cluster/afr: mark new entry changelog for create/mknod failures

Problem:
When create/mknod fails on some of the nodes, appropriate pending
data/metadata changelogs are not assigned. This was not considered
to be an issue because entry self-heal would do the assigning of
appropriate changelog after creating new entries. But using
the combination of rebalance and remove brick we can construct a
case where a file with same name and gfid can be created in a dir
with different data and link-to xattr without any changelog.

Fix:
When a create/mknod failure is observed mark the appropriate
changelog on the new file created.

Change-Id: I4c32cbf5594a13fb14deaf97ff30b2fff11cbfd6
BUG: 858212
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4207
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

5 files changed, 228 insertions, 67 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 46f492bc..c38e5272 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -767,6 +767,13 @@ out:
 }
 
 void
+afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count)
+{
+        afr_reset_xattr (xattr, child_count);
+        GF_FREE (xattr);
+}
+
+void
 afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
 {
         afr_self_heal_t *sh = NULL;
@@ -782,10 +789,7 @@ afr_local_sh_cleanup (afr_local_t *local, xlator_t *this)
         if (sh->inode)
                 inode_unref (sh->inode);
 
-        if (sh->xattr) {
-                afr_reset_xattr (sh->xattr, priv->child_count);
-                GF_FREE (sh->xattr);
-        }
+        afr_xattr_array_destroy (sh->xattr, priv->child_count);
 
         GF_FREE (sh->child_errno);
 
@@ -4233,3 +4237,42 @@ xlator_subvolume_count (xlator_t *this)
                 i++;
         return i;
 }
+
+inline gf_boolean_t
+afr_is_errno_set (int *child_errno, int child)
+{
+        return child_errno[child];
+}
+
+inline gf_boolean_t
+afr_is_errno_unset (int *child_errno, int child)
+{
+        return !afr_is_errno_set (child_errno, child);
+}
+
+void
+afr_prepare_new_entry_pending_matrix (int32_t **pending,
+                                      gf_boolean_t (*is_pending) (int *, int),
+                                      int *ctx, struct iatt *buf,
+                                      unsigned int child_count)
+{
+        int midx = 0;
+        int idx  = 0;
+        int i    = 0;
+
+        midx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
+        if (IA_ISDIR (buf->ia_type))
+                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
+        else if (IA_ISREG (buf->ia_type))
+                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
+        else
+                idx = -1;
+        for (i = 0; i < child_count; i++) {
+                if (is_pending (ctx, i)) {
+                        pending[i][midx] = hton32 (1);
+                        if (idx == -1)
+                                continue;
+                        pending[i][idx] = hton32 (1);
+                }
+        }
+}
diff --git a/xlators/cluster/afr/src/afr-dir-write.c b/xlators/cluster/afr/src/afr-dir-write.c
index 49cf8d45..d86b08b8 100644
--- a/xlators/cluster/afr/src/afr-dir-write.c
+++ b/xlators/cluster/afr/src/afr-dir-write.c
@@ -106,11 +106,134 @@ __dir_entry_fop_common_cbk (call_frame_t *frame, int child_index,
 
                 local->fresh_children[local->success_count] = child_index;
                 local->success_count++;
+                local->child_errno[child_index] = 0;
+        } else {
+                local->child_errno[child_index] = op_errno;
         }
 
         local->op_errno = op_errno;
 }
 
+int
+afr_mark_new_entry_changelog_cbk (call_frame_t *frame, void *cookie,
+                                  xlator_t *this,
+                                  int32_t op_ret, int32_t op_errno,
+                                  dict_t *xattr, dict_t *xdata)
+{
+        int     call_count = 0;
+
+        call_count = afr_frame_return (frame);
+        if (call_count == 0) {
+                AFR_STACK_DESTROY (frame);
+        }
+        return 0;
+}
+
+void
+afr_mark_new_entry_changelog (call_frame_t *frame, xlator_t *this)
+{
+        call_frame_t  *new_frame  = NULL;
+        afr_local_t   *local      = NULL;
+        afr_local_t   *new_local  = NULL;
+        afr_private_t *priv       = NULL;
+        dict_t        **xattr     = NULL;
+        int32_t       **changelog = NULL;
+        int           i           = 0;
+        GF_UNUSED int op_errno    = 0;
+
+        local = frame->local;
+        priv = this->private;
+
+        new_frame = copy_frame (frame);
+        if (!new_frame) {
+                goto out;
+        }
+
+        AFR_LOCAL_ALLOC_OR_GOTO (new_frame->local, out);
+        new_local = new_frame->local;
+        changelog = afr_matrix_create (priv->child_count, AFR_NUM_CHANGE_LOGS);
+        if (!changelog)
+                goto out;
+
+        xattr = GF_CALLOC (priv->child_count, sizeof (*xattr),
+                           gf_afr_mt_dict_t);
+        if (!xattr)
+                goto out;
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_errno[i])
+                        continue;
+                xattr[i] = dict_new ();
+                if (!xattr[i])
+                        goto out;
+        }
+
+        afr_prepare_new_entry_pending_matrix (changelog,
+                                              afr_is_errno_set,
+                                              local->child_errno,
+                                              &local->cont.dir_fop.buf,
+                                              priv->child_count);
+
+        new_local->pending = changelog;
+        uuid_copy (new_local->loc.gfid, local->cont.dir_fop.buf.ia_gfid);
+        new_local->loc.inode = inode_ref (local->cont.dir_fop.inode);
+        new_local->call_count = local->success_count;
+
+        for (i = 0; i < priv->child_count; i++) {
+                if (local->child_errno[i])
+                        continue;
+
+                afr_set_pending_dict (priv, xattr[i], changelog, i, LOCAL_LAST);
+                STACK_WIND_COOKIE (new_frame, afr_mark_new_entry_changelog_cbk,
+                                   (void *) (long) i, priv->children[i],
+                                   priv->children[i]->fops->xattrop,
+                                   &new_local->loc, GF_XATTROP_ADD_ARRAY,
+                                   xattr[i], NULL);
+        }
+        new_frame = NULL;
+out:
+        if (new_frame)
+                AFR_STACK_DESTROY (new_frame);
+        afr_xattr_array_destroy (xattr, priv->child_count);
+        return;
+}
+
+gf_boolean_t
+afr_is_new_entry_changelog_needed (glusterfs_fop_t fop)
+{
+        glusterfs_fop_t fops[]   = {GF_FOP_CREATE, GF_FOP_MKNOD, GF_FOP_NULL};
+        int             i        = 0;
+
+        for (i = 0; fops[i] != GF_FOP_NULL; i++) {
+                if (fop == fops[i])
+                        return _gf_true;
+        }
+        return _gf_false;
+}
+
+void
+afr_dir_fop_mark_entry_pending_changelog (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t   *local      = NULL;
+        afr_private_t *priv       = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+
+        if (local->op_ret < 0)
+                goto out;
+
+        if (local->success_count == priv->child_count)
+                goto out;
+
+        if (!afr_is_new_entry_changelog_needed (local->op))
+                goto out;
+
+        afr_mark_new_entry_changelog (frame, this);
+
+out:
+        local->transaction.resume (frame, this);
+}
+
 void
 afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
 {
@@ -129,7 +252,7 @@ afr_dir_fop_done (call_frame_t *frame, xlator_t *this)
                                       local->cont.dir_fop.buf.ia_gfid);
 done:
         local->transaction.unwind (frame, this);
-        local->transaction.resume (frame, this);
+        afr_dir_fop_mark_entry_pending_changelog (frame, this);
 }
 
 /* {{{ create */
@@ -331,6 +454,7 @@ afr_create (call_frame_t *frame, xlator_t *this,
         }
         UNLOCK (&priv->read_child_lock);
 
+        local->op                = GF_FOP_CREATE;
         local->cont.create.flags = flags;
         local->cont.create.mode  = mode;
         local->cont.create.fd    = fd_ref (fd);
@@ -524,6 +648,7 @@ afr_mknod (call_frame_t *frame, xlator_t *this, loc_t *loc, mode_t mode,
         }
         UNLOCK (&priv->read_child_lock);
 
+        local->op               = GF_FOP_MKNOD;
         local->cont.mknod.mode  = mode;
         local->cont.mknod.dev   = dev;
         local->umask = umask;
diff --git a/xlators/cluster/afr/src/afr-self-heal-entry.c b/xlators/cluster/afr/src/afr-self-heal-entry.c
index 0c70d4b6..c3c9f9fc 100644
--- a/xlators/cluster/afr/src/afr-self-heal-entry.c
+++ b/xlators/cluster/afr/src/afr-self-heal-entry.c
@@ -1018,33 +1018,6 @@ out:
         return 0;
 }
 
-void
-afr_sh_prepare_new_entry_pending_matrix (int32_t **pending,
-                                         int *child_errno,
-                                         struct iatt *buf,
-                                         unsigned int child_count)
-{
-        int              midx             = 0;
-        int              idx              = 0;
-        int              i                = 0;
-
-        midx = afr_index_for_transaction_type (AFR_METADATA_TRANSACTION);
-        if (IA_ISDIR (buf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_ENTRY_TRANSACTION);
-        else if (IA_ISREG (buf->ia_type))
-                idx = afr_index_for_transaction_type (AFR_DATA_TRANSACTION);
-        else
-                idx = -1;
-        for (i = 0; i < child_count; i++) {
-                if (child_errno[i])
-                        continue;
-                pending[i][midx] = hton32 (1);
-                if (idx == -1)
-                        continue;
-                pending[i][idx] = hton32 (1);
-        }
-}
-
 int
 afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame,
                                       xlator_t *this)
@@ -1061,10 +1034,11 @@ afr_sh_entry_impunge_perform_xattrop (call_frame_t *impunge_frame,
         impunge_sh = &impunge_local->self_heal;
         active_src = impunge_sh->active_source;
 
-        afr_sh_prepare_new_entry_pending_matrix (impunge_local->pending,
-                                                 impunge_sh->child_errno,
-                                                 &impunge_sh->entrybuf,
-                                                 priv->child_count);
+        afr_prepare_new_entry_pending_matrix (impunge_local->pending,
+                                              afr_is_errno_unset,
+                                              impunge_sh->child_errno,
+                                              &impunge_sh->entrybuf,
+                                              priv->child_count);
         xattr = dict_new ();
         if (!xattr) {
                 op_errno = ENOMEM;
diff --git a/xlators/cluster/afr/src/afr-transaction.c b/xlators/cluster/afr/src/afr-transaction.c
index 3ba8d9b5..5e636d8d 100644
--- a/xlators/cluster/afr/src/afr-transaction.c
+++ b/xlators/cluster/afr/src/afr-transaction.c
@@ -203,6 +203,40 @@ __mark_all_success (int32_t *pending[], int child_count,
         }
 }
 
+void
+_set_all_child_errno (int *child_errno, unsigned int child_count)
+{
+        int     i = 0;
+
+        for (i = 0; i < child_count; i++)
+                if (child_errno[i] == 0)
+                        child_errno[i] = ENOTCONN;
+}
+
+void
+afr_transaction_perform_fop (call_frame_t *frame, xlator_t *this)
+{
+        afr_local_t     *local = NULL;
+        afr_private_t   *priv = NULL;
+
+        local = frame->local;
+        priv  = this->private;
+        __mark_all_success (local->pending, priv->child_count,
+                            local->transaction.type);
+
+        _set_all_child_errno (local->child_errno, priv->child_count);
+
+        /*  Perform fops with the lk-owner from top xlator.
+         *  Eg: lk-owner of posix-lk and flush should be same,
+         *  flush cant clear the  posix-lks without that lk-owner.
+         */
+        afr_save_lk_owner (frame);
+        frame->root->lk_owner =
+                local->transaction.main_frame->root->lk_owner;
+
+        local->transaction.fop (frame, this);
+}
+
 
 static int
 __changelog_enabled (afr_private_t *priv, afr_transaction_type type)
@@ -778,18 +812,7 @@ afr_changelog_pre_op_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                     (local->op_errno == ENOTSUP)) {
                         local->transaction.resume (frame, this);
                 } else {
-                        __mark_all_success (local->pending, priv->child_count,
-                                            local->transaction.type);
-
-			/*  Perform fops with the lk-owner from top xlator.
-			 *  Eg: lk-owner of posix-lk and flush should be same,
-			 *  flush cant clear the  posix-lks without that lk-owner.
-			 */
-			afr_save_lk_owner (frame);
-			frame->root->lk_owner =
-				local->transaction.main_frame->root->lk_owner;
-
-                        local->transaction.fop (frame, this);
+                        afr_transaction_perform_fop (frame, this);
                 }
         }
 
@@ -1218,28 +1241,10 @@ afr_lock (call_frame_t *frame, xlator_t *this)
 int
 afr_internal_lock_finish (call_frame_t *frame, xlator_t *this)
 {
-        afr_local_t   *local = NULL;
-        afr_private_t *priv  = NULL;
-
-        priv  = this->private;
-        local = frame->local;
-
         if (__fop_changelog_needed (frame, this)) {
                 afr_changelog_pre_op (frame, this);
         } else {
-                __mark_all_success (local->pending, priv->child_count,
-                                    local->transaction.type);
-
-
-		/*  Perform fops with the lk-owner from top xlator.
-		 *  Eg: lk-owner of posix-lk and flush should be same,
-		 *  flush cant clear the  posix-lks without that lk-owner.
-		 */
-		afr_save_lk_owner (frame);
-		frame->root->lk_owner =
-			local->transaction.main_frame->root->lk_owner;
-
-                local->transaction.fop (frame, this);
+                afr_transaction_perform_fop (frame, this);
         }
 
         return 0;
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index d22daf51..da9cc67c 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -1023,6 +1023,20 @@ afr_matrix_cleanup (int32_t **pending, unsigned int m);
 
 int32_t**
 afr_matrix_create (unsigned int m, unsigned int n);
+
+gf_boolean_t
+afr_is_errno_set (int *child_errno, int child);
+
+gf_boolean_t
+afr_is_errno_unset (int *child_errno, int child);
+
+void
+afr_prepare_new_entry_pending_matrix (int32_t **pending,
+                                      gf_boolean_t (*is_pending) (int *, int),
+                                      int *ctx, struct iatt *buf,
+                                      unsigned int child_count);
+void
+afr_xattr_array_destroy (dict_t **xattr, unsigned int child_count);
 /*
  * Special value indicating we should use the "auto" quorum method instead of
  * a fixed value (including zero to turn off quorum enforcement).