summaryrefslogtreecommitdiffstats
path: root/xlators/cluster/afr/src/afr-self-heal-entry.c
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster/afr/src/afr-self-heal-entry.c')
-rw-r--r--xlators/cluster/afr/src/afr-self-heal-entry.c272
1 files changed, 221 insertions, 51 deletions
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index 619558e94b7..64893f441e3 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -10,60 +10,176 @@
#include "afr.h"
#include "afr-self-heal.h"
-#include "byte-order.h"
+#include <glusterfs/byte-order.h>
#include "afr-transaction.h"
#include "afr-messages.h"
-#include "syncop-utils.h"
-#include "events.h"
+#include <glusterfs/syncop-utils.h>
+#include <glusterfs/events.h>
-static int
-afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
- inode_t *inode, int child, struct afr_reply *replies)
+int
+afr_selfheal_entry_anon_inode(xlator_t *this, inode_t *dir, const char *name,
+ inode_t *inode, int child,
+ struct afr_reply *replies,
+ gf_boolean_t *anon_inode)
{
afr_private_t *priv = NULL;
+ afr_local_t *local = NULL;
xlator_t *subvol = NULL;
int ret = 0;
+ int i = 0;
+ char g[64] = {0};
+ unsigned char *lookup_success = NULL;
+ call_frame_t *frame = NULL;
+ loc_t loc2 = {
+ 0,
+ };
loc_t loc = {
0,
};
- char g[64];
priv = this->private;
-
subvol = priv->children[child];
+ lookup_success = alloca0(priv->child_count);
+ uuid_utoa_r(replies[child].poststat.ia_gfid, g);
+ loc.inode = inode_new(inode->table);
+ if (!loc.inode) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ if (replies[child].poststat.ia_type == IA_IFDIR) {
+ /* This directory may have sub-directory hierarchy which may need to
+ * be preserved for subsequent heals. So unconditionally move the
+ * directory to anonymous-inode directory*/
+ *anon_inode = _gf_true;
+ goto anon_inode;
+ }
+
+ frame = afr_frame_create(this, &ret);
+ if (!frame) {
+ ret = -ret;
+ goto out;
+ }
+ local = frame->local;
+ gf_uuid_copy(loc.gfid, replies[child].poststat.ia_gfid);
+ AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup, &loc,
+ NULL);
+ for (i = 0; i < priv->child_count; i++) {
+ if (local->replies[i].op_ret == 0) {
+ lookup_success[i] = 1;
+ } else if (local->replies[i].op_errno != ENOENT &&
+ local->replies[i].op_errno != ESTALE) {
+ ret = -local->replies[i].op_errno;
+ }
+ }
+
+ if (priv->quorum_count) {
+ if (afr_has_quorum(lookup_success, this, NULL)) {
+ *anon_inode = _gf_true;
+ }
+ } else if (AFR_COUNT(lookup_success, priv->child_count) > 1) {
+ *anon_inode = _gf_true;
+ } else if (ret) {
+ goto out;
+ }
+
+anon_inode:
+ if (!*anon_inode) {
+ ret = 0;
+ goto out;
+ }
loc.parent = inode_ref(dir);
gf_uuid_copy(loc.pargfid, dir->gfid);
loc.name = name;
- loc.inode = inode_ref(inode);
- if (replies[child].valid && replies[child].op_ret == 0) {
- switch (replies[child].poststat.ia_type) {
- case IA_IFDIR:
- gf_msg(this->name, GF_LOG_WARNING, 0,
- AFR_MSG_EXPUNGING_FILE_OR_DIR,
- "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid),
- name, uuid_utoa_r(replies[child].poststat.ia_gfid, g),
- subvol->name);
- ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
- break;
- default:
- gf_msg(this->name, GF_LOG_WARNING, 0,
- AFR_MSG_EXPUNGING_FILE_OR_DIR,
- "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid),
- name, uuid_utoa_r(replies[child].poststat.ia_gfid, g),
- subvol->name);
- ret = syncop_unlink(subvol, &loc, NULL, NULL);
- break;
- }
+ ret = afr_anon_inode_create(this, child, &loc2.parent);
+ if (ret < 0)
+ goto out;
+
+ loc2.name = g;
+ ret = syncop_rename(subvol, &loc, &loc2, NULL, NULL);
+ if (ret < 0) {
+ gf_msg(this->name, GF_LOG_WARNING, -ret, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+ "Rename to %s dir %s/%s (%s) on %s failed",
+ priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+ subvol->name);
+ } else {
+ gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+ "Rename to %s dir %s/%s (%s) on %s successful",
+ priv->anon_inode_name, uuid_utoa(dir->gfid), name, g,
+ subvol->name);
}
+out:
loc_wipe(&loc);
+ loc_wipe(&loc2);
+ if (frame) {
+ AFR_STACK_DESTROY(frame);
+ }
return ret;
}
int
+afr_selfheal_entry_delete(xlator_t *this, inode_t *dir, const char *name,
+ inode_t *inode, int child, struct afr_reply *replies)
+{
+ char g[64] = {0};
+ afr_private_t *priv = NULL;
+ xlator_t *subvol = NULL;
+ int ret = 0;
+ loc_t loc = {
+ 0,
+ };
+ gf_boolean_t anon_inode = _gf_false;
+
+ priv = this->private;
+ subvol = priv->children[child];
+
+ if ((!replies[child].valid) || (replies[child].op_ret < 0)) {
+ /*Nothing to do*/
+ ret = 0;
+ goto out;
+ }
+
+ if (priv->use_anon_inode) {
+ ret = afr_selfheal_entry_anon_inode(this, dir, name, inode, child,
+ replies, &anon_inode);
+ if (ret < 0 || anon_inode)
+ goto out;
+ }
+
+ loc.parent = inode_ref(dir);
+ loc.inode = inode_new(inode->table);
+ if (!loc.inode) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ loc.name = name;
+ switch (replies[child].poststat.ia_type) {
+ case IA_IFDIR:
+ gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+ "expunging dir %s/%s (%s) on %s", uuid_utoa(dir->gfid), name,
+ uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+ subvol->name);
+ ret = syncop_rmdir(subvol, &loc, 1, NULL, NULL);
+ break;
+ default:
+ gf_msg(this->name, GF_LOG_WARNING, 0, AFR_MSG_EXPUNGING_FILE_OR_DIR,
+ "expunging file %s/%s (%s) on %s", uuid_utoa(dir->gfid),
+ name, uuid_utoa_r(replies[child].poststat.ia_gfid, g),
+ subvol->name);
+ ret = syncop_unlink(subvol, &loc, NULL, NULL);
+ break;
+ }
+
+out:
+ loc_wipe(&loc);
+ return ret;
+}
+
+int
afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
unsigned char *sources, inode_t *dir,
const char *name, inode_t *inode,
@@ -76,6 +192,9 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
loc_t srcloc = {
0,
};
+ loc_t anonloc = {
+ 0,
+ };
xlator_t *this = frame->this;
afr_private_t *priv = NULL;
dict_t *xdata = NULL;
@@ -86,15 +205,18 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
0,
};
unsigned char *newentry = NULL;
+ char iatt_uuid_str[64] = {0};
+ char dir_uuid_str[64] = {0};
priv = this->private;
iatt = &replies[source].poststat;
+ uuid_utoa_r(iatt->ia_gfid, iatt_uuid_str);
if (iatt->ia_type == IA_INVAL || gf_uuid_is_null(iatt->ia_gfid)) {
gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_SELF_HEAL_FAILED,
"Invalid ia_type (%d) or gfid(%s). source brick=%d, "
"pargfid=%s, name=%s",
- iatt->ia_type, uuid_utoa(iatt->ia_gfid), source,
- uuid_utoa(dir->gfid), name);
+ iatt->ia_type, iatt_uuid_str, source,
+ uuid_utoa_r(dir->gfid, dir_uuid_str), name);
ret = -EINVAL;
goto out;
}
@@ -119,14 +241,24 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
srcloc.inode = inode_ref(inode);
gf_uuid_copy(srcloc.gfid, iatt->ia_gfid);
- if (iatt->ia_type != IA_IFDIR)
- ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0);
- if (iatt->ia_type == IA_IFDIR || ret == -ENOENT || ret == -ESTALE) {
+ ret = syncop_lookup(priv->children[dst], &srcloc, 0, 0, 0, 0);
+ if (ret == -ENOENT || ret == -ESTALE) {
newentry[dst] = 1;
ret = afr_selfheal_newentry_mark(frame, this, inode, source, replies,
sources, newentry);
if (ret)
goto out;
+ } else if (ret == 0 && iatt->ia_type == IA_IFDIR && priv->use_anon_inode) {
+ // Try rename from hidden directory
+ ret = afr_anon_inode_create(this, dst, &anonloc.parent);
+ if (ret < 0)
+ goto out;
+ anonloc.inode = inode_ref(inode);
+ anonloc.name = iatt_uuid_str;
+ ret = syncop_rename(priv->children[dst], &anonloc, &loc, NULL, NULL);
+ if (ret == -ENOENT || ret == -ESTALE)
+ ret = -1; /*This sets 'mismatch' to true*/
+ goto out;
}
mode = st_mode_from_ia(iatt->ia_prot, iatt->ia_type);
@@ -149,7 +281,7 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
}
break;
default:
- ret = dict_set_int32(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
+ ret = dict_set_int32_sizen(xdata, GLUSTERFS_INTERNAL_FOP_KEY, 1);
if (ret)
goto out;
ret = syncop_mknod(
@@ -165,6 +297,7 @@ out:
GF_FREE(linkname);
loc_wipe(&loc);
loc_wipe(&srcloc);
+ loc_wipe(&anonloc);
return ret;
}
@@ -192,7 +325,7 @@ __afr_selfheal_heal_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
if (replies[source].op_ret == 0) {
ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
source, sources,
- &replies[source].poststat.ia_gfid);
+ &replies[source].poststat.ia_gfid, NULL);
if (ret)
return ret;
}
@@ -246,6 +379,19 @@ afr_selfheal_detect_gfid_and_type_mismatch(xlator_t *this,
if (replies[i].op_ret != 0)
continue;
+ if (gf_uuid_is_null(replies[i].poststat.ia_gfid))
+ continue;
+
+ if (replies[i].poststat.ia_type == IA_INVAL)
+ continue;
+
+ if (ia_type == IA_INVAL || gf_uuid_is_null(gfid)) {
+ src_idx = i;
+ ia_type = replies[src_idx].poststat.ia_type;
+ gfid = &replies[src_idx].poststat.ia_gfid;
+ continue;
+ }
+
if (gf_uuid_compare(gfid, replies[i].poststat.ia_gfid) &&
(ia_type == replies[i].poststat.ia_type)) {
ret = afr_gfid_split_brain_source(this, replies, inode, pargfid,
@@ -269,11 +415,12 @@ afr_selfheal_detect_gfid_and_type_mismatch(xlator_t *this,
gf_inode_type_to_str(replies[src_idx].poststat.ia_type),
priv->children[src_idx]->name);
gf_event(EVENT_AFR_SPLIT_BRAIN,
+ "client-pid=%d;"
"subvol=%s;type=file;"
"file=<gfid:%s>/%s>;count=2;child-%d=%s;type-"
"%d=%s;child-%d=%s;type-%d=%s",
- this->name, uuid_utoa(pargfid), bname, i,
- priv->children[i]->name, i,
+ this->ctx->cmd_args.client_pid, this->name,
+ uuid_utoa(pargfid), bname, i, priv->children[i]->name, i,
gf_inode_type_to_str(replies[i].poststat.ia_type), src_idx,
priv->children[src_idx]->name, src_idx,
gf_inode_type_to_str(replies[src_idx].poststat.ia_type));
@@ -319,7 +466,7 @@ __afr_selfheal_merge_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
ret = afr_lookup_and_heal_gfid(this, fd->inode, name, inode, replies,
source, sources,
- &replies[source].poststat.ia_gfid);
+ &replies[source].poststat.ia_gfid, NULL);
if (ret)
return ret;
@@ -465,6 +612,7 @@ __afr_selfheal_entry_finalize_source(xlator_t *this, unsigned char *sources,
afr_private_t *priv = NULL;
int source = -1;
int sources_count = 0;
+ int i = 0;
priv = this->private;
@@ -478,6 +626,20 @@ __afr_selfheal_entry_finalize_source(xlator_t *this, unsigned char *sources,
}
source = afr_choose_source_by_policy(priv, sources, AFR_ENTRY_TRANSACTION);
+
+ /*If the selected source does not blame any other brick, then mark
+ * everything as sink to trigger conservative merge.
+ */
+ if (source != -1 && !AFR_COUNT(healed_sinks, priv->child_count)) {
+ for (i = 0; i < priv->child_count; i++) {
+ if (locked_on[i]) {
+ sources[i] = 0;
+ healed_sinks[i] = 1;
+ }
+ }
+ return -1;
+ }
+
return source;
}
@@ -548,10 +710,15 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
priv = this->private;
+ if (afr_is_private_directory(priv, fd->inode->gfid, name,
+ GF_CLIENT_PID_SELF_HEALD)) {
+ return 0;
+ }
+
xattr = dict_new();
if (!xattr)
return -ENOMEM;
- ret = dict_set_int32(xattr, GF_GFIDLESS_LOOKUP, 1);
+ ret = dict_set_int32_sizen(xattr, GF_GFIDLESS_LOOKUP, 1);
if (ret) {
dict_unref(xattr);
return -1;
@@ -568,7 +735,7 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
locked_on);
{
- if (ret < AFR_SH_MIN_PARTICIPANTS) {
+ if (ret < priv->child_count) {
gf_msg_debug(this->name, 0,
"%s: Skipping "
"entry self-heal as only %d sub-volumes "
@@ -596,7 +763,7 @@ afr_selfheal_entry_dirent(call_frame_t *frame, xlator_t *this, fd_t *fd,
replies);
if ((ret == 0) && (priv->esh_granular) && parent_idx_inode) {
- ret = afr_shd_index_purge(subvol, parent_idx_inode, name,
+ ret = afr_shd_entry_purge(subvol, parent_idx_inode, name,
inode->ia_type);
/* Why is ret force-set to 0? We do not care about
* index purge failing for full heal as it is quite
@@ -726,10 +893,6 @@ afr_selfheal_entry_do_subvol(call_frame_t *frame, xlator_t *this, fd_t *fd,
if (!strcmp(entry->d_name, ".") || !strcmp(entry->d_name, ".."))
continue;
- if (__is_root_gfid(fd->inode->gfid) &&
- !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR))
- continue;
-
ret = afr_selfheal_entry_dirent(iter_frame, this, fd, entry->d_name,
loc.inode, subvol,
local->need_full_crawl);
@@ -792,7 +955,7 @@ afr_selfheal_entry_granular_dirent(xlator_t *subvol, gf_dirent_t *entry,
/* The name indices under the pgfid index dir are guaranteed
* to be regular files. Hence the hardcoding.
*/
- afr_shd_index_purge(subvol, parent->inode, entry->d_name, IA_IFREG);
+ afr_shd_entry_purge(subvol, parent->inode, entry->d_name, IA_IFREG);
ret = 0;
goto out;
}
@@ -831,6 +994,8 @@ afr_selfheal_entry_granular(call_frame_t *frame, xlator_t *this, fd_t *fd,
subvol = priv->children[subvol_idx];
args.frame = afr_copy_frame(frame);
+ if (!args.frame)
+ goto out;
args.xl = this;
/* args.heal_fd represents the fd associated with the original directory
* on which entry heal is being attempted.
@@ -849,9 +1014,10 @@ afr_selfheal_entry_granular(call_frame_t *frame, xlator_t *this, fd_t *fd,
* do not treat heal as failure.
*/
if (is_src)
- return -errno;
+ ret = -errno;
else
- return 0;
+ ret = 0;
+ goto out;
}
ret = syncop_dir_scan(subvol, &loc, GF_CLIENT_PID_SELF_HEALD, &args,
@@ -861,7 +1027,9 @@ afr_selfheal_entry_granular(call_frame_t *frame, xlator_t *this, fd_t *fd,
if (args.mismatch == _gf_true)
ret = -1;
-
+out:
+ if (args.frame)
+ AFR_STACK_DESTROY(args.frame);
return ret;
}
@@ -957,7 +1125,7 @@ __afr_selfheal_entry(call_frame_t *frame, xlator_t *this, fd_t *fd,
ret = afr_selfheal_entrylk(frame, this, fd->inode, this->name, NULL,
data_lock);
{
- if (ret < AFR_SH_MIN_PARTICIPANTS) {
+ if (ret < priv->child_count) {
gf_msg_debug(this->name, 0,
"%s: Skipping "
"entry self-heal as only %d sub-volumes could "
@@ -1013,6 +1181,8 @@ unlock:
goto postop_unlock;
}
+ afr_selfheal_restore_time(frame, this, fd->inode, source, healed_sinks,
+ locked_replies);
ret = afr_selfheal_undo_pending(
frame, this, fd->inode, sources, sinks, healed_sinks, undid_pending,
AFR_ENTRY_TRANSACTION, locked_replies, postop_lock);
@@ -1079,7 +1249,7 @@ afr_selfheal_entry(call_frame_t *frame, xlator_t *this, inode_t *inode)
ret = afr_selfheal_tie_breaker_entrylk(frame, this, inode, priv->sh_domain,
NULL, locked_on);
{
- if (ret < AFR_SH_MIN_PARTICIPANTS) {
+ if (ret < priv->child_count) {
gf_msg_debug(this->name, 0,
"%s: Skipping "
"entry self-heal as only %d sub-volumes could "
generated by cgit (git 2.34.1) at 2026-01-05 20:54:18 +0000