summaryrefslogtreecommitdiffstats
path: root/xlators/cluster
diff options
context:
space:
mode:
Diffstat (limited to 'xlators/cluster')
-rw-r--r--xlators/cluster/afr/src/afr-common.c49
-rw-r--r--xlators/cluster/afr/src/afr-dir-read.c12
-rw-r--r--xlators/cluster/afr/src/afr-self-heal-common.c182
-rw-r--r--xlators/cluster/afr/src/afr-self-heal-entry.c205
-rw-r--r--xlators/cluster/afr/src/afr-self-heal-name.c33
-rw-r--r--xlators/cluster/afr/src/afr-self-heal.h5
-rw-r--r--xlators/cluster/afr/src/afr-self-heald.c178
-rw-r--r--xlators/cluster/afr/src/afr-self-heald.h2
-rw-r--r--xlators/cluster/afr/src/afr.c41
-rw-r--r--xlators/cluster/afr/src/afr.h11
-rw-r--r--xlators/cluster/dht/src/dht-common.c108
-rw-r--r--xlators/cluster/dht/src/dht-common.h105
-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.c278
-rw-r--r--xlators/cluster/dht/src/dht-selfheal.c49
-rw-r--r--xlators/cluster/dht/src/dht-shared.c81
-rw-r--r--xlators/cluster/dht/src/nufa.c1
19 files changed, 776 insertions, 635 deletions
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 4c8fa31b679..032ab5c8001 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -45,6 +45,41 @@ afr_quorum_errno(afr_private_t *priv)
return ENOTCONN;
}
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+ pid_t pid)
+{
+ if (!__is_root_gfid(pargfid)) {
+ return _gf_false;
+ }
+
+ if (strcmp(name, GF_REPLICATE_TRASH_DIR) == 0) {
+ /*For backward compatibility /.landfill is private*/
+ return _gf_true;
+ }
+
+ if (pid == GF_CLIENT_PID_GSYNCD) {
+ /*geo-rep needs to create/sync private directory on slave because
+ * it appears in changelog*/
+ return _gf_false;
+ }
+
+ if (pid == GF_CLIENT_PID_GLFS_HEAL || pid == GF_CLIENT_PID_SELF_HEALD) {
+ if (strcmp(name, priv->anon_inode_name) == 0) {
+ /* anonymous-inode dir is private*/
+ return _gf_true;
+ }
+ } else {
+ if (strncmp(name, AFR_ANON_DIR_PREFIX, strlen(AFR_ANON_DIR_PREFIX)) ==
+ 0) {
+ /* anonymous-inode dir prefix is private for geo-rep to work*/
+ return _gf_true;
+ }
+ }
+
+ return _gf_false;
+}
+
void
afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
unsigned char *replies)
@@ -2271,8 +2306,9 @@ afr_hash_child(afr_read_subvol_args_t *args, afr_private_t *priv,
* need is a low probability that multiple clients
* won't converge on the same subvolume.
*/
+ gf_uuid_copy(gfid_copy, args->gfid);
pid = getpid();
- memcpy(gfid_copy, &pid, sizeof(pid));
+ *(pid_t *)gfid_copy ^= pid;
}
child = SuperFastHash((char *)gfid_copy, sizeof(gfid_copy)) %
priv->child_count;
@@ -3978,11 +4014,10 @@ afr_lookup(call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
return 0;
}
- if (__is_root_gfid(loc->parent->gfid)) {
- if (!strcmp(loc->name, GF_REPLICATE_TRASH_DIR)) {
- op_errno = EPERM;
- goto out;
- }
+ if (afr_is_private_directory(this->private, loc->parent->gfid, loc->name,
+ frame->root->pid)) {
+ op_errno = EPERM;
+ goto out;
}
local = AFR_FRAME_INIT(frame, op_errno);
@@ -5660,6 +5695,7 @@ afr_priv_dump(xlator_t *this)
priv->background_self_heal_count);
gf_proc_dump_write("healers", "%d", priv->healers);
gf_proc_dump_write("read-hash-mode", "%d", priv->hash_mode);
+ gf_proc_dump_write("use-anonymous-inode", "%d", priv->use_anon_inode);
if (priv->quorum_count == AFR_QUORUM_AUTO) {
gf_proc_dump_write("quorum-type", "auto");
} else if (priv->quorum_count == 0) {
@@ -6653,6 +6689,7 @@ afr_priv_destroy(afr_private_t *priv)
GF_FREE(priv->local);
GF_FREE(priv->pending_key);
GF_FREE(priv->children);
+ GF_FREE(priv->anon_inode);
GF_FREE(priv->child_up);
GF_FREE(priv->halo_child_up);
GF_FREE(priv->child_latency);
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index f69013f3e0a..f8bf8340dab 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -164,8 +164,8 @@ afr_validate_read_subvol(inode_t *inode, xlator_t *this, int par_read_subvol)
}
static void
-afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol,
- gf_dirent_t *entries, fd_t *fd)
+afr_readdir_transform_entries(call_frame_t *frame, gf_dirent_t *subvol_entries,
+ int subvol, gf_dirent_t *entries, fd_t *fd)
{
int ret = -1;
gf_dirent_t *entry = NULL;
@@ -183,8 +183,8 @@ afr_readdir_transform_entries(gf_dirent_t *subvol_entries, int subvol,
list_for_each_entry_safe(entry, tmp, &subvol_entries->list, list)
{
- if (__is_root_gfid(fd->inode->gfid) &&
- !strcmp(entry->d_name, GF_REPLICATE_TRASH_DIR)) {
+ if (afr_is_private_directory(priv, fd->inode->gfid, entry->d_name,
+ frame->root->pid)) {
continue;
}
@@ -228,8 +228,8 @@ afr_readdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
}
if (op_ret >= 0)
- afr_readdir_transform_entries(subvol_entries, (long)cookie, &entries,
- local->fd);
+ afr_readdir_transform_entries(frame, subvol_entries, (long)cookie,
+ &entries, local->fd);
AFR_STACK_UNWIND(readdir, frame, op_ret, op_errno, &entries, xdata);
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index f35c41df274..a580a1584cc 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -2750,3 +2750,185 @@ afr_choose_source_by_policy(afr_private_t *priv, unsigned char *sources,
out:
return source;
}
+
+static int
+afr_anon_inode_mkdir_cbk(call_frame_t *frame, void *cookie, xlator_t *this,
+ int32_t op_ret, int32_t op_errno, inode_t *inode,
+ struct iatt *buf, struct iatt *preparent,
+ struct iatt *postparent, dict_t *xdata)
+{
+ afr_local_t *local = frame->local;
+ int i = (long)cookie;
+
+ local->replies[i].valid = 1;
+ local->replies[i].op_ret = op_ret;
+ local->replies[i].op_errno = op_errno;
+ if (op_ret == 0) {
+ local->op_ret = 0;
+ local->replies[i].poststat = *buf;
+ local->replies[i].preparent = *preparent;
+ local->replies[i].postparent = *postparent;
+ }
+ if (xdata) {
+ local->replies[i].xdata = dict_ref(xdata);
+ }
+
+ syncbarrier_wake(&local->barrier);
+ return 0;
+}
+
+int
+afr_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode)
+{
+ call_frame_t *frame = NULL;
+ afr_local_t *local = NULL;
+ afr_private_t *priv = this->private;
+ unsigned char *mkdir_on = alloca0(priv->child_count);
+ unsigned char *lookup_on = alloca0(priv->child_count);
+ loc_t loc = {0};
+ int32_t op_errno = 0;
+ int32_t child_op_errno = 0;
+ struct iatt iatt = {0};
+ dict_t *xdata = NULL;
+ uuid_t anon_inode_gfid = {0};
+ int mkdir_count = 0;
+ int i = 0;
+
+ /*Try to mkdir everywhere and return success if the dir exists on 'child'
+ */
+
+ if (!priv->use_anon_inode) {
+ op_errno = EINVAL;
+ goto out;
+ }
+
+ frame = afr_frame_create(this, &op_errno);
+ if (op_errno) {
+ goto out;
+ }
+ local = frame->local;
+ if (!local->child_up[child]) {
+ /*Other bricks may need mkdir so don't error out yet*/
+ child_op_errno = ENOTCONN;
+ }
+ gf_uuid_parse(priv->anon_gfid_str, anon_inode_gfid);
+ for (i = 0; i < priv->child_count; i++) {
+ if (!local->child_up[i])
+ continue;
+
+ if (priv->anon_inode[i]) {
+ mkdir_on[i] = 0;
+ } else {
+ mkdir_on[i] = 1;
+ mkdir_count++;
+ }
+ }
+
+ if (mkdir_count == 0) {
+ *linked_inode = inode_find(this->itable, anon_inode_gfid);
+ if (*linked_inode) {
+ op_errno = 0;
+ goto out;
+ }
+ }
+
+ loc.parent = inode_ref(this->itable->root);
+ loc.name = priv->anon_inode_name;
+ loc.inode = inode_new(this->itable);
+ if (!loc.inode) {
+ op_errno = ENOMEM;
+ goto out;
+ }
+
+ xdata = dict_new();
+ if (!xdata) {
+ op_errno = ENOMEM;
+ goto out;
+ }
+
+ op_errno = -dict_set_gfuuid(xdata, "gfid-req", anon_inode_gfid, _gf_true);
+ if (op_errno) {
+ goto out;
+ }
+
+ if (mkdir_count == 0) {
+ memcpy(lookup_on, local->child_up, priv->child_count);
+ goto lookup;
+ }
+
+ AFR_ONLIST(mkdir_on, frame, afr_anon_inode_mkdir_cbk, mkdir, &loc, 0755, 0,
+ xdata);
+
+ for (i = 0; i < priv->child_count; i++) {
+ if (!mkdir_on[i]) {
+ continue;
+ }
+
+ if (local->replies[i].op_ret == 0) {
+ priv->anon_inode[i] = 1;
+ iatt = local->replies[i].poststat;
+ } else if (local->replies[i].op_ret < 0 &&
+ local->replies[i].op_errno == EEXIST) {
+ lookup_on[i] = 1;
+ } else if (i == child) {
+ child_op_errno = local->replies[i].op_errno;
+ }
+ }
+
+ if (AFR_COUNT(lookup_on, priv->child_count) == 0) {
+ goto link;
+ }
+
+lookup:
+ AFR_ONLIST(lookup_on, frame, afr_selfheal_discover_cbk, lookup, &loc,
+ xdata);
+ for (i = 0; i < priv->child_count; i++) {
+ if (!lookup_on[i]) {
+ continue;
+ }
+
+ if (local->replies[i].op_ret == 0) {
+ if (gf_uuid_compare(anon_inode_gfid,
+ local->replies[i].poststat.ia_gfid) == 0) {
+ priv->anon_inode[i] = 1;
+ iatt = local->replies[i].poststat;
+ } else {
+ if (i == child)
+ child_op_errno = EINVAL;
+ gf_msg(this->name, GF_LOG_ERROR, 0, AFR_MSG_INVALID_DATA,
+ "%s has gfid: %s", priv->anon_inode_name,
+ uuid_utoa(local->replies[i].poststat.ia_gfid));
+ }
+ } else if (i == child) {
+ child_op_errno = local->replies[i].op_errno;
+ }
+ }
+link:
+ if (!gf_uuid_is_null(iatt.ia_gfid)) {
+ *linked_inode = inode_link(loc.inode, loc.parent, loc.name, &iatt);
+ if (*linked_inode) {
+ op_errno = 0;
+ inode_lookup(*linked_inode);
+ } else {
+ op_errno = ENOMEM;
+ }
+ goto out;
+ }
+
+out:
+ if (xdata)
+ dict_unref(xdata);
+ loc_wipe(&loc);
+ /*child_op_errno takes precedence*/
+ if (child_op_errno == 0) {
+ child_op_errno = op_errno;
+ }
+
+ if (child_op_errno && *linked_inode) {
+ inode_unref(*linked_inode);
+ *linked_inode = NULL;
+ }
+ if (frame)
+ AFR_STACK_DESTROY(frame);
+ return -child_op_errno;
+}
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index ac31751997f..64893f441e3 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -16,54 +16,170 @@
#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 +205,17 @@ afr_selfheal_recreate_entry(call_frame_t *frame, int dst, int source,
0,
};
unsigned char *newentry = NULL;
- char dir_uuid_str[64] = {0}, iatt_uuid_str[64] = {0};
+ 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_r(iatt->ia_gfid, iatt_uuid_str), source,
+ iatt->ia_type, iatt_uuid_str, source,
uuid_utoa_r(dir->gfid, dir_uuid_str), name);
ret = -EINVAL;
goto out;
@@ -120,14 +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);
@@ -166,6 +297,7 @@ out:
GF_FREE(linkname);
loc_wipe(&loc);
loc_wipe(&srcloc);
+ loc_wipe(&anonloc);
return ret;
}
@@ -578,6 +710,11 @@ 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;
@@ -626,7 +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
@@ -756,10 +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);
@@ -822,7 +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;
}
diff --git a/xlators/cluster/afr/src/afr-self-heal-name.c b/xlators/cluster/afr/src/afr-self-heal-name.c
index dd40c57ab12..834aac86d48 100644
--- a/xlators/cluster/afr/src/afr-self-heal-name.c
+++ b/xlators/cluster/afr/src/afr-self-heal-name.c
@@ -98,21 +98,12 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid,
const char *bname, inode_t *inode,
struct afr_reply *replies)
{
- loc_t loc = {
- 0,
- };
int i = 0;
afr_private_t *priv = NULL;
- char g[64];
int ret = 0;
priv = this->private;
- loc.parent = inode_ref(parent);
- gf_uuid_copy(loc.pargfid, pargfid);
- loc.name = bname;
- loc.inode = inode_ref(inode);
-
for (i = 0; i < priv->child_count; i++) {
if (!replies[i].valid)
continue;
@@ -120,30 +111,10 @@ __afr_selfheal_name_expunge(xlator_t *this, inode_t *parent, uuid_t pargfid,
if (replies[i].op_ret)
continue;
- switch (replies[i].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(pargfid),
- bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g),
- priv->children[i]->name);
-
- ret |= syncop_rmdir(priv->children[i], &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(pargfid),
- bname, uuid_utoa_r(replies[i].poststat.ia_gfid, g),
- priv->children[i]->name);
-
- ret |= syncop_unlink(priv->children[i], &loc, NULL, NULL);
- break;
- }
+ ret |= afr_selfheal_entry_delete(this, parent, bname, inode, i,
+ replies);
}
- loc_wipe(&loc);
-
return ret;
}
diff --git a/xlators/cluster/afr/src/afr-self-heal.h b/xlators/cluster/afr/src/afr-self-heal.h
index 7a038fa7fe3..48e6dbcfb18 100644
--- a/xlators/cluster/afr/src/afr-self-heal.h
+++ b/xlators/cluster/afr/src/afr-self-heal.h
@@ -369,4 +369,9 @@ gf_boolean_t
afr_is_file_empty_on_all_children(afr_private_t *priv,
struct afr_reply *replies);
+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_anon_inode_create(xlator_t *this, int child, inode_t **linked_inode);
#endif /* !_AFR_SELFHEAL_H */
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
index f2e08908c24..109fd4b7421 100644
--- a/xlators/cluster/afr/src/afr-self-heald.c
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -222,7 +222,7 @@ out:
}
int
-afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name,
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
ia_type_t type)
{
int ret = 0;
@@ -424,7 +424,7 @@ afr_shd_index_heal(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
ret = afr_shd_selfheal(healer, healer->subvol, gfid);
if (ret == -ENOENT || ret == -ESTALE)
- afr_shd_index_purge(subvol, parent->inode, entry->d_name, val);
+ afr_shd_entry_purge(subvol, parent->inode, entry->d_name, val);
if (ret == 2)
/* If bricks crashed in pre-op after creating indices/xattrop
@@ -843,6 +843,176 @@ out:
return need_heal;
}
+static int
+afr_shd_anon_inode_cleaner(xlator_t *subvol, gf_dirent_t *entry, loc_t *parent,
+ void *data)
+{
+ struct subvol_healer *healer = data;
+ afr_private_t *priv = healer->this->private;
+ call_frame_t *frame = NULL;
+ afr_local_t *local = NULL;
+ int ret = 0;
+ loc_t loc = {0};
+ int count = 0;
+ int i = 0;
+ int op_errno = 0;
+ struct iatt *iatt = NULL;
+ gf_boolean_t multiple_links = _gf_false;
+ unsigned char *gfid_present = alloca0(priv->child_count);
+ unsigned char *entry_present = alloca0(priv->child_count);
+ char *type = "file";
+
+ frame = afr_frame_create(healer->this, &ret);
+ if (!frame) {
+ ret = -ret;
+ goto out;
+ }
+ local = frame->local;
+ if (AFR_COUNT(local->child_up, priv->child_count) != priv->child_count) {
+ gf_msg_debug(healer->this->name, 0,
+ "Not all bricks are up. Skipping "
+ "cleanup of %s on %s",
+ entry->d_name, subvol->name);
+ ret = 0;
+ goto out;
+ }
+
+ loc.inode = inode_new(parent->inode->table);
+ if (!loc.inode) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ ret = gf_uuid_parse(entry->d_name, loc.gfid);
+ if (ret) {
+ ret = 0;
+ goto out;
+ }
+ 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) {
+ count++;
+ gfid_present[i] = 1;
+ iatt = &local->replies[i].poststat;
+ if (iatt->ia_type == IA_IFDIR) {
+ type = "dir";
+ }
+
+ if (i == healer->subvol) {
+ if (local->replies[i].poststat.ia_nlink > 1) {
+ multiple_links = _gf_true;
+ }
+ }
+ } else if (local->replies[i].op_errno != ENOENT &&
+ local->replies[i].op_errno != ESTALE) {
+ /*We don't have complete view. Skip the entry*/
+ gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+ "Skipping cleanup of %s on %s", entry->d_name,
+ subvol->name);
+ ret = 0;
+ goto out;
+ }
+ }
+
+ /*Inode is deleted from subvol*/
+ if (count == 1 || (iatt->ia_type != IA_IFDIR && multiple_links)) {
+ gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+ AFR_MSG_EXPUNGING_FILE_OR_DIR, "expunging %s %s/%s on %s", type,
+ priv->anon_inode_name, entry->d_name, subvol->name);
+ ret = afr_shd_entry_purge(subvol, parent->inode, entry->d_name,
+ iatt->ia_type);
+ if (ret == -ENOENT || ret == -ESTALE)
+ ret = 0;
+ } else if (count > 1) {
+ loc_wipe(&loc);
+ loc.parent = inode_ref(parent->inode);
+ loc.name = entry->d_name;
+ loc.inode = inode_new(parent->inode->table);
+ if (!loc.inode) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ AFR_ONLIST(local->child_up, frame, afr_selfheal_discover_cbk, lookup,
+ &loc, NULL);
+ count = 0;
+ for (i = 0; i < priv->child_count; i++) {
+ if (local->replies[i].op_ret == 0) {
+ count++;
+ entry_present[i] = 1;
+ iatt = &local->replies[i].poststat;
+ } else if (local->replies[i].op_errno != ENOENT &&
+ local->replies[i].op_errno != ESTALE) {
+ /*We don't have complete view. Skip the entry*/
+ gf_msg_debug(healer->this->name, local->replies[i].op_errno,
+ "Skipping cleanup of %s on %s", entry->d_name,
+ subvol->name);
+ ret = 0;
+ goto out;
+ }
+ }
+ for (i = 0; i < priv->child_count; i++) {
+ if (gfid_present[i] && !entry_present[i]) {
+ /*Entry is not anonymous on at least one subvol*/
+ gf_msg_debug(healer->this->name, 0,
+ "Valid entry present on %s "
+ "Skipping cleanup of %s on %s",
+ priv->children[i]->name, entry->d_name,
+ subvol->name);
+ ret = 0;
+ goto out;
+ }
+ }
+
+ gf_msg(healer->this->name, GF_LOG_WARNING, 0,
+ AFR_MSG_EXPUNGING_FILE_OR_DIR,
+ "expunging %s %s/%s on all subvols", type, priv->anon_inode_name,
+ entry->d_name);
+ ret = 0;
+ for (i = 0; i < priv->child_count; i++) {
+ op_errno = -afr_shd_entry_purge(priv->children[i], loc.parent,
+ entry->d_name, iatt->ia_type);
+ if (op_errno != ENOENT && op_errno != ESTALE) {
+ ret |= -op_errno;
+ }
+ }
+ }
+
+out:
+ if (frame)
+ AFR_STACK_DESTROY(frame);
+ loc_wipe(&loc);
+ return ret;
+}
+
+static void
+afr_cleanup_anon_inode_dir(struct subvol_healer *healer)
+{
+ int ret = 0;
+ call_frame_t *frame = NULL;
+ afr_private_t *priv = healer->this->private;
+ loc_t loc = {0};
+
+ ret = afr_anon_inode_create(healer->this, healer->subvol, &loc.inode);
+ if (ret)
+ goto out;
+
+ frame = afr_frame_create(healer->this, &ret);
+ if (!frame) {
+ ret = -ret;
+ goto out;
+ }
+
+ ret = syncop_mt_dir_scan(frame, priv->children[healer->subvol], &loc,
+ GF_CLIENT_PID_SELF_HEALD, healer,
+ afr_shd_anon_inode_cleaner, NULL,
+ priv->shd.max_threads, priv->shd.wait_qlength);
+out:
+ if (frame)
+ AFR_STACK_DESTROY(frame);
+ loc_wipe(&loc);
+ return;
+}
+
void *
afr_shd_index_healer(void *data)
{
@@ -900,6 +1070,10 @@ afr_shd_index_healer(void *data)
sleep(1);
} while (ret > 0);
+ if (ret == 0) {
+ afr_cleanup_anon_inode_dir(healer);
+ }
+
if (ret == 0 && pre_crawl_xdata &&
!healer->crawl_event.heal_failed_count) {
afr_shd_ta_check_and_unset_xattrs(this, &loc, healer,
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
index 687c28e6472..18db728ea7b 100644
--- a/xlators/cluster/afr/src/afr-self-heald.h
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -70,6 +70,6 @@ afr_shd_gfid_to_path(xlator_t *this, xlator_t *subvol, uuid_t gfid,
char **path_p);
int
-afr_shd_index_purge(xlator_t *subvol, inode_t *inode, char *name,
+afr_shd_entry_purge(xlator_t *subvol, inode_t *inode, char *name,
ia_type_t type);
#endif /* !_AFR_SELF_HEALD_H */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index 67e0a4d10be..df7366f0a65 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -135,6 +135,27 @@ set_data_self_heal_algorithm(afr_private_t *priv, char *algo)
}
}
+void
+afr_handle_anon_inode_options(afr_private_t *priv, dict_t *options)
+{
+ char *volfile_id_str = NULL;
+ uuid_t anon_inode_gfid = {0};
+
+ /*If volume id is not present don't enable anything*/
+ if (dict_get_str(options, "volume-id", &volfile_id_str))
+ return;
+ GF_ASSERT(strlen(AFR_ANON_DIR_PREFIX) + strlen(volfile_id_str) <= NAME_MAX);
+ /*anon_inode_name is not supposed to change once assigned*/
+ if (!priv->anon_inode_name[0]) {
+ snprintf(priv->anon_inode_name, sizeof(priv->anon_inode_name), "%s-%s",
+ AFR_ANON_DIR_PREFIX, volfile_id_str);
+ gf_uuid_parse(volfile_id_str, anon_inode_gfid);
+ /*Flip a bit to make sure volfile-id and anon-gfid are not same*/
+ anon_inode_gfid[0] ^= 1;
+ uuid_utoa_r(anon_inode_gfid, priv->anon_gfid_str);
+ }
+}
+
int
reconfigure(xlator_t *this, dict_t *options)
{
@@ -290,6 +311,10 @@ reconfigure(xlator_t *this, dict_t *options)
consistent_io = _gf_false;
priv->consistent_io = consistent_io;
+ afr_handle_anon_inode_options(priv, options);
+
+ GF_OPTION_RECONF("use-anonymous-inode", priv->use_anon_inode, options, bool,
+ out);
if (priv->shd.enabled) {
if ((priv->shd.enabled != enabled_old) ||
(timeout_old != priv->shd.timeout))
@@ -541,7 +566,9 @@ init(xlator_t *this)
GF_OPTION_INIT("consistent-metadata", priv->consistent_metadata, bool, out);
GF_OPTION_INIT("consistent-io", priv->consistent_io, bool, out);
+ afr_handle_anon_inode_options(priv, this->options);
+ GF_OPTION_INIT("use-anonymous-inode", priv->use_anon_inode, bool, out);
if (priv->quorum_count != 0)
priv->consistent_io = _gf_false;
@@ -553,6 +580,9 @@ init(xlator_t *this)
goto out;
}
+ priv->anon_inode = GF_CALLOC(sizeof(unsigned char), child_count,
+ gf_afr_mt_char);
+
priv->child_up = GF_CALLOC(sizeof(unsigned char), child_count,
gf_afr_mt_char);
@@ -561,7 +591,8 @@ init(xlator_t *this)
priv->halo_child_up = GF_CALLOC(sizeof(unsigned char), child_count,
gf_afr_mt_char);
- if (!priv->child_up || !priv->child_latency || !priv->halo_child_up) {
+ if (!priv->child_up || !priv->child_latency || !priv->halo_child_up ||
+ !priv->anon_inode) {
ret = -ENOMEM;
goto out;
}
@@ -1286,6 +1317,14 @@ struct volume_options options[] = {
.tags = {"replicate"},
.description = "This option exists only for backward compatibility "
"and configuring it doesn't have any effect"},
+ {.key = {"use-anonymous-inode"},
+ .type = GF_OPTION_TYPE_BOOL,
+ .default_value = "no",
+ .op_version = {GD_OP_VERSION_8_0},
+ .flags = OPT_FLAG_CLIENT_OPT | OPT_FLAG_SETTABLE,
+ .tags = {"replicate"},
+ .description = "Setting this option heals directory renames efficiently"},
+
{.key = {NULL}},
};
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 2e60708accf..d62f9a9caf2 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -42,6 +42,7 @@
#define AFR_LK_HEAL_DOM "afr.lock-heal.domain"
#define AFR_HALO_MAX_LATENCY 99999
+#define AFR_ANON_DIR_PREFIX ".glusterfs-anonymous-inode"
#define PFLAG_PENDING (1 << 0)
#define PFLAG_SBRAIN (1 << 1)
@@ -190,6 +191,7 @@ typedef struct _afr_private {
struct list_head ta_waitq;
struct list_head ta_onwireq;
+ unsigned char *anon_inode;
unsigned char *child_up;
unsigned char *halo_child_up;
int64_t *child_latency;
@@ -275,10 +277,15 @@ typedef struct _afr_private {
gf_boolean_t esh_granular;
gf_boolean_t consistent_io;
gf_boolean_t data_self_heal; /* on/off */
+ gf_boolean_t use_anon_inode;
/*For lock healing.*/
struct list_head saved_locks;
struct list_head lk_healq;
+
+ /*For anon-inode handling */
+ char anon_inode_name[NAME_MAX + 1];
+ char anon_gfid_str[UUID_SIZE + 1];
} afr_private_t;
typedef enum {
@@ -1409,4 +1416,8 @@ afr_dom_lock_release(call_frame_t *frame);
void
afr_fill_success_replies(afr_local_t *local, afr_private_t *priv,
unsigned char *replies);
+
+gf_boolean_t
+afr_is_private_directory(afr_private_t *priv, uuid_t pargfid, const char *name,
+ pid_t pid);
#endif /* __AFR_H__ */
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index d874415699c..8ba0cc4c732 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -382,7 +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;
@@ -2161,31 +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
@@ -4314,6 +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) {
@@ -4604,18 +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);
@@ -5893,22 +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)
@@ -6720,10 +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;
@@ -10855,15 +10818,12 @@ dht_notify(xlator_t *this, int event, void *data, ...)
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++) {
@@ -11086,15 +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);
}
}
}
@@ -11239,28 +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
@@ -11295,14 +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)
{
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 07bbe204997..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)
@@ -242,19 +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;
@@ -272,7 +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;
@@ -405,14 +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,
@@ -463,75 +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*/
@@ -563,13 +471,6 @@ struct gf_defrag_info_ {
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;
@@ -605,7 +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);
};
@@ -1316,9 +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);
@@ -1482,4 +1379,6 @@ 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-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 358e1c6d533..8ba8082bd86 100644
--- a/xlators/cluster/dht/src/dht-rebalance.c
+++ b/xlators/cluster/dht/src/dht-rebalance.c
@@ -609,26 +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);
@@ -1096,7 +1093,7 @@ __dht_rebalance_migrate_data(xlator_t *this, gf_defrag_info_t *defrag,
break;
}
- if (!conf->force_migration && !dht_is_tier_xlator(this)) {
+ if (!conf->force_migration) {
if (!xdata) {
xdata = dict_new();
if (!xdata) {
@@ -1536,21 +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);
@@ -2300,14 +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)
@@ -3073,7 +3053,7 @@ 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 = 0;
char is_linkfile = 0;
@@ -3277,7 +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);
@@ -3290,6 +3270,9 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
goto out;
}
+ 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");
@@ -3454,7 +3437,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf,
defrag, dir_dfmeta->lfd[dfc_index],
migrate_data, dir_dfmeta, xattr_req,
- &should_commit_hash, perrno);
+ perrno);
if (defrag->defrag_status == GF_DEFRAG_STATUS_STOPPED) {
goto out;
@@ -3505,16 +3488,12 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
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 (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.
@@ -3522,6 +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)
@@ -3536,7 +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;
}
@@ -3582,114 +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)
@@ -3709,7 +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;
@@ -3812,16 +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);
@@ -3838,8 +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;
}
}
@@ -3902,7 +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;
}
}
@@ -3920,7 +3777,7 @@ gf_defrag_fix_layout(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
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);
@@ -3987,11 +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;
@@ -4007,18 +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,
@@ -4042,10 +3893,6 @@ out:
if (fd)
fd_unref(fd);
- if (ret == 0 && should_commit_hash == 0) {
- ret = 2;
- }
-
return ret;
}
@@ -4054,31 +3901,26 @@ 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) {
@@ -4252,7 +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) {
@@ -4309,7 +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. ",
@@ -4552,14 +4394,13 @@ gf_defrag_start_crawl(void *data)
}
ret = gf_defrag_fix_layout(this, defrag, &loc, fix_layout, migrate_data);
- if (ret && ret != 2) {
+ if (ret) {
defrag->total_failures++;
ret = -1;
goto out;
}
- if (ret != 2 &&
- gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
+ if (gf_defrag_settle_hash(this, defrag, &loc, fix_layout) != 0) {
defrag->total_failures++;
ret = -1;
goto out;
@@ -4731,8 +4572,6 @@ 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;
uint64_t time_to_complete = 0;
@@ -4751,15 +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;
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))
@@ -4774,14 +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");
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index 1b6571cd43c..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);
@@ -1330,9 +1326,11 @@ dht_selfheal_dir_mkdir(call_frame_t *frame, loc_t *loc, dht_layout_t *layout,
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;
@@ -1372,15 +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];
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
index 96caf40d0b1..bb72b0ffbb5 100644
--- a/xlators/cluster/dht/src/dht-shared.c
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -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 2024-05-13 09:08:35 +0000