|
 |
3604df |
From 90f7e532ea9ab8136d9f9bbd8e80d61390cd667e Mon Sep 17 00:00:00 2001
|
|
|
3604df |
From: Poornima G <pgurusid@redhat.com>
|
|
|
3604df |
Date: Mon, 21 Nov 2016 11:49:35 +0530
|
|
|
3604df |
Subject: [PATCH 202/206] afr: Fix the EIO that can occur in afr_inode_refresh
|
|
|
3604df |
as a result of cache invalidation(upcall).
|
|
|
3604df |
|
|
|
3604df |
Backport of http://review.gluster.org/15892
|
|
|
3604df |
Conflicts:
|
|
|
3604df |
xlators/cluster/afr/src/afr.h
|
|
|
3604df |
|
|
|
3604df |
Issue:
|
|
|
3604df |
------
|
|
|
3604df |
When a cache invalidation is recieved as a result of changing
|
|
|
3604df |
pending xattr, the read_subvol is reset. Consider the below chain
|
|
|
3604df |
of execution:
|
|
|
3604df |
|
|
|
3604df |
CHILD_DOWN
|
|
|
3604df |
...
|
|
|
3604df |
afr_readv
|
|
|
3604df |
...
|
|
|
3604df |
afr_inode_refresh
|
|
|
3604df |
...
|
|
|
3604df |
afr_inode_read_subvol_reset <- as a result of pending xattr set by
|
|
|
3604df |
some other client GF_EVENT_UPCALL will
|
|
|
3604df |
be sent
|
|
|
3604df |
afr_refresh_done -> this results in an EIO, as the read subvol was
|
|
|
3604df |
reset by the end of the afr_inode_refresh
|
|
|
3604df |
|
|
|
3604df |
Solution:
|
|
|
3604df |
---------
|
|
|
3604df |
When GF_EVENT_UPCALL is recieved, instead of resetting read_subvol,
|
|
|
3604df |
set a variable need_refresh in inode_ctx, the next time some one
|
|
|
3604df |
starts a txn, along with event gen, need_rrefresh also needs to
|
|
|
3604df |
be checked.
|
|
|
3604df |
|
|
|
3604df |
>Reviewed-on: http://review.gluster.org/15892
|
|
|
3604df |
>Reviewed-by: Ravishankar N <ravishankar@redhat.com>
|
|
|
3604df |
>Smoke: Gluster Build System <jenkins@build.gluster.org>
|
|
|
3604df |
>NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
|
|
|
3604df |
>CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
|
|
|
3604df |
>Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
3604df |
|
|
|
3604df |
Change-Id: Ifda21a7a8039b8874215e1afa4bdf20f7d991b58
|
|
|
3604df |
BUG: 1393758
|
|
|
3604df |
Signed-off-by: Poornima G <pgurusid@redhat.com>
|
|
|
3604df |
Reviewed-on: https://code.engineering.redhat.com/gerrit/91495
|
|
|
3604df |
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
3604df |
Tested-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
|
|
|
3604df |
---
|
|
|
3604df |
xlators/cluster/afr/src/afr-common.c | 90 +++++++++++++++++++++++++------
|
|
|
3604df |
xlators/cluster/afr/src/afr-read-txn.c | 3 +-
|
|
|
3604df |
xlators/cluster/afr/src/afr-transaction.c | 4 +-
|
|
|
3604df |
xlators/cluster/afr/src/afr.h | 4 ++
|
|
|
3604df |
4 files changed, 83 insertions(+), 18 deletions(-)
|
|
|
3604df |
|
|
|
3604df |
diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
|
|
|
3604df |
index 2d9b496..1d1882f 100644
|
|
|
3604df |
--- a/xlators/cluster/afr/src/afr-common.c
|
|
|
3604df |
+++ b/xlators/cluster/afr/src/afr-common.c
|
|
|
3604df |
@@ -595,6 +595,62 @@ out:
|
|
|
3604df |
}
|
|
|
3604df |
|
|
|
3604df |
|
|
|
3604df |
+/* The caller of this should perform afr_inode_refresh, if this function
|
|
|
3604df |
+ * returns _gf_true
|
|
|
3604df |
+ */
|
|
|
3604df |
+gf_boolean_t
|
|
|
3604df |
+afr_is_inode_refresh_reqd (inode_t *inode, xlator_t *this,
|
|
|
3604df |
+ int event_gen1, int event_gen2)
|
|
|
3604df |
+{
|
|
|
3604df |
+ gf_boolean_t need_refresh = _gf_false;
|
|
|
3604df |
+ afr_inode_ctx_t *ctx = NULL;
|
|
|
3604df |
+ int ret = -1;
|
|
|
3604df |
+
|
|
|
3604df |
+ GF_VALIDATE_OR_GOTO (this->name, inode, out);
|
|
|
3604df |
+
|
|
|
3604df |
+ LOCK(&inode->lock);
|
|
|
3604df |
+ {
|
|
|
3604df |
+ ret = __afr_inode_ctx_get (this, inode, &ctx;;
|
|
|
3604df |
+ if (ret)
|
|
|
3604df |
+ goto unlock;
|
|
|
3604df |
+
|
|
|
3604df |
+ need_refresh = ctx->need_refresh;
|
|
|
3604df |
+ /* Hoping that the caller will do inode_refresh followed by
|
|
|
3604df |
+ * this, hence setting the need_refresh to false */
|
|
|
3604df |
+ ctx->need_refresh = _gf_false;
|
|
|
3604df |
+ }
|
|
|
3604df |
+unlock:
|
|
|
3604df |
+ UNLOCK(&inode->lock);
|
|
|
3604df |
+
|
|
|
3604df |
+ if (event_gen1 != event_gen2)
|
|
|
3604df |
+ need_refresh = _gf_true;
|
|
|
3604df |
+out:
|
|
|
3604df |
+ return need_refresh;
|
|
|
3604df |
+}
|
|
|
3604df |
+
|
|
|
3604df |
+
|
|
|
3604df |
+static int
|
|
|
3604df |
+afr_inode_need_refresh_set (inode_t *inode, xlator_t *this)
|
|
|
3604df |
+{
|
|
|
3604df |
+ int ret = -1;
|
|
|
3604df |
+ afr_inode_ctx_t *ctx = NULL;
|
|
|
3604df |
+
|
|
|
3604df |
+ GF_VALIDATE_OR_GOTO (this->name, inode, out);
|
|
|
3604df |
+
|
|
|
3604df |
+ LOCK(&inode->lock);
|
|
|
3604df |
+ {
|
|
|
3604df |
+ ret = __afr_inode_ctx_get (this, inode, &ctx;;
|
|
|
3604df |
+ if (ret)
|
|
|
3604df |
+ goto unlock;
|
|
|
3604df |
+
|
|
|
3604df |
+ ctx->need_refresh = _gf_true;
|
|
|
3604df |
+ }
|
|
|
3604df |
+unlock:
|
|
|
3604df |
+ UNLOCK(&inode->lock);
|
|
|
3604df |
+out:
|
|
|
3604df |
+ return ret;
|
|
|
3604df |
+}
|
|
|
3604df |
+
|
|
|
3604df |
int
|
|
|
3604df |
afr_inode_read_subvol_reset (inode_t *inode, xlator_t *this)
|
|
|
3604df |
{
|
|
|
3604df |
@@ -2701,7 +2757,8 @@ afr_discover (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req
|
|
|
3604df |
afr_read_subvol_get (loc->inode, this, NULL, NULL, &event,
|
|
|
3604df |
AFR_DATA_TRANSACTION, NULL);
|
|
|
3604df |
|
|
|
3604df |
- if (event != local->event_generation)
|
|
|
3604df |
+ if (afr_is_inode_refresh_reqd (loc->inode, this, event,
|
|
|
3604df |
+ local->event_generation))
|
|
|
3604df |
afr_inode_refresh (frame, this, loc->inode, NULL,
|
|
|
3604df |
afr_discover_do);
|
|
|
3604df |
else
|
|
|
3604df |
@@ -2852,7 +2909,8 @@ afr_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
|
|
|
3604df |
afr_read_subvol_get (loc->parent, this, NULL, NULL, &event,
|
|
|
3604df |
AFR_DATA_TRANSACTION, NULL);
|
|
|
3604df |
|
|
|
3604df |
- if (event != local->event_generation)
|
|
|
3604df |
+ if (afr_is_inode_refresh_reqd (loc->inode, this, event,
|
|
|
3604df |
+ local->event_generation))
|
|
|
3604df |
afr_inode_refresh (frame, this, loc->parent, NULL,
|
|
|
3604df |
afr_lookup_do);
|
|
|
3604df |
else
|
|
|
3604df |
@@ -4627,22 +4685,22 @@ afr_notify (xlator_t *this, int32_t event,
|
|
|
3604df |
* pronounced. Hence when a pending xattr is set notify
|
|
|
3604df |
* all the md-cache clients to invalidate the existing
|
|
|
3604df |
* stat cache and send the lookup next time */
|
|
|
3604df |
- if (up_ci->dict) {
|
|
|
3604df |
- for (i = 0; i < priv->child_count; i++) {
|
|
|
3604df |
- if (dict_get (up_ci->dict, priv->pending_key[i])) {
|
|
|
3604df |
- ret = dict_set_int8 (up_ci->dict,
|
|
|
3604df |
- MDC_INVALIDATE_IATT , 0);
|
|
|
3604df |
- break;
|
|
|
3604df |
- }
|
|
|
3604df |
+ if (!up_ci->dict)
|
|
|
3604df |
+ break;
|
|
|
3604df |
+ for (i = 0; i < priv->child_count; i++) {
|
|
|
3604df |
+ if (dict_get (up_ci->dict, priv->pending_key[i])) {
|
|
|
3604df |
+ ret = dict_set_int8 (up_ci->dict,
|
|
|
3604df |
+ MDC_INVALIDATE_IATT, 0);
|
|
|
3604df |
+ itable = ((xlator_t *)this->graph->top)->itable;
|
|
|
3604df |
+ /*Internal processes may not have itable for top xlator*/
|
|
|
3604df |
+ if (itable)
|
|
|
3604df |
+ inode = inode_find (itable, up_data->gfid);
|
|
|
3604df |
+ if (inode)
|
|
|
3604df |
+ afr_inode_need_refresh_set (inode, this);
|
|
|
3604df |
+
|
|
|
3604df |
+ break;
|
|
|
3604df |
}
|
|
|
3604df |
}
|
|
|
3604df |
- itable = ((xlator_t *)this->graph->top)->itable;
|
|
|
3604df |
- /*Internal processes may not have itable for top xlator*/
|
|
|
3604df |
- if (itable)
|
|
|
3604df |
- inode = inode_find (itable, up_data->gfid);
|
|
|
3604df |
- if (inode)
|
|
|
3604df |
- afr_inode_read_subvol_reset (inode, this);
|
|
|
3604df |
-
|
|
|
3604df |
break;
|
|
|
3604df |
default:
|
|
|
3604df |
propagate = 1;
|
|
|
3604df |
diff --git a/xlators/cluster/afr/src/afr-read-txn.c b/xlators/cluster/afr/src/afr-read-txn.c
|
|
|
3604df |
index 1c7b4f9..fa98d37 100644
|
|
|
3604df |
--- a/xlators/cluster/afr/src/afr-read-txn.c
|
|
|
3604df |
+++ b/xlators/cluster/afr/src/afr-read-txn.c
|
|
|
3604df |
@@ -221,7 +221,8 @@ afr_read_txn (call_frame_t *frame, xlator_t *this, inode_t *inode,
|
|
|
3604df |
gf_msg_debug (this->name, 0, "%s: generation now vs cached: %d, "
|
|
|
3604df |
"%d", uuid_utoa (inode->gfid), local->event_generation,
|
|
|
3604df |
event_generation);
|
|
|
3604df |
- if (local->event_generation != event_generation)
|
|
|
3604df |
+ if (afr_is_inode_refresh_reqd (inode, this, local->event_generation,
|
|
|
3604df |
+ event_generation))
|
|
|
3604df |
/* servers have disconnected / reconnected, and possibly
|
|
|
3604df |
rebooted, very likely changing the state of freshness
|
|
|
3604df |
of copies */
|
|
|
3604df |
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
|
|
|
3604df |
index 02a968e..62b680c 100644
|
|
|
3604df |
--- a/xlators/cluster/afr/src/afr-transaction.c
|
|
|
3604df |
+++ b/xlators/cluster/afr/src/afr-transaction.c
|
|
|
3604df |
@@ -2575,7 +2575,9 @@ afr_transaction (call_frame_t *frame, xlator_t *this, afr_transaction_type type)
|
|
|
3604df |
|
|
|
3604df |
ret = afr_inode_get_readable (frame, local->inode, this,
|
|
|
3604df |
local->readable, &event_generation, type);
|
|
|
3604df |
- if (ret < 0 || event_generation != priv->event_generation) {
|
|
|
3604df |
+ if (ret < 0 || afr_is_inode_refresh_reqd (local->inode, this,
|
|
|
3604df |
+ priv->event_generation,
|
|
|
3604df |
+ event_generation)) {
|
|
|
3604df |
afr_inode_refresh (frame, this, local->inode, local->loc.gfid,
|
|
|
3604df |
afr_write_txn_refresh_done);
|
|
|
3604df |
} else {
|
|
|
3604df |
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
|
|
|
3604df |
index a9cab13..c84c3af 100644
|
|
|
3604df |
--- a/xlators/cluster/afr/src/afr.h
|
|
|
3604df |
+++ b/xlators/cluster/afr/src/afr.h
|
|
|
3604df |
@@ -813,6 +813,7 @@ typedef struct _afr_inode_ctx {
|
|
|
3604df |
uint64_t read_subvol;
|
|
|
3604df |
int spb_choice;
|
|
|
3604df |
gf_timer_t *timer;
|
|
|
3604df |
+ gf_boolean_t need_refresh;
|
|
|
3604df |
} afr_inode_ctx_t;
|
|
|
3604df |
|
|
|
3604df |
typedef struct afr_spbc_timeout {
|
|
|
3604df |
@@ -1221,4 +1222,7 @@ afr_compound_cleanup (compound_args_t *args, dict_t *xdata,
|
|
|
3604df |
int32_t
|
|
|
3604df |
afr_quorum_errno (afr_private_t *priv);
|
|
|
3604df |
|
|
|
3604df |
+gf_boolean_t
|
|
|
3604df |
+afr_is_inode_refresh_reqd (inode_t *inode, xlator_t *this,
|
|
|
3604df |
+ int event_gen1, int event_gen2);
|
|
|
3604df |
#endif /* __AFR_H__ */
|
|
|
3604df |
--
|
|
|
3604df |
2.9.3
|
|
|
3604df |
|