cluster/afr: Don't queue transactions during open-fd fix

Before Anonymous fds are available, afr had to queue up
transactions if the file is not opened on one of its
subvolumes. This happens until the attempt to open the
file either succeeds or fails. These attempts happen
until the file is successfully opened on the subvolume.
Now client xlator uses anonymous fds to perform the fops
if the fd used for the fop is not 'opened'.
Fops will be successful even when the file is not opened
so there is no need to queue up the transactions anymore in afr.
Open is attempted on the subvolume where it is not
opened independent of the fop.

Change-Id: Id1a4b4ebe6f89f9efe8f6a8247918b91247d0819
BUG: 913051
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/4568
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

1 files changed, 71 insertions, 181 deletions

diff --git a/xlators/cluster/afr/src/afr-inode-write.c b/xlators/cluster/afr/src/afr-inode-write.c
index 99fa027d0..a83223569 100644
--- a/xlators/cluster/afr/src/afr-inode-write.c
+++ b/xlators/cluster/afr/src/afr-inode-write.c
@@ -328,153 +328,74 @@ out:
         return 0;
 }
 
-static int
-afr_prepare_loc (call_frame_t *frame, fd_t *fd)
+static void
+afr_trigger_open_fd_self_heal (fd_t *fd, xlator_t *this)
 {
-        afr_local_t    *local = NULL;
-        char           *name = NULL;
-        char           *path = NULL;
-        int             ret = 0;
-
-        if ((!fd) || (!fd->inode))
-                return -1;
-
-        local = frame->local;
-        ret = inode_path (fd->inode, NULL, (char **)&path);
-        if (ret <= 0) {
-                gf_log (frame->this->name, GF_LOG_DEBUG,
-                        "Unable to get path for gfid: %s",
-                        uuid_utoa (fd->inode->gfid));
-                return -1;
-        }
-
-        if (local->loc.path) {
-                if (strcmp (path, local->loc.path))
-                        gf_log (frame->this->name, GF_LOG_DEBUG,
-                                "overwriting old loc->path %s with %s",
-                                local->loc.path, path);
-                GF_FREE ((char *)local->loc.path);
-        }
-        local->loc.path = path;
-
-        name = strrchr (local->loc.path, '/');
-        if (name)
-                name++;
-        local->loc.name = name;
-
-        if (local->loc.inode) {
-                inode_unref (local->loc.inode);
-        }
-        local->loc.inode = inode_ref (fd->inode);
-
-        if (local->loc.parent) {
-                inode_unref (local->loc.parent);
+        call_frame_t    *frame   = NULL;
+        afr_local_t     *local   = NULL;
+        afr_self_heal_t *sh      = NULL;
+        char            *reason  = NULL;
+        int32_t         op_errno = 0;
+        int             ret      = 0;
+
+        if (!fd || !fd->inode || uuid_is_null (fd->inode->gfid)) {
+                gf_log_callingfn (this->name, GF_LOG_ERROR, "Invalid args: "
+                                  "fd: %p, inode: %p", fd,
+                                  fd ? fd->inode : NULL);
+                goto out;
         }
 
-        local->loc.parent = inode_parent (local->loc.inode, 0, NULL);
-
-        return 0;
-}
-
-afr_fd_paused_call_t*
-afr_paused_call_create (call_frame_t *frame)
-{
-        afr_local_t             *local = NULL;
-        afr_fd_paused_call_t   *paused_call = NULL;
+        frame = create_frame (this, this->ctx->pool);
+        if (!frame)
+                goto out;
 
+        AFR_LOCAL_ALLOC_OR_GOTO (frame->local, out);
         local = frame->local;
-        GF_ASSERT (local->fop_call_continue);
-
-        paused_call = GF_CALLOC (1, sizeof (*paused_call),
-                                  gf_afr_fd_paused_call_t);
-        if (paused_call) {
-                INIT_LIST_HEAD (&paused_call->call_list);
-                paused_call->frame = frame;
-        }
-
-        return paused_call;
-}
-
-static int
-afr_pause_fd_fop (call_frame_t *frame, xlator_t *this, afr_fd_ctx_t *fd_ctx)
-{
-        afr_fd_paused_call_t *paused_call = NULL;
-        int                    ret = 0;
-
-        paused_call = afr_paused_call_create (frame);
-        if (paused_call)
-                list_add (&paused_call->call_list, &fd_ctx->paused_calls);
-        else
-                ret = -ENOMEM;
-
-        return ret;
-}
+        ret = afr_local_init (local, this->private, &op_errno);
+        if (ret < 0)
+                goto out;
 
-static void
-afr_trigger_open_fd_self_heal (call_frame_t *frame, xlator_t *this)
-{
-        afr_local_t             *local = NULL;
-        afr_self_heal_t         *sh = NULL;
-        inode_t                 *inode = NULL;
-        char                    *reason = NULL;
+        local->loc.inode = inode_ref (fd->inode);
+        ret = loc_path (&local->loc, NULL);
+        if (ret < 0)
+                goto out;
 
-        local = frame->local;
         sh    = &local->self_heal;
-        inode = local->fd->inode;
-
-        sh->do_missing_entry_self_heal = _gf_true;
-        sh->do_gfid_self_heal = _gf_true;
-        sh->do_data_self_heal = _gf_true;
+        sh->do_metadata_self_heal = _gf_true;
+        if (fd->inode->ia_type == IA_IFREG)
+                sh->do_data_self_heal = _gf_true;
+        else if (fd->inode->ia_type == IA_IFDIR)
+                sh->do_entry_self_heal = _gf_true;
 
         reason = "subvolume came online";
-        afr_launch_self_heal (frame, this, inode, _gf_true, inode->ia_type,
-                              reason, NULL, NULL);
+        afr_launch_self_heal (frame, this, fd->inode, _gf_true,
+                              fd->inode->ia_type, reason, NULL, NULL);
+        return;
+out:
+        AFR_STACK_DESTROY (frame);
 }
 
-int
-afr_open_fd_fix (call_frame_t *frame, xlator_t *this, gf_boolean_t pause_fop)
+void
+afr_open_fd_fix (fd_t *fd, xlator_t *this)
 {
-        int                     ret = 0;
-        int                     i   = 0;
-        afr_fd_ctx_t            *fd_ctx = NULL;
-        inode_t                 *inode = NULL;
-        gf_boolean_t            need_self_heal = _gf_false;
-        int                     *need_open = NULL;
-        int                     need_open_count = 0;
-        afr_local_t             *local = NULL;
-        afr_private_t           *priv = NULL;
-        gf_boolean_t            fop_continue = _gf_true;
+        int           ret             = 0;
+        int           i               = 0;
+        afr_fd_ctx_t  *fd_ctx         = NULL;
+        gf_boolean_t  need_self_heal  = _gf_false;
+        int           *need_open      = NULL;
+        size_t        need_open_count = 0;
+        afr_private_t *priv           = NULL;
 
-        local = frame->local;
         priv  = this->private;
 
-        GF_ASSERT (local->fd);
-
-        inode = local->fd->inode;
-        //gfid is not set in rebalance, that case needs to be handled.
-        if (fd_is_anonymous (local->fd) ||
-            !inode || uuid_is_null (inode->gfid)) {
-                fop_continue = _gf_true;
-                goto out;
-        }
-
-        if (pause_fop)
-                GF_ASSERT (local->fop_call_continue);
-
-        ret = afr_prepare_loc (frame, local->fd);
-        if (ret < 0) {
-                //File does not exist we cant open it.
-                ret = 0;
+        if (!afr_is_fd_fixable (fd))
                 goto out;
-        }
 
-        fd_ctx = afr_fd_ctx_get (local->fd, this);
-        if (!fd_ctx) {
-                ret = -EINVAL;
+        fd_ctx = afr_fd_ctx_get (fd, this);
+        if (!fd_ctx)
                 goto out;
-        }
 
-        LOCK (&local->fd->lock);
+        LOCK (&fd->lock);
         {
                 if (fd_ctx->up_count < priv->up_count) {
                         need_self_heal = _gf_true;
@@ -482,55 +403,34 @@ afr_open_fd_fix (call_frame_t *frame, xlator_t *this, gf_boolean_t pause_fop)
                         fd_ctx->down_count = priv->down_count;
                 }
 
+                need_open = alloca (priv->child_count * sizeof (*need_open));
                 for (i = 0; i < priv->child_count; i++) {
-                        if ((fd_ctx->opened_on[i] == AFR_FD_NOT_OPENED) &&
-                            local->child_up[i]) {
-                                fd_ctx->opened_on[i] = AFR_FD_OPENING;
-                                if (!need_open)
-                                        need_open = GF_CALLOC (priv->child_count,
-                                                               sizeof (*need_open),
-                                                               gf_afr_mt_int32_t);
-                                need_open[i] = 1;
-                                need_open_count++;
-                        } else if (pause_fop && local->child_up[i] &&
-                                   (fd_ctx->opened_on[i] == AFR_FD_OPENING)) {
-                                local->fop_paused = _gf_true;
-                        }
-                }
+                        need_open[i] = 0;
+                        if (fd_ctx->opened_on[i] != AFR_FD_NOT_OPENED)
+                                continue;
+
+                        if (!priv->child_up[i])
+                                continue;
+
+                        fd_ctx->opened_on[i] = AFR_FD_OPENING;
 
-                if (local->fop_paused) {
-                        GF_ASSERT (pause_fop);
-                        gf_log (this->name, GF_LOG_INFO, "Pause fd %p",
-                                local->fd);
-                        ret = afr_pause_fd_fop (frame, this, fd_ctx);
-                        if (ret)
-                                goto unlock;
-                        fop_continue = _gf_false;
+                        need_open[i] = 1;
+                        need_open_count++;
                 }
         }
-unlock:
-        UNLOCK (&local->fd->lock);
-        if (ret) {
-                gf_log (this->name, GF_LOG_ERROR, "Failed to fix fd for %s",
-                        local->loc.path);
-                fop_continue = _gf_false;
+        UNLOCK (&fd->lock);
+        if (ret)
                 goto out;
-        }
 
         if (need_self_heal)
-                afr_trigger_open_fd_self_heal (frame, this);
+                afr_trigger_open_fd_self_heal (fd, this);
 
         if (!need_open_count)
                 goto out;
 
-        gf_log (this->name, GF_LOG_INFO, "Opening fd %p", local->fd);
-        afr_fix_open (frame, this, fd_ctx, need_open_count, need_open);
-        fop_continue = _gf_false;
+        afr_fix_open (this, fd, need_open_count, need_open);
 out:
-        GF_FREE (need_open);
-        if (fop_continue && local->fop_call_continue)
-                local->fop_call_continue (frame, this);
-        return ret;
+        return;
 }
 
 int
@@ -570,13 +470,10 @@ afr_writev (call_frame_t *frame, xlator_t *this, fd_t *fd,
         local->cont.writev.iobref     = iobref_ref (iobref);
 
         local->fd                = fd_ref (fd);
-        local->fop_call_continue = afr_do_writev;
 
-        ret = afr_open_fd_fix (frame, this, _gf_true);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
+        afr_open_fd_fix (fd, this);
+
+        afr_do_writev (frame, this);
 
         ret = 0;
 out:
@@ -1026,13 +923,10 @@ afr_ftruncate (call_frame_t *frame, xlator_t *this,
         local->cont.ftruncate.offset  = offset;
 
         local->fd = fd_ref (fd);
-        local->fop_call_continue = afr_do_ftruncate;
 
-        ret = afr_open_fd_fix (frame, this, _gf_true);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
+        afr_open_fd_fix (fd, this);
+
+        afr_do_ftruncate (frame, this);
 
         ret = 0;
 out:
@@ -1448,11 +1342,7 @@ afr_fsetattr (call_frame_t *frame, xlator_t *this,
 
         local->fd                 = fd_ref (fd);
 
-        ret = afr_open_fd_fix (transaction_frame, this, _gf_false);
-        if (ret) {
-                op_errno = -ret;
-                goto out;
-        }
+        afr_open_fd_fix (fd, this);
 
         local->transaction.main_frame = frame;
         local->transaction.start   = LLONG_MAX - 1;