cluster/dht: Optimize readdirp calls in DHT

Bring in option which is supported by posix xlator
to filter out directory's entries from being returned.
DHT would now request non-first subvols to filter out
directory entries.

dht xlator-option readdir-optimize will enable this
optimization

Change-Id: I35224bc81c9657f54f952efac02790276c35ded5
BUG: 838199
Signed-off-by: shishir gowda <sgowda@redhat.com>
Reviewed-on: http://review.gluster.com/3772
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

7 files changed, 71 insertions, 2 deletions

diff --git a/libglusterfs/src/gf-dirent.h b/libglusterfs/src/gf-dirent.h
index 36a5a629cf2..26cb5a66872 100644
--- a/libglusterfs/src/gf-dirent.h
+++ b/libglusterfs/src/gf-dirent.h
@@ -48,6 +48,7 @@ struct _gf_dirent_t {
 	char                                 d_name[0];
 };
 
+#define DT_ISDIR(mode) (mode == DT_DIR)
 
 gf_dirent_t *gf_dirent_for_name (const char *name);
 void gf_dirent_free (gf_dirent_t *entries);
diff --git a/libglusterfs/src/glusterfs.h b/libglusterfs/src/glusterfs.h
index 5c2843851fe..8fc49d3d2c5 100644
--- a/libglusterfs/src/glusterfs.h
+++ b/libglusterfs/src/glusterfs.h
@@ -82,6 +82,8 @@
 #define GF_XATTR_NODE_UUID_KEY  "trusted.glusterfs.node-uuid"
 #define GF_XATTR_VOL_ID_KEY   "trusted.glusterfs.volume-id"
 
+#define GF_READDIR_SKIP_DIRS       "readdir-filter-directories"
+
 #define XATTR_IS_PATHINFO(x)  (strncmp (x, GF_XATTR_PATHINFO_KEY,       \
                                         strlen (GF_XATTR_PATHINFO_KEY)) == 0)
 #define XATTR_IS_NODE_UUID(x) (strncmp (x, GF_XATTR_NODE_UUID_KEY,      \
diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 5b57552c388..fbe18959a43 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -2858,6 +2858,16 @@ done:
                         goto unwind;
                 }
 
+		if (conf->readdir_optimize == _gf_true) {
+                        if (next_subvol != dht_first_up_subvol (this)) {
+                                ret = dict_set_int32 (local->xattr,
+                                                      GF_READDIR_SKIP_DIRS, 1);
+                                if (ret)
+                                        gf_log (this->name, GF_LOG_ERROR,
+					         "dict set failed");
+		        }
+                }
+
                 STACK_WIND (frame, dht_readdirp_cbk,
                             next_subvol, next_subvol->fops->readdirp,
                             local->fd, local->size, next_offset,
@@ -2983,11 +2993,14 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
         xlator_t     *xvol = NULL;
         off_t         xoff = 0;
         int           ret = 0;
+        dht_conf_t   *conf = NULL;
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
         VALIDATE_OR_GOTO (fd, err);
 
+        conf = this->private;
+
         local = dht_local_init (frame, NULL, NULL, whichop);
         if (!local) {
                 op_errno = ENOMEM;
@@ -3015,6 +3028,16 @@ dht_do_readdir (call_frame_t *frame, xlator_t *this, fd_t *fd, size_t size,
                                 gf_log (this->name, GF_LOG_WARNING,
                                         "failed to set 'glusterfs.dht.linkto'"
                                         " key");
+			if (conf->readdir_optimize == _gf_true) {
+                                if (xvol != dht_first_up_subvol (this)) {
+				        ret = dict_set_int32 (local->xattr,
+			                               GF_READDIR_SKIP_DIRS, 1);
+				        if (ret)
+					        gf_log (this->name,
+                                                        GF_LOG_ERROR,
+						        "Dict set failed");
+                                }
+			}
                 }
 
                 STACK_WIND (frame, dht_readdirp_cbk, xvol, xvol->fops->readdirp,
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index d3ccacdb8c6..c02c0b83d21 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -249,6 +249,10 @@ struct dht_conf {
 
         /* defrag related */
         gf_defrag_info_t *defrag;
+
+        /* Request to filter directory entries in readdir request */
+
+        gf_boolean_t    readdir_optimize;
 };
 typedef struct dht_conf dht_conf_t;
 
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 6b156c1d354..a802d895d03 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -329,6 +329,8 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF ("directory-layout-spread", conf->dir_spread_cnt,
                           options, uint32, out);
 
+        GF_OPTION_RECONF ("readdir-optimize", conf->readdir_optimize, options,
+                          bool, out);
         if (conf->defrag) {
                 GF_OPTION_RECONF ("rebalance-stats", conf->defrag->stats,
                                   options, bool, out);
@@ -435,6 +437,8 @@ init (xlator_t *this)
         GF_OPTION_INIT ("assert-no-child-down", conf->assert_no_child_down,
                         bool, err);
 
+        GF_OPTION_INIT ("readdir-optimize", conf->readdir_optimize, bool, err);
+
         if (defrag) {
                 GF_OPTION_INIT ("rebalance-stats", defrag->stats, bool, err);
         }
@@ -603,6 +607,10 @@ struct volume_options options[] = {
           .type = GF_OPTION_TYPE_BOOL,
           .default_value = "off",
         },
+        { .key = {"readdir-optimize"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "off",
+        },
 
         { .key  = {NULL} },
 };
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 7444db0b523..1bdf5a19ac0 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -121,6 +121,7 @@ static struct volopt_map_entry glusterd_volopt_map[] = {
         {"cluster.min-free-inodes",              "cluster/distribute", NULL, NULL, NO_DOC, 0    },
         {"cluster.rebalance-stats",              "cluster/distribute", NULL, NULL, NO_DOC, 0    },
         {"cluster.subvols-per-directory",        "cluster/distribute", "directory-layout-spread", NULL, NO_DOC, 0    },
+        {"cluster.readdir-optimize",             "cluster/distribute", NULL, NULL, NO_DOC, 0    },
 
         {"cluster.entry-change-log",             "cluster/replicate",  NULL, NULL, NO_DOC, 0     },
         {"cluster.read-subvolume",               "cluster/replicate",  NULL, NULL, NO_DOC, 0    },
diff --git a/xlators/storage/posix/src/posix.c b/xlators/storage/posix/src/posix.c
index e2f1bbd5f97..6d0f8caa339 100644
--- a/xlators/storage/posix/src/posix.c
+++ b/xlators/storage/posix/src/posix.c
@@ -3547,7 +3547,7 @@ posix_fentrylk (call_frame_t *frame, xlator_t *this,
 
 int
 posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t size,
-                    gf_dirent_t *entries)
+                    gf_dirent_t *entries, xlator_t *this, int32_t skip_dirs)
 {
         off_t     in_case = -1;
         size_t    filled = 0;
@@ -3557,6 +3557,18 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t size,
         int32_t               this_size      = -1;
         gf_dirent_t          *this_entry     = NULL;
         uuid_t                rootgfid = {0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1};
+        struct stat           stbuf = {0,};
+        char                 *hpath = NULL;
+        int                   len = 0;
+        int                   ret = 0;
+
+        if (skip_dirs) {
+                len = posix_handle_path (this, fd->inode->gfid, NULL, NULL, 0);
+                hpath = alloca (len + 256); /* NAME_MAX */
+                posix_handle_path (this, fd->inode->gfid, NULL, hpath, len);
+                len = strlen (hpath);
+                hpath[len] = '/';
+        }
 
         if (!off) {
                 rewinddir (dir);
@@ -3607,6 +3619,17 @@ posix_fill_readdir (fd_t *fd, DIR *dir, off_t off, size_t size,
                         continue;
                 }
 
+                if (skip_dirs) {
+                        if (DT_ISDIR (entry->d_type)) {
+                                continue;
+                        } else if (hpath) {
+                                strcpy (&hpath[len+1],entry->d_name);
+                                ret = lstat (hpath, &stbuf);
+                                if (!ret && S_ISDIR (stbuf.st_mode))
+                                        continue;
+                        }
+                }
+
                 this_size = max (sizeof (gf_dirent_t),
                                  sizeof (gfs3_dirplist))
                         + strlen (entry->d_name) + 1;
@@ -3728,6 +3751,7 @@ posix_do_readdir (call_frame_t *frame, xlator_t *this,
         int32_t               op_ret         = -1;
         int32_t               op_errno       = 0;
         gf_dirent_t           entries;
+        int32_t               skip_dirs      = 0;
 
 
         VALIDATE_OR_GOTO (frame, out);
@@ -3753,7 +3777,13 @@ posix_do_readdir (call_frame_t *frame, xlator_t *this,
                 goto out;
 	}
 
-        count = posix_fill_readdir (fd, dir, off, size, &entries);
+        /* When READDIR_FILTER option is set to on, we can filter out
+         * directory's entry from the entry->list.
+         */
+        ret = dict_get_int32 (dict, GF_READDIR_SKIP_DIRS, &skip_dirs);
+
+        count = posix_fill_readdir (fd, dir, off, size, &entries, this,
+                                    skip_dirs);
 
         /* pick ENOENT to indicate EOF */
         op_errno = errno;