summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/dht
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/dht')
-rw-r--r--xlators/cluster/dht/src/dht-common.c181
-rw-r--r--xlators/cluster/dht/src/dht-common.h138
-rw-r--r--xlators/cluster/dht/src/dht-diskusage.c12
-rw-r--r--xlators/cluster/dht/src/dht-helper.c12
-rw-r--r--xlators/cluster/dht/src/dht-inode-write.c32
-rw-r--r--xlators/cluster/dht/src/dht-mem-types.h3
-rw-r--r--xlators/cluster/dht/src/dht-messages.h36
-rw-r--r--xlators/cluster/dht/src/dht-rebalance.c1146
-rw-r--r--xlators/cluster/dht/src/dht-selfheal.c68
-rw-r--r--xlators/cluster/dht/src/dht-shared.c85
-rw-r--r--xlators/cluster/dht/src/nufa.c1
11 files changed, 466 insertions, 1248 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;
+}
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 871b7aed9b3..fe0dc3db34a 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -24,7 +24,6 @@
#define _DHT_H
#define GF_XATTR_FIX_LAYOUT_KEY "distribute.fix.layout"
-#define GF_XATTR_TIER_LAYOUT_FIXED_KEY "trusted.tier.fix.layout.complete"
#define GF_XATTR_FILE_MIGRATE_KEY "trusted.distribute.migrate-data"
#define DHT_MDS_STR "mds"
#define GF_DHT_LOOKUP_UNHASHED_OFF 0
@@ -36,7 +35,6 @@
#define DHT_LAYOUT_HEAL_DOMAIN "dht.layout.heal"
/* Namespace synchronization */
#define DHT_ENTRY_SYNC_DOMAIN "dht.entry.sync"
-#define TIERING_MIGRATION_KEY "tiering.migration"
#define DHT_LAYOUT_HASH_INVALID 1
#define MAX_REBAL_THREADS sysconf(_SC_NPROCESSORS_ONLN)
@@ -52,10 +50,6 @@
#define DHT_DBG_HASHED_SUBVOL_PATTERN "dht.file.hashed-subvol.*"
#define DHT_DBG_HASHED_SUBVOL_KEY "dht.file.hashed-subvol."
-/* Array to hold custom xattr keys
- */
-extern char *xattrs_to_heal[];
-
/* Rebalance nodeuuid flags */
#define REBAL_NODEUUID_MINE 0x01
@@ -246,19 +240,6 @@ typedef gf_boolean_t (*dht_need_heal_t)(call_frame_t *frame,
dht_layout_t **inmem,
dht_layout_t **ondisk);
-typedef struct {
- uint64_t blocks_used;
- uint64_t pblocks_used;
- uint64_t files_used;
- uint64_t pfiles_used;
- uint64_t unhashed_blocks_used;
- uint64_t unhashed_pblocks_used;
- uint64_t unhashed_files_used;
- uint64_t unhashed_pfiles_used;
- uint64_t unhashed_fsid;
- uint64_t hashed_fsid;
-} tier_statvfs_t;
-
struct dht_local {
loc_t loc;
loc_t loc2;
@@ -276,7 +257,6 @@ struct dht_local {
struct iatt preparent;
struct iatt postparent;
struct statvfs statvfs;
- tier_statvfs_t tier_statvfs;
fd_t *fd;
inode_t *inode;
dict_t *params;
@@ -409,14 +389,7 @@ enum gf_defrag_type {
GF_DEFRAG_CMD_STATUS = 1 + 2,
GF_DEFRAG_CMD_START_LAYOUT_FIX = 1 + 3,
GF_DEFRAG_CMD_START_FORCE = 1 + 4,
- GF_DEFRAG_CMD_START_TIER = 1 + 5,
- GF_DEFRAG_CMD_STATUS_TIER = 1 + 6,
- GF_DEFRAG_CMD_START_DETACH_TIER = 1 + 7,
- GF_DEFRAG_CMD_STOP_DETACH_TIER = 1 + 8,
- GF_DEFRAG_CMD_PAUSE_TIER = 1 + 9,
- GF_DEFRAG_CMD_RESUME_TIER = 1 + 10,
GF_DEFRAG_CMD_DETACH_STATUS = 1 + 11,
- GF_DEFRAG_CMD_STOP_TIER = 1 + 12,
GF_DEFRAG_CMD_DETACH_START = 1 + 13,
GF_DEFRAG_CMD_DETACH_COMMIT = 1 + 14,
GF_DEFRAG_CMD_DETACH_COMMIT_FORCE = 1 + 15,
@@ -467,75 +440,6 @@ struct dht_container {
int local_subvol_index;
};
-typedef enum tier_mode_ {
- TIER_MODE_NONE = 0,
- TIER_MODE_TEST,
- TIER_MODE_WM
-} tier_mode_t;
-
-typedef enum tier_pause_state_ {
- TIER_RUNNING = 0,
- TIER_REQUEST_PAUSE,
- TIER_PAUSED
-} tier_pause_state_t;
-
-/* This Structure is only used in tiering fixlayout */
-typedef struct gf_tier_fix_layout_arg {
- xlator_t *this;
- dict_t *fix_layout;
- pthread_t thread_id;
-} gf_tier_fix_layout_arg_t;
-
-typedef struct gf_tier_conf {
- int is_tier;
- int watermark_hi;
- int watermark_low;
- int watermark_last;
- unsigned long block_size;
- fsblkcnt_t blocks_total;
- fsblkcnt_t blocks_used;
- uint64_t max_migrate_bytes;
- int max_migrate_files;
- int query_limit;
- tier_mode_t mode;
- int percent_full;
- /* These flags are only used for tier-compact */
- gf_boolean_t compact_active;
- /* These 3 flags are set to true when the client changes the */
- /* compaction mode on the command line. */
- /* When they are set, the daemon will trigger compaction as */
- /* soon as possible to activate or deactivate compaction. */
- /* If in the middle of a compaction, then the switches take */
- /* effect on the next compaction, not the current one. */
- /* If the user switches it off, we want to avoid needless */
- /* compactions. */
- /* If the user switches it on, they want to compact as soon */
- /* as possible. */
- gf_boolean_t compact_mode_switched;
- gf_boolean_t compact_mode_switched_hot;
- gf_boolean_t compact_mode_switched_cold;
- int tier_max_promote_size;
- int tier_promote_frequency;
- int tier_demote_frequency;
- int tier_compact_hot_frequency;
- int tier_compact_cold_frequency;
- uint64_t st_last_promoted_size;
- uint64_t st_last_demoted_size;
- struct synctask *pause_synctask;
- gf_timer_t *pause_timer;
- pthread_mutex_t pause_mutex;
- int promote_in_progress;
- int demote_in_progress;
- /* This Structure is only used in tiering fixlayout */
- gf_tier_fix_layout_arg_t tier_fix_layout_arg;
- /* Indicates the index of the first queryfile picked
- * in the last cycle of promote or demote */
- int32_t last_promote_qfile_index;
- int32_t last_demote_qfile_index;
- tier_pause_state_t pause_state;
- char volname[GD_VOLUME_NAME_MAX + 1];
-} gf_tier_conf_t;
-
typedef struct nodeuuid_info {
char info; /* Set to 1 is this is my node's uuid*/
uuid_t uuid; /* Store the nodeuuid as well for debugging*/
@@ -563,17 +467,10 @@ struct gf_defrag_info_ {
int cmd;
inode_t *root_inode;
uuid_t node_uuid;
- struct timeval start_time;
+ time_t start_time;
uint32_t new_commit_hash;
gf_defrag_status_t defrag_status;
gf_defrag_pattern_list_t *defrag_pattern;
- gf_tier_conf_t tier_conf;
-
- /*Data Tiering params for scanner*/
- uint64_t total_files_promoted;
- uint64_t total_files_demoted;
- int write_freq_threshold;
- int read_freq_threshold;
pthread_cond_t parallel_migration_cond;
pthread_mutex_t dfq_mutex;
@@ -609,7 +506,6 @@ typedef struct gf_defrag_info_ gf_defrag_info_t;
struct dht_methods_s {
int32_t (*migration_get_dst_subvol)(xlator_t *this, dht_local_t *local);
int32_t (*migration_other)(xlator_t *this, gf_defrag_info_t *defrag);
- int32_t (*migration_needed)(xlator_t *this);
xlator_t *(*layout_search)(xlator_t *this, dht_layout_t *layout,
const char *name);
};
@@ -630,7 +526,7 @@ struct dht_conf {
int subvolume_cnt;
int32_t refresh_interval;
gf_lock_t subvolume_lock;
- struct timeval last_stat_fetch;
+ time_t last_stat_fetch;
gf_lock_t layout_lock;
dict_t *leaf_to_subvol;
void *private; /* Can be used by wrapper xlators over
@@ -752,6 +648,8 @@ struct dir_dfmeta {
struct list_head **head;
struct list_head **iterator;
int *fetch_entries;
+ /* fds corresponding to local subvols only */
+ fd_t **lfd;
};
typedef struct dht_migrate_info {
@@ -1238,24 +1136,6 @@ dht_common_xattrop_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
int
gf_defrag_status_get(dht_conf_t *conf, dict_t *dict);
-void
-gf_defrag_set_pause_state(gf_tier_conf_t *tier_conf, tier_pause_state_t state);
-
-tier_pause_state_t
-gf_defrag_get_pause_state(gf_tier_conf_t *tier_conf);
-
-int
-gf_defrag_pause_tier(xlator_t *this, gf_defrag_info_t *defrag);
-
-tier_pause_state_t
-gf_defrag_check_pause_tier(gf_tier_conf_t *defrag);
-
-int
-gf_defrag_resume_tier(xlator_t *this, gf_defrag_info_t *defrag);
-
-int
-gf_defrag_start_detach_tier(gf_defrag_info_t *defrag);
-
int
gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output);
@@ -1336,9 +1216,6 @@ dht_layout_missing_dirs(dht_layout_t *layout);
int
dht_refresh_layout(call_frame_t *frame);
-gf_boolean_t
-dht_is_tier_xlator(xlator_t *this);
-
int
dht_build_parent_loc(xlator_t *this, loc_t *parent, loc_t *child,
int32_t *op_errno);
@@ -1451,7 +1328,7 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
dict_t *src, int *uret, int *uflag);
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);
int
dht_common_mark_mdsxattr(call_frame_t *frame, int *errst, int flag);
@@ -1499,4 +1376,9 @@ dht_create_lock(call_frame_t *frame, xlator_t *subvol);
int
dht_set_parent_layout_in_dict(loc_t *loc, xlator_t *this, dht_local_t *local);
+int
+dht_dir_layout_error_check(xlator_t *this, inode_t *inode);
+
+int
+dht_inode_ctx_mdsvol_set(inode_t *inode, xlator_t *this, xlator_t *mds_subvol);
#endif /* _DHT_H */
diff --git a/xlators/cluster/dht/src/dht-diskusage.c b/xlators/cluster/dht/src/dht-diskusage.c
index 27097ca2475..c0588828fdb 100644
--- a/xlators/cluster/dht/src/dht-diskusage.c
+++ b/xlators/cluster/dht/src/dht-diskusage.c
@@ -151,22 +151,18 @@ dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
dht_conf_t *conf = NULL;
call_frame_t *statfs_frame = NULL;
dht_local_t *statfs_local = NULL;
- struct timeval tv = {
- 0,
- };
loc_t tmp_loc = {
0,
};
+ time_t now;
conf = this->private;
-
- gettimeofday(&tv, NULL);
-
+ now = gf_time();
/* make it root gfid, should be enough to get the proper
info back */
tmp_loc.gfid[15] = 1;
- if (tv.tv_sec > (conf->refresh_interval + conf->last_stat_fetch.tv_sec)) {
+ if (now > (conf->refresh_interval + conf->last_stat_fetch)) {
statfs_frame = copy_frame(frame);
if (!statfs_frame) {
goto err;
@@ -198,7 +194,7 @@ dht_get_du_info(call_frame_t *frame, xlator_t *this, loc_t *loc)
statfs_local->params);
}
- conf->last_stat_fetch.tv_sec = tv.tv_sec;
+ conf->last_stat_fetch = now;
}
return 0;
err:
diff --git a/xlators/cluster/dht/src/dht-helper.c b/xlators/cluster/dht/src/dht-helper.c
index 73a89399efd..3f2fe43d5f3 100644
--- a/xlators/cluster/dht/src/dht-helper.c
+++ b/xlators/cluster/dht/src/dht-helper.c
@@ -2083,6 +2083,7 @@ dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data)
dht_local_t *local = NULL;
xlator_t *this = NULL;
int ret = -1;
+ int op_errno = 0;
local = heal_frame->local;
main_frame = local->main_frame;
@@ -2092,11 +2093,12 @@ dht_heal_full_path_done(int op_ret, call_frame_t *heal_frame, void *data)
dht_set_fixed_dir_stat(&local->postparent);
if (local->need_xattr_heal) {
local->need_xattr_heal = 0;
- ret = dht_dir_xattr_heal(this, local);
- if (ret)
- gf_smsg(this->name, GF_LOG_ERROR, ret,
+ ret = dht_dir_xattr_heal(this, local, &op_errno);
+ if (ret) {
+ gf_smsg(this->name, GF_LOG_ERROR, op_errno,
DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s", local->loc.path,
NULL);
+ }
}
DHT_STACK_UNWIND(lookup, main_frame, 0, 0, local->inode, &local->stbuf,
@@ -2265,6 +2267,7 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
int luret = -1;
int luflag = -1;
int i = 0;
+ char **xattrs_to_heal;
if (!src || !dst) {
gf_smsg(this->name, GF_LOG_WARNING, EINVAL, DHT_MSG_DST_NULL_SET_FAILED,
@@ -2279,6 +2282,9 @@ dht_dir_set_heal_xattr(xlator_t *this, dht_local_t *local, dict_t *dst,
and set it to dst dict, here index start from 1 because
user xattr already checked in previous statement
*/
+
+ xattrs_to_heal = get_xattrs_to_heal();
+
for (i = 1; xattrs_to_heal[i]; i++) {
keyval = dict_get(src, xattrs_to_heal[i]);
if (keyval) {
diff --git a/xlators/cluster/dht/src/dht-inode-write.c b/xlators/cluster/dht/src/dht-inode-write.c
index eda2491e0ff..2f23ce90fbd 100644
--- a/xlators/cluster/dht/src/dht-inode-write.c
+++ b/xlators/cluster/dht/src/dht-inode-write.c
@@ -93,30 +93,28 @@ dht_writev_cbk(call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
/* Check if the rebalance phase1 is true */
if (IS_DHT_MIGRATION_PHASE1(postbuf)) {
- if (!dht_is_tier_xlator(this)) {
+ if (!local->xattr_req) {
+ local->xattr_req = dict_new();
if (!local->xattr_req) {
- local->xattr_req = dict_new();
- if (!local->xattr_req) {
- gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
- "insufficient memory");
- local->op_errno = ENOMEM;
- local->op_ret = -1;
- goto out;
- }
- }
-
- ret = dict_set_uint32(local->xattr_req,
- GF_PROTECT_FROM_EXTERNAL_WRITES, 1);
- if (ret) {
- gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0,
- "Failed to set key %s in dictionary",
- GF_PROTECT_FROM_EXTERNAL_WRITES);
+ gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_NO_MEMORY, ENOMEM,
+ "insufficient memory");
local->op_errno = ENOMEM;
local->op_ret = -1;
goto out;
}
}
+ ret = dict_set_uint32(local->xattr_req, GF_PROTECT_FROM_EXTERNAL_WRITES,
+ 1);
+ if (ret) {
+ gf_msg(this->name, GF_LOG_ERROR, DHT_MSG_DICT_SET_FAILED, 0,
+ "Failed to set key %s in dictionary",
+ GF_PROTECT_FROM_EXTERNAL_WRITES);
+ local->op_errno = ENOMEM;
+ local->op_ret = -1;
+ goto out;
+ }
+
dht_iatt_merge(this, &local->stbuf, postbuf);
dht_iatt_merge(this, &local->prebuf, prebuf);
diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h
index 33f9832395b..e3c4471334a 100644
--- a/xlators/cluster/dht/src/dht-mem-types.h
+++ b/xlators/cluster/dht/src/dht-mem-types.h
@@ -30,10 +30,7 @@ enum gf_dht_mem_types_ {
gf_dht_mt_container_t,
gf_dht_mt_octx_t,
gf_dht_mt_miginfo_t,
- gf_tier_mt_bricklist_t,
- gf_tier_mt_ipc_ctr_params_t,
gf_dht_mt_fd_ctx_t,
- gf_tier_mt_qfile_array_t,
gf_dht_ret_cache_t,
gf_dht_nodeuuids_t,
gf_dht_mt_end
diff --git a/xlators/cluster/dht/src/dht-messages.h b/xlators/cluster/dht/src/dht-messages.h
index 026879e14af..601f8dad78b 100644
--- a/xlators/cluster/dht/src/dht-messages.h
+++ b/xlators/cluster/dht/src/dht-messages.h
@@ -38,12 +38,11 @@ GLFS_MSGID(
DHT_MSG_REBALANCE_STATUS, DHT_MSG_REBALANCE_STOPPED, DHT_MSG_RENAME_FAILED,
DHT_MSG_SETATTR_FAILED, DHT_MSG_SUBVOL_INSUFF_INODES,
DHT_MSG_SUBVOL_INSUFF_SPACE, DHT_MSG_UNLINK_FAILED,
- DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT, DHT_MSG_LOG_TIER_ERROR,
- DHT_MSG_LOG_TIER_STATUS, DHT_MSG_GET_XATTR_FAILED,
- DHT_MSG_FILE_LOOKUP_FAILED, DHT_MSG_OPEN_FD_FAILED,
- DHT_MSG_SET_INODE_CTX_FAILED, DHT_MSG_UNLOCKING_FAILED,
- DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO, DHT_MSG_CHUNK_SIZE_INFO,
- DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR,
+ DHT_MSG_LAYOUT_SET_FAILED, DHT_MSG_LOG_FIXED_LAYOUT,
+ DHT_MSG_GET_XATTR_FAILED, DHT_MSG_FILE_LOOKUP_FAILED,
+ DHT_MSG_OPEN_FD_FAILED, DHT_MSG_SET_INODE_CTX_FAILED,
+ DHT_MSG_UNLOCKING_FAILED, DHT_MSG_DISK_LAYOUT_NULL, DHT_MSG_SUBVOL_INFO,
+ DHT_MSG_CHUNK_SIZE_INFO, DHT_MSG_LAYOUT_FORM_FAILED, DHT_MSG_SUBVOL_ERROR,
DHT_MSG_LAYOUT_SORT_FAILED, DHT_MSG_REGEX_INFO, DHT_MSG_FOPEN_FAILED,
DHT_MSG_SET_HOSTNAME_FAILED, DHT_MSG_BRICK_ERROR, DHT_MSG_SYNCOP_FAILED,
DHT_MSG_MIGRATE_INFO, DHT_MSG_SOCKET_ERROR, DHT_MSG_CREATE_FD_FAILED,
@@ -69,8 +68,7 @@ GLFS_MSGID(
DHT_MSG_INIT_LOCAL_SUBVOL_FAILED, DHT_MSG_SYS_CALL_GET_TIME_FAILED,
DHT_MSG_NO_DISK_USAGE_STATUS, DHT_MSG_SUBVOL_DOWN_ERROR,
DHT_MSG_REBAL_THROTTLE_INFO, DHT_MSG_COMMIT_HASH_INFO,
- DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_LOG_IPC_TIER_ERROR,
- DHT_MSG_TIER_PAUSED, DHT_MSG_TIER_RESUME, DHT_MSG_SETTLE_HASH_FAILED,
+ DHT_MSG_REBAL_STRUCT_SET, DHT_MSG_HAS_MIGINFO, DHT_MSG_SETTLE_HASH_FAILED,
DHT_MSG_DEFRAG_PROCESS_DIR_FAILED, DHT_MSG_FD_CTX_SET_FAILED,
DHT_MSG_STALE_LOOKUP, DHT_MSG_PARENT_LAYOUT_CHANGED,
DHT_MSG_LOCK_MIGRATION_FAILED, DHT_MSG_LOCK_INODE_UNREF_FAILED,
@@ -96,15 +94,13 @@ GLFS_MSGID(
DHT_MSG_UNLOCK_FILE_FAILED, DHT_MSG_REMOVE_XATTR_FAILED,
DHT_MSG_DATA_MIGRATE_ABORT, DHT_MSG_DEFRAG_NULL, DHT_MSG_PARENT_NULL,
DHT_MSG_GFID_NOT_PRESENT, DHT_MSG_CHILD_LOC_FAILED,
- DHT_MSG_SET_LOOKUP_FAILED, DHT_MSG_DIR_REMOVED,
- DHT_MSG_TIER_FIX_LAYOUT_STARTED, DHT_MSG_FIX_NOT_COMP,
- DHT_MSG_REMOVE_TIER_FAILED, DHT_MSG_SUBVOL_DETER_FAILED,
- DHT_MSG_LOCAL_SUBVOL, DHT_MSG_NODE_UUID, DHT_MSG_SIZE_FILE,
- DHT_MSG_GET_DATA_SIZE_FAILED, DHT_MSG_PTHREAD_JOIN_FAILED,
- DHT_MSG_COUNTER_THREAD_CREATE_FAILED, DHT_MSG_MIGRATION_INIT_QUEUE_FAILED,
- DHT_MSG_PAUSED_TIMEOUT, DHT_MSG_WOKE, DHT_MSG_ABORT_REBALANCE,
- DHT_MSG_CREATE_TASK_REBAL_FAILED, DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL,
- DHT_MSG_MIG_TIER_PAUSED, DHT_MSG_ADD_CHOICES_ERROR,
+ DHT_MSG_SET_LOOKUP_FAILED, DHT_MSG_DIR_REMOVED, DHT_MSG_FIX_NOT_COMP,
+ DHT_MSG_SUBVOL_DETER_FAILED, DHT_MSG_LOCAL_SUBVOL, DHT_MSG_NODE_UUID,
+ DHT_MSG_SIZE_FILE, DHT_MSG_GET_DATA_SIZE_FAILED,
+ DHT_MSG_PTHREAD_JOIN_FAILED, DHT_MSG_COUNTER_THREAD_CREATE_FAILED,
+ DHT_MSG_MIGRATION_INIT_QUEUE_FAILED, DHT_MSG_PAUSED_TIMEOUT, DHT_MSG_WOKE,
+ DHT_MSG_ABORT_REBALANCE, DHT_MSG_CREATE_TASK_REBAL_FAILED,
+ DHT_MSG_REBAL_ESTIMATE_NOT_AVAIL, DHT_MSG_ADD_CHOICES_ERROR,
DHT_MSG_GET_CHOICES_ERROR, DHT_MSG_PREPARE_STATUS_ERROR,
DHT_MSG_SET_CHOICE_FAILED, DHT_MSG_SET_HASHED_SUBVOL_FAILED,
DHT_MSG_XATTR_HEAL_NOT_POSS, DHT_MSG_LINKTO_FILE_FAILED,
@@ -180,7 +176,6 @@ GLFS_MSGID(
"adding bricks"
#define DHT_MSG_NEW_TARGET_FOUND_STR "New target found for file"
#define DHT_MSG_INSUFF_MEMORY_STR "insufficient memory"
-#define DHT_MSG_MIG_TIER_PAUSED_STR "Migrate file paused"
#define DHT_MSG_SET_XATTR_FAILED_STR "failed to set xattr"
#define DHT_MSG_SET_MODE_FAILED_STR "failed to set mode"
#define DHT_MSG_FILE_EXISTS_IN_DEST_STR "file exists in destination"
@@ -222,17 +217,14 @@ GLFS_MSGID(
#define DHT_MSG_GFID_NOT_PRESENT_STR "gfid not present"
#define DHT_MSG_CHILD_LOC_FAILED_STR "Child loc build failed"
#define DHT_MSG_SET_LOOKUP_FAILED_STR "Failed to set lookup"
-#define DHT_MSG_LOG_TIER_STATUS_STR "lookup to cold tier on attach heal failed"
#define DHT_MSG_DIR_LOOKUP_FAILED_STR "lookup failed"
#define DHT_MSG_DIR_REMOVED_STR "Dir renamed or removed. Skipping"
#define DHT_MSG_READDIR_ERROR_STR "readdir failed, Aborting fix-layout"
#define DHT_MSG_SETTLE_HASH_FAILED_STR "Settle hash failed"
#define DHT_MSG_DEFRAG_PROCESS_DIR_FAILED_STR "gf_defrag_process_dir failed"
-#define DHT_MSG_TIER_FIX_LAYOUT_STARTED_STR "Tiering fix layout started"
#define DHT_MSG_FIX_NOT_COMP_STR \
"Unable to retrieve fixlayout xattr. Assume background fix layout not " \
"complete"
-#define DHT_MSG_REMOVE_TIER_FAILED_STR "Failed removing tier fix layout xattr"
#define DHT_MSG_SUBVOL_DETER_FAILED_STR \
"local subvolume determination failed with error"
#define DHT_MSG_LOCAL_SUBVOL_STR "local subvol"
@@ -248,8 +240,6 @@ GLFS_MSGID(
#define DHT_MSG_MIGRATION_INIT_QUEUE_FAILED_STR \
"Failed to initialise migration queue"
#define DHT_MSG_REBALANCE_STOPPED_STR "Received stop command on rebalance"
-#define DHT_MSG_TIER_RESUME_STR "Pause end. Resume tiering"
-#define DHT_MSG_TIER_PAUSED_STR "Pause tiering"
#define DHT_MSG_PAUSED_TIMEOUT_STR "Request pause timer timeout"
#define DHT_MSG_WOKE_STR "woken"
#define DHT_MSG_ABORT_REBALANCE_STR "Aborting rebalance"
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
index abe2afae5dc..8ba8082bd86 100644
--- a/xlators/cluster/dht/src/dht-rebalance.c
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -16,8 +16,8 @@
#include "glusterfs/compat-errno.h" // for ENODATA on BSD
#define GF_DISK_SECTOR_SIZE 512
-#define DHT_REBALANCE_PID 4242 /* Change it if required */
-#define DHT_REBALANCE_BLKSIZE (1024 * 1024) /* 1 MB */
+#define DHT_REBALANCE_PID 4242 /* Change it if required */
+#define DHT_REBALANCE_BLKSIZE 1048576 /* 1 MB */
#define MAX_MIGRATE_QUEUE_COUNT 500
#define MIN_MIGRATE_QUEUE_COUNT 200
#define MAX_REBAL_TYPE_SIZE 16
@@ -45,7 +45,10 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
if (meta) {
for (i = 0; i < local_subvols_cnt; i++) {
- gf_dirent_free(&meta->equeue[i]);
+ if (meta->equeue)
+ gf_dirent_free(&meta->equeue[i]);
+ if (meta->lfd && meta->lfd[i])
+ fd_unref(meta->lfd[i]);
}
GF_FREE(meta->equeue);
@@ -53,6 +56,7 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
GF_FREE(meta->iterator);
GF_FREE(meta->offset_var);
GF_FREE(meta->fetch_entries);
+ GF_FREE(meta->lfd);
GF_FREE(meta);
}
}
@@ -84,26 +88,6 @@ dht_set_global_defrag_error(gf_defrag_info_t *defrag, int ret)
return;
}
-static gf_boolean_t
-dht_is_tier_command(int cmd)
-{
- gf_boolean_t is_tier = _gf_false;
-
- switch (cmd) {
- case GF_DEFRAG_CMD_START_TIER:
- case GF_DEFRAG_CMD_STATUS_TIER:
- case GF_DEFRAG_CMD_START_DETACH_TIER:
- case GF_DEFRAG_CMD_STOP_DETACH_TIER:
- case GF_DEFRAG_CMD_PAUSE_TIER:
- case GF_DEFRAG_CMD_RESUME_TIER:
- is_tier = _gf_true;
- break;
- default:
- break;
- }
- return is_tier;
-}
-
static int
dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status)
{
@@ -112,8 +96,6 @@ dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status)
char *tmpstr = NULL;
char *ptr = NULL;
char *suffix = "-dht";
- dht_conf_t *conf = NULL;
- gf_defrag_info_t *defrag = NULL;
int len = 0;
eventtypes_t event = EVENT_LAST;
@@ -132,21 +114,14 @@ dht_send_rebalance_event(xlator_t *this, int cmd, gf_defrag_status_t status)
break;
}
- if (dht_is_tier_command(cmd)) {
- /* We should have the tier volume name*/
- conf = this->private;
- defrag = conf->defrag;
- volname = defrag->tier_conf.volname;
- } else {
- /* DHT volume */
- len = strlen(this->name) - strlen(suffix);
- tmpstr = gf_strdup(this->name);
- if (tmpstr) {
- ptr = tmpstr + len;
- if (!strcmp(ptr, suffix)) {
- tmpstr[len] = '\0';
- volname = tmpstr;
- }
+ /* DHT volume */
+ len = strlen(this->name) - strlen(suffix);
+ tmpstr = gf_strdup(this->name);
+ if (tmpstr) {
+ ptr = tmpstr + len;
+ if (!strcmp(ptr, suffix)) {
+ tmpstr[len] = '\0';
+ volname = tmpstr;
}
}
@@ -172,75 +147,6 @@ dht_strip_out_acls(dict_t *dict)
}
}
-static int
-dht_write_with_holes(xlator_t *to, fd_t *fd, struct iovec *vec, int count,
- int32_t size, off_t offset, struct iobref *iobref,
- int *fop_errno)
-{
- int i = 0;
- int ret = -1;
- int start_idx = 0;
- int tmp_offset = 0;
- int write_needed = 0;
- int buf_len = 0;
- int size_pending = 0;
- char *buf = NULL;
-
- /* loop through each vector */
- for (i = 0; i < count; i++) {
- buf = vec[i].iov_base;
- buf_len = vec[i].iov_len;
-
- for (start_idx = 0; (start_idx + GF_DISK_SECTOR_SIZE) <= buf_len;
- start_idx += GF_DISK_SECTOR_SIZE) {
- if (mem_0filled(buf + start_idx, GF_DISK_SECTOR_SIZE) != 0) {
- write_needed = 1;
- continue;
- }
-
- if (write_needed) {
- ret = syncop_write(
- to, fd, (buf + tmp_offset), (start_idx - tmp_offset),
- (offset + tmp_offset), iobref, 0, NULL, NULL);
- /* 'path' will be logged in calling function */
- if (ret < 0) {
- gf_log(THIS->name, GF_LOG_WARNING, "failed to write (%s)",
- strerror(-ret));
- *fop_errno = -ret;
- ret = -1;
- goto out;
- }
-
- write_needed = 0;
- }
- tmp_offset = start_idx + GF_DISK_SECTOR_SIZE;
- }
-
- if ((start_idx < buf_len) || write_needed) {
- /* This means, last chunk is not yet written.. write it */
- ret = syncop_write(to, fd, (buf + tmp_offset),
- (buf_len - tmp_offset), (offset + tmp_offset),
- iobref, 0, NULL, NULL);
- if (ret < 0) {
- /* 'path' will be logged in calling function */
- gf_log(THIS->name, GF_LOG_WARNING, "failed to write (%s)",
- strerror(-ret));
- *fop_errno = -ret;
- ret = -1;
- goto out;
- }
- }
-
- size_pending = (size - buf_len);
- if (!size_pending)
- break;
- }
-
- ret = size;
-out:
- return ret;
-}
-
/*
return values:
-1 : failure
@@ -648,7 +554,7 @@ out:
static int
__dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
loc_t *loc, struct iatt *stbuf, fd_t **dst_fd,
- int *fop_errno)
+ int *fop_errno, int file_has_holes)
{
int ret = -1;
int ret2 = -1;
@@ -703,26 +609,23 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
goto out;
}
- if (!!dht_is_tier_xlator(this)) {
- xdata = dict_new();
- if (!xdata) {
- *fop_errno = ENOMEM;
- ret = -1;
- gf_msg(this->name, GF_LOG_ERROR, ENOMEM,
- DHT_MSG_MIGRATE_FILE_FAILED, "%s: dict_new failed)",
- loc->path);
- goto out;
- }
+ xdata = dict_new();
+ if (!xdata) {
+ *fop_errno = ENOMEM;
+ ret = -1;
+ gf_msg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_MIGRATE_FILE_FAILED,
+ "%s: dict_new failed)", loc->path);
+ goto out;
+ }
- ret = dict_set_int32(xdata, GF_CLEAN_WRITE_PROTECTION, 1);
- if (ret) {
- *fop_errno = ENOMEM;
- ret = -1;
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
- "%s: failed to set dictionary value: key = %s ", loc->path,
- GF_CLEAN_WRITE_PROTECTION);
- goto out;
- }
+ ret = dict_set_int32_sizen(xdata, GF_CLEAN_WRITE_PROTECTION, 1);
+ if (ret) {
+ *fop_errno = ENOMEM;
+ ret = -1;
+ gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_DICT_SET_FAILED,
+ "%s: failed to set dictionary value: key = %s ", loc->path,
+ GF_CLEAN_WRITE_PROTECTION);
+ goto out;
}
ret = syncop_lookup(to, loc, &new_stbuf, NULL, xdata, NULL);
@@ -817,7 +720,7 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
/* No need to bother about 0 byte size files */
if (stbuf->ia_size > 0) {
- if (conf->use_fallocate) {
+ if (conf->use_fallocate && !file_has_holes) {
ret = syncop_fallocate(to, fd, 0, 0, stbuf->ia_size, NULL, NULL);
if (ret < 0) {
if (ret == -EOPNOTSUPP || ret == -EINVAL || ret == -ENOSYS) {
@@ -844,9 +747,7 @@ __dht_rebalance_create_dst_file(xlator_t *this, xlator_t *to, xlator_t *from,
goto out;
}
}
- }
-
- if (!conf->use_fallocate) {
+ } else {
ret = syncop_ftruncate(to, fd, stbuf->ia_size, NULL, NULL, NULL,
NULL);
if (ret < 0) {
@@ -1097,22 +998,90 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag,
int ret = 0;
int count = 0;
off_t offset = 0;
+ off_t data_offset = 0;
+ off_t hole_offset = 0;
struct iovec *vector = NULL;
struct iobref *iobref = NULL;
uint64_t total = 0;
size_t read_size = 0;
+ size_t data_block_size = 0;
dict_t *xdata = NULL;
dht_conf_t *conf = NULL;
conf = this->private;
+
/* if file size is '0', no need to enter this loop */
while (total < ia_size) {
- read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE)
- ? DHT_REBALANCE_BLKSIZE
- : (ia_size - total));
+ /* This is a regular file - read it sequentially */
+ if (!hole_exists) {
+ read_size = (((ia_size - total) > DHT_REBALANCE_BLKSIZE)
+ ? DHT_REBALANCE_BLKSIZE
+ : (ia_size - total));
+ } else {
+ /* This is a sparse file - read only the data segments in the file
+ */
+
+ /* If the previous data block is fully copied, find the next data
+ * segment
+ * starting at the offset of the last read and written byte, */
+ if (data_block_size <= 0) {
+ ret = syncop_seek(from, src, offset, GF_SEEK_DATA, NULL,
+ &data_offset);
+ if (ret) {
+ if (ret == -ENXIO)
+ ret = 0; /* No more data segments */
+ else
+ *fop_errno = -ret; /* Error occurred */
+
+ break;
+ }
+
+ /* If the position of the current data segment is greater than
+ * the position of the next hole, find the next hole in order to
+ * calculate the length of the new data segment */
+ if (data_offset > hole_offset) {
+ /* Starting at the offset of the last data segment, find the
+ * next hole */
+ ret = syncop_seek(from, src, data_offset, GF_SEEK_HOLE,
+ NULL, &hole_offset);
+ if (ret) {
+ /* If an error occurred here it's a real error because
+ * if the seek for a data segment was successful then
+ * necessarily another hole must exist (EOF is a hole)
+ */
+ *fop_errno = -ret;
+ break;
+ }
+
+ /* Calculate the total size of the current data block */
+ data_block_size = hole_offset - data_offset;
+ }
+ } else {
+ /* There is still data in the current segment, move the
+ * data_offset to the position of the last written byte */
+ data_offset = offset;
+ }
+
+ /* Calculate how much data needs to be read and written. If the data
+ * segment's length is bigger than DHT_REBALANCE_BLKSIZE, read and
+ * write DHT_REBALANCE_BLKSIZE data length and the rest in the
+ * next iteration(s) */
+ read_size = ((data_block_size > DHT_REBALANCE_BLKSIZE)
+ ? DHT_REBALANCE_BLKSIZE
+ : data_block_size);
+
+ /* Calculate the remaining size of the data block - maybe there's no
+ * need to seek for data in the next iteration */
+ data_block_size -= read_size;
+
+ /* Set offset to the offset of the data segment so read and write
+ * will have the correct position */
+ offset = data_offset;
+ }
ret = syncop_readv(from, src, read_size, offset, 0, &vector, &count,
&iobref, NULL, NULL, NULL);
+
if (!ret || (ret < 0)) {
if (!ret) {
/* File was probably truncated*/
@@ -1124,57 +1093,42 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag,
break;
}
- if (hole_exists) {
- ret = dht_write_with_holes(to, dst, vector, count, ret, offset,
- iobref, fop_errno);
- } else {
- if (!conf->force_migration && !dht_is_tier_xlator(this)) {
+ if (!conf->force_migration) {
+ if (!xdata) {
+ xdata = dict_new();
if (!xdata) {
- xdata = dict_new();
- if (!xdata) {
- gf_msg("dht", GF_LOG_ERROR, 0,
- DHT_MSG_MIGRATE_FILE_FAILED,
- "insufficient memory");
- ret = -1;
- *fop_errno = ENOMEM;
- break;
- }
+ gf_msg("dht", GF_LOG_ERROR, 0, DHT_MSG_MIGRATE_FILE_FAILED,
+ "insufficient memory");
+ ret = -1;
+ *fop_errno = ENOMEM;
+ break;
+ }
- /* Fail this write and abort rebalance if we
- * detect a write from client since migration of
- * this file started. This is done to avoid
- * potential data corruption due to out of order
- * writes from rebalance and client to the same
- * region (as compared between src and dst
- * files). See
- * https://github.com/gluster/glusterfs/issues/308
- * for more details.
- */
- ret = dict_set_int32(xdata, GF_AVOID_OVERWRITE, 1);
- if (ret) {
- gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM,
- "failed to set dict");
- ret = -1;
- *fop_errno = ENOMEM;
- break;
- }
+ /* Fail this write and abort rebalance if we
+ * detect a write from client since migration of
+ * this file started. This is done to avoid
+ * potential data corruption due to out of order
+ * writes from rebalance and client to the same
+ * region (as compared between src and dst
+ * files). See
+ * https://github.com/gluster/glusterfs/issues/308
+ * for more details.
+ */
+ ret = dict_set_int32_sizen(xdata, GF_AVOID_OVERWRITE, 1);
+ if (ret) {
+ gf_msg("dht", GF_LOG_ERROR, 0, ENOMEM,
+ "failed to set dict");
+ ret = -1;
+ *fop_errno = ENOMEM;
+ break;
}
}
- ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL,
- NULL, xdata, NULL);
- if (ret < 0) {
- *fop_errno = -ret;
- }
- }
-
- if ((defrag && defrag->cmd == GF_DEFRAG_CMD_START_TIER) &&
- (gf_defrag_get_pause_state(&defrag->tier_conf) != TIER_RUNNING)) {
- gf_msg("tier", GF_LOG_INFO, 0, DHT_MSG_TIER_PAUSED,
- "Migrate file paused");
- ret = -1;
}
+ ret = syncop_writev(to, dst, vector, count, offset, iobref, 0, NULL,
+ NULL, xdata, NULL);
if (ret < 0) {
+ *fop_errno = -ret;
break;
}
@@ -1568,6 +1522,7 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
xlator_t *old_target = NULL;
xlator_t *hashed_subvol = NULL;
fd_t *linkto_fd = NULL;
+ dict_t *xdata = NULL;
if (from == to) {
gf_msg_debug(this->name, 0,
@@ -1578,21 +1533,6 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
goto out;
}
- /* If defrag is NULL, it should be assumed that migration is triggered
- * from client using the trusted.distribute.migrate-data virtual xattr
- */
- defrag = conf->defrag;
-
- /* migration of files from clients is restricted to non-tiered clients
- * for now */
- if (!defrag && dht_is_tier_xlator(this)) {
- ret = ENOTSUP;
- goto out;
- }
-
- if (defrag && defrag->tier_conf.is_tier)
- log_level = GF_LOG_TRACE;
-
gf_log(this->name, log_level, "%s: attempting to move from %s to %s",
loc->path, from->name, to->name);
@@ -1739,9 +1679,13 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
goto out;
}
+ /* Try to preserve 'holes' while migrating data */
+ if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
+ file_has_holes = 1;
+
/* create the destination, with required modes/xattr */
ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf, &dst_fd,
- fop_errno);
+ fop_errno, file_has_holes);
if (ret) {
gf_msg(this->name, GF_LOG_ERROR, 0, 0,
"Create dst failed"
@@ -1785,8 +1729,8 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
* destination. We need to do update this only post migration
* as in case of failure the linkto needs to point to the source
* subvol */
- ret = __dht_rebalance_create_dst_file(this, to, from, loc, &stbuf,
- &dst_fd, fop_errno);
+ ret = __dht_rebalance_create_dst_file(
+ this, to, from, loc, &stbuf, &dst_fd, fop_errno, file_has_holes);
if (ret) {
gf_log(this->name, GF_LOG_ERROR,
"Create dst failed"
@@ -1873,9 +1817,6 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
ret = 0;
goto out;
}
- /* Try to preserve 'holes' while migrating data */
- if (stbuf.ia_size > (stbuf.ia_blocks * GF_DISK_SECTOR_SIZE))
- file_has_holes = 1;
ret = __dht_rebalance_migrate_data(this, defrag, from, to, src_fd, dst_fd,
stbuf.ia_size, file_has_holes,
@@ -1890,7 +1831,15 @@ dht_migrate_file(xlator_t *this, loc_t *loc, xlator_t *from, xlator_t *to,
/* TODO: Sync the locks */
- ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, NULL, NULL);
+ xdata = dict_new();
+ if (!xdata || dict_set_int8(xdata, "last-fsync", 1)) {
+ gf_log(this->name, GF_LOG_ERROR,
+ "%s: failed to set last-fsync flag on "
+ "%s (%s)",
+ loc->path, to->name, strerror(ENOMEM));
+ }
+
+ ret = syncop_fsync(to, dst_fd, 0, NULL, NULL, xdata, NULL);
if (ret) {
gf_log(this->name, GF_LOG_WARNING, "%s: failed to fsync on %s (%s)",
loc->path, to->name, strerror(-ret));
@@ -2333,14 +2282,12 @@ out:
}
}
- if (!dht_is_tier_xlator(this)) {
- lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES,
- NULL, NULL);
- if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) {
- gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0,
- "%s: removexattr failed key %s", loc->path,
- GF_PROTECT_FROM_EXTERNAL_WRITES);
- }
+ lk_ret = syncop_removexattr(to, loc, GF_PROTECT_FROM_EXTERNAL_WRITES, NULL,
+ NULL);
+ if (lk_ret && (lk_ret != -ENODATA) && (lk_ret != -ENOATTR)) {
+ gf_msg(this->name, GF_LOG_WARNING, -lk_ret, 0,
+ "%s: removexattr failed key %s", loc->path,
+ GF_PROTECT_FROM_EXTERNAL_WRITES);
}
if (dict)
@@ -2353,11 +2300,15 @@ out:
if (dst_fd)
syncop_close(dst_fd);
+
if (src_fd)
syncop_close(src_fd);
if (linkto_fd)
syncop_close(linkto_fd);
+ if (xdata)
+ dict_unref(xdata);
+
loc_wipe(&tmp_loc);
loc_wipe(&parent_loc);
@@ -2587,10 +2538,10 @@ out:
* all hardlinks.
*/
-int
+gf_boolean_t
gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid)
{
- int ret = 0;
+ gf_boolean_t ret = _gf_false;
int i = local_subvol_index;
char *str = NULL;
uint32_t hashval = 0;
@@ -2612,12 +2563,11 @@ gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid)
}
str = uuid_utoa_r(gfid, buf);
- ret = dht_hash_compute(this, 0, str, &hashval);
- if (ret == 0) {
+ if (dht_hash_compute(this, 0, str, &hashval) == 0) {
index = (hashval % entry->count);
if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
/* Index matches this node's nodeuuid.*/
- ret = 1;
+ ret = _gf_true;
goto out;
}
@@ -2630,12 +2580,12 @@ gf_defrag_should_i_migrate(xlator_t *this, int local_subvol_index, uuid_t gfid)
/* None of the bricks in the subvol are up.
* CHILD_DOWN will kill the process soon */
- return 0;
+ return _gf_false;
}
if (entry->elements[index].info == REBAL_NODEUUID_MINE) {
/* Index matches this node's nodeuuid.*/
- ret = 1;
+ ret = _gf_true;
goto out;
}
}
@@ -2684,6 +2634,7 @@ gf_defrag_migrate_single_file(void *opaque)
struct iatt *iatt_ptr = NULL;
gf_boolean_t update_skippedcount = _gf_true;
int i = 0;
+ gf_boolean_t should_i_migrate = 0;
rebal_entry = (struct dht_container *)opaque;
if (!rebal_entry) {
@@ -2738,11 +2689,29 @@ gf_defrag_migrate_single_file(void *opaque)
goto out;
}
+ should_i_migrate = gf_defrag_should_i_migrate(
+ this, rebal_entry->local_subvol_index, entry->d_stat.ia_gfid);
+
gf_uuid_copy(entry_loc.gfid, entry->d_stat.ia_gfid);
gf_uuid_copy(entry_loc.pargfid, loc->gfid);
ret = syncop_lookup(this, &entry_loc, &iatt, NULL, NULL, NULL);
+
+ if (!should_i_migrate) {
+ /* this node isn't supposed to migrate the file. suppressing any
+ * potential error from lookup as this file is under migration by
+ * another node */
+ if (ret) {
+ gf_msg_debug(this->name, -ret,
+ "Ignoring lookup failure: node isn't migrating %s",
+ entry_loc.path);
+ ret = 0;
+ }
+ gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path);
+ goto out;
+ }
+
if (ret) {
gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_MIGRATE_FILE_FAILED,
"Migrate file failed: %s lookup failed", entry_loc.path);
@@ -2763,12 +2732,6 @@ gf_defrag_migrate_single_file(void *opaque)
goto out;
}
- if (!gf_defrag_should_i_migrate(this, rebal_entry->local_subvol_index,
- entry->d_stat.ia_gfid)) {
- gf_msg_debug(this->name, 0, "Don't migrate %s ", entry_loc.path);
- goto out;
- }
-
iatt_ptr = &iatt;
hashed_subvol = dht_subvol_get_hashed(this, &entry_loc);
@@ -2911,8 +2874,7 @@ gf_defrag_migrate_single_file(void *opaque)
if (defrag->stats == _gf_true) {
gettimeofday(&end, NULL);
- elapsed = (end.tv_sec - start.tv_sec) * 1e6 +
- (end.tv_usec - start.tv_usec);
+ elapsed = gf_tvdiff(&start, &end);
gf_log(this->name, GF_LOG_INFO,
"Migration of "
"file:%s size:%" PRIu64
@@ -3091,9 +3053,9 @@ int static gf_defrag_get_entry(xlator_t *this, int i,
dht_conf_t *conf, gf_defrag_info_t *defrag,
fd_t *fd, dict_t *migrate_data,
struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req,
- int *should_commit_hash, int *perrno)
+ int *perrno)
{
- int ret = -1;
+ int ret = 0;
char is_linkfile = 0;
gf_dirent_t *df_entry = NULL;
struct dht_container *tmp_container = NULL;
@@ -3109,6 +3071,13 @@ int static gf_defrag_get_entry(xlator_t *this, int i,
}
if (dir_dfmeta->fetch_entries[i] == 1) {
+ if (!fd) {
+ dir_dfmeta->fetch_entries[i] = 0;
+ dir_dfmeta->offset_var[i].readdir_done = 1;
+ ret = 0;
+ goto out;
+ }
+
ret = syncop_readdirp(conf->local_subvols[i], fd, 131072,
dir_dfmeta->offset_var[i].offset,
&(dir_dfmeta->equeue[i]), xattr_req, NULL);
@@ -3268,7 +3237,6 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
dict_t *migrate_data, int *perrno)
{
int ret = -1;
- fd_t *fd = NULL;
dht_conf_t *conf = NULL;
gf_dirent_t entries;
dict_t *xattr_req = NULL;
@@ -3289,7 +3257,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
int dfc_index = 0;
int throttle_up = 0;
struct dir_dfmeta *dir_dfmeta = NULL;
- int should_commit_hash = 1;
+ xlator_t *old_THIS = NULL;
gf_log(this->name, GF_LOG_INFO, "migrate data called on %s", loc->path);
gettimeofday(&dir_start, NULL);
@@ -3302,28 +3270,53 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
goto out;
}
- fd = fd_create(loc->inode, defrag->pid);
- if (!fd) {
- gf_log(this->name, GF_LOG_ERROR, "Failed to create fd");
+ old_THIS = THIS;
+ THIS = this;
+
+ dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
+ if (!dir_dfmeta) {
+ gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
ret = -1;
goto out;
}
- ret = syncop_opendir(this, loc, fd, NULL, NULL);
- if (ret) {
- gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_DATA_FAILED,
- "Migrate data failed: Failed to open dir %s", loc->path);
- *perrno = -ret;
+ dir_dfmeta->lfd = GF_CALLOC(local_subvols_cnt, sizeof(fd_t *),
+ gf_common_mt_pointer);
+ if (!dir_dfmeta->lfd) {
+ gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+ "for dir_dfmeta", NULL);
ret = -1;
+ *perrno = ENOMEM;
goto out;
}
- fd_bind(fd);
- dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
- if (!dir_dfmeta) {
- gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
- ret = -1;
- goto out;
+ for (i = 0; i < local_subvols_cnt; i++) {
+ dir_dfmeta->lfd[i] = fd_create(loc->inode, defrag->pid);
+ if (!dir_dfmeta->lfd[i]) {
+ gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_FD_CREATE_FAILED,
+ NULL);
+ *perrno = ENOMEM;
+ ret = -1;
+ goto out;
+ }
+
+ ret = syncop_opendir(conf->local_subvols[i], loc, dir_dfmeta->lfd[i],
+ NULL, NULL);
+ if (ret) {
+ fd_unref(dir_dfmeta->lfd[i]);
+ dir_dfmeta->lfd[i] = NULL;
+ gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_FAILED_TO_OPEN,
+ "dir: %s", loc->path, "subvol: %s",
+ conf->local_subvols[i]->name, NULL);
+
+ if (conf->decommission_in_progress) {
+ *perrno = -ret;
+ ret = -1;
+ goto out;
+ }
+ } else {
+ fd_bind(dir_dfmeta->lfd[i]);
+ }
}
dir_dfmeta->head = GF_CALLOC(local_subvols_cnt, sizeof(*(dir_dfmeta->head)),
@@ -3358,6 +3351,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
ret = -1;
goto out;
}
+
ret = gf_defrag_ctx_subvols_init(dir_dfmeta->offset_var, this);
if (ret) {
gf_log(this->name, GF_LOG_ERROR,
@@ -3370,7 +3364,8 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
dir_dfmeta->fetch_entries = GF_CALLOC(local_subvols_cnt, sizeof(int),
gf_common_mt_int);
if (!dir_dfmeta->fetch_entries) {
- gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->fetch_entries is NULL");
+ gf_smsg(this->name, GF_LOG_ERROR, ENOMEM, DHT_MSG_INSUFF_MEMORY,
+ "for dir_dfmeta->fetch_entries", NULL);
ret = -1;
goto out;
}
@@ -3440,8 +3435,9 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
ldfq_count <= MAX_MIGRATE_QUEUE_COUNT &&
!dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf,
- defrag, fd, migrate_data, dir_dfmeta,
- xattr_req, &should_commit_hash, perrno);
+ defrag, dir_dfmeta->lfd[dfc_index],
+ migrate_data, dir_dfmeta, xattr_req,
+ perrno);
if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) {
goto out;
@@ -3485,27 +3481,19 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
}
gettimeofday(&end, NULL);
- elapsed = (end.tv_sec - dir_start.tv_sec) * 1e6 +
- (end.tv_usec - dir_start.tv_usec);
+ elapsed = gf_tvdiff(&dir_start, &end);
gf_log(this->name, GF_LOG_INFO,
"Migration operation on dir %s took "
"%.2f secs",
loc->path, elapsed / 1e6);
ret = 0;
out:
-
+ THIS = old_THIS;
gf_defrag_free_dir_dfmeta(dir_dfmeta, local_subvols_cnt);
if (xattr_req)
dict_unref(xattr_req);
- if (fd)
- fd_unref(fd);
-
- if (ret == 0 && should_commit_hash == 0) {
- ret = 2;
- }
-
/* It does not matter if it errored out - this number is
* used to calculate rebalance estimated time to complete.
* No locking required as dirs are processed by a single thread.
@@ -3513,6 +3501,7 @@ out:
defrag->num_dirs_processed++;
return ret;
}
+
int
gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
dict_t *fix_layout)
@@ -3527,7 +3516,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
* rebalance is complete.
*/
if (defrag->cmd == GF_DEFRAG_CMD_START_LAYOUT_FIX ||
- defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER ||
defrag->cmd == GF_DEFRAG_CMD_DETACH_START) {
return 0;
}
@@ -3573,114 +3561,6 @@ gf_defrag_settle_hash(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
return 0;
}
-/* Function for doing a named lookup on file inodes during an attach tier
- * So that a hardlink lookup heal i.e gfid to parent gfid lookup heal
- * happens on pre-existing data. This is required so that the ctr database has
- * hardlinks of all the exisitng file in the volume. CTR xlator on the
- * brick/server side does db update/insert of the hardlink on a namelookup.
- * Currently the namedlookup is done synchronous to the fixlayout that is
- * triggered by attach tier. This is not performant, adding more time to
- * fixlayout. The performant approach is record the hardlinks on a compressed
- * datastore and then do the namelookup asynchronously later, giving the ctr db
- * eventual consistency
- * */
-int
-gf_fix_layout_tier_attach_lookup(xlator_t *this, loc_t *parent_loc,
- gf_dirent_t *file_dentry)
-{
- int ret = -1;
- dict_t *lookup_xdata = NULL;
- dht_conf_t *conf = NULL;
- loc_t file_loc = {
- 0,
- };
- struct iatt iatt = {
- 0,
- };
-
- GF_VALIDATE_OR_GOTO("tier", this, out);
-
- GF_VALIDATE_OR_GOTO(this->name, parent_loc, out);
-
- GF_VALIDATE_OR_GOTO(this->name, file_dentry, out);
-
- GF_VALIDATE_OR_GOTO(this->name, this->private, out);
-
- if (!parent_loc->inode) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "%s/%s parent is NULL", parent_loc->path, file_dentry->d_name);
- goto out;
- }
-
- conf = this->private;
-
- loc_wipe(&file_loc);
-
- if (gf_uuid_is_null(file_dentry->d_stat.ia_gfid)) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "%s/%s gfid not present", parent_loc->path, file_dentry->d_name);
- goto out;
- }
-
- gf_uuid_copy(file_loc.gfid, file_dentry->d_stat.ia_gfid);
-
- if (gf_uuid_is_null(parent_loc->gfid)) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "%s/%s"
- " gfid not present",
- parent_loc->path, file_dentry->d_name);
- goto out;
- }
-
- gf_uuid_copy(file_loc.pargfid, parent_loc->gfid);
-
- ret = dht_build_child_loc(this, &file_loc, parent_loc, file_dentry->d_name);
- if (ret) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "Child loc build failed");
- ret = -1;
- goto out;
- }
-
- lookup_xdata = dict_new();
- if (!lookup_xdata) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "Failed creating lookup dict for %s", file_dentry->d_name);
- goto out;
- }
-
- ret = dict_set_int32(lookup_xdata, CTR_ATTACH_TIER_LOOKUP, 1);
- if (ret) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LOG_TIER_ERROR,
- "Failed to set lookup flag");
- goto out;
- }
-
- gf_uuid_copy(file_loc.parent->gfid, parent_loc->gfid);
-
- /* Sending lookup to cold tier only */
- ret = syncop_lookup(conf->subvolumes[0], &file_loc, &iatt, NULL,
- lookup_xdata, NULL);
- if (ret) {
- /* If the file does not exist on the cold tier than it must */
- /* have been discovered on the hot tier. This is not an error. */
- gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS,
- "%s lookup to cold tier on attach heal failed", file_loc.path);
- goto out;
- }
-
- ret = 0;
-
-out:
-
- loc_wipe(&file_loc);
-
- if (lookup_xdata)
- dict_unref(lookup_xdata);
-
- return ret;
-}
-
int
gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
dict_t *fix_layout, dict_t *migrate_data)
@@ -3700,7 +3580,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
};
inode_t *linked_inode = NULL, *inode = NULL;
dht_conf_t *conf = NULL;
- int should_commit_hash = 1;
int perrno = 0;
conf = this->private;
@@ -3803,16 +3682,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
continue;
if (!IA_ISDIR(entry->d_stat.ia_type)) {
- /* If its a fix layout during the attach
- * tier operation do lookups on files
- * on cold subvolume so that there is a
- * CTR DB Lookup Heal triggered on existing
- * data.
- * */
- if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) {
- gf_fix_layout_tier_attach_lookup(this, loc, entry);
- }
-
continue;
}
loc_wipe(&entry_loc);
@@ -3829,8 +3698,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
goto out;
} else {
- should_commit_hash = 0;
-
continue;
}
}
@@ -3893,7 +3760,6 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
ret = -1;
goto out;
} else {
- should_commit_hash = 0;
continue;
}
}
@@ -3906,11 +3772,12 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
ret = gf_defrag_fix_layout(this, defrag, &entry_loc, fix_layout,
migrate_data);
- if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) {
+ if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED ||
+ defrag->defrag_status == GF_DEFRAG_STATUS_FAILED) {
goto out;
}
- if (ret && ret != 2) {
+ if (ret) {
gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_LAYOUT_FIX_FAILED,
"Fix layout failed for %s", entry_loc.path);
@@ -3941,6 +3808,17 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
*/
ret = syncop_setxattr(this, loc, fix_layout, 0, NULL, NULL);
+
+ /* In case of a race where the directory is deleted just before
+ * layout setxattr, the errors are updated in the layout structure.
+ * We can use this information to make a decision whether the directory
+ * is deleted entirely.
+ */
+ if (ret == 0) {
+ ret = dht_dir_layout_error_check(this, loc->inode);
+ ret = -ret;
+ }
+
if (ret) {
if (-ret == ENOENT || -ret == ESTALE) {
gf_msg(this->name, GF_LOG_INFO, -ret, DHT_MSG_LAYOUT_FIX_FAILED,
@@ -3966,11 +3844,10 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
}
}
- if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) &&
- (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX)) {
+ if (defrag->cmd != GF_DEFRAG_CMD_START_LAYOUT_FIX) {
ret = gf_defrag_process_dir(this, defrag, loc, migrate_data, &perrno);
- if (ret && (ret != 2)) {
+ if (ret) {
if (perrno == ENOENT || perrno == ESTALE) {
ret = 0;
goto out;
@@ -3986,18 +3863,13 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
if (conf->decommission_in_progress) {
goto out;
}
-
- should_commit_hash = 0;
}
- } else if (ret == 2) {
- should_commit_hash = 0;
}
}
gf_msg_trace(this->name, 0, "fix layout called on %s", loc->path);
- if (should_commit_hash &&
- gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) {
+ if (gf_defrag_settle_hash(this, defrag, loc, fix_layout) != 0) {
defrag->total_failures++;
gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SETTLE_HASH_FAILED,
@@ -4021,245 +3893,34 @@ out:
if (fd)
fd_unref(fd);
- if (ret == 0 && should_commit_hash == 0) {
- ret = 2;
- }
-
return ret;
}
-/******************************************************************************
- * Tier background Fix layout functions
- ******************************************************************************/
-/* This is the background tier fixlayout thread */
-void *
-gf_tier_do_fix_layout(void *args)
-{
- gf_tier_fix_layout_arg_t *tier_fix_layout_arg = args;
- int ret = -1;
- xlator_t *this = NULL;
- dht_conf_t *conf = NULL;
- gf_defrag_info_t *defrag = NULL;
- dict_t *dict = NULL;
- loc_t loc = {
- 0,
- };
- struct iatt iatt = {
- 0,
- };
- struct iatt parent = {
- 0,
- };
-
- GF_VALIDATE_OR_GOTO("tier", tier_fix_layout_arg, out);
- GF_VALIDATE_OR_GOTO("tier", tier_fix_layout_arg->this, out);
- this = tier_fix_layout_arg->this;
-
- conf = this->private;
- GF_VALIDATE_OR_GOTO(this->name, conf, out);
-
- defrag = conf->defrag;
- GF_VALIDATE_OR_GOTO(this->name, defrag, out);
- GF_VALIDATE_OR_GOTO(this->name, defrag->root_inode, out);
-
- GF_VALIDATE_OR_GOTO(this->name, tier_fix_layout_arg->fix_layout, out);
-
- /* Get Root loc_t */
- dht_build_root_loc(defrag->root_inode, &loc);
- ret = syncop_lookup(this, &loc, &iatt, &parent, NULL, NULL);
- if (ret) {
- gf_msg(this->name, GF_LOG_ERROR, -ret, DHT_MSG_REBALANCE_START_FAILED,
- "Lookup on root failed.");
- ret = -1;
- goto out;
- }
-
- /* Start the crawl */
- gf_msg(this->name, GF_LOG_INFO, 0, DHT_MSG_LOG_TIER_STATUS,
- "Tiering Fixlayout started");
-
- ret = gf_defrag_fix_layout(this, defrag, &loc,
- tier_fix_layout_arg->fix_layout, NULL);
- if (ret && ret != 2) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED,
- "Tiering fixlayout failed.");
- ret = -1;
- goto out;
- }
-
- if (ret != 2 &&
- gf_defrag_settle_hash(this, defrag, &loc,
- tier_fix_layout_arg->fix_layout) != 0) {
- defrag->total_failures++;
- ret = -1;
- goto out;
- }
-
- dict = dict_new();
- if (!dict) {
- ret = -1;
- goto out;
- }
-
- ret = dict_set_str(dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY, "yes");
- if (ret) {
- gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_REBALANCE_FAILED,
- "Failed to set dictionary value: key = %s",
- GF_XATTR_TIER_LAYOUT_FIXED_KEY);
- ret = -1;
- goto out;
- }
-
- /* Marking the completion of tiering fix layout via a xattr on root */
- ret = syncop_setxattr(this, &loc, dict, 0, NULL, NULL);
- if (ret) {
- gf_log(this->name, GF_LOG_ERROR,
- "Failed to set tiering fix "
- "layout completed xattr on %s",
- loc.path);
- ret = -1;
- goto out;
- }
-
- ret = 0;
-out:
- if (ret && defrag)
- defrag->total_failures++;
-
- if (dict)
- dict_unref(dict);
-
- return NULL;
-}
-
-int
-gf_tier_start_fix_layout(xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag,
- dict_t *fix_layout)
-{
- int ret = -1;
- dict_t *tier_dict = NULL;
- gf_tier_fix_layout_arg_t *tier_fix_layout_arg = NULL;
-
- tier_dict = dict_new();
- if (!tier_dict) {
- gf_log("tier", GF_LOG_ERROR,
- "Tier fix layout failed :"
- "Creation of tier_dict failed");
- ret = -1;
- goto out;
- }
-
- /* Check if layout is fixed already */
- ret = syncop_getxattr(this, loc, &tier_dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY,
- NULL, NULL);
- if (ret != 0) {
- tier_fix_layout_arg = &defrag->tier_conf.tier_fix_layout_arg;
-
- /*Fill crawl arguments */
- tier_fix_layout_arg->this = this;
- tier_fix_layout_arg->fix_layout = fix_layout;
-
- /* Spawn the fix layout thread so that its done in the
- * background */
- ret = gf_thread_create(&tier_fix_layout_arg->thread_id, NULL,
- gf_tier_do_fix_layout, tier_fix_layout_arg,
- "tierfixl");
- if (ret) {
- gf_log("tier", GF_LOG_ERROR,
- "Thread creation failed. "
- "Background fix layout for tiering will not "
- "work.");
- defrag->total_failures++;
- goto out;
- }
- }
- ret = 0;
-out:
- if (tier_dict)
- dict_unref(tier_dict);
-
- return ret;
-}
-
-void
-gf_tier_clear_fix_layout(xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag)
-{
- int ret = -1;
- dict_t *dict = NULL;
-
- GF_VALIDATE_OR_GOTO("tier", this, out);
- GF_VALIDATE_OR_GOTO(this->name, loc, out);
- GF_VALIDATE_OR_GOTO(this->name, defrag, out);
-
- /* Check if background fixlayout is completed. This is not
- * multi-process safe i.e there is a possibility that by the time
- * we move to remove the xattr there it might have been cleared by some
- * other detach process from other node. We ignore the error if such
- * a thing happens */
- ret = syncop_getxattr(this, loc, &dict, GF_XATTR_TIER_LAYOUT_FIXED_KEY,
- NULL, NULL);
- if (ret) {
- /* Background fixlayout not complete - nothing to clear*/
- gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS,
- "Unable to retrieve fixlayout xattr."
- "Assume background fix layout not complete");
- goto out;
- }
-
- ret = syncop_removexattr(this, loc, GF_XATTR_TIER_LAYOUT_FIXED_KEY, NULL,
- NULL);
- if (ret) {
- gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_LOG_TIER_STATUS,
- "Failed removing tier fix layout "
- "xattr from %s",
- loc->path);
- goto out;
- }
- ret = 0;
-out:
- if (dict)
- dict_unref(dict);
-}
-
-void
-gf_tier_wait_fix_lookup(gf_defrag_info_t *defrag)
-{
- if (defrag->tier_conf.tier_fix_layout_arg.thread_id) {
- pthread_join(defrag->tier_conf.tier_fix_layout_arg.thread_id, NULL);
- }
-}
-/******************Tier background Fix layout functions END********************/
-
int
dht_init_local_subvols_and_nodeuuids(xlator_t *this, dht_conf_t *conf,
loc_t *loc)
{
dict_t *dict = NULL;
- gf_defrag_info_t *defrag = NULL;
uuid_t *uuid_ptr = NULL;
int ret = -1;
int i = 0;
int j = 0;
- defrag = conf->defrag;
-
- if (defrag->cmd != GF_DEFRAG_CMD_START_TIER) {
- /* Find local subvolumes */
- ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL,
- NULL, NULL);
- if (ret && (ret != -ENODATA)) {
- gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
- "local "
- "subvolume determination failed with error: %d",
- -ret);
- ret = -1;
- goto out;
- }
-
- if (!ret)
- goto out;
+ /* Find local subvolumes */
+ ret = syncop_getxattr(this, loc, &dict, GF_REBAL_FIND_LOCAL_SUBVOL, NULL,
+ NULL);
+ if (ret && (ret != -ENODATA)) {
+ gf_msg(this->name, GF_LOG_ERROR, -ret, 0,
+ "local "
+ "subvolume determination failed with error: %d",
+ -ret);
+ ret = -1;
+ goto out;
}
+ if (!ret)
+ goto out;
+
ret = syncop_getxattr(this, loc, &dict, GF_REBAL_OLD_FIND_LOCAL_SUBVOL,
NULL, NULL);
if (ret) {
@@ -4350,9 +4011,6 @@ dht_file_counter_thread(void *args)
struct timespec time_to_wait = {
0,
};
- struct timeval now = {
- 0,
- };
uint64_t tmp_size = 0;
if (!args)
@@ -4362,9 +4020,8 @@ dht_file_counter_thread(void *args)
dht_build_root_loc(defrag->root_inode, &root_loc);
while (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
- gettimeofday(&now, NULL);
- time_to_wait.tv_sec = now.tv_sec + 600;
- time_to_wait.tv_nsec = 0;
+ timespec_now(&time_to_wait);
+ time_to_wait.tv_sec += 600;
pthread_mutex_lock(&defrag->fc_mutex);
pthread_cond_timedwait(&defrag->fc_wakeup_cond, &defrag->fc_mutex,
@@ -4437,7 +4094,7 @@ gf_defrag_estimates_init(xlator_t *this, loc_t *loc, pthread_t *filecnt_thread)
goto out;
}
- ret = gf_thread_create(filecnt_thread, NULL, &dht_file_counter_thread,
+ ret = gf_thread_create(filecnt_thread, NULL, dht_file_counter_thread,
(void *)defrag, "dhtfcnt");
if (ret) {
@@ -4494,7 +4151,7 @@ gf_defrag_parallel_migration_init(xlator_t *this, gf_defrag_info_t *defrag,
/*Spawn Threads Here*/
while (index < thread_spawn_count) {
- ret = gf_thread_create(&(tid[index]), NULL, &gf_defrag_task,
+ ret = gf_thread_create(&(tid[index]), NULL, gf_defrag_task,
(void *)defrag, "dhtmig%d", (index + 1) & 0x3ff);
if (ret != 0) {
gf_msg("DHT", GF_LOG_ERROR, ret, 0, "Thread[%d] creation failed. ",
@@ -4568,7 +4225,6 @@ gf_defrag_start_crawl(void *data)
dict_t *migrate_data = NULL;
dict_t *status = NULL;
glusterfs_ctx_t *ctx = NULL;
- dht_methods_t *methods = NULL;
call_frame_t *statfs_frame = NULL;
xlator_t *old_THIS = NULL;
int ret = -1;
@@ -4584,7 +4240,6 @@ gf_defrag_start_crawl(void *data)
int thread_index = 0;
pthread_t *tid = NULL;
pthread_t filecnt_thread;
- gf_boolean_t is_tier_detach = _gf_false;
gf_boolean_t fc_thread_started = _gf_false;
this = data;
@@ -4603,7 +4258,8 @@ gf_defrag_start_crawl(void *data)
if (!defrag)
goto exit;
- gettimeofday(&defrag->start_time, NULL);
+ defrag->start_time = gf_time();
+
dht_build_root_inode(this, &defrag->root_inode);
if (!defrag->root_inode)
goto out;
@@ -4737,43 +4393,17 @@ gf_defrag_start_crawl(void *data)
}
}
- if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) {
- /* Fix layout for attach tier */
- ret = gf_tier_start_fix_layout(this, &loc, defrag, fix_layout);
- if (ret) {
- goto out;
- }
-
- methods = &(conf->methods);
-
- /* Calling tier_start of tier.c */
- methods->migration_other(this, defrag);
- if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER ||
- defrag->cmd == GF_DEFRAG_CMD_DETACH_START) {
- ret = dict_set_str(migrate_data, GF_XATTR_FILE_MIGRATE_KEY,
- "force");
- if (ret)
- goto out;
- }
- } else {
- ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout,
- migrate_data);
- if (ret && ret != 2) {
- defrag->total_failures++;
- ret = -1;
- goto out;
- }
-
- if (ret != 2 &&
- gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
- defrag->total_failures++;
- ret = -1;
- goto out;
- }
+ ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data);
+ if (ret) {
+ defrag->total_failures++;
+ ret = -1;
+ goto out;
+ }
- if (defrag->cmd == GF_DEFRAG_CMD_START_DETACH_TIER ||
- defrag->cmd == GF_DEFRAG_CMD_DETACH_START)
- is_tier_detach = _gf_true;
+ if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
+ defrag->total_failures++;
+ ret = -1;
+ goto out;
}
gf_log("DHT", GF_LOG_INFO, "crawling file-system completed");
@@ -4787,19 +4417,6 @@ out:
defrag->defrag_status = GF_DEFRAG_STATUS_FAILED;
}
- if (defrag->cmd == GF_DEFRAG_CMD_START_TIER) {
- /* Wait for the tier fixlayout to
- * complete if its was started.*/
- gf_tier_wait_fix_lookup(defrag);
- }
-
- if (is_tier_detach && ret == 0) {
- /* If it was a detach remove the tier fix-layout
- * xattr on root. Ignoring the failure, as nothing has to be
- * done, logging is done in gf_tier_clear_fix_layout */
- gf_tier_clear_fix_layout(this, &loc, defrag);
- }
-
gf_defrag_parallel_migration_cleanup(defrag, tid, thread_index);
if ((defrag->defrag_status != GF_DEFRAG_STATUS_STOPPED) &&
@@ -4898,9 +4515,6 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf)
uint64_t total_processed = 0;
uint64_t tmp_count = 0;
uint64_t time_to_complete = 0;
- struct timeval now = {
- 0,
- };
double elapsed = 0;
defrag = conf->defrag;
@@ -4908,8 +4522,7 @@ gf_defrag_get_estimates_based_on_size(dht_conf_t *conf)
if (!g_totalsize)
goto out;
- gettimeofday(&now, NULL);
- elapsed = now.tv_sec - defrag->start_time.tv_sec;
+ elapsed = gf_time() - defrag->start_time;
/* Don't calculate the estimates for the first 10 minutes.
* It is unlikely to be accurate and estimates are not required
@@ -4959,13 +4572,8 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict)
uint64_t lookup = 0;
uint64_t failures = 0;
uint64_t skipped = 0;
- uint64_t promoted = 0;
- uint64_t demoted = 0;
char *status = "";
double elapsed = 0;
- struct timeval end = {
- 0,
- };
uint64_t time_to_complete = 0;
uint64_t time_left = 0;
gf_defrag_info_t *defrag = conf->defrag;
@@ -4982,17 +4590,12 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict)
lookup = defrag->num_files_lookedup;
failures = defrag->total_failures;
skipped = defrag->skipped;
- promoted = defrag->total_files_promoted;
- demoted = defrag->total_files_demoted;
-
- gettimeofday(&end, NULL);
- elapsed = end.tv_sec - defrag->start_time.tv_sec;
+ elapsed = gf_time() - defrag->start_time;
/* The rebalance is still in progress */
- if ((defrag->cmd != GF_DEFRAG_CMD_START_TIER) &&
- (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED)) {
+ if (defrag->defrag_status == GF_DEFRAG_STATUS_STARTED) {
time_to_complete = gf_defrag_get_estimates_based_on_size(conf);
if (time_to_complete && (time_to_complete > elapsed))
@@ -5007,14 +4610,6 @@ gf_defrag_status_get(dht_conf_t *conf, dict_t *dict)
if (!dict)
goto log;
- ret = dict_set_uint64(dict, "promoted", promoted);
- if (ret)
- gf_log(THIS->name, GF_LOG_WARNING, "failed to set promoted count");
-
- ret = dict_set_uint64(dict, "demoted", demoted);
- if (ret)
- gf_log(THIS->name, GF_LOG_WARNING, "failed to set demoted count");
-
ret = dict_set_uint64(dict, "files", files);
if (ret)
gf_log(THIS->name, GF_LOG_WARNING, "failed to set file count");
@@ -5080,159 +4675,6 @@ out:
return 0;
}
-void
-gf_defrag_set_pause_state(gf_tier_conf_t *tier_conf, tier_pause_state_t state)
-{
- pthread_mutex_lock(&tier_conf->pause_mutex);
- tier_conf->pause_state = state;
- pthread_mutex_unlock(&tier_conf->pause_mutex);
-}
-
-tier_pause_state_t
-gf_defrag_get_pause_state(gf_tier_conf_t *tier_conf)
-{
- int state;
-
- pthread_mutex_lock(&tier_conf->pause_mutex);
- state = tier_conf->pause_state;
- pthread_mutex_unlock(&tier_conf->pause_mutex);
-
- return state;
-}
-
-tier_pause_state_t
-gf_defrag_check_pause_tier(gf_tier_conf_t *tier_conf)
-{
- int woke = 0;
- int state = -1;
-
- pthread_mutex_lock(&tier_conf->pause_mutex);
-
- if (tier_conf->pause_state == TIER_RUNNING)
- goto out;
-
- if (tier_conf->pause_state == TIER_PAUSED)
- goto out;
-
- if (tier_conf->promote_in_progress || tier_conf->demote_in_progress)
- goto out;
-
- tier_conf->pause_state = TIER_PAUSED;
-
- if (tier_conf->pause_synctask) {
- synctask_wake(tier_conf->pause_synctask);
- tier_conf->pause_synctask = 0;
- woke = 1;
- }
-
- gf_msg("tier", GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED, "woken %d", woke);
-
- gf_event(EVENT_TIER_PAUSE, "vol=%s", tier_conf->volname);
-out:
- state = tier_conf->pause_state;
-
- pthread_mutex_unlock(&tier_conf->pause_mutex);
-
- return state;
-}
-
-void
-gf_defrag_pause_tier_timeout(void *data)
-{
- xlator_t *this = NULL;
- dht_conf_t *conf = NULL;
- gf_defrag_info_t *defrag = NULL;
-
- this = (xlator_t *)data;
- GF_VALIDATE_OR_GOTO("tier", this, out);
-
- conf = this->private;
- GF_VALIDATE_OR_GOTO(this->name, conf, out);
-
- defrag = conf->defrag;
- GF_VALIDATE_OR_GOTO(this->name, defrag, out);
-
- gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED,
- "Request pause timer timeout");
-
- gf_defrag_check_pause_tier(&defrag->tier_conf);
-
-out:
- return;
-}
-
-int
-gf_defrag_pause_tier(xlator_t *this, gf_defrag_info_t *defrag)
-{
- int ret = 0;
- struct timespec delta = {
- 0,
- };
- int delay = 2;
-
- if (defrag->defrag_status != GF_DEFRAG_STATUS_STARTED)
- goto out;
-
- /*
- * Set flag requesting to pause tiering. Wait 'delay' seconds for
- * tiering to actually stop as indicated by the pause state
- * before returning success or failure.
- */
- gf_defrag_set_pause_state(&defrag->tier_conf, TIER_REQUEST_PAUSE);
-
- /*
- * If migration is not underway, can pause immediately.
- */
- gf_defrag_check_pause_tier(&defrag->tier_conf);
- if (gf_defrag_get_pause_state(&defrag->tier_conf) == TIER_PAUSED)
- goto out;
-
- gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED,
- "Request pause tier");
-
- defrag->tier_conf.pause_synctask = synctask_get();
- delta.tv_sec = delay;
- delta.tv_nsec = 0;
- defrag->tier_conf.pause_timer = gf_timer_call_after(
- this->ctx, delta, gf_defrag_pause_tier_timeout, this);
-
- synctask_yield(defrag->tier_conf.pause_synctask);
-
- if (gf_defrag_get_pause_state(&defrag->tier_conf) == TIER_PAUSED)
- goto out;
-
- gf_defrag_set_pause_state(&defrag->tier_conf, TIER_RUNNING);
-
- ret = -1;
-out:
-
- gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_PAUSED,
- "Pause tiering ret=%d", ret);
-
- return ret;
-}
-
-int
-gf_defrag_resume_tier(xlator_t *this, gf_defrag_info_t *defrag)
-{
- gf_msg(this->name, GF_LOG_DEBUG, 0, DHT_MSG_TIER_RESUME,
- "Pause end. Resume tiering");
-
- gf_defrag_set_pause_state(&defrag->tier_conf, TIER_RUNNING);
-
- gf_event(EVENT_TIER_RESUME, "vol=%s", defrag->tier_conf.volname);
-
- return 0;
-}
-
-int
-gf_defrag_start_detach_tier(gf_defrag_info_t *defrag)
-{
- defrag->cmd = GF_DEFRAG_CMD_START_DETACH_TIER;
-
- return 0;
-}
-
int
gf_defrag_stop(dht_conf_t *conf, gf_defrag_status_t status, dict_t *output)
{
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index 8657a22fd82..3e24065227c 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -1271,10 +1271,6 @@ dht_selfheal_dir_mkdir_lock_cbk(call_frame_t *frame, void *cookie,
local->call_cnt = conf->subvolume_cnt;
if (op_ret < 0) {
- /* We get this error when the directory entry was not created
- * on a newky attached tier subvol. Hence proceed and do mkdir
- * on the tier subvol.
- */
if (op_errno == EINVAL) {
local->call_cnt = 1;
dht_selfheal_dir_mkdir_lookup_done(frame, this);
@@ -1326,12 +1322,15 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
{
int missing_dirs = 0;
int i = 0;
+ int op_errno = 0;
int ret = -1;
dht_local_t *local = NULL;
xlator_t *this = NULL;
+ dht_conf_t *conf = NULL;
local = frame->local;
this = frame->this;
+ conf = this->private;
local->selfheal.force_mkdir = force;
local->selfheal.hole_cnt = 0;
@@ -1348,11 +1347,12 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
if (!__is_root_gfid(local->stbuf.ia_gfid)) {
if (local->need_xattr_heal) {
local->need_xattr_heal = 0;
- ret = dht_dir_xattr_heal(this, local);
- if (ret)
- gf_smsg(this->name, GF_LOG_ERROR, ret,
+ ret = dht_dir_xattr_heal(this, local, &op_errno);
+ if (ret) {
+ gf_smsg(this->name, GF_LOG_ERROR, op_errno,
DHT_MSG_DIR_XATTR_HEAL_FAILED, "path=%s",
local->loc.path, "gfid=%s", local->gfid, NULL);
+ }
} else {
if (!gf_uuid_is_null(local->gfid))
gf_uuid_copy(loc->gfid, local->gfid);
@@ -1370,15 +1370,44 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
return 0;
}
- if (local->hashed_subvol == NULL)
- local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+ /* MDS xattr is populated only while DHT is having more than one
+ subvol.In case of graph switch while adding more dht subvols need to
+ consider hash subvol as a MDS to avoid MDS check failure at the time
+ of running fop on directory
+ */
+ if (!dict_get(local->xattr, conf->mds_xattr_key) &&
+ (conf->subvolume_cnt > 1)) {
+ if (local->hashed_subvol == NULL) {
+ local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+ if (local->hashed_subvol == NULL) {
+ local->op_errno = EINVAL;
+ gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+ DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s",
+ loc->pargfid, "name=%s", loc->name, "path=%s",
+ loc->path, NULL);
+ goto err;
+ }
+ }
+ ret = dht_inode_ctx_mdsvol_set(local->inode, this,
+ local->hashed_subvol);
+ if (ret) {
+ gf_msg(this->name, GF_LOG_ERROR, 0, DHT_MSG_SET_INODE_CTX_FAILED,
+ "Failed to set hashed subvol for %s on inode vol is %s",
+ local->loc.path,
+ local->hashed_subvol ? local->hashed_subvol->name : "NULL");
+ goto err;
+ }
+ }
if (local->hashed_subvol == NULL) {
- local->op_errno = EINVAL;
- gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
- DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", loc->pargfid,
- "name=%s", loc->name, "path=%s", loc->path, NULL);
- goto err;
+ local->hashed_subvol = dht_subvol_get_hashed(this, loc);
+ if (local->hashed_subvol == NULL) {
+ local->op_errno = EINVAL;
+ gf_smsg(this->name, GF_LOG_WARNING, local->op_errno,
+ DHT_MSG_HASHED_SUBVOL_GET_FAILED, "gfid=%s", loc->pargfid,
+ "name=%s", loc->name, "path=%s", loc->path, NULL);
+ goto err;
+ }
}
local->current = &local->lock[0];
@@ -2154,6 +2183,15 @@ dht_dir_heal_xattrs(void *data)
if (subvol == mds_subvol)
continue;
if (uret || uflag) {
+ /* Custom xattr heal is required - let posix handle it */
+ ret = dict_set_int8(xdata, "sync_backend_xattrs", _gf_true);
+ if (ret) {
+ gf_smsg(this->name, GF_LOG_WARNING, 0, DHT_MSG_DICT_SET_FAILED,
+ "path=%s", local->loc.path, "key=%s",
+ "sync_backend_xattrs", NULL);
+ goto out;
+ }
+
ret = syncop_setxattr(subvol, &local->loc, user_xattr, 0, xdata,
NULL);
if (ret) {
@@ -2162,6 +2200,8 @@ dht_dir_heal_xattrs(void *data)
DHT_MSG_DIR_XATTR_HEAL_FAILED,
"set-user-xattr-failed path=%s", local->loc.path,
"subvol=%s", subvol->name, "gfid=%s", gfid, NULL);
+ } else {
+ dict_del(xdata, "sync_backend_xattrs");
}
}
}
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
index d85b4d1ce13..bb72b0ffbb5 100644
--- a/xlators/cluster/dht/src/dht-shared.c
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -140,9 +140,9 @@ dht_priv_dump(xlator_t *this)
}
}
- if (conf->last_stat_fetch.tv_sec)
+ if (conf->last_stat_fetch)
gf_proc_dump_write("last_stat_fetch", "%s",
- ctime(&conf->last_stat_fetch.tv_sec));
+ ctime(&conf->last_stat_fetch));
UNLOCK(&conf->subvolume_lock);
@@ -537,6 +537,8 @@ gf_defrag_pattern_list_fill(xlator_t *this, gf_defrag_info_t *defrag,
pattern_str = strtok_r(data, ",", &tmp_str);
while (pattern_str) {
dup_str = gf_strdup(pattern_str);
+ if (!dup_str)
+ goto out;
pattern_list = GF_CALLOC(1, sizeof(gf_defrag_pattern_list_t), 1);
if (!pattern_list) {
goto out;
@@ -596,7 +598,6 @@ dht_init_methods(xlator_t *this)
methods = &(conf->methods);
methods->migration_get_dst_subvol = dht_migration_get_dst_subvol;
- methods->migration_needed = dht_migration_needed;
methods->migration_other = NULL;
methods->layout_search = dht_layout_search;
@@ -1045,84 +1046,6 @@ struct volume_options dht_options[] = {
/* NUFA option */
{.key = {"local-volume-name"}, .type = GF_OPTION_TYPE_XLATOR},
- /* tier options */
- {
- .key = {"tier-pause"},
- .type = GF_OPTION_TYPE_BOOL,
- .default_value = "off",
- },
-
- {
- .key = {"tier-promote-frequency"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "120",
- },
-
- {
- .key = {"tier-demote-frequency"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "3600",
- },
-
- {
- .key = {"write-freq-threshold"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "0",
- },
-
- {
- .key = {"read-freq-threshold"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "0",
- },
- {
- .key = {"watermark-hi"},
- .type = GF_OPTION_TYPE_PERCENT,
- .default_value = "90",
- },
- {
- .key = {"watermark-low"},
- .type = GF_OPTION_TYPE_PERCENT,
- .default_value = "75",
- },
- {
- .key = {"tier-mode"},
- .type = GF_OPTION_TYPE_STR,
- .default_value = "test",
- },
- {
- .key = {"tier-compact"},
- .type = GF_OPTION_TYPE_BOOL,
- .default_value = "off",
- },
- {.key = {"tier-hot-compact-frequency"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "604800",
- .description = "Frequency to compact DBs on hot tier in system"},
- {.key = {"tier-cold-compact-frequency"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "604800",
- .description = "Frequency to compact DBs on cold tier in system"},
- {
- .key = {"tier-max-mb"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "4000",
- },
- {
- .key = {"tier-max-promote-file-size"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "0",
- },
- {
- .key = {"tier-max-files"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "10000",
- },
- {
- .key = {"tier-query-limit"},
- .type = GF_OPTION_TYPE_INT,
- .default_value = "100",
- },
/* switch option */
{.key = {"pattern.switch.case"}, .type = GF_OPTION_TYPE_ANY},
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index 59313639c45..3648a564840 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -595,7 +595,6 @@ nufa_init(xlator_t *this)
dht_methods_t dht_methods = {
.migration_get_dst_subvol = dht_migration_get_dst_subvol,
- .migration_needed = dht_migration_needed,
.layout_search = dht_layout_search,
};
generated by cgit (git 2.34.1) at 2024-06-10 22:41:46 +0000