summaryrefslogtreecommitdiffstats
path: root/xlators
diff options
context:
space:
mode:
authorvmallika <vmallika@redhat.com>2015-03-20 14:25:58 +0530
committerVijay Bellur <vbellur@redhat.com>2015-03-31 23:38:42 -0700
commit3d4203fe3acad47f0a29ebe094b3e427b13f363f (patch)
treeae8b9ff2290cc57c75a8d273cd0fa2d908445933 /xlators
parentd6b6c32ac7f0a6b3d77009aec8bdd0cd289bd0ff (diff)
quota: enhancement of build ancestry
There can a small race window where marker accounting will be missed. Consider below example where NFS mount is used and bricks are restarted and a write operation is performed Currently: T1: lookup from protocol server, quota will initiate ancestry build T2: opendir FOP from ancestry build T3: write FOP from client T4: write_cbk T5: marker initiate accounting. There will be a problem here and txn aborts, because build_ancestry is not complete and parent not found for an inode T6: opendir_cbk T7: initiate readdirp, posix builds ancestry in readdirp With this patch, now calling readdirp on anonoymous fd during build ancestry: T1: lookup from protocol server, quota will initiate ancestry build T2: readirp FOP from ancestry build, posix builds ancestry in readdirp T3: write FOP from client T4: write_cbk T5: marker initiate accounting. No problem here as build ancestry was performed at T1 Change-Id: I2c262c86f34f6c574940a6b8772d94f2bade9465 BUG: 1184885 Signed-off-by: vmallika <vmallika@redhat.com> Reviewed-on: http://review.gluster.org/9954 Reviewed-by: Vijay Bellur <vbellur@redhat.com> Tested-by: Vijay Bellur <vbellur@redhat.com>
Diffstat (limited to 'xlators')
-rw-r--r--xlators/features/quota/src/quota.c135
1 files changed, 45 insertions, 90 deletions
diff --git a/xlators/features/quota/src/quota.c b/xlators/features/quota/src/quota.c
index 2afe58f6210..3f94498f5ad 100644
--- a/xlators/features/quota/src/quota.c
+++ b/xlators/features/quota/src/quota.c
@@ -751,130 +751,85 @@ cleanup:
return 0;
}
-int32_t
-quota_build_ancestry_open_cbk (call_frame_t *frame, void *cookie,
- xlator_t *this, int32_t op_ret, int32_t op_errno,
- fd_t *fd, dict_t *xdata)
-{
- dict_t *xdata_req = NULL;
- quota_local_t *local = NULL;
-
- if (op_ret < 0) {
- goto err;
- }
-
- xdata_req = dict_new ();
- if (xdata_req == NULL) {
- op_ret = -ENOMEM;
- goto err;
- }
-
- op_ret = dict_set_int8 (xdata_req, QUOTA_LIMIT_KEY, 1);
- if (op_ret < 0) {
- op_errno = -op_ret;
- goto err;
- }
-
- op_ret = dict_set_int8 (xdata_req, GET_ANCESTRY_DENTRY_KEY, 1);
- if (op_ret < 0) {
- op_errno = -op_ret;
- goto err;
- }
-
- /* This would ask posix layer to construct dentry chain till root */
- STACK_WIND (frame, quota_build_ancestry_cbk, FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->readdirp, fd, 0, 0, xdata_req);
-
- op_ret = 0;
-
-err:
- fd_unref (fd);
-
- if (xdata_req)
- dict_unref (xdata_req);
-
- if (op_ret < 0) {
- local = frame->local;
- frame->local = NULL;
-
- local->ancestry_cbk (NULL, NULL, -1, op_errno,
- local->ancestry_data);
- quota_local_cleanup (this, local);
- STACK_DESTROY (frame->root);
- }
-
- return 0;
-}
-
int
quota_build_ancestry (inode_t *inode, quota_ancestry_built_t ancestry_cbk,
void *data)
{
- loc_t loc = {0, };
fd_t *fd = NULL;
quota_local_t *local = NULL;
call_frame_t *new_frame = NULL;
- int op_errno = EINVAL;
+ int op_errno = ENOMEM;
+ int op_ret = -1;
xlator_t *this = NULL;
+ dict_t *xdata_req = NULL;
this = THIS;
- loc.inode = inode_ref (inode);
- uuid_copy (loc.gfid, inode->gfid);
+ xdata_req = dict_new ();
+ if (xdata_req == NULL)
+ goto err;
- fd = fd_create (inode, 0);
+ fd = fd_anonymous (inode);
+ if (fd == NULL)
+ goto err;
new_frame = create_frame (this, this->ctx->pool);
- if (new_frame == NULL) {
- op_errno = ENOMEM;
+ if (new_frame == NULL)
goto err;
- }
-
- new_frame->root->uid = new_frame->root->gid = 0;
local = quota_local_new ();
- if (local == NULL) {
- op_errno = ENOMEM;
+ if (local == NULL)
goto err;
- }
+ new_frame->root->uid = new_frame->root->gid = 0;
new_frame->local = local;
local->ancestry_cbk = ancestry_cbk;
local->ancestry_data = data;
local->loc.inode = inode_ref (inode);
- if (IA_ISDIR (inode->ia_type)) {
- STACK_WIND (new_frame, quota_build_ancestry_open_cbk,
- FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->opendir, &loc, fd,
- NULL);
- } else {
- STACK_WIND (new_frame, quota_build_ancestry_open_cbk,
- FIRST_CHILD(this),
- FIRST_CHILD(this)->fops->open, &loc, 0, fd,
- NULL);
+ op_ret = dict_set_int8 (xdata_req, QUOTA_LIMIT_KEY, 1);
+ if (op_ret < 0) {
+ op_errno = -op_ret;
+ goto err;
}
- loc_wipe (&loc);
- return 0;
+ op_ret = dict_set_int8 (xdata_req, GET_ANCESTRY_DENTRY_KEY, 1);
+ if (op_ret < 0) {
+ op_errno = -op_ret;
+ goto err;
+ }
+
+ /* This would ask posix layer to construct dentry chain till root
+ * We don't need to do a opendir, we can use the anonymous fd
+ * here for the readidrp.
+ * avoiding opendir also reduces the window size where another FOP
+ * can be executed before completion of build ancestry
+ */
+ STACK_WIND (new_frame, quota_build_ancestry_cbk, FIRST_CHILD(this),
+ FIRST_CHILD(this)->fops->readdirp, fd, 0, 0, xdata_req);
+
+ op_ret = 0;
err:
- ancestry_cbk (NULL, NULL, -1, op_errno, data);
+ if (fd)
+ fd_unref (fd);
- fd_unref (fd);
+ if (xdata_req)
+ dict_unref (xdata_req);
- local = new_frame->local;
- new_frame->local = NULL;
+ if (op_ret < 0) {
+ ancestry_cbk (NULL, NULL, -1, op_errno, data);
- if (local != NULL) {
- quota_local_cleanup (this, local);
- }
+ if (new_frame) {
+ local = new_frame->local;
+ new_frame->local = NULL;
+ STACK_DESTROY (new_frame->root);
+ }
- if (new_frame != NULL) {
- STACK_DESTROY (new_frame->root);
+ if (local)
+ quota_local_cleanup (this, local);
}
- loc_wipe (&loc);
return 0;
}