|
|
1df6c8 |
From 3d230880aed85737365deafe3c9a32c67da2a79e Mon Sep 17 00:00:00 2001
|
|
|
1df6c8 |
From: Susant Palai <spalai@redhat.com>
|
|
|
1df6c8 |
Date: Mon, 4 May 2020 19:09:00 +0530
|
|
|
1df6c8 |
Subject: [PATCH 370/375] dht: Do opendir selectively in gf_defrag_process_dir
|
|
|
1df6c8 |
|
|
|
1df6c8 |
Currently opendir is done from the cluster view. Hence, even if
|
|
|
1df6c8 |
one opendir is successful, the opendir operation as a whole is considered
|
|
|
1df6c8 |
successful.
|
|
|
1df6c8 |
|
|
|
1df6c8 |
But since in gf_defrag_get_entry we fetch entries selectively from
|
|
|
1df6c8 |
local_subvols, we need to opendir individually on those local subvols
|
|
|
1df6c8 |
and keep track of fds separately. Otherwise it is possible that opendir
|
|
|
1df6c8 |
failed on one of the subvol and we wind readdirp call on the fd to the
|
|
|
1df6c8 |
corresponding subvol, which will ultimately result in EINVAL error.
|
|
|
1df6c8 |
|
|
|
1df6c8 |
> fixes: #1218
|
|
|
1df6c8 |
> Change-Id: I50dd88b9597852a15579f4ee325918979417f570
|
|
|
1df6c8 |
> Signed-off-by: Susant Palai <spalai@redhat.com>
|
|
|
1df6c8 |
(Backport of https://review.gluster.org/#/c/glusterfs/+/24404/)
|
|
|
1df6c8 |
|
|
|
1df6c8 |
BUG: 1831403
|
|
|
1df6c8 |
Change-Id: I96e19fdd630279c3ef44f361c1d1fc5c1c429821
|
|
|
1df6c8 |
Signed-off-by: Susant Palai <spalai@redhat.com>
|
|
|
1df6c8 |
Reviewed-on: https://code.engineering.redhat.com/gerrit/200306
|
|
|
1df6c8 |
Tested-by: RHGS Build Bot <nigelb@redhat.com>
|
|
|
1df6c8 |
Reviewed-by: Sunil Kumar Heggodu Gopala Acharya <sheggodu@redhat.com>
|
|
|
1df6c8 |
---
|
|
|
1df6c8 |
xlators/cluster/dht/src/dht-common.h | 2 +
|
|
|
1df6c8 |
xlators/cluster/dht/src/dht-rebalance.c | 74 +++++++++++++++++++++++----------
|
|
|
1df6c8 |
2 files changed, 54 insertions(+), 22 deletions(-)
|
|
|
1df6c8 |
|
|
|
1df6c8 |
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
|
|
|
1df6c8 |
index 4d2aae6..8e65111 100644
|
|
|
1df6c8 |
--- a/xlators/cluster/dht/src/dht-common.h
|
|
|
1df6c8 |
+++ b/xlators/cluster/dht/src/dht-common.h
|
|
|
1df6c8 |
@@ -742,6 +742,8 @@ struct dir_dfmeta {
|
|
|
1df6c8 |
struct list_head **head;
|
|
|
1df6c8 |
struct list_head **iterator;
|
|
|
1df6c8 |
int *fetch_entries;
|
|
|
1df6c8 |
+ /* fds corresponding to local subvols only */
|
|
|
1df6c8 |
+ fd_t **lfd;
|
|
|
1df6c8 |
};
|
|
|
1df6c8 |
|
|
|
1df6c8 |
typedef struct dht_migrate_info {
|
|
|
1df6c8 |
diff --git a/xlators/cluster/dht/src/dht-rebalance.c b/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
1df6c8 |
index 33cacfe..c692119 100644
|
|
|
1df6c8 |
--- a/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
1df6c8 |
+++ b/xlators/cluster/dht/src/dht-rebalance.c
|
|
|
1df6c8 |
@@ -48,6 +48,8 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
|
|
|
1df6c8 |
if (meta) {
|
|
|
1df6c8 |
for (i = 0; i < local_subvols_cnt; i++) {
|
|
|
1df6c8 |
gf_dirent_free(&meta->equeue[i]);
|
|
|
1df6c8 |
+ if (meta->lfd && meta->lfd[i])
|
|
|
1df6c8 |
+ fd_unref(meta->lfd[i]);
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
GF_FREE(meta->equeue);
|
|
|
1df6c8 |
@@ -55,6 +57,7 @@ gf_defrag_free_dir_dfmeta(struct dir_dfmeta *meta, int local_subvols_cnt)
|
|
|
1df6c8 |
GF_FREE(meta->iterator);
|
|
|
1df6c8 |
GF_FREE(meta->offset_var);
|
|
|
1df6c8 |
GF_FREE(meta->fetch_entries);
|
|
|
1df6c8 |
+ GF_FREE(meta->lfd);
|
|
|
1df6c8 |
GF_FREE(meta);
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
@@ -3095,7 +3098,7 @@ int static gf_defrag_get_entry(xlator_t *this, int i,
|
|
|
1df6c8 |
struct dir_dfmeta *dir_dfmeta, dict_t *xattr_req,
|
|
|
1df6c8 |
int *should_commit_hash, int *perrno)
|
|
|
1df6c8 |
{
|
|
|
1df6c8 |
- int ret = -1;
|
|
|
1df6c8 |
+ int ret = 0;
|
|
|
1df6c8 |
char is_linkfile = 0;
|
|
|
1df6c8 |
gf_dirent_t *df_entry = NULL;
|
|
|
1df6c8 |
struct dht_container *tmp_container = NULL;
|
|
|
1df6c8 |
@@ -3111,6 +3114,13 @@ int static gf_defrag_get_entry(xlator_t *this, int i,
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
if (dir_dfmeta->fetch_entries[i] == 1) {
|
|
|
1df6c8 |
+ if (!fd) {
|
|
|
1df6c8 |
+ dir_dfmeta->fetch_entries[i] = 0;
|
|
|
1df6c8 |
+ dir_dfmeta->offset_var[i].readdir_done = 1;
|
|
|
1df6c8 |
+ ret = 0;
|
|
|
1df6c8 |
+ goto out;
|
|
|
1df6c8 |
+ }
|
|
|
1df6c8 |
+
|
|
|
1df6c8 |
ret = syncop_readdirp(conf->local_subvols[i], fd, 131072,
|
|
|
1df6c8 |
dir_dfmeta->offset_var[i].offset,
|
|
|
1df6c8 |
&(dir_dfmeta->equeue[i]), xattr_req, NULL);
|
|
|
1df6c8 |
@@ -3270,7 +3280,6 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
|
|
|
1df6c8 |
dict_t *migrate_data, int *perrno)
|
|
|
1df6c8 |
{
|
|
|
1df6c8 |
int ret = -1;
|
|
|
1df6c8 |
- fd_t *fd = NULL;
|
|
|
1df6c8 |
dht_conf_t *conf = NULL;
|
|
|
1df6c8 |
gf_dirent_t entries;
|
|
|
1df6c8 |
dict_t *xattr_req = NULL;
|
|
|
1df6c8 |
@@ -3304,28 +3313,49 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
|
|
|
1df6c8 |
goto out;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
- fd = fd_create(loc->inode, defrag->pid);
|
|
|
1df6c8 |
- if (!fd) {
|
|
|
1df6c8 |
- gf_log(this->name, GF_LOG_ERROR, "Failed to create fd");
|
|
|
1df6c8 |
+ dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
|
|
|
1df6c8 |
+ if (!dir_dfmeta) {
|
|
|
1df6c8 |
+ gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
|
|
|
1df6c8 |
ret = -1;
|
|
|
1df6c8 |
goto out;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
- ret = syncop_opendir(this, loc, fd, NULL, NULL);
|
|
|
1df6c8 |
- if (ret) {
|
|
|
1df6c8 |
- gf_msg(this->name, GF_LOG_WARNING, -ret, DHT_MSG_MIGRATE_DATA_FAILED,
|
|
|
1df6c8 |
- "Migrate data failed: Failed to open dir %s", loc->path);
|
|
|
1df6c8 |
- *perrno = -ret;
|
|
|
1df6c8 |
+ dir_dfmeta->lfd = GF_CALLOC(local_subvols_cnt, sizeof(fd_t *),
|
|
|
1df6c8 |
+ gf_common_mt_pointer);
|
|
|
1df6c8 |
+ if (!dir_dfmeta->lfd) {
|
|
|
1df6c8 |
+ gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
|
|
|
1df6c8 |
+ "could not allocate memory for dir_dfmeta");
|
|
|
1df6c8 |
ret = -1;
|
|
|
1df6c8 |
+ *perrno = ENOMEM;
|
|
|
1df6c8 |
goto out;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
- fd_bind(fd);
|
|
|
1df6c8 |
- dir_dfmeta = GF_CALLOC(1, sizeof(*dir_dfmeta), gf_common_mt_pointer);
|
|
|
1df6c8 |
- if (!dir_dfmeta) {
|
|
|
1df6c8 |
- gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta is NULL");
|
|
|
1df6c8 |
- ret = -1;
|
|
|
1df6c8 |
- goto out;
|
|
|
1df6c8 |
+ for (i = 0; i < local_subvols_cnt; i++) {
|
|
|
1df6c8 |
+ dir_dfmeta->lfd[i] = fd_create(loc->inode, defrag->pid);
|
|
|
1df6c8 |
+ if (!dir_dfmeta->lfd[i]) {
|
|
|
1df6c8 |
+ gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0, "failed to create fd");
|
|
|
1df6c8 |
+ *perrno = ENOMEM;
|
|
|
1df6c8 |
+ ret = -1;
|
|
|
1df6c8 |
+ goto out;
|
|
|
1df6c8 |
+ }
|
|
|
1df6c8 |
+
|
|
|
1df6c8 |
+ ret = syncop_opendir(conf->local_subvols[i], loc, dir_dfmeta->lfd[i],
|
|
|
1df6c8 |
+ NULL, NULL);
|
|
|
1df6c8 |
+ if (ret) {
|
|
|
1df6c8 |
+ fd_unref(dir_dfmeta->lfd[i]);
|
|
|
1df6c8 |
+ dir_dfmeta->lfd[i] = NULL;
|
|
|
1df6c8 |
+ gf_smsg(this->name, GF_LOG_WARNING, 0, 0,
|
|
|
1df6c8 |
+ "failed to open dir: %s subvol: %s", loc->path,
|
|
|
1df6c8 |
+ conf->local_subvols[i]->name);
|
|
|
1df6c8 |
+
|
|
|
1df6c8 |
+ if (conf->decommission_in_progress) {
|
|
|
1df6c8 |
+ *perrno = -ret;
|
|
|
1df6c8 |
+ ret = -1;
|
|
|
1df6c8 |
+ goto out;
|
|
|
1df6c8 |
+ }
|
|
|
1df6c8 |
+ } else {
|
|
|
1df6c8 |
+ fd_bind(dir_dfmeta->lfd[i]);
|
|
|
1df6c8 |
+ }
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
|
|
|
1df6c8 |
dir_dfmeta->head = GF_CALLOC(local_subvols_cnt, sizeof(*(dir_dfmeta->head)),
|
|
|
1df6c8 |
@@ -3360,6 +3390,7 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
|
|
|
1df6c8 |
ret = -1;
|
|
|
1df6c8 |
goto out;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
+
|
|
|
1df6c8 |
ret = gf_defrag_ctx_subvols_init(dir_dfmeta->offset_var, this);
|
|
|
1df6c8 |
if (ret) {
|
|
|
1df6c8 |
gf_log(this->name, GF_LOG_ERROR,
|
|
|
1df6c8 |
@@ -3372,7 +3403,8 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
|
|
|
1df6c8 |
dir_dfmeta->fetch_entries = GF_CALLOC(local_subvols_cnt, sizeof(int),
|
|
|
1df6c8 |
gf_common_mt_int);
|
|
|
1df6c8 |
if (!dir_dfmeta->fetch_entries) {
|
|
|
1df6c8 |
- gf_log(this->name, GF_LOG_ERROR, "dir_dfmeta->fetch_entries is NULL");
|
|
|
1df6c8 |
+ gf_msg(this->name, GF_LOG_ERROR, ENOMEM, 0,
|
|
|
1df6c8 |
+ "could not allocate memory for dir_dfmeta->fetch_entries");
|
|
|
1df6c8 |
ret = -1;
|
|
|
1df6c8 |
goto out;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
@@ -3442,8 +3474,9 @@ gf_defrag_process_dir(xlator_t *this, gf_defrag_info_t *defrag, loc_t *loc,
|
|
|
1df6c8 |
ldfq_count <= MAX_MIGRATE_QUEUE_COUNT &&
|
|
|
1df6c8 |
!dht_dfreaddirp_done(dir_dfmeta->offset_var, local_subvols_cnt)) {
|
|
|
1df6c8 |
ret = gf_defrag_get_entry(this, dfc_index, &container, loc, conf,
|
|
|
1df6c8 |
- defrag, fd, migrate_data, dir_dfmeta,
|
|
|
1df6c8 |
- xattr_req, &should_commit_hash, perrno);
|
|
|
1df6c8 |
+ defrag, dir_dfmeta->lfd[dfc_index],
|
|
|
1df6c8 |
+ migrate_data, dir_dfmeta, xattr_req,
|
|
|
1df6c8 |
+ &should_commit_hash, perrno);
|
|
|
1df6c8 |
|
|
|
1df6c8 |
if (ret) {
|
|
|
1df6c8 |
gf_log(this->name, GF_LOG_WARNING,
|
|
|
1df6c8 |
@@ -3497,9 +3530,6 @@ out:
|
|
|
1df6c8 |
if (xattr_req)
|
|
|
1df6c8 |
dict_unref(xattr_req);
|
|
|
1df6c8 |
|
|
|
1df6c8 |
- if (fd)
|
|
|
1df6c8 |
- fd_unref(fd);
|
|
|
1df6c8 |
-
|
|
|
1df6c8 |
if (ret == 0 && should_commit_hash == 0) {
|
|
|
1df6c8 |
ret = 2;
|
|
|
1df6c8 |
}
|
|
|
1df6c8 |
--
|
|
|
1df6c8 |
1.8.3.1
|
|
|
1df6c8 |
|