cluster/afr: Unwind xdata_rsp even in case of failures

DHT expects GF_PREOP_CHECK_FAILED to be present in xdata_rsp in case of mkdir
failures because of stale layout. But AFR was unwinding null xdata_rsp in case
of failures. This was leading to mkdir failures just after remove-brick. Unwind
the xdata_rsp in case of failures to make sure the response from brick reaches
dht.

BUG: 1340623
Change-Id: Idd3f7b95730e8ea987b608e892011ff190e181d1
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/14553
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
Reviewed-by: Ravishankar N <ravishankar@redhat.com>
Smoke: Gluster Build System <jenkins@build.gluster.com>
CentOS-regression: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anuradha Talur <atalur@redhat.com>
Reviewed-by: Krutika Dhananjay <kdhananj@redhat.com>

7 files changed, 120 insertions, 21 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 2043e11d1e8..c20c53218f9 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -1269,6 +1269,7 @@ afr_inode_read_subvol_type_get (inode_t *inode, xlator_t *this,
 
 int
 afr_read_subvol_get (inode_t *inode, xlator_t *this, int *subvol_p,
+                     unsigned char *readables,
 		     int *event_p, afr_transaction_type type,
                      afr_read_subvol_args_t *args)
 {
@@ -1305,6 +1306,9 @@ afr_read_subvol_get (inode_t *inode, xlator_t *this, int *subvol_p,
 		*subvol_p = subvol;
 	if (event_p)
 		*event_p = event;
+        if (readables)
+                memcpy (readables, readable,
+                        sizeof (*readables) * priv->child_count);
 	return subvol;
 }
 
@@ -1434,6 +1438,7 @@ afr_local_cleanup (afr_local_t *local, xlator_t *this)
         GF_FREE (local->read_attempted);
 
         GF_FREE (local->readable);
+        GF_FREE (local->readable2);
 
 	if (local->inode)
 		inode_unref (local->inode);
@@ -1593,8 +1598,8 @@ afr_get_parent_read_subvol (xlator_t *this, inode_t *parent,
         priv = this->private;
 
         if (parent)
-                par_read_subvol = afr_data_subvol_get (parent, this, 0, 0,
-                                                       NULL);
+                par_read_subvol = afr_data_subvol_get (parent, this, NULL, NULL,
+                                                       NULL, NULL);
 
         for (i = 0; i < priv->child_count; i++) {
                 if (!replies[i].valid)
@@ -1633,8 +1638,7 @@ afr_read_subvol_decide (inode_t *inode, xlator_t *this,
         int data_subvol  = -1;
         int mdata_subvol = -1;
 
-        data_subvol = afr_data_subvol_get (inode, this,
-                                           0, 0, args);
+        data_subvol = afr_data_subvol_get (inode, this, NULL, NULL, NULL, args);
         mdata_subvol = afr_metadata_subvol_get (inode, this,
                                                 0, 0, args);
         if (data_subvol == -1 || mdata_subvol == -1)
@@ -1782,7 +1786,7 @@ afr_lookup_done (call_frame_t *frame, xlator_t *this)
 			goto cant_interpret;
 		} else {
                         read_subvol = afr_data_subvol_get (local->inode, this,
-                                                           0, 0, &args);
+                                                       NULL, NULL, NULL, &args);
 		}
 	} else {
 	cant_interpret:
@@ -2409,7 +2413,7 @@ afr_discover (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req
 		return 0;
 	}
 
-	afr_read_subvol_get (loc->inode, this, NULL, &event,
+	afr_read_subvol_get (loc->inode, this, NULL, NULL, &event,
 			     AFR_DATA_TRANSACTION, NULL);
 
 	if (event != local->event_generation)
@@ -2560,7 +2564,7 @@ afr_lookup (call_frame_t *frame, xlator_t *this, loc_t *loc, dict_t *xattr_req)
                 }
         }
 
-	afr_read_subvol_get (loc->parent, this, NULL, &event,
+	afr_read_subvol_get (loc->parent, this, NULL, NULL, &event,
 			     AFR_DATA_TRANSACTION, NULL);
 
 	if (event != local->event_generation)
@@ -2883,7 +2887,8 @@ afr_fsync_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         local = frame->local;
 
-	read_subvol = afr_data_subvol_get (local->inode, this, 0, 0, NULL);
+	read_subvol = afr_data_subvol_get (local->inode, this, NULL, NULL,
+                                           NULL, NULL);
 
         LOCK (&frame->lock);
         {
@@ -4283,6 +4288,14 @@ afr_local_init (afr_local_t *local, afr_private_t *priv, int32_t *op_errno)
 		goto out;
 	}
 
+        local->readable2 = GF_CALLOC (priv->child_count, sizeof (char),
+                                      gf_afr_mt_char);
+        if (!local->readable2) {
+                if (op_errno)
+                        *op_errno = ENOMEM;
+                goto out;
+        }
+
 	local->replies = GF_CALLOC(priv->child_count, sizeof(*local->replies),
 				   gf_afr_mt_reply_t);
 	if (!local->replies) {
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 841c64361cf..2260e5dac26 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -153,7 +153,8 @@ afr_validate_read_subvol (inode_t *inode, xlator_t *this, int par_read_subvol)
          * -1 above due to gen being 0, which is why it is OK to pass NULL for
          *  read_subvol_args here.
          */
-        entry_read_subvol = afr_data_subvol_get (inode, this, 0, 0, NULL);
+        entry_read_subvol = afr_data_subvol_get (inode, this, NULL, NULL,
+                                                 NULL, NULL);
         if (entry_read_subvol != par_read_subvol)
                 return -1;
 
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 55aec7429a7..f3de5352d7e 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -101,18 +101,21 @@ __afr_dir_write_finalize (call_frame_t *frame, xlator_t *this)
 	if (local->inode) {
 		afr_replies_interpret (frame, this, local->inode, NULL);
 		inode_read_subvol = afr_data_subvol_get (local->inode, this,
-							 NULL, NULL, &args);
+                                                       NULL, NULL, NULL, &args);
 	}
 
 	if (local->parent)
 		parent_read_subvol = afr_data_subvol_get (local->parent, this,
-							  NULL, NULL, NULL);
+                                             NULL, local->readable, NULL, NULL);
+
 	if (local->parent2)
 		parent2_read_subvol = afr_data_subvol_get (local->parent2, this,
-							   NULL, NULL, NULL);
+                                            NULL, local->readable2, NULL, NULL);
 
 	local->op_ret = -1;
 	local->op_errno = afr_final_errno (local, priv);
+        afr_pick_error_xdata (local, priv, local->parent, local->readable,
+                              local->parent2, local->readable2);
 
 	for (i = 0; i < priv->child_count; i++) {
 		if (!local->replies[i].valid)
@@ -144,6 +147,11 @@ __afr_dir_write_finalize (call_frame_t *frame, xlator_t *this)
 				local->replies[i].preparent2;
 			local->cont.dir_fop.postnewparent =
 				local->replies[i].postparent2;
+                        if (local->xdata_rsp) {
+                                dict_unref (local->xdata_rsp);
+                                local->xdata_rsp = NULL;
+                        }
+
 			if (local->replies[i].xdata)
 				local->xdata_rsp =
 					dict_ref (local->replies[i].xdata);
@@ -196,6 +204,9 @@ __afr_dir_write_fill (call_frame_t *frame, xlator_t *this, int child_index,
 	local->replies[child_index].valid = 1;
 	local->replies[child_index].op_ret = op_ret;
 	local->replies[child_index].op_errno = op_errno;
+        if (xdata)
+                local->replies[child_index].xdata = dict_ref (xdata);
+
 
 	if (op_ret >= 0) {
 		if (poststat)
@@ -208,9 +219,6 @@ __afr_dir_write_fill (call_frame_t *frame, xlator_t *this, int child_index,
 			local->replies[child_index].preparent2 = *preparent2;
 		if (postparent2)
 			local->replies[child_index].postparent2 = *postparent2;
-		if (xdata)
-			local->replies[child_index].xdata = dict_ref (xdata);
-
 		if (fd_ctx)
 			fd_ctx->opened_on[child_index] = AFR_FD_OPENED;
 	} else {
diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 36889429657..47320ce9412 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -84,7 +84,7 @@ __afr_inode_write_finalize (call_frame_t *frame, xlator_t *this)
                                                                &args);
 		else
 			read_subvol = afr_data_subvol_get (local->inode, this,
-							   NULL, NULL, &args);
+                                                       NULL, NULL, NULL, &args);
 	}
 
 	local->op_ret = -1;
@@ -164,8 +164,8 @@ __afr_inode_write_fill (call_frame_t *frame, xlator_t *this, int child_index,
 			local->replies[child_index].poststat = *postbuf;
 		if (xattr)
 			local->replies[child_index].xattr = dict_ref (xattr);
-		if (xdata)
-			local->replies[child_index].xdata = dict_ref (xdata);
+                if (xdata)
+                        local->replies[child_index].xdata = dict_ref (xdata);
 	} else {
 		afr_transaction_fop_failed (frame, this, child_index);
 	}
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index f5ff6a0428e..2760563e0ae 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -57,6 +57,66 @@ afr_zero_fill_stat (afr_local_t *local)
         }
 }
 
+/* In case of errors afr needs to choose which xdata from lower xlators it needs
+ * to unwind with. The way it is done is by checking if there are
+ * any good subvols which failed. Give preference to errnos other than
+ * ENOTCONN even if the child is source */
+void
+afr_pick_error_xdata (afr_local_t *local, afr_private_t *priv,
+                      inode_t *inode1, unsigned char *readable1,
+                      inode_t *inode2, unsigned char *readable2)
+{
+        int     s = -1;/*selection*/
+        int     i = 0;
+        unsigned char *readable = NULL;
+
+        if (local->xdata_rsp) {
+                dict_unref (local->xdata_rsp);
+                local->xdata_rsp = NULL;
+        }
+
+        readable = alloca0 (priv->child_count * sizeof (*readable));
+        if (inode2 && readable2) {/*rename fop*/
+                AFR_INTERSECT (readable, readable1, readable2,
+                               priv->child_count);
+        } else {
+                memcpy (readable, readable1,
+                        sizeof (*readable) * priv->child_count);
+        }
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (!local->replies[i].valid)
+                        continue;
+
+                if (local->replies[i].op_ret >= 0)
+                        continue;
+
+                if (local->replies[i].op_errno == ENOTCONN)
+                        continue;
+
+                /*Order is important in the following condition*/
+                if ((s < 0) || (!readable[s] && readable[i]))
+                        s = i;
+        }
+
+        if (s != -1 && local->replies[s].xdata) {
+                local->xdata_rsp = dict_ref (local->replies[s].xdata);
+        } else if (s == -1) {
+                for (i = 0; i < priv->child_count; i++) {
+                        if (!local->replies[i].valid)
+                                continue;
+
+                        if (local->replies[i].op_ret >= 0)
+                                continue;
+
+                        if (!local->replies[i].xdata)
+                                continue;
+                        local->xdata_rsp = dict_ref (local->replies[i].xdata);
+                        break;
+                }
+        }
+}
+
 gf_boolean_t
 afr_needs_changelog_update (afr_local_t *local)
 {
@@ -747,6 +807,17 @@ afr_handle_quorum (call_frame_t *frame)
         local->op_errno = afr_final_errno (local, priv);
         if (local->op_errno == 0)
                 local->op_errno = afr_quorum_errno (priv);
+        switch (local->transaction.type) {
+        case AFR_ENTRY_TRANSACTION:
+        case AFR_ENTRY_RENAME_TRANSACTION:
+                afr_pick_error_xdata (local, priv, local->parent,
+                                      local->readable, local->parent2,
+                                      local->readable2);
+                break;
+        default:
+                /*TBD*/
+                break;
+        }
 }
 
 int
diff --git a/xlators/cluster/afr/src/afr-transaction.h b/xlators/cluster/afr/src/afr-transaction.h
index c58531eff44..ca8fcfefa89 100644
--- a/xlators/cluster/afr/src/afr-transaction.h
+++ b/xlators/cluster/afr/src/afr-transaction.h
@@ -55,4 +55,8 @@ gf_boolean_t afr_has_quorum (unsigned char *subvols, xlator_t *this);
 gf_boolean_t afr_needs_changelog_update (afr_local_t *local);
 void afr_zero_fill_stat (afr_local_t *local);
 
+void
+afr_pick_error_xdata (afr_local_t *local, afr_private_t *priv,
+                      inode_t *inode1, unsigned char *readable1,
+                      inode_t *inode2, unsigned char *readable2);
 #endif /* __TRANSACTION_H__ */
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 5482dab25b2..c6afbbe20e0 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -426,6 +426,7 @@ typedef struct _afr_local {
 	   performed. This is the output of afr_inode_refresh()
 	*/
 	unsigned char *readable;
+	unsigned char *readable2; /*For rename transaction*/
 
 	afr_inode_refresh_cbk_t refreshfn;
 
@@ -860,14 +861,15 @@ afr_inode_read_subvol_type_get (inode_t *inode, xlator_t *this,
 				int type);
 int
 afr_read_subvol_get (inode_t *inode, xlator_t *this, int *subvol_p,
+                     unsigned char *readables,
 		     int *event_p, afr_transaction_type type,
                      afr_read_subvol_args_t *args);
 
-#define afr_data_subvol_get(i, t, s, e, a) \
-	afr_read_subvol_get(i, t, s, e, AFR_DATA_TRANSACTION, a)
+#define afr_data_subvol_get(i, t, s, r, e, a) \
+	afr_read_subvol_get(i, t, s, r, e, AFR_DATA_TRANSACTION, a)
 
 #define afr_metadata_subvol_get(i, t, s, e, a) \
-	afr_read_subvol_get(i, t, s, e, AFR_METADATA_TRANSACTION, a)
+	afr_read_subvol_get(i, t, s, NULL, e, AFR_METADATA_TRANSACTION, a)
 
 int
 afr_inode_refresh (call_frame_t *frame, xlator_t *this, inode_t *inode,