op_errno);
+void
+entrylk_trace_out(xlator_t *this, call_frame_t *frame, const char *volume,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type, int op_ret, int op_errno);
-void entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type);
+void
+entrylk_trace_block(xlator_t *this, call_frame_t *frame, const char *volume,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type);
void
-pl_print_verdict (char *str, int size, int op_ret, int op_errno);
+pl_print_verdict(char *str, int size, int op_ret, int op_errno);
void
-pl_print_lockee (char *str, int size, fd_t *fd, loc_t *loc);
+pl_print_lockee(char *str, int size, fd_t *fd, loc_t *loc);
void
-pl_print_locker (char *str, int size, xlator_t *this, call_frame_t *frame);
+pl_print_locker(char *str, int size, xlator_t *this, call_frame_t *frame);
void
-pl_print_inodelk (char *str, int size, int cmd, struct gf_flock *flock, const char *domain);
+pl_print_inodelk(char *str, int size, int cmd, struct gf_flock *flock,
+ const char *domain);
void
-pl_trace_release (xlator_t *this, fd_t *fd);
+pl_trace_release(xlator_t *this, fd_t *fd);
unsigned long
-fd_to_fdnum (fd_t *fd);
+fd_to_fdnum(fd_t *fd);
fd_t *
-fd_from_fdnum (posix_lock_t *lock);
+fd_from_fdnum(posix_lock_t *lock);
+
+int
+pl_reserve_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ int can_block);
+int
+reservelks_equal(posix_lock_t *l1, posix_lock_t *l2);
int
-pl_reserve_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
- int can_block);
+pl_verify_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ int can_block);
int
-reservelks_equal (posix_lock_t *l1, posix_lock_t *l2);
+pl_reserve_unlock(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *reqlock);
+
+int32_t
+check_entrylk_on_basename(xlator_t *this, inode_t *parent, char *basename);
+
+void
+__pl_inodelk_unref(pl_inode_lock_t *lock);
+void
+__pl_entrylk_unref(pl_entry_lock_t *lock);
int
-pl_verify_reservelk (xlator_t *this, pl_inode_t *pl_inode,
- posix_lock_t *lock, int can_block);
+pl_metalock_is_active(pl_inode_t *pl_inode);
+
+void
+__pl_queue_lock(pl_inode_t *pl_inode, posix_lock_t *reqlock);
+
+void
+inodelk_contention_notify_check(xlator_t *xl, pl_inode_lock_t *lock,
+ struct timespec *now,
+ struct list_head *contend);
+
+void
+entrylk_contention_notify_check(xlator_t *xl, pl_entry_lock_t *lock,
+ struct timespec *now,
+ struct list_head *contend);
+
+gf_boolean_t
+pl_does_monkey_want_stuck_lock();
+
+gf_boolean_t
+pl_is_mandatory_locking_enabled(pl_inode_t *pl_inode);
+
+void
+pl_clean_local(pl_local_t *local);
+
int
-pl_reserve_unlock (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *reqlock);
+pl_local_init(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd);
+
+gf_boolean_t
+pl_is_lk_owner_valid(gf_lkowner_t *owner, client_t *client);
+
+int32_t
+pl_inode_remove_prepare(xlator_t *xl, call_frame_t *frame, loc_t *loc,
+ pl_inode_t **ppl_inode, struct list_head *contend);
+
+int32_t
+pl_inode_remove_complete(xlator_t *xl, pl_inode_t *pl_inode, call_stub_t *stub,
+ struct list_head *contend);
-uint32_t
-check_entrylk_on_basename (xlator_t *this, inode_t *parent, char *basename);
+void
+pl_inode_remove_wake(struct list_head *list);
+
+void
+pl_inode_remove_cbk(xlator_t *xl, pl_inode_t *pl_inode, int32_t error);
-void __pl_inodelk_unref (pl_inode_lock_t *lock);
-void __pl_entrylk_unref (pl_entry_lock_t *lock);
+void
+pl_inode_remove_unlocked(xlator_t *xl, pl_inode_t *pl_inode,
+ struct list_head *list);
+
+int32_t
+pl_inode_remove_inodelk(pl_inode_t *pl_inode, pl_inode_lock_t *lock);
#endif /* __COMMON_H__ */
diff --git a/xlators/features/locks/src/entrylk.c b/xlators/features/locks/src/entrylk.c
index 8496d9d8dce..fd772c850dd 100644
--- a/xlators/features/locks/src/entrylk.c
+++ b/xlators/features/locks/src/entrylk.c
@@ -7,77 +7,77 @@
later), or the GNU General Public License, version 2 (GPLv2), in all
cases as published by the Free Software Foundation.
*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
+#include <glusterfs/upcall-utils.h>
#include "locks.h"
+#include "clear.h"
#include "common.h"
-
+#include "pl-messages.h"
void
-__pl_entrylk_unref (pl_entry_lock_t *lock)
+__pl_entrylk_unref(pl_entry_lock_t *lock)
{
- lock->ref--;
- if (!lock->ref) {
- GF_FREE ((char *)lock->basename);
- GF_FREE (lock->connection_id);
- GF_FREE (lock);
- }
+ lock->ref--;
+ if (!lock->ref) {
+ GF_FREE((char *)lock->basename);
+ GF_FREE(lock->connection_id);
+ GF_FREE(lock);
+ }
}
-
static void
-__pl_entrylk_ref (pl_entry_lock_t *lock)
+__pl_entrylk_ref(pl_entry_lock_t *lock)
{
- lock->ref++;
+ lock->ref++;
}
-
static pl_entry_lock_t *
-new_entrylk_lock (pl_inode_t *pinode, const char *basename, entrylk_type type,
- const char *domain, call_frame_t *frame, char *conn_id)
+new_entrylk_lock(pl_inode_t *pinode, const char *basename, entrylk_type type,
+ const char *domain, call_frame_t *frame, char *conn_id,
+ int32_t *op_errno)
{
- pl_entry_lock_t *newlock = NULL;
-
- newlock = GF_CALLOC (1, sizeof (pl_entry_lock_t),
- gf_locks_mt_pl_entry_lock_t);
- if (!newlock) {
- goto out;
- }
-
- newlock->basename = basename ? gf_strdup (basename) : NULL;
- newlock->type = type;
- newlock->client = frame->root->client;
- newlock->client_pid = frame->root->pid;
- newlock->volume = domain;
- newlock->owner = frame->root->lk_owner;
- newlock->frame = frame;
- newlock->this = frame->this;
-
- if (conn_id) {
- newlock->connection_id = gf_strdup (conn_id);
- }
-
- INIT_LIST_HEAD (&newlock->domain_list);
- INIT_LIST_HEAD (&newlock->blocked_locks);
- INIT_LIST_HEAD (&newlock->client_list);
-
- __pl_entrylk_ref (newlock);
+ pl_entry_lock_t *newlock = NULL;
+
+ if (!pl_is_lk_owner_valid(&frame->root->lk_owner, frame->root->client)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ newlock = GF_CALLOC(1, sizeof(pl_entry_lock_t),
+ gf_locks_mt_pl_entry_lock_t);
+ if (!newlock) {
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ newlock->basename = basename ? gf_strdup(basename) : NULL;
+ newlock->type = type;
+ newlock->client = frame->root->client;
+ newlock->client_pid = frame->root->pid;
+ newlock->volume = domain;
+ newlock->owner = frame->root->lk_owner;
+ newlock->frame = frame;
+ newlock->this = frame->this;
+
+ if (conn_id) {
+ newlock->connection_id = gf_strdup(conn_id);
+ }
+
+ INIT_LIST_HEAD(&newlock->domain_list);
+ INIT_LIST_HEAD(&newlock->blocked_locks);
+ INIT_LIST_HEAD(&newlock->client_list);
+
+ __pl_entrylk_ref(newlock);
out:
- return newlock;
+ return newlock;
}
-
/**
* all_names - does a basename represent all names?
* @basename: name to check
@@ -92,20 +92,221 @@ out:
*/
static int
-names_conflict (const char *n1, const char *n2)
+names_conflict(const char *n1, const char *n2)
+{
+ return all_names(n1) || all_names(n2) || !strcmp(n1, n2);
+}
+
+static int
+__same_entrylk_owner(pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+{
+ return (is_same_lkowner(&l1->owner, &l2->owner) &&
+ (l1->client == l2->client));
+}
+
+/* Just as in inodelk, allow conflicting name locks from same (lk_owner, conn)*/
+static int
+__conflicting_entrylks(pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+{
+ if (names_conflict(l1->basename, l2->basename) &&
+ !__same_entrylk_owner(l1, l2))
+ return 1;
+
+ return 0;
+}
+
+/* See comments in inodelk.c for details */
+static inline gf_boolean_t
+__stale_entrylk(xlator_t *this, pl_entry_lock_t *candidate_lock,
+ pl_entry_lock_t *requested_lock, time_t *lock_age_sec)
{
- return all_names (n1) || all_names (n2) || !strcmp (n1, n2);
+ posix_locks_private_t *priv = NULL;
+
+ priv = this->private;
+
+ /* Question: Should we just prune them all given the
+ * chance? Or just the locks we are attempting to acquire?
+ */
+ if (names_conflict(candidate_lock->basename, requested_lock->basename)) {
+ *lock_age_sec = gf_time() - candidate_lock->granted_time;
+ if (*lock_age_sec > priv->revocation_secs)
+ return _gf_true;
+ }
+ return _gf_false;
}
+/* See comments in inodelk.c for details */
+static gf_boolean_t
+__entrylk_prune_stale(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+ pl_entry_lock_t *lock)
+{
+ posix_locks_private_t *priv = NULL;
+ pl_entry_lock_t *tmp = NULL;
+ pl_entry_lock_t *lk = NULL;
+ gf_boolean_t revoke_lock = _gf_false;
+ int bcount = 0;
+ int gcount = 0;
+ int op_errno = 0;
+ clrlk_args args;
+ args.opts = NULL;
+ time_t lk_age_sec = 0;
+ uint32_t max_blocked = 0;
+ char *reason_str = NULL;
+
+ priv = this->private;
+ args.type = CLRLK_ENTRY;
+ if (priv->revocation_clear_all == _gf_true)
+ args.kind = CLRLK_ALL;
+ else
+ args.kind = CLRLK_GRANTED;
+
+ if (list_empty(&dom->entrylk_list))
+ goto out;
+
+ pthread_mutex_lock(&pinode->mutex);
+ lock->pinode = pinode;
+ list_for_each_entry_safe(lk, tmp, &dom->entrylk_list, domain_list)
+ {
+ if (__stale_entrylk(this, lk, lock, &lk_age_sec) == _gf_true) {
+ revoke_lock = _gf_true;
+ reason_str = "age";
+ break;
+ }
+ }
+ max_blocked = priv->revocation_max_blocked;
+ if (max_blocked != 0 && revoke_lock == _gf_false) {
+ list_for_each_entry_safe(lk, tmp, &dom->blocked_entrylks, blocked_locks)
+ {
+ max_blocked--;
+ if (max_blocked == 0) {
+ revoke_lock = _gf_true;
+ reason_str = "max blocked";
+ break;
+ }
+ }
+ }
+ pthread_mutex_unlock(&pinode->mutex);
+
+out:
+ if (revoke_lock == _gf_true) {
+ clrlk_clear_entrylk(this, pinode, dom, &args, &bcount, &gcount,
+ &op_errno);
+ gf_log(this->name, GF_LOG_WARNING,
+ "Lock revocation [reason: %s; gfid: %s; domain: %s; "
+ "age: %ld sec] - Entry lock revoked: %d granted & %d "
+ "blocked locks cleared",
+ reason_str, uuid_utoa(pinode->gfid), dom->domain, lk_age_sec,
+ gcount, bcount);
+ }
+
+ return revoke_lock;
+}
-static inline int
-__same_entrylk_owner (pl_entry_lock_t *l1, pl_entry_lock_t *l2)
+void
+entrylk_contention_notify_check(xlator_t *this, pl_entry_lock_t *lock,
+ struct timespec *now, struct list_head *contend)
{
+ posix_locks_private_t *priv;
+ int64_t elapsed;
+
+ priv = this->private;
+
+ /* If this lock is in a list, it means that we are about to send a
+ * notification for it, so no need to do anything else. */
+ if (!list_empty(&lock->contend)) {
+ return;
+ }
+
+ elapsed = now->tv_sec;
+ elapsed -= lock->contention_time.tv_sec;
+ if (now->tv_nsec < lock->contention_time.tv_nsec) {
+ elapsed--;
+ }
+ if (elapsed < priv->notify_contention_delay) {
+ return;
+ }
- return (is_same_lkowner (&l1->owner, &l2->owner) &&
- (l1->client == l2->client));
+ /* All contention notifications will be sent outside of the locked
+ * region. This means that currently granted locks might have already
+ * been unlocked by that time. To avoid the lock or the inode to be
+ * destroyed before we process them, we take an additional reference
+ * on both. */
+ inode_ref(lock->pinode->inode);
+ __pl_entrylk_ref(lock);
+
+ lock->contention_time = *now;
+
+ list_add_tail(&lock->contend, contend);
}
+void
+entrylk_contention_notify(xlator_t *this, struct list_head *contend)
+{
+ struct gf_upcall up;
+ struct gf_upcall_entrylk_contention lc;
+ pl_entry_lock_t *lock;
+ pl_inode_t *pl_inode;
+ client_t *client;
+ gf_boolean_t notify;
+
+ while (!list_empty(contend)) {
+ lock = list_first_entry(contend, pl_entry_lock_t, contend);
+
+ pl_inode = lock->pinode;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ /* If the lock has already been released, no notification is
+ * sent. We clear the notification time in this case. */
+ notify = !list_empty(&lock->domain_list);
+ if (!notify) {
+ lock->contention_time.tv_sec = 0;
+ lock->contention_time.tv_nsec = 0;
+ } else {
+ lc.type = lock->type;
+ lc.name = lock->basename;
+ lc.pid = lock->client_pid;
+ lc.domain = lock->volume;
+ lc.xdata = NULL;
+
+ gf_uuid_copy(up.gfid, lock->pinode->gfid);
+ client = (client_t *)lock->client;
+ if (client == NULL) {
+ /* A NULL client can be found if the entrylk
+ * was issued by a server side xlator. */
+ up.client_uid = NULL;
+ } else {
+ up.client_uid = client->client_uid;
+ }
+ }
+
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ if (notify) {
+ up.event_type = GF_UPCALL_ENTRYLK_CONTENTION;
+ up.data = &lc;
+
+ if (this->notify(this, GF_EVENT_UPCALL, &up) < 0) {
+ gf_msg_debug(this->name, 0,
+ "Entrylk contention notification "
+ "failed");
+ } else {
+ gf_msg_debug(this->name, 0,
+ "Entrylk contention notification "
+ "sent");
+ }
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ list_del_init(&lock->contend);
+ __pl_entrylk_unref(lock);
+
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ inode_unref(pl_inode->inode);
+ }
+}
/**
* entrylk_grantable - is this lock grantable?
@@ -114,184 +315,188 @@ __same_entrylk_owner (pl_entry_lock_t *l1, pl_entry_lock_t *l2)
* @type: type of lock
*/
static pl_entry_lock_t *
-__entrylk_grantable (pl_dom_list_t *dom, pl_entry_lock_t *lock)
+__entrylk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_entry_lock_t *lock,
+ struct timespec *now, struct list_head *contend)
{
- pl_entry_lock_t *tmp = NULL;
-
- if (list_empty (&dom->entrylk_list))
- return NULL;
-
- list_for_each_entry (tmp, &dom->entrylk_list, domain_list) {
- if (names_conflict (tmp->basename, lock->basename))
- return tmp;
+ pl_entry_lock_t *tmp = NULL;
+ pl_entry_lock_t *ret = NULL;
+
+ list_for_each_entry(tmp, &dom->entrylk_list, domain_list)
+ {
+ if (__conflicting_entrylks(tmp, lock)) {
+ if (ret == NULL) {
+ ret = tmp;
+ if (contend == NULL) {
+ break;
+ }
+ }
+ entrylk_contention_notify_check(this, tmp, now, contend);
}
+ }
- return NULL;
+ return ret;
}
static pl_entry_lock_t *
-__blocked_entrylk_conflict (pl_dom_list_t *dom, pl_entry_lock_t *lock)
+__blocked_entrylk_conflict(pl_dom_list_t *dom, pl_entry_lock_t *lock)
{
- pl_entry_lock_t *tmp = NULL;
-
- if (list_empty (&dom->blocked_entrylks))
- return NULL;
+ pl_entry_lock_t *tmp = NULL;
- list_for_each_entry (tmp, &dom->blocked_entrylks, blocked_locks) {
- if (names_conflict (tmp->basename, lock->basename))
- return lock;
- }
+ list_for_each_entry(tmp, &dom->blocked_entrylks, blocked_locks)
+ {
+ if (names_conflict(tmp->basename, lock->basename))
+ return lock;
+ }
- return NULL;
+ return NULL;
}
static int
-__owner_has_lock (pl_dom_list_t *dom, pl_entry_lock_t *newlock)
+__owner_has_lock(pl_dom_list_t *dom, pl_entry_lock_t *newlock)
{
- pl_entry_lock_t *lock = NULL;
+ pl_entry_lock_t *lock = NULL;
- list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
- if (__same_entrylk_owner (lock, newlock))
- return 1;
- }
+ list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+ {
+ if (__same_entrylk_owner(lock, newlock))
+ return 1;
+ }
- list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
- if (__same_entrylk_owner (lock, newlock))
- return 1;
- }
+ list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
+ {
+ if (__same_entrylk_owner(lock, newlock))
+ return 1;
+ }
- return 0;
+ return 0;
}
static int
-names_equal (const char *n1, const char *n2)
+names_equal(const char *n1, const char *n2)
{
- return (n1 == NULL && n2 == NULL) || (n1 && n2 && !strcmp (n1, n2));
+ return (n1 == NULL && n2 == NULL) || (n1 && n2 && !strcmp(n1, n2));
}
void
-pl_print_entrylk (char *str, int size, entrylk_cmd cmd, entrylk_type type,
- const char *basename, const char *domain)
+pl_print_entrylk(char *str, int size, entrylk_cmd cmd, entrylk_type type,
+ const char *basename, const char *domain)
{
- char *cmd_str = NULL;
- char *type_str = NULL;
+ char *cmd_str = NULL;
+ char *type_str = NULL;
- switch (cmd) {
+ switch (cmd) {
case ENTRYLK_LOCK:
- cmd_str = "LOCK";
- break;
+ cmd_str = "LOCK";
+ break;
case ENTRYLK_LOCK_NB:
- cmd_str = "LOCK_NB";
- break;
+ cmd_str = "LOCK_NB";
+ break;
case ENTRYLK_UNLOCK:
- cmd_str = "UNLOCK";
- break;
+ cmd_str = "UNLOCK";
+ break;
default:
- cmd_str = "UNKNOWN";
- break;
- }
+ cmd_str = "UNKNOWN";
+ break;
+ }
- switch (type) {
+ switch (type) {
case ENTRYLK_RDLCK:
- type_str = "READ";
- break;
+ type_str = "READ";
+ break;
case ENTRYLK_WRLCK:
- type_str = "WRITE";
- break;
+ type_str = "WRITE";
+ break;
default:
- type_str = "UNKNOWN";
- break;
- }
+ type_str = "UNKNOWN";
+ break;
+ }
- snprintf (str, size, "lock=ENTRYLK, cmd=%s, type=%s, basename=%s, domain: %s",
- cmd_str, type_str, basename, domain);
+ snprintf(str, size,
+ "lock=ENTRYLK, cmd=%s, type=%s, basename=%s, domain: %s", cmd_str,
+ type_str, basename, domain);
}
-
void
-entrylk_trace_in (xlator_t *this, call_frame_t *frame, const char *domain,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type)
+entrylk_trace_in(xlator_t *this, call_frame_t *frame, const char *domain,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_entrylk[256];
+ posix_locks_private_t *priv = NULL;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_entrylk[256];
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, domain);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, domain);
- gf_log (this->name, GF_LOG_INFO,
- "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}",
- pl_locker, pl_lockee, pl_entrylk);
+ gf_log(this->name, GF_LOG_INFO,
+ "[REQUEST] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+ pl_lockee, pl_entrylk);
}
-
void
-entrylk_trace_out (xlator_t *this, call_frame_t *frame, const char *domain,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type, int op_ret, int op_errno)
+entrylk_trace_out(xlator_t *this, call_frame_t *frame, const char *domain,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type, int op_ret, int op_errno)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_entrylk[256];
- char verdict[32];
+ posix_locks_private_t *priv = NULL;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_entrylk[256];
+ char verdict[32];
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, domain);
- pl_print_verdict (verdict, 32, op_ret, op_errno);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, domain);
+ pl_print_verdict(verdict, 32, op_ret, op_errno);
- gf_log (this->name, GF_LOG_INFO,
- "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}",
- verdict, pl_locker, pl_lockee, pl_entrylk);
+ gf_log(this->name, GF_LOG_INFO,
+ "[%s] Locker = {%s} Lockee = {%s} Lock = {%s}", verdict, pl_locker,
+ pl_lockee, pl_entrylk);
}
-
void
-entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
- fd_t *fd, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type)
+entrylk_trace_block(xlator_t *this, call_frame_t *frame, const char *volume,
+ fd_t *fd, loc_t *loc, const char *basename, entrylk_cmd cmd,
+ entrylk_type type)
{
- posix_locks_private_t *priv = NULL;
- char pl_locker[256];
- char pl_lockee[256];
- char pl_entrylk[256];
+ posix_locks_private_t *priv = NULL;
+ char pl_locker[256];
+ char pl_lockee[256];
+ char pl_entrylk[256];
- priv = this->private;
+ priv = this->private;
- if (!priv->trace)
- return;
+ if (!priv->trace)
+ return;
- pl_print_locker (pl_locker, 256, this, frame);
- pl_print_lockee (pl_lockee, 256, fd, loc);
- pl_print_entrylk (pl_entrylk, 256, cmd, type, basename, volume);
+ pl_print_locker(pl_locker, 256, this, frame);
+ pl_print_lockee(pl_lockee, 256, fd, loc);
+ pl_print_entrylk(pl_entrylk, 256, cmd, type, basename, volume);
- gf_log (this->name, GF_LOG_INFO,
- "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}",
- pl_locker, pl_lockee, pl_entrylk);
+ gf_log(this->name, GF_LOG_INFO,
+ "[BLOCKED] Locker = {%s} Lockee = {%s} Lock = {%s}", pl_locker,
+ pl_lockee, pl_entrylk);
}
/**
- * __find_most_matching_lock - find the lock struct which most matches in order of:
- * lock on the exact basename ||
- * an all_names lock
+ * __find_most_matching_lock - find the lock struct which most matches in order
+ * of: lock on the exact basename || an all_names lock
*
*
* @inode: inode in which to look
@@ -299,23 +504,57 @@ entrylk_trace_block (xlator_t *this, call_frame_t *frame, const char *volume,
*/
static pl_entry_lock_t *
-__find_most_matching_lock (pl_dom_list_t *dom, const char *basename)
+__find_most_matching_lock(pl_dom_list_t *dom, const char *basename)
{
- pl_entry_lock_t *lock;
- pl_entry_lock_t *all = NULL;
- pl_entry_lock_t *exact = NULL;
+ pl_entry_lock_t *lock;
+ pl_entry_lock_t *all = NULL;
+ pl_entry_lock_t *exact = NULL;
- if (list_empty (&dom->entrylk_list))
- return NULL;
+ if (list_empty(&dom->entrylk_list))
+ return NULL;
- list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
- if (all_names (lock->basename))
- all = lock;
- else if (names_equal (lock->basename, basename))
- exact = lock;
- }
+ list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+ {
+ if (all_names(lock->basename))
+ all = lock;
+ else if (names_equal(lock->basename, basename))
+ exact = lock;
+ }
+
+ return (exact ? exact : all);
+}
+
+static pl_entry_lock_t *
+__find_matching_lock(pl_dom_list_t *dom, pl_entry_lock_t *lock)
+{
+ pl_entry_lock_t *tmp = NULL;
+
+ list_for_each_entry(tmp, &dom->entrylk_list, domain_list)
+ {
+ if (names_equal(lock->basename, tmp->basename) &&
+ __same_entrylk_owner(lock, tmp) && (lock->type == tmp->type))
+ return tmp;
+ }
+ return NULL;
+}
+
+static int
+__lock_blocked_add(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+ pl_entry_lock_t *lock, int nonblock)
+{
+ if (nonblock)
+ goto out;
+
+ lock->blkd_time = gf_time();
+ list_add_tail(&lock->blocked_locks, &dom->blocked_entrylks);
+
+ gf_msg_trace(this->name, 0, "Blocking lock: {pinode=%p, basename=%s}",
+ pinode, lock->basename);
- return (exact ? exact : all);
+ entrylk_trace_block(this, lock->frame, NULL, NULL, NULL, lock->basename,
+ ENTRYLK_LOCK, lock->type);
+out:
+ return -EAGAIN;
}
/**
@@ -330,52 +569,49 @@ __find_most_matching_lock (pl_dom_list_t *dom, const char *basename)
*/
int
-__lock_entrylk (xlator_t *this, pl_inode_t *pinode, pl_entry_lock_t *lock,
- int nonblock, pl_dom_list_t *dom)
+__lock_entrylk(xlator_t *this, pl_inode_t *pinode, pl_entry_lock_t *lock,
+ int nonblock, pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend)
{
- pl_entry_lock_t *conf = NULL;
- int ret = -EAGAIN;
-
- conf = __entrylk_grantable (dom, lock);
- if (conf) {
- ret = -EAGAIN;
- if (nonblock)
- goto out;
-
- gettimeofday (&lock->blkd_time, NULL);
- list_add_tail (&lock->blocked_locks, &dom->blocked_entrylks);
-
- gf_log (this->name, GF_LOG_TRACE,
- "Blocking lock: {pinode=%p, basename=%s}",
- pinode, lock->basename);
-
- goto out;
+ pl_entry_lock_t *conf = NULL;
+ int ret = -EAGAIN;
+
+ conf = __entrylk_grantable(this, dom, lock, now, contend);
+ if (conf) {
+ ret = __lock_blocked_add(this, pinode, dom, lock, nonblock);
+ goto out;
+ }
+
+ /* To prevent blocked locks starvation, check if there are any blocked
+ * locks thay may conflict with this lock. If there is then don't grant
+ * the lock. BUT grant the lock if the owner already has lock to allow
+ * nested locks.
+ * Example: SHD from Machine1 takes (gfid, basename=257-length-name)
+ * and is granted.
+ * SHD from machine2 takes (gfid, basename=NULL) and is blocked.
+ * When SHD from Machine1 takes (gfid, basename=NULL) it needs to be
+ * granted, without which self-heal can't progress.
+ * TODO: Find why 'owner_has_lock' is checked even for blocked locks.
+ */
+ if (__blocked_entrylk_conflict(dom, lock) &&
+ !(__owner_has_lock(dom, lock))) {
+ if (nonblock == 0) {
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Lock is grantable, but blocking to prevent "
+ "starvation");
}
- if (__blocked_entrylk_conflict (dom, lock) && !(__owner_has_lock (dom, lock))) {
- ret = -EAGAIN;
- if (nonblock)
- goto out;
+ ret = __lock_blocked_add(this, pinode, dom, lock, nonblock);
+ goto out;
+ }
- gettimeofday (&lock->blkd_time, NULL);
- list_add_tail (&lock->blocked_locks, &dom->blocked_entrylks);
+ __pl_entrylk_ref(lock);
+ lock->granted_time = gf_time();
+ list_add(&lock->domain_list, &dom->entrylk_list);
- gf_log (this->name, GF_LOG_DEBUG,
- "Lock is grantable, but blocking to prevent starvation");
- gf_log (this->name, GF_LOG_TRACE,
- "Blocking lock: {pinode=%p, basename=%s}",
- pinode, lock->basename);
-
- goto out;
- }
-
- __pl_entrylk_ref (lock);
- gettimeofday (&lock->granted_time, NULL);
- list_add (&lock->domain_list, &dom->entrylk_list);
-
- ret = 0;
+ ret = 0;
out:
- return ret;
+ return ret;
}
/**
@@ -386,262 +622,322 @@ out:
*/
pl_entry_lock_t *
-__unlock_entrylk (pl_dom_list_t *dom, pl_entry_lock_t *lock)
+__unlock_entrylk(pl_dom_list_t *dom, pl_entry_lock_t *lock)
{
- pl_entry_lock_t *tmp = NULL;
- pl_entry_lock_t *ret_lock = NULL;
-
- tmp = __find_most_matching_lock (dom, lock->basename);
-
- if (!tmp) {
- gf_log ("locks", GF_LOG_ERROR,
- "unlock on %s (type=ENTRYLK_WRLCK) attempted but no matching lock found",
- lock->basename);
- goto out;
- }
-
- if (names_equal (tmp->basename, lock->basename)
- && tmp->type == lock->type) {
+ pl_entry_lock_t *ret_lock = NULL;
- list_del_init (&tmp->domain_list);
- ret_lock = tmp;
+ ret_lock = __find_matching_lock(dom, lock);
- } else {
- gf_log ("locks", GF_LOG_ERROR,
- "Unlock on %s for a non-existing lock!", lock->basename);
- goto out;
- }
+ if (ret_lock) {
+ list_del_init(&ret_lock->domain_list);
+ } else {
+ gf_log("locks", GF_LOG_ERROR,
+ "unlock on %s "
+ "(type=ENTRYLK_WRLCK) attempted but no matching lock "
+ "found",
+ lock->basename);
+ }
-out:
- return ret_lock;
+ return ret_lock;
}
-uint32_t
-check_entrylk_on_basename (xlator_t *this, inode_t *parent, char *basename)
+int32_t
+check_entrylk_on_basename(xlator_t *this, inode_t *parent, char *basename)
{
- uint32_t entrylk = 0;
- pl_inode_t *pinode = 0;
- pl_dom_list_t *dom = NULL;
- pl_entry_lock_t *conf = NULL;
-
- pinode = pl_inode_get (this, parent);
- if (!pinode)
- goto out;
- pthread_mutex_lock (&pinode->mutex);
+ int32_t entrylk = 0;
+ pl_dom_list_t *dom = NULL;
+ pl_entry_lock_t *conf = NULL;
+
+ pl_inode_t *pinode = pl_inode_get(this, parent, NULL);
+ if (!pinode)
+ goto out;
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ list_for_each_entry(dom, &pinode->dom_list, inode_list)
{
- list_for_each_entry (dom, &pinode->dom_list, inode_list) {
- conf = __find_most_matching_lock (dom, basename);
- if (conf && conf->basename) {
- entrylk = 1;
- break;
- }
- }
+ conf = __find_most_matching_lock(dom, basename);
+ if (conf && conf->basename) {
+ entrylk = 1;
+ break;
+ }
}
- pthread_mutex_unlock (&pinode->mutex);
+ }
+ pthread_mutex_unlock(&pinode->mutex);
out:
- return entrylk;
+ return entrylk;
}
void
-__grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
- pl_dom_list_t *dom, struct list_head *granted)
+__grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+ pl_dom_list_t *dom, struct list_head *granted,
+ struct timespec *now, struct list_head *contend)
{
- int bl_ret = 0;
- pl_entry_lock_t *bl = NULL;
- pl_entry_lock_t *tmp = NULL;
-
- struct list_head blocked_list;
+ int bl_ret = 0;
+ pl_entry_lock_t *bl = NULL;
+ pl_entry_lock_t *tmp = NULL;
- INIT_LIST_HEAD (&blocked_list);
- list_splice_init (&dom->blocked_entrylks, &blocked_list);
+ struct list_head blocked_list;
- list_for_each_entry_safe (bl, tmp, &blocked_list, blocked_locks) {
+ INIT_LIST_HEAD(&blocked_list);
+ list_splice_init(&dom->blocked_entrylks, &blocked_list);
- list_del_init (&bl->blocked_locks);
+ list_for_each_entry_safe(bl, tmp, &blocked_list, blocked_locks)
+ {
+ list_del_init(&bl->blocked_locks);
- bl_ret = __lock_entrylk (bl->this, pl_inode, bl, 0, dom);
+ bl_ret = __lock_entrylk(bl->this, pl_inode, bl, 0, dom, now, contend);
- if (bl_ret == 0) {
- list_add (&bl->blocked_locks, granted);
- }
+ if (bl_ret == 0) {
+ list_add_tail(&bl->blocked_locks, granted);
}
- return;
+ }
}
/* Grants locks if possible which are blocked on a lock */
void
-grant_blocked_entry_locks (xlator_t *this, pl_inode_t *pl_inode,
- pl_dom_list_t *dom)
+grant_blocked_entry_locks(xlator_t *this, pl_inode_t *pl_inode,
+ pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend)
{
- struct list_head granted_list;
- pl_entry_lock_t *tmp = NULL;
- pl_entry_lock_t *lock = NULL;
-
- INIT_LIST_HEAD (&granted_list);
-
- pthread_mutex_lock (&pl_inode->mutex);
+ struct list_head granted_list;
+ pl_entry_lock_t *tmp = NULL;
+ pl_entry_lock_t *lock = NULL;
+
+ INIT_LIST_HEAD(&granted_list);
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __grant_blocked_entry_locks(this, pl_inode, dom, &granted_list, now,
+ contend);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ list_for_each_entry_safe(lock, tmp, &granted_list, blocked_locks)
+ {
+ entrylk_trace_out(this, lock->frame, NULL, NULL, NULL, lock->basename,
+ ENTRYLK_LOCK, lock->type, 0, 0);
+
+ STACK_UNWIND_STRICT(entrylk, lock->frame, 0, 0, NULL);
+ lock->frame = NULL;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(lock, tmp, &granted_list, blocked_locks)
{
- __grant_blocked_entry_locks (this, pl_inode, dom,
- &granted_list);
+ list_del_init(&lock->blocked_locks);
+ __pl_entrylk_unref(lock);
}
- pthread_mutex_unlock (&pl_inode->mutex);
-
- list_for_each_entry_safe (lock, tmp, &granted_list, blocked_locks) {
- entrylk_trace_out (this, lock->frame, NULL, NULL, NULL,
- lock->basename, ENTRYLK_LOCK, lock->type,
- 0, 0);
-
- STACK_UNWIND_STRICT (entrylk, lock->frame, 0, 0, NULL);
- lock->frame = NULL;
- }
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- list_for_each_entry_safe (lock, tmp, &granted_list, blocked_locks) {
- list_del_init (&lock->blocked_locks);
- __pl_entrylk_unref (lock);
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
-
- return;
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
}
-
/* Common entrylk code called by pl_entrylk and pl_fentrylk */
int
-pl_common_entrylk (call_frame_t *frame, xlator_t *this,
- const char *volume, inode_t *inode, const char *basename,
- entrylk_cmd cmd, entrylk_type type, loc_t *loc, fd_t *fd,
- dict_t *xdata)
-
-{
- int32_t op_ret = -1;
- int32_t op_errno = 0;
- int ret = -1;
- char unwind = 1;
- GF_UNUSED int dict_ret = -1;
- pl_inode_t *pinode = NULL;
- pl_entry_lock_t *reqlock = NULL;
- pl_entry_lock_t *unlocked = NULL;
- pl_dom_list_t *dom = NULL;
- char *conn_id = NULL;
- pl_ctx_t *ctx = NULL;
- int nonblock = 0;
-
- if (xdata)
- dict_ret = dict_get_str (xdata, "connection-id", &conn_id);
-
- pinode = pl_inode_get (this, inode);
- if (!pinode) {
- op_errno = ENOMEM;
- goto out;
- }
+pl_common_entrylk(call_frame_t *frame, xlator_t *this, const char *volume,
+ inode_t *inode, const char *basename, entrylk_cmd cmd,
+ entrylk_type type, loc_t *loc, fd_t *fd, dict_t *xdata)
- if (frame->root->client) {
- ctx = pl_ctx_get (frame->root->client, this);
- if (!ctx) {
- op_errno = ENOMEM;
- gf_log (this->name, GF_LOG_INFO, "pl_ctx_get() failed");
- goto unwind;
- }
- }
-
- dom = get_domain (pinode, volume);
- if (!dom){
- op_errno = ENOMEM;
+{
+ int32_t op_ret = -1;
+ int32_t op_errno = 0;
+ int ret = -1;
+ char unwind = 1;
+ GF_UNUSED int dict_ret = -1;
+ pl_inode_t *pinode = NULL;
+ pl_entry_lock_t *reqlock = NULL;
+ pl_entry_lock_t *unlocked = NULL;
+ pl_dom_list_t *dom = NULL;
+ char *conn_id = NULL;
+ pl_ctx_t *ctx = NULL;
+ int nonblock = 0;
+ gf_boolean_t need_inode_unref = _gf_false;
+ posix_locks_private_t *priv = NULL;
+ struct list_head *pcontend = NULL;
+ struct list_head contend;
+ struct timespec now = {};
+
+ priv = this->private;
+
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ if (xdata)
+ dict_ret = dict_get_str(xdata, "connection-id", &conn_id);
+
+ pinode = pl_inode_get(this, inode, NULL);
+ if (!pinode) {
+ op_errno = ENOMEM;
+ goto out;
+ }
+
+ if (frame->root->client) {
+ ctx = pl_ctx_get(frame->root->client, this);
+ if (!ctx) {
+ op_errno = ENOMEM;
+ gf_log(this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+ goto unwind;
+ }
+ }
+
+ dom = get_domain(pinode, volume);
+ if (!dom) {
+ op_errno = ENOMEM;
+ goto out;
+ }
+
+ entrylk_trace_in(this, frame, volume, fd, loc, basename, cmd, type);
+
+ reqlock = new_entrylk_lock(pinode, basename, type, dom->domain, frame,
+ conn_id, &op_errno);
+ if (!reqlock) {
+ op_ret = -1;
+ goto unwind;
+ }
+
+ /* Ideally, AFTER a successful lock (both blocking and non-blocking) or
+ * an unsuccessful blocking lock operation, the inode needs to be ref'd.
+ *
+ * But doing so might give room to a race where the lock-requesting
+ * client could send a DISCONNECT just before this thread refs the inode
+ * after the locking is done, and the epoll thread could unref the inode
+ * in cleanup which means the inode's refcount would come down to 0, and
+ * the call to pl_forget() at this point destroys @pinode. Now when
+ * the io-thread executing this function tries to access pinode,
+ * it could crash on account of illegal memory access.
+ *
+ * To get around this problem, the inode is ref'd once even before
+ * adding the lock into client_list as a precautionary measure.
+ * This way even if there are DISCONNECTs, there will always be 1 extra
+ * ref on the inode, so @pinode is still alive until after the
+ * current stack unwinds.
+ */
+ pinode->inode = inode_ref(inode);
+ if (priv->revocation_secs != 0) {
+ if (cmd != ENTRYLK_UNLOCK) {
+ __entrylk_prune_stale(this, pinode, dom, reqlock);
+ } else if (priv->monkey_unlocking == _gf_true) {
+ if (pl_does_monkey_want_stuck_lock()) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "MONKEY LOCKING (forcing stuck lock)!");
+ op_ret = 0;
+ need_inode_unref = _gf_true;
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ __pl_entrylk_unref(reqlock);
+ }
+ pthread_mutex_unlock(&pinode->mutex);
goto out;
+ }
}
+ }
- entrylk_trace_in (this, frame, volume, fd, loc, basename, cmd, type);
+ switch (cmd) {
+ case ENTRYLK_LOCK_NB:
+ nonblock = 1;
+ /* fall through */
+ case ENTRYLK_LOCK:
+ if (ctx)
+ pthread_mutex_lock(&ctx->lock);
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ reqlock->pinode = pinode;
+
+ ret = __lock_entrylk(this, pinode, reqlock, nonblock, dom, &now,
+ pcontend);
+ if (ret == 0) {
+ reqlock->frame = NULL;
+ op_ret = 0;
+ } else {
+ op_errno = -ret;
+ }
- reqlock = new_entrylk_lock (pinode, basename, type, dom->domain, frame,
- conn_id);
- if (!reqlock) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+ if (ctx && (!ret || !nonblock))
+ list_add(&reqlock->client_list, &ctx->entrylk_lockers);
- switch (cmd) {
- case ENTRYLK_LOCK_NB:
- nonblock = 1;
- /* fall through */
- case ENTRYLK_LOCK:
- if (ctx)
- pthread_mutex_lock (&ctx->lock);
- pthread_mutex_lock (&pinode->mutex);
- {
- reqlock->pinode = pinode;
-
- ret = __lock_entrylk (this, pinode, reqlock, nonblock, dom);
- if (ret == 0) {
- reqlock->frame = NULL;
- op_ret = 0;
- } else {
- op_errno = -ret;
- }
-
- if (ctx && (!ret || !nonblock))
- list_add (&reqlock->client_list,
- &ctx->entrylk_lockers);
-
- if (ret == -EAGAIN && !nonblock) {
- /* blocked */
- unwind = 0;
- } else {
- __pl_entrylk_unref (reqlock);
- }
+ if (ret == -EAGAIN && !nonblock) {
+ /* blocked */
+ unwind = 0;
+ } else {
+ __pl_entrylk_unref(reqlock);
}
- pthread_mutex_unlock (&pinode->mutex);
- if (ctx)
- pthread_mutex_unlock (&ctx->lock);
- break;
+
+ /* For all but the case where a non-blocking lock
+ * attempt fails, the extra ref taken before the switch
+ * block must be negated.
+ */
+ if ((ret == -EAGAIN) && (nonblock))
+ need_inode_unref = _gf_true;
+ }
+ pthread_mutex_unlock(&pinode->mutex);
+ if (ctx)
+ pthread_mutex_unlock(&ctx->lock);
+ break;
case ENTRYLK_UNLOCK:
- if (ctx)
- pthread_mutex_lock (&ctx->lock);
- pthread_mutex_lock (&pinode->mutex);
- {
- unlocked = __unlock_entrylk (dom, reqlock);
- if (unlocked) {
- list_del_init (&unlocked->client_list);
- __pl_entrylk_unref (unlocked);
- op_ret = 0;
- } else {
- op_errno = EINVAL;
- }
- __pl_entrylk_unref (reqlock);
+ if (ctx)
+ pthread_mutex_lock(&ctx->lock);
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ /* Irrespective of whether unlock succeeds or not,
+ * the extra inode ref that was done before the switch
+ * block must be negated. Towards this,
+ * @need_inode_unref flag is set unconditionally here.
+ */
+ need_inode_unref = _gf_true;
+ unlocked = __unlock_entrylk(dom, reqlock);
+ if (unlocked) {
+ list_del_init(&unlocked->client_list);
+ __pl_entrylk_unref(unlocked);
+ op_ret = 0;
+ } else {
+ op_errno = EINVAL;
}
- pthread_mutex_unlock (&pinode->mutex);
- if (ctx)
- pthread_mutex_unlock (&ctx->lock);
+ __pl_entrylk_unref(reqlock);
+ }
+ pthread_mutex_unlock(&pinode->mutex);
+ if (ctx)
+ pthread_mutex_unlock(&ctx->lock);
- grant_blocked_entry_locks (this, pinode, dom);
+ grant_blocked_entry_locks(this, pinode, dom, &now, pcontend);
- break;
+ break;
default:
- gf_log (this->name, GF_LOG_ERROR,
- "Unexpected case in entrylk (cmd=%d). Please file"
- "a bug report at http://bugs.gluster.com", cmd);
- goto out;
- }
+ need_inode_unref = _gf_true;
+ gf_log(this->name, GF_LOG_ERROR,
+ "Unexpected case in entrylk (cmd=%d). Please file"
+ "a bug report at http://bugs.gluster.com",
+ cmd);
+ goto out;
+ }
+ /* The following (extra) unref corresponds to the ref that
+ * was done at the time the lock was granted.
+ */
+ if ((cmd == ENTRYLK_UNLOCK) && (op_ret == 0))
+ inode_unref(pinode->inode);
+
out:
- pl_update_refkeeper (this, inode);
- if (unwind) {
- entrylk_trace_out (this, frame, volume, fd, loc, basename,
- cmd, type, op_ret, op_errno);
-unwind:
- STACK_UNWIND_STRICT (entrylk, frame, op_ret, op_errno, NULL);
- } else {
- entrylk_trace_block (this, frame, volume, fd, loc, basename,
- cmd, type);
- }
+ if (need_inode_unref)
+ inode_unref(pinode->inode);
+
+ if (unwind) {
+ entrylk_trace_out(this, frame, volume, fd, loc, basename, cmd, type,
+ op_ret, op_errno);
+ unwind:
+ STACK_UNWIND_STRICT(entrylk, frame, op_ret, op_errno, NULL);
+ }
- return 0;
+ if (pcontend != NULL) {
+ entrylk_contention_notify(this, pcontend);
+ }
+
+ return 0;
}
/**
@@ -651,17 +947,16 @@ unwind:
*/
int
-pl_entrylk (call_frame_t *frame, xlator_t *this,
- const char *volume, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+pl_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+ const char *basename, entrylk_cmd cmd, entrylk_type type,
+ dict_t *xdata)
{
- pl_common_entrylk (frame, this, volume, loc->inode, basename, cmd,
- type, loc, NULL, xdata);
+ pl_common_entrylk(frame, this, volume, loc->inode, basename, cmd, type, loc,
+ NULL, xdata);
- return 0;
+ return 0;
}
-
/**
* pl_fentrylk:
*
@@ -669,185 +964,190 @@ pl_entrylk (call_frame_t *frame, xlator_t *this,
*/
int
-pl_fentrylk (call_frame_t *frame, xlator_t *this,
- const char *volume, fd_t *fd, const char *basename,
- entrylk_cmd cmd, entrylk_type type, dict_t *xdata)
+pl_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+ const char *basename, entrylk_cmd cmd, entrylk_type type,
+ dict_t *xdata)
{
- pl_common_entrylk (frame, this, volume, fd->inode, basename, cmd,
- type, NULL, fd, xdata);
+ pl_common_entrylk(frame, this, volume, fd->inode, basename, cmd, type, NULL,
+ fd, xdata);
- return 0;
+ return 0;
}
-
static void
-pl_entrylk_log_cleanup (pl_entry_lock_t *lock)
+pl_entrylk_log_cleanup(pl_entry_lock_t *lock)
{
- pl_inode_t *pinode = NULL;
- char *path = NULL;
- char *file = NULL;
-
- pinode = lock->pinode;
-
- inode_path (pinode->refkeeper, NULL, &path);
+ pl_inode_t *pinode = NULL;
- if (path)
- file = path;
- else
- file = uuid_utoa (pinode->refkeeper->gfid);
+ pinode = lock->pinode;
- gf_log (THIS->name, GF_LOG_WARNING,
- "releasing lock on %s held by "
- "{client=%p, pid=%"PRId64" lk-owner=%s}",
- file, lock->client, (uint64_t) lock->client_pid,
- lkowner_utoa (&lock->owner));
- GF_FREE (path);
+ gf_log(THIS->name, GF_LOG_WARNING,
+ "releasing lock on %s held by "
+ "{client=%p, pid=%" PRId64 " lk-owner=%s}",
+ uuid_utoa(pinode->gfid), lock->client, (uint64_t)lock->client_pid,
+ lkowner_utoa(&lock->owner));
}
-
/* Release all entrylks from this client */
int
-pl_entrylk_client_cleanup (xlator_t *this, pl_ctx_t *ctx)
+pl_entrylk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
{
- pl_entry_lock_t *tmp = NULL;
- pl_entry_lock_t *l = NULL;
- pl_dom_list_t *dom = NULL;
- pl_inode_t *pinode = NULL;
-
- struct list_head released;
- struct list_head unwind;
-
- INIT_LIST_HEAD (&released);
- INIT_LIST_HEAD (&unwind);
-
- pthread_mutex_lock (&ctx->lock);
+ posix_locks_private_t *priv;
+ pl_entry_lock_t *tmp = NULL;
+ pl_entry_lock_t *l = NULL;
+ pl_dom_list_t *dom = NULL;
+ pl_inode_t *pinode = NULL;
+ struct list_head *pcontend = NULL;
+ struct list_head released;
+ struct list_head unwind;
+ struct list_head contend;
+ struct timespec now = {};
+
+ INIT_LIST_HEAD(&released);
+ INIT_LIST_HEAD(&unwind);
+
+ priv = this->private;
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ pthread_mutex_lock(&ctx->lock);
+ {
+ list_for_each_entry_safe(l, tmp, &ctx->entrylk_lockers, client_list)
{
- list_for_each_entry_safe (l, tmp, &ctx->entrylk_lockers,
- client_list) {
- list_del_init (&l->client_list);
-
- pl_entrylk_log_cleanup (l);
-
- pinode = l->pinode;
-
- pthread_mutex_lock (&pinode->mutex);
- {
- /* If the entrylk object is part of granted list but not
- * blocked list, then perform the following actions:
- * i. delete the object from granted list;
- * ii. grant other locks (from other clients) that may
- * have been blocked on this entrylk; and
- * iii. unref the object.
- *
- * If the entrylk object (L1) is part of both granted
- * and blocked lists, then this means that a parallel
- * unlock on another entrylk (L2 say) may have 'granted'
- * L1 and added it to 'granted' list in
- * __grant_blocked_entry_locks() (although using the
- * 'blocked_locks' member). In that case, the cleanup
- * codepath must try and grant other overlapping
- * blocked entrylks from other clients, now that L1 is
- * out of their way and then unref L1 in the end, and
- * leave it to the other thread (the one executing
- * unlock codepath) to unwind L1's frame, delete it from
- * blocked_locks list, and perform the last unref on L1.
- *
- * If the entrylk object (L1) is part of blocked list
- * only, the cleanup code path must:
- * i. delete it from the blocked_locks list inside
- * this critical section,
- * ii. unwind its frame with EAGAIN,
- * iii. try and grant blocked entry locks from other
- * clients that were otherwise grantable, but were
- * blocked to avoid leaving L1 to starve forever.
- * iv. unref the object.
- */
- if (!list_empty (&l->domain_list)) {
- list_del_init (&l->domain_list);
- list_add_tail (&l->client_list,
- &released);
- } else {
- list_del_init (&l->blocked_locks);
- list_add_tail (&l->client_list,
- &unwind);
- }
- }
- pthread_mutex_unlock (&pinode->mutex);
+ pl_entrylk_log_cleanup(l);
+
+ pinode = l->pinode;
+
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ /* If the entrylk object is part of granted list but not
+ * blocked list, then perform the following actions:
+ * i. delete the object from granted list;
+ * ii. grant other locks (from other clients) that may
+ * have been blocked on this entrylk; and
+ * iii. unref the object.
+ *
+ * If the entrylk object (L1) is part of both granted
+ * and blocked lists, then this means that a parallel
+ * unlock on another entrylk (L2 say) may have 'granted'
+ * L1 and added it to 'granted' list in
+ * __grant_blocked_entry_locks() (although using the
+ * 'blocked_locks' member). In that case, the cleanup
+ * codepath must try and grant other overlapping
+ * blocked entrylks from other clients, now that L1 is
+ * out of their way and then unref L1 in the end, and
+ * leave it to the other thread (the one executing
+ * unlock codepath) to unwind L1's frame, delete it from
+ * blocked_locks list, and perform the last unref on L1.
+ *
+ * If the entrylk object (L1) is part of blocked list
+ * only, the cleanup code path must:
+ * i. delete it from the blocked_locks list inside
+ * this critical section,
+ * ii. unwind its frame with EAGAIN,
+ * iii. try and grant blocked entry locks from other
+ * clients that were otherwise grantable, but were
+ * blocked to avoid leaving L1 to starve forever.
+ * iv. unref the object.
+ */
+ list_del_init(&l->client_list);
+
+ if (!list_empty(&l->domain_list)) {
+ list_del_init(&l->domain_list);
+ list_add_tail(&l->client_list, &released);
+ } else {
+ list_del_init(&l->blocked_locks);
+ list_add_tail(&l->client_list, &unwind);
}
- }
- pthread_mutex_unlock (&ctx->lock);
+ }
+ pthread_mutex_unlock(&pinode->mutex);
+ }
+ }
+ pthread_mutex_unlock(&ctx->lock);
- list_for_each_entry_safe (l, tmp, &unwind, client_list) {
- list_del_init (&l->client_list);
+ if (!list_empty(&unwind)) {
+ list_for_each_entry_safe(l, tmp, &unwind, client_list)
+ {
+ list_del_init(&l->client_list);
- if (l->frame)
- STACK_UNWIND_STRICT (entrylk, l->frame, -1, EAGAIN,
- NULL);
- list_add_tail (&l->client_list, &released);
+ if (l->frame)
+ STACK_UNWIND_STRICT(entrylk, l->frame, -1, EAGAIN, NULL);
+ list_add_tail(&l->client_list, &released);
}
+ }
+
+ if (!list_empty(&released)) {
+ list_for_each_entry_safe(l, tmp, &released, client_list)
+ {
+ list_del_init(&l->client_list);
- list_for_each_entry_safe (l, tmp, &released, client_list) {
- list_del_init (&l->client_list);
+ pinode = l->pinode;
- pinode = l->pinode;
+ dom = get_domain(pinode, l->volume);
- dom = get_domain (pinode, l->volume);
+ grant_blocked_entry_locks(this, pinode, dom, &now, pcontend);
- grant_blocked_entry_locks (this, pinode, dom);
+ pthread_mutex_lock(&pinode->mutex);
+ {
+ __pl_entrylk_unref(l);
+ }
+ pthread_mutex_unlock(&pinode->mutex);
- pthread_mutex_lock (&pinode->mutex);
- {
- __pl_entrylk_unref (l);
- }
- pthread_mutex_unlock (&pinode->mutex);
+ inode_unref(pinode->inode);
}
+ }
- return 0;
-}
+ if (pcontend != NULL) {
+ entrylk_contention_notify(this, pcontend);
+ }
+ return 0;
+}
int32_t
-__get_entrylk_count (xlator_t *this, pl_inode_t *pl_inode)
+__get_entrylk_count(xlator_t *this, pl_inode_t *pl_inode)
{
- int32_t count = 0;
- pl_entry_lock_t *lock = NULL;
- pl_dom_list_t *dom = NULL;
+ int32_t count = 0;
+ pl_entry_lock_t *lock = NULL;
+ pl_dom_list_t *dom = NULL;
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
- list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
- count++;
- }
-
- list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
- count++;
- }
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ list_for_each_entry(lock, &dom->entrylk_list, domain_list) { count++; }
+ list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
+ {
+ count++;
}
+ }
- return count;
+ return count;
}
int32_t
-get_entrylk_count (xlator_t *this, inode_t *inode)
+get_entrylk_count(xlator_t *this, inode_t *inode)
{
- pl_inode_t *pl_inode = NULL;
- uint64_t tmp_pl_inode = 0;
- int ret = 0;
- int32_t count = 0;
+ pl_inode_t *pl_inode = NULL;
+ uint64_t tmp_pl_inode = 0;
+ int ret = 0;
+ int32_t count = 0;
- ret = inode_ctx_get (inode, this, &tmp_pl_inode);
- if (ret != 0) {
- goto out;
- }
+ ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+ if (ret != 0) {
+ goto out;
+ }
- pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
- pthread_mutex_lock (&pl_inode->mutex);
- {
- count = __get_entrylk_count (this, pl_inode);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ count = __get_entrylk_count(this, pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
out:
- return count;
+ return count;
}
diff --git a/xlators/features/locks/src/inodelk.c b/xlators/features/locks/src/inodelk.c
index c76cb7f9199..d4e51d6e0a1 100644
--- a/xlators/features/locks/src/inodelk.c
+++ b/xlators/features/locks/src/inodelk.c
@@ -7,848 +7,1168 @@
later), or the GNU General Public License, version 2 (GPLv2), in all
cases as published by the Free Software Foundation.
*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/dict.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/list.h>
+#include <glusterfs/upcall-utils.h>
#include "locks.h"
+#include "clear.h"
#include "common.h"
-inline void
-__delete_inode_lock (pl_inode_lock_t *lock)
+void
+__delete_inode_lock(pl_inode_lock_t *lock)
{
- list_del_init (&lock->list);
+ list_del_init(&lock->list);
}
-static inline void
-__pl_inodelk_ref (pl_inode_lock_t *lock)
+static void
+__pl_inodelk_ref(pl_inode_lock_t *lock)
{
- lock->ref++;
+ lock->ref++;
}
-inline void
-__pl_inodelk_unref (pl_inode_lock_t *lock)
+void
+__pl_inodelk_unref(pl_inode_lock_t *lock)
{
- lock->ref--;
- if (!lock->ref) {
- GF_FREE (lock->connection_id);
- GF_FREE (lock);
- }
+ lock->ref--;
+ if (!lock->ref) {
+ GF_FREE(lock->connection_id);
+ GF_FREE(lock);
+ }
}
-/* Check if 2 inodelks are conflicting on type. Only 2 shared locks don't conflict */
-static inline int
-inodelk_type_conflict (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+/* Check if 2 inodelks are conflicting on type. Only 2 shared locks don't
+ * conflict */
+static int
+inodelk_type_conflict(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
{
- if (l2->fl_type == F_WRLCK || l1->fl_type == F_WRLCK)
- return 1;
+ if (l2->fl_type == F_WRLCK || l1->fl_type == F_WRLCK)
+ return 1;
- return 0;
+ return 0;
}
void
-pl_print_inodelk (char *str, int size, int cmd, struct gf_flock *flock, const char *domain)
+pl_print_inodelk(char *str, int size, int cmd, struct gf_flock *flock,
+ const char *domain)
{
- char *cmd_str = NULL;
- char *type_str = NULL;
+ char *cmd_str = NULL;
+ char *type_str = NULL;
- switch (cmd) {
+ switch (cmd) {
#if F_GETLK != F_GETLK64
case F_GETLK64:
#endif
case F_GETLK:
- cmd_str = "GETLK";
- break;
+ cmd_str = "GETLK";
+ break;
#if F_SETLK != F_SETLK64
case F_SETLK64:
#endif
case F_SETLK:
- cmd_str = "SETLK";
- break;
+ cmd_str = "SETLK";
+ break;
#if F_SETLKW != F_SETLKW64
case F_SETLKW64:
#endif
case F_SETLKW:
- cmd_str = "SETLKW";
- break;
+ cmd_str = "SETLKW";
+ break;
default:
- cmd_str = "UNKNOWN";
- break;
- }
+ cmd_str = "UNKNOWN";
+ break;
+ }
- switch (flock->l_type) {
+ switch (flock->l_type) {
case F_RDLCK:
- type_str = "READ";
- break;
+ type_str = "READ";
+ break;
case F_WRLCK:
- type_str = "WRITE";
- break;
+ type_str = "WRITE";
+ break;
case F_UNLCK:
- type_str = "UNLOCK";
- break;
+ type_str = "UNLOCK";
+ break;
default:
- type_str = "UNKNOWN";
- break;
- }
-
- snprintf (str, size, "lock=INODELK, cmd=%s, type=%s, "
- "domain: %s, start=%llu, len=%llu, pid=%llu",
- cmd_str, type_str, domain,
- (unsigned long long) flock->l_start,
- (unsigned long long) flock->l_len,
- (unsigned long long) flock->l_pid);
+ type_str = "UNKNOWN";
+ break;
+ }
+
+ snprintf(str, size,
+ "lock=INODELK, cmd=%s, type=%s, "
+ "domain: %s, start=%llu, len=%llu, pid=%llu",
+ cmd_str, type_str, domain, (unsigned long long)flock->l_start,
+ (unsigned long long)flock->l_len,
+ (unsigned long long)flock->l_pid);
}
/* Determine if the two inodelks overlap reach other's lock regions */
static int
-inodelk_overlap (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelk_overlap(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
{
- return ((l1->fl_end >= l2->fl_start) &&
- (l2->fl_end >= l1->fl_start));
+ return ((l1->fl_end >= l2->fl_start) && (l2->fl_end >= l1->fl_start));
}
/* Returns true if the 2 inodelks have the same owner */
-static inline int
-same_inodelk_owner (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+static int
+same_inodelk_owner(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
{
- return (is_same_lkowner (&l1->owner, &l2->owner) &&
- (l1->client == l2->client));
+ return (is_same_lkowner(&l1->owner, &l2->owner) &&
+ (l1->client == l2->client));
}
/* Returns true if the 2 inodelks conflict with each other */
static int
-inodelk_conflict (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelk_conflict(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
{
- return (inodelk_overlap (l1, l2) &&
- inodelk_type_conflict (l1, l2));
+ return (inodelk_overlap(l1, l2) && inodelk_type_conflict(l1, l2));
}
-/* Determine if lock is grantable or not */
-static pl_inode_lock_t *
-__inodelk_grantable (pl_dom_list_t *dom, pl_inode_lock_t *lock)
+/*
+ * Check to see if the candidate lock overlaps/conflicts with the
+ * requested lock. If so, determine how old the lock is and return
+ * true if it exceeds the configured threshold, false otherwise.
+ */
+static inline gf_boolean_t
+__stale_inodelk(xlator_t *this, pl_inode_lock_t *candidate_lock,
+ pl_inode_lock_t *requested_lock, time_t *lock_age_sec)
{
- pl_inode_lock_t *l = NULL;
- pl_inode_lock_t *ret = NULL;
- if (list_empty (&dom->inodelk_list))
- goto out;
- list_for_each_entry (l, &dom->inodelk_list, list){
- if (inodelk_conflict (lock, l) &&
- !same_inodelk_owner (lock, l)) {
- ret = l;
- goto out;
- }
- }
-out:
- return ret;
+ posix_locks_private_t *priv = NULL;
+
+ priv = this->private;
+ /* Question: Should we just prune them all given the
+ * chance? Or just the locks we are attempting to acquire?
+ */
+ if (inodelk_conflict(candidate_lock, requested_lock)) {
+ *lock_age_sec = gf_time() - candidate_lock->granted_time;
+ if (*lock_age_sec > priv->revocation_secs)
+ return _gf_true;
+ }
+ return _gf_false;
}
-static pl_inode_lock_t *
-__blocked_lock_conflict (pl_dom_list_t *dom, pl_inode_lock_t *lock)
+/* Examine any locks held on this inode and potentially revoke the lock
+ * if the age exceeds revocation_secs. We will clear _only_ those locks
+ * which are granted, and then grant those locks which are blocked.
+ *
+ * Depending on how this patch works in the wild, we may expand this and
+ * introduce a heuristic which clears blocked locks as well if they
+ * are beyond a threshold.
+ */
+static gf_boolean_t
+__inodelk_prune_stale(xlator_t *this, pl_inode_t *pinode, pl_dom_list_t *dom,
+ pl_inode_lock_t *lock)
{
- pl_inode_lock_t *l = NULL;
- pl_inode_lock_t *ret = NULL;
-
- if (list_empty (&dom->blocked_inodelks))
- return NULL;
+ posix_locks_private_t *priv = NULL;
+ pl_inode_lock_t *tmp = NULL;
+ pl_inode_lock_t *lk = NULL;
+ gf_boolean_t revoke_lock = _gf_false;
+ int bcount = 0;
+ int gcount = 0;
+ int op_errno = 0;
+ clrlk_args args;
+ args.opts = NULL;
+ time_t lk_age_sec = 0;
+ uint32_t max_blocked = 0;
+ char *reason_str = NULL;
+
+ priv = this->private;
+
+ args.type = CLRLK_INODE;
+ if (priv->revocation_clear_all == _gf_true)
+ args.kind = CLRLK_ALL;
+ else
+ args.kind = CLRLK_GRANTED;
+
+ if (list_empty(&dom->inodelk_list))
+ goto out;
+
+ pthread_mutex_lock(&pinode->mutex);
+ list_for_each_entry_safe(lk, tmp, &dom->inodelk_list, list)
+ {
+ if (__stale_inodelk(this, lk, lock, &lk_age_sec) == _gf_true) {
+ revoke_lock = _gf_true;
+ reason_str = "age";
+ break;
+ }
+ }
- list_for_each_entry (l, &dom->blocked_inodelks, blocked_locks) {
- if (inodelk_conflict (lock, l)) {
- ret = l;
- goto out;
- }
+ max_blocked = priv->revocation_max_blocked;
+ if (max_blocked != 0 && revoke_lock == _gf_false) {
+ list_for_each_entry_safe(lk, tmp, &dom->blocked_inodelks, blocked_locks)
+ {
+ max_blocked--;
+ if (max_blocked == 0) {
+ revoke_lock = _gf_true;
+ reason_str = "max blocked";
+ break;
+ }
}
+ }
+ pthread_mutex_unlock(&pinode->mutex);
out:
- return ret;
+ if (revoke_lock == _gf_true) {
+ clrlk_clear_inodelk(this, pinode, dom, &args, &bcount, &gcount,
+ &op_errno);
+ gf_log(this->name, GF_LOG_WARNING,
+ "Lock revocation [reason: %s; gfid: %s; domain: %s; "
+ "age: %ld sec] - Inode lock revoked: %d granted & %d "
+ "blocked locks cleared",
+ reason_str, uuid_utoa(pinode->gfid), dom->domain, lk_age_sec,
+ gcount, bcount);
+ }
+ return revoke_lock;
}
-static int
-__owner_has_lock (pl_dom_list_t *dom, pl_inode_lock_t *newlock)
+void
+inodelk_contention_notify_check(xlator_t *this, pl_inode_lock_t *lock,
+ struct timespec *now, struct list_head *contend)
{
- pl_inode_lock_t *lock = NULL;
+ posix_locks_private_t *priv;
+ int64_t elapsed;
- list_for_each_entry (lock, &dom->inodelk_list, list) {
- if (same_inodelk_owner (lock, newlock))
- return 1;
- }
+ priv = this->private;
- list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
- if (same_inodelk_owner (lock, newlock))
- return 1;
- }
+ /* If this lock is in a list, it means that we are about to send a
+ * notification for it, so no need to do anything else. */
+ if (!list_empty(&lock->contend)) {
+ return;
+ }
+
+ elapsed = now->tv_sec;
+ elapsed -= lock->contention_time.tv_sec;
+ if (now->tv_nsec < lock->contention_time.tv_nsec) {
+ elapsed--;
+ }
+ if (elapsed < priv->notify_contention_delay) {
+ return;
+ }
- return 0;
-}
+ /* All contention notifications will be sent outside of the locked
+ * region. This means that currently granted locks might have already
+ * been unlocked by that time. To avoid the lock or the inode to be
+ * destroyed before we process them, we take an additional reference
+ * on both. */
+ inode_ref(lock->pl_inode->inode);
+ __pl_inodelk_ref(lock);
+ lock->contention_time = *now;
-/* Determines if lock can be granted and adds the lock. If the lock
- * is blocking, adds it to the blocked_inodelks list of the domain.
- */
-static int
-__lock_inodelk (xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock,
- int can_block, pl_dom_list_t *dom)
+ list_add_tail(&lock->contend, contend);
+}
+
+void
+inodelk_contention_notify(xlator_t *this, struct list_head *contend)
{
- pl_inode_lock_t *conf = NULL;
- int ret = -EINVAL;
+ struct gf_upcall up;
+ struct gf_upcall_inodelk_contention lc;
+ pl_inode_lock_t *lock;
+ pl_inode_t *pl_inode;
+ client_t *client;
+ gf_boolean_t notify;
+
+ while (!list_empty(contend)) {
+ lock = list_first_entry(contend, pl_inode_lock_t, contend);
+
+ pl_inode = lock->pl_inode;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ /* If the lock has already been released, no notification is
+ * sent. We clear the notification time in this case. */
+ notify = !list_empty(&lock->list);
+ if (!notify) {
+ lock->contention_time.tv_sec = 0;
+ lock->contention_time.tv_nsec = 0;
+ } else {
+ memcpy(&lc.flock, &lock->user_flock, sizeof(lc.flock));
+ lc.pid = lock->client_pid;
+ lc.domain = lock->volume;
+ lc.xdata = NULL;
+
+ gf_uuid_copy(up.gfid, lock->pl_inode->gfid);
+ client = (client_t *)lock->client;
+ if (client == NULL) {
+ /* A NULL client can be found if the inodelk
+ * was issued by a server side xlator. */
+ up.client_uid = NULL;
+ } else {
+ up.client_uid = client->client_uid;
+ }
+ }
- conf = __inodelk_grantable (dom, lock);
- if (conf) {
- ret = -EAGAIN;
- if (can_block == 0)
- goto out;
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ if (notify) {
+ up.event_type = GF_UPCALL_INODELK_CONTENTION;
+ up.data = &lc;
+
+ if (this->notify(this, GF_EVENT_UPCALL, &up) < 0) {
+ gf_msg_debug(this->name, 0,
+ "Inodelk contention notification "
+ "failed");
+ } else {
+ gf_msg_debug(this->name, 0,
+ "Inodelk contention notification "
+ "sent");
+ }
+ }
- gettimeofday (&lock->blkd_time, NULL);
- list_add_tail (&lock->blocked_locks, &dom->blocked_inodelks);
+ pthread_mutex_lock(&pl_inode->mutex);
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => Blocked",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
+ list_del_init(&lock->contend);
+ __pl_inodelk_unref(lock);
+ pthread_mutex_unlock(&pl_inode->mutex);
- goto out;
+ inode_unref(pl_inode->inode);
+ }
+}
+
+/* Determine if lock is grantable or not */
+static pl_inode_lock_t *
+__inodelk_grantable(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock,
+ struct timespec *now, struct list_head *contend)
+{
+ pl_inode_lock_t *l = NULL;
+ pl_inode_lock_t *ret = NULL;
+
+ list_for_each_entry(l, &dom->inodelk_list, list)
+ {
+ if (inodelk_conflict(lock, l) && !same_inodelk_owner(lock, l)) {
+ if (ret == NULL) {
+ ret = l;
+ if (contend == NULL) {
+ break;
+ }
+ }
+ inodelk_contention_notify_check(this, l, now, contend);
}
+ }
- if (__blocked_lock_conflict (dom, lock) && !(__owner_has_lock (dom, lock))) {
- ret = -EAGAIN;
- if (can_block == 0)
- goto out;
+ return ret;
+}
- gettimeofday (&lock->blkd_time, NULL);
- list_add_tail (&lock->blocked_locks, &dom->blocked_inodelks);
+static pl_inode_lock_t *
+__blocked_lock_conflict(pl_dom_list_t *dom, pl_inode_lock_t *lock)
+{
+ pl_inode_lock_t *l = NULL;
- gf_log (this->name, GF_LOG_DEBUG,
- "Lock is grantable, but blocking to prevent starvation");
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => Blocked",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
+ list_for_each_entry(l, &dom->blocked_inodelks, blocked_locks)
+ {
+ if (inodelk_conflict(lock, l)) {
+ return l;
+ }
+ }
+ return NULL;
+}
- goto out;
- }
- __pl_inodelk_ref (lock);
- gettimeofday (&lock->granted_time, NULL);
- list_add (&lock->list, &dom->inodelk_list);
+static int
+__owner_has_lock(pl_dom_list_t *dom, pl_inode_lock_t *newlock)
+{
+ pl_inode_lock_t *lock = NULL;
+
+ list_for_each_entry(lock, &dom->inodelk_list, list)
+ {
+ if (same_inodelk_owner(lock, newlock))
+ return 1;
+ }
- ret = 0;
+ list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+ {
+ if (same_inodelk_owner(lock, newlock))
+ return 1;
+ }
+ return 0;
+}
+
+static int
+__lock_blocked_add(xlator_t *this, pl_dom_list_t *dom, pl_inode_lock_t *lock,
+ int can_block)
+{
+ if (can_block == 0) {
+ goto out;
+ }
+
+ lock->blkd_time = gf_time();
+ list_add_tail(&lock->blocked_locks, &dom->blocked_inodelks);
+
+ gf_msg_trace(this->name, 0,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64
+ " - "
+ "%" PRId64 " => Blocked",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+ lock->user_flock.l_len);
+
+ pl_trace_block(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+ lock->volume);
out:
+ return -EAGAIN;
+}
+
+/* Determines if lock can be granted and adds the lock. If the lock
+ * is blocking, adds it to the blocked_inodelks list of the domain.
+ */
+static int
+__lock_inodelk(xlator_t *this, pl_inode_t *pl_inode, pl_inode_lock_t *lock,
+ int can_block, pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend)
+{
+ pl_inode_lock_t *conf = NULL;
+ int ret;
+
+ ret = pl_inode_remove_inodelk(pl_inode, lock);
+ if (ret < 0) {
return ret;
+ }
+ if (ret == 0) {
+ conf = __inodelk_grantable(this, dom, lock, now, contend);
+ }
+ if ((ret > 0) || (conf != NULL)) {
+ return __lock_blocked_add(this, dom, lock, can_block);
+ }
+
+ /* To prevent blocked locks starvation, check if there are any blocked
+ * locks thay may conflict with this lock. If there is then don't grant
+ * the lock. BUT grant the lock if the owner already has lock to allow
+ * nested locks.
+ * Example:
+ * SHD from Machine1 takes (gfid, 0-infinity) and is granted.
+ * SHD from machine2 takes (gfid, 0-infinity) and is blocked.
+ * When SHD from Machine1 takes (gfid, 0-128KB) it
+ * needs to be granted, without which the earlier lock on 0-infinity
+ * will not be unlocked by SHD from Machine1.
+ * TODO: Find why 'owner_has_lock' is checked even for blocked locks.
+ */
+ if (__blocked_lock_conflict(dom, lock) && !(__owner_has_lock(dom, lock))) {
+ if (can_block != 0) {
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Lock is grantable, but blocking to prevent "
+ "starvation");
+ }
+
+ return __lock_blocked_add(this, dom, lock, can_block);
+ }
+ __pl_inodelk_ref(lock);
+ lock->granted_time = gf_time();
+ list_add(&lock->list, &dom->inodelk_list);
+
+ return 0;
}
/* Return true if the two inodelks have exactly same lock boundaries */
static int
-inodelks_equal (pl_inode_lock_t *l1, pl_inode_lock_t *l2)
+inodelks_equal(pl_inode_lock_t *l1, pl_inode_lock_t *l2)
{
- if ((l1->fl_start == l2->fl_start) &&
- (l1->fl_end == l2->fl_end))
- return 1;
+ if ((l1->fl_start == l2->fl_start) && (l1->fl_end == l2->fl_end))
+ return 1;
- return 0;
+ return 0;
}
-
static pl_inode_lock_t *
-find_matching_inodelk (pl_inode_lock_t *lock, pl_dom_list_t *dom)
+find_matching_inodelk(pl_inode_lock_t *lock, pl_dom_list_t *dom)
{
- pl_inode_lock_t *l = NULL;
- list_for_each_entry (l, &dom->inodelk_list, list) {
- if (inodelks_equal (l, lock) &&
- same_inodelk_owner (l, lock))
- return l;
- }
- return NULL;
+ pl_inode_lock_t *l = NULL;
+ list_for_each_entry(l, &dom->inodelk_list, list)
+ {
+ if (inodelks_equal(l, lock) && same_inodelk_owner(l, lock))
+ return l;
+ }
+ return NULL;
}
/* Set F_UNLCK removes a lock which has the exact same lock boundaries
* as the UNLCK lock specifies. If such a lock is not found, returns invalid
*/
static pl_inode_lock_t *
-__inode_unlock_lock (xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
+__inode_unlock_lock(xlator_t *this, pl_inode_lock_t *lock, pl_dom_list_t *dom)
{
-
- pl_inode_lock_t *conf = NULL;
-
- conf = find_matching_inodelk (lock, dom);
- if (!conf) {
- gf_log (this->name, GF_LOG_ERROR,
- " Matching lock not found for unlock %llu-%llu, by %s "
- "on %p", (unsigned long long)lock->fl_start,
- (unsigned long long)lock->fl_end,
- lkowner_utoa (&lock->owner), lock->client);
- goto out;
- }
- __delete_inode_lock (conf);
- gf_log (this->name, GF_LOG_DEBUG,
- " Matching lock found for unlock %llu-%llu, by %s on %p",
- (unsigned long long)lock->fl_start,
- (unsigned long long)lock->fl_end, lkowner_utoa (&lock->owner),
- lock->client);
+ pl_inode_lock_t *conf = NULL;
+ inode_t *inode = NULL;
+
+ inode = lock->pl_inode->inode;
+
+ conf = find_matching_inodelk(lock, dom);
+ if (!conf) {
+ gf_log(this->name, GF_LOG_ERROR,
+ " Matching lock not found for unlock %llu-%llu, by %s "
+ "on %p for gfid:%s",
+ (unsigned long long)lock->fl_start,
+ (unsigned long long)lock->fl_end, lkowner_utoa(&lock->owner),
+ lock->client, inode ? uuid_utoa(inode->gfid) : "UNKNOWN");
+ goto out;
+ }
+ __delete_inode_lock(conf);
+ gf_log(this->name, GF_LOG_DEBUG,
+ " Matching lock found for unlock %llu-%llu, by %s on %p for gfid:%s",
+ (unsigned long long)lock->fl_start, (unsigned long long)lock->fl_end,
+ lkowner_utoa(&lock->owner), lock->client,
+ inode ? uuid_utoa(inode->gfid) : "UNKNOWN");
out:
- return conf;
+ return conf;
}
-
-static void
-__grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
- struct list_head *granted, pl_dom_list_t *dom)
+void
+__grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted, pl_dom_list_t *dom,
+ struct timespec *now, struct list_head *contend)
{
- int bl_ret = 0;
- pl_inode_lock_t *bl = NULL;
- pl_inode_lock_t *tmp = NULL;
-
- struct list_head blocked_list;
+ pl_inode_lock_t *bl = NULL;
+ pl_inode_lock_t *tmp = NULL;
- INIT_LIST_HEAD (&blocked_list);
- list_splice_init (&dom->blocked_inodelks, &blocked_list);
+ struct list_head blocked_list;
- list_for_each_entry_safe (bl, tmp, &blocked_list, blocked_locks) {
+ INIT_LIST_HEAD(&blocked_list);
+ list_splice_init(&dom->blocked_inodelks, &blocked_list);
- list_del_init (&bl->blocked_locks);
+ list_for_each_entry_safe(bl, tmp, &blocked_list, blocked_locks)
+ {
+ list_del_init(&bl->blocked_locks);
- bl_ret = __lock_inodelk (this, pl_inode, bl, 1, dom);
+ bl->status = __lock_inodelk(this, pl_inode, bl, 1, dom, now, contend);
- if (bl_ret == 0) {
- list_add (&bl->blocked_locks, granted);
- }
+ if (bl->status != -EAGAIN) {
+ list_add_tail(&bl->blocked_locks, granted);
}
- return;
+ }
}
-/* Grant all inodelks blocked on a lock */
void
-grant_blocked_inode_locks (xlator_t *this, pl_inode_t *pl_inode,
- pl_dom_list_t *dom)
+unwind_granted_inodes(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted)
{
- struct list_head granted;
- pl_inode_lock_t *lock;
- pl_inode_lock_t *tmp;
-
- INIT_LIST_HEAD (&granted);
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __grant_blocked_inode_locks (this, pl_inode, &granted, dom);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
-
- list_for_each_entry_safe (lock, tmp, &granted, blocked_locks) {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => Granted",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
-
- pl_trace_out (this, lock->frame, NULL, NULL, F_SETLKW,
- &lock->user_flock, 0, 0, lock->volume);
-
- STACK_UNWIND_STRICT (inodelk, lock->frame, 0, 0, NULL);
- lock->frame = NULL;
+ pl_inode_lock_t *lock;
+ pl_inode_lock_t *tmp;
+ int32_t op_ret;
+ int32_t op_errno;
+
+ list_for_each_entry_safe(lock, tmp, granted, blocked_locks)
+ {
+ if (lock->status == 0) {
+ op_ret = 0;
+ op_errno = 0;
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+ " => Granted",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->user_flock.l_start, lock->user_flock.l_len);
+ } else {
+ op_ret = -1;
+ op_errno = -lock->status;
}
+ pl_trace_out(this, lock->frame, NULL, NULL, F_SETLKW, &lock->user_flock,
+ op_ret, op_errno, lock->volume);
+
+ STACK_UNWIND_STRICT(inodelk, lock->frame, op_ret, op_errno, NULL);
+ lock->frame = NULL;
+ }
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(lock, tmp, granted, blocked_locks)
{
- list_for_each_entry_safe (lock, tmp, &granted, blocked_locks) {
- list_del_init (&lock->blocked_locks);
- __pl_inodelk_unref (lock);
- }
+ list_del_init(&lock->blocked_locks);
+ __pl_inodelk_unref(lock);
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
}
-
-static void
-pl_inodelk_log_cleanup (pl_inode_lock_t *lock)
+/* Grant all inodelks blocked on a lock */
+void
+grant_blocked_inode_locks(xlator_t *this, pl_inode_t *pl_inode,
+ pl_dom_list_t *dom, struct timespec *now,
+ struct list_head *contend)
{
- pl_inode_t *pl_inode = NULL;
- char *path = NULL;
- char *file = NULL;
-
- pl_inode = lock->pl_inode;
+ struct list_head granted;
- inode_path (pl_inode->refkeeper, NULL, &path);
+ INIT_LIST_HEAD(&granted);
- if (path)
- file = path;
- else
- file = uuid_utoa (pl_inode->refkeeper->gfid);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __grant_blocked_inode_locks(this, pl_inode, &granted, dom, now,
+ contend);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- gf_log (THIS->name, GF_LOG_WARNING,
- "releasing lock on %s held by "
- "{client=%p, pid=%"PRId64" lk-owner=%s}",
- file, lock->client, (uint64_t) lock->client_pid,
- lkowner_utoa (&lock->owner));
- GF_FREE (path);
+ unwind_granted_inodes(this, pl_inode, &granted);
}
-
-/* Release all inodelks from this client */
-int
-pl_inodelk_client_cleanup (xlator_t *this, pl_ctx_t *ctx)
+static void
+pl_inodelk_log_cleanup(pl_inode_lock_t *lock)
{
- pl_inode_lock_t *tmp = NULL;
- pl_inode_lock_t *l = NULL;
- pl_dom_list_t *dom = NULL;
- pl_inode_t *pl_inode = NULL;
+ pl_inode_t *pl_inode = NULL;
- struct list_head released;
- struct list_head unwind;
+ pl_inode = lock->pl_inode;
- INIT_LIST_HEAD (&released);
- INIT_LIST_HEAD (&unwind);
+ gf_log(THIS->name, GF_LOG_WARNING,
+ "releasing lock on %s held by "
+ "{client=%p, pid=%" PRId64 " lk-owner=%s}",
+ uuid_utoa(pl_inode->gfid), lock->client, (uint64_t)lock->client_pid,
+ lkowner_utoa(&lock->owner));
+}
- pthread_mutex_lock (&ctx->lock);
+/* Release all inodelks from this client */
+int
+pl_inodelk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
+{
+ posix_locks_private_t *priv;
+ pl_inode_lock_t *tmp = NULL;
+ pl_inode_lock_t *l = NULL;
+ pl_dom_list_t *dom = NULL;
+ pl_inode_t *pl_inode = NULL;
+ struct list_head *pcontend = NULL;
+ struct list_head released;
+ struct list_head unwind;
+ struct list_head contend;
+ struct timespec now = {};
+
+ priv = this->private;
+
+ INIT_LIST_HEAD(&released);
+ INIT_LIST_HEAD(&unwind);
+
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ pthread_mutex_lock(&ctx->lock);
+ {
+ list_for_each_entry_safe(l, tmp, &ctx->inodelk_lockers, client_list)
{
- list_for_each_entry_safe (l, tmp, &ctx->inodelk_lockers,
- client_list) {
- list_del_init (&l->client_list);
-
- pl_inodelk_log_cleanup (l);
-
- pl_inode = l->pl_inode;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- /* If the inodelk object is part of granted list but not
- * blocked list, then perform the following actions:
- * i. delete the object from granted list;
- * ii. grant other locks (from other clients) that may
- * have been blocked on this inodelk; and
- * iii. unref the object.
- *
- * If the inodelk object (L1) is part of both granted
- * and blocked lists, then this means that a parallel
- * unlock on another inodelk (L2 say) may have 'granted'
- * L1 and added it to 'granted' list in
- * __grant_blocked_node_locks() (although using the
- * 'blocked_locks' member). In that case, the cleanup
- * codepath must try and grant other overlapping
- * blocked inodelks from other clients, now that L1 is
- * out of their way and then unref L1 in the end, and
- * leave it to the other thread (the one executing
- * unlock codepath) to unwind L1's frame, delete it from
- * blocked_locks list, and perform the last unref on L1.
- *
- * If the inodelk object (L1) is part of blocked list
- * only, the cleanup code path must:
- * i. delete it from the blocked_locks list inside
- * this critical section,
- * ii. unwind its frame with EAGAIN,
- * iii. try and grant blocked inode locks from other
- * clients that were otherwise grantable, but just
- * got blocked to avoid leaving L1 to starve
- * forever.
- * iv. unref the object.
- */
- if (!list_empty (&l->list)) {
- __delete_inode_lock (l);
- list_add_tail (&l->client_list,
- &released);
- } else {
- list_del_init(&l->blocked_locks);
- list_add_tail (&l->client_list,
- &unwind);
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pl_inodelk_log_cleanup(l);
+
+ pl_inode = l->pl_inode;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ /* If the inodelk object is part of granted list but not
+ * blocked list, then perform the following actions:
+ * i. delete the object from granted list;
+ * ii. grant other locks (from other clients) that may
+ * have been blocked on this inodelk; and
+ * iii. unref the object.
+ *
+ * If the inodelk object (L1) is part of both granted
+ * and blocked lists, then this means that a parallel
+ * unlock on another inodelk (L2 say) may have 'granted'
+ * L1 and added it to 'granted' list in
+ * __grant_blocked_inode_locks() (although using the
+ * 'blocked_locks' member). In that case, the cleanup
+ * codepath must try and grant other overlapping
+ * blocked inodelks from other clients, now that L1 is
+ * out of their way and then unref L1 in the end, and
+ * leave it to the other thread (the one executing
+ * unlock codepath) to unwind L1's frame, delete it from
+ * blocked_locks list, and perform the last unref on L1.
+ *
+ * If the inodelk object (L1) is part of blocked list
+ * only, the cleanup code path must:
+ * i. delete it from the blocked_locks list inside
+ * this critical section,
+ * ii. unwind its frame with EAGAIN,
+ * iii. try and grant blocked inode locks from other
+ * clients that were otherwise grantable, but just
+ * got blocked to avoid leaving L1 to starve
+ * forever.
+ * iv. unref the object.
+ */
+ list_del_init(&l->client_list);
+
+ if (!list_empty(&l->list)) {
+ __delete_inode_lock(l);
+ list_add_tail(&l->client_list, &released);
+ } else {
+ list_del_init(&l->blocked_locks);
+ list_add_tail(&l->client_list, &unwind);
}
- }
- pthread_mutex_unlock (&ctx->lock);
-
- list_for_each_entry_safe (l, tmp, &unwind, client_list) {
- list_del_init (&l->client_list);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+ }
+ pthread_mutex_unlock(&ctx->lock);
- if (l->frame)
- STACK_UNWIND_STRICT (inodelk, l->frame, -1, EAGAIN,
- NULL);
- list_add_tail (&l->client_list, &released);
+ if (!list_empty(&unwind)) {
+ list_for_each_entry_safe(l, tmp, &unwind, client_list)
+ {
+ list_del_init(&l->client_list);
+ if (l->frame)
+ STACK_UNWIND_STRICT(inodelk, l->frame, -1, EAGAIN, NULL);
+ list_add_tail(&l->client_list, &released);
}
+ }
- list_for_each_entry_safe (l, tmp, &released, client_list) {
- list_del_init (&l->client_list);
+ if (!list_empty(&released)) {
+ list_for_each_entry_safe(l, tmp, &released, client_list)
+ {
+ list_del_init(&l->client_list);
- pl_inode = l->pl_inode;
+ pl_inode = l->pl_inode;
- dom = get_domain (pl_inode, l->volume);
+ dom = get_domain(pl_inode, l->volume);
- grant_blocked_inode_locks (this, pl_inode, dom);
+ grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __pl_inodelk_unref (l);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __pl_inodelk_unref(l);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ inode_unref(pl_inode->inode);
}
+ }
- return 0;
-}
+ if (pcontend != NULL) {
+ inodelk_contention_notify(this, pcontend);
+ }
+ return 0;
+}
static int
-pl_inode_setlk (xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode,
- pl_inode_lock_t *lock, int can_block, pl_dom_list_t *dom)
+pl_inode_setlk(xlator_t *this, pl_ctx_t *ctx, pl_inode_t *pl_inode,
+ pl_inode_lock_t *lock, int can_block, pl_dom_list_t *dom,
+ inode_t *inode)
{
- int ret = -EINVAL;
- pl_inode_lock_t *retlock = NULL;
- gf_boolean_t unref = _gf_true;
-
- lock->pl_inode = pl_inode;
-
- if (ctx)
- pthread_mutex_lock (&ctx->lock);
- pthread_mutex_lock (&pl_inode->mutex);
- {
- if (lock->fl_type != F_UNLCK) {
- ret = __lock_inodelk (this, pl_inode, lock, can_block, dom);
- if (ret == 0) {
- lock->frame = NULL;
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => OK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->fl_start,
- lock->fl_end);
- } else if (ret == -EAGAIN) {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => NOK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
- if (can_block)
- unref = _gf_false;
- }
-
- if (ctx && (!ret || can_block))
- list_add_tail (&lock->client_list,
- &ctx->inodelk_lockers);
- } else {
- retlock = __inode_unlock_lock (this, lock, dom);
- if (!retlock) {
- gf_log (this->name, GF_LOG_DEBUG,
- "Bad Unlock issued on Inode lock");
- ret = -EINVAL;
- goto out;
- }
- list_del_init (&retlock->client_list);
- __pl_inodelk_unref (retlock);
-
- ret = 0;
+ posix_locks_private_t *priv = NULL;
+ int ret = -EINVAL;
+ pl_inode_lock_t *retlock = NULL;
+ gf_boolean_t unref = _gf_true;
+ gf_boolean_t need_inode_unref = _gf_false;
+ struct list_head *pcontend = NULL;
+ struct list_head contend;
+ struct list_head wake;
+ struct timespec now = {};
+ short fl_type;
+
+ lock->pl_inode = pl_inode;
+ fl_type = lock->fl_type;
+
+ priv = this->private;
+
+ /* Ideally, AFTER a successful lock (both blocking and non-blocking) or
+ * an unsuccessful blocking lock operation, the inode needs to be ref'd.
+ *
+ * But doing so might give room to a race where the lock-requesting
+ * client could send a DISCONNECT just before this thread refs the inode
+ * after the locking is done, and the epoll thread could unref the inode
+ * in cleanup which means the inode's refcount would come down to 0, and
+ * the call to pl_forget() at this point destroys @pl_inode. Now when
+ * the io-thread executing this function tries to access pl_inode,
+ * it could crash on account of illegal memory access.
+ *
+ * To get around this problem, the inode is ref'd once even before
+ * adding the lock into client_list as a precautionary measure.
+ * This way even if there are DISCONNECTs, there will always be 1 extra
+ * ref on the inode, so @pl_inode is still alive until after the
+ * current stack unwinds.
+ */
+ pl_inode->inode = inode_ref(inode);
+
+ if (priv->revocation_secs != 0) {
+ if (lock->fl_type != F_UNLCK) {
+ __inodelk_prune_stale(this, pl_inode, dom, lock);
+ } else if (priv->monkey_unlocking == _gf_true) {
+ if (pl_does_monkey_want_stuck_lock()) {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __pl_inodelk_unref(lock);
}
-out:
- if (unref)
- __pl_inodelk_unref (lock);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ inode_unref(pl_inode->inode);
+ gf_log(this->name, GF_LOG_WARNING,
+ "MONKEY LOCKING (forcing stuck lock)!");
+ return 0;
+ }
}
- pthread_mutex_unlock (&pl_inode->mutex);
- if (ctx)
- pthread_mutex_unlock (&ctx->lock);
+ }
+
+ if (priv->notify_contention) {
+ pcontend = &contend;
+ INIT_LIST_HEAD(pcontend);
+ timespec_now(&now);
+ }
+
+ INIT_LIST_HEAD(&wake);
+
+ if (ctx)
+ pthread_mutex_lock(&ctx->lock);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (lock->fl_type != F_UNLCK) {
+ ret = __lock_inodelk(this, pl_inode, lock, can_block, dom, &now,
+ pcontend);
+ if (ret == 0) {
+ lock->frame = NULL;
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+ " => OK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->fl_start, lock->fl_end);
+ } else if (ret == -EAGAIN) {
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+ " => NOK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
+ lock->client_pid, lkowner_utoa(&lock->owner),
+ lock->user_flock.l_start, lock->user_flock.l_len);
+ if (can_block) {
+ unref = _gf_false;
+ }
+ }
+ /* For all but the case where a non-blocking lock attempt fails
+ * with -EAGAIN, the extra ref taken at the start of this function
+ * must be negated. */
+ need_inode_unref = (ret != 0) && ((ret != -EAGAIN) || !can_block);
+ if (ctx && !need_inode_unref) {
+ list_add_tail(&lock->client_list, &ctx->inodelk_lockers);
+ }
+ } else {
+ /* Irrespective of whether unlock succeeds or not,
+ * the extra inode ref that was done at the start of
+ * this function must be negated. Towards this,
+ * @need_inode_unref flag is set unconditionally here.
+ */
+ need_inode_unref = _gf_true;
+ retlock = __inode_unlock_lock(this, lock, dom);
+ if (!retlock) {
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Bad Unlock issued on Inode lock");
+ ret = -EINVAL;
+ goto out;
+ }
+ list_del_init(&retlock->client_list);
+ __pl_inodelk_unref(retlock);
- grant_blocked_inode_locks (this, pl_inode, dom);
+ pl_inode_remove_unlocked(this, pl_inode, &wake);
- return ret;
+ ret = 0;
+ }
+ out:
+ if (unref)
+ __pl_inodelk_unref(lock);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ if (ctx)
+ pthread_mutex_unlock(&ctx->lock);
+
+ pl_inode_remove_wake(&wake);
+
+ /* The following (extra) unref corresponds to the ref that
+ * was done at the time the lock was granted.
+ */
+ if ((fl_type == F_UNLCK) && (ret == 0)) {
+ inode_unref(pl_inode->inode);
+ grant_blocked_inode_locks(this, pl_inode, dom, &now, pcontend);
+ }
+
+ if (need_inode_unref) {
+ inode_unref(pl_inode->inode);
+ }
+
+ if (pcontend != NULL) {
+ inodelk_contention_notify(this, pcontend);
+ }
+
+ return ret;
}
/* Create a new inode_lock_t */
-pl_inode_lock_t *
-new_inode_lock (struct gf_flock *flock, client_t *client, pid_t client_pid,
- call_frame_t *frame, xlator_t *this, const char *volume,
- char *conn_id)
+static pl_inode_lock_t *
+new_inode_lock(struct gf_flock *flock, client_t *client, pid_t client_pid,
+ call_frame_t *frame, xlator_t *this, const char *volume,
+ char *conn_id, int32_t *op_errno)
{
- pl_inode_lock_t *lock = NULL;
+ pl_inode_lock_t *lock = NULL;
+
+ if (!pl_is_lk_owner_valid(&frame->root->lk_owner, frame->root->client)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ lock = GF_CALLOC(1, sizeof(*lock), gf_locks_mt_pl_inode_lock_t);
+ if (!lock) {
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ lock->fl_start = flock->l_start;
+ lock->fl_type = flock->l_type;
+
+ if (flock->l_len == 0)
+ lock->fl_end = LLONG_MAX;
+ else
+ lock->fl_end = flock->l_start + flock->l_len - 1;
+
+ lock->client = client;
+ lock->client_pid = client_pid;
+ lock->volume = volume;
+ lock->owner = frame->root->lk_owner;
+ lock->frame = frame;
+ lock->this = this;
+
+ if (conn_id) {
+ lock->connection_id = gf_strdup(conn_id);
+ }
+
+ INIT_LIST_HEAD(&lock->list);
+ INIT_LIST_HEAD(&lock->blocked_locks);
+ INIT_LIST_HEAD(&lock->client_list);
+ INIT_LIST_HEAD(&lock->contend);
+ __pl_inodelk_ref(lock);
- lock = GF_CALLOC (1, sizeof (*lock),
- gf_locks_mt_pl_inode_lock_t);
- if (!lock) {
- return NULL;
- }
-
- lock->fl_start = flock->l_start;
- lock->fl_type = flock->l_type;
-
- if (flock->l_len == 0)
- lock->fl_end = LLONG_MAX;
- else
- lock->fl_end = flock->l_start + flock->l_len - 1;
-
- lock->client = client;
- lock->client_pid = client_pid;
- lock->volume = volume;
- lock->owner = frame->root->lk_owner;
- lock->frame = frame;
- lock->this = this;
-
- if (conn_id) {
- lock->connection_id = gf_strdup (conn_id);
- }
-
- INIT_LIST_HEAD (&lock->list);
- INIT_LIST_HEAD (&lock->blocked_locks);
- INIT_LIST_HEAD (&lock->client_list);
- __pl_inodelk_ref (lock);
-
- return lock;
+out:
+ return lock;
}
int32_t
-_pl_convert_volume (const char *volume, char **res)
+_pl_convert_volume(const char *volume, char **res)
{
- char *mdata_vol = NULL;
- int ret = 0;
+ char *mdata_vol = NULL;
+ int ret = 0;
- mdata_vol = strrchr (volume, ':');
- //if the volume already ends with :metadata don't bother
- if (mdata_vol && (strcmp (mdata_vol, ":metadata") == 0))
- return 0;
-
- ret = gf_asprintf (res, "%s:metadata", volume);
- if (ret <= 0)
- return ENOMEM;
+ mdata_vol = strrchr(volume, ':');
+ // if the volume already ends with :metadata don't bother
+ if (mdata_vol && (strcmp(mdata_vol, ":metadata") == 0))
return 0;
+
+ ret = gf_asprintf(res, "%s:metadata", volume);
+ if (ret <= 0)
+ return ENOMEM;
+ return 0;
}
int32_t
-_pl_convert_volume_for_special_range (struct gf_flock *flock,
- const char *volume, char **res)
+_pl_convert_volume_for_special_range(struct gf_flock *flock, const char *volume,
+ char **res)
{
- int32_t ret = 0;
+ int32_t ret = 0;
- if ((flock->l_start == LLONG_MAX -1) &&
- (flock->l_len == 0)) {
- ret = _pl_convert_volume (volume, res);
- }
+ if ((flock->l_start == LLONG_MAX - 1) && (flock->l_len == 0)) {
+ ret = _pl_convert_volume(volume, res);
+ }
- return ret;
+ return ret;
}
/* Common inodelk code called from pl_inodelk and pl_finodelk */
int
-pl_common_inodelk (call_frame_t *frame, xlator_t *this,
- const char *volume, inode_t *inode, int32_t cmd,
- struct gf_flock *flock, loc_t *loc, fd_t *fd, dict_t *xdata)
-{
- int32_t op_ret = -1;
- int32_t op_errno = 0;
- int ret = -1;
- GF_UNUSED int dict_ret = -1;
- int can_block = 0;
- pl_inode_t * pinode = NULL;
- pl_inode_lock_t * reqlock = NULL;
- pl_dom_list_t * dom = NULL;
- char *res = NULL;
- char *res1 = NULL;
- char *conn_id = NULL;
- pl_ctx_t *ctx = NULL;
-
- if (xdata)
- dict_ret = dict_get_str (xdata, "connection-id", &conn_id);
-
- VALIDATE_OR_GOTO (frame, out);
- VALIDATE_OR_GOTO (inode, unwind);
- VALIDATE_OR_GOTO (flock, unwind);
-
- if ((flock->l_start < 0) || (flock->l_len < 0)) {
- op_errno = EINVAL;
- goto unwind;
- }
-
- op_errno = _pl_convert_volume_for_special_range (flock, volume, &res);
- if (op_errno)
- goto unwind;
- if (res)
- volume = res;
-
- pl_trace_in (this, frame, fd, loc, cmd, flock, volume);
-
- if (frame->root->client) {
- ctx = pl_ctx_get (frame->root->client, this);
- if (!ctx) {
- op_errno = ENOMEM;
- gf_log (this->name, GF_LOG_INFO, "pl_ctx_get() failed");
- goto unwind;
- }
- }
-
- pinode = pl_inode_get (this, inode);
- if (!pinode) {
- op_errno = ENOMEM;
- goto unwind;
+pl_common_inodelk(call_frame_t *frame, xlator_t *this, const char *volume,
+ inode_t *inode, int32_t cmd, struct gf_flock *flock,
+ loc_t *loc, fd_t *fd, dict_t *xdata)
+{
+ int32_t op_ret = -1;
+ int32_t op_errno = 0;
+ int ret = -1;
+ GF_UNUSED int dict_ret = -1;
+ int can_block = 0;
+ short lock_type = 0;
+ pl_inode_t *pinode = NULL;
+ pl_inode_lock_t *reqlock = NULL;
+ pl_dom_list_t *dom = NULL;
+ char *res = NULL;
+ char *res1 = NULL;
+ char *conn_id = NULL;
+ pl_ctx_t *ctx = NULL;
+
+ if (xdata)
+ dict_ret = dict_get_str(xdata, "connection-id", &conn_id);
+
+ VALIDATE_OR_GOTO(frame, out);
+ VALIDATE_OR_GOTO(inode, unwind);
+ VALIDATE_OR_GOTO(flock, unwind);
+
+ if ((flock->l_start < 0) || (flock->l_len < 0)) {
+ op_errno = EINVAL;
+ goto unwind;
+ }
+
+ op_errno = _pl_convert_volume_for_special_range(flock, volume, &res);
+ if (op_errno)
+ goto unwind;
+ if (res)
+ volume = res;
+
+ pl_trace_in(this, frame, fd, loc, cmd, flock, volume);
+
+ if (frame->root->client) {
+ ctx = pl_ctx_get(frame->root->client, this);
+ if (!ctx) {
+ op_errno = ENOMEM;
+ gf_log(this->name, GF_LOG_INFO, "pl_ctx_get() failed");
+ goto unwind;
}
+ }
- dom = get_domain (pinode, volume);
- if (!dom) {
- op_errno = ENOMEM;
- goto unwind;
- }
+ pinode = pl_inode_get(this, inode, NULL);
+ if (!pinode) {
+ op_errno = ENOMEM;
+ goto unwind;
+ }
- reqlock = new_inode_lock (flock, frame->root->client, frame->root->pid,
- frame, this, volume, conn_id);
+ dom = get_domain(pinode, volume);
+ if (!dom) {
+ op_errno = ENOMEM;
+ goto unwind;
+ }
- if (!reqlock) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+ reqlock = new_inode_lock(flock, frame->root->client, frame->root->pid,
+ frame, this, dom->domain, conn_id, &op_errno);
+ if (!reqlock) {
+ op_ret = -1;
+ goto unwind;
+ }
- switch (cmd) {
+ switch (cmd) {
case F_SETLKW:
- can_block = 1;
+ can_block = 1;
- /* fall through */
+ /* fall through */
case F_SETLK:
- memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
- ret = pl_inode_setlk (this, ctx, pinode, reqlock, can_block,
- dom);
-
- if (ret < 0) {
- if ((can_block) && (F_UNLCK != flock->l_type)) {
- pl_trace_block (this, frame, fd, loc,
- cmd, flock, volume);
- goto out;
- }
- gf_log (this->name, GF_LOG_TRACE, "returning EAGAIN");
- op_errno = -ret;
- goto unwind;
+ lock_type = flock->l_type;
+ memcpy(&reqlock->user_flock, flock, sizeof(struct gf_flock));
+ ret = pl_inode_setlk(this, ctx, pinode, reqlock, can_block, dom,
+ inode);
+
+ if (ret < 0) {
+ if (ret == -EAGAIN) {
+ if (can_block && (F_UNLCK != lock_type)) {
+ goto out;
+ }
+ gf_log(this->name, GF_LOG_TRACE, "returning EAGAIN");
+ } else {
+ gf_log(this->name, GF_LOG_TRACE, "returning %d", ret);
}
- break;
+ op_errno = -ret;
+ goto unwind;
+ }
+ break;
default:
- op_errno = ENOTSUP;
- gf_log (this->name, GF_LOG_DEBUG,
- "Lock command F_GETLK not supported for [f]inodelk "
- "(cmd=%d)",
- cmd);
- goto unwind;
- }
+ op_errno = ENOTSUP;
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Lock command F_GETLK not supported for [f]inodelk "
+ "(cmd=%d)",
+ cmd);
+ goto unwind;
+ }
- op_ret = 0;
+ op_ret = 0;
unwind:
- if ((inode != NULL) && (flock !=NULL)) {
- pl_update_refkeeper (this, inode);
- pl_trace_out (this, frame, fd, loc, cmd, flock, op_ret, op_errno, volume);
- }
+ if (flock != NULL)
+ pl_trace_out(this, frame, fd, loc, cmd, flock, op_ret, op_errno,
+ volume);
- STACK_UNWIND_STRICT (inodelk, frame, op_ret, op_errno, NULL);
+ STACK_UNWIND_STRICT(inodelk, frame, op_ret, op_errno, NULL);
out:
- GF_FREE (res);
- GF_FREE (res1);
- return 0;
+ GF_FREE(res);
+ GF_FREE(res1);
+ return 0;
}
int
-pl_inodelk (call_frame_t *frame, xlator_t *this,
- const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock,
- dict_t *xdata)
+pl_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+ int32_t cmd, struct gf_flock *flock, dict_t *xdata)
{
- pl_common_inodelk (frame, this, volume, loc->inode, cmd, flock,
- loc, NULL, xdata);
+ pl_common_inodelk(frame, this, volume, loc->inode, cmd, flock, loc, NULL,
+ xdata);
- return 0;
+ return 0;
}
int
-pl_finodelk (call_frame_t *frame, xlator_t *this,
- const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
- dict_t *xdata)
+pl_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+ int32_t cmd, struct gf_flock *flock, dict_t *xdata)
{
- pl_common_inodelk (frame, this, volume, fd->inode, cmd, flock,
- NULL, fd, xdata);
-
- return 0;
+ pl_common_inodelk(frame, this, volume, fd->inode, cmd, flock, NULL, fd,
+ xdata);
+ return 0;
}
-static inline int32_t
-__get_inodelk_dom_count (pl_dom_list_t *dom)
+static int32_t
+__get_inodelk_dom_count(pl_dom_list_t *dom)
{
- pl_inode_lock_t *lock = NULL;
- int32_t count = 0;
-
- list_for_each_entry (lock, &dom->inodelk_list, list) {
- count++;
- }
- list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
- count++;
- }
- return count;
+ pl_inode_lock_t *lock = NULL;
+ int32_t count = 0;
+
+ list_for_each_entry(lock, &dom->inodelk_list, list) { count++; }
+ list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+ {
+ count++;
+ }
+ return count;
}
/* Returns the no. of locks (blocked/granted) held on a given domain name
* If @domname is NULL, returns the no. of locks in all the domains present.
* If @domname is non-NULL and non-existent, returns 0 */
int32_t
-__get_inodelk_count (xlator_t *this, pl_inode_t *pl_inode, char *domname)
+__get_inodelk_count(xlator_t *this, pl_inode_t *pl_inode, char *domname)
{
- int32_t count = 0;
- pl_dom_list_t *dom = NULL;
-
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
- if (domname) {
- if (strcmp (domname, dom->domain) == 0) {
- count = __get_inodelk_dom_count (dom);
- goto out;
- }
-
- } else {
- /* Counting locks from all domains */
- count += __get_inodelk_dom_count (dom);
+ int32_t count = 0;
+ pl_dom_list_t *dom = NULL;
+
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ if (domname) {
+ if (strcmp(domname, dom->domain) == 0) {
+ count = __get_inodelk_dom_count(dom);
+ goto out;
+ }
- }
+ } else {
+ /* Counting locks from all domains */
+ count += __get_inodelk_dom_count(dom);
}
+ }
out:
- return count;
+ return count;
}
int32_t
-get_inodelk_count (xlator_t *this, inode_t *inode, char *domname)
+get_inodelk_count(xlator_t *this, inode_t *inode, char *domname)
{
- pl_inode_t *pl_inode = NULL;
- uint64_t tmp_pl_inode = 0;
- int ret = 0;
- int32_t count = 0;
+ pl_inode_t *pl_inode = NULL;
+ uint64_t tmp_pl_inode = 0;
+ int ret = 0;
+ int32_t count = 0;
- ret = inode_ctx_get (inode, this, &tmp_pl_inode);
- if (ret != 0) {
- goto out;
- }
+ ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+ if (ret != 0) {
+ goto out;
+ }
- pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
- pthread_mutex_lock (&pl_inode->mutex);
- {
- count = __get_inodelk_count (this, pl_inode, domname);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ count = __get_inodelk_count(this, pl_inode, domname);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
out:
- return count;
+ return count;
}
diff --git a/xlators/features/locks/src/locks-mem-types.h b/xlators/features/locks/src/locks-mem-types.h
index 08aeb0a7925..a76605027b3 100644
--- a/xlators/features/locks/src/locks-mem-types.h
+++ b/xlators/features/locks/src/locks-mem-types.h
@@ -11,19 +11,18 @@
#ifndef __LOCKS_MEM_TYPES_H__
#define __LOCKS_MEM_TYPES_H__
-#include "mem-types.h"
+#include <glusterfs/mem-types.h>
enum gf_locks_mem_types_ {
- gf_locks_mt_pl_dom_list_t = gf_common_mt_end + 1,
- gf_locks_mt_pl_inode_t,
- gf_locks_mt_posix_lock_t,
- gf_locks_mt_pl_entry_lock_t,
- gf_locks_mt_pl_inode_lock_t,
- gf_locks_mt_truncate_ops,
- gf_locks_mt_pl_rw_req_t,
- gf_locks_mt_posix_locks_private_t,
- gf_locks_mt_pl_fdctx_t,
- gf_locks_mt_end
+ gf_locks_mt_pl_dom_list_t = gf_common_mt_end + 1,
+ gf_locks_mt_pl_inode_t,
+ gf_locks_mt_posix_lock_t,
+ gf_locks_mt_pl_entry_lock_t,
+ gf_locks_mt_pl_inode_lock_t,
+ gf_locks_mt_pl_rw_req_t,
+ gf_locks_mt_posix_locks_private_t,
+ gf_locks_mt_pl_fdctx_t,
+ gf_locks_mt_pl_meta_lock_t,
+ gf_locks_mt_end
};
#endif
-
diff --git a/xlators/features/locks/src/locks.h b/xlators/features/locks/src/locks.h
index 8c2a6f867ee..c868eb494a2 100644
--- a/xlators/features/locks/src/locks.h
+++ b/xlators/features/locks/src/locks.h
@@ -1,5 +1,5 @@
/*
- Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+ Copyright (c) 2006-2012, 2015-2016 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
This file is licensed to you under your choice of the GNU Lesser
@@ -10,195 +10,283 @@
#ifndef __POSIX_LOCKS_H__
#define __POSIX_LOCKS_H__
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "compat-errno.h"
-#include "stack.h"
-#include "call-stub.h"
+#include <glusterfs/compat-errno.h>
+#include <glusterfs/stack.h>
+#include <glusterfs/call-stub.h>
#include "locks-mem-types.h"
-#include "client_t.h"
+#include <glusterfs/client_t.h>
+
+#include <glusterfs/lkowner.h>
-#include "lkowner.h"
+typedef enum {
+ MLK_NONE,
+ MLK_FILE_BASED,
+ MLK_FORCED,
+ MLK_OPTIMAL
+} mlk_mode_t; /* defines different mandatory locking modes*/
struct __pl_fd;
struct __posix_lock {
- struct list_head list;
+ struct list_head list;
+
+ off_t fl_start;
+ off_t fl_end;
+ uint32_t lk_flags;
+
+ short fl_type;
+ short blocked; /* waiting to acquire */
+ struct gf_flock user_flock; /* the flock supplied by the user */
+ xlator_t *this; /* required for blocked locks */
+ unsigned long fd_num;
- short fl_type;
- off_t fl_start;
- off_t fl_end;
+ fd_t *fd;
+ call_frame_t *frame;
- short blocked; /* waiting to acquire */
- struct gf_flock user_flock; /* the flock supplied by the user */
- xlator_t *this; /* required for blocked locks */
- unsigned long fd_num;
+ time_t blkd_time; /* time at which lock was queued into blkd list */
+ time_t granted_time; /* time at which lock was queued into active list */
- fd_t *fd;
- call_frame_t *frame;
+ /* These two together serve to uniquely identify each process
+ across nodes */
- struct timeval blkd_time; /*time at which lock was queued into blkd list*/
- struct timeval granted_time; /*time at which lock was queued into active list*/
+ void *client; /* to identify client node */
- /* These two together serve to uniquely identify each process
- across nodes */
+ /* This field uniquely identifies the client the lock belongs to. As
+ * lock migration is handled by rebalance, the client_t object will be
+ * overwritten by rebalance and can't be deemed as the owner of the
+ * lock on destination. Hence, the below field is migrated from
+ * source to destination by lock_migration_info_t and updated on the
+ * destination. So that on client-server disconnection, server can
+ * cleanup the locks proper;y. */
- void *client; /* to identify client node */
- gf_lkowner_t owner;
- pid_t client_pid; /* pid of client process */
+ char *client_uid;
+ gf_lkowner_t owner;
+ pid_t client_pid; /* pid of client process */
+
+ int blocking;
};
typedef struct __posix_lock posix_lock_t;
struct __pl_inode_lock {
- struct list_head list;
- struct list_head blocked_locks; /* list_head pointing to blocked_inodelks */
- int ref;
+ struct list_head list;
+ struct list_head blocked_locks; /* list_head pointing to blocked_inodelks */
+ struct list_head contend; /* list of contending locks */
+ int ref;
+
+ off_t fl_start;
+ off_t fl_end;
- short fl_type;
- off_t fl_start;
- off_t fl_end;
+ const char *volume;
- const char *volume;
+ struct gf_flock user_flock; /* the flock supplied by the user */
+ xlator_t *this; /* required for blocked locks */
+ struct __pl_inode *pl_inode;
- struct gf_flock user_flock; /* the flock supplied by the user */
- xlator_t *this; /* required for blocked locks */
- struct __pl_inode *pl_inode;
+ call_frame_t *frame;
- call_frame_t *frame;
+ time_t blkd_time; /* time at which lock was queued into blkd list */
+ time_t granted_time; /* time at which lock was queued into active list */
- struct timeval blkd_time; /*time at which lock was queued into blkd list*/
- struct timeval granted_time; /*time at which lock was queued into active list*/
+ /*last time at which lock contention was detected and notified*/
+ struct timespec contention_time;
- /* These two together serve to uniquely identify each process
- across nodes */
+ /* These two together serve to uniquely identify each process
+ across nodes */
- void *client; /* to identify client node */
- gf_lkowner_t owner;
- pid_t client_pid; /* pid of client process */
+ void *client; /* to identify client node */
+ gf_lkowner_t owner;
+ pid_t client_pid; /* pid of client process */
- char *connection_id; /* stores the client connection id */
+ char *connection_id; /* stores the client connection id */
- struct list_head client_list; /* list of all locks from a client */
+ struct list_head client_list; /* list of all locks from a client */
+ short fl_type;
+
+ int32_t status; /* Error code when we try to grant a lock in blocked
+ state */
};
typedef struct __pl_inode_lock pl_inode_lock_t;
-struct __pl_rw_req_t {
- struct list_head list;
- call_stub_t *stub;
- posix_lock_t region;
+struct _pl_rw_req {
+ struct list_head list;
+ call_stub_t *stub;
+ posix_lock_t region;
};
-typedef struct __pl_rw_req_t pl_rw_req_t;
-
-struct __pl_dom_list_t {
- struct list_head inode_list; /* list_head back to pl_inode_t */
- const char *domain;
- struct list_head entrylk_list; /* List of entry locks */
- struct list_head blocked_entrylks; /* List of all blocked entrylks */
- struct list_head inodelk_list; /* List of inode locks */
- struct list_head blocked_inodelks; /* List of all blocked inodelks */
+typedef struct _pl_rw_req pl_rw_req_t;
+
+struct _pl_dom_list {
+ struct list_head inode_list; /* list_head back to pl_inode_t */
+ const char *domain;
+ struct list_head entrylk_list; /* List of entry locks */
+ struct list_head blocked_entrylks; /* List of all blocked entrylks */
+ struct list_head inodelk_list; /* List of inode locks */
+ struct list_head blocked_inodelks; /* List of all blocked inodelks */
};
-typedef struct __pl_dom_list_t pl_dom_list_t;
+typedef struct _pl_dom_list pl_dom_list_t;
struct __entry_lock {
- struct list_head domain_list; /* list_head back to pl_dom_list_t */
- struct list_head blocked_locks; /* list_head back to blocked_entrylks */
- int ref;
+ struct list_head domain_list; /* list_head back to pl_dom_list_t */
+ struct list_head blocked_locks; /* list_head back to blocked_entrylks */
+ struct list_head contend; /* list of contending locks */
+ int ref;
+
+ call_frame_t *frame;
+ xlator_t *this;
+ struct __pl_inode *pinode;
- call_frame_t *frame;
- xlator_t *this;
- struct __pl_inode *pinode;
+ const char *volume;
- const char *volume;
+ const char *basename;
- const char *basename;
- entrylk_type type;
+ time_t blkd_time; /* time at which lock was queued into blkd list */
+ time_t granted_time; /* time at which lock was queued into active list */
- struct timeval blkd_time; /*time at which lock was queued into blkd list*/
- struct timeval granted_time; /*time at which lock was queued into active list*/
+ /*last time at which lock contention was detected and notified*/
+ struct timespec contention_time;
- void *client;
- gf_lkowner_t owner;
- pid_t client_pid; /* pid of client process */
+ void *client;
+ gf_lkowner_t owner;
+ pid_t client_pid; /* pid of client process */
- char *connection_id; /* stores the client connection id */
+ char *connection_id; /* stores the client connection id */
- struct list_head client_list; /* list of all locks from a client */
+ struct list_head client_list; /* list of all locks from a client */
+ entrylk_type type;
};
typedef struct __entry_lock pl_entry_lock_t;
-
/* The "simulated" inode. This contains a list of all the locks associated
with this file */
struct __pl_inode {
- pthread_mutex_t mutex;
-
- struct list_head dom_list; /* list of domains */
- struct list_head ext_list; /* list of fcntl locks */
- struct list_head rw_list; /* list of waiting r/w requests */
- struct list_head reservelk_list; /* list of reservelks */
- struct list_head blocked_reservelks; /* list of blocked reservelks */
- struct list_head blocked_calls; /* List of blocked lock calls while a reserve is held*/
- int mandatory; /* if mandatory locking is enabled */
-
- inode_t *refkeeper; /* hold refs on an inode while locks are
- held to prevent pruning */
+ pthread_mutex_t mutex;
+
+ struct list_head dom_list; /* list of domains */
+ struct list_head ext_list; /* list of fcntl locks */
+ struct list_head rw_list; /* list of waiting r/w requests */
+ struct list_head reservelk_list; /* list of reservelks */
+ struct list_head blocked_reservelks; /* list of blocked reservelks */
+ struct list_head blocked_calls; /* List of blocked lock calls while a
+ reserve is held*/
+ struct list_head metalk_list; /* Meta lock list */
+ struct list_head queued_locks; /* This is to store the incoming lock
+ requests while meta lock is enabled */
+ struct list_head waiting; /* List of pending fops waiting to unlink/rmdir
+ the inode. */
+ int mandatory; /* if mandatory locking is enabled */
+
+ inode_t *refkeeper; /* hold refs on an inode while locks are
+ held to prevent pruning */
+ uuid_t gfid; /* placeholder for gfid of the inode */
+ inode_t *inode; /* pointer to be used for ref and unref
+ of inode_t as long as there are
+ locks on it */
+ gf_boolean_t migrated;
+
+ /* Flag to indicate whether to read mlock-enforce xattr from disk */
+ gf_boolean_t check_mlock_info;
+
+ /* Mandatory_lock enforce: IO will be allowed if and only if the lkowner has
+ held the lock.
+
+ Note: An xattr is set on the file to recover this information post
+ reboot. If client does not want mandatory lock to be enforced, then it
+ should remove this xattr explicitly
+ */
+ gf_boolean_t mlock_enforced;
+ /* There are scenarios where mandatory lock is granted but there are IOs
+ pending at posix level. To avoid this before preempting the previous lock
+ owner, we wait for all the fops to be unwound.
+ */
+ int fop_wind_count;
+ pthread_cond_t check_fop_wind_count;
+
+ gf_boolean_t track_fop_wind_count;
+
+ int32_t links; /* Number of hard links the inode has. */
+ uint32_t remove_running; /* Number of remove operations running. */
+ gf_boolean_t is_locked; /* Regular locks will be blocked. */
+ gf_boolean_t removed; /* The inode has been deleted. */
};
typedef struct __pl_inode pl_inode_t;
+struct __pl_metalk {
+ pthread_mutex_t mutex;
+ /* For pl_inode meta lock list */
+ struct list_head list;
+ /* For pl_ctx_t list */
+ struct list_head client_list;
+ char *client_uid;
+
+ pl_inode_t *pl_inode;
+ int ref;
+};
+typedef struct __pl_metalk pl_meta_lock_t;
typedef struct {
- gf_boolean_t mandatory; /* if mandatory locking is enabled */
- gf_boolean_t trace; /* trace lock requests in and out */
- char *brickname;
+ char *brickname;
+ uint32_t revocation_secs;
+ uint32_t revocation_max_blocked;
+ uint32_t notify_contention_delay;
+ mlk_mode_t mandatory_mode; /* holds current mandatory locking mode */
+ gf_boolean_t trace; /* trace lock requests in and out */
+ gf_boolean_t monkey_unlocking;
+ gf_boolean_t revocation_clear_all;
+ gf_boolean_t notify_contention;
+ gf_boolean_t mlock_enforced;
} posix_locks_private_t;
-
typedef struct {
- gf_boolean_t entrylk_count_req;
- gf_boolean_t inodelk_count_req;
- gf_boolean_t inodelk_dom_count_req;
- gf_boolean_t posixlk_count_req;
- gf_boolean_t parent_entrylk_req;
-
- /* used by {f,}truncate */
- loc_t loc;
- fd_t *fd;
- off_t offset;
- dict_t *xdata;
- enum {TRUNCATE, FTRUNCATE} op;
+ data_t *inodelk_dom_count_req;
+
+ dict_t *xdata;
+ loc_t loc[2];
+ fd_t *fd;
+ inode_t *inode;
+ off_t offset;
+ glusterfs_fop_t op;
+ gf_boolean_t entrylk_count_req;
+ gf_boolean_t inodelk_count_req;
+ gf_boolean_t posixlk_count_req;
+ gf_boolean_t parent_entrylk_req;
+ gf_boolean_t multiple_dom_lk_requests;
+ int update_mlock_enforced_flag;
} pl_local_t;
-
typedef struct {
- struct list_head locks_list;
+ struct list_head locks_list;
} pl_fdctx_t;
-
struct _locker {
- struct list_head lockers;
- char *volume;
- inode_t *inode;
- gf_lkowner_t owner;
+ struct list_head lockers;
+ char *volume;
+ inode_t *inode;
+ gf_lkowner_t owner;
};
typedef struct _locks_ctx {
- pthread_mutex_t lock;
- struct list_head inodelk_lockers;
- struct list_head entrylk_lockers;
+ pthread_mutex_t lock;
+ struct list_head inodelk_lockers;
+ struct list_head entrylk_lockers;
+ struct list_head metalk_list;
} pl_ctx_t;
+typedef struct _multi_dom_lk_data {
+ xlator_t *this;
+ inode_t *inode;
+ dict_t *xdata_rsp;
+ gf_boolean_t keep_max;
+} multi_dom_lk_data;
+
+typedef enum { DECREMENT, INCREMENT } pl_count_op_t;
pl_ctx_t *
-pl_ctx_get (client_t *client, xlator_t *xlator);
+pl_ctx_get(client_t *client, xlator_t *xlator);
int
-pl_inodelk_client_cleanup (xlator_t *this, pl_ctx_t *ctx);
+pl_inodelk_client_cleanup(xlator_t *this, pl_ctx_t *ctx);
int
-pl_entrylk_client_cleanup (xlator_t *this, pl_ctx_t *ctx);
+pl_entrylk_client_cleanup(xlator_t *this, pl_ctx_t *ctx);
#endif /* __POSIX_LOCKS_H__ */
diff --git a/xlators/features/locks/src/pl-messages.h b/xlators/features/locks/src/pl-messages.h
new file mode 100644
index 00000000000..e2d3d7ca974
--- /dev/null
+++ b/xlators/features/locks/src/pl-messages.h
@@ -0,0 +1,29 @@
+/*
+ Copyright (c) 2016 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+*/
+
+#ifndef _PL_MESSAGES_H_
+#define _PL_MESSAGES_H_
+
+#include <glusterfs/glfs-message-id.h>
+
+/* To add new message IDs, append new identifiers at the end of the list.
+ *
+ * Never remove a message ID. If it's not used anymore, you can rename it or
+ * leave it as it is, but not delete it. This is to prevent reutilization of
+ * IDs by other messages.
+ *
+ * The component name must match one of the entries defined in
+ * glfs-message-id.h.
+ */
+
+GLFS_MSGID(PL, PL_MSG_LOCK_NUMBER, PL_MSG_INODELK_CONTENTION_FAILED,
+ PL_MSG_ENTRYLK_CONTENTION_FAILED);
+
+#endif /* !_PL_MESSAGES_H_ */
diff --git a/xlators/features/locks/src/posix.c b/xlators/features/locks/src/posix.c
index 337623d6577..cf0ae4c57dd 100644
--- a/xlators/features/locks/src/posix.c
+++ b/xlators/features/locks/src/posix.c
@@ -1,5 +1,5 @@
/*
- Copyright (c) 2006-2012 Red Hat, Inc. <http://www.redhat.com>
+ Copyright (c) 2006-2012, 2016 Red Hat, Inc. <http://www.redhat.com>
This file is part of GlusterFS.
This file is licensed to you under your choice of the GNU Lesser
@@ -12,2718 +12,5084 @@
#include <limits.h>
#include <pthread.h>
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
+#include <glusterfs/compat.h>
+#include <glusterfs/logging.h>
#include "locks.h"
#include "common.h"
-#include "statedump.h"
+#include <glusterfs/statedump.h>
#include "clear.h"
-#include "defaults.h"
-#include "syncop.h"
+#include <glusterfs/defaults.h>
+#include <glusterfs/syncop.h>
#ifndef LLONG_MAX
#define LLONG_MAX LONG_LONG_MAX /* compat with old gcc */
-#endif /* LLONG_MAX */
+#endif /* LLONG_MAX */
/* Forward declarations */
+void
+do_blocked_rw(pl_inode_t *);
+static int
+__rw_allowable(pl_inode_t *, posix_lock_t *, glusterfs_fop_t);
+static int
+format_brickname(char *);
+int
+pl_lockinfo_get_brickname(xlator_t *, inode_t *, int32_t *);
+static int
+fetch_pathinfo(xlator_t *, inode_t *, int32_t *, char **);
-void do_blocked_rw (pl_inode_t *);
-static int __rw_allowable (pl_inode_t *, posix_lock_t *, glusterfs_fop_t);
-static int format_brickname(char *);
-int pl_lockinfo_get_brickname (xlator_t *, inode_t *, int32_t *);
-static int fetch_pathinfo(xlator_t *, inode_t *, int32_t *, char **);
+/*
+ * The client is always requesting data, but older
+ * servers were not returning it. Newer ones are, so
+ * the client is receiving a mix of NULL and non-NULL
+ * xdata in the answers when bricks are of different
+ * versions. This triggers a bug in older clients.
+ * To prevent that, we avoid returning extra xdata to
+ * older clients (making the newer brick to behave as
+ * an old brick).
+ */
+#define PL_STACK_UNWIND_FOR_CLIENT(fop, xdata, frame, op_ret, params...) \
+ do { \
+ pl_local_t *__local = NULL; \
+ if (frame->root->client && \
+ (frame->root->client->opversion < GD_OP_VERSION_3_10_0)) { \
+ __local = frame->local; \
+ PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params); \
+ } else { \
+ PL_STACK_UNWIND(fop, xdata, frame, op_ret, params); \
+ } \
+ } while (0)
+
+#define PL_STACK_UNWIND(fop, xdata, frame, op_ret, params...) \
+ do { \
+ pl_local_t *__local = NULL; \
+ inode_t *__parent = NULL; \
+ inode_t *__inode = NULL; \
+ char *__name = NULL; \
+ dict_t *__unref = NULL; \
+ int __i = 0; \
+ __local = frame->local; \
+ if (op_ret >= 0 && pl_needs_xdata_response(frame->local)) { \
+ if (xdata) \
+ dict_ref(xdata); \
+ else \
+ xdata = dict_new(); \
+ if (xdata) { \
+ __unref = xdata; \
+ while (__local->fd || __local->loc[__i].inode) { \
+ pl_get_xdata_rsp_args(__local, #fop, &__parent, &__inode, \
+ &__name, __i); \
+ pl_set_xdata_response(frame->this, __local, __parent, \
+ __inode, __name, xdata, __i > 0); \
+ if (__local->fd || __i == 1) \
+ break; \
+ __i++; \
+ } \
+ } \
+ } \
+ PL_STACK_UNWIND_AND_FREE(__local, fop, frame, op_ret, params); \
+ if (__unref) \
+ dict_unref(__unref); \
+ } while (0)
+
+#define PL_LOCAL_GET_REQUESTS(frame, this, xdata, __fd, __loc, __newloc) \
+ do { \
+ if (pl_has_xdata_requests(xdata)) { \
+ if (!frame->local) \
+ frame->local = mem_get0(this->local_pool); \
+ pl_local_t *__local = frame->local; \
+ if (__local) { \
+ if (__fd) { \
+ __local->fd = fd_ref(__fd); \
+ __local->inode = inode_ref(__fd->inode); \
+ } else { \
+ if (__loc) \
+ loc_copy(&__local->loc[0], __loc); \
+ if (__newloc) \
+ loc_copy(&__local->loc[1], __newloc); \
+ __local->inode = inode_ref(__local->loc[0].inode); \
+ } \
+ pl_get_xdata_requests(__local, xdata); \
+ } \
+ } \
+ } while (0)
+
+#define PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, loc, fd, priv) \
+ do { \
+ if ((dict && (dict_get(dict, GF_ENFORCE_MANDATORY_LOCK))) || \
+ (name && (strcmp(name, GF_ENFORCE_MANDATORY_LOCK) == 0))) { \
+ inode_t *__inode = (loc ? loc->inode : fd->inode); \
+ pl_inode_t *__pl_inode = pl_inode_get(this, __inode, NULL); \
+ if (__pl_inode == NULL) { \
+ op_ret = -1; \
+ op_errno = ENOMEM; \
+ goto unwind; \
+ } \
+ if (!pl_is_mandatory_locking_enabled(__pl_inode) || \
+ !priv->mlock_enforced) { \
+ op_ret = -1; \
+ gf_msg(this->name, GF_LOG_DEBUG, EINVAL, 0, \
+ "option %s would need mandatory lock to be enabled " \
+ "and feature.enforce-mandatory-lock option to be set " \
+ "to on", \
+ GF_ENFORCE_MANDATORY_LOCK); \
+ op_errno = EINVAL; \
+ goto unwind; \
+ } \
+ \
+ op_ret = pl_local_init(frame, this, loc, fd); \
+ if (op_ret) { \
+ op_errno = ENOMEM; \
+ goto unwind; \
+ } \
+ \
+ ((pl_local_t *)(frame->local))->update_mlock_enforced_flag = 1; \
+ } \
+ } while (0)
+
+#define PL_INODE_REMOVE(_fop, _frame, _xl, _loc1, _loc2, _cont, _cbk, \
+ _args...) \
+ ({ \
+ struct list_head contend; \
+ pl_inode_t *__pl_inode; \
+ call_stub_t *__stub; \
+ int32_t __error; \
+ INIT_LIST_HEAD(&contend); \
+ __error = pl_inode_remove_prepare(_xl, _frame, _loc2 ? _loc2 : _loc1, \
+ &__pl_inode, &contend); \
+ if (__error < 0) { \
+ __stub = fop_##_fop##_stub(_frame, _cont, ##_args); \
+ __error = pl_inode_remove_complete(_xl, __pl_inode, __stub, \
+ &contend); \
+ } else if (__error == 0) { \
+ PL_LOCAL_GET_REQUESTS(_frame, _xl, xdata, ((fd_t *)NULL), _loc1, \
+ _loc2); \
+ STACK_WIND_COOKIE(_frame, _cbk, __pl_inode, FIRST_CHILD(_xl), \
+ FIRST_CHILD(_xl)->fops->_fop, ##_args); \
+ } \
+ __error; \
+ })
+
+gf_boolean_t
+pl_has_xdata_requests(dict_t *xdata)
+{
+ static char *reqs[] = {GLUSTERFS_ENTRYLK_COUNT,
+ GLUSTERFS_INODELK_COUNT,
+ GLUSTERFS_INODELK_DOM_COUNT,
+ GLUSTERFS_POSIXLK_COUNT,
+ GLUSTERFS_PARENT_ENTRYLK,
+ GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS,
+ NULL};
+ static int reqs_size[] = {SLEN(GLUSTERFS_ENTRYLK_COUNT),
+ SLEN(GLUSTERFS_INODELK_COUNT),
+ SLEN(GLUSTERFS_INODELK_DOM_COUNT),
+ SLEN(GLUSTERFS_POSIXLK_COUNT),
+ SLEN(GLUSTERFS_PARENT_ENTRYLK),
+ SLEN(GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS),
+ 0};
+ int i = 0;
+
+ if (!xdata)
+ return _gf_false;
+
+ for (i = 0; reqs[i]; i++)
+ if (dict_getn(xdata, reqs[i], reqs_size[i]))
+ return _gf_true;
+
+ return _gf_false;
+}
-static pl_fdctx_t *
-pl_new_fdctx ()
+static int
+dict_delete_domain_key(dict_t *dict, char *key, data_t *value, void *data)
{
- pl_fdctx_t *fdctx = NULL;
+ dict_del(dict, key);
+ return 0;
+}
- fdctx = GF_CALLOC (1, sizeof (*fdctx),
- gf_locks_mt_pl_fdctx_t);
- GF_VALIDATE_OR_GOTO ("posix-locks", fdctx, out);
+void
+pl_get_xdata_requests(pl_local_t *local, dict_t *xdata)
+{
+ if (!local || !xdata)
+ return;
- INIT_LIST_HEAD (&fdctx->locks_list);
+ GF_ASSERT(local->xdata == NULL);
+ local->xdata = dict_copy_with_ref(xdata, NULL);
+
+ if (dict_get_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT)) {
+ local->entrylk_count_req = 1;
+ dict_del_sizen(xdata, GLUSTERFS_ENTRYLK_COUNT);
+ }
+ if (dict_get_sizen(xdata, GLUSTERFS_INODELK_COUNT)) {
+ local->inodelk_count_req = 1;
+ dict_del_sizen(xdata, GLUSTERFS_INODELK_COUNT);
+ }
+ if (dict_get_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS)) {
+ local->multiple_dom_lk_requests = 1;
+ dict_del_sizen(xdata, GLUSTERFS_MULTIPLE_DOM_LK_CNT_REQUESTS);
+ dict_foreach_fnmatch(xdata, GLUSTERFS_INODELK_DOM_PREFIX "*",
+ dict_delete_domain_key, NULL);
+ }
+
+ local->inodelk_dom_count_req = dict_get_sizen(xdata,
+ GLUSTERFS_INODELK_DOM_COUNT);
+ if (local->inodelk_dom_count_req) {
+ data_ref(local->inodelk_dom_count_req);
+ dict_del_sizen(xdata, GLUSTERFS_INODELK_DOM_COUNT);
+ }
+
+ if (dict_get_sizen(xdata, GLUSTERFS_POSIXLK_COUNT)) {
+ local->posixlk_count_req = 1;
+ dict_del_sizen(xdata, GLUSTERFS_POSIXLK_COUNT);
+ }
+
+ if (dict_get_sizen(xdata, GLUSTERFS_PARENT_ENTRYLK)) {
+ local->parent_entrylk_req = 1;
+ dict_del_sizen(xdata, GLUSTERFS_PARENT_ENTRYLK);
+ }
+}
-out:
- return fdctx;
+gf_boolean_t
+pl_needs_xdata_response(pl_local_t *local)
+{
+ if (!local)
+ return _gf_false;
+
+ if (local->parent_entrylk_req || local->entrylk_count_req ||
+ local->inodelk_dom_count_req || local->inodelk_count_req ||
+ local->posixlk_count_req || local->multiple_dom_lk_requests)
+ return _gf_true;
+
+ return _gf_false;
}
-static pl_fdctx_t *
-pl_check_n_create_fdctx (xlator_t *this, fd_t *fd)
+void
+pl_get_xdata_rsp_args(pl_local_t *local, char *fop, inode_t **parent,
+ inode_t **inode, char **name, int i)
+{
+ if (strcmp(fop, "lookup") == 0) {
+ *parent = local->loc[0].parent;
+ *inode = local->loc[0].inode;
+ *name = (char *)local->loc[0].name;
+ } else {
+ if (local->fd) {
+ *inode = local->fd->inode;
+ } else {
+ *inode = local->loc[i].parent;
+ }
+ }
+}
+
+static inline int
+pl_track_io_fop_count(pl_local_t *local, xlator_t *this, pl_count_op_t op)
{
- int ret = 0;
- uint64_t tmp = 0;
- pl_fdctx_t *fdctx = NULL;
+ pl_inode_t *pl_inode = NULL;
- GF_VALIDATE_OR_GOTO ("posix-locks", this, out);
- GF_VALIDATE_OR_GOTO (this->name, fd, out);
+ if (!local)
+ return -1;
- LOCK (&fd->lock);
- {
- ret = __fd_ctx_get (fd, this, &tmp);
- if ((ret != 0) || (tmp == 0)) {
- fdctx = pl_new_fdctx ();
- if (fdctx == NULL) {
- goto unlock;
- }
- }
+ pl_inode = pl_inode_get(this, local->inode, NULL);
+ if (!pl_inode)
+ return -1;
- ret = __fd_ctx_set (fd, this, (uint64_t)(long)fdctx);
- if (ret != 0) {
- GF_FREE (fdctx);
- fdctx = NULL;
- gf_log (this->name, GF_LOG_DEBUG,
- "failed to set fd ctx");
+ if (pl_inode->mlock_enforced && pl_inode->track_fop_wind_count) {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (op == DECREMENT) {
+ pl_inode->fop_wind_count--;
+ /* fop_wind_count can go negative when lock enforcement is
+ * enabled on unwind path of an IO. Hence the "<" comparision.
+ */
+ if (pl_inode->fop_wind_count <= 0) {
+ pthread_cond_broadcast(&pl_inode->check_fop_wind_count);
+ pl_inode->track_fop_wind_count = _gf_false;
+ pl_inode->fop_wind_count = 0;
}
+ } else {
+ pl_inode->fop_wind_count++;
+ }
}
-unlock:
- UNLOCK (&fd->lock);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
-out:
- return fdctx;
+ return 0;
}
-int
-pl_truncate_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
- struct iatt *postbuf, dict_t *xdata)
+static int32_t
+__get_posixlk_count(pl_inode_t *pl_inode)
{
- pl_local_t *local = NULL;
+ posix_lock_t *lock = NULL;
+ int32_t count = 0;
- local = frame->local;
+ list_for_each_entry(lock, &pl_inode->ext_list, list) { count++; }
- if (local->op == TRUNCATE)
- loc_wipe (&local->loc);
+ return count;
+}
- if (local->xdata)
- dict_unref (local->xdata);
- if (local->fd)
- fd_unref (local->fd);
+int32_t
+get_posixlk_count(xlator_t *this, inode_t *inode)
+{
+ pl_inode_t *pl_inode = NULL;
+ uint64_t tmp_pl_inode = 0;
+ int32_t count = 0;
- STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno,
- prebuf, postbuf, xdata);
- return 0;
+ int ret = inode_ctx_get(inode, this, &tmp_pl_inode);
+ if (ret != 0) {
+ goto out;
+ }
+
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ count = __get_posixlk_count(pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+out:
+ return count;
}
+void
+pl_parent_entrylk_xattr_fill(xlator_t *this, inode_t *parent, char *basename,
+ dict_t *dict, gf_boolean_t keep_max)
+{
+ int32_t entrylk = 0;
+ int32_t maxcount = -1;
+ int ret = -1;
+
+ if (!parent || !basename)
+ goto out;
+ if (keep_max) {
+ ret = dict_get_int32_sizen(dict, GLUSTERFS_PARENT_ENTRYLK, &maxcount);
+ if (ret < 0)
+ gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+ GLUSTERFS_PARENT_ENTRYLK);
+ }
+ entrylk = check_entrylk_on_basename(this, parent, basename);
+ if (maxcount >= entrylk)
+ return;
+out:
+ ret = dict_set_int32_sizen(dict, GLUSTERFS_PARENT_ENTRYLK, entrylk);
+ if (ret < 0) {
+ gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+ GLUSTERFS_PARENT_ENTRYLK);
+ }
+}
-static int
-truncate_allowed (pl_inode_t *pl_inode,
- client_t *client, pid_t client_pid,
- gf_lkowner_t *owner, off_t offset)
+void
+pl_entrylk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+ gf_boolean_t keep_max)
+{
+ int32_t count = 0;
+ int32_t maxcount = -1;
+ int ret = -1;
+
+ if (keep_max) {
+ ret = dict_get_int32_sizen(dict, GLUSTERFS_ENTRYLK_COUNT, &maxcount);
+ if (ret < 0)
+ gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+ GLUSTERFS_ENTRYLK_COUNT);
+ }
+ count = get_entrylk_count(this, inode);
+ if (maxcount >= count)
+ return;
+
+ ret = dict_set_int32_sizen(dict, GLUSTERFS_ENTRYLK_COUNT, count);
+ if (ret < 0) {
+ gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+ GLUSTERFS_ENTRYLK_COUNT);
+ }
+}
+
+void
+pl_inodelk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+ char *domname, gf_boolean_t keep_max)
{
- posix_lock_t *l = NULL;
- posix_lock_t region = {.list = {0, }, };
- int ret = 1;
+ int32_t count = 0;
+ int32_t maxcount = -1;
+ int ret = -1;
+
+ if (keep_max) {
+ ret = dict_get_int32_sizen(dict, GLUSTERFS_INODELK_COUNT, &maxcount);
+ if (ret < 0)
+ gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+ GLUSTERFS_INODELK_COUNT);
+ }
+ count = get_inodelk_count(this, inode, domname);
+ if (maxcount >= count)
+ return;
- region.fl_start = offset;
- region.fl_end = LLONG_MAX;
- region.client = client;
- region.client_pid = client_pid;
- region.owner = *owner;
+ ret = dict_set_int32_sizen(dict, GLUSTERFS_INODELK_COUNT, count);
+ if (ret < 0) {
+ gf_msg_debug(this->name, 0,
+ "Failed to set count for "
+ "key %s",
+ GLUSTERFS_INODELK_COUNT);
+ }
- pthread_mutex_lock (&pl_inode->mutex);
- {
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (!l->blocked
- && locks_overlap (&region, l)
- && !same_owner (&region, l)) {
- ret = 0;
- gf_log ("posix-locks", GF_LOG_TRACE, "Truncate "
- "allowed");
- break;
- }
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ return;
+}
- return ret;
+void
+pl_posixlk_xattr_fill(xlator_t *this, inode_t *inode, dict_t *dict,
+ gf_boolean_t keep_max)
+{
+ int32_t count = 0;
+ int32_t maxcount = -1;
+ int ret = -1;
+
+ if (keep_max) {
+ ret = dict_get_int32_sizen(dict, GLUSTERFS_POSIXLK_COUNT, &maxcount);
+ if (ret < 0)
+ gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+ GLUSTERFS_POSIXLK_COUNT);
+ }
+ count = get_posixlk_count(this, inode);
+ if (maxcount >= count)
+ return;
+
+ ret = dict_set_int32_sizen(dict, GLUSTERFS_POSIXLK_COUNT, count);
+ if (ret < 0) {
+ gf_msg_debug(this->name, 0, " dict_set failed on key %s",
+ GLUSTERFS_POSIXLK_COUNT);
+ }
}
+void
+pl_inodelk_xattr_fill_each(xlator_t *this, inode_t *inode, dict_t *dict,
+ char *domname, gf_boolean_t keep_max, char *key)
+{
+ int32_t count = 0;
+ int32_t maxcount = -1;
+ int ret = -1;
+
+ if (keep_max) {
+ ret = dict_get_int32(dict, key, &maxcount);
+ if (ret < 0)
+ gf_msg_debug(this->name, 0, " Failed to fetch the value for key %s",
+ GLUSTERFS_INODELK_COUNT);
+ }
+ count = get_inodelk_count(this, inode, domname);
+ if (maxcount >= count)
+ return;
+
+ ret = dict_set_int32(dict, key, count);
+ if (ret < 0) {
+ gf_msg_debug(this->name, 0,
+ "Failed to set count for "
+ "key %s",
+ key);
+ }
+
+ return;
+}
static int
-truncate_stat_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, struct iatt *buf,
- dict_t *xdata)
+pl_inodelk_xattr_fill_multiple(dict_t *this, char *key, data_t *value,
+ void *data)
{
- posix_locks_private_t *priv = NULL;
- pl_local_t *local = NULL;
- inode_t *inode = NULL;
- pl_inode_t *pl_inode = NULL;
+ multi_dom_lk_data *d = data;
+ char *tmp_key = NULL;
+ char *save_ptr = NULL;
+
+ tmp_key = gf_strdup(key);
+ if (!tmp_key)
+ return -1;
+
+ strtok_r(tmp_key, ":", &save_ptr);
+ if (!*save_ptr) {
+ if (tmp_key)
+ GF_FREE(tmp_key);
+ gf_msg(THIS->name, GF_LOG_ERROR, 0, EINVAL,
+ "Could not tokenize domain string from key %s", key);
+ return -1;
+ }
+
+ pl_inodelk_xattr_fill_each(d->this, d->inode, d->xdata_rsp, save_ptr,
+ d->keep_max, key);
+ if (tmp_key)
+ GF_FREE(tmp_key);
+
+ return 0;
+}
+void
+pl_fill_multiple_dom_lk_requests(xlator_t *this, pl_local_t *local,
+ inode_t *inode, dict_t *dict,
+ gf_boolean_t keep_max)
+{
+ multi_dom_lk_data data;
- priv = this->private;
- local = frame->local;
+ data.this = this;
+ data.inode = inode;
+ data.xdata_rsp = dict;
+ data.keep_max = keep_max;
- if (op_ret != 0) {
- gf_log (this->name, GF_LOG_ERROR,
- "got error (errno=%d, stderror=%s) from child",
- op_errno, strerror (op_errno));
- goto unwind;
- }
+ dict_foreach_fnmatch(local->xdata, GLUSTERFS_INODELK_DOM_PREFIX "*",
+ pl_inodelk_xattr_fill_multiple, &data);
+}
- if (local->op == TRUNCATE)
- inode = local->loc.inode;
- else
- inode = local->fd->inode;
+void
+pl_set_xdata_response(xlator_t *this, pl_local_t *local, inode_t *parent,
+ inode_t *inode, char *name, dict_t *xdata,
+ gf_boolean_t max_lock)
+{
+ if (!xdata || !local)
+ return;
- pl_inode = pl_inode_get (this, inode);
- if (!pl_inode) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+ if (local->parent_entrylk_req && parent && name && name[0] != '\0')
+ pl_parent_entrylk_xattr_fill(this, parent, name, xdata, max_lock);
- if (priv->mandatory
- && pl_inode->mandatory
- && !truncate_allowed (pl_inode, frame->root->client,
- frame->root->pid, &frame->root->lk_owner,
- local->offset)) {
- op_ret = -1;
- op_errno = EAGAIN;
- goto unwind;
- }
+ if (!inode)
+ return;
- switch (local->op) {
- case TRUNCATE:
- STACK_WIND (frame, pl_truncate_cbk, FIRST_CHILD (this),
- FIRST_CHILD (this)->fops->truncate,
- &local->loc, local->offset, local->xdata);
- break;
- case FTRUNCATE:
- STACK_WIND (frame, pl_truncate_cbk, FIRST_CHILD (this),
- FIRST_CHILD (this)->fops->ftruncate,
- local->fd, local->offset, local->xdata);
- break;
- }
+ if (local->entrylk_count_req)
+ pl_entrylk_xattr_fill(this, inode, xdata, max_lock);
- return 0;
+ if (local->inodelk_dom_count_req)
+ pl_inodelk_xattr_fill(this, inode, xdata,
+ data_to_str(local->inodelk_dom_count_req),
+ max_lock);
-unwind:
- gf_log (this->name, GF_LOG_ERROR, "truncate failed with ret: %d, "
- "error: %s", op_ret, strerror (op_errno));
- if (local->op == TRUNCATE)
- loc_wipe (&local->loc);
- if (local->xdata)
- dict_unref (local->xdata);
- if (local->fd)
- fd_unref (local->fd);
-
- STACK_UNWIND_STRICT (truncate, frame, op_ret, op_errno, buf, NULL, xdata);
- return 0;
-}
+ if (local->inodelk_count_req)
+ pl_inodelk_xattr_fill(this, inode, xdata, NULL, max_lock);
+ if (local->posixlk_count_req)
+ pl_posixlk_xattr_fill(this, inode, xdata, max_lock);
+
+ if (local->multiple_dom_lk_requests)
+ pl_fill_multiple_dom_lk_requests(this, local, inode, xdata, max_lock);
+}
+/* Checks whether the region where fop is acting upon conflicts
+ * with existing locks. If there is no conflict function returns
+ * 1 else returns 0 with can_block boolean set accordingly to
+ * indicate block/fail the fop.
+ */
int
-pl_truncate (call_frame_t *frame, xlator_t *this,
- loc_t *loc, off_t offset, dict_t *xdata)
+pl_is_fop_allowed(pl_inode_t *pl_inode, posix_lock_t *region, fd_t *fd,
+ glusterfs_fop_t op, gf_boolean_t *can_block)
{
- pl_local_t *local = NULL;
+ int ret = 0;
+
+ if (!__rw_allowable(pl_inode, region, op)) {
+ if (pl_inode->mlock_enforced) {
+ *can_block = _gf_false;
+ } else if ((!fd) || (fd && (fd->flags & O_NONBLOCK))) {
+ gf_log("locks", GF_LOG_TRACE,
+ "returning EAGAIN"
+ " because fd is O_NONBLOCK");
+ *can_block = _gf_false;
+ } else {
+ *can_block = _gf_true;
+ }
+ } else {
+ ret = 1;
+ }
- local = mem_get0 (this->local_pool);
- GF_VALIDATE_OR_GOTO (this->name, local, unwind);
+ return ret;
+}
- local->op = TRUNCATE;
- local->offset = offset;
- loc_copy (&local->loc, loc);
- if (xdata)
- local->xdata = dict_ref (xdata);
+static pl_fdctx_t *
+pl_new_fdctx()
+{
+ pl_fdctx_t *fdctx = GF_MALLOC(sizeof(*fdctx), gf_locks_mt_pl_fdctx_t);
+ GF_VALIDATE_OR_GOTO("posix-locks", fdctx, out);
- frame->local = local;
+ INIT_LIST_HEAD(&fdctx->locks_list);
- STACK_WIND (frame, truncate_stat_cbk, FIRST_CHILD (this),
- FIRST_CHILD (this)->fops->stat, loc, NULL);
+out:
+ return fdctx;
+}
- return 0;
+static pl_fdctx_t *
+pl_check_n_create_fdctx(xlator_t *this, fd_t *fd)
+{
+ int ret = 0;
+ uint64_t tmp = 0;
+ pl_fdctx_t *fdctx = NULL;
+
+ GF_VALIDATE_OR_GOTO("posix-locks", this, out);
+ GF_VALIDATE_OR_GOTO(this->name, fd, out);
+
+ LOCK(&fd->lock);
+ {
+ ret = __fd_ctx_get(fd, this, &tmp);
+ if ((ret != 0) || (tmp == 0)) {
+ fdctx = pl_new_fdctx();
+ if (fdctx == NULL) {
+ goto unlock;
+ }
+ }
-unwind:
- gf_log (this->name, GF_LOG_ERROR, "truncate for %s failed with ret: %d, "
- "error: %s", loc->path, -1, strerror (ENOMEM));
- STACK_UNWIND_STRICT (truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+ ret = __fd_ctx_set(fd, this, (uint64_t)(long)fdctx);
+ if (ret != 0) {
+ GF_FREE(fdctx);
+ fdctx = NULL;
+ UNLOCK(&fd->lock);
+ gf_log(this->name, GF_LOG_DEBUG, "failed to set fd ctx");
+ goto out;
+ }
+ }
+unlock:
+ UNLOCK(&fd->lock);
- return 0;
+out:
+ return fdctx;
}
+int32_t
+pl_discard_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+ struct iatt *postbuf, dict_t *xdata)
+{
+ pl_track_io_fop_count(frame->local, this, DECREMENT);
+
+ PL_STACK_UNWIND(discard, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+ xdata);
+ return 0;
+}
int
-pl_ftruncate (call_frame_t *frame, xlator_t *this,
- fd_t *fd, off_t offset, dict_t *xdata)
+pl_discard_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ size_t len, dict_t *xdata)
{
- pl_local_t *local = NULL;
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
+
+ STACK_WIND(frame, pl_discard_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
+ return 0;
+}
+
+int32_t
+pl_discard(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ size_t len, dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+ pl_rw_req_t *rw = NULL;
+ posix_lock_t region = {
+ .list =
+ {
+ 0,
+ },
+ };
+ gf_boolean_t enabled = _gf_false;
+ gf_boolean_t can_block = _gf_true;
+ int op_ret = 0;
+ int op_errno = 0;
+ int allowed = 1;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ local = mem_get0(this->local_pool);
+ if (!local) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ frame->local = local;
+ local->inode = inode_ref(fd->inode);
+ local->fd = fd_ref(fd);
+
+ pl_inode = pl_inode_get(this, fd->inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ if (frame->root->pid < 0)
+ enabled = _gf_false;
+ else
+ enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+ if (enabled) {
+ region.fl_start = offset;
+ region.fl_end = offset + len - 1;
+ region.client = frame->root->client;
+ region.fd_num = fd_to_fdnum(fd);
+ region.client_pid = frame->root->pid;
+ region.owner = frame->root->lk_owner;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_DISCARD,
+ &can_block);
+ if (allowed == 1) {
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
+ }
+ goto unlock;
+ } else if (!can_block) {
+ op_errno = EAGAIN;
+ op_ret = -1;
+ goto unlock;
+ }
- local = mem_get0 (this->local_pool);
- GF_VALIDATE_OR_GOTO (this->name, local, unwind);
+ rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+ if (!rw) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unlock;
+ }
- local->op = FTRUNCATE;
- local->offset = offset;
- local->fd = fd_ref (fd);
- if (xdata)
- local->xdata = dict_ref (xdata);
+ rw->stub = fop_discard_stub(frame, pl_discard_cont, fd, offset, len,
+ xdata);
+ if (!rw->stub) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ GF_FREE(rw);
+ goto unlock;
+ }
- frame->local = local;
+ rw->region = region;
- STACK_WIND (frame, truncate_stat_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->fstat, fd, xdata);
- return 0;
+ list_add_tail(&rw->list, &pl_inode->rw_list);
+ }
+ unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+ if (allowed == 1)
+ STACK_WIND(frame, pl_discard_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->discard, fd, offset, len, xdata);
unwind:
- gf_log (this->name, GF_LOG_ERROR, "ftruncate failed with ret: %d, "
- "error: %s", -1, strerror (ENOMEM));
- STACK_UNWIND_STRICT (ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+ if (op_ret == -1)
+ PL_STACK_UNWIND(discard, xdata, frame, op_ret, op_errno, NULL, NULL,
+ NULL);
- return 0;
+ return 0;
}
-int
-pl_locks_by_fd (pl_inode_t *pl_inode, fd_t *fd)
+int32_t
+pl_zerofill_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+ struct iatt *postbuf, dict_t *xdata)
{
- posix_lock_t *l = NULL;
- int found = 0;
+ pl_track_io_fop_count(frame->local, this, DECREMENT);
- pthread_mutex_lock (&pl_inode->mutex);
- {
+ PL_STACK_UNWIND(zerofill, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+ xdata);
+ return 0;
+}
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->fd_num == fd_to_fdnum(fd)) {
- found = 1;
- break;
- }
- }
+int
+pl_zerofill_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ off_t len, dict_t *xdata)
+{
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
- return found;
+ STACK_WIND(frame, pl_zerofill_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+ return 0;
}
-static void
-delete_locks_of_fd (xlator_t *this, pl_inode_t *pl_inode, fd_t *fd)
+int32_t
+pl_zerofill(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ off_t len, dict_t *xdata)
{
- posix_lock_t *tmp = NULL;
- posix_lock_t *l = NULL;
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+ pl_rw_req_t *rw = NULL;
+ posix_lock_t region = {
+ .list =
+ {
+ 0,
+ },
+ };
+ gf_boolean_t enabled = _gf_false;
+ gf_boolean_t can_block = _gf_true;
+ int op_ret = 0;
+ int op_errno = 0;
+ int allowed = 1;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ local = mem_get0(this->local_pool);
+ if (!local) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ frame->local = local;
+ local->inode = inode_ref(fd->inode);
+ local->fd = fd_ref(fd);
+
+ pl_inode = pl_inode_get(this, fd->inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ if (frame->root->pid < 0)
+ enabled = _gf_false;
+ else
+ enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+ if (enabled) {
+ region.fl_start = offset;
+ region.fl_end = offset + len - 1;
+ region.client = frame->root->client;
+ region.fd_num = fd_to_fdnum(fd);
+ region.client_pid = frame->root->pid;
+ region.owner = frame->root->lk_owner;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_ZEROFILL,
+ &can_block);
+ if (allowed == 1) {
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
+ }
+ goto unlock;
+ } else if (!can_block) {
+ op_errno = EAGAIN;
+ op_ret = -1;
+ goto unlock;
+ }
- struct list_head blocked_list;
+ rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+ if (!rw) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unlock;
+ }
- INIT_LIST_HEAD (&blocked_list);
+ rw->stub = fop_zerofill_stub(frame, pl_zerofill_cont, fd, offset,
+ len, xdata);
+ if (!rw->stub) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ GF_FREE(rw);
+ goto unlock;
+ }
- pthread_mutex_lock (&pl_inode->mutex);
- {
+ rw->region = region;
- list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
- if (l->fd_num == fd_to_fdnum(fd)) {
- if (l->blocked) {
- list_move_tail (&l->list, &blocked_list);
- continue;
- }
- __delete_lock (pl_inode, l);
- __destroy_lock (l);
- }
- }
+ list_add_tail(&rw->list, &pl_inode->rw_list);
+ }
+ unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ if (allowed == 1)
+ STACK_WIND(frame, pl_zerofill_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->zerofill, fd, offset, len, xdata);
+unwind:
+ if (op_ret == -1)
+ PL_STACK_UNWIND(zerofill, xdata, frame, op_ret, op_errno, NULL, NULL,
+ NULL);
- list_for_each_entry_safe (l, tmp, &blocked_list, list) {
- list_del_init(&l->list);
- STACK_UNWIND_STRICT (lk, l->frame, -1, EAGAIN, &l->user_flock,
- NULL);
- __destroy_lock (l);
- }
+ return 0;
+}
- grant_blocked_locks (this, pl_inode);
+int
+pl_truncate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
+ struct iatt *postbuf, dict_t *xdata)
+{
+ pl_local_t *local = frame->local;
- do_blocked_rw (pl_inode);
+ pl_track_io_fop_count(local, this, DECREMENT);
+ if (local->op == GF_FOP_TRUNCATE)
+ PL_STACK_UNWIND(truncate, xdata, frame, op_ret, op_errno, prebuf,
+ postbuf, xdata);
+ else
+ PL_STACK_UNWIND(ftruncate, xdata, frame, op_ret, op_errno, prebuf,
+ postbuf, xdata);
+ return 0;
}
-static void
-__delete_locks_of_owner (pl_inode_t *pl_inode,
- client_t *client, gf_lkowner_t *owner)
-{
- posix_lock_t *tmp = NULL;
- posix_lock_t *l = NULL;
-
- /* TODO: what if it is a blocked lock with pending l->frame */
-
- list_for_each_entry_safe (l, tmp, &pl_inode->ext_list, list) {
- if (l->blocked)
- continue;
- if ((l->client == client) &&
- is_same_lkowner (&l->owner, owner)) {
- gf_log ("posix-locks", GF_LOG_TRACE,
- " Flushing lock"
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" state: %s",
- l->fl_type == F_UNLCK ? "Unlock" : "Lock",
- l->client_pid,
- lkowner_utoa (&l->owner),
- l->user_flock.l_start,
- l->user_flock.l_len,
- l->blocked == 1 ? "Blocked" : "Active");
-
- __delete_lock (pl_inode, l);
- __destroy_lock (l);
- }
- }
+int
+pl_ftruncate_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ dict_t *xdata)
+{
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
- return;
+ STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->ftruncate, fd, offset, xdata);
+ return 0;
}
-
-int32_t
-pl_getxattr_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+int
+pl_truncate_cont(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+ dict_t *xdata)
{
- STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict, xdata);
- return 0;
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
+ STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->truncate, loc, offset, xdata);
+ return 0;
}
-int32_t
-pl_getxattr (call_frame_t *frame, xlator_t *this, loc_t *loc,
- const char *name, dict_t *xdata)
-{
- int32_t op_errno = EINVAL;
- int op_ret = -1;
- int32_t bcount = 0;
- int32_t gcount = 0;
- char key[PATH_MAX] = {0, };
- char *lk_summary = NULL;
- pl_inode_t *pl_inode = NULL;
- dict_t *dict = NULL;
- clrlk_args args = {0,};
- char *brickname = NULL;
-
- if (!name)
- goto usual;
-
- if (strncmp (name, GF_XATTR_CLRLK_CMD, strlen (GF_XATTR_CLRLK_CMD)))
- goto usual;
-
- if (clrlk_parse_args (name, &args)) {
- op_errno = EINVAL;
- goto out;
- }
+static int
+truncate_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *buf,
+ dict_t *xdata)
+{
+ pl_local_t *local = frame->local;
+ inode_t *inode = NULL;
+ pl_inode_t *pl_inode = NULL;
+ pl_rw_req_t *rw = NULL;
+ posix_lock_t region = {
+ .list =
+ {
+ 0,
+ },
+ };
+ gf_boolean_t enabled = _gf_false;
+ gf_boolean_t can_block = _gf_true;
+ int allowed = 1;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ if (op_ret != 0) {
+ gf_log(this->name, GF_LOG_ERROR,
+ "got error (errno=%d, stderror=%s) from child", op_errno,
+ strerror(op_errno));
+ goto unwind;
+ }
+
+ if (local->op == GF_FOP_TRUNCATE)
+ inode = local->loc[0].inode;
+ else
+ inode = local->fd->inode;
+
+ local->inode = inode_ref(inode);
+
+ pl_inode = pl_inode_get(this, inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ if (frame->root->pid < 0)
+ enabled = _gf_false;
+ else
+ enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+ if (enabled) {
+ region.fl_start = local->offset;
+ region.fl_end = LLONG_MAX;
+ region.client = frame->root->client;
+ region.fd_num = fd_to_fdnum(local->fd);
+ region.client_pid = frame->root->pid;
+ region.owner = frame->root->lk_owner;
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ allowed = pl_is_fop_allowed(pl_inode, &region, local->fd, local->op,
+ &can_block);
- dict = dict_new ();
- if (!dict) {
- op_errno = ENOMEM;
- goto out;
- }
+ if (allowed == 1) {
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
+ }
+ goto unlock;
+ } else if (!can_block) {
+ op_errno = EAGAIN;
+ op_ret = -1;
+ goto unlock;
+ }
- pl_inode = pl_inode_get (this, loc->inode);
- if (!pl_inode) {
+ rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+ if (!rw) {
op_errno = ENOMEM;
- goto out;
- }
+ op_ret = -1;
+ goto unlock;
+ }
+
+ if (local->op == GF_FOP_TRUNCATE)
+ rw->stub = fop_truncate_stub(frame, pl_truncate_cont,
+ &local->loc[0], local->offset,
+ local->xdata);
+ else
+ rw->stub = fop_ftruncate_stub(frame, pl_ftruncate_cont,
+ local->fd, local->offset,
+ local->xdata);
+ if (!rw->stub) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ GF_FREE(rw);
+ goto unlock;
+ }
- switch (args.type) {
- case CLRLK_INODE:
- case CLRLK_ENTRY:
- op_ret = clrlk_clear_lks_in_all_domains (this, pl_inode,
- &args, &bcount,
- &gcount,
- &op_errno);
- if (op_ret)
- goto out;
- break;
- case CLRLK_POSIX:
- op_ret = clrlk_clear_posixlk (this, pl_inode, &args,
- &bcount, &gcount,
- &op_errno);
- if (op_ret)
- goto out;
- break;
- case CLRLK_TYPE_MAX:
- op_errno = EINVAL;
- goto out;
- }
+ rw->region = region;
- op_ret = fetch_pathinfo (this, loc->inode, &op_errno, &brickname);
- if (op_ret) {
- gf_log (this->name, GF_LOG_WARNING,
- "Couldn't get brickname");
- } else {
- op_ret = format_brickname(brickname);
- if (op_ret) {
- gf_log (this->name, GF_LOG_WARNING,
- "Couldn't format brickname");
- GF_FREE(brickname);
- brickname = NULL;
- }
+ list_add_tail(&rw->list, &pl_inode->rw_list);
}
+ unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- if (!gcount && !bcount) {
- if (gf_asprintf (&lk_summary, "No locks cleared.") == -1) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto out;
- }
- } else if (gf_asprintf (&lk_summary, "%s: %s blocked locks=%d "
- "granted locks=%d",
- (brickname == NULL)? this->name : brickname,
- (args.type == CLRLK_INODE)? "inode":
- (args.type == CLRLK_ENTRY)? "entry":
- (args.type == CLRLK_POSIX)? "posix": " ",
- bcount, gcount) == -1) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto out;
+ if (allowed == 1) {
+ switch (local->op) {
+ case GF_FOP_TRUNCATE:
+ STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->truncate, &local->loc[0],
+ local->offset, local->xdata);
+ break;
+ case GF_FOP_FTRUNCATE:
+ STACK_WIND(frame, pl_truncate_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->ftruncate, local->fd,
+ local->offset, local->xdata);
+ break;
+ default:
+ break;
}
+ }
+unwind:
+ if (op_ret == -1) {
+ gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+ "truncate failed with "
+ "ret: %d, error: %s",
+ op_ret, strerror(op_errno));
- strncpy (key, name, strlen (name));
- if (dict_set_dynstr (dict, key, lk_summary)) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto out;
+ switch (local->op) {
+ case GF_FOP_TRUNCATE:
+ PL_STACK_UNWIND(truncate, xdata, frame, op_ret, op_errno, buf,
+ NULL, xdata);
+ break;
+ case GF_FOP_FTRUNCATE:
+ PL_STACK_UNWIND(ftruncate, xdata, frame, op_ret, op_errno, buf,
+ NULL, xdata);
+ break;
+ default:
+ break;
}
+ }
+ return 0;
+}
- op_ret = 0;
-out:
- GF_FREE(brickname);
- STACK_UNWIND_STRICT (getxattr, frame, op_ret, op_errno, dict, xdata);
+int
+pl_truncate(call_frame_t *frame, xlator_t *this, loc_t *loc, off_t offset,
+ dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ int ret = -1;
- GF_FREE (args.opts);
- if (op_ret && lk_summary)
- GF_FREE (lk_summary);
- if (dict)
- dict_unref (dict);
- return 0;
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
-usual:
- STACK_WIND (frame, pl_getxattr_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
- return 0;
+ local = mem_get0(this->local_pool);
+ GF_VALIDATE_OR_GOTO(this->name, local, unwind);
+
+ local->op = GF_FOP_TRUNCATE;
+ local->offset = offset;
+ loc_copy(&local->loc[0], loc);
+ if (xdata)
+ local->xdata = dict_ref(xdata);
+
+ frame->local = local;
+
+ STACK_WIND(frame, truncate_stat_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->stat, loc, NULL);
+ ret = 0;
+
+unwind:
+ if (ret == -1) {
+ gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+ "truncate on %s failed with"
+ " ret: %d, error: %s",
+ loc->path, -1, strerror(ENOMEM));
+ STACK_UNWIND_STRICT(truncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+ }
+ return 0;
}
-static int
-format_brickname(char *brickname)
+int
+pl_ftruncate(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ dict_t *xdata)
{
- int ret = -1;
- char *hostname = NULL;
- char *volume = NULL;
- char *saveptr = NULL;
+ pl_local_t *local = NULL;
+ int ret = -1;
- if (!brickname)
- goto out;
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+ local = mem_get0(this->local_pool);
+ GF_VALIDATE_OR_GOTO(this->name, local, unwind);
- strtok_r(brickname, ":", &saveptr);
- hostname = gf_strdup(strtok_r(NULL, ":", &saveptr));
- if (hostname == NULL)
- goto out;
- volume = gf_strdup(strtok_r(NULL, ".", &saveptr));
- if (volume == NULL)
- goto out;
+ local->op = GF_FOP_FTRUNCATE;
+ local->offset = offset;
+ local->fd = fd_ref(fd);
+ if (xdata)
+ local->xdata = dict_ref(xdata);
- sprintf(brickname, "%s:%s", hostname, volume);
+ frame->local = local;
- ret = 0;
-out:
- GF_FREE(hostname);
- GF_FREE(volume);
- return ret;
+ STACK_WIND(frame, truncate_stat_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fstat, fd, xdata);
+ ret = 0;
+unwind:
+ if (ret == -1) {
+ gf_log(this ? this->name : "locks", GF_LOG_ERROR,
+ "ftruncate failed with"
+ " ret: %d, error: %s",
+ -1, strerror(ENOMEM));
+ STACK_UNWIND_STRICT(ftruncate, frame, -1, ENOMEM, NULL, NULL, NULL);
+ }
+ return 0;
}
-static int
-fetch_pathinfo (xlator_t *this, inode_t *inode, int32_t *op_errno,
- char **brickname)
+int
+pl_locks_by_fd(pl_inode_t *pl_inode, fd_t *fd)
{
- int ret = -1;
- loc_t loc = {0, };
- dict_t *dict = NULL;
+ posix_lock_t *l = NULL;
+ int found = 0;
- if (!brickname)
- goto out;
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (l->fd_num == fd_to_fdnum(fd)) {
+ found = 1;
+ break;
+ }
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ return found;
+}
- if (!op_errno)
- goto out;
+static void
+delete_locks_of_fd(xlator_t *this, pl_inode_t *pl_inode, fd_t *fd)
+{
+ posix_lock_t *tmp = NULL;
+ posix_lock_t *l = NULL;
- uuid_copy (loc.gfid, inode->gfid);
- loc.inode = inode_ref (inode);
+ struct list_head blocked_list;
- ret = syncop_getxattr (FIRST_CHILD(this), &loc, &dict,
- GF_XATTR_PATHINFO_KEY);
- if (ret < 0) {
- *op_errno = -ret;
- ret = -1;
- goto out;
+ INIT_LIST_HEAD(&blocked_list);
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+ {
+ if (l->fd_num == fd_to_fdnum(fd)) {
+ if (l->blocked) {
+ list_move_tail(&l->list, &blocked_list);
+ continue;
+ }
+ __delete_lock(l);
+ __destroy_lock(l);
+ }
}
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, brickname);
- if (ret)
- goto out;
+ list_for_each_entry_safe(l, tmp, &blocked_list, list)
+ {
+ list_del_init(&l->list);
+ STACK_UNWIND_STRICT(lk, l->frame, -1, EAGAIN, &l->user_flock, NULL);
+ __destroy_lock(l);
+ }
- *brickname = gf_strdup(*brickname);
- if (*brickname == NULL) {
- ret = -1;
- goto out;
- }
+ grant_blocked_locks(this, pl_inode);
- ret = 0;
-out:
- if (dict != NULL) {
- dict_unref (dict);
+ do_blocked_rw(pl_inode);
+}
+
+static void
+__delete_locks_of_owner(pl_inode_t *pl_inode, client_t *client,
+ gf_lkowner_t *owner)
+{
+ posix_lock_t *tmp = NULL;
+ posix_lock_t *l = NULL;
+
+ /* TODO: what if it is a blocked lock with pending l->frame */
+
+ list_for_each_entry_safe(l, tmp, &pl_inode->ext_list, list)
+ {
+ if (l->blocked)
+ continue;
+ if ((l->client == client) && is_same_lkowner(&l->owner, owner)) {
+ gf_log("posix-locks", GF_LOG_TRACE,
+ " Flushing lock"
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64
+ " state: %s",
+ l->fl_type == F_UNLCK ? "Unlock" : "Lock", l->client_pid,
+ lkowner_utoa(&l->owner), l->user_flock.l_start,
+ l->user_flock.l_len, l->blocked == 1 ? "Blocked" : "Active");
+
+ __delete_lock(l);
+ __destroy_lock(l);
}
- loc_wipe(&loc);
+ }
- return ret;
+ return;
}
+int32_t
+pl_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+ STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+ return 0;
+}
-int
-pl_lockinfo_get_brickname (xlator_t *this, inode_t *inode, int32_t *op_errno)
+static int32_t
+pl_getxattr_clrlk(xlator_t *this, const char *name, inode_t *inode,
+ dict_t **dict, int32_t *op_errno)
{
- int ret = -1;
- posix_locks_private_t *priv = NULL;
- char *brickname = NULL;
- char *end = NULL;
- char *tmp = NULL;
+ int32_t bcount = 0;
+ int32_t gcount = 0;
+ char *key = NULL;
+ char *lk_summary = NULL;
+ pl_inode_t *pl_inode = NULL;
+ clrlk_args args = {
+ 0,
+ };
+ char *brickname = NULL;
+ int32_t op_ret = -1;
+
+ *op_errno = EINVAL;
+
+ if (clrlk_parse_args(name, &args)) {
+ *op_errno = EINVAL;
+ goto out;
+ }
+
+ *dict = dict_new();
+ if (!*dict) {
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (!pl_inode) {
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ switch (args.type) {
+ case CLRLK_INODE:
+ case CLRLK_ENTRY:
+ op_ret = clrlk_clear_lks_in_all_domains(this, pl_inode, &args,
+ &bcount, &gcount, op_errno);
+ break;
+ case CLRLK_POSIX:
+ op_ret = clrlk_clear_posixlk(this, pl_inode, &args, &bcount,
+ &gcount, op_errno);
+ break;
+ default:
+ op_ret = -1;
+ *op_errno = EINVAL;
+ }
+ if (op_ret) {
+ if (args.type >= CLRLK_TYPE_MAX) {
+ gf_log(this->name, GF_LOG_ERROR,
+ "clear locks: invalid lock type %d", args.type);
+ } else {
+ gf_log(this->name, GF_LOG_ERROR,
+ "clear locks of type %s failed: %s",
+ clrlk_type_names[args.type], strerror(*op_errno));
+ }
- priv = this->private;
+ goto out;
+ }
- ret = fetch_pathinfo (this, inode, op_errno, &brickname);
- if (ret)
- goto out;
-
- end = strrchr (brickname, ':');
- if (!end) {
- GF_FREE(brickname);
- ret = -1;
- goto out;
+ op_ret = fetch_pathinfo(this, inode, op_errno, &brickname);
+ if (op_ret) {
+ gf_log(this->name, GF_LOG_WARNING, "Couldn't get brickname");
+ } else {
+ op_ret = format_brickname(brickname);
+ if (op_ret) {
+ gf_log(this->name, GF_LOG_WARNING, "Couldn't format brickname");
+ GF_FREE(brickname);
+ brickname = NULL;
}
+ }
- tmp = brickname;
- brickname = gf_strndup (brickname, (end - brickname));
- if (brickname == NULL) {
- ret = -1;
- goto out;
+ if (!gcount && !bcount) {
+ if (gf_asprintf(&lk_summary, "No locks cleared.") == -1) {
+ op_ret = -1;
+ *op_errno = ENOMEM;
+ goto out;
}
+ } else if (gf_asprintf(&lk_summary,
+ "%s: %s blocked locks=%d "
+ "granted locks=%d",
+ (brickname == NULL) ? this->name : brickname,
+ clrlk_type_names[args.type], bcount, gcount) == -1) {
+ op_ret = -1;
+ *op_errno = ENOMEM;
+ goto out;
+ }
+ gf_log(this->name, GF_LOG_DEBUG, "%s", lk_summary);
+
+ key = gf_strdup(name);
+ if (!key) {
+ op_ret = -1;
+ goto out;
+ }
+ if (dict_set_dynstr(*dict, key, lk_summary)) {
+ op_ret = -1;
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ op_ret = 0;
- priv->brickname = brickname;
- ret = 0;
out:
- GF_FREE(tmp);
- return ret;
+ GF_FREE(brickname);
+ GF_FREE(args.opts);
+ GF_FREE(key);
+ if (op_ret) {
+ GF_FREE(lk_summary);
+ }
+
+ return op_ret;
}
-char *
-pl_lockinfo_key (xlator_t *this, inode_t *inode, int32_t *op_errno)
+int32_t
+pl_getxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, const char *name,
+ dict_t *xdata)
{
- posix_locks_private_t *priv = NULL;
- char *key = NULL;
- int ret = 0;
+ int32_t op_errno = EINVAL;
+ int32_t op_ret = -1;
+ dict_t *dict = NULL;
- priv = this->private;
+ if (!name)
+ goto usual;
- if (priv->brickname == NULL) {
- ret = pl_lockinfo_get_brickname (this, inode, op_errno);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_WARNING,
- "cannot get brickname");
- goto out;
- }
- }
+ if (strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)))
+ goto usual;
- key = priv->brickname;
-out:
- return key;
+ op_ret = pl_getxattr_clrlk(this, name, loc->inode, &dict, &op_errno);
+
+ STACK_UNWIND_STRICT(getxattr, frame, op_ret, op_errno, dict, xdata);
+
+ if (dict)
+ dict_unref(dict);
+ return 0;
+
+usual:
+ STACK_WIND(frame, pl_getxattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->getxattr, loc, name, xdata);
+ return 0;
}
-int32_t
-pl_fgetxattr_handle_lockinfo (xlator_t *this, fd_t *fd,
- dict_t *dict, int32_t *op_errno)
+static int
+format_brickname(char *brickname)
{
- pl_inode_t *pl_inode = NULL;
- char *key = NULL, *buf = NULL;
- int32_t op_ret = 0;
- unsigned long fdnum = 0;
- int32_t len = 0;
- dict_t *tmp = NULL;
+ int ret = -1;
+ char *hostname = NULL;
+ char *volume = NULL;
+ char *saveptr = NULL;
- pl_inode = pl_inode_get (this, fd->inode);
+ if (!brickname)
+ goto out;
- if (!pl_inode) {
- gf_log (this->name, GF_LOG_DEBUG, "Could not get inode.");
- *op_errno = EBADFD;
- op_ret = -1;
- goto out;
- }
+ strtok_r(brickname, ":", &saveptr);
+ hostname = gf_strdup(strtok_r(NULL, ":", &saveptr));
+ if (hostname == NULL)
+ goto out;
+ volume = gf_strdup(strtok_r(NULL, ".", &saveptr));
+ if (volume == NULL)
+ goto out;
- if (!pl_locks_by_fd (pl_inode, fd)) {
- op_ret = 0;
- goto out;
- }
+ sprintf(brickname, "%s:%s", hostname, volume);
- fdnum = fd_to_fdnum (fd);
+ ret = 0;
+out:
+ GF_FREE(hostname);
+ GF_FREE(volume);
+ return ret;
+}
- key = pl_lockinfo_key (this, fd->inode, op_errno);
- if (key == NULL) {
- op_ret = -1;
- goto out;
- }
+static int
+fetch_pathinfo(xlator_t *this, inode_t *inode, int32_t *op_errno,
+ char **brickname)
+{
+ int ret = -1;
+ loc_t loc = {
+ 0,
+ };
+ dict_t *dict = NULL;
+
+ if (!brickname)
+ goto out;
+
+ if (!op_errno)
+ goto out;
+
+ gf_uuid_copy(loc.gfid, inode->gfid);
+ loc.inode = inode_ref(inode);
+
+ ret = syncop_getxattr(FIRST_CHILD(this), &loc, &dict, GF_XATTR_PATHINFO_KEY,
+ NULL, NULL);
+ if (ret < 0) {
+ *op_errno = -ret;
+ ret = -1;
+ goto out;
+ }
+
+ ret = dict_get_str_sizen(dict, GF_XATTR_PATHINFO_KEY, brickname);
+ if (ret)
+ goto out;
+
+ *brickname = gf_strdup(*brickname);
+ if (*brickname == NULL) {
+ ret = -1;
+ goto out;
+ }
+
+ ret = 0;
+out:
+ if (dict != NULL) {
+ dict_unref(dict);
+ }
+ loc_wipe(&loc);
- tmp = dict_new ();
- if (tmp == NULL) {
- op_ret = -1;
- *op_errno = ENOMEM;
- goto out;
- }
+ return ret;
+}
- op_ret = dict_set_uint64 (tmp, key, fdnum);
- if (op_ret < 0) {
- *op_errno = -op_ret;
- op_ret = -1;
- gf_log (this->name, GF_LOG_WARNING, "setting lockinfo value "
- "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
- fdnum, fd, uuid_utoa (fd->inode->gfid),
- strerror (*op_errno));
- goto out;
- }
+int
+pl_lockinfo_get_brickname(xlator_t *this, inode_t *inode, int32_t *op_errno)
+{
+ posix_locks_private_t *priv = this->private;
+ char *brickname = NULL;
+ char *end = NULL;
+ char *tmp = NULL;
- len = dict_serialized_length (tmp);
- if (len < 0) {
- *op_errno = -op_ret;
- op_ret = -1;
- gf_log (this->name, GF_LOG_WARNING,
- "dict_serialized_length failed (%s) while handling "
- "lockinfo for fd (ptr:%p inode-gfid:%s)",
- strerror (*op_errno), fd, uuid_utoa (fd->inode->gfid));
- goto out;
- }
+ int ret = fetch_pathinfo(this, inode, op_errno, &brickname);
+ if (ret)
+ goto out;
- buf = GF_CALLOC (1, len, gf_common_mt_char);
- if (buf == NULL) {
- op_ret = -1;
- *op_errno = ENOMEM;
- goto out;
- }
+ end = strrchr(brickname, ':');
+ if (!end) {
+ GF_FREE(brickname);
+ ret = -1;
+ goto out;
+ }
+
+ tmp = brickname;
+ brickname = gf_strndup(brickname, (end - brickname));
+ if (brickname == NULL) {
+ ret = -1;
+ goto out;
+ }
+
+ priv->brickname = brickname;
+ ret = 0;
+out:
+ GF_FREE(tmp);
+ return ret;
+}
- op_ret = dict_serialize (tmp, buf);
- if (op_ret < 0) {
- *op_errno = -op_ret;
- op_ret = -1;
- gf_log (this->name, GF_LOG_WARNING,
- "dict_serialize failed (%s) while handling lockinfo "
- "for fd (ptr: %p inode-gfid:%s)", strerror (*op_errno),
- fd, uuid_utoa (fd->inode->gfid));
- goto out;
- }
+char *
+pl_lockinfo_key(xlator_t *this, inode_t *inode, int32_t *op_errno)
+{
+ posix_locks_private_t *priv = this->private;
+ char *key = NULL;
+ int ret = 0;
- op_ret = dict_set_dynptr (dict, GF_XATTR_LOCKINFO_KEY, buf, len);
- if (op_ret < 0) {
- *op_errno = -op_ret;
- op_ret = -1;
- gf_log (this->name, GF_LOG_WARNING, "setting lockinfo value "
- "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
- fdnum, fd, uuid_utoa (fd->inode->gfid),
- strerror (*op_errno));
- goto out;
+ if (priv->brickname == NULL) {
+ ret = pl_lockinfo_get_brickname(this, inode, op_errno);
+ if (ret < 0) {
+ gf_log(this->name, GF_LOG_WARNING, "cannot get brickname");
+ goto out;
}
+ }
- buf = NULL;
+ key = priv->brickname;
out:
- if (tmp != NULL) {
- dict_unref (tmp);
- }
+ return key;
+}
- if (buf != NULL) {
- GF_FREE (buf);
- }
+int32_t
+pl_fgetxattr_handle_lockinfo(xlator_t *this, fd_t *fd, dict_t *dict,
+ int32_t *op_errno)
+{
+ char *key = NULL, *buf = NULL;
+ int32_t op_ret = 0;
+ unsigned long fdnum = 0;
+ int32_t len = 0;
+ dict_t *tmp = NULL;
+
+ pl_inode_t *pl_inode = pl_inode_get(this, fd->inode, NULL);
+
+ if (!pl_inode) {
+ gf_log(this->name, GF_LOG_DEBUG, "Could not get inode.");
+ *op_errno = EBADFD;
+ op_ret = -1;
+ goto out;
+ }
+
+ if (!pl_locks_by_fd(pl_inode, fd)) {
+ op_ret = 0;
+ goto out;
+ }
+
+ fdnum = fd_to_fdnum(fd);
+
+ key = pl_lockinfo_key(this, fd->inode, op_errno);
+ if (key == NULL) {
+ op_ret = -1;
+ goto out;
+ }
+
+ tmp = dict_new();
+ if (tmp == NULL) {
+ op_ret = -1;
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ op_ret = dict_set_uint64(tmp, key, fdnum);
+ if (op_ret < 0) {
+ *op_errno = -op_ret;
+ op_ret = -1;
+ gf_log(this->name, GF_LOG_WARNING,
+ "setting lockinfo value "
+ "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
+ fdnum, fd, uuid_utoa(fd->inode->gfid), strerror(*op_errno));
+ goto out;
+ }
+
+ op_ret = dict_allocate_and_serialize(tmp, (char **)&buf,
+ (unsigned int *)&len);
+ if (op_ret != 0) {
+ *op_errno = -op_ret;
+ op_ret = -1;
+ gf_log(this->name, GF_LOG_WARNING,
+ "dict_serialized_length failed (%s) while handling "
+ "lockinfo for fd (ptr:%p inode-gfid:%s)",
+ strerror(*op_errno), fd, uuid_utoa(fd->inode->gfid));
+ goto out;
+ }
+
+ op_ret = dict_set_dynptr(dict, GF_XATTR_LOCKINFO_KEY, buf, len);
+ if (op_ret < 0) {
+ *op_errno = -op_ret;
+ op_ret = -1;
+ gf_log(this->name, GF_LOG_WARNING,
+ "setting lockinfo value "
+ "(%lu) for fd (ptr:%p inode-gfid:%s) failed (%s)",
+ fdnum, fd, uuid_utoa(fd->inode->gfid), strerror(*op_errno));
+ goto out;
+ }
+
+ buf = NULL;
+out:
+ if (tmp != NULL) {
+ dict_unref(tmp);
+ }
- return op_ret;
-}
+ if (buf != NULL) {
+ GF_FREE(buf);
+ }
+ return op_ret;
+}
int32_t
-pl_fgetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd,
- const char *name, dict_t *xdata)
+pl_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+ dict_t *xdata)
{
- int32_t op_ret = 0, op_errno = 0;
- dict_t *dict = NULL;
-
- if (!name) {
- goto usual;
+ int32_t op_ret = 0, op_errno = 0;
+ dict_t *dict = NULL;
+
+ if (!name) {
+ goto usual;
+ }
+
+ if (strcmp(name, GF_XATTR_LOCKINFO_KEY) == 0) {
+ dict = dict_new();
+ if (dict == NULL) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
}
- if (strcmp (name, GF_XATTR_LOCKINFO_KEY) == 0) {
- dict = dict_new ();
- if (dict == NULL) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+ op_ret = pl_fgetxattr_handle_lockinfo(this, fd, dict, &op_errno);
+ if (op_ret < 0) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "getting lockinfo on fd (ptr:%p inode-gfid:%s) "
+ "failed (%s)",
+ fd, uuid_utoa(fd->inode->gfid), strerror(op_errno));
+ }
- op_ret = pl_fgetxattr_handle_lockinfo (this, fd, dict,
- &op_errno);
- if (op_ret < 0) {
- gf_log (this->name, GF_LOG_WARNING,
- "getting lockinfo on fd (ptr:%p inode-gfid:%s) "
- "failed (%s)", fd, uuid_utoa (fd->inode->gfid),
- strerror (op_errno));
- }
+ goto unwind;
+ } else if (strncmp(name, GF_XATTR_CLRLK_CMD, SLEN(GF_XATTR_CLRLK_CMD)) ==
+ 0) {
+ op_ret = pl_getxattr_clrlk(this, name, fd->inode, &dict, &op_errno);
- goto unwind;
- } else {
- goto usual;
- }
+ goto unwind;
+ } else {
+ goto usual;
+ }
unwind:
- STACK_UNWIND_STRICT (fgetxattr, frame, op_ret, op_errno, dict, NULL);
- if (dict != NULL) {
- dict_unref (dict);
- }
+ STACK_UNWIND_STRICT(fgetxattr, frame, op_ret, op_errno, dict, NULL);
+ if (dict != NULL) {
+ dict_unref(dict);
+ }
- return 0;
+ return 0;
usual:
- STACK_WIND (frame, default_fgetxattr_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
- return 0;
+ STACK_WIND(frame, default_fgetxattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fgetxattr, fd, name, xdata);
+ return 0;
}
int32_t
-pl_migrate_locks (call_frame_t *frame, fd_t *newfd, uint64_t oldfd_num,
- int32_t *op_errno)
+pl_migrate_locks(call_frame_t *frame, fd_t *newfd, uint64_t oldfd_num,
+ int32_t *op_errno)
{
- pl_inode_t *pl_inode = NULL;
- uint64_t newfd_num = 0;
- posix_lock_t *l = NULL;
- int32_t op_ret = 0;
-
- newfd_num = fd_to_fdnum (newfd);
-
- pl_inode = pl_inode_get (frame->this, newfd->inode);
- if (pl_inode == NULL) {
- op_ret = -1;
- *op_errno = EBADFD;
- goto out;
- }
-
- pthread_mutex_lock (&pl_inode->mutex);
+ posix_lock_t *l = NULL;
+ int32_t op_ret = 0;
+ uint64_t newfd_num = fd_to_fdnum(newfd);
+
+ pl_inode_t *pl_inode = pl_inode_get(frame->this, newfd->inode, NULL);
+ if (pl_inode == NULL) {
+ op_ret = -1;
+ *op_errno = EBADFD;
+ goto out;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry(l, &pl_inode->ext_list, list)
{
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->fd_num == oldfd_num) {
- l->fd_num = newfd_num;
- l->client = frame->root->client;
- }
- }
+ if (l->fd_num == oldfd_num) {
+ l->fd_num = newfd_num;
+ l->client = frame->root->client;
+ }
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- op_ret = 0;
+ op_ret = 0;
out:
- return op_ret;
+ return op_ret;
}
int32_t
-pl_fsetxattr_handle_lockinfo (call_frame_t *frame, fd_t *fd, char *lockinfo_buf,
- int len, int32_t *op_errno)
+pl_fsetxattr_handle_lockinfo(call_frame_t *frame, fd_t *fd, char *lockinfo_buf,
+ int len, int32_t *op_errno)
{
- int32_t op_ret = -1;
- dict_t *lockinfo = NULL;
- uint64_t oldfd_num = 0;
- char *key = NULL;
-
- lockinfo = dict_new ();
- if (lockinfo == NULL) {
- op_ret = -1;
- *op_errno = ENOMEM;
- goto out;
- }
+ int32_t op_ret = -1;
+ uint64_t oldfd_num = 0;
+ char *key = NULL;
+
+ dict_t *lockinfo = dict_new();
+ if (lockinfo == NULL) {
+ op_ret = -1;
+ *op_errno = ENOMEM;
+ goto out;
+ }
+
+ op_ret = dict_unserialize(lockinfo_buf, len, &lockinfo);
+ if (op_ret < 0) {
+ *op_errno = -op_ret;
+ op_ret = -1;
+ goto out;
+ }
+
+ key = pl_lockinfo_key(frame->this, fd->inode, op_errno);
+ if (key == NULL) {
+ op_ret = -1;
+ goto out;
+ }
+
+ op_ret = dict_get_uint64(lockinfo, key, &oldfd_num);
+
+ if (oldfd_num == 0) {
+ op_ret = 0;
+ goto out;
+ }
+
+ op_ret = pl_migrate_locks(frame, fd, oldfd_num, op_errno);
+ if (op_ret < 0) {
+ gf_log(frame->this->name, GF_LOG_WARNING,
+ "migration of locks from oldfd (ptr:%p) to newfd "
+ "(ptr:%p) (inode-gfid:%s)",
+ (void *)(uintptr_t)oldfd_num, fd, uuid_utoa(fd->inode->gfid));
+ goto out;
+ }
- op_ret = dict_unserialize (lockinfo_buf, len, &lockinfo);
- if (op_ret < 0) {
- *op_errno = -op_ret;
- op_ret = -1;
- goto out;
- }
+out:
+ dict_unref(lockinfo);
- key = pl_lockinfo_key (frame->this, fd->inode, op_errno);
- if (key == NULL) {
- op_ret = -1;
- goto out;
- }
+ return op_ret;
+}
- op_ret = dict_get_uint64 (lockinfo, key, &oldfd_num);
+int32_t
+pl_fsetxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
- if (oldfd_num == 0) {
- op_ret = 0;
- goto out;
+ local = frame->local;
+ if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+ pl_inode = pl_inode_get(this, local->inode, NULL);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
}
- op_ret = pl_migrate_locks (frame, fd, oldfd_num, op_errno);
- if (op_ret < 0) {
- gf_log (frame->this->name, GF_LOG_WARNING,
- "migration of locks from oldfd (ptr:%p) to newfd "
- "(ptr:%p) (inode-gfid:%s)", (void *)oldfd_num, fd,
- uuid_utoa (fd->inode->gfid));
- goto out;
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ pl_inode->mlock_enforced = _gf_true;
+ pl_inode->check_mlock_info = _gf_false;
}
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
-out:
- dict_unref (lockinfo);
-
- return op_ret;
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(fsetxattr, xdata, frame, op_ret, op_errno,
+ xdata);
+ return 0;
}
int32_t
-pl_fsetxattr (call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
- int32_t flags, dict_t *xdata)
+pl_fsetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *dict,
+ int32_t flags, dict_t *xdata)
{
- int32_t op_ret = 0, op_errno = 0;
- void *lockinfo_buf = NULL;
- int len = 0;
+ int32_t op_errno = 0;
+ void *lockinfo_buf = NULL;
+ int len = 0;
+ char *name = NULL;
+ posix_locks_private_t *priv = this->private;
+
+ int32_t op_ret = dict_get_ptr_and_len(dict, GF_XATTR_LOCKINFO_KEY,
+ &lockinfo_buf, &len);
+ if (lockinfo_buf == NULL) {
+ goto usual;
+ }
+
+ op_ret = pl_fsetxattr_handle_lockinfo(frame, fd, lockinfo_buf, len,
+ &op_errno);
+ if (op_ret < 0) {
+ goto unwind;
+ }
- op_ret = dict_get_ptr_and_len (dict, GF_XATTR_LOCKINFO_KEY,
- &lockinfo_buf, &len);
- if (lockinfo_buf == NULL) {
- goto usual;
- }
+usual:
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
- op_ret = pl_fsetxattr_handle_lockinfo (frame, fd, lockinfo_buf, len,
- &op_errno);
- if (op_ret < 0) {
- goto unwind;
- }
+ PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, ((loc_t *)NULL), fd,
+ priv);
-usual:
- STACK_WIND (frame, default_fsetxattr_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
- return 0;
+ STACK_WIND(frame, pl_fsetxattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fsetxattr, fd, dict, flags, xdata);
+ return 0;
unwind:
- STACK_UNWIND_STRICT (fsetxattr, frame, op_ret, op_errno, NULL);
- return 0;
+ PL_STACK_UNWIND_FOR_CLIENT(fsetxattr, xdata, frame, op_ret, op_errno, NULL);
+
+ return 0;
}
int32_t
-pl_opendir_cbk (call_frame_t *frame,
- void *cookie,
- xlator_t *this,
- int32_t op_ret,
- int32_t op_errno,
- fd_t *fd, dict_t *xdata)
+pl_opendir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
{
- pl_fdctx_t *fdctx = NULL;
+ pl_fdctx_t *fdctx = NULL;
- if (op_ret < 0)
- goto unwind;
+ if (op_ret < 0)
+ goto unwind;
- fdctx = pl_check_n_create_fdctx (this, fd);
- if (!fdctx) {
- op_errno = ENOMEM;
- op_ret = -1;
- goto unwind;
- }
+ fdctx = pl_check_n_create_fdctx(this, fd);
+ if (!fdctx) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unwind;
+ }
unwind:
- STACK_UNWIND_STRICT (opendir,
- frame,
- op_ret,
- op_errno,
- fd, xdata);
- return 0;
+ PL_STACK_UNWIND(opendir, xdata, frame, op_ret, op_errno, fd, xdata);
+
+ return 0;
}
int32_t
-pl_opendir (call_frame_t *frame, xlator_t *this,
- loc_t *loc, fd_t *fd, dict_t *xdata)
-{
- STACK_WIND (frame,
- pl_opendir_cbk,
- FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->opendir,
- loc, fd, xdata);
- return 0;
-
+pl_opendir(call_frame_t *frame, xlator_t *this, loc_t *loc, fd_t *fd,
+ dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_opendir_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->opendir, loc, fd, xdata);
+ return 0;
}
int
-pl_flush_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, dict_t *xdata)
+pl_flush_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, dict_t *xdata)
{
- STACK_UNWIND_STRICT (flush, frame, op_ret, op_errno, xdata);
+ PL_STACK_UNWIND_FOR_CLIENT(flush, xdata, frame, op_ret, op_errno, xdata);
- return 0;
+ return 0;
}
-
int
-pl_flush (call_frame_t *frame, xlator_t *this,
- fd_t *fd, dict_t *xdata)
+pl_flush(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
{
- pl_inode_t *pl_inode = NULL;
-
- pl_inode = pl_inode_get (this, fd->inode);
-
- if (!pl_inode) {
- gf_log (this->name, GF_LOG_DEBUG, "Could not get inode.");
- STACK_UNWIND_STRICT (flush, frame, -1, EBADFD, NULL);
- return 0;
- }
-
- pl_trace_flush (this, frame, fd);
-
- if (frame->root->lk_owner.len == 0) {
- /* Handle special case when protocol/server sets lk-owner to zero.
- * This usually happens due to a client disconnection. Hence, free
- * all locks opened with this fd.
- */
- gf_log (this->name, GF_LOG_TRACE,
- "Releasing all locks with fd %p", fd);
- delete_locks_of_fd (this, pl_inode, fd);
- goto wind;
-
- }
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __delete_locks_of_owner (pl_inode, frame->root->client,
- &frame->root->lk_owner);
+ pl_inode_t *pl_inode = pl_inode_get(this, fd->inode, NULL);
+ if (!pl_inode) {
+ gf_log(this->name, GF_LOG_DEBUG, "Could not get inode.");
+ STACK_UNWIND_STRICT(flush, frame, -1, EBADFD, NULL);
+ return 0;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (pl_inode->migrated) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ STACK_UNWIND_STRICT(flush, frame, -1, EREMOTE, NULL);
+ return 0;
}
- pthread_mutex_unlock (&pl_inode->mutex);
-
- grant_blocked_locks (this, pl_inode);
-
- do_blocked_rw (pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ pl_trace_flush(this, frame, fd);
+
+ if (frame->root->lk_owner.len == 0) {
+ /* Handle special case when protocol/server sets lk-owner to zero.
+ * This usually happens due to a client disconnection. Hence, free
+ * all locks opened with this fd.
+ */
+ gf_log(this->name, GF_LOG_TRACE, "Releasing all locks with fd %p", fd);
+ delete_locks_of_fd(this, pl_inode, fd);
+ goto wind;
+ }
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __delete_locks_of_owner(pl_inode, frame->root->client,
+ &frame->root->lk_owner);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ grant_blocked_locks(this, pl_inode);
+
+ do_blocked_rw(pl_inode);
wind:
- STACK_WIND (frame, pl_flush_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->flush, fd, xdata);
- return 0;
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_flush_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->flush, fd, xdata);
+ return 0;
}
-
int
-pl_open_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, fd_t *fd, dict_t *xdata)
+pl_open_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, fd_t *fd, dict_t *xdata)
{
- pl_fdctx_t *fdctx = NULL;
+ pl_fdctx_t *fdctx = NULL;
- if (op_ret < 0)
- goto unwind;
+ if (op_ret < 0)
+ goto unwind;
- fdctx = pl_check_n_create_fdctx (this, fd);
- if (!fdctx) {
- op_errno = ENOMEM;
- op_ret = -1;
- goto unwind;
- }
+ fdctx = pl_check_n_create_fdctx(this, fd);
+ if (!fdctx) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unwind;
+ }
unwind:
- STACK_UNWIND_STRICT (open, frame, op_ret, op_errno, fd, xdata);
+ STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, fd, xdata);
- return 0;
+ return 0;
}
-
int
-pl_open (call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
- fd_t *fd, dict_t *xdata)
+pl_open(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+ fd_t *fd, dict_t *xdata)
{
- STACK_WIND (frame, pl_open_cbk,
- FIRST_CHILD(this), FIRST_CHILD(this)->fops->open,
- loc, flags, fd, xdata);
+ int op_ret = -1;
+ int op_errno = EINVAL;
+ pl_inode_t *pl_inode = NULL;
+ posix_lock_t *l = NULL;
+ posix_locks_private_t *priv = this->private;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ op_ret = 0, op_errno = 0;
+ pl_inode = pl_inode_get(this, fd->inode, NULL);
+ if (!pl_inode) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM, "Could not get inode");
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ /* As per design, under forced and file-based mandatory locking modes
+ * it doesn't matter whether inodes's lock list contain advisory or
+ * mandatory type locks. So we just check whether inode's lock list is
+ * empty or not to make sure that no locks are being held for the file.
+ * Whereas under optimal mandatory locking mode, we strictly fail open
+ * if and only if lock list contain mandatory locks.
+ */
+ if (((priv->mandatory_mode == MLK_FILE_BASED) && pl_inode->mandatory) ||
+ priv->mandatory_mode == MLK_FORCED) {
+ if (fd->flags & O_TRUNC) {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (!list_empty(&pl_inode->ext_list)) {
+ op_ret = -1;
+ op_errno = EAGAIN;
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+ } else if (priv->mandatory_mode == MLK_OPTIMAL) {
+ if (fd->flags & O_TRUNC) {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if ((l->lk_flags & GF_LK_MANDATORY)) {
+ op_ret = -1;
+ op_errno = EAGAIN;
+ break;
+ }
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+ }
- return 0;
+unwind:
+ if (op_ret == -1)
+ STACK_UNWIND_STRICT(open, frame, op_ret, op_errno, NULL, NULL);
+ else
+ STACK_WIND(frame, pl_open_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->open, loc, flags, fd, xdata);
+ return 0;
}
-
int
-pl_create_cbk (call_frame_t *frame, void *cookie,
- xlator_t *this, int32_t op_ret, int32_t op_errno,
- fd_t *fd, inode_t *inode, struct iatt *buf,
- struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+pl_create_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, fd_t *fd, inode_t *inode, struct iatt *buf,
+ struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
{
- pl_fdctx_t *fdctx = NULL;
+ pl_fdctx_t *fdctx = NULL;
- if (op_ret < 0)
- goto unwind;
+ if (op_ret < 0)
+ goto unwind;
- fdctx = pl_check_n_create_fdctx (this, fd);
- if (!fdctx) {
- op_errno = ENOMEM;
- op_ret = -1;
- goto unwind;
- }
+ fdctx = pl_check_n_create_fdctx(this, fd);
+ if (!fdctx) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unwind;
+ }
unwind:
- STACK_UNWIND_STRICT (create, frame, op_ret, op_errno, fd, inode, buf,
- preparent, postparent, xdata);
+ PL_STACK_UNWIND(create, xdata, frame, op_ret, op_errno, fd, inode, buf,
+ preparent, postparent, xdata);
- return 0;
+ return 0;
}
-
int
-pl_create (call_frame_t *frame, xlator_t *this,
- loc_t *loc, int32_t flags, mode_t mode, mode_t umask, fd_t *fd,
- dict_t *xdata)
+pl_create(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t flags,
+ mode_t mode, mode_t umask, fd_t *fd, dict_t *xdata)
{
- STACK_WIND (frame, pl_create_cbk,
- FIRST_CHILD (this), FIRST_CHILD (this)->fops->create,
- loc, flags, mode, umask, fd, xdata);
- return 0;
-}
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_create_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->create, loc, flags, mode, umask, fd,
+ xdata);
+ return 0;
+}
int
-pl_readv_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno,
- struct iovec *vector, int32_t count, struct iatt *stbuf,
- struct iobref *iobref, dict_t *xdata)
+pl_readv_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iovec *vector, int32_t count,
+ struct iatt *stbuf, struct iobref *iobref, dict_t *xdata)
{
- STACK_UNWIND_STRICT (readv, frame, op_ret, op_errno,
- vector, count, stbuf, iobref, xdata);
+ pl_track_io_fop_count(frame->local, this, DECREMENT);
- return 0;
+ PL_STACK_UNWIND(readv, xdata, frame, op_ret, op_errno, vector, count, stbuf,
+ iobref, xdata);
+
+ return 0;
}
int
-pl_writev_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int32_t op_ret, int32_t op_errno, struct iatt *prebuf,
- struct iatt *postbuf, dict_t *xdata)
+pl_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+ dict_t *xdata)
{
- STACK_UNWIND_STRICT (writev, frame, op_ret, op_errno, prebuf, postbuf,
- xdata);
+ pl_track_io_fop_count(frame->local, this, DECREMENT);
- return 0;
-}
+ PL_STACK_UNWIND(writev, xdata, frame, op_ret, op_errno, prebuf, postbuf,
+ xdata);
+ return 0;
+}
void
-do_blocked_rw (pl_inode_t *pl_inode)
+do_blocked_rw(pl_inode_t *pl_inode)
{
- struct list_head wind_list;
- pl_rw_req_t *rw = NULL;
- pl_rw_req_t *tmp = NULL;
+ struct list_head wind_list;
+ pl_rw_req_t *rw = NULL;
+ pl_rw_req_t *tmp = NULL;
- INIT_LIST_HEAD (&wind_list);
+ INIT_LIST_HEAD(&wind_list);
- pthread_mutex_lock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_for_each_entry_safe(rw, tmp, &pl_inode->rw_list, list)
{
- list_for_each_entry_safe (rw, tmp, &pl_inode->rw_list, list) {
- if (__rw_allowable (pl_inode, &rw->region,
- rw->stub->fop)) {
- list_del_init (&rw->list);
- list_add_tail (&rw->list, &wind_list);
- }
+ if (__rw_allowable(pl_inode, &rw->region, rw->stub->fop)) {
+ list_del_init(&rw->list);
+ list_add_tail(&rw->list, &wind_list);
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
}
+ }
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- list_for_each_entry_safe (rw, tmp, &wind_list, list) {
- list_del_init (&rw->list);
- call_resume (rw->stub);
- GF_FREE (rw);
- }
+ list_for_each_entry_safe(rw, tmp, &wind_list, list)
+ {
+ list_del_init(&rw->list);
+ call_resume(rw->stub);
+ GF_FREE(rw);
+ }
- return;
+ return;
}
+/* when mandatory lock is enforced:
+ If an IO request comes on a region which is out of the boundary of the
+ granted mandatory lock, it will be rejected.
+
+ Note: There is no IO blocking with mandatory lock enforced as it may be
+ a stale data from an old client.
+ */
+gf_boolean_t static within_range(posix_lock_t *existing, posix_lock_t *new)
+{
+ if (existing->fl_start <= new->fl_start && existing->fl_end >= new->fl_end)
+ return _gf_true;
+
+ return _gf_false;
+}
static int
-__rw_allowable (pl_inode_t *pl_inode, posix_lock_t *region,
- glusterfs_fop_t op)
+__rw_allowable(pl_inode_t *pl_inode, posix_lock_t *region, glusterfs_fop_t op)
{
- posix_lock_t *l = NULL;
- int ret = 1;
+ posix_lock_t *l = NULL;
+ posix_locks_private_t *priv = THIS->private;
+ int ret = 1;
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (locks_overlap (l, region) && !same_owner (l, region)) {
- if ((op == GF_FOP_READ) && (l->fl_type != F_WRLCK))
- continue;
- ret = 0;
- break;
+ if (pl_inode->mlock_enforced) {
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ /*
+ with lock enforced (fencing) there should not be any blocking
+ lock coexisting.
+ */
+ if (same_owner(l, region)) {
+ /* Should range check be strict for same owner with fencing? */
+ if (locks_overlap(l, region)) {
+ if (within_range(l, region)) {
+ return 1;
+ } else {
+ /*
+ Should we allow read fop if it does not fit it in the
+ range?
+ if (op == GF_FOP_READ && l->fl_type != F_WRLCK) {
+ return 1;
+ }
+ */
+ return 0;
+ }
+ }
+ } else {
+ if (locks_overlap(l, region)) {
+ /*
+ with fencing should a read from a different owner be
+ allowed if the mandatory lock taken is F_RDLCK?
+ if (op == GF_FOP_READ && l->fl_type != F_WRLCK) {
+ return 1;
+ }
+ */
+ return 0;
}
+ }
}
- return ret;
-}
+ /* No lock has been taken by this owner */
+ return 0;
+ }
+
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (!l->blocked && locks_overlap(l, region) && !same_owner(l, region)) {
+ if ((op == GF_FOP_READ) && (l->fl_type != F_WRLCK))
+ continue;
+ /* Check for mandatory lock under optimal
+ * mandatory-locking mode */
+ if (priv->mandatory_mode == MLK_OPTIMAL &&
+ !(l->lk_flags & GF_LK_MANDATORY))
+ continue;
+ ret = 0;
+ break;
+ }
+ }
+ return ret;
+}
int
-pl_readv_cont (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
- off_t offset, uint32_t flags, dict_t *xdata)
+pl_readv_cont(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, uint32_t flags, dict_t *xdata)
{
- STACK_WIND (frame, pl_readv_cbk,
- FIRST_CHILD (this), FIRST_CHILD (this)->fops->readv,
- fd, size, offset, flags, xdata);
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
- return 0;
-}
+ STACK_WIND(frame, pl_readv_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readv, fd, size, offset, flags, xdata);
+ return 0;
+}
int
-pl_readv (call_frame_t *frame, xlator_t *this,
- fd_t *fd, size_t size, off_t offset, uint32_t flags, dict_t *xdata)
-{
- posix_locks_private_t *priv = NULL;
- pl_inode_t *pl_inode = NULL;
- pl_rw_req_t *rw = NULL;
- posix_lock_t region = {.list = {0, }, };
- int op_ret = 0;
- int op_errno = 0;
- char wind_needed = 1;
-
-
- priv = this->private;
- pl_inode = pl_inode_get (this, fd->inode);
-
- if (priv->mandatory && pl_inode->mandatory) {
- region.fl_start = offset;
- region.fl_end = offset + size - 1;
- region.client = frame->root->client;
- region.fd_num = fd_to_fdnum(fd);
- region.client_pid = frame->root->pid;
- region.owner = frame->root->lk_owner;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- wind_needed = __rw_allowable (pl_inode, &region,
- GF_FOP_READ);
- if (wind_needed) {
- goto unlock;
- }
-
- if (fd->flags & O_NONBLOCK) {
- gf_log (this->name, GF_LOG_TRACE,
- "returning EAGAIN as fd is O_NONBLOCK");
- op_errno = EAGAIN;
- op_ret = -1;
- goto unlock;
- }
-
- rw = GF_CALLOC (1, sizeof (*rw),
- gf_locks_mt_pl_rw_req_t);
- if (!rw) {
- op_errno = ENOMEM;
- op_ret = -1;
- goto unlock;
- }
-
- rw->stub = fop_readv_stub (frame, pl_readv_cont,
- fd, size, offset, flags,
- xdata);
- if (!rw->stub) {
- op_errno = ENOMEM;
- op_ret = -1;
- GF_FREE (rw);
- goto unlock;
- }
+pl_readv(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, uint32_t flags, dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+ pl_rw_req_t *rw = NULL;
+ posix_lock_t region = {
+ .list =
+ {
+ 0,
+ },
+ };
+ gf_boolean_t enabled = _gf_false;
+ gf_boolean_t can_block = _gf_true;
+ int op_ret = 0;
+ int op_errno = 0;
+ int allowed = 1;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+ if (!frame->local) {
+ frame->local = mem_get0(this->local_pool);
+ local = frame->local;
+ local->inode = inode_ref(fd->inode);
+ local->fd = fd_ref(fd);
+ }
+
+ pl_inode = pl_inode_get(this, fd->inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ if (frame->root->pid < 0)
+ enabled = _gf_false;
+ else
+ enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+ if (enabled) {
+ region.fl_start = offset;
+ region.fl_end = offset + size - 1;
+ region.client = frame->root->client;
+ region.fd_num = fd_to_fdnum(fd);
+ region.client_pid = frame->root->pid;
+ region.owner = frame->root->lk_owner;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_READ,
+ &can_block);
+ if (allowed == 1) {
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
+ }
+ goto unlock;
+ } else if (!can_block) {
+ op_errno = EAGAIN;
+ op_ret = -1;
+ goto unlock;
+ }
- rw->region = region;
+ rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+ if (!rw) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unlock;
+ }
- list_add_tail (&rw->list, &pl_inode->rw_list);
- }
- unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
- }
+ rw->stub = fop_readv_stub(frame, pl_readv_cont, fd, size, offset,
+ flags, xdata);
+ if (!rw->stub) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ GF_FREE(rw);
+ goto unlock;
+ }
+ rw->region = region;
- if (wind_needed) {
- STACK_WIND (frame, pl_readv_cbk,
- FIRST_CHILD (this), FIRST_CHILD (this)->fops->readv,
- fd, size, offset, flags, xdata);
+ list_add_tail(&rw->list, &pl_inode->rw_list);
}
+ unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+
+ if (allowed == 1) {
+ STACK_WIND(frame, pl_readv_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readv, fd, size, offset, flags,
+ xdata);
+ }
+unwind:
+ if (op_ret == -1)
+ PL_STACK_UNWIND(readv, xdata, frame, op_ret, op_errno, NULL, 0, NULL,
+ NULL, NULL);
- if (op_ret == -1)
- STACK_UNWIND_STRICT (readv, frame, -1, op_errno,
- NULL, 0, NULL, NULL, NULL);
-
- return 0;
+ return 0;
}
-
int
-pl_writev_cont (call_frame_t *frame, xlator_t *this, fd_t *fd,
- struct iovec *vector, int count, off_t offset,
- uint32_t flags, struct iobref *iobref, dict_t *xdata)
+pl_writev_cont(call_frame_t *frame, xlator_t *this, fd_t *fd,
+ struct iovec *vector, int count, off_t offset, uint32_t flags,
+ struct iobref *iobref, dict_t *xdata)
{
- STACK_WIND (frame, pl_writev_cbk,
- FIRST_CHILD (this), FIRST_CHILD (this)->fops->writev,
- fd, vector, count, offset, flags, iobref, xdata);
+ pl_track_io_fop_count(frame->local, this, INCREMENT);
- return 0;
-}
+ STACK_WIND(frame, pl_writev_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+ flags, iobref, xdata);
+ return 0;
+}
int
-pl_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
- struct iovec *vector, int32_t count, off_t offset,
- uint32_t flags, struct iobref *iobref, dict_t *xdata)
-{
- posix_locks_private_t *priv = NULL;
- pl_inode_t *pl_inode = NULL;
- pl_rw_req_t *rw = NULL;
- posix_lock_t region = {.list = {0, }, };
- int op_ret = 0;
- int op_errno = 0;
- char wind_needed = 1;
-
- priv = this->private;
- pl_inode = pl_inode_get (this, fd->inode);
-
- if (priv->mandatory && pl_inode->mandatory) {
- region.fl_start = offset;
- region.fl_end = offset + iov_length (vector, count) - 1;
- region.client = frame->root->client;
- region.fd_num = fd_to_fdnum(fd);
- region.client_pid = frame->root->pid;
- region.owner = frame->root->lk_owner;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- wind_needed = __rw_allowable (pl_inode, &region,
- GF_FOP_WRITE);
- if (wind_needed)
- goto unlock;
-
- if (fd->flags & O_NONBLOCK) {
- gf_log (this->name, GF_LOG_TRACE,
- "returning EAGAIN because fd is "
- "O_NONBLOCK");
- op_errno = EAGAIN;
- op_ret = -1;
- goto unlock;
- }
-
- rw = GF_CALLOC (1, sizeof (*rw),
- gf_locks_mt_pl_rw_req_t);
- if (!rw) {
- op_errno = ENOMEM;
- op_ret = -1;
- goto unlock;
- }
-
- rw->stub = fop_writev_stub (frame, pl_writev_cont,
- fd, vector, count, offset,
- flags, iobref, xdata);
- if (!rw->stub) {
- op_errno = ENOMEM;
- op_ret = -1;
- GF_FREE (rw);
- goto unlock;
- }
+pl_writev(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iovec *vector,
+ int32_t count, off_t offset, uint32_t flags, struct iobref *iobref,
+ dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+ pl_rw_req_t *rw = NULL;
+ posix_lock_t region = {
+ .list =
+ {
+ 0,
+ },
+ };
+ gf_boolean_t enabled = _gf_false;
+ gf_boolean_t can_block = _gf_true;
+ int op_ret = 0;
+ int op_errno = 0;
+ int allowed = 1;
+
+ GF_VALIDATE_OR_GOTO("locks", this, unwind);
+
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+ if (!frame->local) {
+ frame->local = mem_get0(this->local_pool);
+ local = frame->local;
+ local->inode = inode_ref(fd->inode);
+ local->fd = fd_ref(fd);
+ }
+
+ pl_inode = pl_inode_get(this, fd->inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ if (frame->root->pid < 0)
+ enabled = _gf_false;
+ else
+ enabled = pl_is_mandatory_locking_enabled(pl_inode);
+
+ if (enabled) {
+ region.fl_start = offset;
+ region.fl_end = offset + iov_length(vector, count) - 1;
+ region.client = frame->root->client;
+ region.fd_num = fd_to_fdnum(fd);
+ region.client_pid = frame->root->pid;
+ region.owner = frame->root->lk_owner;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ allowed = pl_is_fop_allowed(pl_inode, &region, fd, GF_FOP_WRITE,
+ &can_block);
+ if (allowed == 1) {
+ if (pl_inode->mlock_enforced &&
+ pl_inode->track_fop_wind_count) {
+ pl_inode->fop_wind_count++;
+ }
+ goto unlock;
+ } else if (!can_block) {
+ if (pl_inode->mlock_enforced) {
+ op_errno = EBUSY;
+ } else {
+ op_errno = EAGAIN;
+ }
- rw->region = region;
+ op_ret = -1;
+ goto unlock;
+ }
- list_add_tail (&rw->list, &pl_inode->rw_list);
- }
- unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
- }
+ rw = GF_MALLOC(sizeof(*rw), gf_locks_mt_pl_rw_req_t);
+ if (!rw) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ goto unlock;
+ }
+ rw->stub = fop_writev_stub(frame, pl_writev_cont, fd, vector, count,
+ offset, flags, iobref, xdata);
+ if (!rw->stub) {
+ op_errno = ENOMEM;
+ op_ret = -1;
+ GF_FREE(rw);
+ goto unlock;
+ }
- if (wind_needed)
- STACK_WIND (frame, pl_writev_cbk,
- FIRST_CHILD (this), FIRST_CHILD (this)->fops->writev,
- fd, vector, count, offset, flags, iobref, xdata);
+ rw->region = region;
- if (op_ret == -1)
- STACK_UNWIND_STRICT (writev, frame, -1, op_errno, NULL, NULL,
- NULL);
+ list_add_tail(&rw->list, &pl_inode->rw_list);
+ }
+ unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+
+ if (allowed == 1) {
+ STACK_WIND(frame, pl_writev_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->writev, fd, vector, count, offset,
+ flags, iobref, xdata);
+ }
+unwind:
+ if (op_ret == -1)
+ PL_STACK_UNWIND(writev, xdata, frame, op_ret, op_errno, NULL, NULL,
+ NULL);
- return 0;
+ return 0;
}
static int
-__fd_has_locks (pl_inode_t *pl_inode, fd_t *fd)
+__fd_has_locks(pl_inode_t *pl_inode, fd_t *fd)
{
- int found = 0;
- posix_lock_t *l = NULL;
+ posix_lock_t *l = NULL;
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->fd_num == fd_to_fdnum(fd)) {
- found = 1;
- break;
- }
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (l->fd_num == fd_to_fdnum(fd)) {
+ return 1;
}
+ }
- return found;
+ return 0;
}
static posix_lock_t *
-lock_dup (posix_lock_t *lock)
+lock_dup(posix_lock_t *lock)
{
- posix_lock_t *new_lock = NULL;
-
- new_lock = new_posix_lock (&lock->user_flock, lock->client,
- lock->client_pid, &lock->owner,
- (fd_t *)lock->fd_num);
- return new_lock;
+ int32_t op_errno = 0;
+ return new_posix_lock(&lock->user_flock, lock->client, lock->client_pid,
+ &lock->owner, (fd_t *)lock->fd_num, lock->lk_flags,
+ lock->blocking, &op_errno);
}
static int
-__dup_locks_to_fdctx (pl_inode_t *pl_inode, fd_t *fd,
- pl_fdctx_t *fdctx)
-{
- posix_lock_t *l = NULL;
- posix_lock_t *duplock = NULL;
- int ret = 0;
-
- list_for_each_entry (l, &pl_inode->ext_list, list) {
- if (l->fd_num == fd_to_fdnum(fd)) {
- duplock = lock_dup (l);
- if (!duplock) {
- ret = -1;
- break;
- }
+__dup_locks_to_fdctx(pl_inode_t *pl_inode, fd_t *fd, pl_fdctx_t *fdctx)
+{
+ posix_lock_t *l = NULL;
+ posix_lock_t *duplock = NULL;
+ int ret = 0;
+
+ list_for_each_entry(l, &pl_inode->ext_list, list)
+ {
+ if (l->fd_num == fd_to_fdnum(fd)) {
+ duplock = lock_dup(l);
+ if (!duplock) {
+ ret = -1;
+ break;
+ }
- list_add_tail (&duplock->list, &fdctx->locks_list);
- }
+ list_add_tail(&duplock->list, &fdctx->locks_list);
}
+ }
- return ret;
+ return ret;
}
static int
-__copy_locks_to_fdctx (pl_inode_t *pl_inode, fd_t *fd,
- pl_fdctx_t *fdctx)
+__copy_locks_to_fdctx(pl_inode_t *pl_inode, fd_t *fd, pl_fdctx_t *fdctx)
{
- int ret = 0;
-
- ret = __dup_locks_to_fdctx (pl_inode, fd, fdctx);
- if (ret)
- goto out;
-
-out:
- return ret;
-
+ return __dup_locks_to_fdctx(pl_inode, fd, fdctx);
}
static void
-pl_mark_eol_lock (posix_lock_t *lock)
+pl_mark_eol_lock(posix_lock_t *lock)
{
- lock->user_flock.l_type = GF_LK_EOL;
- return;
+ lock->user_flock.l_type = GF_LK_EOL;
+ return;
}
static posix_lock_t *
-__get_next_fdctx_lock (pl_fdctx_t *fdctx)
+__get_next_fdctx_lock(pl_fdctx_t *fdctx)
{
- posix_lock_t *lock = NULL;
+ posix_lock_t *lock = NULL;
- GF_ASSERT (fdctx);
+ GF_ASSERT(fdctx);
- if (list_empty (&fdctx->locks_list)) {
- gf_log (THIS->name, GF_LOG_DEBUG,
- "fdctx lock list empty");
- goto out;
- }
+ if (list_empty(&fdctx->locks_list)) {
+ gf_log(THIS->name, GF_LOG_DEBUG, "fdctx lock list empty");
+ goto out;
+ }
- lock = list_entry (fdctx->locks_list.next, typeof (*lock),
- list);
+ lock = list_entry(fdctx->locks_list.next, typeof(*lock), list);
- GF_ASSERT (lock);
+ GF_ASSERT(lock);
- list_del_init (&lock->list);
+ list_del_init(&lock->list);
out:
- return lock;
+ return lock;
}
static int
-__set_next_lock_fd (pl_fdctx_t *fdctx, posix_lock_t *reqlock)
+__set_next_lock_fd(pl_fdctx_t *fdctx, posix_lock_t *reqlock)
{
- posix_lock_t *lock = NULL;
- int ret = 0;
+ posix_lock_t *lock = NULL;
+ int ret = 0;
- GF_ASSERT (fdctx);
+ GF_ASSERT(fdctx);
- lock = __get_next_fdctx_lock (fdctx);
- if (!lock) {
- gf_log (THIS->name, GF_LOG_DEBUG,
- "marking EOL in reqlock");
- pl_mark_eol_lock (reqlock);
- goto out;
- }
+ lock = __get_next_fdctx_lock(fdctx);
+ if (!lock) {
+ gf_log(THIS->name, GF_LOG_DEBUG, "marking EOL in reqlock");
+ pl_mark_eol_lock(reqlock);
+ goto out;
+ }
- reqlock->user_flock = lock->user_flock;
- reqlock->fl_start = lock->fl_start;
- reqlock->fl_type = lock->fl_type;
- reqlock->fl_end = lock->fl_end;
- reqlock->owner = lock->owner;
+ reqlock->user_flock = lock->user_flock;
+ reqlock->fl_start = lock->fl_start;
+ reqlock->fl_type = lock->fl_type;
+ reqlock->fl_end = lock->fl_end;
+ reqlock->owner = lock->owner;
out:
- if (lock)
- __destroy_lock (lock);
+ if (lock)
+ __destroy_lock(lock);
- return ret;
+ return ret;
}
static int
-pl_getlk_fd (xlator_t *this, pl_inode_t *pl_inode,
- fd_t *fd, posix_lock_t *reqlock)
+pl_getlk_fd(xlator_t *this, pl_inode_t *pl_inode, fd_t *fd,
+ posix_lock_t *reqlock)
{
- uint64_t tmp = 0;
- pl_fdctx_t *fdctx = NULL;
- int ret = 0;
+ uint64_t tmp = 0;
+ pl_fdctx_t *fdctx = NULL;
+ int ret = 0;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (!__fd_has_locks(pl_inode, fd)) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_log(this->name, GF_LOG_DEBUG, "fd=%p has no active locks", fd);
+ ret = 0;
+ goto out;
+ }
- pthread_mutex_lock (&pl_inode->mutex);
- {
- if (!__fd_has_locks (pl_inode, fd)) {
- gf_log (this->name, GF_LOG_DEBUG,
- "fd=%p has no active locks", fd);
- ret = 0;
- goto unlock;
- }
+ gf_log(this->name, GF_LOG_DEBUG, "There are active locks on fd");
- gf_log (this->name, GF_LOG_DEBUG,
- "There are active locks on fd");
+ ret = fd_ctx_get(fd, this, &tmp);
+ fdctx = (pl_fdctx_t *)(long)tmp;
- ret = fd_ctx_get (fd, this, &tmp);
- fdctx = (pl_fdctx_t *)(long) tmp;
+ if (list_empty(&fdctx->locks_list)) {
+ gf_log(this->name, GF_LOG_TRACE,
+ "no fdctx -> copying all locks on fd");
- if (list_empty (&fdctx->locks_list)) {
- gf_log (this->name, GF_LOG_TRACE,
- "no fdctx -> copying all locks on fd");
+ ret = __copy_locks_to_fdctx(pl_inode, fd, fdctx);
+ if (ret) {
+ goto unlock;
+ }
- ret = __copy_locks_to_fdctx (pl_inode, fd, fdctx);
- if (ret) {
- goto unlock;
- }
+ ret = __set_next_lock_fd(fdctx, reqlock);
- ret = __set_next_lock_fd (fdctx, reqlock);
-
- } else {
- gf_log (this->name, GF_LOG_TRACE,
- "fdctx present -> returning the next lock");
- ret = __set_next_lock_fd (fdctx, reqlock);
- if (ret) {
- gf_log (this->name, GF_LOG_DEBUG,
- "could not get next lock of fd");
- goto unlock;
- }
- }
+ } else {
+ gf_log(this->name, GF_LOG_TRACE,
+ "fdctx present -> returning the next lock");
+ ret = __set_next_lock_fd(fdctx, reqlock);
+ if (ret) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_log(this->name, GF_LOG_DEBUG,
+ "could not get next lock of fd");
+ goto out;
+ }
}
+ }
unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
- return ret;
-
+ pthread_mutex_unlock(&pl_inode->mutex);
+out:
+ return ret;
}
int
-pl_lk (call_frame_t *frame, xlator_t *this,
- fd_t *fd, int32_t cmd, struct gf_flock *flock, dict_t *xdata)
-{
- pl_inode_t *pl_inode = NULL;
- int op_ret = 0;
- int op_errno = 0;
- int can_block = 0;
- posix_lock_t *reqlock = NULL;
- posix_lock_t *conf = NULL;
- int ret = 0;
-
- if ((flock->l_start < 0) || (flock->l_len < 0)) {
- op_ret = -1;
- op_errno = EINVAL;
- goto unwind;
- }
+pl_metalock_is_active(pl_inode_t *pl_inode)
+{
+ if (list_empty(&pl_inode->metalk_list))
+ return 0;
+ else
+ return 1;
+}
- pl_inode = pl_inode_get (this, fd->inode);
- if (!pl_inode) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+void
+__pl_queue_lock(pl_inode_t *pl_inode, posix_lock_t *reqlock)
+{
+ list_add_tail(&reqlock->list, &pl_inode->queued_locks);
+}
- reqlock = new_posix_lock (flock, frame->root->client, frame->root->pid,
- &frame->root->lk_owner, fd);
+int
+pl_lk(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t cmd,
+ struct gf_flock *flock, dict_t *xdata)
+{
+ pl_inode_t *pl_inode = NULL;
+ int op_ret = 0;
+ int op_errno = 0;
+ int can_block = 0;
+ posix_lock_t *reqlock = NULL;
+ posix_lock_t *conf = NULL;
+ uint32_t lk_flags = 0;
+ posix_locks_private_t *priv = this->private;
+ pl_local_t *local = NULL;
+ short lock_type = 0;
+
+ int ret = dict_get_uint32(xdata, GF_LOCK_MODE, &lk_flags);
+ if (ret == 0) {
+ if (priv->mandatory_mode == MLK_NONE)
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Lock flags received "
+ "in a non-mandatory locking environment, "
+ "continuing");
+ else
+ gf_log(this->name, GF_LOG_DEBUG,
+ "Lock flags received, "
+ "continuing");
+ }
+
+ if ((flock->l_start < 0) || ((flock->l_start + flock->l_len) < 0)) {
+ op_ret = -1;
+ op_errno = EINVAL;
+ goto unwind;
+ }
+
+ /* As per 'man 3 fcntl', the value of l_len may be
+ * negative. In such cases, lock request should be
+ * considered for the range starting at 'l_start+l_len'
+ * and ending at 'l_start-1'. Update the fields accordingly.
+ */
+ if (flock->l_len < 0) {
+ flock->l_start += flock->l_len;
+ flock->l_len = labs(flock->l_len);
+ }
+
+ local = mem_get0(this->local_pool);
+ if (!local) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ } else {
+ frame->local = local;
+ local->fd = fd_ref(fd);
+ }
- if (!reqlock) {
- op_ret = -1;
- op_errno = ENOMEM;
- goto unwind;
- }
+ pl_inode = pl_inode_get(this, fd->inode, local);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
+
+ reqlock = new_posix_lock(flock, frame->root->client, frame->root->pid,
+ &frame->root->lk_owner, fd, lk_flags, can_block,
+ &op_errno);
- pl_trace_in (this, frame, fd, NULL, cmd, flock, NULL);
+ if (!reqlock) {
+ op_ret = -1;
+ goto unwind;
+ }
- switch (cmd) {
+ pl_trace_in(this, frame, fd, NULL, cmd, flock, NULL);
+ switch (cmd) {
case F_RESLK_LCKW:
- can_block = 1;
+ can_block = 1;
- /* fall through */
+ /* fall through */
case F_RESLK_LCK:
- memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
- reqlock->frame = frame;
- reqlock->this = this;
+ reqlock->frame = frame;
+ reqlock->this = this;
- ret = pl_reserve_setlk (this, pl_inode, reqlock,
- can_block);
- if (ret < 0) {
- if (can_block)
- goto out;
+ ret = pl_reserve_setlk(this, pl_inode, reqlock, can_block);
+ if (ret < 0) {
+ if (can_block)
+ goto out;
- op_ret = -1;
- op_errno = -ret;
- __destroy_lock (reqlock);
- goto unwind;
- }
- /* Finally a getlk and return the call */
- conf = pl_getlk (pl_inode, reqlock);
- if (conf)
- posix_lock_to_flock (conf, flock);
- break;
+ op_ret = -1;
+ op_errno = -ret;
+ __destroy_lock(reqlock);
+ goto unwind;
+ }
+ /* Finally a getlk and return the call */
+ conf = pl_getlk(pl_inode, reqlock);
+ if (conf)
+ posix_lock_to_flock(conf, flock);
+ break;
case F_RESLK_UNLCK:
- reqlock->frame = frame;
- reqlock->this = this;
- ret = pl_reserve_unlock (this, pl_inode, reqlock);
- if (ret < 0) {
- op_ret = -1;
- op_errno = -ret;
- }
- __destroy_lock (reqlock);
- goto unwind;
+ reqlock->frame = frame;
+ reqlock->this = this;
+ ret = pl_reserve_unlock(this, pl_inode, reqlock);
+ if (ret < 0) {
+ op_ret = -1;
+ op_errno = -ret;
+ }
+ __destroy_lock(reqlock);
+ goto unwind;
- break;
+ break;
case F_GETLK_FD:
- reqlock->frame = frame;
- reqlock->this = this;
- ret = pl_verify_reservelk (this, pl_inode, reqlock, can_block);
- GF_ASSERT (ret >= 0);
-
- ret = pl_getlk_fd (this, pl_inode, fd, reqlock);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_DEBUG,
- "getting locks on fd failed");
- op_ret = -1;
- op_errno = ENOLCK;
- goto unwind;
- }
+ reqlock->frame = frame;
+ reqlock->this = this;
+ ret = pl_verify_reservelk(this, pl_inode, reqlock, can_block);
+ GF_ASSERT(ret >= 0);
+
+ ret = pl_getlk_fd(this, pl_inode, fd, reqlock);
+ if (ret < 0) {
+ gf_log(this->name, GF_LOG_DEBUG, "getting locks on fd failed");
+ op_ret = -1;
+ op_errno = ENOLCK;
+ goto unwind;
+ }
- gf_log (this->name, GF_LOG_TRACE,
- "Replying with a lock on fd for healing");
+ gf_log(this->name, GF_LOG_TRACE,
+ "Replying with a lock on fd for healing");
- posix_lock_to_flock (reqlock, flock);
- __destroy_lock (reqlock);
+ posix_lock_to_flock(reqlock, flock);
+ __destroy_lock(reqlock);
- break;
+ break;
#if F_GETLK != F_GETLK64
case F_GETLK64:
#endif
case F_GETLK:
- conf = pl_getlk (pl_inode, reqlock);
- posix_lock_to_flock (conf, flock);
- __destroy_lock (reqlock);
+ conf = pl_getlk(pl_inode, reqlock);
+ posix_lock_to_flock(conf, flock);
+ __destroy_lock(reqlock);
- break;
+ break;
#if F_SETLKW != F_SETLKW64
case F_SETLKW64:
#endif
case F_SETLKW:
- can_block = 1;
- reqlock->frame = frame;
- reqlock->this = this;
-
- /* fall through */
+ can_block = 1;
+ reqlock->frame = frame;
+ reqlock->this = this;
+ reqlock->blocking = can_block;
+ /* fall through */
#if F_SETLK != F_SETLK64
case F_SETLK64:
#endif
case F_SETLK:
- memcpy (&reqlock->user_flock, flock, sizeof (struct gf_flock));
- ret = pl_verify_reservelk (this, pl_inode, reqlock, can_block);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_TRACE,
- "Lock blocked due to conflicting reserve lock");
- goto out;
+ reqlock->frame = frame;
+ reqlock->this = this;
+ lock_type = flock->l_type;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (pl_inode->migrated) {
+ op_errno = EREMOTE;
+ pthread_mutex_unlock(&pl_inode->mutex);
+ STACK_UNWIND_STRICT(lk, frame, -1, op_errno, flock, xdata);
+
+ __destroy_lock(reqlock);
+ goto out;
}
- ret = pl_setlk (this, pl_inode, reqlock,
- can_block);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ ret = pl_verify_reservelk(this, pl_inode, reqlock, can_block);
+ if (ret < 0) {
+ gf_log(this->name, GF_LOG_TRACE,
+ "Lock blocked due to conflicting reserve lock");
+ goto out;
+ }
+ if (reqlock->fl_type != F_UNLCK && pl_inode->mlock_enforced) {
+ ret = pl_lock_preempt(pl_inode, reqlock);
if (ret == -1) {
- if ((can_block) && (F_UNLCK != flock->l_type)) {
- pl_trace_block (this, frame, fd, NULL, cmd, flock, NULL);
- goto out;
- }
- gf_log (this->name, GF_LOG_DEBUG, "returning EAGAIN");
- op_ret = -1;
- op_errno = EAGAIN;
- __destroy_lock (reqlock);
-
- } else if ((0 == ret) && (F_UNLCK == flock->l_type)) {
- /* For NLM's last "unlock on fd" detection */
- if (pl_locks_by_fd (pl_inode, fd))
- flock->l_type = F_RDLCK;
- else
- flock->l_type = F_UNLCK;
+ gf_log(this->name, GF_LOG_ERROR, "lock preempt failed");
+ op_ret = -1;
+ op_errno = EAGAIN;
+ __destroy_lock(reqlock);
+ goto out;
}
- }
-unwind:
- pl_trace_out (this, frame, fd, NULL, cmd, flock, op_ret, op_errno, NULL);
- pl_update_refkeeper (this, fd->inode);
+ pl_trace_block(this, frame, fd, NULL, cmd, flock, NULL);
+ goto unwind;
+ }
+ ret = pl_setlk(this, pl_inode, reqlock, can_block);
+ if (ret == -1) {
+ if ((can_block) && (F_UNLCK != lock_type)) {
+ goto out;
+ }
+ gf_log(this->name, GF_LOG_DEBUG, "returning EAGAIN");
+ op_ret = -1;
+ op_errno = EAGAIN;
+ __destroy_lock(reqlock);
+ } else if (ret == -2) {
+ goto out;
+ } else if ((0 == ret) && (F_UNLCK == flock->l_type)) {
+ /* For NLM's last "unlock on fd" detection */
+ if (pl_locks_by_fd(pl_inode, fd))
+ flock->l_type = F_RDLCK;
+ else
+ flock->l_type = F_UNLCK;
+ }
+ }
- STACK_UNWIND_STRICT (lk, frame, op_ret, op_errno, flock, xdata);
+unwind:
+ pl_trace_out(this, frame, fd, NULL, cmd, flock, op_ret, op_errno, NULL);
+ pl_update_refkeeper(this, fd->inode);
+
+ PL_STACK_UNWIND(lk, xdata, frame, op_ret, op_errno, flock, xdata);
out:
- return 0;
+ return 0;
}
-
/* TODO: this function just logs, no action required?? */
int
-pl_forget (xlator_t *this,
- inode_t *inode)
+pl_forget(xlator_t *this, inode_t *inode)
{
- pl_inode_t *pl_inode = NULL;
+ pl_inode_t *pl_inode = NULL;
+
+ posix_lock_t *ext_tmp = NULL;
+ posix_lock_t *ext_l = NULL;
+ struct list_head posixlks_released;
- posix_lock_t *ext_tmp = NULL;
- posix_lock_t *ext_l = NULL;
- struct list_head posixlks_released;
+ pl_inode_lock_t *ino_tmp = NULL;
+ pl_inode_lock_t *ino_l = NULL;
+ struct list_head inodelks_released;
- pl_inode_lock_t *ino_tmp = NULL;
- pl_inode_lock_t *ino_l = NULL;
- struct list_head inodelks_released;
+ pl_rw_req_t *rw_tmp = NULL;
+ pl_rw_req_t *rw_req = NULL;
- pl_rw_req_t *rw_tmp = NULL;
- pl_rw_req_t *rw_req = NULL;
+ pl_entry_lock_t *entry_tmp = NULL;
+ pl_entry_lock_t *entry_l = NULL;
+ struct list_head entrylks_released;
- pl_entry_lock_t *entry_tmp = NULL;
- pl_entry_lock_t *entry_l = NULL;
- struct list_head entrylks_released;
+ pl_dom_list_t *dom = NULL;
+ pl_dom_list_t *dom_tmp = NULL;
- pl_dom_list_t *dom = NULL;
- pl_dom_list_t *dom_tmp = NULL;
+ INIT_LIST_HEAD(&posixlks_released);
+ INIT_LIST_HEAD(&inodelks_released);
+ INIT_LIST_HEAD(&entrylks_released);
- INIT_LIST_HEAD (&posixlks_released);
- INIT_LIST_HEAD (&inodelks_released);
- INIT_LIST_HEAD (&entrylks_released);
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (!pl_inode)
+ return 0;
- pl_inode = pl_inode_get (this, inode);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (!list_empty(&pl_inode->rw_list)) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "Pending R/W requests found, releasing.");
+
+ list_for_each_entry_safe(rw_req, rw_tmp, &pl_inode->rw_list, list)
+ {
+ list_del(&rw_req->list);
+ call_stub_destroy(rw_req->stub);
+ GF_FREE(rw_req);
+ }
+ }
- pthread_mutex_lock (&pl_inode->mutex);
- {
+ if (!list_empty(&pl_inode->ext_list)) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "Pending fcntl locks found, releasing.");
- if (!list_empty (&pl_inode->rw_list)) {
- gf_log (this->name, GF_LOG_WARNING,
- "Pending R/W requests found, releasing.");
+ list_for_each_entry_safe(ext_l, ext_tmp, &pl_inode->ext_list, list)
+ {
+ __delete_lock(ext_l);
+ if (ext_l->blocked) {
+ list_add_tail(&ext_l->list, &posixlks_released);
+ continue;
+ }
+ __destroy_lock(ext_l);
+ }
+ }
- list_for_each_entry_safe (rw_req, rw_tmp, &pl_inode->rw_list,
- list) {
+ list_for_each_entry_safe(dom, dom_tmp, &pl_inode->dom_list, inode_list)
+ {
+ if (!list_empty(&dom->inodelk_list)) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "Pending inode locks found, releasing.");
- list_del (&rw_req->list);
- GF_FREE (rw_req);
- }
+ list_for_each_entry_safe(ino_l, ino_tmp, &dom->inodelk_list,
+ list)
+ {
+ __delete_inode_lock(ino_l);
+ __pl_inodelk_unref(ino_l);
}
- if (!list_empty (&pl_inode->ext_list)) {
- gf_log (this->name, GF_LOG_WARNING,
- "Pending fcntl locks found, releasing.");
+ list_splice_init(&dom->blocked_inodelks, &inodelks_released);
+ }
+ if (!list_empty(&dom->entrylk_list)) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "Pending entry locks found, releasing.");
- list_for_each_entry_safe (ext_l, ext_tmp, &pl_inode->ext_list,
- list) {
+ list_for_each_entry_safe(entry_l, entry_tmp, &dom->entrylk_list,
+ domain_list)
+ {
+ list_del_init(&entry_l->domain_list);
- __delete_lock (pl_inode, ext_l);
- if (ext_l->blocked) {
- list_add_tail (&ext_l->list, &posixlks_released);
- continue;
- }
- __destroy_lock (ext_l);
- }
+ GF_FREE((char *)entry_l->basename);
+ GF_FREE(entry_l->connection_id);
+ GF_FREE(entry_l);
}
+ list_splice_init(&dom->blocked_entrylks, &entrylks_released);
+ }
- list_for_each_entry_safe (dom, dom_tmp, &pl_inode->dom_list, inode_list) {
+ list_del(&dom->inode_list);
+ gf_log("posix-locks", GF_LOG_TRACE, " Cleaning up domain: %s",
+ dom->domain);
+ GF_FREE((char *)(dom->domain));
+ GF_FREE(dom);
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- if (!list_empty (&dom->inodelk_list)) {
- gf_log (this->name, GF_LOG_WARNING,
- "Pending inode locks found, releasing.");
+ if (!list_empty(&posixlks_released)) {
+ list_for_each_entry_safe(ext_l, ext_tmp, &posixlks_released, list)
+ {
+ STACK_UNWIND_STRICT(lk, ext_l->frame, -1, 0, &ext_l->user_flock,
+ NULL);
+ __destroy_lock(ext_l);
+ }
+ }
- list_for_each_entry_safe (ino_l, ino_tmp, &dom->inodelk_list, list) {
- __delete_inode_lock (ino_l);
- __pl_inodelk_unref (ino_l);
- }
+ if (!list_empty(&inodelks_released)) {
+ list_for_each_entry_safe(ino_l, ino_tmp, &inodelks_released,
+ blocked_locks)
+ {
+ STACK_UNWIND_STRICT(inodelk, ino_l->frame, -1, 0, NULL);
+ __pl_inodelk_unref(ino_l);
+ }
+ }
- list_splice_init (&dom->blocked_inodelks, &inodelks_released);
+ if (!list_empty(&entrylks_released)) {
+ list_for_each_entry_safe(entry_l, entry_tmp, &entrylks_released,
+ blocked_locks)
+ {
+ STACK_UNWIND_STRICT(entrylk, entry_l->frame, -1, 0, NULL);
+ GF_FREE((char *)entry_l->basename);
+ GF_FREE(entry_l->connection_id);
+ GF_FREE(entry_l);
+ }
+ }
+ pthread_mutex_destroy(&pl_inode->mutex);
- }
- if (!list_empty (&dom->entrylk_list)) {
- gf_log (this->name, GF_LOG_WARNING,
- "Pending entry locks found, releasing.");
+ GF_FREE(pl_inode);
- list_for_each_entry_safe (entry_l, entry_tmp, &dom->entrylk_list, domain_list) {
- list_del_init (&entry_l->domain_list);
+ return 0;
+}
- GF_FREE ((char *)entry_l->basename);
- GF_FREE (entry_l->connection_id);
- GF_FREE (entry_l);
- }
+int
+pl_release(xlator_t *this, fd_t *fd)
+{
+ pl_inode_t *pl_inode = NULL;
+ uint64_t tmp_pl_inode = 0;
+ int ret = -1;
+ uint64_t tmp = 0;
+ pl_fdctx_t *fdctx = NULL;
- list_splice_init (&dom->blocked_entrylks, &entrylks_released);
- }
+ if (fd == NULL) {
+ goto out;
+ }
- list_del (&dom->inode_list);
- gf_log ("posix-locks", GF_LOG_TRACE,
- " Cleaning up domain: %s", dom->domain);
- GF_FREE ((char *)(dom->domain));
- GF_FREE (dom);
- }
+ ret = inode_ctx_get(fd->inode, this, &tmp_pl_inode);
+ if (ret != 0)
+ goto clean;
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
- list_for_each_entry_safe (ext_l, ext_tmp, &posixlks_released, list) {
+ pl_trace_release(this, fd);
- STACK_UNWIND_STRICT (lk, ext_l->frame, -1, 0,
- &ext_l->user_flock, NULL);
- __destroy_lock (ext_l);
- }
+ gf_log(this->name, GF_LOG_TRACE, "Releasing all locks with fd %p", fd);
- list_for_each_entry_safe (ino_l, ino_tmp, &inodelks_released, blocked_locks) {
+ delete_locks_of_fd(this, pl_inode, fd);
+ pl_update_refkeeper(this, fd->inode);
- STACK_UNWIND_STRICT (inodelk, ino_l->frame, -1, 0, NULL);
- __pl_inodelk_unref (ino_l);
- }
+clean:
+ ret = fd_ctx_del(fd, this, &tmp);
+ if (ret) {
+ gf_log(this->name, GF_LOG_DEBUG, "Could not get fdctx");
+ goto out;
+ }
+
+ fdctx = (pl_fdctx_t *)(long)tmp;
- list_for_each_entry_safe (entry_l, entry_tmp, &entrylks_released, blocked_locks) {
+ GF_FREE(fdctx);
+out:
+ return ret;
+}
- STACK_UNWIND_STRICT (entrylk, entry_l->frame, -1, 0, NULL);
- GF_FREE ((char *)entry_l->basename);
- GF_FREE (entry_l->connection_id);
- GF_FREE (entry_l);
+int
+pl_releasedir(xlator_t *this, fd_t *fd)
+{
+ int ret = -1;
+ uint64_t tmp = 0;
+ pl_fdctx_t *fdctx = NULL;
- }
+ if (fd == NULL) {
+ goto out;
+ }
- GF_FREE (pl_inode);
+ ret = fd_ctx_del(fd, this, &tmp);
+ if (ret) {
+ gf_log(this->name, GF_LOG_DEBUG, "Could not get fdctx");
+ goto out;
+ }
- return 0;
+ fdctx = (pl_fdctx_t *)(long)tmp;
+
+ GF_FREE(fdctx);
+out:
+ return ret;
}
-int
-pl_release (xlator_t *this, fd_t *fd)
+static int32_t
+pl_request_link_count(dict_t **pxdata)
{
- pl_inode_t *pl_inode = NULL;
- uint64_t tmp_pl_inode = 0;
- int ret = -1;
- uint64_t tmp = 0;
- pl_fdctx_t *fdctx = NULL;
+ dict_t *xdata;
- if (fd == NULL) {
- goto out;
+ xdata = *pxdata;
+ if (xdata == NULL) {
+ xdata = dict_new();
+ if (xdata == NULL) {
+ return ENOMEM;
}
+ } else {
+ dict_ref(xdata);
+ }
- ret = inode_ctx_get (fd->inode, this, &tmp_pl_inode);
- if (ret != 0)
- goto out;
+ if (dict_set_uint32(xdata, GET_LINK_COUNT, 0) != 0) {
+ dict_unref(xdata);
+ return ENOMEM;
+ }
- pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+ *pxdata = xdata;
- pl_trace_release (this, fd);
+ return 0;
+}
- gf_log (this->name, GF_LOG_TRACE,
- "Releasing all locks with fd %p", fd);
+static int32_t
+pl_check_link_count(dict_t *xdata)
+{
+ int32_t count;
- delete_locks_of_fd (this, pl_inode, fd);
- pl_update_refkeeper (this, fd->inode);
+ /* In case we are unable to read the link count from xdata, we take a
+ * conservative approach and return -2, which will prevent the inode from
+ * being considered deleted. In fact it will cause link tracking for this
+ * inode to be disabled completely to avoid races. */
- ret = fd_ctx_del (fd, this, &tmp);
- if (ret) {
- gf_log (this->name, GF_LOG_DEBUG,
- "Could not get fdctx");
- goto out;
- }
+ if (xdata == NULL) {
+ return -2;
+ }
- fdctx = (pl_fdctx_t *)(long)tmp;
+ if (dict_get_int32(xdata, GET_LINK_COUNT, &count) != 0) {
+ return -2;
+ }
- GF_FREE (fdctx);
-out:
- return ret;
+ return count;
}
-int
-pl_releasedir (xlator_t *this, fd_t *fd)
+int32_t
+pl_lookup_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, inode_t *inode, struct iatt *buf, dict_t *xdata,
+ struct iatt *postparent)
{
- int ret = -1;
- uint64_t tmp = 0;
- pl_fdctx_t *fdctx = NULL;
+ pl_inode_t *pl_inode;
- if (fd == NULL) {
- goto out;
+ if (op_ret >= 0) {
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (pl_inode == NULL) {
+ PL_STACK_UNWIND(lookup, xdata, frame, -1, ENOMEM, NULL, NULL, NULL,
+ NULL);
+ return 0;
}
- ret = fd_ctx_del (fd, this, &tmp);
- if (ret) {
- gf_log (this->name, GF_LOG_DEBUG,
- "Could not get fdctx");
- goto out;
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ /* We only update the link count if we previously didn't know it.
+ * Doing it always can lead to races since lookup is not executed
+ * atomically most of the times. */
+ if (pl_inode->links == -2) {
+ pl_inode->links = pl_check_link_count(xdata);
+ if (buf->ia_type == IA_IFDIR) {
+ /* Directories have at least 2 links. To avoid special handling
+ * for directories, we simply decrement the value here to make
+ * them equivalent to regular files. */
+ pl_inode->links--;
+ }
}
- fdctx = (pl_fdctx_t *)(long)tmp;
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- GF_FREE (fdctx);
-out:
- return ret;
+ PL_STACK_UNWIND(lookup, xdata, frame, op_ret, op_errno, inode, buf, xdata,
+ postparent);
+ return 0;
}
int32_t
-__get_posixlk_count (xlator_t *this, pl_inode_t *pl_inode)
+pl_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
{
- posix_lock_t *lock = NULL;
- int32_t count = 0;
-
- list_for_each_entry (lock, &pl_inode->ext_list, list) {
-
- count++;
- }
+ int32_t error;
+
+ error = pl_request_link_count(&xdata);
+ if (error == 0) {
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_lookup_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->lookup, loc, xdata);
+ dict_unref(xdata);
+ } else {
+ STACK_UNWIND_STRICT(lookup, frame, -1, error, NULL, NULL, NULL, NULL);
+ }
+ return 0;
+}
- return count;
+int32_t
+pl_fstat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+ PL_STACK_UNWIND(fstat, xdata, frame, op_ret, op_errno, buf, xdata);
+ return 0;
}
int32_t
-get_posixlk_count (xlator_t *this, inode_t *inode)
+pl_fstat(call_frame_t *frame, xlator_t *this, fd_t *fd, dict_t *xdata)
{
- pl_inode_t *pl_inode = NULL;
- uint64_t tmp_pl_inode = 0;
- int ret = 0;
- int32_t count = 0;
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fstat_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fstat, fd, xdata);
+ return 0;
+}
- ret = inode_ctx_get (inode, this, &tmp_pl_inode);
- if (ret != 0) {
- goto out;
- }
+int
+pl_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
+ int op_errno, gf_dirent_t *entries, dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ gf_dirent_t *entry = NULL;
- pl_inode = (pl_inode_t *)(long) tmp_pl_inode;
+ if (op_ret <= 0)
+ goto unwind;
- pthread_mutex_lock (&pl_inode->mutex);
- {
- count =__get_posixlk_count (this, pl_inode);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ local = frame->local;
+ if (!local)
+ goto unwind;
-out:
- return count;
+ list_for_each_entry(entry, &entries->list, list)
+ {
+ pl_set_xdata_response(this, local, local->fd->inode, entry->inode,
+ entry->d_name, entry->dict, 0);
+ }
+
+unwind:
+ PL_STACK_UNWIND(readdirp, xdata, frame, op_ret, op_errno, entries, xdata);
+
+ return 0;
}
-void
-pl_parent_entrylk_xattr_fill (xlator_t *this, inode_t *parent,
- char *basename, dict_t *dict)
+int
+pl_readdirp(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, dict_t *xdata)
{
- uint32_t entrylk = 0;
- int ret = -1;
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_readdirp_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readdirp, fd, size, offset, xdata);
+
+ return 0;
+}
+
+lock_migration_info_t *
+gf_mig_info_for_lock(posix_lock_t *lock)
+{
+ lock_migration_info_t *new = GF_MALLOC(sizeof(lock_migration_info_t),
+ gf_common_mt_lock_mig);
+ if (new == NULL) {
+ goto out;
+ }
+
+ INIT_LIST_HEAD(&new->list);
+
+ posix_lock_to_flock(lock, &new->flock);
+
+ new->lk_flags = lock->lk_flags;
+
+ new->client_uid = gf_strdup(lock->client_uid);
- if (!parent || !basename || !strlen (basename))
- goto out;
- entrylk = check_entrylk_on_basename (this, parent, basename);
out:
- ret = dict_set_uint32 (dict, GLUSTERFS_PARENT_ENTRYLK, entrylk);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_DEBUG,
- " dict_set failed on key %s", GLUSTERFS_PARENT_ENTRYLK);
- }
+ return new;
}
-void
-pl_entrylk_xattr_fill (xlator_t *this, inode_t *inode,
- dict_t *dict)
+int
+pl_fill_active_locks(pl_inode_t *pl_inode, lock_migration_info_t *lmi)
{
- int32_t count = 0;
- int ret = -1;
+ posix_lock_t *temp = NULL;
+ lock_migration_info_t *newlock = NULL;
+ int count = 0;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (list_empty(&pl_inode->ext_list)) {
+ count = 0;
+ goto unlock;
+ }
- count = get_entrylk_count (this, inode);
- ret = dict_set_int32 (dict, GLUSTERFS_ENTRYLK_COUNT, count);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_DEBUG,
- " dict_set failed on key %s", GLUSTERFS_ENTRYLK_COUNT);
+ list_for_each_entry(temp, &pl_inode->ext_list, list)
+ {
+ if (temp->blocked)
+ continue;
+
+ newlock = gf_mig_info_for_lock(temp);
+ if (!newlock) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "lock_dup failed");
+ count = -1;
+ goto out;
+ }
+
+ list_add_tail(&newlock->list, &lmi->list);
+ count++;
}
+ }
+unlock:
+ pthread_mutex_unlock(&pl_inode->mutex);
+out:
+ return count;
}
-void
-pl_inodelk_xattr_fill (xlator_t *this, inode_t *inode, dict_t *dict,
- gf_boolean_t per_dom)
+/* This function reads only active locks */
+static int
+pl_getactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
{
- int32_t count = 0;
- int ret = -1;
- char *domname = NULL;
+ pl_inode_t *pl_inode = NULL;
+ lock_migration_info_t locks;
+ int op_ret = 0;
+ int op_errno = 0;
+ int count = 0;
+ INIT_LIST_HEAD(&locks.list);
- if (per_dom){
- ret = dict_get_str (dict, GLUSTERFS_INODELK_DOM_COUNT,
- &domname);
- if (ret) {
- gf_log (this->name, GF_LOG_ERROR, "Failed to get "
- "value for key %s",GLUSTERFS_INODELK_DOM_COUNT);
- goto out;
- }
- }
+ pl_inode = pl_inode_get(this, loc->inode, NULL);
+ if (!pl_inode) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_inode_get failed");
- count = get_inodelk_count (this, inode, domname);
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto out;
+ }
- ret = dict_set_int32 (dict, GLUSTERFS_INODELK_COUNT, count);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_DEBUG, "Failed to set count for "
- "key %s", GLUSTERFS_INODELK_COUNT);
- }
+ count = pl_fill_active_locks(pl_inode, &locks);
+
+ op_ret = count;
out:
- return;
+ STACK_UNWIND_STRICT(getactivelk, frame, op_ret, op_errno, &locks, NULL);
+
+ gf_free_mig_locks(&locks);
+
+ return 0;
}
void
-pl_posixlk_xattr_fill (xlator_t *this, inode_t *inode,
- dict_t *dict)
+pl_metalk_unref(pl_meta_lock_t *lock)
{
- int32_t count = 0;
- int ret = -1;
+ lock->ref--;
+ if (!lock->ref) {
+ GF_FREE(lock->client_uid);
+ GF_FREE(lock);
+ }
+}
- count = get_posixlk_count (this, inode);
- ret = dict_set_int32 (dict, GLUSTERFS_POSIXLK_COUNT, count);
- if (ret < 0) {
- gf_log (this->name, GF_LOG_DEBUG,
- " dict_set failed on key %s", GLUSTERFS_POSIXLK_COUNT);
- }
+void
+__pl_metalk_ref(pl_meta_lock_t *lock)
+{
+ lock->ref++;
+}
+pl_meta_lock_t *
+new_meta_lock(call_frame_t *frame, xlator_t *this)
+{
+ pl_meta_lock_t *lock = GF_CALLOC(1, sizeof(*lock),
+ gf_locks_mt_pl_meta_lock_t);
+
+ if (!lock) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+ "mem allocation"
+ " failed for meta lock");
+ goto out;
+ }
+
+ INIT_LIST_HEAD(&lock->list);
+ INIT_LIST_HEAD(&lock->client_list);
+
+ lock->client_uid = gf_strdup(frame->root->client->client_uid);
+ if (!lock->client_uid) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+ "mem allocation"
+ " failed for client_uid");
+ GF_FREE(lock);
+ lock = NULL;
+ goto out;
+ }
+
+ __pl_metalk_ref(lock);
+out:
+ return lock;
}
-int32_t
-pl_lookup_cbk (call_frame_t *frame,
- void *cookie,
- xlator_t *this,
- int32_t op_ret,
- int32_t op_errno,
- inode_t *inode,
- struct iatt *buf,
- dict_t *xdata,
- struct iatt *postparent)
+int
+pl_insert_metalk(pl_inode_t *pl_inode, pl_ctx_t *ctx, pl_meta_lock_t *lock)
{
- pl_local_t *local = NULL;
+ int ret = 0;
- GF_VALIDATE_OR_GOTO (this->name, frame->local, out);
+ if (!pl_inode || !ctx || !lock) {
+ gf_msg(THIS->name, GF_LOG_INFO, 0, 0, "NULL parameter");
+ ret = -1;
+ goto out;
+ }
- if (op_ret)
- goto out;
+ lock->pl_inode = pl_inode;
- local = frame->local;
+ /* refer function pl_inode_setlk for more info for this ref.
+ * This should be unrefed on meta-unlock triggered by rebalance or
+ * in cleanup with client disconnect*/
+ /*TODO: unref this in cleanup code for disconnect and meta-unlock*/
+ pl_inode->inode = inode_ref(pl_inode->inode);
- if (local->parent_entrylk_req)
- pl_parent_entrylk_xattr_fill (this, local->loc.parent,
- (char*)local->loc.name, xdata);
- if (local->entrylk_count_req)
- pl_entrylk_xattr_fill (this, inode, xdata);
- if (local->inodelk_count_req)
- pl_inodelk_xattr_fill (this, inode, xdata, _gf_false);
- if (local->inodelk_dom_count_req)
- pl_inodelk_xattr_fill (this, inode, xdata, _gf_true);
- if (local->posixlk_count_req)
- pl_posixlk_xattr_fill (this, inode, xdata);
+ /* NOTE:In case of a client-server disconnect we need to cleanup metalk.
+ * Hence, adding the metalk to pl_ctx_t as well. The mutex lock order
+ * should always be on ctx and then on pl_inode*/
+
+ pthread_mutex_lock(&ctx->lock);
+ {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ list_add_tail(&lock->list, &pl_inode->metalk_list);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ list_add_tail(&lock->client_list, &ctx->metalk_list);
+ }
+ pthread_mutex_unlock(&ctx->lock);
out:
- local = frame->local;
- frame->local = NULL;
-
- if (local != NULL) {
- loc_wipe (&local->loc);
- mem_put (local);
- }
-
- STACK_UNWIND_STRICT (
- lookup,
- frame,
- op_ret,
- op_errno,
- inode,
- buf,
- xdata,
- postparent);
- return 0;
+ return ret;
}
int32_t
-pl_lookup (call_frame_t *frame,
- xlator_t *this,
- loc_t *loc,
- dict_t *xdata)
+pl_metalk(call_frame_t *frame, xlator_t *this, inode_t *inode)
{
- pl_local_t *local = NULL;
- int ret = -1;
+ pl_inode_t *pl_inode = NULL;
+ int ret = 0;
+ pl_meta_lock_t *reqlk = NULL;
+ pl_ctx_t *ctx = NULL;
+
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (!pl_inode) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, ENOMEM,
+ "pl_inode mem allocation failedd");
+
+ ret = -1;
+ goto out;
+ }
+
+ /* Non rebalance process trying to do metalock */
+ if (frame->root->pid != GF_CLIENT_PID_DEFRAG) {
+ ret = -1;
+ goto out;
+ }
+
+ /* Note: In the current scheme of glusterfs where lock migration is
+ * experimental, (ideally) the rebalance process which is migrating
+ * the file should request for a metalock. Hence, the metalock count
+ * should not be more than one for an inode. In future, if there is a
+ * need for meta-lock from other clients, the following block can be
+ * removed.
+ *
+ * Since pl_metalk is called as part of setxattr operation, any client
+ * process(non-rebalance) residing outside trusted network can exhaust
+ * memory of the server node by issuing setxattr repetitively on the
+ * metalock key. The following code makes sure that more than
+ * one metalock cannot be granted on an inode*/
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (pl_metalock_is_active(pl_inode)) {
+ ret = -1;
+ }
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ if (ret == -1) {
+ gf_msg(this->name, GF_LOG_WARNING, EINVAL, 0,
+ "More than one meta-lock cannot be granted on"
+ " the inode");
+ goto out;
+ }
+
+ if (frame->root->client) {
+ ctx = pl_ctx_get(frame->root->client, this);
+ if (!ctx) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_ctx_get failed");
+
+ ret = -1;
+ goto out;
+ }
+ } else {
+ gf_msg(this->name, GF_LOG_INFO, 0, 0,
+ "frame-root-client "
+ "is NULL");
+
+ ret = -1;
+ goto out;
+ }
+
+ reqlk = new_meta_lock(frame, this);
+ if (!reqlk) {
+ ret = -1;
+ goto out;
+ }
+
+ ret = pl_insert_metalk(pl_inode, ctx, reqlk);
+ if (ret < 0) {
+ pl_metalk_unref(reqlk);
+ }
- VALIDATE_OR_GOTO (frame, out);
- VALIDATE_OR_GOTO (this, out);
- VALIDATE_OR_GOTO (loc, out);
+out:
+ return ret;
+}
- local = mem_get0 (this->local_pool);
- GF_VALIDATE_OR_GOTO (this->name, local, out);
+static void
+__unwind_queued_locks(pl_inode_t *pl_inode, struct list_head *tmp_list)
+{
+ if (list_empty(&pl_inode->queued_locks))
+ return;
- if (xdata) {
- if (dict_get (xdata, GLUSTERFS_ENTRYLK_COUNT))
- local->entrylk_count_req = 1;
- if (dict_get (xdata, GLUSTERFS_INODELK_COUNT))
- local->inodelk_count_req = 1;
- if (dict_get (xdata, GLUSTERFS_INODELK_DOM_COUNT))
- local->inodelk_dom_count_req = 1;
- if (dict_get (xdata, GLUSTERFS_POSIXLK_COUNT))
- local->posixlk_count_req = 1;
- if (dict_get (xdata, GLUSTERFS_PARENT_ENTRYLK))
- local->parent_entrylk_req = 1;
- }
+ list_splice_init(&pl_inode->queued_locks, tmp_list);
+}
- frame->local = local;
- loc_copy (&local->loc, loc);
-
- STACK_WIND (frame,
- pl_lookup_cbk,
- FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->lookup,
- loc, xdata);
- ret = 0;
-out:
- if (ret == -1)
- STACK_UNWIND_STRICT (lookup, frame, -1, 0, NULL,
- NULL, NULL, NULL);
+static void
+__unwind_blocked_locks(pl_inode_t *pl_inode, struct list_head *tmp_list)
+{
+ posix_lock_t *lock = NULL;
+ posix_lock_t *tmp = NULL;
- return 0;
+ if (list_empty(&pl_inode->ext_list))
+ return;
+
+ list_for_each_entry_safe(lock, tmp, &pl_inode->ext_list, list)
+ {
+ if (!lock->blocking)
+ continue;
+
+ list_del_init(&lock->list);
+ list_add_tail(&lock->list, tmp_list);
+ }
}
+
int
-pl_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
- int op_ret, int op_errno, gf_dirent_t *entries, dict_t *xdata)
+pl_metaunlock(call_frame_t *frame, xlator_t *this, inode_t *inode, dict_t *dict)
{
- pl_local_t *local = NULL;
- gf_dirent_t *entry = NULL;
+ pl_inode_t *pl_inode = NULL;
+ int ret = 0;
+ pl_meta_lock_t *meta_lock = NULL;
+ pl_meta_lock_t *tmp_metalk = NULL;
+ pl_ctx_t *ctx = NULL;
+ posix_lock_t *posix_lock = NULL;
+ posix_lock_t *tmp_posixlk = NULL;
+ struct list_head tmp_posixlk_list;
+
+ INIT_LIST_HEAD(&tmp_posixlk_list);
+
+ if (frame->root->client) {
+ ctx = pl_ctx_get(frame->root->client, this);
+ if (!ctx) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_ctx_get failed");
+
+ ret = -1;
+ goto out;
+ }
+ } else {
+ gf_msg(this->name, GF_LOG_ERROR, 0, 0,
+ "frame-root-client is "
+ "NULL");
+ ret = -1;
+ goto out;
+ }
+
+ pl_inode = pl_inode_get(this, inode, NULL);
+ if (!pl_inode) {
+ ret = -1;
+ goto out;
+ }
+
+ pthread_mutex_lock(&ctx->lock);
+ {
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ /* Unwind queued locks regardless of migration status */
+ __unwind_queued_locks(pl_inode, &tmp_posixlk_list);
- local = frame->local;
+ /* Unwind blocked locks only for successful migration */
+ if (dict_get_sizen(dict, "status")) {
+ /* unwind all blocked locks */
+ __unwind_blocked_locks(pl_inode, &tmp_posixlk_list);
+ }
- if (op_ret <= 0)
- goto unwind;
+ /* unlock metalk */
+ /* if this list is empty then pl_inode->metalk_list
+ * should be empty too. meta lock should in all cases
+ * be added/removed from both pl_ctx_t and pl_inode */
+
+ if (list_empty(&ctx->metalk_list))
+ goto unlock;
- list_for_each_entry (entry, &entries->list, list) {
- if (local->entrylk_count_req)
- pl_entrylk_xattr_fill (this, entry->inode, entry->dict);
- if (local->inodelk_count_req)
- pl_inodelk_xattr_fill (this, entry->inode, entry->dict,
- _gf_false);
- if (local->inodelk_dom_count_req)
- pl_inodelk_xattr_fill (this, entry->inode, entry->dict,
- _gf_true);
- if (local->posixlk_count_req)
- pl_posixlk_xattr_fill (this, entry->inode, entry->dict);
+ list_for_each_entry_safe(meta_lock, tmp_metalk, &ctx->metalk_list,
+ client_list)
+ {
+ list_del_init(&meta_lock->client_list);
+
+ pl_inode = meta_lock->pl_inode;
+
+ list_del_init(&meta_lock->list);
+
+ pl_metalk_unref(meta_lock);
+
+ /* The corresponding ref is taken in
+ * pl_insert_metalk*/
+ inode_unref(pl_inode->inode);
+ }
+
+ if (dict_get_sizen(dict, "status"))
+ pl_inode->migrated = _gf_true;
+ else
+ pl_inode->migrated = _gf_false;
}
+ unlock:
-unwind:
- frame->local = NULL;
- STACK_UNWIND_STRICT (readdirp, frame, op_ret, op_errno, entries, xdata);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
+ pthread_mutex_unlock(&ctx->lock);
+
+out:
+ list_for_each_entry_safe(posix_lock, tmp_posixlk, &tmp_posixlk_list, list)
+ {
+ list_del_init(&posix_lock->list);
- if (local)
- mem_put (local);
+ STACK_UNWIND_STRICT(lk, posix_lock->frame, -1, EREMOTE,
+ &posix_lock->user_flock, NULL);
- return 0;
+ __destroy_lock(posix_lock);
+ }
+
+ return ret;
}
-int
-pl_readdirp (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
- off_t offset, dict_t *dict)
+int32_t
+pl_setxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *xdata)
{
- pl_local_t *local = NULL;
-
- local = mem_get0 (this->local_pool);
- GF_VALIDATE_OR_GOTO (this->name, local, out);
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+ local = frame->local;
+ if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+ pl_inode = pl_inode_get(this, local->inode, NULL);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
+ }
- if (dict) {
- if (dict_get (dict, GLUSTERFS_ENTRYLK_COUNT))
- local->entrylk_count_req = 1;
- if (dict_get (dict, GLUSTERFS_INODELK_COUNT))
- local->inodelk_count_req = 1;
- if (dict_get (dict, GLUSTERFS_INODELK_DOM_COUNT))
- local->inodelk_dom_count_req = 1;
- if (dict_get (dict, GLUSTERFS_POSIXLK_COUNT))
- local->posixlk_count_req = 1;
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ while (pl_inode->fop_wind_count > 0) {
+ gf_msg(this->name, GF_LOG_INFO, 0, 0,
+ "waiting for existing fops (count %d) to drain for "
+ "gfid %s",
+ pl_inode->fop_wind_count, uuid_utoa(pl_inode->gfid));
+ pthread_cond_wait(&pl_inode->check_fop_wind_count,
+ &pl_inode->mutex);
+ }
+ pl_inode->mlock_enforced = _gf_true;
+ pl_inode->check_mlock_info = _gf_false;
}
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- frame->local = local;
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata, frame, op_ret, op_errno, xdata);
+ return 0;
+}
- STACK_WIND (frame, pl_readdirp_cbk,
- FIRST_CHILD(this), FIRST_CHILD(this)->fops->readdirp,
- fd, size, offset, dict);
+int32_t
+pl_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *dict,
+ int flags, dict_t *xdata)
+{
+ int op_ret = 0;
+ int op_errno = EINVAL;
+ dict_t *xdata_rsp = NULL;
+ char *name = NULL;
+ posix_locks_private_t *priv = this->private;
- return 0;
-out:
- STACK_UNWIND_STRICT (readdirp, frame, -1, ENOMEM, NULL, NULL);
- return 0;
-}
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+
+ if (dict_get_sizen(dict, GF_META_LOCK_KEY)) {
+ op_ret = pl_metalk(frame, this, loc->inode);
+
+ } else if (dict_get_sizen(dict, GF_META_UNLOCK_KEY)) {
+ op_ret = pl_metaunlock(frame, this, loc->inode, dict);
+ } else {
+ goto usual;
+ }
+ PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata_rsp, frame, op_ret, op_errno,
+ xdata_rsp);
+ return 0;
+
+usual:
+ PL_CHECK_LOCK_ENFORCE_KEY(frame, dict, name, this, loc, ((fd_t *)NULL),
+ priv);
+
+ STACK_WIND(frame, pl_setxattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->setxattr, loc, dict, flags, xdata);
+ return 0;
+
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(setxattr, xdata, frame, op_ret, op_errno, xdata);
+
+ return 0;
+}
void
-pl_dump_lock (char *str, int size, struct gf_flock *flock,
- gf_lkowner_t *owner, void *trans, char *conn_id,
- time_t *granted_time, time_t *blkd_time, gf_boolean_t active)
+pl_dump_lock(char *str, int size, struct gf_flock *flock, gf_lkowner_t *owner,
+ void *trans, char *conn_id, time_t *granted_time,
+ time_t *blkd_time, gf_boolean_t active)
{
- char *type_str = NULL;
- char granted[32] = {0,};
- char blocked[32] = {0,};
-
- switch (flock->l_type) {
+ char *type_str = NULL;
+ char granted[GF_TIMESTR_SIZE] = {
+ 0,
+ };
+ char blocked[GF_TIMESTR_SIZE] = {
+ 0,
+ };
+
+ if (granted_time)
+ gf_time_fmt(granted, sizeof(granted), *granted_time, gf_timefmt_FT);
+ if (blkd_time)
+ gf_time_fmt(blocked, sizeof(blocked), *blkd_time, gf_timefmt_FT);
+ switch (flock->l_type) {
case F_RDLCK:
- type_str = "READ";
- break;
+ type_str = "READ";
+ break;
case F_WRLCK:
- type_str = "WRITE";
- break;
+ type_str = "WRITE";
+ break;
case F_UNLCK:
- type_str = "UNLOCK";
- break;
+ type_str = "UNLOCK";
+ break;
default:
- type_str = "UNKNOWN";
- break;
+ type_str = "UNKNOWN";
+ break;
+ }
+
+ if (active) {
+ if (blkd_time && *blkd_time == 0) {
+ snprintf(str, size, RANGE_GRNTD_FMT, type_str, flock->l_whence,
+ (unsigned long long)flock->l_start,
+ (unsigned long long)flock->l_len,
+ (unsigned long long)flock->l_pid, lkowner_utoa(owner),
+ trans, conn_id, granted);
+ } else {
+ snprintf(str, size, RANGE_BLKD_GRNTD_FMT, type_str, flock->l_whence,
+ (unsigned long long)flock->l_start,
+ (unsigned long long)flock->l_len,
+ (unsigned long long)flock->l_pid, lkowner_utoa(owner),
+ trans, conn_id, blocked, granted);
}
+ } else {
+ snprintf(str, size, RANGE_BLKD_FMT, type_str, flock->l_whence,
+ (unsigned long long)flock->l_start,
+ (unsigned long long)flock->l_len,
+ (unsigned long long)flock->l_pid, lkowner_utoa(owner), trans,
+ conn_id, blocked);
+ }
+}
- if (active) {
- if (blkd_time && *blkd_time == 0) {
- snprintf (str, size, RANGE_GRNTD_FMT,
- type_str, flock->l_whence,
- (unsigned long long) flock->l_start,
- (unsigned long long) flock->l_len,
- (unsigned long long) flock->l_pid,
- lkowner_utoa (owner), trans, conn_id,
- ctime_r (granted_time, granted));
- } else {
- snprintf (str, size, RANGE_BLKD_GRNTD_FMT,
- type_str, flock->l_whence,
- (unsigned long long) flock->l_start,
- (unsigned long long) flock->l_len,
- (unsigned long long) flock->l_pid,
- lkowner_utoa (owner), trans, conn_id,
- ctime_r (blkd_time, blocked),
- ctime_r (granted_time, granted));
- }
- }
- else {
- snprintf (str, size, RANGE_BLKD_FMT,
- type_str, flock->l_whence,
- (unsigned long long) flock->l_start,
- (unsigned long long) flock->l_len,
- (unsigned long long) flock->l_pid,
- lkowner_utoa (owner), trans, conn_id,
- ctime_r (blkd_time, blocked));
+void
+__dump_entrylks(pl_inode_t *pl_inode)
+{
+ pl_dom_list_t *dom = NULL;
+ pl_entry_lock_t *lock = NULL;
+ char blocked[GF_TIMESTR_SIZE] = {
+ 0,
+ };
+ char granted[GF_TIMESTR_SIZE] = {
+ 0,
+ };
+ int count = 0;
+ char key[GF_DUMP_MAX_BUF_LEN] = {
+ 0,
+ };
+ char *k = "xlator.feature.locks.lock-dump.domain.entrylk";
+
+ char tmp[4098];
+
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ count = 0;
+
+ gf_proc_dump_build_key(key, "lock-dump.domain", "domain");
+ gf_proc_dump_write(key, "%s", dom->domain);
+
+ list_for_each_entry(lock, &dom->entrylk_list, domain_list)
+ {
+ gf_time_fmt(granted, sizeof(granted), lock->granted_time,
+ gf_timefmt_FT);
+ gf_proc_dump_build_key(key, k, "entrylk[%d](ACTIVE)", count);
+ if (lock->blkd_time == 0) {
+ snprintf(tmp, sizeof(tmp), ENTRY_GRNTD_FMT,
+ lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK"
+ : "ENTRYLK_WRLCK",
+ lock->basename, (unsigned long long)lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->client,
+ lock->connection_id, granted);
+ } else {
+ gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time,
+ gf_timefmt_FT);
+ snprintf(tmp, sizeof(tmp), ENTRY_BLKD_GRNTD_FMT,
+ lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK"
+ : "ENTRYLK_WRLCK",
+ lock->basename, (unsigned long long)lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->client,
+ lock->connection_id, blocked, granted);
+ }
+
+ gf_proc_dump_write(key, "%s", tmp);
+
+ count++;
}
+ list_for_each_entry(lock, &dom->blocked_entrylks, blocked_locks)
+ {
+ gf_time_fmt(blocked, sizeof(blocked), lock->blkd_time,
+ gf_timefmt_FT);
+
+ gf_proc_dump_build_key(key, k, "entrylk[%d](BLOCKED)", count);
+ snprintf(
+ tmp, sizeof(tmp), ENTRY_BLKD_FMT,
+ lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" : "ENTRYLK_WRLCK",
+ lock->basename, (unsigned long long)lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->client, lock->connection_id,
+ blocked);
+
+ gf_proc_dump_write(key, "%s", tmp);
+
+ count++;
+ }
+ }
}
void
-__dump_entrylks (pl_inode_t *pl_inode)
-{
- pl_dom_list_t *dom = NULL;
- pl_entry_lock_t *lock = NULL;
- char blocked[32] = {0,};
- char granted[32] = {0,};
- int count = 0;
- char key[GF_DUMP_MAX_BUF_LEN] = {0,};
-
- char tmp[256];
-
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
-
- count = 0;
-
- gf_proc_dump_build_key(key,
- "lock-dump.domain",
- "domain");
- gf_proc_dump_write(key, "%s", dom->domain);
-
- list_for_each_entry (lock, &dom->entrylk_list, domain_list) {
-
- gf_proc_dump_build_key(key,
- "xlator.feature.locks.lock-dump.domain.entrylk",
- "entrylk[%d](ACTIVE)", count );
- if (lock->blkd_time.tv_sec == 0 && lock->blkd_time.tv_usec == 0) {
- snprintf (tmp, 256, ENTRY_GRNTD_FMT,
- lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" :
- "ENTRYLK_WRLCK", lock->basename,
- (unsigned long long) lock->client_pid,
- lkowner_utoa (&lock->owner), lock->client,
- lock->connection_id,
- ctime_r (&lock->granted_time.tv_sec, granted));
- } else {
- snprintf (tmp, 256, ENTRY_BLKD_GRNTD_FMT,
- lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" :
- "ENTRYLK_WRLCK", lock->basename,
- (unsigned long long) lock->client_pid,
- lkowner_utoa (&lock->owner), lock->client,
- lock->connection_id,
- ctime_r (&lock->blkd_time.tv_sec, blocked),
- ctime_r (&lock->granted_time.tv_sec, granted));
- }
+dump_entrylks(pl_inode_t *pl_inode)
+{
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __dump_entrylks(pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+}
- gf_proc_dump_write(key, tmp);
+void
+__dump_inodelks(pl_inode_t *pl_inode)
+{
+ pl_dom_list_t *dom = NULL;
+ pl_inode_lock_t *lock = NULL;
+ int count = 0;
+ char key[GF_DUMP_MAX_BUF_LEN];
- count++;
- }
+ char tmp[4098];
- list_for_each_entry (lock, &dom->blocked_entrylks, blocked_locks) {
+ list_for_each_entry(dom, &pl_inode->dom_list, inode_list)
+ {
+ count = 0;
- gf_proc_dump_build_key(key,
- "xlator.feature.locks.lock-dump.domain.entrylk",
- "entrylk[%d](BLOCKED)", count );
- snprintf (tmp, 256, ENTRY_BLKD_FMT,
- lock->type == ENTRYLK_RDLCK ? "ENTRYLK_RDLCK" :
- "ENTRYLK_WRLCK", lock->basename,
- (unsigned long long) lock->client_pid,
- lkowner_utoa (&lock->owner), lock->client,
- lock->connection_id,
- ctime_r (&lock->blkd_time.tv_sec, blocked));
+ gf_proc_dump_build_key(key, "lock-dump.domain", "domain");
+ gf_proc_dump_write(key, "%s", dom->domain);
- gf_proc_dump_write(key, tmp);
+ list_for_each_entry(lock, &dom->inodelk_list, list)
+ {
+ gf_proc_dump_build_key(key, "inodelk", "inodelk[%d](ACTIVE)",
+ count);
- count++;
- }
+ SET_FLOCK_PID(&lock->user_flock, lock);
+ pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+ lock->client, lock->connection_id, &lock->granted_time,
+ &lock->blkd_time, _gf_true);
+ gf_proc_dump_write(key, "%s", tmp);
+ count++;
}
+ list_for_each_entry(lock, &dom->blocked_inodelks, blocked_locks)
+ {
+ gf_proc_dump_build_key(key, "inodelk", "inodelk[%d](BLOCKED)",
+ count);
+ SET_FLOCK_PID(&lock->user_flock, lock);
+ pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+ lock->client, lock->connection_id, 0, &lock->blkd_time,
+ _gf_false);
+ gf_proc_dump_write(key, "%s", tmp);
+
+ count++;
+ }
+ }
}
void
-dump_entrylks (pl_inode_t *pl_inode)
+dump_inodelks(pl_inode_t *pl_inode)
{
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __dump_entrylks (pl_inode);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __dump_inodelks(pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+}
+void
+__dump_posixlks(pl_inode_t *pl_inode)
+{
+ posix_lock_t *lock = NULL;
+ int count = 0;
+ char key[GF_DUMP_MAX_BUF_LEN];
+
+ char tmp[4098];
+
+ list_for_each_entry(lock, &pl_inode->ext_list, list)
+ {
+ SET_FLOCK_PID(&lock->user_flock, lock);
+ gf_proc_dump_build_key(key, "posixlk", "posixlk[%d](%s)", count,
+ lock->blocked ? "BLOCKED" : "ACTIVE");
+ pl_dump_lock(tmp, sizeof(tmp), &lock->user_flock, &lock->owner,
+ lock->client, lock->client_uid, &lock->granted_time,
+ &lock->blkd_time, (lock->blocked) ? _gf_false : _gf_true);
+ gf_proc_dump_write(key, "%s", tmp);
+
+ count++;
+ }
}
void
-__dump_inodelks (pl_inode_t *pl_inode)
+dump_posixlks(pl_inode_t *pl_inode)
{
- pl_dom_list_t *dom = NULL;
- pl_inode_lock_t *lock = NULL;
- int count = 0;
- char key[GF_DUMP_MAX_BUF_LEN];
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __dump_posixlks(pl_inode);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+}
- char tmp[256];
+int32_t
+pl_dump_inode_priv(xlator_t *this, inode_t *inode)
+{
+ int ret = -1;
+ uint64_t tmp_pl_inode = 0;
+ pl_inode_t *pl_inode = NULL;
+ char *pathname = NULL;
+ gf_boolean_t section_added = _gf_false;
+
+ int count = 0;
+
+ if (!inode) {
+ errno = EINVAL;
+ goto out;
+ }
+
+ ret = TRY_LOCK(&inode->lock);
+ if (ret)
+ goto out;
+ {
+ ret = __inode_ctx_get(inode, this, &tmp_pl_inode);
+ if (ret)
+ goto unlock;
+ }
+unlock:
+ UNLOCK(&inode->lock);
+ if (ret)
+ goto out;
+
+ pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
+ if (!pl_inode) {
+ ret = -1;
+ goto out;
+ }
+
+ gf_proc_dump_add_section("xlator.features.locks.%s.inode", this->name);
+ section_added = _gf_true;
+
+ /*We are safe to call __inode_path since we have the
+ * inode->table->lock */
+ __inode_path(inode, NULL, &pathname);
+ if (pathname)
+ gf_proc_dump_write("path", "%s", pathname);
+
+ gf_proc_dump_write("mandatory", "%d", pl_inode->mandatory);
+
+ ret = pthread_mutex_trylock(&pl_inode->mutex);
+ if (ret)
+ goto out;
+ {
+ count = __get_entrylk_count(this, pl_inode);
+ if (count) {
+ gf_proc_dump_write("entrylk-count", "%d", count);
+ __dump_entrylks(pl_inode);
+ }
- list_for_each_entry (dom, &pl_inode->dom_list, inode_list) {
+ count = __get_inodelk_count(this, pl_inode, NULL);
+ if (count) {
+ gf_proc_dump_write("inodelk-count", "%d", count);
+ __dump_inodelks(pl_inode);
+ }
- count = 0;
+ count = __get_posixlk_count(pl_inode);
+ if (count) {
+ gf_proc_dump_write("posixlk-count", "%d", count);
+ __dump_posixlks(pl_inode);
+ }
- gf_proc_dump_build_key(key,
- "lock-dump.domain",
- "domain");
- gf_proc_dump_write(key, "%s", dom->domain);
+ gf_proc_dump_write("links", "%d", pl_inode->links);
+ gf_proc_dump_write("removes_pending", "%u", pl_inode->remove_running);
+ gf_proc_dump_write("removed", "%u", pl_inode->removed);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
- list_for_each_entry (lock, &dom->inodelk_list, list) {
+out:
+ GF_FREE(pathname);
+
+ if (ret && inode) {
+ if (!section_added)
+ gf_proc_dump_add_section(
+ "xlator.features.locks.%s."
+ "inode",
+ this->name);
+ gf_proc_dump_write("Unable to print lock state",
+ "(Lock "
+ "acquisition failure) %s",
+ uuid_utoa(inode->gfid));
+ }
+ return ret;
+}
- gf_proc_dump_build_key(key,
- "inodelk",
- "inodelk[%d](ACTIVE)",count );
+int32_t
+mem_acct_init(xlator_t *this)
+{
+ int ret = -1;
- SET_FLOCK_PID (&lock->user_flock, lock);
- pl_dump_lock (tmp, 256, &lock->user_flock,
- &lock->owner,
- lock->client, lock->connection_id,
- &lock->granted_time.tv_sec,
- &lock->blkd_time.tv_sec,
- _gf_true);
- gf_proc_dump_write(key, tmp);
+ if (!this)
+ return ret;
- count++;
- }
+ ret = xlator_mem_acct_init(this, gf_locks_mt_end + 1);
- list_for_each_entry (lock, &dom->blocked_inodelks, blocked_locks) {
+ if (ret != 0) {
+ gf_log(this->name, GF_LOG_ERROR,
+ "Memory accounting init"
+ "failed");
+ return ret;
+ }
- gf_proc_dump_build_key(key,
- "inodelk",
- "inodelk[%d](BLOCKED)",count );
- SET_FLOCK_PID (&lock->user_flock, lock);
- pl_dump_lock (tmp, 256, &lock->user_flock,
- &lock->owner,
- lock->client, lock->connection_id,
- 0, &lock->blkd_time.tv_sec,
- _gf_false);
- gf_proc_dump_write(key, tmp);
+ return ret;
+}
- count++;
- }
+pl_ctx_t *
+pl_ctx_get(client_t *client, xlator_t *xlator)
+{
+ void *tmp = NULL;
+ pl_ctx_t *ctx = NULL;
+ pl_ctx_t *setted_ctx = NULL;
- }
+ client_ctx_get(client, xlator, &tmp);
+
+ ctx = tmp;
+
+ if (ctx != NULL)
+ goto out;
+
+ ctx = GF_CALLOC(1, sizeof(pl_ctx_t), gf_locks_mt_posix_lock_t);
+
+ if (ctx == NULL)
+ goto out;
+ pthread_mutex_init(&ctx->lock, NULL);
+ INIT_LIST_HEAD(&ctx->inodelk_lockers);
+ INIT_LIST_HEAD(&ctx->entrylk_lockers);
+ INIT_LIST_HEAD(&ctx->metalk_list);
+
+ setted_ctx = client_ctx_set(client, xlator, ctx);
+ if (ctx != setted_ctx) {
+ pthread_mutex_destroy(&ctx->lock);
+ GF_FREE(ctx);
+ ctx = setted_ctx;
+ }
+out:
+ return ctx;
}
-void
-dump_inodelks (pl_inode_t *pl_inode)
+int
+pl_metalk_client_cleanup(xlator_t *this, pl_ctx_t *ctx)
{
- pthread_mutex_lock (&pl_inode->mutex);
+ pl_meta_lock_t *meta_lock = NULL;
+ pl_meta_lock_t *tmp_metalk = NULL;
+ pl_inode_t *pl_inode = NULL;
+ posix_lock_t *posix_lock = NULL;
+ posix_lock_t *tmp_posixlk = NULL;
+ struct list_head tmp_posixlk_list;
+
+ INIT_LIST_HEAD(&tmp_posixlk_list);
+
+ pthread_mutex_lock(&ctx->lock);
+ {
+ /* if this list is empty then pl_inode->metalk_list should be
+ * empty too. meta lock should in all cases be added/removed
+ * from both pl_ctx_t and pl_inode */
+ if (list_empty(&ctx->metalk_list))
+ goto unlock;
+
+ list_for_each_entry_safe(meta_lock, tmp_metalk, &ctx->metalk_list,
+ client_list)
{
- __dump_inodelks (pl_inode);
+ list_del_init(&meta_lock->client_list);
+
+ pl_inode = meta_lock->pl_inode;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+
+ {
+ /* Since the migration status is unknown here
+ * unwind all queued and blocked locks to check
+ * migration status and find the correct
+ * destination */
+ __unwind_queued_locks(pl_inode, &tmp_posixlk_list);
+
+ __unwind_blocked_locks(pl_inode, &tmp_posixlk_list);
+
+ list_del_init(&meta_lock->list);
+
+ pl_metalk_unref(meta_lock);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ /* The corresponding ref is taken in
+ * pl_insert_metalk*/
+ inode_unref(pl_inode->inode);
}
- pthread_mutex_unlock (&pl_inode->mutex);
+ }
+
+unlock:
+ pthread_mutex_unlock(&ctx->lock);
+ list_for_each_entry_safe(posix_lock, tmp_posixlk, &tmp_posixlk_list, list)
+ {
+ list_del_init(&posix_lock->list);
+
+ STACK_UNWIND_STRICT(lk, posix_lock->frame, -1, EREMOTE,
+ &posix_lock->user_flock, NULL);
+
+ __destroy_lock(posix_lock);
+ }
+ return 0;
}
-void
-__dump_posixlks (pl_inode_t *pl_inode)
+static int
+pl_client_disconnect_cbk(xlator_t *this, client_t *client)
{
- posix_lock_t *lock = NULL;
- int count = 0;
- char key[GF_DUMP_MAX_BUF_LEN];
+ pl_ctx_t *pl_ctx = pl_ctx_get(client, this);
+ if (pl_ctx) {
+ pl_inodelk_client_cleanup(this, pl_ctx);
+ pl_entrylk_client_cleanup(this, pl_ctx);
+ pl_metalk_client_cleanup(this, pl_ctx);
+ }
+
+ return 0;
+}
- char tmp[256];
+static int
+pl_client_destroy_cbk(xlator_t *this, client_t *client)
+{
+ void *tmp = NULL;
+ pl_ctx_t *pl_ctx = NULL;
- list_for_each_entry (lock, &pl_inode->ext_list, list) {
+ pl_client_disconnect_cbk(this, client);
- SET_FLOCK_PID (&lock->user_flock, lock);
- gf_proc_dump_build_key(key,
- "posixlk",
- "posixlk[%d](%s)",
- count,
- lock->blocked ? "BLOCKED" : "ACTIVE");
- pl_dump_lock (tmp, 256, &lock->user_flock,
- &lock->owner, lock->client, NULL,
- &lock->granted_time.tv_sec, &lock->blkd_time.tv_sec,
- (lock->blocked)? _gf_false: _gf_true);
- gf_proc_dump_write(key, tmp);
+ client_ctx_del(client, this, &tmp);
- count++;
- }
+ if (tmp == NULL)
+ return 0;
+
+ pl_ctx = tmp;
+
+ GF_ASSERT(list_empty(&pl_ctx->inodelk_lockers));
+ GF_ASSERT(list_empty(&pl_ctx->entrylk_lockers));
+
+ pthread_mutex_destroy(&pl_ctx->lock);
+ GF_FREE(pl_ctx);
+
+ return 0;
+}
+
+int
+reconfigure(xlator_t *this, dict_t *options)
+{
+ posix_locks_private_t *priv = this->private;
+ int ret = -1;
+ char *tmp_str = NULL;
+
+ GF_OPTION_RECONF("trace", priv->trace, options, bool, out);
+
+ GF_OPTION_RECONF("monkey-unlocking", priv->monkey_unlocking, options, bool,
+ out);
+
+ GF_OPTION_RECONF("revocation-secs", priv->revocation_secs, options, uint32,
+ out);
+
+ GF_OPTION_RECONF("revocation-clear-all", priv->revocation_clear_all,
+ options, bool, out);
+
+ GF_OPTION_RECONF("revocation-max-blocked", priv->revocation_max_blocked,
+ options, uint32, out);
+
+ GF_OPTION_RECONF("notify-contention", priv->notify_contention, options,
+ bool, out);
+
+ GF_OPTION_RECONF("notify-contention-delay", priv->notify_contention_delay,
+ options, uint32, out);
+
+ GF_OPTION_RECONF("mandatory-locking", tmp_str, options, str, out);
+
+ GF_OPTION_RECONF("enforce-mandatory-lock", priv->mlock_enforced, options,
+ bool, out);
+
+ if (!strcmp(tmp_str, "forced"))
+ priv->mandatory_mode = MLK_FORCED;
+ else if (!strcmp(tmp_str, "file"))
+ priv->mandatory_mode = MLK_FILE_BASED;
+ else if (!strcmp(tmp_str, "optimal"))
+ priv->mandatory_mode = MLK_OPTIMAL;
+ else
+ priv->mandatory_mode = MLK_NONE;
+
+ ret = 0;
+
+out:
+ return ret;
+}
+
+int
+init(xlator_t *this)
+{
+ posix_locks_private_t *priv = NULL;
+ xlator_list_t *trav = NULL;
+ char *tmp_str = NULL;
+ int ret = -1;
+
+ if (!this->children || this->children->next) {
+ gf_log(this->name, GF_LOG_CRITICAL,
+ "FATAL: posix-locks should have exactly one child");
+ goto out;
+ }
+
+ if (!this->parents) {
+ gf_log(this->name, GF_LOG_WARNING,
+ "Volume is dangling. Please check the volume file.");
+ }
+
+ trav = this->children;
+ while (trav->xlator->children)
+ trav = trav->xlator->children;
+
+ if (strncmp("storage/", trav->xlator->type, 8)) {
+ gf_log(this->name, GF_LOG_CRITICAL,
+ "'locks' translator is not loaded over a storage "
+ "translator");
+ goto out;
+ }
+
+ priv = GF_CALLOC(1, sizeof(*priv), gf_locks_mt_posix_locks_private_t);
+
+ GF_OPTION_INIT("mandatory-locking", tmp_str, str, out);
+ if (!strcmp(tmp_str, "forced"))
+ priv->mandatory_mode = MLK_FORCED;
+ else if (!strcmp(tmp_str, "file"))
+ priv->mandatory_mode = MLK_FILE_BASED;
+ else if (!strcmp(tmp_str, "optimal"))
+ priv->mandatory_mode = MLK_OPTIMAL;
+ else
+ priv->mandatory_mode = MLK_NONE;
+
+ tmp_str = NULL;
+
+ GF_OPTION_INIT("trace", priv->trace, bool, out);
+
+ GF_OPTION_INIT("monkey-unlocking", priv->monkey_unlocking, bool, out);
+
+ GF_OPTION_INIT("revocation-secs", priv->revocation_secs, uint32, out);
+
+ GF_OPTION_INIT("revocation-clear-all", priv->revocation_clear_all, bool,
+ out);
+
+ GF_OPTION_INIT("revocation-max-blocked", priv->revocation_max_blocked,
+ uint32, out);
+
+ GF_OPTION_INIT("notify-contention", priv->notify_contention, bool, out);
+
+ GF_OPTION_INIT("notify-contention-delay", priv->notify_contention_delay,
+ uint32, out);
+
+ GF_OPTION_INIT("enforce-mandatory-lock", priv->mlock_enforced, bool, out);
+
+ this->local_pool = mem_pool_new(pl_local_t, 32);
+ if (!this->local_pool) {
+ ret = -1;
+ gf_log(this->name, GF_LOG_ERROR,
+ "failed to create local_t's memory pool");
+ goto out;
+ }
+
+ this->private = priv;
+ ret = 0;
+
+out:
+ if (ret) {
+ GF_FREE(priv);
+ }
+ return ret;
}
void
-dump_posixlks (pl_inode_t *pl_inode)
+fini(xlator_t *this)
{
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __dump_posixlks (pl_inode);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ posix_locks_private_t *priv = this->private;
+ if (!priv)
+ return;
+ this->private = NULL;
+ if (this->local_pool) {
+ mem_pool_destroy(this->local_pool);
+ this->local_pool = NULL;
+ }
+ GF_FREE(priv->brickname);
+ GF_FREE(priv);
+
+ return;
+}
+
+int
+pl_inodelk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+ int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int
+pl_finodelk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+ int32_t cmd, struct gf_flock *flock, dict_t *xdata);
+
+int
+pl_entrylk(call_frame_t *frame, xlator_t *this, const char *volume, loc_t *loc,
+ const char *basename, entrylk_cmd cmd, entrylk_type type,
+ dict_t *xdata);
+int
+pl_fentrylk(call_frame_t *frame, xlator_t *this, const char *volume, fd_t *fd,
+ const char *basename, entrylk_cmd cmd, entrylk_type type,
+ dict_t *xdata);
+
+int32_t
+pl_rename_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *buf, struct iatt *preoldparent,
+ struct iatt *postoldparent, struct iatt *prenewparent,
+ struct iatt *postnewparent, dict_t *xdata)
+{
+ pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
+
+ PL_STACK_UNWIND(rename, xdata, frame, op_ret, op_errno, buf, preoldparent,
+ postoldparent, prenewparent, postnewparent, xdata);
+
+ return 0;
}
int32_t
-pl_dump_inode_priv (xlator_t *this, inode_t *inode)
+pl_rename(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+ dict_t *xdata)
{
+ int32_t error;
- int ret = -1;
- uint64_t tmp_pl_inode = 0;
- pl_inode_t *pl_inode = NULL;
- char *pathname = NULL;
- gf_boolean_t section_added = _gf_false;
+ error = PL_INODE_REMOVE(rename, frame, this, oldloc, newloc, pl_rename,
+ pl_rename_cbk, oldloc, newloc, xdata);
+ if (error > 0) {
+ STACK_UNWIND_STRICT(rename, frame, -1, error, NULL, NULL, NULL, NULL,
+ NULL, NULL);
+ }
- int count = 0;
+ return 0;
+}
- if (!inode) {
- errno = EINVAL;
- goto out;
+posix_lock_t *
+gf_lkmig_info_to_posix_lock(call_frame_t *frame, lock_migration_info_t *lmi)
+{
+ posix_lock_t *lock = GF_CALLOC(1, sizeof(posix_lock_t),
+ gf_locks_mt_posix_lock_t);
+ if (!lock)
+ goto out;
+
+ lock->fl_start = lmi->flock.l_start;
+ lock->fl_type = lmi->flock.l_type;
+
+ if (lmi->flock.l_len == 0)
+ lock->fl_end = LLONG_MAX;
+ else
+ lock->fl_end = lmi->flock.l_start + lmi->flock.l_len - 1;
+
+ lock->client = frame->root->client;
+
+ lock->lk_flags = lmi->lk_flags;
+
+ lock->client_uid = gf_strdup(lmi->client_uid);
+ if (lock->client_uid == NULL) {
+ GF_FREE(lock);
+ lock = NULL;
+ goto out;
+ }
+
+ lock->client_pid = lmi->flock.l_pid;
+ lock->owner = lmi->flock.l_owner;
+
+ INIT_LIST_HEAD(&lock->list);
+
+out:
+ return lock;
+}
+
+/* This function is supposed to write the active locks from the source brick(in
+ * rebalance context) and write here. Hence, will add the locks directly to the
+ * pl_inode->ext_list*/
+int
+pl_write_active_locks(call_frame_t *frame, pl_inode_t *pl_inode,
+ lock_migration_info_t *locklist)
+{
+ posix_lock_t *newlock = NULL;
+ lock_migration_info_t *temp = NULL;
+ int ret = 0;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ /* Just making sure the activelk list is empty. Should not
+ * happen though*/
+ if (!list_empty(&pl_inode->ext_list)) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "invalid locks found");
+
+ ret = -1;
+ goto out;
}
- ret = TRY_LOCK (&inode->lock);
- if (ret)
- goto out;
- {
- ret = __inode_ctx_get (inode, this, &tmp_pl_inode);
- if (ret)
- goto unlock;
+ /* This list also should not be empty */
+ if (list_empty(&locklist->list)) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_msg(THIS->name, GF_LOG_ERROR, 0, 0, "empty lock list");
+
+ ret = -1;
+ goto out;
}
-unlock:
- UNLOCK (&inode->lock);
- if (ret)
- goto out;
- pl_inode = (pl_inode_t *)(long)tmp_pl_inode;
- if (!pl_inode) {
+ list_for_each_entry(temp, &locklist->list, list)
+ {
+ newlock = gf_lkmig_info_to_posix_lock(frame, temp);
+ if (!newlock) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_msg(THIS->name, GF_LOG_ERROR, 0, 0,
+ "mem allocation failed for newlock");
+
ret = -1;
goto out;
+ }
+ list_add_tail(&newlock->list, &pl_inode->ext_list);
}
+ }
+ /*TODO: What if few lock add failed with ENOMEM. Should the already
+ * added locks be clearted */
+ pthread_mutex_unlock(&pl_inode->mutex);
+out:
+ return ret;
+}
- gf_proc_dump_add_section("xlator.features.locks.%s.inode", this->name);
- section_added = _gf_true;
+static int
+pl_setactivelk(call_frame_t *frame, xlator_t *this, loc_t *loc,
+ lock_migration_info_t *locklist, dict_t *xdata)
+{
+ int op_ret = 0;
+ int op_errno = 0;
+ int ret = 0;
- /*We are safe to call __inode_path since we have the
- * inode->table->lock */
- __inode_path (inode, NULL, &pathname);
- if (pathname)
- gf_proc_dump_write ("path", "%s", pathname);
+ pl_inode_t *pl_inode = pl_inode_get(this, loc->inode, NULL);
+ if (!pl_inode) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, 0, "pl_inode_get failed");
- gf_proc_dump_write("mandatory", "%d", pl_inode->mandatory);
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto out;
+ }
+ ret = pl_write_active_locks(frame, pl_inode, locklist);
- ret = pthread_mutex_trylock (&pl_inode->mutex);
- if (ret)
- goto out;
- {
- count = __get_entrylk_count (this, pl_inode);
- if (count) {
- gf_proc_dump_write("entrylk-count", "%d", count);
- __dump_entrylks (pl_inode);
- }
+ op_ret = ret;
- count = __get_inodelk_count (this, pl_inode, NULL);
- if (count) {
- gf_proc_dump_write("inodelk-count", "%d", count);
- __dump_inodelks (pl_inode);
- }
+out:
+ STACK_UNWIND_STRICT(setactivelk, frame, op_ret, op_errno, NULL);
- count = __get_posixlk_count (this, pl_inode);
- if (count) {
- gf_proc_dump_write("posixlk-count", "%d", count);
- __dump_posixlks (pl_inode);
- }
- }
- pthread_mutex_unlock (&pl_inode->mutex);
+ return 0;
+}
-out:
- GF_FREE (pathname);
+int32_t
+pl_unlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+ dict_t *xdata)
+{
+ pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
- if (ret && inode) {
- if (!section_added)
- gf_proc_dump_add_section ("xlator.features.locks.%s."
- "inode", this->name);
- gf_proc_dump_write ("Unable to print lock state", "(Lock "
- "acquisition failure) %s",
- uuid_utoa (inode->gfid));
- }
- return ret;
+ PL_STACK_UNWIND(unlink, xdata, frame, op_ret, op_errno, preparent,
+ postparent, xdata);
+
+ return 0;
}
int32_t
-mem_acct_init (xlator_t *this)
+pl_unlink(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflag,
+ dict_t *xdata)
{
- int ret = -1;
+ int32_t error;
- if (!this)
- return ret;
+ error = PL_INODE_REMOVE(unlink, frame, this, loc, NULL, pl_unlink,
+ pl_unlink_cbk, loc, xflag, xdata);
+ if (error > 0) {
+ STACK_UNWIND_STRICT(unlink, frame, -1, error, NULL, NULL, NULL);
+ }
- ret = xlator_mem_acct_init (this, gf_locks_mt_end + 1);
+ return 0;
+}
- if (ret != 0) {
- gf_log (this->name, GF_LOG_ERROR, "Memory accounting init"
- "failed");
- return ret;
- }
+int32_t
+pl_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, inode_t *inode, struct iatt *buf,
+ struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(mkdir, xdata, frame, op_ret, op_errno, inode,
+ buf, preparent, postparent, xdata);
+ return 0;
+}
- return ret;
+int
+pl_mkdir(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+ mode_t umask, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_mkdir_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->mkdir, loc, mode, umask, xdata);
+ return 0;
}
+int32_t
+pl_stat_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *buf, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(stat, xdata, frame, op_ret, op_errno, buf,
+ xdata);
+ return 0;
+}
-pl_ctx_t*
-pl_ctx_get (client_t *client, xlator_t *xlator)
+int
+pl_stat(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
{
- void *tmp = NULL;
- pl_ctx_t *ctx = NULL;
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_stat_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->stat, loc, xdata);
+ return 0;
+}
- client_ctx_get (client, xlator, &tmp);
+int32_t
+pl_mknod_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, inode_t *inode, struct iatt *buf,
+ struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(mknod, xdata, frame, op_ret, op_errno, inode,
+ buf, preparent, postparent, xdata);
+ return 0;
+}
- ctx = tmp;
+int
+pl_mknod(call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
+ dev_t rdev, mode_t umask, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_mknod_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->mknod, loc, mode, rdev, umask, xdata);
+ return 0;
+}
- if (ctx != NULL)
- goto out;
+int32_t
+pl_rmdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *preparent, struct iatt *postparent,
+ dict_t *xdata)
+{
+ pl_inode_remove_cbk(this, cookie, op_ret < 0 ? op_errno : 0);
- ctx = GF_CALLOC (1, sizeof (pl_ctx_t), gf_locks_mt_posix_lock_t);
+ PL_STACK_UNWIND_FOR_CLIENT(rmdir, xdata, frame, op_ret, op_errno, preparent,
+ postparent, xdata);
- if (ctx == NULL)
- goto out;
+ return 0;
+}
- pthread_mutex_init (&ctx->lock, NULL);
- INIT_LIST_HEAD (&ctx->inodelk_lockers);
- INIT_LIST_HEAD (&ctx->entrylk_lockers);
+int
+pl_rmdir(call_frame_t *frame, xlator_t *this, loc_t *loc, int xflags,
+ dict_t *xdata)
+{
+ int32_t error;
- if (client_ctx_set (client, xlator, ctx) != 0) {
- pthread_mutex_destroy (&ctx->lock);
- GF_FREE (ctx);
- ctx = NULL;
- }
-out:
- return ctx;
+ error = PL_INODE_REMOVE(rmdir, frame, this, loc, NULL, pl_rmdir,
+ pl_rmdir_cbk, loc, xflags, xdata);
+ if (error > 0) {
+ STACK_UNWIND_STRICT(rmdir, frame, -1, error, NULL, NULL, NULL);
+ }
+
+ return 0;
+}
+
+int32_t
+pl_symlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, inode_t *inode,
+ struct iatt *buf, struct iatt *preparent,
+ struct iatt *postparent, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(symlink, xdata, frame, op_ret, op_errno, inode,
+ buf, preparent, postparent, xdata);
+ return 0;
}
+int
+pl_symlink(call_frame_t *frame, xlator_t *this, const char *linkname,
+ loc_t *loc, mode_t umask, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_symlink_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->symlink, linkname, loc, umask, xdata);
+ return 0;
+}
-static int
-pl_client_disconnect_cbk (xlator_t *this, client_t *client)
+int32_t
+pl_link_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, inode_t *inode, struct iatt *buf,
+ struct iatt *preparent, struct iatt *postparent, dict_t *xdata)
{
- pl_ctx_t *pl_ctx = NULL;
+ pl_inode_t *pl_inode = (pl_inode_t *)cookie;
- pl_ctx = pl_ctx_get (client, this);
+ if (op_ret >= 0) {
+ pthread_mutex_lock(&pl_inode->mutex);
- pl_inodelk_client_cleanup (this, pl_ctx);
+ /* TODO: can happen pl_inode->links == 0 ? */
+ if (pl_inode->links >= 0) {
+ pl_inode->links++;
+ }
- pl_entrylk_client_cleanup (this, pl_ctx);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- return 0;
+ PL_STACK_UNWIND_FOR_CLIENT(link, xdata, frame, op_ret, op_errno, inode, buf,
+ preparent, postparent, xdata);
+ return 0;
}
-
-static int
-pl_client_destroy_cbk (xlator_t *this, client_t *client)
+int
+pl_link(call_frame_t *frame, xlator_t *this, loc_t *oldloc, loc_t *newloc,
+ dict_t *xdata)
{
- void *tmp = NULL;
- pl_ctx_t *pl_ctx = NULL;
+ pl_inode_t *pl_inode;
- pl_client_disconnect_cbk (this, client);
+ pl_inode = pl_inode_get(this, oldloc->inode, NULL);
+ if (pl_inode == NULL) {
+ STACK_UNWIND_STRICT(link, frame, -1, ENOMEM, NULL, NULL, NULL, NULL,
+ NULL);
+ return 0;
+ }
- client_ctx_del (client, this, &tmp);
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), oldloc, newloc);
+ STACK_WIND_COOKIE(frame, pl_link_cbk, pl_inode, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->link, oldloc, newloc, xdata);
+ return 0;
+}
- if (tmp == NULL)
- return 0;
+int32_t
+pl_fsync_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct iatt *prebuf, struct iatt *postbuf,
+ dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(fsync, xdata, frame, op_ret, op_errno, prebuf,
+ postbuf, xdata);
+ return 0;
+}
- pl_ctx = tmp;
+int
+pl_fsync(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+ dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fsync_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fsync, fd, datasync, xdata);
+ return 0;
+}
- GF_ASSERT (list_empty(&pl_ctx->inodelk_lockers));
- GF_ASSERT (list_empty(&pl_ctx->entrylk_lockers));
+int32_t
+pl_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, gf_dirent_t *entries,
+ dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(readdir, xdata, frame, op_ret, op_errno, entries,
+ xdata);
+ return 0;
+}
- pthread_mutex_destroy (&pl_ctx->lock);
- GF_FREE (pl_ctx);
+int
+pl_readdir(call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
+ off_t offset, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_readdir_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readdir, fd, size, offset, xdata);
+ return 0;
+}
- return 0;
+int32_t
+pl_fsyncdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(fsyncdir, xdata, frame, op_ret, op_errno, xdata);
+ return 0;
}
+int
+pl_fsyncdir(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t datasync,
+ dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fsyncdir_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fsyncdir, fd, datasync, xdata);
+ return 0;
+}
+
+int32_t
+pl_statfs_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, struct statvfs *buf, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(statfs, xdata, frame, op_ret, op_errno, buf,
+ xdata);
+ return 0;
+}
int
-init (xlator_t *this)
+pl_statfs(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_statfs_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->statfs, loc, xdata);
+ return 0;
+}
+
+int32_t
+pl_removexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *xdata)
{
- posix_locks_private_t *priv = NULL;
- xlator_list_t *trav = NULL;
- data_t *mandatory = NULL;
- data_t *trace = NULL;
- int ret = -1;
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
- if (!this->children || this->children->next) {
- gf_log (this->name, GF_LOG_CRITICAL,
- "FATAL: posix-locks should have exactly one child");
- goto out;
+ local = frame->local;
+ if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+ pl_inode = pl_inode_get(this, local->inode, NULL);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
}
- if (!this->parents) {
- gf_log (this->name, GF_LOG_WARNING,
- "Volume is dangling. Please check the volume file.");
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ pl_inode->mlock_enforced = _gf_false;
+ pl_inode->check_mlock_info = _gf_false;
+ pl_inode->track_fop_wind_count = _gf_true;
}
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- trav = this->children;
- while (trav->xlator->children)
- trav = trav->xlator->children;
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(removexattr, xdata, frame, op_ret, op_errno,
+ xdata);
+ return 0;
+}
- if (strncmp ("storage/", trav->xlator->type, 8)) {
- gf_log (this->name, GF_LOG_CRITICAL,
- "'locks' translator is not loaded over a storage "
- "translator");
- goto out;
- }
+int
+pl_removexattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
+ const char *name, dict_t *xdata)
+{
+ int op_ret = 0;
+ int op_errno = EINVAL;
+ posix_locks_private_t *priv = this->private;
- priv = GF_CALLOC (1, sizeof (*priv),
- gf_locks_mt_posix_locks_private_t);
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
- mandatory = dict_get (this->options, "mandatory-locks");
- if (mandatory)
- gf_log (this->name, GF_LOG_WARNING,
- "mandatory locks not supported in this minor release.");
+ PL_CHECK_LOCK_ENFORCE_KEY(frame, ((dict_t *)NULL), name, this, loc,
+ ((fd_t *)NULL), priv);
- trace = dict_get (this->options, "trace");
- if (trace) {
- if (gf_string2boolean (trace->data,
- &priv->trace) == -1) {
- gf_log (this->name, GF_LOG_ERROR,
- "'trace' takes on only boolean values.");
- goto out;
- }
+ STACK_WIND(frame, pl_removexattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->removexattr, loc, name, xdata);
+ return 0;
+
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(removexattr, xdata, frame, op_ret, op_errno,
+ NULL);
+
+ return 0;
+}
+
+int32_t
+pl_fremovexattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *xdata)
+{
+ pl_local_t *local = NULL;
+ pl_inode_t *pl_inode = NULL;
+
+ local = frame->local;
+ if (local && local->update_mlock_enforced_flag && op_ret != -1) {
+ pl_inode = pl_inode_get(this, local->inode, NULL);
+ if (!pl_inode) {
+ op_ret = -1;
+ op_errno = ENOMEM;
+ goto unwind;
}
- this->local_pool = mem_pool_new (pl_local_t, 32);
- if (!this->local_pool) {
- ret = -1;
- gf_log (this->name, GF_LOG_ERROR,
- "failed to create local_t's memory pool");
- goto out;
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ pl_inode->mlock_enforced = _gf_false;
+ pl_inode->check_mlock_info = _gf_false;
}
+ pthread_mutex_unlock(&pl_inode->mutex);
+ }
- this->private = priv;
- ret = 0;
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(fremovexattr, xdata, frame, op_ret, op_errno,
+ xdata);
+ return 0;
+}
-out:
- if (ret) {
- GF_FREE (priv);
- }
- return ret;
+int
+pl_fremovexattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *name,
+ dict_t *xdata)
+{
+ int op_ret = -1;
+ int op_errno = EINVAL;
+ posix_locks_private_t *priv = this->private;
+
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+
+ PL_CHECK_LOCK_ENFORCE_KEY(frame, ((dict_t *)NULL), name, this,
+ ((loc_t *)NULL), fd, priv);
+
+ STACK_WIND(frame, pl_fremovexattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fremovexattr, fd, name, xdata);
+ return 0;
+
+unwind:
+ PL_STACK_UNWIND_FOR_CLIENT(fremovexattr, xdata, frame, op_ret, op_errno,
+ NULL);
+ return 0;
}
+int32_t
+pl_rchecksum_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, uint32_t weak_cksum,
+ uint8_t *strong_cksum, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(rchecksum, xdata, frame, op_ret, op_errno,
+ weak_cksum, strong_cksum, xdata);
+ return 0;
+}
int
-fini (xlator_t *this)
+pl_rchecksum(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ int32_t len, dict_t *xdata)
{
- posix_locks_private_t *priv = NULL;
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_rchecksum_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->rchecksum, fd, offset, len, xdata);
+ return 0;
+}
- priv = this->private;
- if (!priv)
- return 0;
- this->private = NULL;
- GF_FREE (priv->brickname);
- GF_FREE (priv);
+int32_t
+pl_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(xattrop, xdata, frame, op_ret, op_errno, dict,
+ xdata);
+ return 0;
+}
- return 0;
+int
+pl_xattrop(call_frame_t *frame, xlator_t *this, loc_t *loc,
+ gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_xattrop_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->xattrop, loc, optype, xattr, xdata);
+ return 0;
}
+int32_t
+pl_fxattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, dict_t *dict, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(fxattrop, xdata, frame, op_ret, op_errno, dict,
+ xdata);
+ return 0;
+}
int
-pl_inodelk (call_frame_t *frame, xlator_t *this,
- const char *volume, loc_t *loc, int32_t cmd, struct gf_flock *flock,
- dict_t *xdata);
+pl_fxattrop(call_frame_t *frame, xlator_t *this, fd_t *fd,
+ gf_xattrop_flags_t optype, dict_t *xattr, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fxattrop_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fxattrop, fd, optype, xattr, xdata);
+ return 0;
+}
+
+int32_t
+pl_setattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+ struct iatt *statpost, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(setattr, xdata, frame, op_ret, op_errno, statpre,
+ statpost, xdata);
+ return 0;
+}
int
-pl_finodelk (call_frame_t *frame, xlator_t *this,
- const char *volume, fd_t *fd, int32_t cmd, struct gf_flock *flock,
- dict_t *xdata);
+pl_setattr(call_frame_t *frame, xlator_t *this, loc_t *loc, struct iatt *stbuf,
+ int32_t valid, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_setattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->setattr, loc, stbuf, valid, xdata);
+ return 0;
+}
+
+int32_t
+pl_fsetattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *statpre,
+ struct iatt *statpost, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(fsetattr, xdata, frame, op_ret, op_errno,
+ statpre, statpost, xdata);
+ return 0;
+}
int
-pl_entrylk (call_frame_t *frame, xlator_t *this,
- const char *volume, loc_t *loc, const char *basename,
- entrylk_cmd cmd, entrylk_type type, dict_t *xdata);
+pl_fsetattr(call_frame_t *frame, xlator_t *this, fd_t *fd, struct iatt *stbuf,
+ int32_t valid, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fsetattr_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fsetattr, fd, stbuf, valid, xdata);
+ return 0;
+}
+
+int32_t
+pl_fallocate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, struct iatt *pre,
+ struct iatt *post, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(fallocate, xdata, frame, op_ret, op_errno, pre,
+ post, xdata);
+ return 0;
+}
+
+int
+pl_fallocate(call_frame_t *frame, xlator_t *this, fd_t *fd, int32_t keep_size,
+ off_t offset, size_t len, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_fallocate_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->fallocate, fd, keep_size, offset, len,
+ xdata);
+ return 0;
+}
+
+int32_t
+pl_readlink_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, const char *path,
+ struct iatt *buf, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(readlink, xdata, frame, op_ret, op_errno, path,
+ buf, xdata);
+ return 0;
+}
int
-pl_fentrylk (call_frame_t *frame, xlator_t *this,
- const char *volume, fd_t *fd, const char *basename,
- entrylk_cmd cmd, entrylk_type type, dict_t *xdata);
+pl_readlink(call_frame_t *frame, xlator_t *this, loc_t *loc, size_t size,
+ dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_readlink_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readlink, loc, size, xdata);
+ return 0;
+}
+
+int32_t
+pl_access_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(access, xdata, frame, op_ret, op_errno, xdata);
+ return 0;
+}
+
+int
+pl_access(call_frame_t *frame, xlator_t *this, loc_t *loc, int32_t mask,
+ dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, ((fd_t *)NULL), loc, NULL);
+ STACK_WIND(frame, pl_access_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->access, loc, mask, xdata);
+ return 0;
+}
+
+int32_t
+pl_seek_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int32_t op_ret,
+ int32_t op_errno, off_t offset, dict_t *xdata)
+{
+ PL_STACK_UNWIND_FOR_CLIENT(seek, xdata, frame, op_ret, op_errno, offset,
+ xdata);
+ return 0;
+}
+
+int32_t
+pl_seek(call_frame_t *frame, xlator_t *this, fd_t *fd, off_t offset,
+ gf_seek_what_t what, dict_t *xdata)
+{
+ PL_LOCAL_GET_REQUESTS(frame, this, xdata, fd, NULL, NULL);
+ STACK_WIND(frame, pl_seek_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->seek, fd, offset, what, xdata);
+ return 0;
+}
struct xlator_fops fops = {
- .lookup = pl_lookup,
- .create = pl_create,
- .truncate = pl_truncate,
- .ftruncate = pl_ftruncate,
- .open = pl_open,
- .readv = pl_readv,
- .writev = pl_writev,
- .lk = pl_lk,
- .inodelk = pl_inodelk,
- .finodelk = pl_finodelk,
- .entrylk = pl_entrylk,
- .fentrylk = pl_fentrylk,
- .flush = pl_flush,
- .opendir = pl_opendir,
- .readdirp = pl_readdirp,
- .getxattr = pl_getxattr,
- .fgetxattr = pl_fgetxattr,
- .fsetxattr = pl_fsetxattr,
+ .lookup = pl_lookup,
+ .create = pl_create,
+ .fstat = pl_fstat,
+ .truncate = pl_truncate,
+ .ftruncate = pl_ftruncate,
+ .discard = pl_discard,
+ .zerofill = pl_zerofill,
+ .open = pl_open,
+ .readv = pl_readv,
+ .writev = pl_writev,
+ .lk = pl_lk,
+ .inodelk = pl_inodelk,
+ .finodelk = pl_finodelk,
+ .entrylk = pl_entrylk,
+ .fentrylk = pl_fentrylk,
+ .flush = pl_flush,
+ .opendir = pl_opendir,
+ .readdirp = pl_readdirp,
+ .setxattr = pl_setxattr,
+ .fsetxattr = pl_fsetxattr,
+ .getxattr = pl_getxattr,
+ .fgetxattr = pl_fgetxattr,
+ .removexattr = pl_removexattr,
+ .fremovexattr = pl_fremovexattr,
+ .rename = pl_rename,
+ .getactivelk = pl_getactivelk,
+ .setactivelk = pl_setactivelk,
+ .unlink = pl_unlink,
+ .access = pl_access,
+ .readlink = pl_readlink,
+ .fallocate = pl_fallocate,
+ .fsetattr = pl_fsetattr,
+ .setattr = pl_setattr,
+ .fxattrop = pl_fxattrop,
+ .xattrop = pl_xattrop,
+ .rchecksum = pl_rchecksum,
+ .statfs = pl_statfs,
+ .fsyncdir = pl_fsyncdir,
+ .readdir = pl_readdir,
+ .symlink = pl_symlink,
+ .link = pl_link,
+ .rmdir = pl_rmdir,
+ .mknod = pl_mknod,
+ .stat = pl_stat,
+ .seek = pl_seek,
};
struct xlator_dumpops dumpops = {
- .inodectx = pl_dump_inode_priv,
+ .inodectx = pl_dump_inode_priv,
};
struct xlator_cbks cbks = {
- .forget = pl_forget,
- .release = pl_release,
- .releasedir = pl_releasedir,
- .client_destroy = pl_client_destroy_cbk,
- .client_disconnect = pl_client_disconnect_cbk,
+ .forget = pl_forget,
+ .release = pl_release,
+ .releasedir = pl_releasedir,
+ .client_destroy = pl_client_destroy_cbk,
+ .client_disconnect = pl_client_disconnect_cbk,
};
-
struct volume_options options[] = {
- { .key = { "mandatory-locks", "mandatory" },
- .type = GF_OPTION_TYPE_BOOL
- },
- { .key = { "trace" },
- .type = GF_OPTION_TYPE_BOOL
- },
- { .key = {NULL} },
+ {.key = {"mandatory-locking"},
+ .type = GF_OPTION_TYPE_STR,
+ .default_value = "off",
+ .op_version = {GD_OP_VERSION_3_8_0},
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .tags = {"locks"},
+ .description = "Specifies the mandatory-locking mode. Valid options "
+ "are 'file' to use linux style mandatory locks, "
+ "'forced' to use volume strictly under mandatory lock "
+ "semantics only and 'optimal' to treat advisory and "
+ "mandatory locks separately on their own."},
+ {.key = {"trace"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "off",
+ .op_version = {GD_OP_VERSION_3_7_0},
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .tags = {"locks"},
+ .description = "Trace the different lock requests "
+ "to logs."},
+ {.key = {"monkey-unlocking"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "false",
+ .op_version = {GD_OP_VERSION_3_9_0},
+ .flags = OPT_FLAG_SETTABLE,
+ .tags = {"locks"},
+ .description = "Ignore a random number of unlock requests. Useful "
+ "for testing/creating robust lock recovery mechanisms."},
+ {
+ .key = {"revocation-secs"},
+ .type = GF_OPTION_TYPE_INT,
+ .min = 0,
+ .max = INT_MAX,
+ .default_value = "0",
+ .op_version = {GD_OP_VERSION_3_9_0},
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .tags = {"locks"},
+ .description = "Maximum time a lock can be taken out, before"
+ "being revoked.",
+ },
+ {
+ .key = {"revocation-clear-all"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "false",
+ .op_version = {GD_OP_VERSION_3_9_0},
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .tags = {"locks"},
+ .description = "If set to true, will revoke BOTH granted and blocked "
+ "(pending) lock requests if a revocation threshold is "
+ "hit.",
+ },
+ {.key = {"revocation-max-blocked"},
+ .type = GF_OPTION_TYPE_INT,
+ .min = 0,
+ .max = INT_MAX,
+ .default_value = "0",
+ .op_version = {GD_OP_VERSION_3_9_0},
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .tags = {"locks"},
+ .description = "A number of blocked lock requests after which a lock "
+ "will be revoked to allow the others to proceed. Can "
+ "be used in conjunction w/ revocation-clear-all."},
+ {.key = {"notify-contention"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "yes",
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .op_version = {GD_OP_VERSION_4_0_0},
+ .tags = {"locks", "contention"},
+ .description = "When this option is enabled and a lock request "
+ "conflicts with a currently granted lock, an upcall "
+ "notification will be sent to the current owner of "
+ "the lock to request it to be released as soon as "
+ "possible."},
+ {.key = {"notify-contention-delay"},
+ .type = GF_OPTION_TYPE_INT,
+ .min = 0, /* An upcall notification is sent every time a conflict is
+ * detected. */
+ .max = 60,
+ .default_value = "5",
+ .flags = OPT_FLAG_SETTABLE | OPT_FLAG_DOC,
+ .op_version = {GD_OP_VERSION_4_0_0},
+ .tags = {"locks", "contention", "timeout"},
+ .description = "This value determines the minimum amount of time "
+ "(in seconds) between upcall contention notifications "
+ "on the same inode. If multiple lock requests are "
+ "received during this period, only one upcall will "
+ "be sent."},
+ {.key = {"enforce-mandatory-lock"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "off",
+ .flags = OPT_FLAG_SETTABLE,
+ .op_version = {GD_OP_VERSION_6_0},
+ .description = "option to enable lock enforcement"},
+ {.key = {NULL}},
+};
+
+xlator_api_t xlator_api = {
+ .init = init,
+ .fini = fini,
+ .reconfigure = reconfigure,
+ .mem_acct_init = mem_acct_init,
+ .op_version = {1}, /* Present from the initial version */
+ .dumpops = &dumpops,
+ .fops = &fops,
+ .cbks = &cbks,
+ .options = options,
+ .identifier = "locks",
+ .category = GF_MAINTAINED,
};
diff --git a/xlators/features/locks/src/reservelk.c b/xlators/features/locks/src/reservelk.c
index 11abd26d85f..604691fd887 100644
--- a/xlators/features/locks/src/reservelk.c
+++ b/xlators/features/locks/src/reservelk.c
@@ -7,437 +7,376 @@
later), or the GNU General Public License, version 2 (GPLv2), in all
cases as published by the Free Software Foundation.
*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
#include "locks.h"
#include "common.h"
-void
-__delete_reserve_lock (posix_lock_t *lock)
-{
- list_del (&lock->list);
-}
-
-void
-__destroy_reserve_lock (posix_lock_t *lock)
-{
- GF_FREE (lock);
-}
-
/* Return true if the two reservelks have exactly same lock boundaries */
int
-reservelks_equal (posix_lock_t *l1, posix_lock_t *l2)
+reservelks_equal(posix_lock_t *l1, posix_lock_t *l2)
{
- if ((l1->fl_start == l2->fl_start) &&
- (l1->fl_end == l2->fl_end))
- return 1;
+ if ((l1->fl_start == l2->fl_start) && (l1->fl_end == l2->fl_end))
+ return 1;
- return 0;
+ return 0;
}
/* Determine if lock is grantable or not */
static posix_lock_t *
-__reservelk_grantable (pl_inode_t *pl_inode, posix_lock_t *lock)
+__reservelk_grantable(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- xlator_t *this = NULL;
- posix_lock_t *l = NULL;
- posix_lock_t *ret_lock = NULL;
-
- this = THIS;
-
- if (list_empty (&pl_inode->reservelk_list)) {
- gf_log (this->name, GF_LOG_TRACE,
- "No reservelks in list");
- goto out;
- }
- list_for_each_entry (l, &pl_inode->reservelk_list, list){
- if (reservelks_equal (lock, l)) {
- ret_lock = l;
- break;
- }
+ xlator_t *this = THIS;
+ posix_lock_t *l = NULL;
+ posix_lock_t *ret_lock = NULL;
+
+ if (list_empty(&pl_inode->reservelk_list)) {
+ gf_log(this->name, GF_LOG_TRACE, "No reservelks in list");
+ goto out;
+ }
+ list_for_each_entry(l, &pl_inode->reservelk_list, list)
+ {
+ if (reservelks_equal(lock, l)) {
+ ret_lock = l;
+ break;
}
+ }
out:
- return ret_lock;
+ return ret_lock;
}
-static inline int
-__same_owner_reservelk (posix_lock_t *l1, posix_lock_t *l2)
+static int
+__same_owner_reservelk(posix_lock_t *l1, posix_lock_t *l2)
{
- return (is_same_lkowner (&l1->owner, &l2->owner));
-
+ return (is_same_lkowner(&l1->owner, &l2->owner));
}
static posix_lock_t *
-__matching_reservelk (pl_inode_t *pl_inode, posix_lock_t *lock)
+__matching_reservelk(pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *l = NULL;
+ posix_lock_t *l = NULL;
- if (list_empty (&pl_inode->reservelk_list)) {
- gf_log ("posix-locks", GF_LOG_TRACE,
- "reservelk list empty");
- return NULL;
- }
+ if (list_empty(&pl_inode->reservelk_list)) {
+ gf_log("posix-locks", GF_LOG_TRACE, "reservelk list empty");
+ return NULL;
+ }
- list_for_each_entry (l, &pl_inode->reservelk_list, list) {
- if (reservelks_equal (l, lock)) {
- gf_log ("posix-locks", GF_LOG_TRACE,
- "equal reservelk found");
- break;
- }
+ list_for_each_entry(l, &pl_inode->reservelk_list, list)
+ {
+ if (reservelks_equal(l, lock)) {
+ gf_log("posix-locks", GF_LOG_TRACE, "equal reservelk found");
+ break;
}
+ }
- return l;
+ return l;
}
static int
-__reservelk_conflict (xlator_t *this, pl_inode_t *pl_inode,
- posix_lock_t *lock)
+__reservelk_conflict(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *conf = NULL;
- int ret = 0;
-
- conf = __matching_reservelk (pl_inode, lock);
- if (conf) {
- gf_log (this->name, GF_LOG_TRACE,
- "Matching reservelk found");
- if (__same_owner_reservelk (lock, conf)) {
- list_del_init (&conf->list);
- gf_log (this->name, GF_LOG_TRACE,
- "Removing the matching reservelk for setlk to progress");
- GF_FREE (conf);
- ret = 0;
- } else {
- gf_log (this->name, GF_LOG_TRACE,
- "Conflicting reservelk found");
- ret = 1;
- }
-
+ int ret = 0;
+
+ posix_lock_t *conf = __matching_reservelk(pl_inode, lock);
+ if (conf) {
+ gf_log(this->name, GF_LOG_TRACE, "Matching reservelk found");
+ if (__same_owner_reservelk(lock, conf)) {
+ list_del_init(&conf->list);
+ gf_log(this->name, GF_LOG_TRACE,
+ "Removing the matching reservelk for setlk to progress");
+ __destroy_lock(conf);
+ ret = 0;
+ } else {
+ gf_log(this->name, GF_LOG_TRACE, "Conflicting reservelk found");
+ ret = 1;
}
- return ret;
-
+ }
+ return ret;
}
int
-pl_verify_reservelk (xlator_t *this, pl_inode_t *pl_inode,
- posix_lock_t *lock, int can_block)
+pl_verify_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ const int can_block)
{
- int ret = 0;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- if (__reservelk_conflict (this, pl_inode, lock)) {
- gf_log (this->name, GF_LOG_TRACE,
- "Found conflicting reservelk. Blocking until reservelk is unlocked.");
- lock->blocked = can_block;
- list_add_tail (&lock->list, &pl_inode->blocked_calls);
- ret = -1;
- goto unlock;
- }
-
- gf_log (this->name, GF_LOG_TRACE,
- "no conflicting reservelk found. Call continuing");
- ret = 0;
-
+ int ret = 0;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ if (__reservelk_conflict(this, pl_inode, lock)) {
+ lock->blocked = can_block;
+ list_add_tail(&lock->list, &pl_inode->blocked_calls);
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_log(this->name, GF_LOG_TRACE,
+ "Found conflicting reservelk. Blocking until reservelk is "
+ "unlocked.");
+ ret = -1;
+ goto out;
}
-unlock:
- pthread_mutex_unlock (&pl_inode->mutex);
-
- return ret;
-
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_log(this->name, GF_LOG_TRACE,
+ "no conflicting reservelk found. Call continuing");
+ ret = 0;
+out:
+ return ret;
}
-
/* Determines if lock can be granted and adds the lock. If the lock
* is blocking, adds it to the blocked_reservelks.
*/
static int
-__lock_reservelk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
- int can_block)
+__lock_reservelk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ const int can_block)
{
- posix_lock_t *conf = NULL;
- int ret = -EINVAL;
-
- conf = __reservelk_grantable (pl_inode, lock);
- if (conf){
- ret = -EAGAIN;
- if (can_block == 0)
- goto out;
+ int ret = -EINVAL;
- list_add_tail (&lock->list, &pl_inode->blocked_reservelks);
+ posix_lock_t *conf = __reservelk_grantable(pl_inode, lock);
+ if (conf) {
+ ret = -EAGAIN;
+ if (can_block == 0)
+ goto out;
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) lk-owner:%s %"PRId64" - %"PRId64" => Blocked",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
+ list_add_tail(&lock->list, &pl_inode->blocked_reservelks);
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) lk-owner:%s %" PRId64 " - %" PRId64 " => Blocked",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+ lock->user_flock.l_len);
- goto out;
- }
+ goto out;
+ }
- list_add (&lock->list, &pl_inode->reservelk_list);
+ list_add(&lock->list, &pl_inode->reservelk_list);
- ret = 0;
+ ret = 0;
out:
- return ret;
+ return ret;
}
static posix_lock_t *
-find_matching_reservelk (posix_lock_t *lock, pl_inode_t *pl_inode)
+find_matching_reservelk(posix_lock_t *lock, pl_inode_t *pl_inode)
{
- posix_lock_t *l = NULL;
- list_for_each_entry (l, &pl_inode->reservelk_list, list) {
- if (reservelks_equal (l, lock))
- return l;
- }
- return NULL;
+ posix_lock_t *l = NULL;
+ list_for_each_entry(l, &pl_inode->reservelk_list, list)
+ {
+ if (reservelks_equal(l, lock))
+ return l;
+ }
+ return NULL;
}
/* Set F_UNLCK removes a lock which has the exact same lock boundaries
* as the UNLCK lock specifies. If such a lock is not found, returns invalid
*/
static posix_lock_t *
-__reserve_unlock_lock (xlator_t *this, posix_lock_t *lock, pl_inode_t *pl_inode)
+__reserve_unlock_lock(xlator_t *this, posix_lock_t *lock, pl_inode_t *pl_inode)
{
-
- posix_lock_t *conf = NULL;
-
- conf = find_matching_reservelk (lock, pl_inode);
- if (!conf) {
- gf_log (this->name, GF_LOG_DEBUG,
- " Matching lock not found for unlock");
- goto out;
- }
- __delete_reserve_lock (conf);
- gf_log (this->name, GF_LOG_DEBUG,
- " Matching lock found for unlock");
+ posix_lock_t *conf = find_matching_reservelk(lock, pl_inode);
+ if (!conf) {
+ gf_log(this->name, GF_LOG_DEBUG, " Matching lock not found for unlock");
+ goto out;
+ }
+ __delete_lock(conf);
+ gf_log(this->name, GF_LOG_DEBUG, " Matching lock found for unlock");
out:
- return conf;
-
-
+ return conf;
}
static void
-__grant_blocked_reserve_locks (xlator_t *this, pl_inode_t *pl_inode,
- struct list_head *granted)
+__grant_blocked_reserve_locks(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted)
{
- int bl_ret = 0;
- posix_lock_t *bl = NULL;
- posix_lock_t *tmp = NULL;
-
- struct list_head blocked_list;
+ int bl_ret = 0;
+ posix_lock_t *bl = NULL;
+ posix_lock_t *tmp = NULL;
- INIT_LIST_HEAD (&blocked_list);
- list_splice_init (&pl_inode->blocked_reservelks, &blocked_list);
+ struct list_head blocked_list;
- list_for_each_entry_safe (bl, tmp, &blocked_list, list) {
+ INIT_LIST_HEAD(&blocked_list);
+ list_splice_init(&pl_inode->blocked_reservelks, &blocked_list);
- list_del_init (&bl->list);
+ list_for_each_entry_safe(bl, tmp, &blocked_list, list)
+ {
+ list_del_init(&bl->list);
- bl_ret = __lock_reservelk (this, pl_inode, bl, 1);
+ bl_ret = __lock_reservelk(this, pl_inode, bl, 1);
- if (bl_ret == 0) {
- list_add (&bl->list, granted);
- }
+ if (bl_ret == 0) {
+ list_add(&bl->list, granted);
}
- return;
+ }
+ return;
}
/* Grant all reservelks blocked on lock(s) */
void
-grant_blocked_reserve_locks (xlator_t *this, pl_inode_t *pl_inode)
+grant_blocked_reserve_locks(xlator_t *this, pl_inode_t *pl_inode)
{
- struct list_head granted;
- posix_lock_t *lock = NULL;
- posix_lock_t *tmp = NULL;
+ struct list_head granted;
+ posix_lock_t *lock = NULL;
+ posix_lock_t *tmp = NULL;
- INIT_LIST_HEAD (&granted);
-
- if (list_empty (&pl_inode->blocked_reservelks)) {
- gf_log (this->name, GF_LOG_TRACE,
- "No blocked locks to be granted");
- return;
- }
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __grant_blocked_reserve_locks (this, pl_inode, &granted);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
-
- list_for_each_entry_safe (lock, tmp, &granted, list) {
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => Granted",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
-
- STACK_UNWIND_STRICT (lk, lock->frame, 0, 0, &lock->user_flock,
- NULL);
- }
+ INIT_LIST_HEAD(&granted);
+ if (list_empty(&pl_inode->blocked_reservelks)) {
+ gf_log(this->name, GF_LOG_TRACE, "No blocked locks to be granted");
+ return;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __grant_blocked_reserve_locks(this, pl_inode, &granted);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ list_for_each_entry_safe(lock, tmp, &granted, list)
+ {
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => Granted",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+ lock->user_flock.l_len);
+
+ STACK_UNWIND_STRICT(lk, lock->frame, 0, 0, &lock->user_flock, NULL);
+ }
}
static void
-__grant_blocked_lock_calls (xlator_t *this, pl_inode_t *pl_inode,
- struct list_head *granted)
+__grant_blocked_lock_calls(xlator_t *this, pl_inode_t *pl_inode,
+ struct list_head *granted)
{
- int bl_ret = 0;
- posix_lock_t *bl = NULL;
- posix_lock_t *tmp = NULL;
+ int bl_ret = 0;
+ posix_lock_t *bl = NULL;
+ posix_lock_t *tmp = NULL;
- struct list_head blocked_list;
+ struct list_head blocked_list;
- INIT_LIST_HEAD (&blocked_list);
- list_splice_init (&pl_inode->blocked_reservelks, &blocked_list);
+ INIT_LIST_HEAD(&blocked_list);
+ list_splice_init(&pl_inode->blocked_reservelks, &blocked_list);
- list_for_each_entry_safe (bl, tmp, &blocked_list, list) {
+ list_for_each_entry_safe(bl, tmp, &blocked_list, list)
+ {
+ list_del_init(&bl->list);
- list_del_init (&bl->list);
+ bl_ret = pl_verify_reservelk(this, pl_inode, bl, bl->blocked);
- bl_ret = pl_verify_reservelk (this, pl_inode, bl, bl->blocked);
-
- if (bl_ret == 0) {
- list_add_tail (&bl->list, granted);
- }
+ if (bl_ret == 0) {
+ list_add_tail(&bl->list, granted);
}
- return;
+ }
+ return;
}
void
-grant_blocked_lock_calls (xlator_t *this, pl_inode_t *pl_inode)
+grant_blocked_lock_calls(xlator_t *this, pl_inode_t *pl_inode)
{
- struct list_head granted;
- posix_lock_t *lock = NULL;
- posix_lock_t *tmp = NULL;
- fd_t *fd = NULL;
-
- int can_block = 0;
- int32_t cmd = 0;
- int ret = 0;
-
- if (list_empty (&pl_inode->blocked_calls)) {
- gf_log (this->name, GF_LOG_TRACE,
- "No blocked lock calls to be granted");
- return;
- }
+ struct list_head granted;
+ posix_lock_t *lock = NULL;
+ posix_lock_t *tmp = NULL;
+ fd_t *fd = NULL;
- pthread_mutex_lock (&pl_inode->mutex);
- {
- __grant_blocked_lock_calls (this, pl_inode, &granted);
- }
- pthread_mutex_unlock (&pl_inode->mutex);
-
- list_for_each_entry_safe (lock, tmp, &granted, list) {
- fd = fd_from_fdnum (lock);
-
- if (lock->blocked) {
- can_block = 1;
- cmd = F_SETLKW;
- }
- else
- cmd = F_SETLK;
-
- lock->blocked = 0;
- ret = pl_setlk (this, pl_inode, lock, can_block);
- if (ret == -1) {
- if (can_block) {
- pl_trace_block (this, lock->frame, fd, NULL,
- cmd, &lock->user_flock, NULL);
- continue;
- } else {
- gf_log (this->name, GF_LOG_DEBUG, "returning EAGAIN");
- pl_trace_out (this, lock->frame, fd, NULL, cmd,
- &lock->user_flock, -1, EAGAIN, NULL);
- pl_update_refkeeper (this, fd->inode);
- STACK_UNWIND_STRICT (lk, lock->frame, -1,
- EAGAIN, &lock->user_flock,
- NULL);
- __destroy_lock (lock);
- }
- }
+ int can_block = 0;
+ int32_t cmd = 0;
+ int ret = 0;
+ if (list_empty(&pl_inode->blocked_calls)) {
+ gf_log(this->name, GF_LOG_TRACE, "No blocked lock calls to be granted");
+ return;
+ }
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ __grant_blocked_lock_calls(this, pl_inode, &granted);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ list_for_each_entry_safe(lock, tmp, &granted, list)
+ {
+ fd = fd_from_fdnum(lock);
+
+ if (lock->blocked) {
+ can_block = 1;
+ cmd = F_SETLKW;
+ } else
+ cmd = F_SETLK;
+
+ lock->blocked = 0;
+ ret = pl_setlk(this, pl_inode, lock, can_block);
+ if (ret == -1) {
+ if (can_block) {
+ continue;
+ } else {
+ gf_log(this->name, GF_LOG_DEBUG, "returning EAGAIN");
+ pl_trace_out(this, lock->frame, fd, NULL, cmd,
+ &lock->user_flock, -1, EAGAIN, NULL);
+ pl_update_refkeeper(this, fd->inode);
+ STACK_UNWIND_STRICT(lk, lock->frame, -1, EAGAIN,
+ &lock->user_flock, NULL);
+ __destroy_lock(lock);
+ }
}
-
+ }
}
-
int
-pl_reserve_unlock (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
+pl_reserve_unlock(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock)
{
- posix_lock_t *retlock = NULL;
- int ret = -1;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
- retlock = __reserve_unlock_lock (this, lock, pl_inode);
- if (!retlock) {
- gf_log (this->name, GF_LOG_DEBUG,
- "Bad Unlock issued on Inode lock");
- ret = -EINVAL;
- goto out;
- }
-
- gf_log (this->name, GF_LOG_TRACE,
- "Reservelk Unlock successful");
- __destroy_reserve_lock (retlock);
- ret = 0;
+ posix_lock_t *retlock = NULL;
+ int ret = -1;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ retlock = __reserve_unlock_lock(this, lock, pl_inode);
+ if (!retlock) {
+ pthread_mutex_unlock(&pl_inode->mutex);
+ gf_log(this->name, GF_LOG_DEBUG, "Bad Unlock issued on Inode lock");
+ ret = -EINVAL;
+ goto out;
}
-out:
- pthread_mutex_unlock (&pl_inode->mutex);
-
- grant_blocked_reserve_locks (this, pl_inode);
- grant_blocked_lock_calls (this, pl_inode);
- return ret;
+ gf_log(this->name, GF_LOG_TRACE, "Reservelk Unlock successful");
+ __destroy_lock(retlock);
+ ret = 0;
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+out:
+ grant_blocked_reserve_locks(this, pl_inode);
+ grant_blocked_lock_calls(this, pl_inode);
+ return ret;
}
int
-pl_reserve_setlk (xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
- int can_block)
+pl_reserve_setlk(xlator_t *this, pl_inode_t *pl_inode, posix_lock_t *lock,
+ int can_block)
{
- int ret = -EINVAL;
-
- pthread_mutex_lock (&pl_inode->mutex);
- {
-
- ret = __lock_reservelk (this, pl_inode, lock, can_block);
- if (ret < 0)
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => NOK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->user_flock.l_start,
- lock->user_flock.l_len);
- else
- gf_log (this->name, GF_LOG_TRACE,
- "%s (pid=%d) (lk-owner=%s) %"PRId64" - %"PRId64" => OK",
- lock->fl_type == F_UNLCK ? "Unlock" : "Lock",
- lock->client_pid,
- lkowner_utoa (&lock->owner),
- lock->fl_start,
- lock->fl_end);
-
- }
- pthread_mutex_unlock (&pl_inode->mutex);
- return ret;
+ int ret = -EINVAL;
+
+ pthread_mutex_lock(&pl_inode->mutex);
+ {
+ ret = __lock_reservelk(this, pl_inode, lock, can_block);
+ }
+ pthread_mutex_unlock(&pl_inode->mutex);
+
+ if (ret < 0)
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => NOK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->user_flock.l_start,
+ lock->user_flock.l_len);
+ else
+ gf_log(this->name, GF_LOG_TRACE,
+ "%s (pid=%d) (lk-owner=%s) %" PRId64 " - %" PRId64 " => OK",
+ lock->fl_type == F_UNLCK ? "Unlock" : "Lock", lock->client_pid,
+ lkowner_utoa(&lock->owner), lock->fl_start, lock->fl_end);
+
+ return ret;
}
diff --git a/xlators/features/locks/tests/unit-test.c b/xlators/features/locks/tests/unit-test.c
index d2cca32dec3..d285b12b5aa 100644
--- a/xlators/features/locks/tests/unit-test.c
+++ b/xlators/features/locks/tests/unit-test.c
@@ -7,59 +7,71 @@
later), or the GNU General Public License, version 2 (GPLv2), in all
cases as published by the Free Software Foundation.
*/
-#ifndef _CONFIG_H
-#define _CONFIG_H
-#include "config.h"
-#endif
-
-#include "glusterfs.h"
-#include "compat.h"
-#include "xlator.h"
-#include "inode.h"
-#include "logging.h"
-#include "common-utils.h"
-#include "list.h"
+#include <glusterfs/glusterfs.h>
+#include <glusterfs/compat.h>
+#include <glusterfs/xlator.h>
+#include <glusterfs/logging.h>
+#include <glusterfs/common-utils.h>
+#include <glusterfs/list.h>
#include "locks.h"
#include "common.h"
-#define expect(cond) if (!(cond)) { goto out; }
+#define expect(cond) \
+ if (!(cond)) { \
+ goto out; \
+ }
-extern int lock_name (pl_inode_t *, const char *, entrylk_type);
-extern int unlock_name (pl_inode_t *, const char *, entrylk_type);
+extern int
+lock_name(pl_inode_t *, const char *, entrylk_type);
+extern int
+unlock_name(pl_inode_t *, const char *, entrylk_type);
-int main (int argc, char **argv)
+int
+main(int argc, char **argv)
{
- int ret = 1;
- int r = -1;
+ int ret = 1;
+ int r = -1;
+
+ pl_inode_t *pinode = CALLOC(sizeof(pl_inode_t), 1);
+ pthread_mutex_init(&pinode->dir_lock_mutex, NULL);
+ INIT_LIST_HEAD(&pinode->gf_dir_locks);
- pl_inode_t *pinode = CALLOC (sizeof (pl_inode_t), 1);
- pthread_mutex_init (&pinode->dir_lock_mutex, NULL);
- INIT_LIST_HEAD (&pinode->gf_dir_locks);
+ r = lock_name(pinode, NULL, ENTRYLK_WRLCK);
+ expect(r == 0);
+ {
+ r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+ expect(r == -EAGAIN);
+ }
+ r = unlock_name(pinode, NULL, ENTRYLK_WRLCK);
+ expect(r == 0);
- r = lock_name (pinode, NULL, ENTRYLK_WRLCK); expect (r == 0);
- {
- r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == -EAGAIN);
- }
- r = unlock_name (pinode, NULL, ENTRYLK_WRLCK); expect (r == 0);
+ r = lock_name(pinode, "foo", ENTRYLK_RDLCK);
+ expect(r == 0);
+ {
+ r = lock_name(pinode, "foo", ENTRYLK_RDLCK);
+ expect(r == 0);
+ {
+ r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+ expect(r == -EAGAIN);
+ }
+ r = unlock_name(pinode, "foo", ENTRYLK_RDLCK);
+ expect(r == 0);
+ }
+ r = unlock_name(pinode, "foo", ENTRYLK_RDLCK);
+ expect(r == 0);
- r = lock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
- {
- r = lock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
- {
- r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == -EAGAIN);
- }
- r = unlock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
- }
- r = unlock_name (pinode, "foo", ENTRYLK_RDLCK); expect (r == 0);
-
- r = lock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == 0);
- r = unlock_name (pinode, "foo", ENTRYLK_WRLCK); expect (r == 0);
+ r = lock_name(pinode, "foo", ENTRYLK_WRLCK);
+ expect(r == 0);
+ r = unlock_name(pinode, "foo", ENTRYLK_WRLCK);
+ expect(r == 0);
- r = lock_name (pinode, "baz", ENTRYLK_WRLCK); expect (r == 0);
- r = lock_name (pinode, "baz", ENTRYLK_RDLCK); expect (r == -EAGAIN);
+ r = lock_name(pinode, "baz", ENTRYLK_WRLCK);
+ expect(r == 0);
+ r = lock_name(pinode, "baz", ENTRYLK_RDLCK);
+ expect(r == -EAGAIN);
- ret = 0;
+ ret = 0;
out:
- return ret;
+ return ret;
}
generated by cgit (git 2.34.1) at 2025-12-28 13:59:38 +0000