libglusterfs/client_t client_t implementation, phase 1

Implementation of client_t

The feature page for client_t is at
http://www.gluster.org/community/documentation/index.php/Planning34/client_t

In addition to adding libglusterfs/client_t.[ch] it also extracts/moves
the locktable functionality from xlators/protocol/server to libglusterfs,
where it is used; thus it may now be shared by other xlators too.

This patch is large as it is. Hooking up the state dump is left to do
in phase 2 of this patch set.

(N.B. this change/patch-set supercedes previous change 3689, which was
corrupted during a rebase. That change will be abandoned.)

BUG: 849630
Change-Id: I1433743190630a6d8119a72b81439c0c4c990340
Signed-off-by: Kaleb S. KEITHLEY <kkeithle@redhat.com>
Reviewed-on: http://review.gluster.org/3957
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Amar Tumballi <amarts@redhat.com>

7 files changed, 1292 insertions, 52 deletions

diff --git a/libglusterfs/src/Makefile.am b/libglusterfs/src/Makefile.am
index 80753db3008..fc54f49a359 100644
--- a/libglusterfs/src/Makefile.am
+++ b/libglusterfs/src/Makefile.am
@@ -4,7 +4,7 @@ libglusterfs_la_CFLAGS = -Wall $(GF_CFLAGS) \
 
 libglusterfs_la_CPPFLAGS = $(GF_CPPFLAGS) -D__USE_FILE_OFFSET64 \
 	-DXLATORDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/xlator\" \
-	-I$(CONTRIBDIR)/rbtree
+	-I$(top_srcdir)/rpc/rpc-lib/src/ -I$(CONTRIBDIR)/rbtree
 
 libglusterfs_la_LIBADD = @LEXLIB@
 
@@ -23,7 +23,7 @@ libglusterfs_la_SOURCES = dict.c xlator.c logging.c \
 	$(CONTRIBDIR)/uuid/uuid_time.c $(CONTRIBDIR)/uuid/compare.c \
 	$(CONTRIBDIR)/uuid/isnull.c $(CONTRIBDIR)/uuid/unpack.c syncop.c \
 	graph-print.c trie.c run.c options.c fd-lk.c circ-buff.c \
-	event-history.c gidcache.c ctx.c \
+	event-history.c gidcache.c ctx.c client_t.c lock-table.c \
 	$(CONTRIBDIR)/libgen/basename_r.c $(CONTRIBDIR)/libgen/dirname_r.c \
 	$(CONTRIBDIR)/stdlib/gf_mkostemp.c \
 	event-poll.c event-epoll.c
@@ -40,8 +40,9 @@ noinst_HEADERS = common-utils.h defaults.h dict.h glusterfs.h hashfn.h \
 	checksum.h daemon.h $(CONTRIBDIR)/rbtree/rb.h store.h\
 	rbthash.h iatt.h latency.h mem-types.h $(CONTRIBDIR)/uuid/uuidd.h \
 	$(CONTRIBDIR)/uuid/uuid.h $(CONTRIBDIR)/uuid/uuidP.h \
-	$(CONTRIB_BUILDDIR)/uuid/uuid_types.h syncop.h graph-utils.h trie.h run.h \
-	options.h lkowner.h fd-lk.h circ-buff.h event-history.h gidcache.h
+	$(CONTRIB_BUILDDIR)/uuid/uuid_types.h syncop.h graph-utils.h trie.h \
+	run.h options.h lkowner.h fd-lk.h circ-buff.h event-history.h \
+	gidcache.h client_t.h lock-table.h
 
 EXTRA_DIST = graph.l graph.y
 
diff --git a/libglusterfs/src/client_t.c b/libglusterfs/src/client_t.c
new file mode 100644
index 00000000000..f0d66da3a3d
--- /dev/null
+++ b/libglusterfs/src/client_t.c
@@ -0,0 +1,881 @@
+/*
+  Copyright (c) 2008-2013 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "glusterfs.h"
+#include "dict.h"
+#include "statedump.h"
+#include "lock-table.h"
+#include "rpcsvc.h"
+#include "client_t.h"
+
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+
+static int
+gf_client_chain_client_entries (cliententry_t *entries, uint32_t startidx,
+                        uint32_t endcount)
+{
+        uint32_t        i = 0;
+
+        if (!entries) {
+                gf_log_callingfn ("client_t", GF_LOG_WARNING, "!entries");
+                return -1;
+        }
+
+        /* Chain only till the second to last entry because we want to
+         * ensure that the last entry has GF_CLIENTTABLE_END.
+         */
+        for (i = startidx; i < (endcount - 1); i++)
+                entries[i].next_free = i + 1;
+
+        /* i has already been incremented upto the last entry. */
+        entries[i].next_free = GF_CLIENTTABLE_END;
+
+        return 0;
+}
+
+
+static int
+gf_client_clienttable_expand (clienttable_t *clienttable, uint32_t nr)
+{
+        cliententry_t   *oldclients     = NULL;
+        uint32_t         oldmax_clients = -1;
+        int              ret            = -1;
+
+        if (clienttable == NULL || nr < 0) {
+                gf_log_callingfn ("client_t", GF_LOG_ERROR, "invalid argument");
+                ret = EINVAL;
+                goto out;
+        }
+
+        /* expand size by power-of-two...
+           this originally came from .../xlators/protocol/server/src/server.c
+           where it was not commented */
+        nr /= (1024 / sizeof (cliententry_t));
+        nr = gf_roundup_next_power_of_two (nr + 1);
+        nr *= (1024 / sizeof (cliententry_t));
+
+        oldclients = clienttable->cliententries;
+        oldmax_clients = clienttable->max_clients;
+
+        clienttable->cliententries = GF_CALLOC (nr, sizeof (cliententry_t),
+                                                gf_common_mt_cliententry_t);
+        if (!clienttable->cliententries) {
+                ret = ENOMEM;
+                goto out;
+        }
+        clienttable->max_clients = nr;
+
+        if (oldclients) {
+                uint32_t cpy = oldmax_clients * sizeof (cliententry_t);
+                memcpy (clienttable->cliententries, oldclients, cpy);
+        }
+
+        gf_client_chain_client_entries (clienttable->cliententries, oldmax_clients,
+                                clienttable->max_clients);
+
+        /* Now that expansion is done, we must update the client list
+         * head pointer so that the client allocation functions can continue
+         * using the expanded table.
+         */
+        clienttable->first_free = oldmax_clients;
+        GF_FREE (oldclients);
+        ret = 0;
+out:
+        return ret;
+}
+
+
+clienttable_t *
+gf_clienttable_alloc (void)
+{
+        clienttable_t *clienttable = NULL;
+
+        clienttable =
+                GF_CALLOC (1, sizeof (*clienttable), gf_common_mt_clienttable_t);
+        if (!clienttable)
+                return NULL;
+
+        LOCK_INIT (&clienttable->lock);
+        gf_client_clienttable_expand (clienttable, GF_CLIENTTABLE_INITIAL_SIZE);
+        return clienttable;
+}
+
+
+void
+gf_client_clienttable_destroy (clienttable_t *clienttable)
+{
+        struct list_head  list          = {0, };
+        client_t         *client        = NULL;
+        cliententry_t    *cliententries = NULL;
+        uint32_t          client_count  = 0;
+        int32_t           i             = 0;
+
+        INIT_LIST_HEAD (&list);
+
+        if (!clienttable) {
+                gf_log_callingfn ("client_t", GF_LOG_WARNING, "!clienttable");
+                return;
+        }
+
+        LOCK (&clienttable->lock);
+        {
+                client_count = clienttable->max_clients;
+                clienttable->max_clients = 0;
+                cliententries = clienttable->cliententries;
+                clienttable->cliententries = NULL;
+        }
+        UNLOCK (&clienttable->lock);
+
+        if (cliententries != NULL) {
+                for (i = 0; i < client_count; i++) {
+                        client = cliententries[i].client;
+                        if (client != NULL) {
+                                gf_client_unref (client);
+                        }
+                }
+
+                GF_FREE (cliententries);
+                LOCK_DESTROY (&clienttable->lock);
+                GF_FREE (clienttable);
+        }
+}
+
+
+client_t *
+gf_client_get (xlator_t *this, rpcsvc_auth_data_t *cred, char *client_uid)
+{
+        client_t      *client      = NULL;
+        cliententry_t *cliententry = NULL;
+        clienttable_t *clienttable = NULL;
+        unsigned int   i           = 0;
+
+        if (this == NULL || client_uid == NULL) {
+                gf_log_callingfn ("client_t", GF_LOG_ERROR, "invalid argument");
+                errno = EINVAL;
+                return NULL;
+        }
+
+        gf_log (this->name, GF_LOG_INFO, "client_uid=%s", client_uid);
+
+        clienttable = this->ctx->clienttable;
+
+        LOCK (&clienttable->lock);
+        {
+                for (; i < clienttable->max_clients; i++) {
+                        client = clienttable->cliententries[i].client;
+                        if (client == NULL)
+                                continue;
+                        /*
+                         * look for matching client_uid, _and_
+                         * if auth was used, matching auth flavour and data
+                         */
+                        if (strcmp (client_uid, client->server_ctx.client_uid) == 0 &&
+                                (cred->flavour != AUTH_NONE &&
+                                        (cred->flavour == client->server_ctx.auth.flavour &&
+                                        (size_t) cred->datalen == client->server_ctx.auth.len &&
+                                        memcmp (cred->authdata,
+                                                client->server_ctx.auth.data,
+                                                client->server_ctx.auth.len) == 0))) {
+#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)
+                                __sync_add_and_fetch(&client->ref.bind, 1);
+#else
+                                LOCK (&client->ref.lock);
+                                {
+                                        ++client->ref.bind;
+                                }
+                                UNLOCK (&client->ref.lock);
+#endif
+                                break;
+                        }
+                }
+                if (client) {
+                        gf_client_ref (client);
+                        goto unlock;
+                }
+                client = GF_CALLOC (1, sizeof(client_t), gf_common_mt_client_t);
+                if (client == NULL) {
+                        errno = ENOMEM;
+                        goto unlock;
+                }
+
+                client->this = this;
+             /* client->server_ctx.lk_version = 0; redundant */
+
+                LOCK_INIT (&client->server_ctx.fdtable_lock);
+                LOCK_INIT (&client->locks_ctx.ltable_lock);
+                LOCK_INIT (&client->scratch_ctx.lock);
+                LOCK_INIT (&client->ref.lock);
+
+                client->server_ctx.client_uid = gf_strdup (client_uid);
+                if (client->server_ctx.client_uid == NULL) {
+                        errno = ENOMEM;
+                        GF_FREE (client);
+                        client = NULL;
+                        goto unlock;
+                }
+                client->server_ctx.fdtable = gf_fd_fdtable_alloc ();
+                if (client->server_ctx.fdtable == NULL) {
+                        errno = ENOMEM;
+                        GF_FREE (client->server_ctx.client_uid);
+                        GF_FREE (client);
+                        client = NULL;
+                        goto unlock;
+                }
+
+                client->locks_ctx.ltable = gf_lock_table_new ();
+                if (client->locks_ctx.ltable == NULL) {
+                        errno = ENOMEM;
+                        GF_FREE (client->server_ctx.fdtable);
+                        GF_FREE (client->server_ctx.client_uid);
+                        GF_FREE (client);
+                        client = NULL;
+                        goto unlock;
+                }
+
+                /* no need to do these atomically here */
+                client->ref.bind = client->ref.count = 1;
+
+                client->server_ctx.auth.flavour = cred->flavour;
+                if (cred->flavour != AUTH_NONE) {
+                        client->server_ctx.auth.data =
+                                GF_CALLOC (1, cred->datalen, gf_common_mt_client_t);
+                        if (client->server_ctx.auth.data == NULL) {
+                                errno = ENOMEM;
+                                GF_FREE (client->locks_ctx.ltable);
+                                GF_FREE (client->server_ctx.fdtable);
+                                GF_FREE (client->server_ctx.client_uid);
+                                GF_FREE (client);
+                                client = NULL;
+                                goto unlock;
+                        }
+                        memcpy (client->server_ctx.auth.data, cred->authdata,
+                                cred->datalen);
+                        client->server_ctx.auth.len = cred->datalen;
+                }
+
+                client->tbl_index = clienttable->first_free;
+                cliententry = &clienttable->cliententries[client->tbl_index];
+                cliententry->client = client;
+                clienttable->first_free = cliententry->next_free;
+                cliententry->next_free = GF_CLIENTENTRY_ALLOCATED;
+                gf_client_ref (client);
+        }
+unlock:
+        UNLOCK (&clienttable->lock);
+
+        return client;
+}
+
+void
+gf_client_put (client_t *client, gf_boolean_t *detached)
+{
+        gf_boolean_t unref = _gf_false;
+        int bind_ref;
+
+        if (detached)
+                *detached = _gf_false;
+
+#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)
+        bind_ref = __sync_sub_and_fetch(&client->ref.bind, 1);
+#else
+        LOCK (&client->ref.lock);
+        {
+                bind_ref = --client->ref.bind;
+        }
+        UNLOCK (&client->ref.lock);
+#endif
+        if (bind_ref == 0)
+                unref = _gf_true;
+
+        if (unref) {
+                gf_log (THIS->name, GF_LOG_INFO, "Shutting down connection %s",
+                        client->server_ctx.client_uid);
+                if (detached)
+                        *detached = _gf_true;
+                gf_client_unref (client);
+        }
+}
+
+
+client_t *
+gf_client_ref (client_t *client)
+{
+        if (!client) {
+                gf_log_callingfn ("client_t", GF_LOG_ERROR, "null client");
+                return NULL;
+        }
+
+#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)
+        __sync_add_and_fetch(&client->ref.count, 1);
+#else
+        LOCK (&client->ref.lock);
+        {
+                ++client->ref.count;
+        }
+        UNLOCK (&client->ref.lock);
+#endif
+        return client;
+}
+
+
+static void
+client_destroy (client_t *client)
+{
+        clienttable_t *clienttable = NULL;
+
+        if (client == NULL){
+                gf_log_callingfn ("xlator", GF_LOG_ERROR, "invalid argument");
+                goto out;
+        }
+
+        clienttable = client->this->ctx->clienttable;
+
+        LOCK_DESTROY (&client->server_ctx.fdtable_lock);
+        LOCK_DESTROY (&client->locks_ctx.ltable_lock);
+        LOCK_DESTROY (&client->scratch_ctx.lock);
+        LOCK_DESTROY (&client->ref.lock);
+
+        LOCK (&clienttable->lock);
+        {
+                clienttable->cliententries[client->tbl_index].client = NULL;
+                clienttable->cliententries[client->tbl_index].next_free =
+                        clienttable->first_free;
+                clienttable->first_free = client->tbl_index;
+        }
+        UNLOCK (&clienttable->lock);
+
+        GF_FREE (client->server_ctx.auth.data);
+        GF_FREE (client->scratch_ctx.ctx);
+        GF_FREE (client->locks_ctx.ltable);
+        GF_FREE (client->server_ctx.fdtable);
+        GF_FREE (client->server_ctx.client_uid);
+        GF_FREE (client);
+out:
+        return;
+}
+
+
+void
+gf_client_unref (client_t *client)
+{
+        int refcount;
+
+        if (!client) {
+                gf_log_callingfn ("client_t", GF_LOG_ERROR, "client is NULL");
+                return;
+        }
+
+#if __GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)
+        refcount = __sync_sub_and_fetch(&client->ref.count, 1);
+#else
+        LOCK (&client->ref.lock);
+        {
+                refcount = --client->ref.count;
+        }
+        UNLOCK (&client->ref.lock);
+#endif
+        if (refcount == 0) {
+                client_destroy (client);
+        }
+}
+
+
+int
+__client_ctx_set (client_t *client, xlator_t *xlator, uint64_t value)
+{
+        int index   = 0;
+        int ret     = 0;
+        int set_idx = -1;
+
+        if (!client || !xlator)
+                return -1;
+
+        for (index = 0; index < client->scratch_ctx.count; index++) {
+                if (!client->scratch_ctx.ctx[index].key) {
+                        if (set_idx == -1)
+                                set_idx = index;
+                        /* dont break, to check if key already exists
+                           further on */
+                }
+                if (client->scratch_ctx.ctx[index].xl_key == xlator) {
+                        set_idx = index;
+                        break;
+                }
+        }
+
+        if (set_idx == -1) {
+                gf_log_callingfn ("", GF_LOG_WARNING, "%p %s", client, xlator->name);
+                ret = -1;
+                goto out;
+        }
+
+        client->scratch_ctx.ctx[set_idx].xl_key = xlator;
+        client->scratch_ctx.ctx[set_idx].value  = value;
+
+out:
+        return ret;
+}
+
+
+int
+client_ctx_set (client_t *client, xlator_t *xlator, uint64_t value)
+{
+        int ret = 0;
+
+        if (!client || !xlator) {
+                gf_log_callingfn ("", GF_LOG_WARNING, "%p %p", client, xlator);
+                return -1;
+        }
+
+        LOCK (&client->scratch_ctx.lock);
+        {
+                ret = __client_ctx_set (client, xlator, value);
+        }
+        UNLOCK (&client->scratch_ctx.lock);
+
+        return ret;
+}
+
+
+int
+__client_ctx_get (client_t *client, xlator_t *xlator, uint64_t *value)
+{
+        int index = 0;
+        int ret   = 0;
+
+        if (!client || !xlator)
+                return -1;
+
+        for (index = 0; index < client->scratch_ctx.count; index++) {
+                if (client->scratch_ctx.ctx[index].xl_key == xlator)
+                        break;
+        }
+
+        if (index == client->scratch_ctx.count) {
+                ret = -1;
+                goto out;
+        }
+
+        if (value)
+                *value = client->scratch_ctx.ctx[index].value;
+
+out:
+        return ret;
+}
+
+
+int
+client_ctx_get (client_t *client, xlator_t *xlator, uint64_t *value)
+{
+        int ret = 0;
+
+        if (!client || !xlator)
+                return -1;
+
+        LOCK (&client->scratch_ctx.lock);
+        {
+                ret = __client_ctx_get (client, xlator, value);
+        }
+        UNLOCK (&client->scratch_ctx.lock);
+
+        return ret;
+}
+
+
+int
+__client_ctx_del (client_t *client, xlator_t *xlator, uint64_t *value)
+{
+        int index = 0;
+        int ret   = 0;
+
+        if (!client || !xlator)
+                return -1;
+
+        for (index = 0; index < client->scratch_ctx.count; index++) {
+                if (client->scratch_ctx.ctx[index].xl_key == xlator)
+                        break;
+        }
+
+        if (index == client->scratch_ctx.count) {
+                ret = -1;
+                goto out;
+        }
+
+        if (value)
+                *value = client->scratch_ctx.ctx[index].value;
+
+        client->scratch_ctx.ctx[index].key   = 0;
+        client->scratch_ctx.ctx[index].value = 0;
+
+out:
+        return ret;
+}
+
+
+int
+client_ctx_del (client_t *client, xlator_t *xlator, uint64_t *value)
+{
+        int ret = 0;
+
+        if (!client || !xlator)
+                return -1;
+
+        LOCK (&client->scratch_ctx.lock);
+        {
+                ret = __client_ctx_del (client, xlator, value);
+        }
+        UNLOCK (&client->scratch_ctx.lock);
+
+        return ret;
+}
+
+
+void
+client_dump (client_t *client, char *prefix)
+{
+        char key[GF_DUMP_MAX_BUF_LEN];
+
+        if (!client)
+                return;
+
+        memset(key, 0, sizeof key);
+        gf_proc_dump_write("refcount", "%d", client->ref.count);
+}
+
+
+void
+cliententry_dump (cliententry_t *cliententry, char *prefix)
+{
+        if (!cliententry)
+                return;
+
+        if (GF_CLIENTENTRY_ALLOCATED != cliententry->next_free)
+                return;
+
+        if (cliententry->client)
+                client_dump(cliententry->client, prefix);
+}
+
+
+void
+clienttable_dump (clienttable_t *clienttable, char *prefix)
+{
+        int     i   = 0;
+        int     ret = -1;
+        char    key[GF_DUMP_MAX_BUF_LEN] = {0};
+
+        if (!clienttable)
+                return;
+
+        ret = TRY_LOCK (&clienttable->lock);
+        {
+                if (ret) {
+                        gf_log ("client_t", GF_LOG_WARNING,
+                                "Unable to acquire lock");
+                        return;
+                }
+                memset(key, 0, sizeof key);
+                gf_proc_dump_build_key(key, prefix, "maxclients");
+                gf_proc_dump_write(key, "%d", clienttable->max_clients);
+                gf_proc_dump_build_key(key, prefix, "first_free");
+                gf_proc_dump_write(key, "%d", clienttable->first_free);
+                for ( i = 0 ; i < clienttable->max_clients; i++) {
+                        if (GF_CLIENTENTRY_ALLOCATED ==
+                            clienttable->cliententries[i].next_free) {
+                                gf_proc_dump_build_key(key, prefix,
+                                                       "cliententry[%d]", i);
+                                gf_proc_dump_add_section(key);
+                                cliententry_dump(&clienttable->cliententries[i],
+                                                 key);
+                        }
+                }
+        }
+        UNLOCK(&clienttable->lock);
+}
+
+
+void
+client_ctx_dump (client_t *client, char *prefix)
+{
+#if 0 /* TBD, FIXME */
+        struct client_ctx   *client_ctx = NULL;
+        xlator_t            *xl = NULL;
+        int                  i = 0;
+
+        if ((client == NULL) || (client->ctx == NULL)) {
+                goto out;
+        }
+
+        LOCK (&client->ctx_lock);
+        if (client->ctx != NULL) {
+                client_ctx = GF_CALLOC (client->inode->table->xl->graph->ctx_count,
+                                        sizeof (*client_ctx),
+                                        gf_common_mt_client_ctx);
+                if (client_ctx == NULL) {
+                        goto unlock;
+                }
+
+                for (i = 0; i < client->inode->table->xl->graph->ctx_count; i++) {
+                        client_ctx[i] = client->ctx[i];
+                }
+        }
+unlock:
+        UNLOCK (&client->ctx_lock);
+
+        if (client_ctx == NULL) {
+                goto out;
+        }
+
+        for (i = 0; i < client->inode->table->xl->graph->ctx_count; i++) {
+                if (client_ctx[i].xl_key) {
+                        xl = (xlator_t *)(long)client_ctx[i].xl_key;
+                        if (xl->dumpops && xl->dumpops->clientctx)
+                                xl->dumpops->clientctx (xl, client);
+                }
+        }
+out:
+        GF_FREE (client_ctx);
+#endif
+}
+
+
+/*
+ * the following functions are here to preserve legacy behavior of the
+ * protocol/server xlator dump, but perhaps they should just be folded
+ * into the client dump instead?
+ */
+int
+gf_client_dump_fdtables_to_dict (xlator_t *this, dict_t *dict)
+{
+        client_t       *client      = NULL;
+        clienttable_t  *clienttable = NULL;
+        int             count       = 0;
+        int             ret         = -1;
+        char            key[GF_DUMP_MAX_BUF_LEN] = {0,};
+
+        GF_VALIDATE_OR_GOTO (THIS->name, this, out);
+        GF_VALIDATE_OR_GOTO (this->name, dict, out);
+
+        clienttable = this->ctx->clienttable;
+
+        if (!clienttable)
+                return -1;
+
+        ret = TRY_LOCK (&clienttable->lock);
+        {
+                if (ret) {
+                        gf_log ("client_t", GF_LOG_WARNING,
+                                "Unable to acquire lock");
+                        return -1;
+                }
+                for ( ; count < clienttable->max_clients; count++) {
+                        if (GF_CLIENTENTRY_ALLOCATED !=
+                            clienttable->cliententries[count].next_free)
+                                continue;
+                        client = clienttable->cliententries[count].client;
+                        memset(key, 0, sizeof key);
+                        snprintf (key, sizeof key, "conn%d", count++);
+                        fdtable_dump_to_dict (client->server_ctx.fdtable,
+                                              key, dict);
+                }
+        }
+        UNLOCK(&clienttable->lock);
+
+        ret = dict_set_int32 (dict, "conncount", count);
+out:
+        return ret;
+}
+
+int
+gf_client_dump_fdtables (xlator_t *this)
+{
+        client_t       *client = NULL;
+        clienttable_t  *clienttable = NULL;
+        int             count = 1;
+        int             ret = -1;
+        char            key[GF_DUMP_MAX_BUF_LEN] = {0,};
+
+        GF_VALIDATE_OR_GOTO (THIS->name, this, out);
+
+        clienttable = this->ctx->clienttable;
+
+        if (!clienttable)
+                return -1;
+
+        ret = TRY_LOCK (&clienttable->lock);
+        {
+                if (ret) {
+                        gf_log ("client_t", GF_LOG_WARNING,
+                                "Unable to acquire lock");
+                        return -1;
+                }
+
+
+                for ( ; count < clienttable->max_clients; count++) {
+                        if (GF_CLIENTENTRY_ALLOCATED !=
+                            clienttable->cliententries[count].next_free)
+                                continue;
+                        client = clienttable->cliententries[count].client;
+                        memset(key, 0, sizeof key);
+                        if (client->server_ctx.client_uid) {
+                                gf_proc_dump_build_key (key, "conn",
+                                                        "%d.id", count);
+                                gf_proc_dump_write (key, "%s",
+                                                    client->server_ctx.client_uid);
+                        }
+
+                        gf_proc_dump_build_key (key, "conn", "%d.ref",
+                                                        count);
+                        gf_proc_dump_write (key, "%d", client->ref.count);
+                        if (client->bound_xl) {
+                                gf_proc_dump_build_key (key, "conn",
+                                                        "%d.bound_xl", count);
+                                gf_proc_dump_write (key, "%s",
+                                                    client->bound_xl->name);
+                        }
+
+                        gf_proc_dump_build_key (key, "conn","%d.id", count);
+                        fdtable_dump (client->server_ctx.fdtable, key);
+                }
+        }
+
+        UNLOCK(&clienttable->lock);
+
+        ret = 0;
+out:
+        return ret;
+}
+
+
+int
+gf_client_dump_inodes_to_dict (xlator_t *this, dict_t *dict)
+{
+        client_t       *client        = NULL;
+        clienttable_t  *clienttable   = NULL;
+        xlator_t       *prev_bound_xl = NULL;
+        char            key[32]       = {0,};
+        int             count         = 0;
+        int             ret           = -1;
+
+        GF_VALIDATE_OR_GOTO (THIS->name, this, out);
+        GF_VALIDATE_OR_GOTO (this->name, dict, out);
+
+        clienttable = this->ctx->clienttable;
+
+        if (!clienttable)
+                return -1;
+
+        ret = TRY_LOCK (&clienttable->lock);
+        {
+                if (ret) {
+                        gf_log ("client_t", GF_LOG_WARNING,
+                                "Unable to acquire lock");
+                        return -1;
+                }
+                for ( ; count < clienttable->max_clients; count++) {
+                        if (GF_CLIENTENTRY_ALLOCATED !=
+                            clienttable->cliententries[count].next_free)
+                                continue;
+                        client = clienttable->cliententries[count].client;
+                        memset(key, 0, sizeof key);
+                        if (client->bound_xl && client->bound_xl->itable) {
+                                /* Presently every brick contains only
+                                 * one bound_xl for all connections.
+                                 * This will lead to duplicating of
+                                 *  the inode lists, if listing is
+                                 * done for every connection. This
+                                 * simple check prevents duplication
+                                 * in the present case. If need arises
+                                 * the check can be improved.
+                                 */
+                                if (client->bound_xl == prev_bound_xl)
+                                        continue;
+                                prev_bound_xl = client->bound_xl;
+
+                                memset (key, 0, sizeof (key));
+                                snprintf (key, sizeof (key), "conn%d", count);
+                                inode_table_dump_to_dict (client->bound_xl->itable,
+                                                          key, dict);
+                        }
+                }
+        }
+        UNLOCK(&clienttable->lock);
+
+        ret = dict_set_int32 (dict, "conncount", count);
+
+out:
+        if (prev_bound_xl)
+                prev_bound_xl = NULL;
+        return ret;
+}
+
+int
+gf_client_dump_inodes (xlator_t *this)
+{
+        client_t       *client        = NULL;
+        clienttable_t  *clienttable   = NULL;
+        xlator_t       *prev_bound_xl = NULL;
+        int             count         = 1;
+        int             ret           = -1;
+        char            key[GF_DUMP_MAX_BUF_LEN] = {0,};
+
+        GF_VALIDATE_OR_GOTO (THIS->name, this, out);
+
+        clienttable = this->ctx->clienttable;
+
+        if (!clienttable)
+                return -1;
+
+        ret = TRY_LOCK (&clienttable->lock);
+        {
+                if (ret) {
+                        gf_log ("client_t", GF_LOG_WARNING,
+                                "Unable to acquire lock");
+                        return -1;
+                }
+
+                for ( ; count < clienttable->max_clients; count++) {
+                        if (GF_CLIENTENTRY_ALLOCATED !=
+                            clienttable->cliententries[count].next_free)
+                                continue;
+                        client = clienttable->cliententries[count].client;
+                        memset(key, 0, sizeof key);
+                        if (client->bound_xl && client->bound_xl->itable) {
+                                /* Presently every brick contains only
+                                 * one bound_xl for all connections.
+                                 * This will lead to duplicating of
+                                 * the inode lists, if listing is
+                                 * done for every connection. This
+                                 * simple check prevents duplication
+                                 * in the present case. If need arises
+                                 * the check can be improved.
+                                 */
+                                if (client->bound_xl == prev_bound_xl)
+                                        continue;
+                                prev_bound_xl = client->bound_xl;
+
+                                gf_proc_dump_build_key(key, "conn",
+                                                       "%d.bound_xl.%s", count,
+                                                       client->bound_xl->name);
+                                inode_table_dump(client->bound_xl->itable,key);
+                        }
+                }
+        }
+        UNLOCK(&clienttable->lock);
+
+        ret = 0;
+out:
+        return ret;
+}
diff --git a/libglusterfs/src/client_t.h b/libglusterfs/src/client_t.h
new file mode 100644
index 00000000000..7b3fcf0dd3a
--- /dev/null
+++ b/libglusterfs/src/client_t.h
@@ -0,0 +1,168 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CLIENT_T_H
+#define _CLIENT_T_H
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "locking.h"  /* for gf_lock_t, not included by glusterfs.h */
+
+struct client_ctx {
+        union {
+                uint64_t  key;
+                void     *xl_key;
+        };
+        union {
+                uint64_t  value;
+                void     *ptr1;
+        };
+};
+
+struct _client_t {
+        struct {
+                /* ctx for .../xlators/protocol/server */
+                gf_lock_t            fdtable_lock;
+                fdtable_t           *fdtable;
+                char                *client_uid;
+                struct _gf_timer    *grace_timer;
+                uint32_t             lk_version;
+                struct {
+                        int          flavour;
+                        size_t       len;
+                        char        *data;
+                }            auth;
+        }            server_ctx;
+        struct {
+                /* ctx for .../xlators/features/locks */
+                gf_lock_t            ltable_lock;
+                struct _lock_table  *ltable;
+        }            locks_ctx;
+        struct {
+                /* e.g. hekafs uidmap can stash stuff here */
+                gf_lock_t            lock;
+                unsigned short       count;
+                struct client_ctx   *ctx;
+        }            scratch_ctx;
+        struct {
+                gf_lock_t            lock;
+                volatile int         bind;
+                volatile int         count;
+        }            ref;
+        xlator_t    *bound_xl;
+        xlator_t    *this;
+        int          tbl_index;
+};
+typedef struct _client_t client_t;
+
+
+struct client_table_entry {
+        client_t            *client;
+        int                  next_free;
+};
+typedef struct client_table_entry cliententry_t;
+
+
+struct _clienttable {
+        unsigned int         max_clients;
+        gf_lock_t            lock;
+        cliententry_t       *cliententries;
+        int                  first_free;
+};
+typedef struct _clienttable clienttable_t;
+
+#define GF_CLIENTTABLE_INITIAL_SIZE 32
+
+/* Signifies no more entries in the client table. */
+#define GF_CLIENTTABLE_END  -1
+
+/* This is used to invalidate
+ * the next_free value in an cliententry that has been allocated
+ */
+#define GF_CLIENTENTRY_ALLOCATED    -2
+
+
+
+client_t *
+gf_client_get (xlator_t *this, rpcsvc_auth_data_t *cred, char *client_uid);
+
+void
+gf_client_put (client_t *client, gf_boolean_t *detached);
+
+clienttable_t *
+gf_clienttable_alloc (void);
+
+
+void
+gf_client_clienttable_destroy (clienttable_t *clienttable);
+
+
+client_t *
+gf_client_ref (client_t *client);
+
+
+void
+gf_client_unref (client_t *client);
+
+int
+gf_client_dump_fdtable_to_dict (xlator_t *this, dict_t *dict);
+
+int
+gf_client_dump_fdtable (xlator_t *this);
+
+int
+gf_client_dump_inodes_to_dict (xlator_t *this, dict_t *dict);
+
+int
+gf_client_dump_inodes (xlator_t *this);
+
+int
+client_ctx_set (client_t *client, xlator_t *xlator, uint64_t value);
+
+
+int
+client_ctx_get (client_t *client, xlator_t *xlator, uint64_t *value);
+
+
+int
+client_ctx_del (client_t *client, xlator_t *xlator, uint64_t *value);
+
+
+int
+_client_ctx_set (client_t *client, xlator_t *xlator, uint64_t value);
+
+
+int
+_client_ctx_get (client_t *client, xlator_t *xlator, uint64_t *value);
+
+
+int
+_client_ctx_del (client_t *client, xlator_t *xlator, uint64_t *value);
+
+void
+client_ctx_dump (client_t *client, char *prefix);
+
+int
+gf_client_dump_fdtables_to_dict (xlator_t *this, dict_t *dict);
+
+int
+gf_client_dump_fdtables (xlator_t *this);
+
+int
+gf_client_dump_inodes_to_dict (xlator_t *this, dict_t *dict);
+
+int
+gf_client_dump_inodes (xlator_t *this);
+
+#endif /* _CLIENT_T_H */
diff --git a/libglusterfs/src/glusterfs.h b/libglusterfs/src/glusterfs.h
index b3ff1fd38de..763968c9eda 100644
--- a/libglusterfs/src/glusterfs.h
+++ b/libglusterfs/src/glusterfs.h
@@ -140,7 +140,7 @@
 #define GLUSTERFS_RPC_REPLY_SIZE               24
 
 #define ZR_FILE_CONTENT_REQUEST(key) (!strncmp(key, ZR_FILE_CONTENT_STR, \
-					       ZR_FILE_CONTENT_STRLEN))
+                                               ZR_FILE_CONTENT_STRLEN))
 
 #define DEFAULT_VAR_RUN_DIRECTORY    DATADIR "/run/gluster"
 #define GF_REPLICATE_TRASH_DIR          ".landfill"
@@ -192,8 +192,8 @@ typedef enum {
         GF_FOP_READDIR,
         GF_FOP_INODELK,
         GF_FOP_FINODELK,
-	GF_FOP_ENTRYLK,
-	GF_FOP_FENTRYLK,
+        GF_FOP_ENTRYLK,
+        GF_FOP_FENTRYLK,
         GF_FOP_XATTROP,
         GF_FOP_FXATTROP,
         GF_FOP_FGETXATTR,
@@ -258,20 +258,20 @@ typedef enum {
 
 
 typedef enum {
-	ENTRYLK_LOCK,
-	ENTRYLK_UNLOCK,
-	ENTRYLK_LOCK_NB
+        ENTRYLK_LOCK,
+        ENTRYLK_UNLOCK,
+        ENTRYLK_LOCK_NB
 } entrylk_cmd;
 
 
 typedef enum {
-	ENTRYLK_RDLCK,
-	ENTRYLK_WRLCK
+        ENTRYLK_RDLCK,
+        ENTRYLK_WRLCK
 } entrylk_type;
 
 
 typedef enum {
-	GF_XATTROP_ADD_ARRAY,
+        GF_XATTROP_ADD_ARRAY,
         GF_XATTROP_ADD_ARRAY64,
         GF_XATTROP_OR_ARRAY,
         GF_XATTROP_AND_ARRAY
@@ -287,10 +287,10 @@ typedef enum {
 #define GF_CONTENT_KEY "glusterfs.content"
 
 struct _xlator_cmdline_option {
-	struct list_head    cmd_args;
-	char               *volume;
-	char               *key;
-	char               *value;
+        struct list_head    cmd_args;
+        char               *volume;
+        char               *key;
+        char               *value;
 };
 typedef struct _xlator_cmdline_option xlator_cmdline_option_t;
 
@@ -300,22 +300,22 @@ typedef struct _xlator_cmdline_option xlator_cmdline_option_t;
 #define GF_OPTION_DEFERRED 2
 
 struct _cmd_args {
-	/* basic options */
-	char            *volfile_server;
-	char            *volfile;
+        /* basic options */
+        char            *volfile_server;
+        char            *volfile;
         char            *log_server;
-	gf_loglevel_t    log_level;
-	char            *log_file;
+        gf_loglevel_t    log_level;
+        char            *log_file;
         int32_t          max_connect_attempts;
-	/* advanced options */
-	uint32_t         volfile_server_port;
-	char            *volfile_server_transport;
+        /* advanced options */
+        uint32_t         volfile_server_port;
+        char            *volfile_server_transport;
         uint32_t         log_server_port;
-	char            *pid_file;
+        char            *pid_file;
         char            *sock_file;
-	int              no_daemon_mode;
-	char            *run_id;
-	int              debug_mode;
+        int              no_daemon_mode;
+        char            *run_id;
+        int              debug_mode;
         int              read_only;
         int              acl;
         int              selinux;
@@ -331,13 +331,13 @@ struct _cmd_args {
 	int              fuse_direct_io_mode;
 	char             *use_readdirp;
         int              volfile_check;
-	double           fuse_entry_timeout;
-	double           fuse_negative_timeout;
-	double           fuse_attribute_timeout;
-	char            *volume_name;
-	int              fuse_nodev;
-	int              fuse_nosuid;
-	char            *dump_fuse;
+        double           fuse_entry_timeout;
+        double           fuse_negative_timeout;
+        double           fuse_attribute_timeout;
+        char            *volume_name;
+        int              fuse_nodev;
+        int              fuse_nosuid;
+        char            *dump_fuse;
         pid_t            client_pid;
         int              client_pid_set;
         unsigned         uid_map_root;
@@ -345,9 +345,9 @@ struct _cmd_args {
         int              congestion_threshold;
         char            *fuse_mountopts;
 
-	/* key args */
-	char            *mount_point;
-	char            *volfile_id;
+        /* key args */
+        char            *mount_point;
+        char            *volfile_id;
 
         /* required for portmap */
         int             brick_port;
@@ -376,16 +376,16 @@ typedef struct _glusterfs_graph glusterfs_graph_t;
 typedef int32_t (*glusterfsd_mgmt_event_notify_fn_t) (int32_t event, void *data,
                                                       ...);
 struct _glusterfs_ctx {
-	cmd_args_t          cmd_args;
-	char               *process_uuid;
-	FILE               *pidfp;
-	char                fin;
-	void               *timer;
-	void               *ib;
-	void               *pool;
-	void               *event_pool;
+        cmd_args_t          cmd_args;
+        char               *process_uuid;
+        FILE               *pidfp;
+        char                fin;
+        void               *timer;
+        void               *ib;
+        void               *pool;
+        void               *event_pool;
         void               *iobuf_pool;
-	pthread_mutex_t     lock;
+        pthread_mutex_t     lock;
         size_t              page_size;
         struct list_head    graphs; /* double linked list of graphs - one per volfile parse */
         glusterfs_graph_t  *active; /* the latest graph in use */
@@ -401,12 +401,12 @@ struct _glusterfs_ctx {
                                          got changed */
         pid_t               mnt_pid; /* pid of the mount agent */
         int                 process_mode; /*mode in which process is runninng*/
-	struct syncenv      *env;         /* The env pointer to the synctasks */
+        struct syncenv     *env;          /* The env pointer to the synctasks */
 
         struct list_head    mempool_list; /* used to keep a global list of
                                              mempools, used to log details of
                                              mempool in statedump */
-        char                *statedump_path;
+        char               *statedump_path;
 
         struct mem_pool    *dict_pool;
         struct mem_pool    *dict_pair_pool;
@@ -416,9 +416,11 @@ struct _glusterfs_ctx {
                                                      call to fsd-mgmt */
         gf_log_handle_t     log; /* all logging related variables */
 
-        int           mem_acct_enable;
+        int                 mem_acct_enable;
 
         int                 daemon_pipe[2];
+
+        struct _clienttable *clienttable;
 };
 typedef struct _glusterfs_ctx glusterfs_ctx_t;
 
diff --git a/libglusterfs/src/lock-table.c b/libglusterfs/src/lock-table.c
new file mode 100644
index 00000000000..a2fff2e3333
--- /dev/null
+++ b/libglusterfs/src/lock-table.c
@@ -0,0 +1,128 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#include "lock-table.h"
+#include "common-utils.h"
+
+
+struct _lock_table *
+gf_lock_table_new (void)
+{
+        struct _lock_table *new = NULL;
+
+        new = GF_CALLOC (1, sizeof (struct _lock_table), gf_common_mt_lock_table);
+        if (new == NULL) {
+                goto out;
+        }
+        INIT_LIST_HEAD (&new->entrylk_lockers);
+        INIT_LIST_HEAD (&new->inodelk_lockers);
+        LOCK_INIT (&new->lock);
+out:
+        return new;
+}
+
+
+int
+gf_add_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc, fd_t *fd, pid_t pid, gf_lkowner_t *owner,
+               glusterfs_fop_t type)
+{
+        int32_t         ret = -1;
+        struct _locker *new = NULL;
+
+        GF_VALIDATE_OR_GOTO ("lock-table", table, out);
+        GF_VALIDATE_OR_GOTO ("lock-table", volume, out);
+
+        new = GF_CALLOC (1, sizeof (struct _locker), gf_common_mt_locker);
+        if (new == NULL) {
+                goto out;
+        }
+        INIT_LIST_HEAD (&new->lockers);
+
+        new->volume = gf_strdup (volume);
+
+        if (fd == NULL) {
+                loc_copy (&new->loc, loc);
+        } else {
+                new->fd = fd_ref (fd);
+        }
+
+        new->pid   = pid;
+        new->owner = *owner;
+
+        LOCK (&table->lock);
+        {
+                if (type == GF_FOP_ENTRYLK)
+                        list_add_tail (&new->lockers, &table->entrylk_lockers);
+                else
+                        list_add_tail (&new->lockers, &table->inodelk_lockers);
+        }
+        UNLOCK (&table->lock);
+out:
+        return ret;
+}
+
+int
+gf_del_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc, fd_t *fd, gf_lkowner_t *owner, glusterfs_fop_t type)
+{
+        struct _locker    *locker = NULL;
+        struct _locker    *tmp = NULL;
+        int32_t            ret = -1;
+        struct list_head  *head = NULL;
+        struct list_head   del;
+
+        GF_VALIDATE_OR_GOTO ("lock-table", table, out);
+        GF_VALIDATE_OR_GOTO ("lock-table", volume, out);
+
+        INIT_LIST_HEAD (&del);
+
+        LOCK (&table->lock);
+        {
+                if (type == GF_FOP_ENTRYLK) {
+                        head = &table->entrylk_lockers;
+                } else {
+                        head = &table->inodelk_lockers;
+                }
+
+                list_for_each_entry_safe (locker, tmp, head, lockers) {
+                        if (!is_same_lkowner (&locker->owner, owner) ||
+                            strcmp (locker->volume, volume))
+                                continue;
+
+                        if (locker->fd && fd && (locker->fd == fd))
+                                list_move_tail (&locker->lockers, &del);
+                        else if (locker->loc.inode && loc &&
+                                 (locker->loc.inode == loc->inode))
+                                list_move_tail (&locker->lockers, &del);
+                }
+        }
+        UNLOCK (&table->lock);
+
+        tmp = NULL;
+        locker = NULL;
+
+        list_for_each_entry_safe (locker, tmp, &del, lockers) {
+                list_del_init (&locker->lockers);
+                if (locker->fd)
+                        fd_unref (locker->fd);
+                else
+                        loc_wipe (&locker->loc);
+
+                GF_FREE (locker->volume);
+                GF_FREE (locker);
+        }
+
+        ret = 0;
+out:
+        return ret;
+
+}
+
diff --git a/libglusterfs/src/lock-table.h b/libglusterfs/src/lock-table.h
new file mode 100644
index 00000000000..4a9083873f3
--- /dev/null
+++ b/libglusterfs/src/lock-table.h
@@ -0,0 +1,54 @@
+/*
+  Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com>
+  This file is part of GlusterFS.
+
+  This file is licensed to you under your choice of the GNU Lesser
+  General Public License, version 3 or any later version (LGPLv3 or
+  later), or the GNU General Public License, version 2 (GPLv2), in all
+  cases as published by the Free Software Foundation.
+*/
+
+#ifndef _LOCK_TABLE_H
+#define _LOCK_TABLE_H
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "xlator.h"
+
+struct _locker {
+        struct list_head  lockers;
+        char             *volume;
+        loc_t             loc;
+        fd_t             *fd;
+        gf_lkowner_t      owner;
+        pid_t             pid;
+};
+
+struct _lock_table {
+        struct list_head  inodelk_lockers;
+        struct list_head  entrylk_lockers;
+        gf_lock_t         lock;
+};
+
+int32_t
+gf_add_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc,
+               fd_t *fd,
+               pid_t pid,
+               gf_lkowner_t *owner,
+               glusterfs_fop_t type);
+
+int32_t
+gf_del_locker (struct _lock_table *table, const char *volume,
+               loc_t *loc,
+               fd_t *fd,
+               gf_lkowner_t *owner,
+               glusterfs_fop_t type);
+
+struct _lock_table *
+gf_lock_table_new (void);
+
+#endif /* _LOCK_TABLE_H */
diff --git a/libglusterfs/src/mem-types.h b/libglusterfs/src/mem-types.h
index fea54c35ebc..7dcbfb3dc80 100644
--- a/libglusterfs/src/mem-types.h
+++ b/libglusterfs/src/mem-types.h
@@ -109,6 +109,12 @@ enum gf_common_mem_types_ {
         gf_common_mt_drc_rbtree_node_t    = 93,
         gf_common_mt_iov_base_t           = 94,
         gf_common_mt_groups_t             = 95,
-        gf_common_mt_end                  = 96
+        gf_common_mt_cliententry_t        = 96,
+        gf_common_mt_clienttable_t        = 97,
+        gf_common_mt_client_t             = 98,
+        gf_common_mt_client_ctx           = 99,
+        gf_common_mt_lock_table           = 100,
+        gf_common_mt_locker               = 101,
+        gf_common_mt_end                  = 102 
 };
 #endif