cluster/afr: Provide option to disable readdir failover

In a replica pair unlike files, directories may not have their
content in same order, so readdir for same (offset, size) may
not give same entries on both the sobvolumes of replica pair.
Switching over from one subvolume to another may not be a good
idea sometimes. It may lead to duplicate entries or fewer entries
or both. This patch provides a way to disable readdir-failover
so that applications like rebalance can retry if they want to.

Change-Id: I2b23eb224a2e84016a561362932613ac824c11a0
BUG: 859387
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4159
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

5 files changed, 42 insertions, 25 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 2e339986621..35201085444 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -2462,6 +2462,7 @@ __afr_fd_ctx_set (xlator_t *this, fd_t *fd)
 	pthread_mutex_init (&fd_ctx->delay_lock, NULL);
         INIT_LIST_HEAD (&fd_ctx->paused_calls);
         INIT_LIST_HEAD (&fd_ctx->entries);
+        fd_ctx->call_child = -1;
 
         ret = __fd_ctx_set (fd, this, (uint64_t)(long) fd_ctx);
         if (ret)
diff --git a/xlators/cluster/afr/src/afr-dir-read.c b/xlators/cluster/afr/src/afr-dir-read.c
index ce91ffba729..c201d45fd68 100644
--- a/xlators/cluster/afr/src/afr-dir-read.c
+++ b/xlators/cluster/afr/src/afr-dir-read.c
@@ -497,6 +497,9 @@ afr_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 
         local = frame->local;
 
+        if ((priv->readdir_failover == _gf_false) && (op_ret < 0))
+                goto out;
+
         read_child = (long) cookie;
         last_index = &local->cont.readdir.last_index;
         fresh_children = local->fresh_children;
@@ -593,15 +596,14 @@ int32_t
 afr_do_readdir (call_frame_t *frame, xlator_t *this,
                 fd_t *fd, size_t size, off_t offset, int whichop, dict_t *dict)
 {
-        afr_private_t *  priv       = NULL;
-        xlator_t **      children   = NULL;
-        int              call_child = 0;
-        afr_local_t     *local      = NULL;
-        uint64_t         ctx        = 0;
-        afr_fd_ctx_t    *fd_ctx     = NULL;
-        int              ret        = -1;
-        int32_t          op_errno   = 0;
-        uint64_t         read_child = 0;
+        afr_private_t *priv      = NULL;
+        xlator_t      **children = NULL;
+        int           call_child = 0;
+        afr_local_t   *local     = NULL;
+        afr_fd_ctx_t  *fd_ctx    = NULL;
+        int           ret        = -1;
+        int32_t       op_errno   = 0;
+        uint64_t      read_child = 0;
 
         VALIDATE_OR_GOTO (frame, out);
         VALIDATE_OR_GOTO (this, out);
@@ -626,29 +628,33 @@ afr_do_readdir (call_frame_t *frame, xlator_t *this,
         read_child = afr_inode_get_read_ctx (this, fd->inode,
                                              local->fresh_children);
         ret = afr_get_call_child (this, local->child_up, read_child,
-                                     local->fresh_children,
-                                     &call_child,
-                                     &local->cont.readdir.last_index);
+                                  local->fresh_children,
+                                  &call_child,
+                                  &local->cont.readdir.last_index);
         if (ret < 0) {
                 op_errno = -ret;
                 goto out;
         }
 
+        fd_ctx  = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx) {
+                op_errno = EBADF;
+                goto out;
+        }
+
+        if ((offset == 0) || (fd_ctx->call_child == -1)) {
+                fd_ctx->call_child = call_child;
+        } else if ((priv->readdir_failover == _gf_false) &&
+                   (call_child != fd_ctx->call_child)) {
+                op_errno = EBADF;
+                goto out;
+        }
+
         local->fd                  = fd_ref (fd);
         local->cont.readdir.size   = size;
         local->cont.readdir.dict   = (dict)? dict_ref (dict) : NULL;
 
         if (priv->strict_readdir) {
-                ret = fd_ctx_get (fd, this, &ctx);
-                if (ret < 0) {
-                        gf_log (this->name, GF_LOG_INFO,
-                                "could not get fd ctx for fd=%p", fd);
-                        op_errno = -ret;
-                        goto out;
-                }
-
-                fd_ctx = (afr_fd_ctx_t *)(long) ctx;
-
                 if (fd_ctx->last_tried != call_child) {
                         gf_log (this->name, GF_LOG_TRACE,
                                 "first up child has changed from %d to %d, "
@@ -675,10 +681,9 @@ afr_do_readdir (call_frame_t *frame, xlator_t *this,
                                    children[call_child]->fops->readdirp, fd,
                                    size, offset, dict);
 
-        ret = 0;
+        return 0;
 out:
-        if (ret < 0)
-                AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
+        AFR_STACK_UNWIND (readdir, frame, -1, op_errno, NULL, NULL);
         return 0;
 }
 
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index c120ba57b5f..cdc7a609b4d 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -189,6 +189,8 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF (AFR_SH_READDIR_SIZE_KEY, priv->sh_readdir_size,
                           options, size, out);
         /* Reset this so we re-discover in case the topology changed.  */
+        GF_OPTION_RECONF ("readdir-failover", priv->readdir_failover, options,
+                          bool, out);
         priv->did_discovery = _gf_false;
 
         ret = 0;
@@ -332,6 +334,7 @@ init (xlator_t *this)
         fix_quorum_options(this,priv,qtype);
 
 	GF_OPTION_INIT ("post-op-delay-secs", priv->post_op_delay_secs, uint32, out);
+        GF_OPTION_INIT ("readdir-failover", priv->readdir_failover, bool, out);
 
         priv->wait_count = 1;
 
@@ -758,5 +761,10 @@ struct volume_options options[] = {
           .max = 131072,
           .default_value = "1KB",
         },
+        { .key = {"readdir-failover"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .description = "readdir(p) will not failover if this option is off",
+          .default_value = "on",
+        },
         { .key  = {NULL} },
 };
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index 48dfbf37eb8..7de8d82436c 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -160,6 +160,7 @@ typedef struct _afr_private {
         afr_self_heald_t       shd;
         gf_boolean_t           choose_local;
         gf_boolean_t           did_discovery;
+        gf_boolean_t           readdir_failover;
         uint64_t               sh_readdir_size;
 } afr_private_t;
 
@@ -710,6 +711,7 @@ typedef struct {
 	pthread_mutex_t    delay_lock;
 	gf_timer_t        *delay_timer;
 	call_frame_t      *delay_frame;
+        int               call_child;
 } afr_fd_ctx_t;
 
 
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index f3de2feed52..da1c2a4ef94 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -130,6 +130,7 @@ static struct volopt_map_entry glusterd_volopt_map[] = {
         {"cluster.quorum-count",                 "cluster/replicate",  "quorum-count", NULL, NO_DOC, 0, 1},
         {"cluster.choose-local",                 "cluster/replicate",  NULL, NULL, DOC, 0, 2},
         {"cluster.self-heal-readdir-size",       "cluster/replicate",  NULL, NULL, DOC, 0, 2},
+        {"cluster.readdir-failover",             "cluster/replicate",  NULL, NULL, DOC, 0, 2},
 
         /* Stripe xlator options */
         {"cluster.stripe-block-size",            "cluster/stripe",     "block-size", NULL, DOC, 0, 1},