replicate: default read_child to a local brick if there is one.

Controlled by the "choose-local" option (on by default).

Change-Id: I560f27c81703f2c9c62fdb51532c8eb763826df7
BUG: 806462
Signed-off-by: Jeff Darcy <jdarcy@redhat.com>
Reviewed-on: http://review.gluster.com/3005
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Anand Avati <avati@redhat.com>

7 files changed, 165 insertions, 5 deletions

diff --git a/xlators/cluster/afr/src/afr-common.c b/xlators/cluster/afr/src/afr-common.c
index 21a2be3dd6f..f24bd8b7d7f 100644
--- a/xlators/cluster/afr/src/afr-common.c
+++ b/xlators/cluster/afr/src/afr-common.c
@@ -2035,12 +2035,79 @@ afr_lookup_handle_first_success (afr_local_t *local, xlator_t *this,
         afr_set_root_inode_on_first_lookup (local, this, inode);
 }
 
+static int32_t
+afr_discovery_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
+                   int32_t op_ret, int32_t op_errno, dict_t *dict,
+                   dict_t *xdata)
+{
+        int              ret            = 0;
+        char            *pathinfo       = NULL;
+        gf_boolean_t     is_local        = _gf_false;
+        afr_private_t   *priv           = NULL;
+        int32_t          child_index    = -1;
+
+        if (op_ret != 0) {
+                goto out;
+        }
+
+        ret = dict_get_str (dict, GF_XATTR_PATHINFO_KEY, &pathinfo);
+        if (ret != 0) {
+                goto out;
+        }
+
+        ret = afr_local_pathinfo (pathinfo, &is_local);
+        if (ret) {
+                goto out;
+        }
+
+        priv = this->private;
+        /*
+         * Note that one local subvolume will override another here.  The only
+         * way to avoid that would be to retain extra information about whether
+         * the previous read_child is local, and it's just not worth it.  Even
+         * the slowest local subvolume is far preferable to a remote one.
+         */
+        if (is_local) {
+                child_index = (int32_t)(long)cookie;
+                gf_log (this->name, GF_LOG_INFO,
+                        "selecting local read_child %s",
+                        priv->children[child_index]->name);
+                priv->read_child = child_index;
+        }
+
+out:
+        STACK_DESTROY(frame->root);
+        return 0;
+}
+
+static void
+afr_attempt_local_discovery (xlator_t *this, int32_t child_index)
+{
+        call_frame_t    *newframe = NULL;
+        loc_t            tmploc = {0,};
+        afr_private_t   *priv = this->private;
+
+        newframe = create_frame(this,this->ctx->pool);
+        if (!newframe) {
+                return;
+        }
+
+        tmploc.gfid[sizeof(tmploc.gfid)-1] = 1;
+        STACK_WIND_COOKIE (newframe, afr_discovery_cbk,
+                           (void *)(long)child_index,
+                           priv->children[child_index],
+                           priv->children[child_index]->fops->getxattr,
+                           &tmploc, GF_XATTR_PATHINFO_KEY, NULL);
+}
+
 static void
 afr_lookup_handle_success (afr_local_t *local, xlator_t *this, int32_t child_index,
                            int32_t op_ret, int32_t op_errno, inode_t *inode,
                            struct iatt *buf, dict_t *xattr,
                            struct iatt *postparent)
 {
+        afr_private_t   *priv   = this->private;
+
         if (local->success_count == 0) {
                 if (local->op_errno != ESTALE) {
                         local->op_ret = op_ret;
@@ -2053,6 +2120,11 @@ afr_lookup_handle_success (afr_local_t *local, xlator_t *this, int32_t child_ind
 
         afr_lookup_cache_args (local, child_index, xattr,
                                buf, postparent);
+
+        if (local->do_discovery && (priv->read_child == (-1))) {
+                afr_attempt_local_discovery(this,child_index);
+        }
+
         local->cont.lookup.success_children[local->success_count] = child_index;
         local->success_count++;
 }
@@ -2214,8 +2286,6 @@ afr_lookup (call_frame_t *frame, xlator_t *this,
         /* By default assume ENOTCONN. On success it will be set to 0. */
         local->op_errno = ENOTCONN;
 
-        local->call_count = afr_up_children_count (local->child_up,
-                                                   priv->child_count);
         ret = afr_lookup_xattr_req_prepare (local, this, xattr_req, &local->loc,
                                             &gfid_req);
         if (ret) {
@@ -2225,6 +2295,12 @@ afr_lookup (call_frame_t *frame, xlator_t *this,
         afr_lookup_save_gfid (local->cont.lookup.gfid_req, gfid_req,
                               &local->loc);
         local->fop = GF_FOP_LOOKUP;
+        if (priv->choose_local && !priv->did_discovery) {
+                if (__is_root_gfid(gfid_req)) {
+                        local->do_discovery = _gf_true;
+                        priv->did_discovery = _gf_true;
+                }
+        }
         for (i = 0; i < priv->child_count; i++) {
                 if (local->child_up[i]) {
                         STACK_WIND_COOKIE (frame, afr_lookup_cbk,
@@ -3626,6 +3702,14 @@ afr_notify (xlator_t *this, int32_t event,
         if (!priv)
                 return 0;
 
+        /*
+         * We need to reset this in case children come up in "staggered"
+         * fashion, so that we discover a late-arriving local subvolume.  Note
+         * that we could end up issuing N lookups to the first subvolume, and
+         * O(N^2) overall, but N is small for AFR so it shouldn't be an issue.
+         */
+        priv->did_discovery = _gf_false;
+
         had_heard_from_all = 1;
         for (i = 0; i < priv->child_count; i++) {
                 if (!priv->last_event[i]) {
diff --git a/xlators/cluster/afr/src/afr-self-heald.c b/xlators/cluster/afr/src/afr-self-heald.c
index 575bf4361d2..c5deb18b8af 100644
--- a/xlators/cluster/afr/src/afr-self-heald.c
+++ b/xlators/cluster/afr/src/afr-self-heald.c
@@ -671,6 +671,62 @@ out:
         return;
 }
 
+static int
+get_pathinfo_host (char *pathinfo, char *hostname, size_t size)
+{
+        char    *start = NULL;
+        char    *end = NULL;
+        int     ret  = -1;
+        int     i    = 0;
+
+        if (!pathinfo)
+                goto out;
+
+        start = strchr (pathinfo, ':');
+        if (!start)
+                goto out;
+        end = strrchr (pathinfo, ':');
+        if (start == end)
+                goto out;
+
+        memset (hostname, 0, size);
+        i = 0;
+        while (++start != end)
+                hostname[i++] = *start;
+        ret = 0;
+out:
+        return ret;
+}
+
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *local)
+{
+        int             ret   = 0;
+        char            pathinfohost[1024] = {0};
+        char            localhost[1024] = {0};
+        xlator_t        *this = THIS;
+
+        *local = _gf_false;
+        ret = get_pathinfo_host (pathinfo, pathinfohost, sizeof (pathinfohost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "Invalid pathinfo: %s",
+                        pathinfo);
+                goto out;
+        }
+
+        ret = gethostname (localhost, sizeof (localhost));
+        if (ret) {
+                gf_log (this->name, GF_LOG_ERROR, "gethostname() failed, "
+                        "reason: %s", strerror (errno));
+                goto out;
+        }
+
+        if (!strcmp (localhost, pathinfohost))
+                *local = _gf_true;
+out:
+        return ret;
+}
+
 int
 afr_crawl_build_start_loc (xlator_t *this, afr_crawl_data_t *crawl_data,
                            loc_t *dirloc)
diff --git a/xlators/cluster/afr/src/afr-self-heald.h b/xlators/cluster/afr/src/afr-self-heald.h
index 8e608459684..32a8aaca50c 100644
--- a/xlators/cluster/afr/src/afr-self-heald.h
+++ b/xlators/cluster/afr/src/afr-self-heald.h
@@ -42,4 +42,11 @@ afr_proactive_self_heal (void *data);
 
 int
 afr_xl_op (xlator_t *this, dict_t *input, dict_t *output);
+
+/*
+ * In addition to its self-heal use, this is used to find a local default
+ * read_child.
+ */
+int
+afr_local_pathinfo (char *pathinfo, gf_boolean_t *local);
 #endif /* __AFR_SELF_HEALD_H__ */
diff --git a/xlators/cluster/afr/src/afr.c b/xlators/cluster/afr/src/afr.c
index b7ba2619711..4f7bf2de004 100644
--- a/xlators/cluster/afr/src/afr.c
+++ b/xlators/cluster/afr/src/afr.c
@@ -170,6 +170,9 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF ("heal-timeout", priv->shd.timeout, options,
                           int32, out);
 
+        /* Reset this so we re-discover in case the topology changed.  */
+        priv->did_discovery = _gf_false;
+
         ret = 0;
 out:
         return ret;
@@ -227,7 +230,6 @@ init (xlator_t *this)
 
         priv->child_count = child_count;
 
-
         priv->read_child = -1;
 
         GF_OPTION_INIT ("read-subvolume", read_subvol, xlator, out);
@@ -239,6 +241,7 @@ init (xlator_t *this)
                         goto out;
                 }
         }
+        GF_OPTION_INIT ("choose-local", priv->choose_local, bool, out);
 
         GF_OPTION_INIT ("read-hash-mode", priv->hash_mode, uint32, out);
 
@@ -508,6 +511,12 @@ struct volume_options options[] = {
                          "1 = hash by GFID (all clients use same subvolume), "
                          "2 = hash by GFID and client PID",
         },
+        { .key  = {"choose-local" },
+          .type = GF_OPTION_TYPE_BOOL,
+          .default_value = "true",
+          .description = "Choose a local subvolume to read from if "
+                         "read-subvolume is not explicitly set.",
+        },
         { .key  = {"favorite-child"},
           .type = GF_OPTION_TYPE_XLATOR
         },
diff --git a/xlators/cluster/afr/src/afr.h b/xlators/cluster/afr/src/afr.h
index a1a30562bf1..c8e01fcb841 100644
--- a/xlators/cluster/afr/src/afr.h
+++ b/xlators/cluster/afr/src/afr.h
@@ -156,6 +156,8 @@ typedef struct _afr_private {
         char                   vol_uuid[UUID_SIZE + 1];
         int32_t                *last_event;
         afr_self_heald_t       shd;
+        gf_boolean_t           choose_local;
+        gf_boolean_t           did_discovery;
 } afr_private_t;
 
 typedef struct {
@@ -697,6 +699,7 @@ typedef struct _afr_local {
 
         mode_t          umask;
         int             xflag;
+        gf_boolean_t    do_discovery;
 } afr_local_t;
 
 typedef enum {
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 515bff359bc..219352c6147 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -136,6 +136,7 @@ static struct volopt_map_entry glusterd_volopt_map[] = {
         {"cluster.eager-lock",                   "cluster/replicate",  NULL, NULL, NO_DOC, 0     },
         {"cluster.quorum-type",                  "cluster/replicate",  "quorum-type", NULL, NO_DOC, 0},
         {"cluster.quorum-count",                 "cluster/replicate",  "quorum-count", NULL, NO_DOC, 0},
+        {"cluster.choose-local",                 "cluster/replicate",  NULL, NULL, DOC, 0},
 
         {"cluster.stripe-block-size",            "cluster/stripe",     "block-size", NULL, DOC, 0},
 
diff --git a/xlators/protocol/client/src/client3_1-fops.c b/xlators/protocol/client/src/client3_1-fops.c
index 266f84a1dfb..d44eb86c46c 100644
--- a/xlators/protocol/client/src/client3_1-fops.c
+++ b/xlators/protocol/client/src/client3_1-fops.c
@@ -4688,7 +4688,7 @@ client3_1_getxattr (call_frame_t *frame, xlator_t *this,
         }
         args = data;
 
-        if (!(args->loc && args->loc->inode)) {
+        if (!args->loc) {
                 op_errno = EINVAL;
                 goto unwind;
         }
@@ -4729,7 +4729,7 @@ client3_1_getxattr (call_frame_t *frame, xlator_t *this,
         rsp_iobuf = NULL;
         rsp_iobref = NULL;
 
-        if (!uuid_is_null (args->loc->inode->gfid))
+        if (args->loc->inode && !uuid_is_null (args->loc->inode->gfid))
                 memcpy (req.gfid,  args->loc->inode->gfid, 16);
         else
                 memcpy (req.gfid, args->loc->gfid, 16);