From 04536e53082cdff5cd10a804502347ee83490c81 Mon Sep 17 00:00:00 2001 From: "Kaleb S. KEITHLEY" Date: Tue, 18 Sep 2012 14:07:40 -0400 Subject: 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 Reviewed-on: http://review.gluster.org/3957 Tested-by: Gluster Build System Reviewed-by: Amar Tumballi --- libglusterfs/src/client_t.c | 881 ++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 881 insertions(+) create mode 100644 libglusterfs/src/client_t.c (limited to 'libglusterfs/src/client_t.c') 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. + 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; +} -- cgit