self-heald: Find self-heal failures, split-brain

Change-Id: Ib967f0fe0b537fe60e51d7d05462b58a7f16596e
BUG: 806745
Signed-off-by: Pranith Kumar K <pranithk@gluster.com>
Reviewed-on: http://review.gluster.com/3077
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-by: Vijay Bellur <vijay@gluster.com>

6 files changed, 36 insertions, 13 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 2cfe92ac..9874b261 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -1301,7 +1301,8 @@ afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
                       void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
                                                    xlator_t *this),
                       int (*unwind) (call_frame_t *frame, xlator_t *this,
-                                     int32_t op_ret, int32_t op_errno))
+                                     int32_t op_ret, int32_t op_errno,
+                                     int32_t sh_failed))
 {
         afr_local_t             *local = NULL;
         char                    sh_type_str[256] = {0,};
@@ -1527,9 +1528,12 @@ afr_lookup_set_self_heal_params (afr_local_t *local, xlator_t *this)
 
 int
 afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
-                             int32_t op_ret, int32_t op_errno)
+                             int32_t op_ret, int32_t op_errno,
+                             int32_t sh_failed)
 {
         afr_local_t *local = NULL;
+        int         ret    = -1;
+        dict_t      *xattr = NULL;
 
         local = frame->local;
 
@@ -1544,6 +1548,13 @@ afr_self_heal_lookup_unwind (call_frame_t *frame, xlator_t *this,
         }
 
         afr_lookup_done_success_action (frame, this, _gf_true);
+        xattr = local->cont.lookup.xattr;
+        if (xattr) {
+                ret = dict_set_int32 (xattr, "sh-failed", sh_failed);
+                if (ret)
+                        gf_log (this->name, GF_LOG_ERROR, "%s: Failed to set "
+                                "sh-failed to %d", local->loc.path, sh_failed);
+        }
 out:
         AFR_STACK_UNWIND (lookup, frame, local->op_ret, local->op_errno,
                           local->cont.lookup.inode, &local->cont.lookup.buf,
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index 5abbd9c1..87282256 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -51,7 +51,7 @@
 
 int
 afr_examine_dir_sh_unwind (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                           int32_t op_errno)
+                           int32_t op_errno, int32_t sh_failed)
 {
         afr_local_t *local  = NULL;
 
diff --git a/xlators/cluster/afr/src/afr-self-heal-common.c b/xlators/cluster/afr/src/afr-self-heal-common.c
index bd0e0462..af5aadc3 100644
--- a/xlators/cluster/afr/src/afr-self-heal-common.c
+++ b/xlators/cluster/afr/src/afr-self-heal-common.c
@@ -2048,7 +2048,8 @@ afr_self_heal_completion_cbk (call_frame_t *bgsh_frame, xlator_t *this)
         FRAME_SU_UNDO (bgsh_frame, afr_local_t);
 
         if (!sh->unwound && sh->unwind) {
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno);
+                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
+                            sh->op_failed);
         }
 
         if (sh->background) {
@@ -2187,7 +2188,7 @@ afr_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode)
 
 out:
         if (op_errno) {
-                orig_sh->unwind (frame, this, -1, op_errno);
+                orig_sh->unwind (frame, this, -1, op_errno, 1);
                 if (sh_frame)
                         AFR_STACK_DESTROY (sh_frame);
         }
diff --git a/xlators/cluster/afr/src/afr-self-heal-data.c b/xlators/cluster/afr/src/afr-self-heal-data.c
index d362d5ad..69494157 100644
--- a/xlators/cluster/afr/src/afr-self-heal-data.c
+++ b/xlators/cluster/afr/src/afr-self-heal-data.c
@@ -769,7 +769,8 @@ afr_sh_data_fix (call_frame_t *frame, xlator_t *this)
         }
 
         if (sh->background && sh->unwind) {
-                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno);
+                sh->unwind (sh->orig_frame, this, sh->op_ret, sh->op_errno,
+                            sh->op_failed);
                 sh->unwound = _gf_true;
         }
 
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
index 3068d5c4..55ede78a 100644
--- a/xlators/cluster/afr/src/afr-self-heald.c
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -277,7 +277,7 @@ out:
 
 void
 _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
-                       int32_t op_ret, int32_t op_errno,
+                       int32_t op_ret, int32_t op_errno, dict_t *xattr_rsp,
                        afr_crawl_data_t *crawl_data)
 {
         int              ret = 0;
@@ -286,6 +286,8 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
         eh_t             *eh = NULL;
         char             *path = NULL;
         shd_event_t      *event = NULL;
+        int32_t          sh_failed = 0;
+        gf_boolean_t     split_brain = 0;
 
         priv = this->private;
         shd  = &priv->shd;
@@ -307,9 +309,12 @@ _crawl_post_sh_action (xlator_t *this, loc_t *parent, loc_t *child,
                 }
         }
 
-        if (op_ret < 0 && op_errno == EIO)
+        if (xattr_rsp)
+                ret = dict_get_int32 (xattr_rsp, "sh-failed", &sh_failed);
+        split_brain = afr_is_split_brain (this, child->inode);
+        if ((op_ret < 0 && op_errno == EIO) || split_brain)
                 eh = shd->split_brain;
-        else if (op_ret < 0)
+        else if ((op_ret < 0) || sh_failed)
                 eh = shd->heal_failed;
         else
                 eh = shd->healed;
@@ -338,6 +343,7 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent
 {
         struct iatt      parentbuf = {0};
         int              ret = 0;
+        dict_t           *xattr_rsp = NULL;
 
         if (uuid_is_null (child->gfid))
                 gf_log (this->name, GF_LOG_DEBUG, "lookup %s", child->path);
@@ -346,8 +352,11 @@ _self_heal_entry (xlator_t *this, afr_crawl_data_t *crawl_data, gf_dirent_t *ent
                         uuid_utoa (child->gfid));
 
         ret = syncop_lookup (this, child, NULL,
-                             iattr, NULL, &parentbuf);
-        _crawl_post_sh_action (this, parent, child, ret, errno, crawl_data);
+                             iattr, &xattr_rsp, &parentbuf);
+        _crawl_post_sh_action (this, parent, child, ret, errno, xattr_rsp,
+                               crawl_data);
+        if (xattr_rsp)
+                dict_unref (xattr_rsp);
         return ret;
 }
 
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 815f4667..fccb39b1 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -191,7 +191,7 @@ typedef struct {
            background, this function will be called as soon as possible. */
 
         int (*unwind) (call_frame_t *frame, xlator_t *this, int32_t op_ret,
-                       int32_t op_errno);
+                       int32_t op_errno, int32_t sh_failed);
 
         /* End of external interface members */
 
@@ -1016,7 +1016,8 @@ afr_launch_self_heal (call_frame_t *frame, xlator_t *this, inode_t *inode,
                       void (*gfid_sh_success_cbk) (call_frame_t *sh_frame,
                                                    xlator_t *this),
                       int (*unwind) (call_frame_t *frame, xlator_t *this,
-                                     int32_t op_ret, int32_t op_errno));
+                                     int32_t op_ret, int32_t op_errno,
+                                     int32_t sh_failed));
 int
 afr_fix_open (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx,
               int need_open_count, int *need_open);