summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht/src/dht-common.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht/src/dht-common.c')
-rw-r--r--xlators/cluster/dht/src/dht-common.c181
1 files changed, 63 insertions, 118 deletions
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 22ef8200911..8ba0cc4c732 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -17,6 +17,7 @@
#include <glusterfs/quota-common-utils.h>
#include <glusterfs/upcall-utils.h>
#include "glusterfs/compat-errno.h" // for ENODATA on BSD
+#include <glusterfs/common-utils.h>
#include <sys/time.h>
#include <libgen.h>
@@ -43,15 +44,6 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
static int
dht_rmdir_unlock(call_frame_t *frame, xlator_t *this);
-char *xattrs_to_heal[] = {"user.",
- POSIX_ACL_ACCESS_XATTR,
- POSIX_ACL_DEFAULT_XATTR,
- QUOTA_LIMIT_KEY,
- QUOTA_LIMIT_OBJECTS_KEY,
- GF_SELINUX_XATTR_KEY,
- GF_XATTR_MDATA_KEY,
- NULL};
-
static const char *dht_dbg_vxattrs[] = {DHT_DBG_HASHED_SUBVOL_PATTERN, NULL};
/* Check the xdata to make sure EBADF has been set by client xlator */
@@ -84,6 +76,8 @@ dht_set_fixed_dir_stat(struct iatt *stat)
static gf_boolean_t
dht_match_xattr(const char *key)
{
+ char **xattrs_to_heal = get_xattrs_to_heal();
+
return gf_get_index_by_elem(xattrs_to_heal, (char *)key) >= 0;
}
@@ -388,7 +382,7 @@ out:
/* Code to save hashed subvol on inode ctx as a mds subvol
*/
-static int
+int
dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol)
{
dht_inode_ctx_t *ctx = NULL;
@@ -619,13 +613,14 @@ dht_discover_complete(xlator_t *this, call_frame_t *discover_frame)
if (local->need_xattr_heal && !heal_path) {
local->need_xattr_heal = 0;
- ret = dht_dir_xattr_heal(this, local);
- if (ret)
- gf_msg(this->name, GF_LOG_ERROR, ret,
+ ret = dht_dir_xattr_heal(this, local, &op_errno);
+ if (ret) {
+ gf_msg(this->name, GF_LOG_ERROR, op_errno,
DHT_MSG_DIR_XATTR_HEAL_FAILED,
"xattr heal failed for "
"directory gfid is %s ",
gfid_local);
+ }
}
}
@@ -695,6 +690,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
int ret = -1;
dht_conf_t *conf = 0;
dht_layout_t *layout = NULL;
+ int32_t mds_heal_fresh_lookup = 0;
GF_VALIDATE_OR_GOTO(this->name, frame, out);
GF_VALIDATE_OR_GOTO(this->name, frame->local, out);
@@ -702,6 +698,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
local = frame->local;
conf = this->private;
layout = local->selfheal.layout;
+ mds_heal_fresh_lookup = local->mds_heal_fresh_lookup;
if (op_ret) {
gf_msg_debug(this->name, op_ret,
@@ -722,7 +719,7 @@ dht_common_mark_mdsxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
layout);
}
out:
- if (local && local->mds_heal_fresh_lookup)
+ if (mds_heal_fresh_lookup)
DHT_STACK_DESTROY(frame);
return 0;
}
@@ -1256,7 +1253,7 @@ err:
to non hashed subvol
*/
int
-dht_dir_xattr_heal(xlator_t *this, dht_local_t *local)
+dht_dir_xattr_heal(xlator_t *this, dht_local_t *local, int *op_errno)
{
dht_local_t *copy_local = NULL;
call_frame_t *copy = NULL;
@@ -1268,6 +1265,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local)
"No gfid exists for path %s "
"so healing xattr is not possible",
local->loc.path);
+ *op_errno = EIO;
goto out;
}
@@ -1281,6 +1279,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local)
"Memory allocation failed "
"for path %s gfid %s ",
local->loc.path, gfid_local);
+ *op_errno = ENOMEM;
DHT_STACK_DESTROY(copy);
} else {
copy_local->stbuf = local->stbuf;
@@ -1295,6 +1294,7 @@ dht_dir_xattr_heal(xlator_t *this, dht_local_t *local)
"Synctask creation failed to heal xattr "
"for path %s gfid %s ",
local->loc.path, gfid_local);
+ *op_errno = ENOMEM;
DHT_STACK_DESTROY(copy);
}
}
@@ -1651,7 +1651,7 @@ dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
uint32_t vol_commit_hash = 0;
xlator_t *subvol = NULL;
int32_t check_mds = 0;
- int errst = 0;
+ int errst = 0, i = 0;
int32_t mds_xattr_val[1] = {0};
GF_VALIDATE_OR_GOTO("dht", frame, err);
@@ -1718,6 +1718,14 @@ dht_revalidate_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
local->need_lookup_everywhere = 1;
} else if (IA_ISDIR(local->loc.inode->ia_type)) {
+ layout = local->layout;
+ for (i = 0; i < layout->cnt; i++) {
+ if (layout->list[i].xlator == prev) {
+ layout->list[i].err = op_errno;
+ break;
+ }
+ }
+
local->need_selfheal = 1;
}
}
@@ -2153,31 +2161,18 @@ static int
dht_fill_dict_to_avoid_unlink_of_migrating_file(dict_t *dict)
{
int ret = 0;
- xlator_t *this = NULL;
- char *linktoskip_key = NULL;
- this = THIS;
- GF_VALIDATE_OR_GOTO("dht", this, err);
-
- if (dht_is_tier_xlator(this))
- linktoskip_key = TIER_SKIP_NON_LINKTO_UNLINK;
- else
- linktoskip_key = DHT_SKIP_NON_LINKTO_UNLINK;
-
- ret = dict_set_int32(dict, linktoskip_key, 1);
+ ret = dict_set_int32_sizen(dict, DHT_SKIP_NON_LINKTO_UNLINK, 1);
if (ret)
- goto err;
+ return -1;
- ret = dict_set_int32(dict, DHT_SKIP_OPEN_FD_UNLINK, 1);
+ ret = dict_set_int32_sizen(dict, DHT_SKIP_OPEN_FD_UNLINK, 1);
if (ret)
- goto err;
+ return -1;
return 0;
-
-err:
- return -1;
}
static int32_t
@@ -4306,6 +4301,8 @@ dht_find_local_subvol_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
index = conf->local_subvols_cnt;
uuid_list_copy = gf_strdup(uuid_list);
+ if (!uuid_list_copy)
+ goto unlock;
for (uuid_str = strtok_r(uuid_list, " ", &saveptr); uuid_str;
uuid_str = next_uuid_str) {
@@ -4596,18 +4593,8 @@ dht_getxattr_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
dict_del(xattr, conf->xattr_name);
dict_del(xattr, conf->mds_xattr_key);
- /* filter out following two xattrs that need not
- * be visible on the mount point for geo-rep -
- * trusted.tier.fix.layout.complete and
- * trusted.tier.tier-dht.commithash
- */
-
dict_del(xattr, conf->commithash_xattr_name);
- if (frame->root->pid >= 0 && dht_is_tier_xlator(this)) {
- dict_del(xattr, GF_XATTR_TIER_LAYOUT_FIXED_KEY);
- }
-
if (frame->root->pid >= 0) {
GF_REMOVE_INTERNAL_XATTR("trusted.glusterfs.quota*", xattr);
GF_REMOVE_INTERNAL_XATTR("trusted.pgfid*", xattr);
@@ -5446,11 +5433,13 @@ dht_dir_common_set_remove_xattr(call_frame_t *frame, xlator_t *this, loc_t *loc,
int call_cnt = 0;
dht_local_t *local = NULL;
char gfid_local[GF_UUID_BUF_SIZE] = {0};
+ char **xattrs_to_heal;
conf = this->private;
local = frame->local;
call_cnt = conf->subvolume_cnt;
local->flags = flags;
+ xattrs_to_heal = get_xattrs_to_heal();
if (!gf_uuid_is_null(local->gfid)) {
gf_uuid_unparse(local->gfid, gfid_local);
@@ -5883,22 +5872,7 @@ dht_setxattr(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr,
if (local->rebalance.target_node) {
local->flags = forced_rebalance;
- /* Flag to suggest its a tiering migration
- * The reason for this dic key-value is that
- * promotions and demotions are multithreaded
- * so the original frame from gf_defrag_start()
- * is not carried. A new frame will be created when
- * we do syncop_setxattr(). This does not have the
- * frame->root->pid of the original frame. So we pass
- * this dic key-value when we do syncop_setxattr() to do
- * data migration and set the frame->root->pid to
- * GF_CLIENT_PID_TIER_DEFRAG in dht_setxattr() just before
- * calling dht_start_rebalance_task() */
- tmp = dict_get(xattr, TIERING_MIGRATION_KEY);
- if (tmp)
- frame->root->pid = GF_CLIENT_PID_TIER_DEFRAG;
- else
- frame->root->pid = GF_CLIENT_PID_DEFRAG;
+ frame->root->pid = GF_CLIENT_PID_DEFRAG;
ret = dht_start_rebalance_task(this, frame);
if (!ret)
@@ -6710,10 +6684,9 @@ dht_readdirp_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
layout = local->layout;
- /* We have seen crashes in while running "rm -rf" on tier volumes
- when the layout was NULL on the hot tier. This will skip the
- entries on the subvol without a layout, hence preventing the crash
- but rmdir might fail with "directory not empty" errors*/
+ /* This will skip the entries on the subvol without a layout,
+ * hence preventing the crash but rmdir might fail with
+ * "directory not empty" errors*/
if (layout == NULL)
goto done;
@@ -10840,23 +10813,17 @@ dht_notify(xlator_t *this, int event, void *data, ...)
int had_heard_from_all = 0;
int have_heard_from_all = 0;
- struct timeval time = {
- 0,
- };
gf_defrag_info_t *defrag = NULL;
dict_t *dict = NULL;
gf_defrag_type cmd = 0;
dict_t *output = NULL;
va_list ap;
- dht_methods_t *methods = NULL;
struct gf_upcall *up_data = NULL;
struct gf_upcall_cache_invalidation *up_ci = NULL;
conf = this->private;
GF_VALIDATE_OR_GOTO(this->name, conf, out);
- methods = &(conf->methods);
-
/* had all subvolumes reported status once till now? */
had_heard_from_all = 1;
for (i = 0; i < conf->subvolume_cnt; i++) {
@@ -10886,12 +10853,11 @@ dht_notify(xlator_t *this, int event, void *data, ...)
break;
}
- gettimeofday(&time, NULL);
LOCK(&conf->subvolume_lock);
{
conf->subvolume_status[cnt] = 1;
conf->last_event[cnt] = event;
- conf->subvol_up_time[cnt] = time.tv_sec;
+ conf->subvol_up_time[cnt] = gf_time();
}
UNLOCK(&conf->subvolume_lock);
@@ -10999,21 +10965,13 @@ dht_notify(xlator_t *this, int event, void *data, ...)
if (defrag->is_exiting)
goto unlock;
if ((cmd == GF_DEFRAG_CMD_STATUS) ||
- (cmd == GF_DEFRAG_CMD_STATUS_TIER) ||
(cmd == GF_DEFRAG_CMD_DETACH_STATUS))
gf_defrag_status_get(conf, output);
- else if (cmd == GF_DEFRAG_CMD_START_DETACH_TIER)
- gf_defrag_start_detach_tier(defrag);
else if (cmd == GF_DEFRAG_CMD_DETACH_START)
defrag->cmd = GF_DEFRAG_CMD_DETACH_START;
else if (cmd == GF_DEFRAG_CMD_STOP ||
- cmd == GF_DEFRAG_CMD_STOP_DETACH_TIER ||
cmd == GF_DEFRAG_CMD_DETACH_STOP)
gf_defrag_stop(conf, GF_DEFRAG_STATUS_STOPPED, output);
- else if (cmd == GF_DEFRAG_CMD_PAUSE_TIER)
- ret = gf_defrag_pause_tier(this, defrag);
- else if (cmd == GF_DEFRAG_CMD_RESUME_TIER)
- ret = gf_defrag_resume_tier(this, defrag);
}
unlock:
UNLOCK(&defrag->lock);
@@ -11088,15 +11046,13 @@ dht_notify(xlator_t *this, int event, void *data, ...)
* thread has already started.
*/
if (conf->defrag && !run_defrag) {
- if (methods->migration_needed(this)) {
- run_defrag = 1;
- ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start,
- this, "dhtdg");
- if (ret) {
- GF_FREE(conf->defrag);
- conf->defrag = NULL;
- kill(getpid(), SIGTERM);
- }
+ run_defrag = 1;
+ ret = gf_thread_create(&conf->defrag->th, NULL, gf_defrag_start,
+ this, "dhtdg");
+ if (ret) {
+ GF_FREE(conf->defrag);
+ conf->defrag = NULL;
+ kill(getpid(), SIGTERM);
}
}
}
@@ -11241,28 +11197,6 @@ out:
return ret;
}
-int32_t
-dht_migration_needed(xlator_t *this)
-{
- gf_defrag_info_t *defrag = NULL;
- dht_conf_t *conf = NULL;
- int ret = 0;
-
- conf = this->private;
-
- GF_VALIDATE_OR_GOTO("dht", conf, out);
- GF_VALIDATE_OR_GOTO("dht", conf->defrag, out);
-
- defrag = conf->defrag;
-
- if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) &&
- (defrag->cmd != GF_DEFRAG_CMD_START_DETACH_TIER))
- ret = 1;
-
-out:
- return ret;
-}
-
/*
This function should not be called more then once during a FOP
handling path. It is valid only for for ops on files
@@ -11297,14 +11231,6 @@ dht_set_local_rebalance(xlator_t *this, dht_local_t *local, struct iatt *stbuf,
return 0;
}
-gf_boolean_t
-dht_is_tier_xlator(xlator_t *this)
-{
- if (strcmp(this->type, "cluster/tier") == 0)
- return _gf_true;
- return _gf_false;
-}
-
int32_t
dht_release(xlator_t *this, fd_t *fd)
{
@@ -11444,3 +11370,22 @@ dht_pt_fgetxattr(call_frame_t *frame, xlator_t *this, fd_t *fd, const char *key,
FIRST_CHILD(this)->fops->fgetxattr, fd, key, xdata);
return 0;
}
+
+/* The job of this function is to check if all the xlators have updated
+ * error in the layout. */
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode)
+{
+ dht_layout_t *layout = NULL;
+ int i = 0;
+
+ layout = dht_layout_get(this, inode);
+ for (i = 0; i < layout->cnt; i++) {
+ if (layout->list[i].err == 0) {
+ return 0;
+ }
+ }
+
+ /* Returning the first xlator error as all xlators have errors */
+ return layout->list[0].err;
+}
generated by cgit (git 2.34.1) at 2025-07-01 03:32:03 +0000