diff options
Diffstat (limited to 'rpc')
32 files changed, 4190 insertions, 2631 deletions
diff --git a/rpc/rpc-lib/src/Makefile.am b/rpc/rpc-lib/src/Makefile.am index ca62a27f9..f19c3c8a4 100644 --- a/rpc/rpc-lib/src/Makefile.am +++ b/rpc/rpc-lib/src/Makefile.am @@ -1,16 +1,18 @@ lib_LTLIBRARIES = libgfrpc.la libgfrpc_la_SOURCES = auth-unix.c rpcsvc-auth.c rpcsvc.c auth-null.c \ - rpc-transport.c xdr-rpc.c xdr-rpcclnt.c rpc-clnt.c auth-glusterfs.c + rpc-transport.c xdr-rpc.c xdr-rpcclnt.c rpc-clnt.c auth-glusterfs.c \ + rpc-drc.c libgfrpc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la noinst_HEADERS = rpcsvc.h rpc-transport.h xdr-common.h xdr-rpc.h xdr-rpcclnt.h \ - rpc-clnt.h rpcsvc-common.h protocol-common.h + rpc-clnt.h rpcsvc-common.h protocol-common.h rpc-drc.h AM_CPPFLAGS = $(GF_CPPFLAGS) -I$(top_srcdir)/libglusterfs/src \ -I$(top_srcdir)/rpc/xdr/src \ - -DRPC_TRANSPORTDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/rpc-transport\" + -DRPC_TRANSPORTDIR=\"$(libdir)/glusterfs/$(PACKAGE_VERSION)/rpc-transport\" \ + -I$(top_srcdir)/contrib/rbtree AM_CFLAGS = -Wall $(GF_CFLAGS) diff --git a/rpc/rpc-lib/src/auth-glusterfs.c b/rpc/rpc-lib/src/auth-glusterfs.c index 9c6f8385b..db488434c 100644 --- a/rpc/rpc-lib/src/auth-glusterfs.c +++ b/rpc/rpc-lib/src/auth-glusterfs.c @@ -96,6 +96,22 @@ int auth_glusterfs_authenticate (rpcsvc_request_t *req, void *priv) goto err; } + if (req->auxgidcount > SMALL_GROUP_COUNT) { + req->auxgidlarge = GF_CALLOC(req->auxgidcount, + sizeof(req->auxgids[0]), + gf_common_mt_auxgids); + req->auxgids = req->auxgidlarge; + } else { + req->auxgids = req->auxgidsmall; + } + + if (!req->auxgids) { + gf_log ("auth-glusterfs", GF_LOG_WARNING, + "cannot allocate gid list"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + for (gidcount = 0; gidcount < au.ngrps; ++gidcount) req->auxgids[gidcount] = au.groups[gidcount]; @@ -203,6 +219,22 @@ int auth_glusterfs_v2_authenticate (rpcsvc_request_t *req, void *priv) goto err; } + if (req->auxgidcount > SMALL_GROUP_COUNT) { + req->auxgidlarge = GF_CALLOC(req->auxgidcount, + sizeof(req->auxgids[0]), + gf_common_mt_auxgids); + req->auxgids = req->auxgidlarge; + } else { + req->auxgids = req->auxgidsmall; + } + + if (!req->auxgids) { + gf_log ("auth-glusterfs-v2", GF_LOG_WARNING, + "cannot allocate gid list"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + for (i = 0; i < req->auxgidcount; ++i) req->auxgids[i] = au.groups.groups_val[i]; diff --git a/rpc/rpc-lib/src/auth-unix.c b/rpc/rpc-lib/src/auth-unix.c index 6251d60a8..fa5f0576e 100644 --- a/rpc/rpc-lib/src/auth-unix.c +++ b/rpc/rpc-lib/src/auth-unix.c @@ -42,6 +42,7 @@ int auth_unix_authenticate (rpcsvc_request_t *req, void *priv) if (!req) return ret; + req->auxgids = req->auxgidsmall; ret = xdr_to_auth_unix_cred (req->cred.authdata, req->cred.datalen, &aup, machname, req->auxgids); if (ret == -1) { diff --git a/rpc/rpc-lib/src/protocol-common.h b/rpc/rpc-lib/src/protocol-common.h index 97017e5fe..8bef906cc 100644 --- a/rpc/rpc-lib/src/protocol-common.h +++ b/rpc/rpc-lib/src/protocol-common.h @@ -56,6 +56,9 @@ enum gf_fop_procnum { GFS3_OP_RELEASE, GFS3_OP_RELEASEDIR, GFS3_OP_FREMOVEXATTR, + GFS3_OP_FALLOCATE, + GFS3_OP_DISCARD, + GFS3_OP_ZEROFILL, GFS3_OP_MAXVALUE, } ; @@ -154,7 +157,10 @@ enum gluster_cli_procnum { GLUSTER_CLI_LIST_VOLUME, GLUSTER_CLI_CLRLOCKS_VOLUME, GLUSTER_CLI_UUID_RESET, - GLUSTER_CLI_BD_OP, + GLUSTER_CLI_UUID_GET, + GLUSTER_CLI_COPY_FILE, + GLUSTER_CLI_SYS_EXEC, + GLUSTER_CLI_SNAP, GLUSTER_CLI_MAXVALUE, }; @@ -186,7 +192,7 @@ enum glusterd_brick_procnum { GLUSTERD_BRICK_XLATOR_DEFRAG, GLUSTERD_NODE_PROFILE, GLUSTERD_NODE_STATUS, - GLUSTERD_BRICK_BD_OP, + GLUSTERD_VOLUME_BARRIER_OP, GLUSTERD_BRICK_MAXVALUE, }; @@ -204,16 +210,22 @@ typedef enum { GF_AFR_OP_INDEX_SUMMARY, GF_AFR_OP_HEALED_FILES, GF_AFR_OP_HEAL_FAILED_FILES, - GF_AFR_OP_SPLIT_BRAIN_FILES + GF_AFR_OP_SPLIT_BRAIN_FILES, + GF_AFR_OP_STATISTICS, + GF_AFR_OP_STATISTICS_HEAL_COUNT, + GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA, } gf_xl_afr_op_t ; -typedef enum { - GF_BD_OP_INVALID, - GF_BD_OP_NEW_BD, - GF_BD_OP_DELETE_BD, - GF_BD_OP_CLONE_BD, - GF_BD_OP_SNAPSHOT_BD, -} gf_xl_bd_op_t ; +enum glusterd_mgmt_v3_procnum { + GLUSTERD_MGMT_V3_NULL, /* 0 */ + GLUSTERD_MGMT_V3_LOCK, + GLUSTERD_MGMT_V3_PRE_VALIDATE, + GLUSTERD_MGMT_V3_BRICK_OP, + GLUSTERD_MGMT_V3_COMMIT, + GLUSTERD_MGMT_V3_POST_VALIDATE, + GLUSTERD_MGMT_V3_UNLOCK, + GLUSTERD_MGMT_V3_MAXVALUE, +}; #define GLUSTER_HNDSK_PROGRAM 14398633 /* Completely random */ #define GLUSTER_HNDSK_VERSION 2 /* 0.0.2 */ @@ -241,6 +253,10 @@ typedef enum { #define GD_BRICK_PROGRAM 4867634 /*Completely random*/ #define GD_BRICK_VERSION 2 +/* Third version */ +#define GD_MGMT_V3_PROGRAM 2210013 /* Completely random */ +#define GD_MGMT_V3_VERSION 3 + /* OP-VERSION handshake */ #define GD_MGMT_HNDSK_PROGRAM 1239873 /* Completely random */ #define GD_MGMT_HNDSK_VERSION 1 diff --git a/rpc/rpc-lib/src/rpc-clnt.c b/rpc/rpc-lib/src/rpc-clnt.c index e6c681df8..ac98a5c91 100644 --- a/rpc/rpc-lib/src/rpc-clnt.c +++ b/rpc/rpc-lib/src/rpc-clnt.c @@ -145,7 +145,7 @@ call_bail (void *data) struct saved_frame *trav = NULL; struct saved_frame *tmp = NULL; char frame_sent[256] = {0,}; - struct timeval timeout = {0,}; + struct timespec timeout = {0,}; struct iovec iov = {0,}; GF_VALIDATE_OR_GOTO ("client", data, out); @@ -163,7 +163,7 @@ call_bail (void *data) call-once timer */ if (conn->timer) { timeout.tv_sec = 10; - timeout.tv_usec = 0; + timeout.tv_nsec = 0; gf_timer_call_cancel (clnt->ctx, conn->timer); conn->timer = gf_timer_call_after (clnt->ctx, @@ -173,7 +173,8 @@ call_bail (void *data) if (conn->timer == NULL) { gf_log (conn->trans->name, GF_LOG_WARNING, - "Cannot create bailout timer"); + "Cannot create bailout timer for %s", + conn->trans->peerinfo.identifier); } } @@ -197,14 +198,14 @@ call_bail (void *data) ".%"GF_PRI_SUSECONDS, trav->saved_at.tv_usec); gf_log (conn->trans->name, GF_LOG_ERROR, - "bailing out frame type(%s) op(%s(%d)) xid = 0x%ux " - "sent = %s. timeout = %d", + "bailing out frame type(%s) op(%s(%d)) xid = 0x%x " + "sent = %s. timeout = %d for %s", trav->rpcreq->prog->progname, (trav->rpcreq->prog->procnames) ? trav->rpcreq->prog->procnames[trav->rpcreq->procnum] : "--", trav->rpcreq->procnum, trav->rpcreq->xid, frame_sent, - conn->frame_timeout); + conn->frame_timeout, conn->trans->peerinfo.identifier); clnt = rpc_clnt_ref (clnt); trav->rpcreq->rpc_status = -1; @@ -226,7 +227,7 @@ __save_frame (struct rpc_clnt *rpc_clnt, call_frame_t *frame, struct rpc_req *rpcreq) { rpc_clnt_connection_t *conn = NULL; - struct timeval timeout = {0, }; + struct timespec timeout = {0, }; struct saved_frame *saved_frame = NULL; conn = &rpc_clnt->conn; @@ -240,7 +241,7 @@ __save_frame (struct rpc_clnt *rpc_clnt, call_frame_t *frame, /* TODO: make timeout configurable */ if (conn->timer == NULL) { timeout.tv_sec = 10; - timeout.tv_usec = 0; + timeout.tv_nsec = 0; conn->timer = gf_timer_call_after (rpc_clnt->ctx, timeout, call_bail, @@ -359,7 +360,7 @@ saved_frames_unwind (struct saved_frames *saved_frames) gf_log_callingfn (trav->rpcreq->conn->trans->name, GF_LOG_ERROR, "forced unwinding frame type(%s) op(%s(%d)) " - "called at %s (xid=0x%ux)", + "called at %s (xid=0x%x)", trav->rpcreq->prog->progname, ((trav->rpcreq->prog->procnames) ? trav->rpcreq->prog->procnames[trav->rpcreq->procnum] @@ -397,7 +398,7 @@ rpc_clnt_reconnect (void *trans_ptr) { rpc_transport_t *trans = NULL; rpc_clnt_connection_t *conn = NULL; - struct timeval tv = {0, 0}; + struct timespec ts = {0, 0}; int32_t ret = 0; struct rpc_clnt *clnt = NULL; @@ -416,14 +417,15 @@ rpc_clnt_reconnect (void *trans_ptr) conn->reconnect = 0; if (conn->connected == 0) { - tv.tv_sec = 3; + ts.tv_sec = 3; + ts.tv_nsec = 0; gf_log (trans->name, GF_LOG_TRACE, "attempting reconnect"); ret = rpc_transport_connect (trans, conn->config.remote_port); conn->reconnect = - gf_timer_call_after (clnt->ctx, tv, + gf_timer_call_after (clnt->ctx, ts, rpc_clnt_reconnect, trans); } else { @@ -661,7 +663,7 @@ rpc_clnt_reply_init (rpc_clnt_connection_t *conn, rpc_transport_pollin_t *msg, } gf_log (conn->trans->name, GF_LOG_TRACE, - "received rpc message (RPC XID: 0x%ux" + "received rpc message (RPC XID: 0x%x" " Program: %s, ProgVers: %d, Proc: %d) from rpc-transport (%s)", saved_frame->rpcreq->xid, saved_frame->rpcreq->prog->progname, @@ -819,6 +821,9 @@ out: return; } +static void +rpc_clnt_destroy (struct rpc_clnt *rpc); + int rpc_clnt_notify (rpc_transport_t *trans, void *mydata, rpc_transport_event_t event, void *data, ...) @@ -828,7 +833,7 @@ rpc_clnt_notify (rpc_transport_t *trans, void *mydata, int ret = -1; rpc_request_info_t *req_info = NULL; rpc_transport_pollin_t *pollin = NULL; - struct timeval tv = {0, }; + struct timespec ts = {0, }; conn = mydata; if (conn == NULL) { @@ -847,10 +852,11 @@ rpc_clnt_notify (rpc_transport_t *trans, void *mydata, { if (!conn->rpc_clnt->disabled && (conn->reconnect == NULL)) { - tv.tv_sec = 10; + ts.tv_sec = 10; + ts.tv_nsec = 0; conn->reconnect = - gf_timer_call_after (clnt->ctx, tv, + gf_timer_call_after (clnt->ctx, ts, rpc_clnt_reconnect, conn->trans); } @@ -864,9 +870,7 @@ rpc_clnt_notify (rpc_transport_t *trans, void *mydata, } case RPC_TRANSPORT_CLEANUP: - /* this event should not be received on a client for, a - * transport is only disconnected, but never destroyed. - */ + rpc_clnt_destroy (clnt); ret = 0; break; @@ -1481,7 +1485,7 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, if (ret == -1) { gf_log (conn->trans->name, GF_LOG_WARNING, "failed to submit rpc-request " - "(XID: 0x%ux Program: %s, ProgVers: %d, " + "(XID: 0x%x Program: %s, ProgVers: %d, " "Proc: %d) to rpc-transport (%s)", rpcreq->xid, rpcreq->prog->progname, rpcreq->prog->progver, rpcreq->procnum, rpc->conn.trans->name); @@ -1492,7 +1496,7 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, __save_frame (rpc, frame, rpcreq); gf_log ("rpc-clnt", GF_LOG_TRACE, "submitted request " - "(XID: 0x%ux Program: %s, ProgVers: %d, " + "(XID: 0x%x Program: %s, ProgVers: %d, " "Proc: %d) to rpc-transport (%s)", rpcreq->xid, rpcreq->prog->progname, rpcreq->prog->progver, rpcreq->procnum, rpc->conn.trans->name); @@ -1541,18 +1545,21 @@ rpc_clnt_ref (struct rpc_clnt *rpc) static void -rpc_clnt_destroy (struct rpc_clnt *rpc) +rpc_clnt_trigger_destroy (struct rpc_clnt *rpc) { if (!rpc) return; - if (rpc->conn.trans) { - rpc_transport_unregister_notify (rpc->conn.trans); - rpc_transport_disconnect (rpc->conn.trans); - rpc_transport_unref (rpc->conn.trans); - } + rpc_clnt_disable (rpc); + rpc_transport_unref (rpc->conn.trans); +} + +static void +rpc_clnt_destroy (struct rpc_clnt *rpc) +{ + if (!rpc) + return; - rpc_clnt_reconnect_cleanup (&rpc->conn); saved_frames_destroy (rpc->conn.saved_frames); pthread_mutex_destroy (&rpc->lock); pthread_mutex_destroy (&rpc->conn.lock); @@ -1579,13 +1586,36 @@ rpc_clnt_unref (struct rpc_clnt *rpc) } pthread_mutex_unlock (&rpc->lock); if (!count) { - rpc_clnt_destroy (rpc); + rpc_clnt_trigger_destroy (rpc); return NULL; } return rpc; } +char +rpc_clnt_is_disabled (struct rpc_clnt *rpc) +{ + + rpc_clnt_connection_t *conn = NULL; + char disabled = 0; + + if (!rpc) { + goto out; + } + + conn = &rpc->conn; + + pthread_mutex_lock (&conn->lock); + { + disabled = rpc->disabled; + } + pthread_mutex_unlock (&conn->lock); + +out: + return disabled; +} + void rpc_clnt_disable (struct rpc_clnt *rpc) { @@ -1668,60 +1698,3 @@ rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config) rpc->conn.config.remote_host = gf_strdup (config->remote_host); } } - -int -rpc_clnt_transport_unix_options_build (dict_t **options, char *filepath, - int frame_timeout) -{ - dict_t *dict = NULL; - char *fpath = NULL; - int ret = -1; - - GF_ASSERT (filepath); - GF_ASSERT (options); - - dict = dict_new (); - if (!dict) - goto out; - - fpath = gf_strdup (filepath); - if (!fpath) { - ret = -1; - goto out; - } - - ret = dict_set_dynstr (dict, "transport.socket.connect-path", fpath); - if (ret) - goto out; - - ret = dict_set_str (dict, "transport.address-family", "unix"); - if (ret) - goto out; - - ret = dict_set_str (dict, "transport.socket.nodelay", "off"); - if (ret) - goto out; - - ret = dict_set_str (dict, "transport-type", "socket"); - if (ret) - goto out; - - ret = dict_set_str (dict, "transport.socket.keepalive", "off"); - if (ret) - goto out; - - if (frame_timeout > 0) { - ret = dict_set_int32 (dict, "frame-timeout", frame_timeout); - if (ret) - goto out; - } - - *options = dict; -out: - if (ret) { - GF_FREE (fpath); - if (dict) - dict_unref (dict); - } - return ret; -} diff --git a/rpc/rpc-lib/src/rpc-clnt.h b/rpc/rpc-lib/src/rpc-clnt.h index 0da165559..584963ad0 100644 --- a/rpc/rpc-lib/src/rpc-clnt.h +++ b/rpc/rpc-lib/src/rpc-clnt.h @@ -237,11 +237,10 @@ void rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config); int rpcclnt_cbk_program_register (struct rpc_clnt *svc, rpcclnt_cb_program_t *program, void *mydata); -int -rpc_clnt_transport_unix_options_build (dict_t **options, char *filepath, - int frame_timeout); - void rpc_clnt_disable (struct rpc_clnt *rpc); +char +rpc_clnt_is_disabled (struct rpc_clnt *rpc); + #endif /* !_RPC_CLNT_H */ diff --git a/rpc/rpc-lib/src/rpc-drc.c b/rpc/rpc-lib/src/rpc-drc.c new file mode 100644 index 000000000..8181e6aee --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.c @@ -0,0 +1,872 @@ +/* + Copyright (c) 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. +*/ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc.h" +#ifndef RPC_DRC_H +#include "rpc-drc.h" +#endif +#include "locking.h" +#include "hashfn.h" +#include "common-utils.h" +#include "statedump.h" +#include "mem-pool.h" + +#include <netinet/in.h> +#include <unistd.h> + +/** + * rpcsvc_drc_op_destroy - Destroys the cached reply + * + * @param drc - the main drc structure + * @param reply - the cached reply to destroy + * @return NULL if reply is destroyed, reply otherwise + */ +static drc_cached_op_t * +rpcsvc_drc_op_destroy (rpcsvc_drc_globals_t *drc, drc_cached_op_t *reply) +{ + GF_ASSERT (drc); + GF_ASSERT (reply); + + if (reply->state == DRC_OP_IN_TRANSIT) + return reply; + + iobref_unref (reply->msg.iobref); + if (reply->msg.rpchdr) + GF_FREE (reply->msg.rpchdr); + if (reply->msg.proghdr) + GF_FREE (reply->msg.proghdr); + if (reply->msg.progpayload) + GF_FREE (reply->msg.progpayload); + + list_del (&reply->global_list); + reply->client->op_count--; + drc->op_count--; + mem_put (reply); + reply = NULL; + + return reply; +} + +/** + * rpcsvc_drc_op_rb_unref - This function is used in rb tree cleanup only + * + * @param reply - the cached reply to unref + * @param drc - the main drc structure + * @return void + */ +static void +rpcsvc_drc_rb_op_destroy (void *reply, void *drc) +{ + rpcsvc_drc_op_destroy (drc, (drc_cached_op_t *)reply); +} + +/** + * rpcsvc_remove_drc_client - Cleanup the drc client + * + * @param client - the drc client to be removed + * @return void + */ +static void +rpcsvc_remove_drc_client (drc_client_t *client) +{ + rb_destroy (client->rbtree, rpcsvc_drc_rb_op_destroy); + list_del (&client->client_list); + GF_FREE (client); +} + +/** + * rpcsvc_client_lookup - Given a sockaddr_storage, find the client if it exists + * + * @param drc - the main drc structure + * @param sockaddr - the network address of the client to be looked up + * @return drc client if it exists, NULL otherwise + */ +static drc_client_t * +rpcsvc_client_lookup (rpcsvc_drc_globals_t *drc, + struct sockaddr_storage *sockaddr) +{ + drc_client_t *client = NULL; + + GF_ASSERT (drc); + GF_ASSERT (sockaddr); + + if (list_empty (&drc->clients_head)) + return NULL; + + list_for_each_entry (client, &drc->clients_head, client_list) { + if (gf_sock_union_equal_addr (&client->sock_union, + (union gf_sock_union *)sockaddr)) + return client; + } + + return NULL; +} + +/** + * drc_compare_reqs - Used by rbtree to determine if incoming req matches with + * an existing node(cached reply) in rbtree + * + * @param item - pointer to the incoming req + * @param rb_node_data - pointer to an rbtree node (cached reply) + * @param param - drc pointer - unused here, but used in *op_destroy + * @return 0 if req matches reply, else (req->xid - reply->xid) + */ +int +drc_compare_reqs (const void *item, const void *rb_node_data, void *param) +{ + int ret = -1; + rpcsvc_request_t *req = NULL; + drc_cached_op_t *reply = NULL; + + GF_ASSERT (item); + GF_ASSERT (rb_node_data); + GF_ASSERT (param); + + req = (rpcsvc_request_t *)item; + reply = (drc_cached_op_t *)rb_node_data; + + ret = req->xid - reply->xid; + if (ret != 0) + return ret; + + if (req->prognum == reply->prognum && + req->procnum == reply->procnum && + req->progver == reply->progversion) + return 0; + + return 1; +} + +/** + * drc_rb_calloc - used by rbtree api to allocate memory for nodes + * + * @param allocator - the libavl_allocator structure used by rbtree + * @param size - not needed by this function + * @return pointer to new cached reply (node in rbtree) + */ +static void * +drc_rb_calloc (struct libavl_allocator *allocator, size_t size) +{ + rpcsvc_drc_globals_t *drc = NULL; + + /* get the drc pointer by simple typecast, since allocator + * is the first member of rpcsvc_drc_globals_t + */ + drc = (rpcsvc_drc_globals_t *)allocator; + + return mem_get (drc->mempool); +} + +/** + * drc_rb_free - used by rbtree api to free a node + * + * @param a - the libavl_allocator structure used by rbtree api + * @param block - node that needs to be freed + * @return void + */ +static void +drc_rb_free (struct libavl_allocator *a, void *block) +{ + mem_put (block); +} + +/** + * drc_init_client_cache - initialize a drc client and its rb tree + * + * @param drc - the main drc structure + * @param client - the drc client to be initialized + * @return 0 on success, -1 on failure + */ +static int +drc_init_client_cache (rpcsvc_drc_globals_t *drc, drc_client_t *client) +{ + GF_ASSERT (drc); + GF_ASSERT (client); + + drc->allocator.libavl_malloc = drc_rb_calloc; + drc->allocator.libavl_free = drc_rb_free; + + client->rbtree = rb_create (drc_compare_reqs, drc, + (struct libavl_allocator *)drc); + if (!client->rbtree) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "rb tree creation failed"); + return -1; + } + + return 0; +} + +/** + * rpcsvc_get_drc_client - find the drc client with given sockaddr, else + * allocate and initialize a new drc client + * + * @param drc - the main drc structure + * @param sockaddr - network address of client + * @return drc client on success, NULL on failure + */ +static drc_client_t * +rpcsvc_get_drc_client (rpcsvc_drc_globals_t *drc, + struct sockaddr_storage *sockaddr) +{ + drc_client_t *client = NULL; + + GF_ASSERT (drc); + GF_ASSERT (sockaddr); + + client = rpcsvc_client_lookup (drc, sockaddr); + if (client) + goto out; + + /* if lookup fails, allocate cache for the new client */ + client = GF_CALLOC (1, sizeof (drc_client_t), + gf_common_mt_drc_client_t); + if (!client) + goto out; + + client->ref = 0; + client->sock_union = (union gf_sock_union)*sockaddr; + client->op_count = 0; + + if (drc_init_client_cache (drc, client)) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, + "initialization of drc client failed"); + GF_FREE (client); + client = NULL; + goto out; + } + drc->client_count++; + + list_add (&client->client_list, &drc->clients_head); + + out: + return client; +} + +/** + * rpcsvc_need_drc - Determine if a request needs DRC service + * + * @param req - incoming request + * @return 1 if DRC is needed for req, 0 otherwise + */ +int +rpcsvc_need_drc (rpcsvc_request_t *req) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT (req); + GF_ASSERT (req->svc); + + drc = req->svc->drc; + + if (!drc || drc->status == DRC_UNINITIATED) + return 0; + + actor = rpcsvc_program_actor (req); + if (!actor) + return 0; + + return (actor->op_type == DRC_NON_IDEMPOTENT + && drc->type != DRC_TYPE_NONE); +} + +/** + * rpcsvc_drc_client_ref - ref the drc client + * + * @param client - the drc client to ref + * @return client + */ +static drc_client_t * +rpcsvc_drc_client_ref (drc_client_t *client) +{ + GF_ASSERT (client); + client->ref++; + return client; +} + +/** + * rpcsvc_drc_client_unref - unref the drc client, and destroy + * the client on last unref + * + * @param drc - the main drc structure + * @param client - the drc client to unref + * @return NULL if it is the last unref, client otherwise + */ +static drc_client_t * +rpcsvc_drc_client_unref (rpcsvc_drc_globals_t *drc, drc_client_t *client) +{ + GF_ASSERT (drc); + GF_ASSERT (client->ref); + + client->ref--; + if (!client->ref) { + drc->client_count--; + rpcsvc_remove_drc_client (client); + client = NULL; + } + + return client; +} + +/** + * rpcsvc_drc_lookup - lookup a request to see if it is already cached + * + * @param req - incoming request + * @return cached reply of req if found, NULL otherwise + */ +drc_cached_op_t * +rpcsvc_drc_lookup (rpcsvc_request_t *req) +{ + drc_client_t *client = NULL; + drc_cached_op_t *reply = NULL; + + GF_ASSERT (req); + + if (!req->trans->drc_client) { + client = rpcsvc_get_drc_client (req->svc->drc, + &req->trans->peerinfo.sockaddr); + if (!client) + goto out; + req->trans->drc_client = client; + } + + client = rpcsvc_drc_client_ref (req->trans->drc_client); + + if (client->op_count == 0) + goto out; + + reply = rb_find (client->rbtree, req); + + out: + if (client) + rpcsvc_drc_client_unref (req->svc->drc, client); + + return reply; +} + +/** + * rpcsvc_send_cached_reply - send the cached reply for the incoming request + * + * @param req - incoming request (which is a duplicate in this case) + * @param reply - the cached reply for req + * @return 0 on successful reply submission, -1 or other non-zero value otherwise + */ +int +rpcsvc_send_cached_reply (rpcsvc_request_t *req, drc_cached_op_t *reply) +{ + int ret = 0; + + GF_ASSERT (req); + GF_ASSERT (reply); + + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "sending cached reply: xid: %d, " + "client: %s", req->xid, req->trans->peerinfo.identifier); + + rpcsvc_drc_client_ref (reply->client); + ret = rpcsvc_transport_submit (req->trans, + reply->msg.rpchdr, reply->msg.rpchdrcount, + reply->msg.proghdr, reply->msg.proghdrcount, + reply->msg.progpayload, reply->msg.progpayloadcount, + reply->msg.iobref, req->trans_private); + rpcsvc_drc_client_unref (req->svc->drc, reply->client); + + return ret; +} + +/** + * rpcsvc_cache_reply - cache the reply for the processed request 'req' + * + * @param req - processed request + * @param iobref - iobref structure of the reply + * @param rpchdr - rpc header of the reply + * @param rpchdrcount - size of rpchdr + * @param proghdr - program header of the reply + * @param proghdrcount - size of proghdr + * @param payload - payload of the reply if any + * @param payloadcount - size of payload + * @return 0 on success, -1 on failure + */ +int +rpcsvc_cache_reply (rpcsvc_request_t *req, struct iobref *iobref, + struct iovec *rpchdr, int rpchdrcount, + struct iovec *proghdr, int proghdrcount, + struct iovec *payload, int payloadcount) +{ + int ret = -1; + drc_cached_op_t *reply = NULL; + + GF_ASSERT (req); + GF_ASSERT (req->reply); + + reply = req->reply; + + reply->state = DRC_OP_CACHED; + + reply->msg.iobref = iobref_ref (iobref); + + reply->msg.rpchdrcount = rpchdrcount; + reply->msg.rpchdr = iov_dup (rpchdr, rpchdrcount); + + reply->msg.proghdrcount = proghdrcount; + reply->msg.proghdr = iov_dup (proghdr, proghdrcount); + + reply->msg.progpayloadcount = payloadcount; + if (payloadcount) + reply->msg.progpayload = iov_dup (payload, payloadcount); + + // rpcsvc_drc_client_unref (req->svc->drc, req->trans->drc_client); + // rpcsvc_drc_op_unref (req->svc->drc, reply); + ret = 0; + + return ret; +} + +/** + * rpcsvc_vacate_drc_entries - free up some percentage of drc cache + * based on the lru factor + * + * @param drc - the main drc structure + * @return void + */ +static void +rpcsvc_vacate_drc_entries (rpcsvc_drc_globals_t *drc) +{ + uint32_t i = 0; + uint32_t n = 0; + drc_cached_op_t *reply = NULL; + drc_cached_op_t *tmp = NULL; + drc_client_t *client = NULL; + + GF_ASSERT (drc); + + n = drc->global_cache_size / drc->lru_factor; + + list_for_each_entry_safe_reverse (reply, tmp, &drc->cache_head, global_list) { + /* Don't delete ops that are in transit */ + if (reply->state == DRC_OP_IN_TRANSIT) + continue; + + client = reply->client; + + (void *)rb_delete (client->rbtree, reply); + + rpcsvc_drc_op_destroy (drc, reply); + rpcsvc_drc_client_unref (drc, client); + i++; + if (i >= n) + break; + } +} + +/** + * rpcsvc_add_op_to_cache - insert the cached op into the client rbtree and drc list + * + * @param drc - the main drc structure + * @param reply - the op to be inserted + * @return 0 on success, -1 on failure + */ +static int +rpcsvc_add_op_to_cache (rpcsvc_drc_globals_t *drc, drc_cached_op_t *reply) +{ + drc_client_t *client = NULL; + drc_cached_op_t **tmp_reply = NULL; + + GF_ASSERT (drc); + GF_ASSERT (reply); + + client = reply->client; + + /* cache is full, free up some space */ + if (drc->op_count >= drc->global_cache_size) + rpcsvc_vacate_drc_entries (drc); + + tmp_reply = (drc_cached_op_t **)rb_probe (client->rbtree, reply); + if (*tmp_reply != reply) { + /* should never happen */ + gf_log (GF_RPCSVC, GF_LOG_ERROR, + "DRC failed to detect duplicates"); + return -1; + } else if (*tmp_reply == NULL) { + /* mem alloc failed */ + return -1; + } + + client->op_count++; + list_add (&reply->global_list, &drc->cache_head); + drc->op_count++; + + return 0; +} + +/** + * rpcsvc_cache_request - cache the in-transition incoming request + * + * @param req - incoming request + * @return 0 on success, -1 on failure + */ +int +rpcsvc_cache_request (rpcsvc_request_t *req) +{ + int ret = -1; + drc_client_t *client = NULL; + drc_cached_op_t *reply = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT (req); + + drc = req->svc->drc; + + client = req->trans->drc_client; + if (!client) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "drc client is NULL"); + goto out; + } + + reply = mem_get (drc->mempool); + if (!reply) + goto out; + + reply->client = rpcsvc_drc_client_ref (client); + reply->xid = req->xid; + reply->prognum = req->prognum; + reply->progversion = req->progver; + reply->procnum = req->procnum; + reply->state = DRC_OP_IN_TRANSIT; + req->reply = reply; + + ret = rpcsvc_add_op_to_cache (drc, reply); + if (ret) { + req->reply = NULL; + rpcsvc_drc_op_destroy (drc, reply); + rpcsvc_drc_client_unref (drc, client); + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Failed to add op to drc cache"); + } + + out: + return ret; +} + +/** + * + * rpcsvc_drc_priv - function which dumps the drc state + * + * @param drc - the main drc structure + * @return 0 on success, -1 on failure + */ +int32_t +rpcsvc_drc_priv (rpcsvc_drc_globals_t *drc) +{ + int i = 0; + char key[GF_DUMP_MAX_BUF_LEN] = {0}; + drc_client_t *client = NULL; + char ip[INET6_ADDRSTRLEN] = {0}; + + if (!drc || drc->status == DRC_UNINITIATED) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "DRC is " + "uninitialized, not dumping its state"); + return 0; + } + + gf_proc_dump_add_section("rpc.drc"); + + if (TRY_LOCK (&drc->lock)) + return -1; + + gf_proc_dump_build_key (key, "drc", "type"); + gf_proc_dump_write (key, "%d", drc->type); + + gf_proc_dump_build_key (key, "drc", "client_count"); + gf_proc_dump_write (key, "%d", drc->client_count); + + gf_proc_dump_build_key (key, "drc", "current_cache_size"); + gf_proc_dump_write (key, "%d", drc->op_count); + + gf_proc_dump_build_key (key, "drc", "max_cache_size"); + gf_proc_dump_write (key, "%d", drc->global_cache_size); + + gf_proc_dump_build_key (key, "drc", "lru_factor"); + gf_proc_dump_write (key, "%d", drc->lru_factor); + + gf_proc_dump_build_key (key, "drc", "duplicate_request_count"); + gf_proc_dump_write (key, "%d", drc->cache_hits); + + gf_proc_dump_build_key (key, "drc", "in_transit_duplicate_requests"); + gf_proc_dump_write (key, "%d", drc->intransit_hits); + + list_for_each_entry (client, &drc->clients_head, client_list) { + gf_proc_dump_build_key (key, "client", "%d.ip-address", i); + memset (ip, 0, INET6_ADDRSTRLEN); + switch (client->sock_union.storage.ss_family) { + case AF_INET: + gf_proc_dump_write (key, "%s", inet_ntop (AF_INET, + &client->sock_union.sin.sin_addr.s_addr, + ip, INET_ADDRSTRLEN)); + break; + case AF_INET6: + gf_proc_dump_write (key, "%s", inet_ntop (AF_INET6, + &client->sock_union.sin6.sin6_addr, + ip, INET6_ADDRSTRLEN)); + break; + default: + gf_proc_dump_write (key, "%s", "N/A"); + } + + gf_proc_dump_build_key (key, "client", "%d.ref_count", i); + gf_proc_dump_write (key, "%d", client->ref); + gf_proc_dump_build_key (key, "client", "%d.op_count", i); + gf_proc_dump_write (key, "%d", client->op_count); + i++; + } + + UNLOCK (&drc->lock); + return 0; +} + +/** + * rpcsvc_drc_notify - function which is notified of RPC transport events + * + * @param svc - pointer to rpcsvc_t structure of the rpc + * @param xl - pointer to the xlator + * @param event - the event which triggered this notify + * @param data - the transport structure + * @return 0 on success, -1 on failure + */ +int +rpcsvc_drc_notify (rpcsvc_t *svc, void *xl, + rpcsvc_event_t event, void *data) +{ + int ret = -1; + rpc_transport_t *trans = NULL; + drc_client_t *client = NULL; + rpcsvc_drc_globals_t *drc = NULL; + + GF_ASSERT (svc); + GF_ASSERT (svc->drc); + GF_ASSERT (data); + + drc = svc->drc; + + if (drc->status == DRC_UNINITIATED || + drc->type == DRC_TYPE_NONE) + return 0; + + LOCK (&drc->lock); + + trans = (rpc_transport_t *)data; + client = rpcsvc_get_drc_client (drc, &trans->peerinfo.sockaddr); + if (!client) + goto out; + + switch (event) { + case RPCSVC_EVENT_ACCEPT: + trans->drc_client = rpcsvc_drc_client_ref (client); + ret = 0; + break; + + case RPCSVC_EVENT_DISCONNECT: + ret = 0; + if (list_empty (&drc->clients_head)) + break; + /* should be the last unref */ + rpcsvc_drc_client_unref (drc, client); + trans->drc_client = NULL; + break; + + default: + break; + } + + out: + UNLOCK (&drc->lock); + return ret; +} + +/** + * rpcsvc_drc_init - Initialize the duplicate request cache service + * + * @param svc - pointer to rpcsvc_t structure of the rpc + * @param options - the options dictionary which configures drc + * @return 0 on success, non-zero integer on failure + */ +int +rpcsvc_drc_init (rpcsvc_t *svc, dict_t *options) +{ + int ret = 0; + uint32_t drc_type = 0; + uint32_t drc_size = 0; + uint32_t drc_factor = 0; + rpcsvc_drc_globals_t *drc = NULL; + static gf_boolean_t drc_inited = _gf_false; + + GF_ASSERT (svc); + GF_ASSERT (options); + + /* Already inited */ + if (drc_inited) + return 0; + + if (!svc->drc) { + drc = GF_CALLOC (1, sizeof (rpcsvc_drc_globals_t), + gf_common_mt_drc_globals_t); + if (!drc) + return -1; + + svc->drc = drc; + LOCK_INIT (&drc->lock); + } else { + drc = svc->drc; + } + + LOCK (&drc->lock); + if (drc->type != DRC_TYPE_NONE) { + ret = 0; + goto out; + } + + /* Toggle DRC on/off, when more drc types(persistent/cluster) + are added, we shouldn't treat this as boolean */ + ret = dict_get_str_boolean (options, "nfs.drc", _gf_true); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_INFO, "drc user options need second look"); + ret = _gf_true; + } + drc->enable_drc = ret; + + if (ret == _gf_false) { + /* drc off */ + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "DRC is off"); + ret = 0; + goto out; + } + + /* Specify type of DRC to be used */ + ret = dict_get_uint32 (options, "nfs.drc-type", &drc_type); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "drc type not set." + " Continuing with default"); + drc_type = DRC_DEFAULT_TYPE; + } + + drc->type = drc_type; + + /* Set the global cache size (no. of ops to cache) */ + ret = dict_get_uint32 (options, "nfs.drc-size", &drc_size); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "drc size not set." + " Continuing with default size"); + drc_size = DRC_DEFAULT_CACHE_SIZE; + } + + drc->global_cache_size = drc_size; + + /* Mempool for cached ops */ + drc->mempool = mem_pool_new (drc_cached_op_t, drc->global_cache_size); + if (!drc->mempool) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get mempool for" + " DRC, drc-size: %d", drc->global_cache_size); + ret = -1; + goto out; + } + + /* What percent of cache to be evicted whenever it fills up */ + ret = dict_get_uint32 (options, "nfs.drc-lru-factor", &drc_factor); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "drc lru factor not set." + " Continuing with policy default"); + drc_factor = DRC_DEFAULT_LRU_FACTOR; + } + + drc->lru_factor = (drc_lru_factor_t) drc_factor; + + INIT_LIST_HEAD (&drc->clients_head); + INIT_LIST_HEAD (&drc->cache_head); + + ret = rpcsvc_register_notify (svc, rpcsvc_drc_notify, THIS); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, + "registration of drc_notify function failed"); + goto out; + } + + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "drc init successful"); + drc->status = DRC_INITIATED; + drc_inited = _gf_true; + + out: + UNLOCK (&drc->lock); + if (ret == -1) { + if (drc->mempool) { + mem_pool_destroy (drc->mempool); + drc->mempool = NULL; + } + GF_FREE (drc); + svc->drc = NULL; + } + return ret; +} + +int +rpcsvc_drc_reconfigure (rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + gf_boolean_t enable_drc = _gf_false; + rpcsvc_drc_globals_t *drc = NULL; + uint32_t drc_size = 0; + + if ((!svc) || (!options)) + return (-1); + + drc = svc->drc; + /* reconfig for drc-size */ + if (dict_get_uint32 (options, "nfs.drc-size", &drc_size)) + drc_size = DRC_DEFAULT_CACHE_SIZE; + + if (drc->global_cache_size != drc_size) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "nfs.drc-size size can not " + "be reconfigured without NFS server restart."); + return (-1); + } + + /* reconfig for nfs.drc */ + ret = dict_get_str_boolean (options, "nfs.drc", _gf_true); + if (ret < 0) { + ret = _gf_true; + } + enable_drc = ret; + + if (drc->enable_drc == enable_drc) + return 0; + + drc->enable_drc = enable_drc; + if (enable_drc) { + if (drc == NULL) + return rpcsvc_drc_init(svc, options); + } else { + if (drc == NULL) + return (0); + + LOCK (&drc->lock); + (void) rpcsvc_unregister_notify (svc, rpcsvc_drc_notify, THIS); + if (drc->mempool) { + mem_pool_destroy (drc->mempool); + drc->mempool = NULL; + } + UNLOCK (&drc->lock); + GF_FREE (drc); + svc->drc = NULL; + } + + return (0); +} diff --git a/rpc/rpc-lib/src/rpc-drc.h b/rpc/rpc-lib/src/rpc-drc.h new file mode 100644 index 000000000..7dfaef978 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.h @@ -0,0 +1,104 @@ +/* + Copyright (c) 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. +*/ + +#ifndef RPC_DRC_H +#define RPC_DRC_H + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc-common.h" +#include "rpcsvc.h" +#include "locking.h" +#include "dict.h" +#include "rb.h" + +/* per-client cache structure */ +struct drc_client { + uint32_t ref; + union gf_sock_union sock_union; + /* pointers to the cache */ + struct rb_table *rbtree; + /* no. of ops currently cached */ + uint32_t op_count; + struct list_head client_list; +}; + +struct drc_cached_op { + drc_op_state_t state; + uint32_t xid; + int prognum; + int progversion; + int procnum; + rpc_transport_msg_t msg; + drc_client_t *client; + struct list_head client_list; + struct list_head global_list; + int32_t ref; +}; + +/* global drc definitions */ +enum drc_status { + DRC_UNINITIATED, + DRC_INITIATED +}; +typedef enum drc_status drc_status_t; + +struct drc_globals { + /* allocator must be the first member since + * it is used so in gf_libavl_allocator + */ + struct libavl_allocator allocator; + drc_type_t type; + /* configurable size parameter */ + uint32_t global_cache_size; + drc_lru_factor_t lru_factor; + gf_lock_t lock; + drc_status_t status; + uint32_t op_count; + uint64_t cache_hits; + uint64_t intransit_hits; + struct mem_pool *mempool; + struct list_head cache_head; + uint32_t client_count; + struct list_head clients_head; + gf_boolean_t enable_drc; +}; + +int +rpcsvc_need_drc (rpcsvc_request_t *req); + +drc_cached_op_t * +rpcsvc_drc_lookup (rpcsvc_request_t *req); + +int +rpcsvc_send_cached_reply (rpcsvc_request_t *req, drc_cached_op_t *reply); + +int +rpcsvc_cache_reply (rpcsvc_request_t *req, struct iobref *iobref, + struct iovec *rpchdr, int rpchdrcount, + struct iovec *proghdr, int proghdrcount, + struct iovec *payload, int payloadcount); + +int +rpcsvc_cache_request (rpcsvc_request_t *req); + +int32_t +rpcsvc_drc_priv (rpcsvc_drc_globals_t *drc); + +int +rpcsvc_drc_init (rpcsvc_t *svc, dict_t *options); + +int +rpcsvc_drc_reconfigure (rpcsvc_t *svc, dict_t *options); + +#endif /* RPC_DRC_H */ diff --git a/rpc/rpc-lib/src/rpc-transport.c b/rpc/rpc-lib/src/rpc-transport.c index 695261a71..c24d41084 100644 --- a/rpc/rpc-lib/src/rpc-transport.c +++ b/rpc/rpc-lib/src/rpc-transport.c @@ -69,6 +69,19 @@ out: return ret; } +int +rpc_transport_throttle (rpc_transport_t *this, gf_boolean_t onoff) +{ + int ret = 0; + + if (!this->ops->throttle) + return -ENOSYS; + + ret = this->ops->throttle (this, onoff); + + return ret; +} + int32_t rpc_transport_get_peeraddr (rpc_transport_t *this, char *peeraddr, int addrlen, struct sockaddr_storage *sa, size_t salen) @@ -145,6 +158,7 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) int8_t is_tcp = 0, is_unix = 0, is_ibsdp = 0; volume_opt_list_t *vol_opt = NULL; gf_boolean_t bind_insecure = _gf_false; + xlator_t *this = NULL; GF_VALIDATE_OR_GOTO("rpc-transport", options, fail); GF_VALIDATE_OR_GOTO("rpc-transport", ctx, fail); @@ -169,7 +183,7 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) gf_log ("dict", GF_LOG_DEBUG, "setting transport-type failed"); else - gf_log ("rpc-transport", GF_LOG_WARNING, + gf_log ("rpc-transport", GF_LOG_DEBUG, "missing 'option transport-type'. defaulting to " "\"socket\""); } else { @@ -281,7 +295,7 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) } *VOID(&(trans->reconfigure)) = dlsym (handle, "reconfigure"); - if (trans->fini == NULL) { + if (trans->reconfigure == NULL) { gf_log ("rpc-transport", GF_LOG_DEBUG, "dlsym (gf_rpc_transport_reconfigure) on %s", dlerror()); } @@ -292,26 +306,26 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) goto fail; } + this = THIS; vol_opt->given_opt = dlsym (handle, "options"); if (vol_opt->given_opt == NULL) { gf_log ("rpc-transport", GF_LOG_DEBUG, "volume option validation not specified"); } else { INIT_LIST_HEAD (&vol_opt->list); - list_add_tail (&vol_opt->list, &(THIS->volume_options)); - if (xlator_options_validate_list (THIS, options, vol_opt, + list_add_tail (&vol_opt->list, &(this->volume_options)); + if (xlator_options_validate_list (this, options, vol_opt, NULL)) { gf_log ("rpc-transport", GF_LOG_ERROR, "volume option validation failed"); goto fail; } - vol_opt = NULL; } trans->options = options; pthread_mutex_init (&trans->lock, NULL); - trans->xl = THIS; + trans->xl = this; ret = trans->init (trans); if (ret != 0) { @@ -324,8 +338,6 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) GF_FREE (name); - GF_FREE (vol_opt); - return return_trans; fail: @@ -340,7 +352,10 @@ fail: GF_FREE (name); - GF_FREE (vol_opt); + if (vol_opt && !list_empty (&vol_opt->list)) { + list_del_init (&vol_opt->list); + GF_FREE (vol_opt); + } return NULL; } @@ -475,6 +490,8 @@ rpc_transport_unref (rpc_transport_t *this) if (this->mydata) this->notify (this, this->mydata, RPC_TRANSPORT_CLEANUP, NULL); + this->mydata = NULL; + this->notify = NULL; rpc_transport_destroy (this); } @@ -518,18 +535,6 @@ out: } -inline int -rpc_transport_unregister_notify (rpc_transport_t *trans) -{ - GF_VALIDATE_OR_GOTO ("rpc-transport", trans, out); - - trans->notify = NULL; - trans->mydata = NULL; - -out: - return 0; -} - //give negative values to skip setting that value //this function asserts if both the values are negative. @@ -557,6 +562,63 @@ out: } int +rpc_transport_unix_options_build (dict_t **options, char *filepath, + int frame_timeout) +{ + dict_t *dict = NULL; + char *fpath = NULL; + int ret = -1; + + GF_ASSERT (filepath); + GF_ASSERT (options); + + dict = dict_new (); + if (!dict) + goto out; + + fpath = gf_strdup (filepath); + if (!fpath) { + ret = -1; + goto out; + } + + ret = dict_set_dynstr (dict, "transport.socket.connect-path", fpath); + if (ret) + goto out; + + ret = dict_set_str (dict, "transport.address-family", "unix"); + if (ret) + goto out; + + ret = dict_set_str (dict, "transport.socket.nodelay", "off"); + if (ret) + goto out; + + ret = dict_set_str (dict, "transport-type", "socket"); + if (ret) + goto out; + + ret = dict_set_str (dict, "transport.socket.keepalive", "off"); + if (ret) + goto out; + + if (frame_timeout > 0) { + ret = dict_set_int32 (dict, "frame-timeout", frame_timeout); + if (ret) + goto out; + } + + *options = dict; +out: + if (ret) { + GF_FREE (fpath); + if (dict) + dict_unref (dict); + } + return ret; +} + +int rpc_transport_inet_options_build (dict_t **options, const char *hostname, int port) { diff --git a/rpc/rpc-lib/src/rpc-transport.h b/rpc/rpc-lib/src/rpc-transport.h index 272de9d7d..2db9072ae 100644 --- a/rpc/rpc-lib/src/rpc-transport.h +++ b/rpc/rpc-lib/src/rpc-transport.h @@ -186,16 +186,19 @@ struct rpc_transport { */ void *private; - void *xl_private; + struct _client_t *xl_private; void *xl; /* Used for THIS */ void *mydata; pthread_mutex_t lock; int32_t refcount; + int32_t outstanding_rpc_count; + glusterfs_ctx_t *ctx; dict_t *options; char *name; void *dnscache; + void *drc_client; data_t *buf; int32_t (*init) (rpc_transport_t *this); void (*fini) (rpc_transport_t *this); @@ -210,7 +213,7 @@ struct rpc_transport { struct list_head list; int bind_insecure; - void *dl_handle; /* handle of dlopen() */ + void *dl_handle; /* handle of dlopen() */ }; struct rpc_transport_ops { @@ -234,6 +237,7 @@ struct rpc_transport_ops { int32_t (*get_myaddr) (rpc_transport_t *this, char *peeraddr, int addrlen, struct sockaddr_storage *sa, socklen_t sasize); + int32_t (*throttle) (rpc_transport_t *this, gf_boolean_t onoff); }; @@ -273,9 +277,6 @@ int rpc_transport_register_notify (rpc_transport_t *trans, rpc_transport_notify_t, void *mydata); -int -rpc_transport_unregister_notify (rpc_transport_t *trans); - int32_t rpc_transport_get_peername (rpc_transport_t *this, char *hostname, int hostlen); @@ -290,6 +291,9 @@ int32_t rpc_transport_get_myaddr (rpc_transport_t *this, char *peeraddr, int addrlen, struct sockaddr_storage *sa, size_t salen); +int +rpc_transport_throttle (rpc_transport_t *this, gf_boolean_t onoff); + rpc_transport_pollin_t * rpc_transport_pollin_alloc (rpc_transport_t *this, struct iovec *vector, int count, struct iobuf *hdr_iobuf, @@ -302,5 +306,9 @@ rpc_transport_keepalive_options_set (dict_t *options, int32_t interval, int32_t time); int +rpc_transport_unix_options_build (dict_t **options, char *filepath, + int frame_timeout); + +int rpc_transport_inet_options_build (dict_t **options, const char *hostname, int port); #endif /* __RPC_TRANSPORT_H__ */ diff --git a/rpc/rpc-lib/src/rpcsvc-auth.c b/rpc/rpc-lib/src/rpcsvc-auth.c index 907ae1ec9..4cb86a758 100644 --- a/rpc/rpc-lib/src/rpcsvc-auth.c +++ b/rpc/rpc-lib/src/rpcsvc-auth.c @@ -178,6 +178,29 @@ err: } int +rpcsvc_set_addr_namelookup (rpcsvc_t *svc, dict_t *options) +{ + int ret; + static char *addrlookup_key = "rpc-auth.addr.namelookup"; + + if (!svc || !options) + return (-1); + + /* By default it's disabled */ + ret = dict_get_str_boolean (options, addrlookup_key, _gf_false); + if (ret < 0) { + svc->addr_namelookup = _gf_false; + } else { + svc->addr_namelookup = ret; + } + + if (svc->addr_namelookup) + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Addr-Name lookup enabled"); + + return (0); +} + +int rpcsvc_set_allow_insecure (rpcsvc_t *svc, dict_t *options) { int ret = -1; @@ -206,11 +229,16 @@ rpcsvc_set_allow_insecure (rpcsvc_t *svc, dict_t *options) int rpcsvc_set_root_squash (rpcsvc_t *svc, dict_t *options) { + int ret = -1; + GF_ASSERT (svc); GF_ASSERT (options); - if (dict_get_str_boolean (options, "root-squash", 0)) - svc->root_squash = _gf_true; + ret = dict_get_str_boolean (options, "root-squash", 0); + if (ret != -1) + svc->root_squash = ret; + else + svc->root_squash = _gf_false; if (svc->root_squash) gf_log (GF_RPCSVC, GF_LOG_DEBUG, "root squashing enabled "); @@ -228,6 +256,7 @@ rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options) (void) rpcsvc_set_allow_insecure (svc, options); (void) rpcsvc_set_root_squash (svc, options); + (void) rpcsvc_set_addr_namelookup (svc, options); ret = rpcsvc_auth_add_initers (svc); if (ret == -1) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to add initers"); @@ -244,6 +273,25 @@ out: return ret; } +int +rpcsvc_auth_reconf (rpcsvc_t *svc, dict_t *options) +{ + int ret = 0; + + if ((!svc) || (!options)) + return (-1); + + ret = rpcsvc_set_allow_insecure (svc, options); + if (ret) + return (-1); + + ret = rpcsvc_set_root_squash (svc, options); + if (ret) + return (-1); + + return rpcsvc_set_addr_namelookup (svc, options); +} + rpcsvc_auth_t * __rpcsvc_auth_get_handler (rpcsvc_request_t *req) @@ -322,6 +370,9 @@ rpcsvc_auth_request_init (rpcsvc_request_t *req) if (!auth->authops->request_init) ret = auth->authops->request_init (req, auth->authprivate); + req->auxgids = req->auxgidsmall; /* reset to auxgidlarge during + unsersialize if necessary */ + req->auxgidlarge = NULL; err: return ret; } @@ -361,14 +412,10 @@ err: int rpcsvc_auth_array (rpcsvc_t *svc, char *volname, int *autharr, int arrlen) { - int count = 0; - int gen = RPCSVC_AUTH_REJECT; - int spec = RPCSVC_AUTH_REJECT; - int final = RPCSVC_AUTH_REJECT; - char *srchstr = NULL; - char *valstr = NULL; - gf_boolean_t boolval = _gf_false; - int ret = 0; + int count = 0; + int result = RPCSVC_AUTH_REJECT; + char *srchstr = NULL; + int ret = 0; struct rpcsvc_auth_list *auth = NULL; struct rpcsvc_auth_list *tmp = NULL; @@ -386,59 +433,27 @@ rpcsvc_auth_array (rpcsvc_t *svc, char *volname, int *autharr, int arrlen) if (count >= arrlen) break; - gen = gf_asprintf (&srchstr, "rpc-auth.%s", auth->name); - if (gen == -1) { + result = gf_asprintf (&srchstr, "rpc-auth.%s.%s", + auth->name, volname); + if (result == -1) { count = -1; goto err; } - gen = RPCSVC_AUTH_REJECT; - if (dict_get (svc->options, srchstr)) { - ret = dict_get_str (svc->options, srchstr, &valstr); - if (ret == 0) { - ret = gf_string2boolean (valstr, &boolval); - if (ret == 0) { - if (boolval == _gf_true) - gen = RPCSVC_AUTH_ACCEPT; - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Faile" - "d to read auth val"); - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Faile" - "d to read auth val"); - } - + ret = dict_get_str_boolean (svc->options, srchstr, 0xC00FFEE); GF_FREE (srchstr); - spec = gf_asprintf (&srchstr, "rpc-auth.%s.%s", auth->name, - volname); - if (spec == -1) { - count = -1; - goto err; - } - - spec = RPCSVC_AUTH_DONTCARE; - if (dict_get (svc->options, srchstr)) { - ret = dict_get_str (svc->options, srchstr, &valstr); - if (ret == 0) { - ret = gf_string2boolean (valstr, &boolval); - if (ret == 0) { - if (boolval == _gf_true) - spec = RPCSVC_AUTH_ACCEPT; - else - spec = RPCSVC_AUTH_REJECT; - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Faile" - "d to read auth val"); - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Faile" - "d to read auth val"); - } - GF_FREE (srchstr); - final = rpcsvc_combine_gen_spec_volume_checks (gen, spec); - if (final == RPCSVC_AUTH_ACCEPT) { + switch (ret) { + case _gf_true: + result = RPCSVC_AUTH_ACCEPT; autharr[count] = auth->auth->authnum; ++count; + break; + case _gf_false: + result = RPCSVC_AUTH_REJECT; + break; + default: + result = RPCSVC_AUTH_DONTCARE; } } diff --git a/rpc/rpc-lib/src/rpcsvc-common.h b/rpc/rpc-lib/src/rpcsvc-common.h index 2c6f07488..aed55e039 100644 --- a/rpc/rpc-lib/src/rpcsvc-common.h +++ b/rpc/rpc-lib/src/rpcsvc-common.h @@ -30,6 +30,8 @@ struct rpcsvc_state; typedef int (*rpcsvc_notify_t) (struct rpcsvc_state *, void *mydata, rpcsvc_event_t, void *data); +struct drc_globals; +typedef struct drc_globals rpcsvc_drc_globals_t; /* Contains global state required for all the RPC services. */ @@ -50,25 +52,75 @@ typedef struct rpcsvc_state { dict_t *options; /* Allow insecure ports. */ - int allow_insecure; + gf_boolean_t allow_insecure; gf_boolean_t register_portmap; gf_boolean_t root_squash; glusterfs_ctx_t *ctx; /* list of connections which will listen for incoming connections */ - struct list_head listeners; + struct list_head listeners; /* list of programs registered with rpcsvc */ - struct list_head programs; + struct list_head programs; /* list of notification callbacks */ - struct list_head notify; - int notify_count; + struct list_head notify; + int notify_count; void *mydata; /* This is xlator */ - rpcsvc_notify_t notifyfn; + rpcsvc_notify_t notifyfn; struct mem_pool *rxpool; + rpcsvc_drc_globals_t *drc; + + /* per-client limit of outstanding rpc requests */ + int outstanding_rpc_limit; + gf_boolean_t addr_namelookup; } rpcsvc_t; +/* DRC START */ +enum drc_op_type { + DRC_NA = 0, + DRC_IDEMPOTENT = 1, + DRC_NON_IDEMPOTENT = 2 +}; +typedef enum drc_op_type drc_op_type_t; + +enum drc_type { + DRC_TYPE_NONE = 0, + DRC_TYPE_IN_MEMORY = 1 +}; +typedef enum drc_type drc_type_t; + +enum drc_lru_factor { + DRC_LRU_5_PC = 20, + DRC_LRU_10_PC = 10, + DRC_LRU_25_PC = 4, + DRC_LRU_50_PC = 2 +}; +typedef enum drc_lru_factor drc_lru_factor_t; + +enum drc_xid_state { + DRC_XID_MONOTONOUS = 0, + DRC_XID_WRAPPED = 1 +}; +typedef enum drc_xid_state drc_xid_state_t; + +enum drc_op_state { + DRC_OP_IN_TRANSIT = 0, + DRC_OP_CACHED = 1 +}; +typedef enum drc_op_state drc_op_state_t; + +enum drc_policy { + DRC_LRU = 0 +}; +typedef enum drc_policy drc_policy_t; + +/* Default policies for DRC */ +#define DRC_DEFAULT_TYPE DRC_TYPE_IN_MEMORY +#define DRC_DEFAULT_CACHE_SIZE 0x20000 +#define DRC_DEFAULT_LRU_FACTOR DRC_LRU_25_PC + +/* DRC END */ #endif /* #ifndef _RPCSVC_COMMON_H */ diff --git a/rpc/rpc-lib/src/rpcsvc.c b/rpc/rpc-lib/src/rpcsvc.c index 078adc062..037c157f2 100644 --- a/rpc/rpc-lib/src/rpcsvc.c +++ b/rpc/rpc-lib/src/rpcsvc.c @@ -28,6 +28,7 @@ #include "xdr-generic.h" #include "rpc-common-xdr.h" #include "syncop.h" +#include "rpc-drc.h" #include <errno.h> #include <pthread.h> @@ -41,8 +42,7 @@ #include <stdio.h> #include "xdr-rpcclnt.h" - -#define ACL_PROGRAM 100227 +#include "glusterfs-acl.h" struct rpcsvc_program gluster_dump_prog; @@ -129,6 +129,37 @@ rpcsvc_get_program_vector_sizer (rpcsvc_t *svc, uint32_t prognum, return NULL; } +int +rpcsvc_request_outstanding (rpcsvc_t *svc, rpc_transport_t *trans, int delta) +{ + int ret = 0; + int old_count = 0; + int new_count = 0; + int limit = 0; + + pthread_mutex_lock (&trans->lock); + { + limit = svc->outstanding_rpc_limit; + if (!limit) + goto unlock; + + old_count = trans->outstanding_rpc_count; + trans->outstanding_rpc_count += delta; + new_count = trans->outstanding_rpc_count; + + if (old_count <= limit && new_count > limit) + ret = rpc_transport_throttle (trans, _gf_true); + + if (old_count > limit && new_count <= limit) + ret = rpc_transport_throttle (trans, _gf_false); + } +unlock: + pthread_mutex_unlock (&trans->lock); + + return ret; +} + + /* This needs to change to returning errors, since * we need to return RPC specific error messages when some * of the pointers below are NULL. @@ -207,7 +238,7 @@ rpcsvc_program_actor (rpcsvc_request_t *req) goto err; } - req->synctask = program->synctask; + req->synctask = program->synctask; err = SUCCESS; gf_log (GF_RPCSVC, GF_LOG_TRACE, "Actor found: %s - %s", @@ -279,8 +310,17 @@ rpcsvc_request_destroy (rpcsvc_request_t *req) if (req->hdr_iobuf) iobuf_unref (req->hdr_iobuf); + /* This marks the "end" of an RPC request. Reply is + completely written to the socket and is on the way + to the client. It is time to decrement the + outstanding request counter by 1. + */ + rpcsvc_request_outstanding (req->svc, req->trans, -1); + rpc_transport_unref (req->trans); + GF_FREE (req->auxgidlarge); + mem_put (req); out: @@ -363,6 +403,12 @@ rpcsvc_request_create (rpcsvc_t *svc, rpc_transport_t *trans, goto err; } + /* We just received a new request from the wire. Account for + it in the outsanding request counter to make sure we don't + ingest too many concurrent requests from the same client. + */ + ret = rpcsvc_request_outstanding (svc, trans, +1); + msgbuf = msg->vector[0].iov_base; msglen = msg->vector[0].iov_len; @@ -422,6 +468,7 @@ rpcsvc_request_create (rpcsvc_t *svc, rpc_transport_t *trans, * since we are not handling authentication failures for now. */ req->rpc_status = MSG_ACCEPTED; + req->reply = NULL; ret = 0; err: if (ret == -1) { @@ -439,9 +486,9 @@ err: int rpcsvc_check_and_reply_error (int ret, call_frame_t *frame, void *opaque) { - rpcsvc_request_t *req = NULL; + rpcsvc_request_t *req = NULL; - req = opaque; + req = opaque; if (ret) gf_log ("rpcsvc", GF_LOG_ERROR, @@ -454,20 +501,22 @@ rpcsvc_check_and_reply_error (int ret, call_frame_t *frame, void *opaque) "failed to queue error reply"); } - return 0; + return 0; } int rpcsvc_handle_rpc_call (rpcsvc_t *svc, rpc_transport_t *trans, rpc_transport_pollin_t *msg) { - rpcsvc_actor_t *actor = NULL; - rpcsvc_actor actor_fn = NULL; - rpcsvc_request_t *req = NULL; - int ret = -1; - uint16_t port = 0; - gf_boolean_t is_unix = _gf_false; - gf_boolean_t unprivileged = _gf_false; + rpcsvc_actor_t *actor = NULL; + rpcsvc_actor actor_fn = NULL; + rpcsvc_request_t *req = NULL; + int ret = -1; + uint16_t port = 0; + gf_boolean_t is_unix = _gf_false; + gf_boolean_t unprivileged = _gf_false; + drc_cached_op_t *reply = NULL; + rpcsvc_drc_globals_t *drc = NULL; if (!trans || !svc) return -1; @@ -503,7 +552,7 @@ rpcsvc_handle_rpc_call (rpcsvc_t *svc, rpc_transport_t *trans, req = rpcsvc_request_create (svc, trans, msg); if (!req) - goto err; + goto out; if (!rpcsvc_request_accepted (req)) goto err_reply; @@ -521,33 +570,65 @@ rpcsvc_handle_rpc_call (rpcsvc_t *svc, rpc_transport_t *trans, return -1; } + /* DRC */ + if (rpcsvc_need_drc (req)) { + drc = req->svc->drc; + + LOCK (&drc->lock); + reply = rpcsvc_drc_lookup (req); + + /* retransmission of completed request, send cached reply */ + if (reply && reply->state == DRC_OP_CACHED) { + gf_log (GF_RPCSVC, GF_LOG_INFO, "duplicate request:" + " XID: 0x%x", req->xid); + ret = rpcsvc_send_cached_reply (req, reply); + drc->cache_hits++; + UNLOCK (&drc->lock); + goto out; + + } /* retransmitted request, original op in transit, drop it */ + else if (reply && reply->state == DRC_OP_IN_TRANSIT) { + gf_log (GF_RPCSVC, GF_LOG_INFO, "op in transit," + " discarding. XID: 0x%x", req->xid); + ret = 0; + drc->intransit_hits++; + rpcsvc_request_destroy (req); + UNLOCK (&drc->lock); + goto out; + + } /* fresh request, cache it as in-transit and proceed */ + else { + ret = rpcsvc_cache_request (req); + } + UNLOCK (&drc->lock); + } + if (req->rpc_err == SUCCESS) { /* Before going to xlator code, set the THIS properly */ THIS = svc->mydata; - actor_fn = actor->actor; + actor_fn = actor->actor; - if (!actor_fn) { - rpcsvc_request_seterr (req, PROC_UNAVAIL); - /* LOG TODO: print more info about procnum, - prognum etc, also print transport info */ - gf_log (GF_RPCSVC, GF_LOG_ERROR, - "No vectored handler present"); - ret = RPCSVC_ACTOR_ERROR; - goto err_reply; - } + if (!actor_fn) { + rpcsvc_request_seterr (req, PROC_UNAVAIL); + /* LOG TODO: print more info about procnum, + prognum etc, also print transport info */ + gf_log (GF_RPCSVC, GF_LOG_ERROR, + "No vectored handler present"); + ret = RPCSVC_ACTOR_ERROR; + goto err_reply; + } - if (req->synctask) { + if (req->synctask) { if (msg->hdr_iobuf) req->hdr_iobuf = iobuf_ref (msg->hdr_iobuf); - ret = synctask_new (THIS->ctx->env, - (synctask_fn_t) actor_fn, - rpcsvc_check_and_reply_error, NULL, + ret = synctask_new (THIS->ctx->env, + (synctask_fn_t) actor_fn, + rpcsvc_check_and_reply_error, NULL, req); } else { - ret = actor_fn (req); - req->hdr_iobuf = NULL; + ret = actor_fn (req); } } @@ -558,7 +639,7 @@ err_reply: * has now been queued. */ ret = 0; -err: +out: return ret; } @@ -905,21 +986,22 @@ out: return ret; } -static inline int -rpcsvc_transport_submit (rpc_transport_t *trans, struct iovec *hdrvec, - int hdrcount, struct iovec *proghdr, int proghdrcount, - struct iovec *progpayload, int progpayloadcount, - struct iobref *iobref, void *priv) +int +rpcsvc_transport_submit (rpc_transport_t *trans, struct iovec *rpchdr, + int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *progpayload, + int progpayloadcount, struct iobref *iobref, + void *priv) { int ret = -1; rpc_transport_reply_t reply = {{0, }}; - if ((!trans) || (!hdrvec) || (!hdrvec->iov_base)) { + if ((!trans) || (!rpchdr) || (!rpchdr->iov_base)) { goto out; } - reply.msg.rpchdr = hdrvec; - reply.msg.rpchdrcount = hdrcount; + reply.msg.rpchdr = rpchdr; + reply.msg.rpchdrcount = rpchdrcount; reply.msg.proghdr = proghdr; reply.msg.proghdrcount = proghdrcount; reply.msg.progpayload = progpayload; @@ -1065,6 +1147,7 @@ rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec *proghdr, size_t msglen = 0; size_t hdrlen = 0; char new_iobref = 0; + rpcsvc_drc_globals_t *drc = NULL; if ((!req) || (!req->trans)) return -1; @@ -1099,20 +1182,31 @@ rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec *proghdr, iobref_add (iobref, replyiob); + /* cache the request in the duplicate request cache for appropriate ops */ + if (req->reply) { + drc = req->svc->drc; + + LOCK (&drc->lock); + ret = rpcsvc_cache_reply (req, iobref, &recordhdr, 1, + proghdr, hdrcount, + payload, payloadcount); + UNLOCK (&drc->lock); + } + ret = rpcsvc_transport_submit (trans, &recordhdr, 1, proghdr, hdrcount, payload, payloadcount, iobref, req->trans_private); if (ret == -1) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "failed to submit message " - "(XID: 0x%ux, Program: %s, ProgVers: %d, Proc: %d) to " + "(XID: 0x%x, Program: %s, ProgVers: %d, Proc: %d) to " "rpc-transport (%s)", req->xid, req->prog ? req->prog->progname : "(not matched)", req->prog ? req->prog->progver : 0, req->procnum, trans->name); } else { gf_log (GF_RPCSVC, GF_LOG_TRACE, - "submitted reply for rpc-message (XID: 0x%ux, " + "submitted reply for rpc-message (XID: 0x%x, " "Program: %s, ProgVers: %d, Proc: %d) to rpc-transport " "(%s)", req->xid, req->prog ? req->prog->progname: "-", req->prog ? req->prog->progver : 0, @@ -1155,12 +1249,13 @@ rpcsvc_error_reply (rpcsvc_request_t *req) inline int rpcsvc_program_register_portmap (rpcsvc_program_t *newprog, uint32_t port) { - int ret = 0; + int ret = -1; /* FAIL */ if (!newprog) { goto out; } + /* pmap_set() returns 0 for FAIL and 1 for SUCCESS */ if (!(pmap_set (newprog->prognum, newprog->progver, IPPROTO_TCP, port))) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "Could not register with" @@ -1168,16 +1263,16 @@ rpcsvc_program_register_portmap (rpcsvc_program_t *newprog, uint32_t port) goto out; } - ret = 0; + ret = 0; /* SUCCESS */ out: return ret; } -static inline int +inline int rpcsvc_program_unregister_portmap (rpcsvc_program_t *prog) { - int ret = 0; + int ret = -1; if (!prog) goto out; @@ -1795,12 +1890,92 @@ rpcsvc_init_options (rpcsvc_t *svc, dict_t *options) gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Portmap registration " "disabled"); - ret = 0; + ret = rpcsvc_set_outstanding_rpc_limit (svc, options); out: return ret; } int +rpcsvc_reconfigure_options (rpcsvc_t *svc, dict_t *options) +{ + xlator_t *xlator = NULL; + xlator_list_t *volentry = NULL; + char *srchkey = NULL; + char *keyval = NULL; + int ret = -1; + + if ((!svc) || (!svc->options) || (!options)) + return (-1); + + /* Fetch the xlator from svc */ + xlator = (xlator_t *) svc->mydata; + if (!xlator) + return (-1); + + /* Reconfigure the volume specific rpc-auth.addr allow part */ + volentry = xlator->children; + while (volentry) { + ret = gf_asprintf (&srchkey, "rpc-auth.addr.%s.allow", + volentry->xlator->name); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return (-1); + } + + /* If found the srchkey, delete old key/val pair + * and set the key with new value. + */ + if (!dict_get_str (options, srchkey, &keyval)) { + dict_del (svc->options, srchkey); + ret = dict_set_str (svc->options, srchkey, keyval); + if (ret < 0) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, + "dict_set_str error"); + GF_FREE (srchkey); + return (-1); + } + } + + GF_FREE (srchkey); + volentry = volentry->next; + } + + /* Reconfigure the volume specific rpc-auth.addr reject part */ + volentry = xlator->children; + while (volentry) { + ret = gf_asprintf (&srchkey, "rpc-auth.addr.%s.reject", + volentry->xlator->name); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return (-1); + } + + /* If found the srchkey, delete old key/val pair + * and set the key with new value. + */ + if (!dict_get_str (options, srchkey, &keyval)) { + dict_del (svc->options, srchkey); + ret = dict_set_str (svc->options, srchkey, keyval); + if (ret < 0) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, + "dict_set_str error"); + GF_FREE (srchkey); + return (-1); + } + } + + GF_FREE (srchkey); + volentry = volentry->next; + } + + ret = rpcsvc_init_options (svc, options); + if (ret) + return (-1); + + return rpcsvc_auth_reconf (svc, options); +} + +int rpcsvc_transport_unix_options_build (dict_t **options, char *filepath) { dict_t *dict = NULL; @@ -1846,6 +2021,48 @@ out: return ret; } +/* + * Reconfigure() the rpc.outstanding-rpc-limit param. + */ +int +rpcsvc_set_outstanding_rpc_limit (rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; /* FAILURE */ + int rpclim = 0; + static char *rpclimkey = "rpc.outstanding-rpc-limit"; + + if ((!svc) || (!options)) + return (-1); + + /* Reconfigure() the rpc.outstanding-rpc-limit param */ + ret = dict_get_int32 (options, rpclimkey, &rpclim); + if (ret < 0) { + /* Fall back to default for FAILURE */ + rpclim = RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT; + } else { + /* SUCCESS: round off to multiple of 8. + * If the input value fails Boundary check, fall back to + * default i.e. RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT. + * NB: value 0 is special, means its unset i.e. unlimited. + */ + rpclim = ((rpclim + 8 - 1) >> 3) * 8; + if (rpclim < RPCSVC_MIN_OUTSTANDING_RPC_LIMIT) { + rpclim = RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT; + } else if (rpclim > RPCSVC_MAX_OUTSTANDING_RPC_LIMIT) { + rpclim = RPCSVC_MAX_OUTSTANDING_RPC_LIMIT; + } + } + + if (svc->outstanding_rpc_limit != rpclim) { + svc->outstanding_rpc_limit = rpclim; + gf_log (GF_RPCSVC, GF_LOG_INFO, + "Configured %s with value %d", + rpclimkey, rpclim); + } + + return (0); +} + /* The global RPC service initializer. */ rpcsvc_t * @@ -1906,6 +2123,7 @@ rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options, "failed to register DUMP program"); goto free_svc; } + ret = 0; free_svc: if (ret == -1) { @@ -1918,21 +2136,16 @@ free_svc: int -rpcsvc_transport_peer_check_search (dict_t *options, char *pattern, char *clstr) +rpcsvc_transport_peer_check_search (dict_t *options, char *pattern, + char *ip, char *hostname) { - int ret = -1; - char *addrtok = NULL; - char *addrstr = NULL; - char *dup_addrstr = NULL; - char *svptr = NULL; - char *fqdn = NULL; + int ret = -1; + char *addrtok = NULL; + char *addrstr = NULL; + char *dup_addrstr = NULL; + char *svptr = NULL; - if ((!options) || (!clstr)) - return -1; - - ret = dict_get_str (options, "fqdn", &fqdn); - - if (!dict_get (options, pattern)) + if ((!options) || (!ip)) return -1; ret = dict_get_str (options, pattern, &addrstr); @@ -1952,19 +2165,19 @@ rpcsvc_transport_peer_check_search (dict_t *options, char *pattern, char *clstr) /* CASEFOLD not present on Solaris */ #ifdef FNM_CASEFOLD - ret = fnmatch (addrtok, clstr, FNM_CASEFOLD); + ret = fnmatch (addrtok, ip, FNM_CASEFOLD); #else - ret = fnmatch (addrtok, clstr, 0); + ret = fnmatch (addrtok, ip, 0); #endif if (ret == 0) goto err; /* compare hostnames if applicable */ - if (fqdn) { + if (hostname) { #ifdef FNM_CASEFOLD - ret = fnmatch (addrtok, fqdn, FNM_CASEFOLD); + ret = fnmatch (addrtok, hostname, FNM_CASEFOLD); #else - ret = fnmatch (addrtok, fqdn, 0); + ret = fnmatch (addrtok, hostname, 0); #endif if (ret == 0) goto err; @@ -1981,66 +2194,56 @@ err: } -int -rpcsvc_transport_peer_check_allow (dict_t *options, char *volname, char *clstr) +static int +rpcsvc_transport_peer_check_allow (dict_t *options, char *volname, + char *ip, char *hostname) { - int ret = RPCSVC_AUTH_DONTCARE; + int ret = RPCSVC_AUTH_DONTCARE; char *srchstr = NULL; - char globalrule[] = "rpc-auth.addr.allow"; - if ((!options) || (!clstr)) + if ((!options) || (!ip) || (!volname)) return ret; - /* If volname is NULL, then we're searching for the general rule to - * determine the current address in clstr is allowed or not for all - * subvolumes. - */ - if (volname) { - ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.allow", volname); - if (ret == -1) { - gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); - ret = RPCSVC_AUTH_DONTCARE; - goto out; - } - } else - srchstr = globalrule; + ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_DONTCARE; + goto out; + } - ret = rpcsvc_transport_peer_check_search (options, srchstr, clstr); - if (volname) - GF_FREE (srchstr); + ret = rpcsvc_transport_peer_check_search (options, srchstr, + ip, hostname); + GF_FREE (srchstr); if (ret == 0) ret = RPCSVC_AUTH_ACCEPT; else - ret = RPCSVC_AUTH_DONTCARE; + ret = RPCSVC_AUTH_REJECT; out: return ret; } -int -rpcsvc_transport_peer_check_reject (dict_t *options, char *volname, char *clstr) +static int +rpcsvc_transport_peer_check_reject (dict_t *options, char *volname, + char *ip, char *hostname) { - int ret = RPCSVC_AUTH_DONTCARE; + int ret = RPCSVC_AUTH_DONTCARE; char *srchstr = NULL; - char generalrule[] = "rpc-auth.addr.reject"; - if ((!options) || (!clstr)) + if ((!options) || (!ip) || (!volname)) return ret; - if (volname) { - ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.reject", - volname); - if (ret == -1) { - gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); - ret = RPCSVC_AUTH_REJECT; - goto out; - } - } else - srchstr = generalrule; + ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.reject", + volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_REJECT; + goto out; + } - ret = rpcsvc_transport_peer_check_search (options, srchstr, clstr); - if (volname) - GF_FREE (srchstr); + ret = rpcsvc_transport_peer_check_search (options, srchstr, + ip, hostname); + GF_FREE (srchstr); if (ret == 0) ret = RPCSVC_AUTH_REJECT; @@ -2051,309 +2254,113 @@ out: } -/* This function tests the results of the allow rule and the reject rule to - * combine them into a single result that can be used to determine if the - * connection should be allowed to proceed. - * Heres the test matrix we need to follow in this function. - * - * A - Allow, the result of the allow test. Never returns R. - * R - Reject, result of the reject test. Never returns A. - * Both can return D or dont care if no rule was given. - * - * | @allow | @reject | Result | - * | A | R | R | - * | D | D | D | - * | A | D | A | - * | D | R | R | +/* Combines rpc auth's allow and reject options. + * Order of checks is important. + * First, REJECT if either rejects. + * If neither rejects, ACCEPT if either accepts. + * If neither accepts, DONTCARE */ int rpcsvc_combine_allow_reject_volume_check (int allow, int reject) { - int final = RPCSVC_AUTH_REJECT; - - /* If allowed rule allows but reject rule rejects, we stay cautious - * and reject. */ - if ((allow == RPCSVC_AUTH_ACCEPT) && (reject == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - /* if both are dont care, that is user did not specify for either allow - * or reject, we leave it up to the general rule to apply, in the hope - * that there is one. - */ - else if ((allow == RPCSVC_AUTH_DONTCARE) && - (reject == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_DONTCARE; - /* If one is dont care, the other one applies. */ - else if ((allow == RPCSVC_AUTH_ACCEPT) && - (reject == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_ACCEPT; - else if ((allow == RPCSVC_AUTH_DONTCARE) && - (reject == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - - return final; -} - - -/* Combines the result of the general rule test against, the specific rule - * to determine final permission for the client's address. - * - * | @gen | @spec | Result | - * | A | A | A | - * | A | R | R | - * | A | D | A | - * | D | A | A | - * | D | R | R | - * | D | D | D | - * | R | A | A | - * | R | D | R | - * | R | R | R | - */ -int -rpcsvc_combine_gen_spec_addr_checks (int gen, int spec) -{ - int final = RPCSVC_AUTH_REJECT; - - if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec== RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_DONTCARE; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - - return final; -} - + if (allow == RPCSVC_AUTH_REJECT || + reject == RPCSVC_AUTH_REJECT) + return RPCSVC_AUTH_REJECT; + if (allow == RPCSVC_AUTH_ACCEPT || + reject == RPCSVC_AUTH_ACCEPT) + return RPCSVC_AUTH_ACCEPT; -/* Combines the result of the general rule test against, the specific rule - * to determine final test for the connection coming in for a given volume. - * - * | @gen | @spec | Result | - * | A | A | A | - * | A | R | R | - * | A | D | A | - * | D | A | A | - * | D | R | R | - * | D | D | R |, special case, we intentionally disallow this. - * | R | A | A | - * | R | D | R | - * | R | R | R | - */ -int -rpcsvc_combine_gen_spec_volume_checks (int gen, int spec) -{ - int final = RPCSVC_AUTH_REJECT; - - if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_ACCEPT) && (spec == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - /* On no rule, we reject. */ - else if ((gen == RPCSVC_AUTH_DONTCARE) && (spec== RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_ACCEPT)) - final = RPCSVC_AUTH_ACCEPT; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_DONTCARE)) - final = RPCSVC_AUTH_REJECT; - else if ((gen == RPCSVC_AUTH_REJECT) && (spec == RPCSVC_AUTH_REJECT)) - final = RPCSVC_AUTH_REJECT; - - return final; + return RPCSVC_AUTH_DONTCARE; } - int -rpcsvc_transport_peer_check_name (dict_t *options, char *volname, - rpc_transport_t *trans) +rpcsvc_auth_check (rpcsvc_t *svc, char *volname, + rpc_transport_t *trans) { - int ret = RPCSVC_AUTH_REJECT; - int aret = RPCSVC_AUTH_REJECT; - int rjret = RPCSVC_AUTH_REJECT; - char clstr[RPCSVC_PEER_STRLEN]; - char *hostname = NULL; - - if (!trans) + int ret = RPCSVC_AUTH_REJECT; + int accept = RPCSVC_AUTH_REJECT; + int reject = RPCSVC_AUTH_REJECT; + char *hostname = NULL; + char *ip = NULL; + char client_ip[RPCSVC_PEER_STRLEN] = {0}; + char *allow_str = NULL; + char *reject_str = NULL; + char *srchstr = NULL; + dict_t *options = NULL; + + if (!svc || !volname || !trans) return ret; - ret = rpcsvc_transport_peername (trans, clstr, RPCSVC_PEER_STRLEN); - if (ret != 0) { - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get remote addr: " - "%s", gai_strerror (ret)); - ret = RPCSVC_AUTH_REJECT; - goto err; - } - - ret = gf_get_hostname_from_ip (clstr, &hostname); - if (!ret) - ret = dict_set_dynstr (options, "fqdn", - hostname); - - aret = rpcsvc_transport_peer_check_allow (options, volname, clstr); - rjret = rpcsvc_transport_peer_check_reject (options, volname, clstr); - - ret = rpcsvc_combine_allow_reject_volume_check (aret, rjret); - -err: - return ret; -} - - -int -rpcsvc_transport_peer_check_addr (dict_t *options, char *volname, - rpc_transport_t *trans) -{ - int ret = RPCSVC_AUTH_REJECT; - int aret = RPCSVC_AUTH_DONTCARE; - int rjret = RPCSVC_AUTH_REJECT; - char clstr[RPCSVC_PEER_STRLEN]; - char *tmp = NULL; - union gf_sock_union sock_union; - - if (!trans) + /* Fetch the options from svc struct and validate */ + options = svc->options; + if (!options) return ret; - ret = rpcsvc_transport_peeraddr (trans, clstr, RPCSVC_PEER_STRLEN, - &sock_union.storage, - sizeof (sock_union.storage)); + ret = rpcsvc_transport_peername (trans, client_ip, RPCSVC_PEER_STRLEN); if (ret != 0) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get remote addr: " "%s", gai_strerror (ret)); - ret = RPCSVC_AUTH_REJECT; - goto err; - } - - switch (sock_union.sa.sa_family) { - - case AF_INET: - case AF_INET6: - tmp = strrchr (clstr, ':'); - if (tmp) - *tmp = '\0'; - break; - } - - aret = rpcsvc_transport_peer_check_allow (options, volname, clstr); - rjret = rpcsvc_transport_peer_check_reject (options, volname, clstr); - - ret = rpcsvc_combine_allow_reject_volume_check (aret, rjret); -err: - return ret; -} - - -int -rpcsvc_transport_check_volume_specific (dict_t *options, char *volname, - rpc_transport_t *trans) -{ - int namechk = RPCSVC_AUTH_REJECT; - int addrchk = RPCSVC_AUTH_REJECT; - gf_boolean_t namelookup = _gf_false; - char *namestr = NULL; - int ret = 0; - - if ((!options) || (!volname) || (!trans)) return RPCSVC_AUTH_REJECT; - - /* Disabled by default */ - if ((dict_get (options, "rpc-auth.addr.namelookup"))) { - ret = dict_get_str (options, "rpc-auth.addr.namelookup" - , &namestr); - if (ret == 0) - ret = gf_string2boolean (namestr, &namelookup); } - /* We need two separate checks because the rules with addresses in them - * can be network addresses which can be general and names can be - * specific which will over-ride the network address rules. + /* Accept if its the default case: Allow all, Reject none + * The default volfile always contains a 'allow *' rule + * for each volume. If allow rule is missing (which implies + * there is some bad volfile generating code doing this), we + * assume no one is allowed mounts, and thus, we reject mounts. */ - if (namelookup) - namechk = rpcsvc_transport_peer_check_name (options, volname, - trans); - addrchk = rpcsvc_transport_peer_check_addr (options, volname, trans); - - if (namelookup) - ret = rpcsvc_combine_gen_spec_addr_checks (addrchk, - namechk); - else - ret = addrchk; - - return ret; -} - - -int -rpcsvc_transport_check_volume_general (dict_t *options, rpc_transport_t *trans) -{ - int addrchk = RPCSVC_AUTH_REJECT; - int namechk = RPCSVC_AUTH_REJECT; - gf_boolean_t namelookup = _gf_false; - char *namestr = NULL; - int ret = 0; - - if ((!options) || (!trans)) + ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); return RPCSVC_AUTH_REJECT; - - /* Disabled by default */ - if ((dict_get (options, "rpc-auth.addr.namelookup"))) { - ret = dict_get_str (options, "rpc-auth.addr.namelookup" - , &namestr); - if (ret == 0) - ret = gf_string2boolean (namestr, &namelookup); } - /* We need two separate checks because the rules with addresses in them - * can be network addresses which can be general and names can be - * specific which will over-ride the network address rules. - */ - if (namelookup) - namechk = rpcsvc_transport_peer_check_name (options, NULL, trans); - addrchk = rpcsvc_transport_peer_check_addr (options, NULL, trans); - - if (namelookup) - ret = rpcsvc_combine_gen_spec_addr_checks (addrchk, namechk); - else - ret = addrchk; + ret = dict_get_str (options, srchstr, &allow_str); + GF_FREE (srchstr); + if (ret < 0) + return RPCSVC_AUTH_REJECT; - return ret; -} + ret = gf_asprintf (&srchstr, "rpc-auth.addr.%s.reject", volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + return RPCSVC_AUTH_REJECT; + } -int -rpcsvc_transport_peer_check (dict_t *options, char *volname, - rpc_transport_t *trans) -{ - int general_chk = RPCSVC_AUTH_REJECT; - int specific_chk = RPCSVC_AUTH_REJECT; + ret = dict_get_str (options, srchstr, &reject_str); + GF_FREE (srchstr); + if (reject_str == NULL && !strcmp ("*", allow_str)) + return RPCSVC_AUTH_ACCEPT; + + /* Non-default rule, authenticate */ + if (!get_host_name (client_ip, &ip)) + ip = client_ip; + + /* addr-namelookup check */ + if (svc->addr_namelookup == _gf_true) { + ret = gf_get_hostname_from_ip (ip, &hostname); + if (ret) { + if (hostname) + GF_FREE (hostname); + /* failed to get hostname, but hostname auth + * is enabled, so authentication will not be + * 100% correct. reject mounts + */ + return RPCSVC_AUTH_REJECT; + } + } - if ((!options) || (!volname) || (!trans)) - return RPCSVC_AUTH_REJECT; + accept = rpcsvc_transport_peer_check_allow (options, volname, + ip, hostname); - general_chk = rpcsvc_transport_check_volume_general (options, trans); - specific_chk = rpcsvc_transport_check_volume_specific (options, volname, - trans); + reject = rpcsvc_transport_peer_check_reject (options, volname, + ip, hostname); - return rpcsvc_combine_gen_spec_volume_checks (general_chk, - specific_chk); + if (hostname) + GF_FREE (hostname); + return rpcsvc_combine_allow_reject_volume_check (accept, reject); } - int rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, rpc_transport_t *trans) @@ -2363,8 +2370,6 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, socklen_t sinsize = sizeof (&sock_union.sin); char *srchstr = NULL; char *valstr = NULL; - int globalinsecure = RPCSVC_AUTH_REJECT; - int exportinsecure = RPCSVC_AUTH_DONTCARE; uint16_t port = 0; gf_boolean_t insecure = _gf_false; @@ -2393,23 +2398,6 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, } /* Disabled by default */ - if ((dict_get (svc->options, "rpc-auth.ports.insecure"))) { - ret = dict_get_str (svc->options, "rpc-auth.ports.insecure" - , &srchstr); - if (ret == 0) { - ret = gf_string2boolean (srchstr, &insecure); - if (ret == 0) { - if (insecure == _gf_true) - globalinsecure = RPCSVC_AUTH_ACCEPT; - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" - " read rpc-auth.ports.insecure value"); - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" - " read rpc-auth.ports.insecure value"); - } - - /* Disabled by default */ ret = gf_asprintf (&srchstr, "rpc-auth.ports.%s.insecure", volname); if (ret == -1) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); @@ -2417,25 +2405,22 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, goto err; } - if (dict_get (svc->options, srchstr)) { - ret = dict_get_str (svc->options, srchstr, &valstr); - if (ret == 0) { - ret = gf_string2boolean (valstr, &insecure); - if (ret == 0) { - if (insecure == _gf_true) - exportinsecure = RPCSVC_AUTH_ACCEPT; - else - exportinsecure = RPCSVC_AUTH_REJECT; - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" - " read rpc-auth.ports.insecure value"); - } else - gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" - " read rpc-auth.ports.insecure value"); - } - - ret = rpcsvc_combine_gen_spec_volume_checks (globalinsecure, - exportinsecure); + ret = dict_get_str (svc->options, srchstr, &valstr); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" + " read rpc-auth.ports.insecure value"); + goto err; + } + + ret = gf_string2boolean (valstr, &insecure); + if (ret) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to" + " convert rpc-auth.ports.insecure value"); + goto err; + } + + ret = insecure ? RPCSVC_AUTH_ACCEPT : RPCSVC_AUTH_REJECT; + if (ret == RPCSVC_AUTH_ACCEPT) gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Unprivileged port allowed"); else @@ -2443,7 +2428,8 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, " allowed"); err: - GF_FREE (srchstr); + if (srchstr) + GF_FREE (srchstr); return ret; } @@ -2479,9 +2465,9 @@ out: rpcsvc_actor_t gluster_dump_actors[] = { - [GF_DUMP_NULL] = {"NULL", GF_DUMP_NULL, NULL, NULL, 0}, - [GF_DUMP_DUMP] = {"DUMP", GF_DUMP_DUMP, rpcsvc_dump, NULL, 0}, - [GF_DUMP_MAXVALUE] = {"MAXVALUE", GF_DUMP_MAXVALUE, NULL, NULL, 0}, + [GF_DUMP_NULL] = {"NULL", GF_DUMP_NULL, NULL, NULL, 0, DRC_NA}, + [GF_DUMP_DUMP] = {"DUMP", GF_DUMP_DUMP, rpcsvc_dump, NULL, 0, DRC_NA}, + [GF_DUMP_MAXVALUE] = {"MAXVALUE", GF_DUMP_MAXVALUE, NULL, NULL, 0, DRC_NA}, }; diff --git a/rpc/rpc-lib/src/rpcsvc.h b/rpc/rpc-lib/src/rpcsvc.h index 1323c8b7a..cbc1f4226 100644 --- a/rpc/rpc-lib/src/rpcsvc.h +++ b/rpc/rpc-lib/src/rpcsvc.h @@ -38,6 +38,10 @@ #define MAX_IOVEC 16 #endif +#define RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT 64 +#define RPCSVC_MAX_OUTSTANDING_RPC_LIMIT 65536 +#define RPCSVC_MIN_OUTSTANDING_RPC_LIMIT 0 /* No limit i.e. Unlimited */ + #define GF_RPCSVC "rpc-service" #define RPCSVC_THREAD_STACK_SIZE ((size_t)(1024 * GF_UNIT_KB)) @@ -140,6 +144,9 @@ typedef struct rpcsvc_auth_data { #define rpcsvc_auth_flavour(au) ((au).flavour) +typedef struct drc_client drc_client_t; +typedef struct drc_cached_op drc_cached_op_t; + /* The container for the RPC call handed up to an actor. * Dynamically allocated. Lives till the call reply is completely * transmitted. @@ -178,7 +185,9 @@ struct rpcsvc_request { /* Might want to move this to AUTH_UNIX specific state since this array * is not available for every authentication scheme. */ - gid_t auxgids[GF_MAX_AUX_GROUPS]; + gid_t *auxgids; + gid_t auxgidsmall[SMALL_GROUP_COUNT]; + gid_t *auxgidlarge; int auxgidcount; @@ -241,6 +250,9 @@ struct rpcsvc_request { /* we need to ref the 'iobuf' in case of 'synctasking' it */ struct iobuf *hdr_iobuf; + + /* pointer to cached reply for use in DRC */ + drc_cached_op_t *reply; }; #define rpcsvc_request_program(req) ((rpcsvc_program_t *)((req)->prog)) @@ -314,7 +326,6 @@ typedef void *(*rpcsvc_encode_reply) (void *msg); */ typedef void (*rpcsvc_deallocate_reply) (void *msg); - #define RPCSVC_NAME_MAX 32 /* The descriptor for each procedure/actor that runs * over the RPC service. @@ -336,6 +347,7 @@ typedef struct rpcsvc_actor_desc { /* Can actor be ran on behalf an unprivileged requestor? */ gf_boolean_t unprivileged; + drc_op_type_t op_type; } rpcsvc_actor_t; /* Describes a program and its version along with the function pointers @@ -429,6 +441,9 @@ extern int rpcsvc_program_register_portmap (rpcsvc_program_t *newprog, uint32_t port); extern int +rpcsvc_program_unregister_portmap (rpcsvc_program_t *newprog); + +extern int rpcsvc_register_portmap_enabled (rpcsvc_t *svc); /* Inits the global RPC service data structures. @@ -438,6 +453,9 @@ extern rpcsvc_t * rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options, uint32_t poolcount); +extern int +rpcsvc_reconfigure_options (rpcsvc_t *svc, dict_t *options); + int rpcsvc_register_notify (rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata); @@ -448,6 +466,13 @@ int rpcsvc_unregister_notify (rpcsvc_t *svc, rpcsvc_notify_t notify, void *mydata); int +rpcsvc_transport_submit (rpc_transport_t *trans, struct iovec *rpchdr, + int rpchdrcount, struct iovec *proghdr, + int proghdrcount, struct iovec *progpayload, + int progpayloadcount, struct iobref *iobref, + void *priv); + +int rpcsvc_submit_message (rpcsvc_request_t *req, struct iovec *proghdr, int hdrcount, struct iovec *payload, int payloadcount, struct iobref *iobref); @@ -473,12 +498,12 @@ rpcsvc_transport_peeraddr (rpc_transport_t *trans, char *addrstr, int addrlen, struct sockaddr_storage *returnsa, socklen_t sasize); extern int -rpcsvc_transport_peer_check (dict_t *options, char *volname, - rpc_transport_t *trans); +rpcsvc_auth_check (rpcsvc_t *svc, char *volname, rpc_transport_t *trans); extern int rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, rpc_transport_t *trans); + #define rpcsvc_request_seterr(req, err) (req)->rpc_err = err #define rpcsvc_request_set_autherr(req, err) (req)->auth_err = err @@ -534,6 +559,9 @@ extern int rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options); extern int +rpcsvc_auth_reconf (rpcsvc_t *svc, dict_t *options); + +extern int rpcsvc_auth_transport_init (rpc_transport_t *xprt); extern int @@ -550,9 +578,6 @@ rpcsvc_auth_array (rpcsvc_t *svc, char *volname, int *autharr, int arrlen); extern gid_t * rpcsvc_auth_unix_auxgids (rpcsvc_request_t *req, int *arrlen); -extern int -rpcsvc_combine_gen_spec_volume_checks (int gen, int spec); - extern char * rpcsvc_volume_allowed (dict_t *options, char *volname); @@ -560,18 +585,22 @@ int rpcsvc_callback_submit (rpcsvc_t *rpc, rpc_transport_t *trans, rpcsvc_cbk_program_t *prog, int procnum, struct iovec *proghdr, int proghdrcount); +rpcsvc_actor_t * +rpcsvc_program_actor (rpcsvc_request_t *req); + int rpcsvc_transport_unix_options_build (dict_t **options, char *filepath); int rpcsvc_set_allow_insecure (rpcsvc_t *svc, dict_t *options); int +rpcsvc_set_addr_namelookup (rpcsvc_t *svc, dict_t *options); +int rpcsvc_set_root_squash (rpcsvc_t *svc, dict_t *options); int +rpcsvc_set_outstanding_rpc_limit (rpcsvc_t *svc, dict_t *options); +int rpcsvc_auth_array (rpcsvc_t *svc, char *volname, int *autharr, int arrlen); -char * -rpcsvc_volume_allowed (dict_t *options, char *volname); rpcsvc_vector_sizer rpcsvc_get_program_vector_sizer (rpcsvc_t *svc, uint32_t prognum, uint32_t progver, uint32_t procnum); - #endif diff --git a/rpc/rpc-lib/src/xdr-rpc.c b/rpc/rpc-lib/src/xdr-rpc.c index ef52764c3..adb48a531 100644 --- a/rpc/rpc-lib/src/xdr-rpc.c +++ b/rpc/rpc-lib/src/xdr-rpc.c @@ -34,7 +34,7 @@ xdr_to_rpc_call (char *msgbuf, size_t len, struct rpc_msg *call, struct iovec *payload, char *credbytes, char *verfbytes) { XDR xdr; - char opaquebytes[MAX_AUTH_BYTES]; + char opaquebytes[GF_MAX_AUTH_BYTES]; struct opaque_auth *oa = NULL; int ret = -1; diff --git a/rpc/rpc-transport/rdma/src/Makefile.am b/rpc/rpc-transport/rdma/src/Makefile.am index 817925dec..2bf7cf238 100644 --- a/rpc/rpc-transport/rdma/src/Makefile.am +++ b/rpc/rpc-transport/rdma/src/Makefile.am @@ -7,7 +7,7 @@ rdma_la_LDFLAGS = -module -avoid-version rdma_la_SOURCES = rdma.c name.c rdma_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la \ - -libverbs + -libverbs -lrdmacm noinst_HEADERS = rdma.h name.h -I$(top_srcdir)/libglusterfs/src -I$(top_srcdir)/rpc/rpc-lib/src/ \ diff --git a/rpc/rpc-transport/rdma/src/name.c b/rpc/rpc-transport/rdma/src/name.c index f6ae818b1..c57428ad6 100644 --- a/rpc/rpc-transport/rdma/src/name.c +++ b/rpc/rpc-transport/rdma/src/name.c @@ -13,6 +13,7 @@ #include <errno.h> #include <netdb.h> #include <string.h> +#include <rdma/rdma_cma.h> #ifndef AF_INET_SDP #define AF_INET_SDP 27 @@ -31,7 +32,8 @@ gf_resolve_ip6 (const char *hostname, struct addrinfo **addr_info); static int32_t -af_inet_bind_to_port_lt_ceiling (int fd, struct sockaddr *sockaddr, +af_inet_bind_to_port_lt_ceiling (struct rdma_cm_id *cm_id, + struct sockaddr *sockaddr, socklen_t sockaddr_len, int ceiling) { int32_t ret = -1; @@ -51,12 +53,14 @@ af_inet_bind_to_port_lt_ceiling (int fd, struct sockaddr *sockaddr, switch (sockaddr->sa_family) { case AF_INET6: - ((struct sockaddr_in6 *)sockaddr)->sin6_port = htons (port); + ((struct sockaddr_in6 *)sockaddr)->sin6_port + = htons (port); break; case AF_INET_SDP: case AF_INET: - ((struct sockaddr_in *)sockaddr)->sin_port = htons (port); + ((struct sockaddr_in *)sockaddr)->sin_port + = htons (port); break; } // ignore the reserved ports @@ -64,7 +68,7 @@ af_inet_bind_to_port_lt_ceiling (int fd, struct sockaddr *sockaddr, port--; continue; } - ret = bind (fd, sockaddr, sockaddr_len); + ret = rdma_bind_addr (cm_id, sockaddr); if (ret == 0) break; @@ -78,11 +82,10 @@ af_inet_bind_to_port_lt_ceiling (int fd, struct sockaddr *sockaddr, return ret; } +#if 0 static int32_t -af_unix_client_bind (rpc_transport_t *this, - struct sockaddr *sockaddr, - socklen_t sockaddr_len, - int sock) +af_unix_client_bind (rpc_transport_t *this, struct sockaddr *sockaddr, + socklen_t sockaddr_len, struct rdma_cm_id *cm_id) { data_t *path_data = NULL; struct sockaddr_un *addr = NULL; @@ -114,6 +117,7 @@ af_unix_client_bind (rpc_transport_t *this, err: return ret; } +#endif static int32_t client_fill_address_family (rpc_transport_t *this, struct sockaddr *sockaddr) @@ -412,10 +416,8 @@ out: } int32_t -gf_rdma_client_bind (rpc_transport_t *this, - struct sockaddr *sockaddr, - socklen_t *sockaddr_len, - int sock) +gf_rdma_client_bind (rpc_transport_t *this, struct sockaddr *sockaddr, + socklen_t *sockaddr_len, struct rdma_cm_id *cm_id) { int ret = 0; @@ -427,22 +429,24 @@ gf_rdma_client_bind (rpc_transport_t *this, *sockaddr_len = sizeof (struct sockaddr_in); case AF_INET6: - ret = af_inet_bind_to_port_lt_ceiling (sock, sockaddr, + ret = af_inet_bind_to_port_lt_ceiling (cm_id, sockaddr, *sockaddr_len, GF_CLIENT_PORT_CEILING); if (ret == -1) { gf_log (this->name, GF_LOG_WARNING, - "cannot bind inet socket (%d) to port " - "less than %d (%s)", - sock, GF_CLIENT_PORT_CEILING, strerror (errno)); + "cannot bind rdma_cm_id to port " + "less than %d (%s)", GF_CLIENT_PORT_CEILING, + strerror (errno)); ret = 0; } break; case AF_UNIX: *sockaddr_len = sizeof (struct sockaddr_un); +#if 0 ret = af_unix_client_bind (this, (struct sockaddr *)sockaddr, *sockaddr_len, sock); +#endif break; default: diff --git a/rpc/rpc-transport/rdma/src/name.h b/rpc/rpc-transport/rdma/src/name.h index 114ed1661..742fc5fc3 100644 --- a/rpc/rpc-transport/rdma/src/name.h +++ b/rpc/rpc-transport/rdma/src/name.h @@ -11,16 +11,13 @@ #ifndef _IB_VERBS_NAME_H #define _IB_VERBS_NAME_H -#include <sys/socket.h> -#include <sys/un.h> +#include <rdma/rdma_cma.h> #include "compat.h" int32_t -gf_rdma_client_bind (rpc_transport_t *this, - struct sockaddr *sockaddr, - socklen_t *sockaddr_len, - int sock); +gf_rdma_client_bind (rpc_transport_t *this, struct sockaddr *sockaddr, + socklen_t *sockaddr_len, struct rdma_cm_id *cm_id); int32_t gf_rdma_client_get_remote_sockaddr (rpc_transport_t *this, diff --git a/rpc/rpc-transport/rdma/src/rdma.c b/rpc/rpc-transport/rdma/src/rdma.c index a44e8995f..6e6099a98 100644 --- a/rpc/rpc-transport/rdma/src/rdma.c +++ b/rpc/rpc-transport/rdma/src/rdma.c @@ -8,7 +8,6 @@ cases as published by the Free Software Foundation. */ - #ifndef _CONFIG_H #define _CONFIG_H #include "config.h" @@ -35,99 +34,29 @@ gf_rdma_post_ref (gf_rdma_post_t *post); int gf_rdma_post_unref (gf_rdma_post_t *post); -int32_t -gf_resolve_ip6 (const char *hostname, - uint16_t port, - int family, - void **dnscache, - struct addrinfo **addr_info); - -static uint16_t -gf_rdma_get_local_lid (struct ibv_context *context, - int32_t port) -{ - struct ibv_port_attr attr; - - if (ibv_query_port (context, port, &attr)) - return 0; +static void * +gf_rdma_send_completion_proc (void *data); - return attr.lid; -} +static void * +gf_rdma_recv_completion_proc (void *data); -static const char * -get_port_state_str(enum ibv_port_state pstate) -{ - switch (pstate) { - case IBV_PORT_DOWN: return "PORT_DOWN"; - case IBV_PORT_INIT: return "PORT_INIT"; - case IBV_PORT_ARMED: return "PORT_ARMED"; - case IBV_PORT_ACTIVE: return "PORT_ACTIVE"; - case IBV_PORT_ACTIVE_DEFER: return "PORT_ACTIVE_DEFER"; - default: return "invalid state"; - } -} +void * +gf_rdma_async_event_thread (void *context); static int32_t -ib_check_active_port (struct ibv_context *ctx, uint8_t port) -{ - struct ibv_port_attr port_attr = {0, }; - int32_t ret = 0; - const char *state_str = NULL; - - if (!ctx) { - gf_log_callingfn (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "Error in supplied context"); - return -1; - } - - ret = ibv_query_port (ctx, port, &port_attr); - - if (ret) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "Failed to query port %u properties", port); - return -1; - } - - state_str = get_port_state_str (port_attr.state); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_TRACE, - "Infiniband PORT: (%u) STATE: (%s)", - port, state_str); - - if (port_attr.state == IBV_PORT_ACTIVE) - return 0; - - return -1; -} +gf_rdma_create_qp (rpc_transport_t *this); static int32_t -ib_get_active_port (struct ibv_context *ib_ctx) -{ - struct ibv_device_attr ib_device_attr = {{0, }, }; - int32_t ret = -1; - uint8_t ib_port = 0; +__gf_rdma_teardown (rpc_transport_t *this); - if (!ib_ctx) { - gf_log_callingfn (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "Error in supplied context"); - return -1; - } - if (ibv_query_device (ib_ctx, &ib_device_attr)) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "Failed to query device properties"); - return -1; - } +static int32_t +gf_rdma_teardown (rpc_transport_t *this); - for (ib_port = 1; ib_port <= ib_device_attr.phys_port_cnt; ++ib_port) { - ret = ib_check_active_port (ib_ctx, ib_port); - if (ret == 0) - return ib_port; +static int32_t +gf_rdma_disconnect (rpc_transport_t *this); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_TRACE, - "Port:(%u) not active", ib_port); - continue; - } - return ret; -} +static void +gf_rdma_cm_handle_disconnect (rpc_transport_t *this); static void @@ -157,7 +86,7 @@ gf_rdma_put_post (gf_rdma_queue_t *queue, gf_rdma_post_t *post) static gf_rdma_post_t * -gf_rdma_new_post (gf_rdma_device_t *device, int32_t len, +gf_rdma_new_post (rpc_transport_t *this, gf_rdma_device_t *device, int32_t len, gf_rdma_post_type_t type) { gf_rdma_post_t *post = NULL; @@ -184,7 +113,7 @@ gf_rdma_new_post (gf_rdma_device_t *device, int32_t len, post->buf_size, IBV_ACCESS_LOCAL_WRITE); if (!post->mr) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + gf_log (this->name, GF_LOG_WARNING, "memory registration failed (%s)", strerror (errno)); goto out; @@ -259,22 +188,6 @@ __gf_rdma_quota_get (gf_rdma_peer_t *peer) return ret; } -/* - static int32_t - gf_rdma_quota_get (gf_rdma_peer_t *peer) - { - int32_t ret = -1; - gf_rdma_private_t *priv = peer->trans->private; - - pthread_mutex_lock (&priv->write_mutex); - { - ret = __gf_rdma_quota_get (peer); - } - pthread_mutex_unlock (&priv->write_mutex); - - return ret; - } -*/ static void __gf_rdma_ioq_entry_free (gf_rdma_ioq_t *entry) @@ -290,6 +203,7 @@ __gf_rdma_ioq_entry_free (gf_rdma_ioq_t *entry) iobref_unref (entry->msg.request.rsp_iobref); entry->msg.request.rsp_iobref = NULL; } + mem_put (entry); } @@ -309,26 +223,898 @@ static int32_t __gf_rdma_disconnect (rpc_transport_t *this) { gf_rdma_private_t *priv = NULL; - int32_t ret = 0; priv = this->private; - if (priv->connected || priv->tcp_connected) { - fcntl (priv->sock, F_SETFL, O_NONBLOCK); - if (shutdown (priv->sock, SHUT_RDWR) != 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_DEBUG, - "shutdown () - error: %s", - strerror (errno)); - ret = -errno; - priv->tcp_connected = 0; + if (priv->connected) { + rdma_disconnect (priv->peer.cm_id); + } + + return 0; +} + + +static void +gf_rdma_queue_init (gf_rdma_queue_t *queue) +{ + pthread_mutex_init (&queue->lock, NULL); + + queue->active_posts.next = &queue->active_posts; + queue->active_posts.prev = &queue->active_posts; + queue->passive_posts.next = &queue->passive_posts; + queue->passive_posts.prev = &queue->passive_posts; +} + + +static void +__gf_rdma_destroy_queue (gf_rdma_post_t *post) +{ + gf_rdma_post_t *tmp = NULL; + + while (post->next != post) { + tmp = post->next; + + post->next = post->next->next; + post->next->prev = post; + + gf_rdma_destroy_post (tmp); + } +} + + +static void +gf_rdma_destroy_queue (gf_rdma_queue_t *queue) +{ + if (queue == NULL) { + goto out; + } + + pthread_mutex_lock (&queue->lock); + { + if (queue->passive_count > 0) { + __gf_rdma_destroy_queue (&queue->passive_posts); + queue->passive_count = 0; + } + + if (queue->active_count > 0) { + __gf_rdma_destroy_queue (&queue->active_posts); + queue->active_count = 0; + } + } + pthread_mutex_unlock (&queue->lock); + +out: + return; +} + + +static void +gf_rdma_destroy_posts (rpc_transport_t *this) +{ + gf_rdma_device_t *device = NULL; + gf_rdma_private_t *priv = NULL; + + if (this == NULL) { + goto out; + } + + priv = this->private; + device = priv->device; + + gf_rdma_destroy_queue (&device->sendq); + gf_rdma_destroy_queue (&device->recvq); + +out: + return; +} + + +static int32_t +__gf_rdma_create_posts (rpc_transport_t *this, int32_t count, int32_t size, + gf_rdma_queue_t *q, gf_rdma_post_type_t type) +{ + int32_t i = 0; + int32_t ret = 0; + gf_rdma_private_t *priv = NULL; + gf_rdma_device_t *device = NULL; + + priv = this->private; + device = priv->device; + + for (i=0 ; i<count ; i++) { + gf_rdma_post_t *post = NULL; + + post = gf_rdma_new_post (this, device, size + 2048, type); + if (!post) { + gf_log (this->name, GF_LOG_ERROR, + "post creation failed"); + ret = -1; + break; + } + + gf_rdma_put_post (q, post); + } + return ret; +} + + +static int32_t +gf_rdma_post_recv (struct ibv_srq *srq, + gf_rdma_post_t *post) +{ + struct ibv_sge list = { + .addr = (unsigned long) post->buf, + .length = post->buf_size, + .lkey = post->mr->lkey + }; + + struct ibv_recv_wr wr = { + .wr_id = (unsigned long) post, + .sg_list = &list, + .num_sge = 1, + }, *bad_wr; + + gf_rdma_post_ref (post); + + return ibv_post_srq_recv (srq, &wr, &bad_wr); +} + + +static int32_t +gf_rdma_create_posts (rpc_transport_t *this) +{ + int32_t i = 0, ret = 0; + gf_rdma_post_t *post = NULL; + gf_rdma_private_t *priv = NULL; + gf_rdma_options_t *options = NULL; + gf_rdma_device_t *device = NULL; + + priv = this->private; + options = &priv->options; + device = priv->device; + + ret = __gf_rdma_create_posts (this, options->send_count, + options->send_size, + &device->sendq, GF_RDMA_SEND_POST); + if (!ret) + ret = __gf_rdma_create_posts (this, options->recv_count, + options->recv_size, + &device->recvq, + GF_RDMA_RECV_POST); + + if (!ret) { + for (i=0 ; i<options->recv_count ; i++) { + post = gf_rdma_get_post (&device->recvq); + if (gf_rdma_post_recv (device->srq, post) != 0) { + ret = -1; + break; + } + } + } + + if (ret) + gf_rdma_destroy_posts (this); + + return ret; +} + + +static void +gf_rdma_destroy_cq (rpc_transport_t *this) +{ + gf_rdma_private_t *priv = NULL; + gf_rdma_device_t *device = NULL; + + priv = this->private; + device = priv->device; + + if (device->recv_cq) + ibv_destroy_cq (device->recv_cq); + device->recv_cq = NULL; + + if (device->send_cq) + ibv_destroy_cq (device->send_cq); + device->send_cq = NULL; + + return; +} + + +static int32_t +gf_rdma_create_cq (rpc_transport_t *this) +{ + gf_rdma_private_t *priv = NULL; + gf_rdma_options_t *options = NULL; + gf_rdma_device_t *device = NULL; + uint64_t send_cqe = 0; + int32_t ret = 0; + struct ibv_device_attr device_attr = {{0}, }; + + priv = this->private; + options = &priv->options; + device = priv->device; + + device->recv_cq = ibv_create_cq (priv->device->context, + options->recv_count * 2, + device, + device->recv_chan, + 0); + if (!device->recv_cq) { + gf_log (this->name, GF_LOG_ERROR, + "creation of CQ for device %s failed", + device->device_name); + ret = -1; + goto out; + } else if (ibv_req_notify_cq (device->recv_cq, 0)) { + gf_log (this->name, GF_LOG_ERROR, + "ibv_req_notify_cq on recv CQ of device %s failed", + device->device_name); + ret = -1; + goto out; + } + + do { + ret = ibv_query_device (priv->device->context, &device_attr); + if (ret != 0) { + gf_log (this->name, GF_LOG_ERROR, + "ibv_query_device on %s returned %d (%s)", + priv->device->device_name, ret, + (ret > 0) ? strerror (ret) : ""); + ret = -1; + goto out; + } + + send_cqe = options->send_count * 128; + send_cqe = (send_cqe > device_attr.max_cqe) + ? device_attr.max_cqe : send_cqe; + + /* TODO: make send_cq size dynamically adaptive */ + device->send_cq = ibv_create_cq (priv->device->context, + send_cqe, device, + device->send_chan, 0); + if (!device->send_cq) { + gf_log (this->name, GF_LOG_ERROR, + "creation of send_cq for device %s failed", + device->device_name); + ret = -1; + goto out; + } + + if (ibv_req_notify_cq (device->send_cq, 0)) { + gf_log (this->name, GF_LOG_ERROR, + "ibv_req_notify_cq on send_cq for device %s" + " failed", device->device_name); + ret = -1; + goto out; + } + } while (0); + +out: + if (ret != 0) + gf_rdma_destroy_cq (this); + + return ret; +} + + +static gf_rdma_device_t * +gf_rdma_get_device (rpc_transport_t *this, struct ibv_context *ibctx, + char *device_name) +{ + glusterfs_ctx_t *ctx = NULL; + gf_rdma_private_t *priv = NULL; + gf_rdma_options_t *options = NULL; + int32_t ret = 0; + int32_t i = 0; + gf_rdma_device_t *trav = NULL, *device = NULL; + gf_rdma_ctx_t *rdma_ctx = NULL; + + priv = this->private; + options = &priv->options; + ctx = this->ctx; + rdma_ctx = ctx->ib; + + trav = rdma_ctx->device; + + while (trav) { + if (!strcmp (trav->device_name, device_name)) + break; + trav = trav->next; + } + + if (!trav) { + trav = GF_CALLOC (1, sizeof (*trav), + gf_common_mt_rdma_device_t); + if (trav == NULL) { + goto out; + } + + priv->device = trav; + trav->context = ibctx; + + trav->request_ctx_pool + = mem_pool_new (gf_rdma_request_context_t, + GF_RDMA_POOL_SIZE); + if (trav->request_ctx_pool == NULL) { + goto out; + } + + trav->ioq_pool + = mem_pool_new (gf_rdma_ioq_t, GF_RDMA_POOL_SIZE); + if (trav->ioq_pool == NULL) { + goto out; + } + + trav->reply_info_pool = mem_pool_new (gf_rdma_reply_info_t, + GF_RDMA_POOL_SIZE); + if (trav->reply_info_pool == NULL) { + goto out; + } + + trav->device_name = gf_strdup (device_name); + + trav->next = rdma_ctx->device; + rdma_ctx->device = trav; + + trav->send_chan = ibv_create_comp_channel (trav->context); + if (!trav->send_chan) { + gf_log (this->name, GF_LOG_ERROR, + "could not create send completion channel for " + "device (%s)", device_name); + goto out; + } + + trav->recv_chan = ibv_create_comp_channel (trav->context); + if (!trav->recv_chan) { + gf_log (this->name, GF_LOG_ERROR, + "could not create recv completion channel for " + "device (%s)", device_name); + + /* TODO: cleanup current mess */ + goto out; + } + + if (gf_rdma_create_cq (this) < 0) { + gf_log (this->name, GF_LOG_ERROR, + "could not create CQ for device (%s)", + device_name); + goto out; + } + + /* protection domain */ + trav->pd = ibv_alloc_pd (trav->context); + + if (!trav->pd) { + gf_log (this->name, GF_LOG_ERROR, + "could not allocate protection domain for " + "device (%s)", device_name); + goto out; + } + + struct ibv_srq_init_attr attr = { + .attr = { + .max_wr = options->recv_count, + .max_sge = 1, + .srq_limit = 10 + } + }; + trav->srq = ibv_create_srq (trav->pd, &attr); + + if (!trav->srq) { + gf_log (this->name, GF_LOG_ERROR, + "could not create SRQ for device (%s)", + device_name); + goto out; + } + + /* queue init */ + gf_rdma_queue_init (&trav->sendq); + gf_rdma_queue_init (&trav->recvq); + + if (gf_rdma_create_posts (this) < 0) { + gf_log (this->name, GF_LOG_ERROR, + "could not allocate posts for device (%s)", + device_name); + goto out; + } + + /* completion threads */ + ret = gf_thread_create (&trav->send_thread, NULL, + gf_rdma_send_completion_proc, + trav->send_chan); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "could not create send completion thread for " + "device (%s)", device_name); + goto out; + } + + ret = gf_thread_create (&trav->recv_thread, NULL, + gf_rdma_recv_completion_proc, + trav->recv_chan); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "could not create recv completion thread " + "for device (%s)", device_name); + return NULL; + } + + ret = gf_thread_create (&trav->async_event_thread, NULL, + gf_rdma_async_event_thread, + ibctx); + if (ret) { + gf_log (this->name, GF_LOG_ERROR, + "could not create async_event_thread"); + return NULL; + } + + /* qpreg */ + pthread_mutex_init (&trav->qpreg.lock, NULL); + for (i=0; i<42; i++) { + trav->qpreg.ents[i].next = &trav->qpreg.ents[i]; + trav->qpreg.ents[i].prev = &trav->qpreg.ents[i]; + } + } + + device = trav; + trav = NULL; +out: + + if (trav != NULL) { + gf_rdma_destroy_posts (this); + mem_pool_destroy (trav->ioq_pool); + mem_pool_destroy (trav->request_ctx_pool); + mem_pool_destroy (trav->reply_info_pool); + ibv_dealloc_pd (trav->pd); + gf_rdma_destroy_cq (this); + ibv_destroy_comp_channel (trav->recv_chan); + ibv_destroy_comp_channel (trav->send_chan); + GF_FREE ((char *)trav->device_name); + GF_FREE (trav); + } + + return device; +} + + +static rpc_transport_t * +gf_rdma_transport_new (rpc_transport_t *listener, struct rdma_cm_id *cm_id) +{ + gf_rdma_private_t *listener_priv = NULL, *priv = NULL; + rpc_transport_t *this = NULL, *new = NULL; + gf_rdma_options_t *options = NULL; + char *device_name = NULL; + + listener_priv = listener->private; + + this = GF_CALLOC (1, sizeof (rpc_transport_t), + gf_common_mt_rpc_transport_t); + if (this == NULL) { + goto out; + } + + this->listener = listener; + + priv = GF_CALLOC (1, sizeof (gf_rdma_private_t), + gf_common_mt_rdma_private_t); + if (priv == NULL) { + goto out; + } + + this->private = priv; + priv->options = listener_priv->options; + + priv->listener = listener; + priv->entity = GF_RDMA_SERVER; + + options = &priv->options; + + this->ops = listener->ops; + this->init = listener->init; + this->fini = listener->fini; + this->ctx = listener->ctx; + this->name = gf_strdup (listener->name); + this->notify = listener->notify; + this->mydata = listener->mydata; + + this->myinfo.sockaddr_len = sizeof (cm_id->route.addr.src_addr); + memcpy (&this->myinfo.sockaddr, &cm_id->route.addr.src_addr, + this->myinfo.sockaddr_len); + + this->peerinfo.sockaddr_len = sizeof (cm_id->route.addr.dst_addr); + memcpy (&this->peerinfo.sockaddr, &cm_id->route.addr.dst_addr, + this->peerinfo.sockaddr_len); + + priv->peer.trans = this; + gf_rdma_get_transport_identifiers (this); + + device_name = (char *)ibv_get_device_name (cm_id->verbs->device); + if (device_name == NULL) { + gf_log (listener->name, GF_LOG_WARNING, + "cannot get device name (peer:%s me:%s)", + this->peerinfo.identifier, this->myinfo.identifier); + goto out; + } + + priv->device = gf_rdma_get_device (this, cm_id->verbs, + device_name); + if (priv->device == NULL) { + gf_log (listener->name, GF_LOG_WARNING, + "cannot get infiniband device %s (peer:%s me:%s)", + device_name, this->peerinfo.identifier, + this->myinfo.identifier); + goto out; + } + + priv->peer.send_count = options->send_count; + priv->peer.recv_count = options->recv_count; + priv->peer.send_size = options->send_size; + priv->peer.recv_size = options->recv_size; + priv->peer.cm_id = cm_id; + INIT_LIST_HEAD (&priv->peer.ioq); + + pthread_mutex_init (&priv->write_mutex, NULL); + pthread_mutex_init (&priv->recv_mutex, NULL); + + cm_id->context = this; + + new = rpc_transport_ref (this); + this = NULL; +out: + if (this != NULL) { + if (this->private != NULL) { + GF_FREE (this->private); + } + + if (this->name != NULL) { + GF_FREE (this->name); + } + + GF_FREE (this); + } + + return new; +} + + +static int +gf_rdma_cm_handle_connect_request (struct rdma_cm_event *event) +{ + int ret = -1; + rpc_transport_t *this = NULL, *listener = NULL; + struct rdma_cm_id *child_cm_id = NULL, *listener_cm_id = NULL; + struct rdma_conn_param conn_param = {0, }; + gf_rdma_private_t *priv = NULL; + gf_rdma_options_t *options = NULL; + + child_cm_id = event->id; + listener_cm_id = event->listen_id; + + listener = listener_cm_id->context; + priv = listener->private; + options = &priv->options; + + this = gf_rdma_transport_new (listener, child_cm_id); + if (this == NULL) { + gf_log (listener->name, GF_LOG_WARNING, + "could not create a transport for incoming connection" + " (me.name:%s me.identifier:%s)", listener->name, + listener->myinfo.identifier); + rdma_destroy_id (child_cm_id); + goto out; + } + + gf_log (listener->name, GF_LOG_TRACE, + "got a connect request (me:%s peer:%s)", + listener->myinfo.identifier, this->peerinfo.identifier); + + ret = gf_rdma_create_qp (this); + if (ret < 0) { + gf_log (listener->name, GF_LOG_WARNING, + "could not create QP (peer:%s me:%s)", + this->peerinfo.identifier, this->myinfo.identifier); + gf_rdma_cm_handle_disconnect (this); + goto out; + } + + conn_param.responder_resources = 1; + conn_param.initiator_depth = 1; + conn_param.retry_count = options->attr_retry_cnt; + conn_param.rnr_retry_count = options->attr_rnr_retry; + + ret = rdma_accept(child_cm_id, &conn_param); + if (ret < 0) { + gf_log (listener->name, GF_LOG_WARNING, "rdma_accept failed " + "peer:%s me:%s (%s)", this->peerinfo.identifier, + this->myinfo.identifier, strerror (errno)); + gf_rdma_cm_handle_disconnect (this); + goto out; + } + + ret = 0; + +out: + return ret; +} + + +static int +gf_rdma_cm_handle_route_resolved (struct rdma_cm_event *event) +{ + struct rdma_conn_param conn_param = {0, }; + int ret = 0; + rpc_transport_t *this = NULL; + gf_rdma_private_t *priv = NULL; + gf_rdma_peer_t *peer = NULL; + gf_rdma_options_t *options = NULL; + + if (event == NULL) { + goto out; + } + + this = event->id->context; + + priv = this->private; + peer = &priv->peer; + options = &priv->options; + + ret = gf_rdma_create_qp (this); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "could not create QP (peer:%s me:%s)", + this->peerinfo.identifier, this->myinfo.identifier); + gf_rdma_cm_handle_disconnect (this); + goto out; + } + + memset(&conn_param, 0, sizeof conn_param); + conn_param.responder_resources = 1; + conn_param.initiator_depth = 1; + conn_param.retry_count = options->attr_retry_cnt; + conn_param.rnr_retry_count = options->attr_rnr_retry; + + ret = rdma_connect(peer->cm_id, &conn_param); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "rdma_connect failed (%s)", strerror (errno)); + gf_rdma_cm_handle_disconnect (this); + goto out; + } + + gf_log (this->name, GF_LOG_TRACE, "route resolved (me:%s peer:%s)", + this->myinfo.identifier, this->peerinfo.identifier); + + ret = 0; +out: + return ret; +} + + +static int +gf_rdma_cm_handle_addr_resolved (struct rdma_cm_event *event) +{ + rpc_transport_t *this = NULL; + gf_rdma_peer_t *peer = NULL; + gf_rdma_private_t *priv = NULL; + int ret = 0; + + this = event->id->context; + + priv = this->private; + peer = &priv->peer; + + GF_ASSERT (peer->cm_id == event->id); + + this->myinfo.sockaddr_len = sizeof (peer->cm_id->route.addr.src_addr); + memcpy (&this->myinfo.sockaddr, &peer->cm_id->route.addr.src_addr, + this->myinfo.sockaddr_len); + + this->peerinfo.sockaddr_len = sizeof (peer->cm_id->route.addr.dst_addr); + memcpy (&this->peerinfo.sockaddr, &peer->cm_id->route.addr.dst_addr, + this->peerinfo.sockaddr_len); + + gf_rdma_get_transport_identifiers (this); + + ret = rdma_resolve_route(peer->cm_id, 2000); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "rdma_resolve_route failed (me:%s peer:%s) (%s)", + this->myinfo.identifier, this->peerinfo.identifier, + strerror (errno)); + gf_rdma_cm_handle_disconnect (this); + } + + gf_log (this->name, GF_LOG_TRACE, "Address resolved (me:%s peer:%s)", + this->myinfo.identifier, this->peerinfo.identifier); + + return ret; +} + + +static void +gf_rdma_cm_handle_disconnect (rpc_transport_t *this) +{ + gf_rdma_private_t *priv = NULL; + char need_unref = 0, connected = 0; + + priv = this->private; + gf_log (this->name, GF_LOG_DEBUG, + "peer disconnected, cleaning up"); + + pthread_mutex_lock (&priv->write_mutex); + { + if (priv->peer.cm_id != NULL) { + need_unref = 1; + connected = priv->connected; priv->connected = 0; } + + __gf_rdma_teardown (this); + } + pthread_mutex_unlock (&priv->write_mutex); + + if (connected) { + rpc_transport_notify (this, RPC_TRANSPORT_DISCONNECT, this); + } + + if (need_unref) + rpc_transport_unref (this); + +} + + +static int +gf_rdma_cm_handle_event_established (struct rdma_cm_event *event) +{ + rpc_transport_t *this = NULL; + gf_rdma_private_t *priv = NULL; + struct rdma_cm_id *cm_id = NULL; + int ret = 0; + + cm_id = event->id; + this = cm_id->context; + priv = this->private; + + priv->connected = 1; + + pthread_mutex_lock (&priv->write_mutex); + { + priv->peer.quota = 1; + priv->peer.quota_set = 0; + } + pthread_mutex_unlock (&priv->write_mutex); + + if (priv->entity == GF_RDMA_CLIENT) { + ret = rpc_transport_notify (this, RPC_TRANSPORT_CONNECT, this); + + } else if (priv->entity == GF_RDMA_SERVER) { + ret = rpc_transport_notify (priv->listener, + RPC_TRANSPORT_ACCEPT, this); + } + + if (ret < 0) { + gf_rdma_disconnect (this); } + gf_log (this->name, GF_LOG_TRACE, + "recieved event RDMA_CM_EVENT_ESTABLISHED (me:%s peer:%s)", + this->myinfo.identifier, this->peerinfo.identifier); + return ret; } +static int +gf_rdma_cm_handle_event_error (rpc_transport_t *this) +{ + gf_rdma_private_t *priv = NULL; + + priv = this->private; + + if (priv->entity != GF_RDMA_SERVER_LISTENER) { + gf_rdma_cm_handle_disconnect (this); + } + + return 0; +} + + +static int +gf_rdma_cm_handle_device_removal (struct rdma_cm_event *event) +{ + return 0; +} + + +static void * +gf_rdma_cm_event_handler (void *data) +{ + struct rdma_cm_event *event = NULL; + int ret = 0; + rpc_transport_t *this = NULL; + struct rdma_event_channel *event_channel = NULL; + + event_channel = data; + + while (1) { + ret = rdma_get_cm_event (event_channel, &event); + if (ret != 0) { + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "rdma_cm_get_event failed (%s)", + strerror (errno)); + break; + } + + switch (event->event) { + case RDMA_CM_EVENT_ADDR_RESOLVED: + gf_rdma_cm_handle_addr_resolved (event); + break; + + case RDMA_CM_EVENT_ROUTE_RESOLVED: + gf_rdma_cm_handle_route_resolved (event); + break; + + case RDMA_CM_EVENT_CONNECT_REQUEST: + gf_rdma_cm_handle_connect_request (event); + break; + + case RDMA_CM_EVENT_ESTABLISHED: + gf_rdma_cm_handle_event_established (event); + break; + + case RDMA_CM_EVENT_ADDR_ERROR: + case RDMA_CM_EVENT_ROUTE_ERROR: + case RDMA_CM_EVENT_CONNECT_ERROR: + case RDMA_CM_EVENT_UNREACHABLE: + case RDMA_CM_EVENT_REJECTED: + this = event->id->context; + + gf_log (this->name, GF_LOG_WARNING, + "cma event %s, error %d (me:%s peer:%s)\n", + rdma_event_str(event->event), event->status, + this->myinfo.identifier, + this->peerinfo.identifier); + + rdma_ack_cm_event (event); + event = NULL; + + gf_rdma_cm_handle_event_error (this); + continue; + + case RDMA_CM_EVENT_DISCONNECTED: + this = event->id->context; + + gf_log (this->name, GF_LOG_DEBUG, + "recieved disconnect (me:%s peer:%s)\n", + this->myinfo.identifier, + this->peerinfo.identifier); + + rdma_ack_cm_event (event); + event = NULL; + + gf_rdma_cm_handle_disconnect (this); + continue; + + case RDMA_CM_EVENT_DEVICE_REMOVAL: + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "device removed"); + gf_rdma_cm_handle_device_removal (event); + break; + + default: + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "unhandled event: %s, ignoring", + rdma_event_str(event->event)); + break; + } + + rdma_ack_cm_event (event); + } + + return NULL; +} + + static int32_t gf_rdma_post_send (struct ibv_qp *qp, gf_rdma_post_t *post, int32_t len) { @@ -832,28 +1618,6 @@ out: } -static int32_t -gf_rdma_post_recv (struct ibv_srq *srq, - gf_rdma_post_t *post) -{ - struct ibv_sge list = { - .addr = (unsigned long) post->buf, - .length = post->buf_size, - .lkey = post->mr->lkey - }; - - struct ibv_recv_wr wr = { - .wr_id = (unsigned long) post, - .sg_list = &list, - .num_sge = 1, - }, *bad_wr; - - gf_rdma_post_ref (post); - - return ibv_post_srq_recv (srq, &wr, &bad_wr); -} - - int gf_rdma_post_unref (gf_rdma_post_t *post) { @@ -1263,7 +2027,7 @@ out: } -static inline int32_t +inline int32_t __gf_rdma_register_local_mr_for_rdma (gf_rdma_peer_t *peer, struct iovec *vector, int count, gf_rdma_post_context_t *ctx) @@ -1715,7 +2479,7 @@ __gf_rdma_ioq_churn_entry (gf_rdma_peer_t *peer, gf_rdma_ioq_t *entry) if (quota > 0) { post = gf_rdma_get_post (&device->sendq); if (post == NULL) { - post = gf_rdma_new_post (device, + post = gf_rdma_new_post (peer->trans, device, (options->send_size + 2048), GF_RDMA_SEND_POST); } @@ -1992,189 +2756,6 @@ out: return ret; } -#if 0 -static int -gf_rdma_receive (rpc_transport_t *this, char **hdr_p, size_t *hdrlen_p, - struct iobuf **iobuf_p) -{ - gf_rdma_private_t *priv = this->private; - /* TODO: return error if !priv->connected, check with locks */ - /* TODO: boundry checks for data_ptr/offset */ - char *copy_from = NULL; - gf_rdma_header_t *header = NULL; - uint32_t size1, size2, data_len = 0; - char *hdr = NULL; - struct iobuf *iobuf = NULL; - int32_t ret = 0; - - pthread_mutex_lock (&priv->recv_mutex); - { -/* - while (!priv->data_ptr) - pthread_cond_wait (&priv->recv_cond, &priv->recv_mutex); -*/ - - copy_from = priv->data_ptr + priv->data_offset; - - priv->data_ptr = NULL; - data_len = priv->data_len; - pthread_cond_broadcast (&priv->recv_cond); - } - pthread_mutex_unlock (&priv->recv_mutex); - - header = (gf_rdma_header_t *)copy_from; - if (strcmp (header->colonO, ":O")) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_DEBUG, - "%s: corrupt header received", this->name); - ret = -1; - goto err; - } - - size1 = ntoh32 (header->size1); - size2 = ntoh32 (header->size2); - - if (data_len != (size1 + size2 + sizeof (*header))) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_DEBUG, - "%s: sizeof data read from transport is not equal " - "to the size specified in the header", - this->name); - ret = -1; - goto err; - } - - copy_from += sizeof (*header); - - if (size1) { - hdr = GF_CALLOC (1, size1, gf_common_mt_char); - if (!hdr) { - gf_log (this->name, GF_LOG_ERROR, - "unable to allocate header for peer %s", - this->peerinfo.identifier); - ret = -ENOMEM; - goto err; - } - memcpy (hdr, copy_from, size1); - copy_from += size1; - *hdr_p = hdr; - } - *hdrlen_p = size1; - - if (size2) { - iobuf = iobuf_get2 (this->ctx->iobuf_pool, size2); - if (!iobuf) { - gf_log (this->name, GF_LOG_ERROR, - "unable to allocate IO buffer for peer %s", - this->peerinfo.identifier); - ret = -ENOMEM; - goto err; - } - memcpy (iobuf->ptr, copy_from, size2); - *iobuf_p = iobuf; - } - -err: - return ret; -} -#endif - - -static void -gf_rdma_destroy_cq (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = NULL; - gf_rdma_device_t *device = NULL; - - priv = this->private; - device = priv->device; - - if (device->recv_cq) - ibv_destroy_cq (device->recv_cq); - device->recv_cq = NULL; - - if (device->send_cq) - ibv_destroy_cq (device->send_cq); - device->send_cq = NULL; - - return; -} - - -static int32_t -gf_rdma_create_cq (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - gf_rdma_device_t *device = NULL; - uint64_t send_cqe = 0; - int32_t ret = 0; - struct ibv_device_attr device_attr = {{0}, }; - - priv = this->private; - options = &priv->options; - device = priv->device; - - device->recv_cq = ibv_create_cq (priv->device->context, - options->recv_count * 2, - device, - device->recv_chan, - 0); - if (!device->recv_cq) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: creation of CQ for device %s failed", - this->name, device->device_name); - ret = -1; - goto out; - } else if (ibv_req_notify_cq (device->recv_cq, 0)) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: ibv_req_notify_cq on recv CQ of device %s failed", - this->name, device->device_name); - ret = -1; - goto out; - } - - do { - ret = ibv_query_device (priv->device->context, &device_attr); - if (ret != 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: ibv_query_device on %s returned %d (%s)", - this->name, priv->device->device_name, ret, - (ret > 0) ? strerror (ret) : ""); - ret = -1; - goto out; - } - - send_cqe = options->send_count * 128; - send_cqe = (send_cqe > device_attr.max_cqe) - ? device_attr.max_cqe : send_cqe; - - /* TODO: make send_cq size dynamically adaptive */ - device->send_cq = ibv_create_cq (priv->device->context, - send_cqe, device, - device->send_chan, 0); - if (!device->send_cq) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: creation of send_cq for device %s failed", - this->name, device->device_name); - ret = -1; - goto out; - } - - if (ibv_req_notify_cq (device->send_cq, 0)) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: ibv_req_notify_cq on send_cq for device %s" - " failed", this->name, device->device_name); - ret = -1; - goto out; - } - } while (0); - -out: - if (ret != 0) - gf_rdma_destroy_cq (this); - - return ret; -} - static int gf_rdma_register_peer (gf_rdma_device_t *device, int32_t qp_num, @@ -2273,25 +2854,6 @@ __gf_rdma_lookup_peer (gf_rdma_device_t *device, int32_t qp_num) return peer; } -/* - static gf_rdma_peer_t * - gf_rdma_lookup_peer (gf_rdma_device_t *device, - int32_t qp_num) - { - gf_rdma_qpreg_t *qpreg = NULL; - gf_rdma_peer_t *peer = NULL; - - qpreg = &device->qpreg; - pthread_mutex_lock (&qpreg->lock); - { - peer = __gf_rdma_lookup_peer (device, qp_num); - } - pthread_mutex_unlock (&qpreg->lock); - - return peer; - } -*/ - static void __gf_rdma_destroy_qp (rpc_transport_t *this) @@ -2301,7 +2863,7 @@ __gf_rdma_destroy_qp (rpc_transport_t *this) priv = this->private; if (priv->peer.qp) { gf_rdma_unregister_peer (priv->device, priv->peer.qp->qp_num); - ibv_destroy_qp (priv->peer.qp); + rdma_destroy_qp (priv->peer.cm_id); } priv->peer.qp = NULL; @@ -2312,18 +2874,36 @@ __gf_rdma_destroy_qp (rpc_transport_t *this) static int32_t gf_rdma_create_qp (rpc_transport_t *this) { - gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - gf_rdma_device_t *device = NULL; - int32_t ret = 0; - gf_rdma_peer_t *peer = NULL; + gf_rdma_private_t *priv = NULL; + gf_rdma_device_t *device = NULL; + int32_t ret = 0; + gf_rdma_peer_t *peer = NULL; + char *device_name = NULL; priv = this->private; - options = &priv->options; - device = priv->device; peer = &priv->peer; + device_name = (char *)ibv_get_device_name (peer->cm_id->verbs->device); + if (device_name == NULL) { + ret = -1; + gf_log (this->name, GF_LOG_WARNING, "cannot get device_name"); + goto out; + } + + device = gf_rdma_get_device (this, peer->cm_id->verbs, + device_name); + if (device == NULL) { + ret = -1; + gf_log (this->name, GF_LOG_WARNING, "cannot get device for " + "device %s", device_name); + goto out; + } + + if (priv->device == NULL) { + priv->device = device; + } + struct ibv_qp_init_attr init_attr = { .send_cq = device->send_cq, .recv_cq = device->recv_cq, @@ -2337,39 +2917,16 @@ gf_rdma_create_qp (rpc_transport_t *this) .qp_type = IBV_QPT_RC }; - struct ibv_qp_attr attr = { - .qp_state = IBV_QPS_INIT, - .pkey_index = 0, - .port_num = options->port, - .qp_access_flags - = IBV_ACCESS_REMOTE_READ | IBV_ACCESS_REMOTE_WRITE - }; - - peer->qp = ibv_create_qp (device->pd, &init_attr); - if (!peer->qp) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "%s: could not create QP", - this->name); - ret = -1; - goto out; - } else if (ibv_modify_qp (peer->qp, &attr, - IBV_QP_STATE | - IBV_QP_PKEY_INDEX | - IBV_QP_PORT | - IBV_QP_ACCESS_FLAGS)) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_ERROR, - "%s: failed to modify QP to INIT state", - this->name); + ret = rdma_create_qp(peer->cm_id, device->pd, &init_attr); + if (ret != 0) { + gf_log (peer->trans->name, GF_LOG_CRITICAL, + "%s: could not create QP (%s)", this->name, + strerror (errno)); ret = -1; goto out; } - peer->local_lid = gf_rdma_get_local_lid (device->context, - options->port); - peer->local_qpn = peer->qp->qp_num; - peer->local_psn = lrand48 () & 0xffffff; + peer->qp = peer->cm_id->qp; ret = gf_rdma_register_peer (device, peer->qp->qp_num, peer); @@ -2381,300 +2938,52 @@ out: } -static void -gf_rdma_destroy_posts (rpc_transport_t *this) -{ - -} - - -static int32_t -__gf_rdma_create_posts (rpc_transport_t *this, int32_t count, int32_t size, - gf_rdma_queue_t *q, gf_rdma_post_type_t type) -{ - int32_t i = 0; - int32_t ret = 0; - gf_rdma_private_t *priv = NULL; - gf_rdma_device_t *device = NULL; - - priv = this->private; - device = priv->device; - - for (i=0 ; i<count ; i++) { - gf_rdma_post_t *post = NULL; - - post = gf_rdma_new_post (device, size + 2048, type); - if (!post) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: post creation failed", - this->name); - ret = -1; - break; - } - - gf_rdma_put_post (q, post); - } - return ret; -} - - static int32_t -gf_rdma_create_posts (rpc_transport_t *this) +__gf_rdma_teardown (rpc_transport_t *this) { - int32_t i = 0, ret = 0; - gf_rdma_post_t *post = NULL; - gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - gf_rdma_device_t *device = NULL; + gf_rdma_private_t *priv = NULL; + gf_rdma_peer_t *peer = NULL; priv = this->private; - options = &priv->options; - device = priv->device; - - ret = __gf_rdma_create_posts (this, options->send_count, - options->send_size, - &device->sendq, GF_RDMA_SEND_POST); - if (!ret) - ret = __gf_rdma_create_posts (this, options->recv_count, - options->recv_size, - &device->recvq, - GF_RDMA_RECV_POST); + peer = &priv->peer; - if (!ret) { - for (i=0 ; i<options->recv_count ; i++) { - post = gf_rdma_get_post (&device->recvq); - if (gf_rdma_post_recv (device->srq, post) != 0) { - ret = -1; - break; - } - } + if (peer->cm_id->qp != NULL) { + __gf_rdma_destroy_qp (this); } - if (ret) - gf_rdma_destroy_posts (this); - - return ret; -} - - -static int32_t -gf_rdma_connect_qp (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = this->private; - gf_rdma_options_t *options = &priv->options; - struct ibv_qp_attr attr = { - .qp_state = IBV_QPS_RTR, - .path_mtu = options->mtu, - .dest_qp_num = priv->peer.remote_qpn, - .rq_psn = priv->peer.remote_psn, - .max_dest_rd_atomic = 1, - .min_rnr_timer = 12, - .qp_access_flags - = IBV_ACCESS_REMOTE_READ | IBV_ACCESS_REMOTE_WRITE, - .ah_attr = { - .is_global = 0, - .dlid = priv->peer.remote_lid, - .sl = 0, - .src_path_bits = 0, - .port_num = options->port - } - }; - if (ibv_modify_qp (priv->peer.qp, &attr, - IBV_QP_STATE | - IBV_QP_AV | - IBV_QP_PATH_MTU | - IBV_QP_DEST_QPN | - IBV_QP_RQ_PSN | - IBV_QP_MAX_DEST_RD_ATOMIC | - IBV_QP_MIN_RNR_TIMER)) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "Failed to modify QP to RTR\n"); - return -1; + if (!list_empty (&priv->peer.ioq)) { + __gf_rdma_ioq_flush (peer); } - attr.qp_state = IBV_QPS_RTS; - attr.timeout = options->attr_timeout; - attr.retry_cnt = options->attr_retry_cnt; - attr.rnr_retry = options->attr_rnr_retry; - attr.sq_psn = priv->peer.local_psn; - attr.max_rd_atomic = 1; - if (ibv_modify_qp (priv->peer.qp, &attr, - IBV_QP_STATE | - IBV_QP_TIMEOUT | - IBV_QP_RETRY_CNT | - IBV_QP_RNR_RETRY | - IBV_QP_SQ_PSN | - IBV_QP_MAX_QP_RD_ATOMIC)) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "Failed to modify QP to RTS\n"); - return -1; + if (peer->cm_id != NULL) { + rdma_destroy_id (peer->cm_id); + peer->cm_id = NULL; } + /* TODO: decrement cq size */ return 0; } + static int32_t -__gf_rdma_teardown (rpc_transport_t *this) +gf_rdma_teardown (rpc_transport_t *this) { + int32_t ret = 0; gf_rdma_private_t *priv = NULL; - priv = this->private; - __gf_rdma_destroy_qp (this); - - if (!list_empty (&priv->peer.ioq)) { - __gf_rdma_ioq_flush (&priv->peer); + if (this == NULL) { + goto out; } - /* TODO: decrement cq size */ - return 0; -} - -/* - * return value: - * 0 = success (completed) - * -1 = error - * > 0 = incomplete - */ - -static int -__tcp_rwv (rpc_transport_t *this, struct iovec *vector, int count, - struct iovec **pending_vector, int *pending_count, - int write) -{ - gf_rdma_private_t *priv = NULL; - int sock = -1; - int ret = -1; - struct iovec *opvector = NULL; - int opcount = 0; - int moved = 0; - priv = this->private; - sock = priv->sock; - opvector = vector; - opcount = count; - while (opcount) + pthread_mutex_lock (&priv->write_mutex); { - if (write) - { - ret = writev (sock, opvector, opcount); - - if (ret == 0 || (ret == -1 && errno == EAGAIN)) - { - /* done for now */ - break; - } - } - else - { - ret = readv (sock, opvector, opcount); - - if (ret == -1 && errno == EAGAIN) - { - /* done for now */ - break; - } - } - - if (ret == 0) - { - gf_log (this->name, GF_LOG_DEBUG, - "EOF from peer %s", this->peerinfo.identifier); - opcount = -1; - errno = ENOTCONN; - break; - } - - if (ret == -1) - { - if (errno == EINTR) - continue; - - gf_log (this->name, GF_LOG_DEBUG, - "%s failed (%s)", write ? "writev" : "readv", - strerror (errno)); - if (write && !priv->connected && - (errno == ECONNREFUSED)) - gf_log (this->name, GF_LOG_ERROR, - "possible mismatch of 'rpc-transport-type'" - " in protocol server and client. " - "check volume file"); - opcount = -1; - break; - } - - moved = 0; - - while (moved < ret) - { - if ((ret - moved) >= opvector[0].iov_len) - { - moved += opvector[0].iov_len; - opvector++; - opcount--; - } - else - { - opvector[0].iov_len -= (ret - moved); - opvector[0].iov_base += (ret - moved); - moved += (ret - moved); - } - while (opcount && !opvector[0].iov_len) - { - opvector++; - opcount--; - } - } - } - - if (pending_vector) - *pending_vector = opvector; - - if (pending_count) - *pending_count = opcount; - - return opcount; -} - - -static int -__tcp_readv (rpc_transport_t *this, struct iovec *vector, int count, - struct iovec **pending_vector, int *pending_count) -{ - int ret = -1; - - ret = __tcp_rwv (this, vector, count, - pending_vector, pending_count, 0); - - return ret; -} - - -static int -__tcp_writev (rpc_transport_t *this, struct iovec *vector, int count, - struct iovec **pending_vector, int *pending_count) -{ - int ret = -1; - gf_rdma_private_t *priv = NULL; - - priv = this->private; - - ret = __tcp_rwv (this, vector, count, pending_vector, - pending_count, 1); - - if (ret > 0) { - /* TODO: Avoid multiple calls when socket is already - registered for POLLOUT */ - priv->idx = event_select_on (this->ctx->event_pool, - priv->sock, priv->idx, -1, 1); - } else if (ret == 0) { - priv->idx = event_select_on (this->ctx->event_pool, - priv->sock, - priv->idx, -1, 0); + ret = __gf_rdma_teardown (this); } + pthread_mutex_unlock (&priv->write_mutex); +out: return ret; } @@ -2765,7 +3074,7 @@ out: } -static inline int32_t +inline int32_t gf_rdma_decode_error_msg (gf_rdma_peer_t *peer, gf_rdma_post_t *post, size_t bytes_in_post) { @@ -3307,7 +3616,7 @@ out: } -static inline int32_t +inline int32_t gf_rdma_recv_request (gf_rdma_peer_t *peer, gf_rdma_post_t *post, gf_rdma_read_chunk_t *readch) { @@ -3340,6 +3649,7 @@ gf_rdma_process_recv (gf_rdma_peer_t *peer, struct ibv_wc *wc) uint32_t *ptr = NULL; enum msg_type msg_type = 0; gf_rdma_header_t *header = NULL; + gf_rdma_private_t *priv = NULL; post = (gf_rdma_post_t *) (long) wc->wr_id; if (post == NULL) { @@ -3357,6 +3667,26 @@ gf_rdma_process_recv (gf_rdma_peer_t *peer, struct ibv_wc *wc) header = (gf_rdma_header_t *)post->buf; + priv = peer->trans->private; + + pthread_mutex_lock (&priv->write_mutex); + { + if (!priv->peer.quota_set) { + priv->peer.quota_set = 1; + + /* Initially peer.quota is set to 1 as per RFC 5666. We + * have to account for the quota used while sending + * first msg (which may or may not be returned to pool + * at this point) while deriving peer.quota from + * header->rm_credit. Hence the arithmatic below, + * instead of directly setting it to header->rm_credit. + */ + priv->peer.quota = header->rm_credit + - ( 1 - priv->peer.quota); + } + } + pthread_mutex_unlock (&priv->write_mutex); + switch (header->rm_type) { case GF_RDMA_MSG: ptr = (uint32_t *)post->ctx.vector[0].iov_base; @@ -3424,6 +3754,42 @@ out: return; } +void * +gf_rdma_async_event_thread (void *context) +{ + struct ibv_async_event event; + int ret; + + while (1) { + do { + ret = ibv_get_async_event((struct ibv_context *)context, + &event); + + if (ret && errno != EINTR) { + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "Error getting event (%s)", + strerror (errno)); + } + } while(ret && errno == EINTR); + + switch (event.event_type) { + case IBV_EVENT_SRQ_LIMIT_REACHED: + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "recieved srq_limit reached"); + break; + + default: + gf_log (GF_RDMA_LOG_NAME, GF_LOG_DEBUG, + "event (%d) recieved", event.event_type); + break; + } + + ibv_ack_async_event(&event); + } + + return 0; +} + static void * gf_rdma_recv_completion_proc (void *data) @@ -3809,364 +4175,87 @@ gf_rdma_options_init (rpc_transport_t *this) return; } -static void -gf_rdma_queue_init (gf_rdma_queue_t *queue) -{ - pthread_mutex_init (&queue->lock, NULL); - - queue->active_posts.next = &queue->active_posts; - queue->active_posts.prev = &queue->active_posts; - queue->passive_posts.next = &queue->passive_posts; - queue->passive_posts.prev = &queue->passive_posts; -} - -static gf_rdma_device_t * -gf_rdma_get_device (rpc_transport_t *this, struct ibv_context *ibctx) +gf_rdma_ctx_t * +__gf_rdma_ctx_create (void) { - glusterfs_ctx_t *ctx = NULL; - gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - char *device_name = NULL; - uint32_t port = 0; - uint8_t active_port = 0; - int32_t ret = 0; - int32_t i = 0; - gf_rdma_device_t *trav = NULL; - - priv = this->private; - options = &priv->options; - device_name = priv->options.device_name; - ctx = this->ctx; - trav = ctx->ib; - port = priv->options.port; + gf_rdma_ctx_t *rdma_ctx = NULL; + int ret = -1; - while (trav) { - if ((!strcmp (trav->device_name, device_name)) && - (trav->port == port)) - break; - trav = trav->next; + rdma_ctx = GF_CALLOC (1, sizeof (*rdma_ctx), gf_common_mt_char); + if (rdma_ctx == NULL) { + goto out; } - if (!trav) { - - trav = GF_CALLOC (1, sizeof (*trav), - gf_common_mt_rdma_device_t); - if (trav == NULL) { - return NULL; - } - - priv->device = trav; - - trav->context = ibctx; - - ret = ib_get_active_port (trav->context); - - if (ret < 0) { - if (!port) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "Failed to find any active ports and " - "none specified in volume file," - " exiting"); - GF_FREE (trav); - return NULL; - } - } - - trav->request_ctx_pool = mem_pool_new (gf_rdma_request_context_t, - GF_RDMA_POOL_SIZE); - if (trav->request_ctx_pool == NULL) { - return NULL; - } - - trav->ioq_pool = mem_pool_new (gf_rdma_ioq_t, GF_RDMA_POOL_SIZE); - if (trav->ioq_pool == NULL) { - mem_pool_destroy (trav->request_ctx_pool); - return NULL; - } - - trav->reply_info_pool = mem_pool_new (gf_rdma_reply_info_t, - GF_RDMA_POOL_SIZE); - if (trav->reply_info_pool == NULL) { - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->ioq_pool); - return NULL; - } - - - active_port = ret; - - if (port) { - ret = ib_check_active_port (trav->context, port); - if (ret < 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, - "On device %s: provided port:%u is " - "found to be offline, continuing to " - "use the same port", device_name, port); - } - } else { - priv->options.port = active_port; - port = active_port; - gf_log (GF_RDMA_LOG_NAME, GF_LOG_TRACE, - "Port unspecified in volume file using active " - "port: %u", port); - } - - trav->device_name = gf_strdup (device_name); - trav->port = port; - - trav->next = ctx->ib; - ctx->ib = trav; - - trav->send_chan = ibv_create_comp_channel (trav->context); - if (!trav->send_chan) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not create send completion channel", - device_name); - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - return NULL; - } - - trav->recv_chan = ibv_create_comp_channel (trav->context); - if (!trav->recv_chan) { - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "could not create recv completion channel"); - /* TODO: cleanup current mess */ - return NULL; - } - - if (gf_rdma_create_cq (this) < 0) { - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not create CQ", - this->name); - return NULL; - } - - /* protection domain */ - trav->pd = ibv_alloc_pd (trav->context); - - if (!trav->pd) { - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - gf_rdma_destroy_cq (this); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not allocate protection domain", - this->name); - return NULL; - } - - struct ibv_srq_init_attr attr = { - .attr = { - .max_wr = options->recv_count, - .max_sge = 1 - } - }; - trav->srq = ibv_create_srq (trav->pd, &attr); - - if (!trav->srq) { - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_dealloc_pd (trav->pd); - gf_rdma_destroy_cq (this); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not create SRQ", - this->name); - return NULL; - } - - /* queue init */ - gf_rdma_queue_init (&trav->sendq); - gf_rdma_queue_init (&trav->recvq); - - if (gf_rdma_create_posts (this) < 0) { - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_dealloc_pd (trav->pd); - gf_rdma_destroy_cq (this); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not allocate posts", - this->name); - return NULL; - } - - /* completion threads */ - ret = pthread_create (&trav->send_thread, - NULL, - gf_rdma_send_completion_proc, - trav->send_chan); - if (ret) { - gf_rdma_destroy_posts (this); - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_dealloc_pd (trav->pd); - gf_rdma_destroy_cq (this); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); + rdma_ctx->rdma_cm_event_channel = rdma_create_event_channel (); + if (rdma_ctx->rdma_cm_event_channel == NULL) { + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "rdma_cm event channel creation failed (%s)", + strerror (errno)); + goto out; + } - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "could not create send completion thread"); - return NULL; - } + ret = gf_thread_create (&rdma_ctx->rdma_cm_thread, NULL, + gf_rdma_cm_event_handler, + rdma_ctx->rdma_cm_event_channel); + if (ret != 0) { + gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, + "creation of thread to handle rdma-cm events " + "failed (%s)", strerror (ret)); + goto out; + } - ret = pthread_create (&trav->recv_thread, - NULL, - gf_rdma_recv_completion_proc, - trav->recv_chan); - if (ret) { - gf_rdma_destroy_posts (this); - mem_pool_destroy (trav->ioq_pool); - mem_pool_destroy (trav->request_ctx_pool); - mem_pool_destroy (trav->reply_info_pool); - ibv_dealloc_pd (trav->pd); - gf_rdma_destroy_cq (this); - ibv_destroy_comp_channel (trav->recv_chan); - ibv_destroy_comp_channel (trav->send_chan); - GF_FREE ((char *)trav->device_name); - GF_FREE (trav); - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "could not create recv completion thread"); - return NULL; +out: + if (ret < 0) { + if (rdma_ctx->rdma_cm_event_channel != NULL) { + rdma_destroy_event_channel (rdma_ctx->rdma_cm_event_channel); } - /* qpreg */ - pthread_mutex_init (&trav->qpreg.lock, NULL); - for (i=0; i<42; i++) { - trav->qpreg.ents[i].next = &trav->qpreg.ents[i]; - trav->qpreg.ents[i].prev = &trav->qpreg.ents[i]; - } + GF_FREE (rdma_ctx); + rdma_ctx = NULL; } - return trav; + + return rdma_ctx; } static int32_t gf_rdma_init (rpc_transport_t *this) { gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - struct ibv_device **dev_list; - struct ibv_context *ib_ctx = NULL; int32_t ret = 0; + glusterfs_ctx_t *ctx = NULL; + gf_rdma_options_t *options = NULL; + + ctx= this->ctx; priv = this->private; - options = &priv->options; ibv_fork_init (); gf_rdma_options_init (this); - { - dev_list = ibv_get_device_list (NULL); - - if (!dev_list) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "Failed to get IB devices"); - ret = -1; - goto cleanup; - } - - if (!*dev_list) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "No IB devices found"); - ret = -1; - goto cleanup; - } - - if (!options->device_name) { - if (*dev_list) { - options->device_name = - gf_strdup (ibv_get_device_name (*dev_list)); - } else { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_CRITICAL, - "IB device list is empty. Check for " - "'ib_uverbs' module"); - return -1; - goto cleanup; - } - } - - while (*dev_list) { - if (!strcmp (ibv_get_device_name (*dev_list), - options->device_name)) { - ib_ctx = ibv_open_device (*dev_list); - - if (!ib_ctx) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_ERROR, - "Failed to get infiniband" - "device context"); - ret = -1; - goto cleanup; - } - break; - } - ++dev_list; - } - - priv->device = gf_rdma_get_device (this, ib_ctx); - - if (!priv->device) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "could not create rdma device for %s", - options->device_name); - ret = -1; - goto cleanup; - } - } + options = &priv->options; + priv->peer.send_count = options->send_count; + priv->peer.recv_count = options->recv_count; + priv->peer.send_size = options->send_size; + priv->peer.recv_size = options->recv_size; priv->peer.trans = this; INIT_LIST_HEAD (&priv->peer.ioq); - pthread_mutex_init (&priv->read_mutex, NULL); pthread_mutex_init (&priv->write_mutex, NULL); pthread_mutex_init (&priv->recv_mutex, NULL); pthread_cond_init (&priv->recv_cond, NULL); -cleanup: - if (-1 == ret) { - if (ib_ctx) - ibv_close_device (ib_ctx); + pthread_mutex_lock (&ctx->lock); + { + if (ctx->ib == NULL) { + ctx->ib = __gf_rdma_ctx_create (); + if (ctx->ib == NULL) { + ret = -1; + } + } } - - if (dev_list) - ibv_free_device_list (dev_list); + pthread_mutex_unlock (&ctx->lock); return ret; } @@ -4194,537 +4283,54 @@ gf_rdma_disconnect (rpc_transport_t *this) static int32_t -__tcp_connect_finish (int fd) -{ - int ret = -1; - int optval = 0; - socklen_t optlen = sizeof (int); - - ret = getsockopt (fd, SOL_SOCKET, SO_ERROR, - (void *)&optval, &optlen); - - if (ret == 0 && optval) - { - errno = optval; - ret = -1; - } - - return ret; -} - -static inline void -gf_rdma_fill_handshake_data (char *buf, struct gf_rdma_nbio *nbio, - gf_rdma_private_t *priv) -{ - sprintf (buf, - "QP1:RECV_BLKSIZE=%08x:SEND_BLKSIZE=%08x\n" - "QP1:LID=%04x:QPN=%06x:PSN=%06x\n", - priv->peer.recv_size, - priv->peer.send_size, - priv->peer.local_lid, - priv->peer.local_qpn, - priv->peer.local_psn); - - nbio->vector.iov_base = buf; - nbio->vector.iov_len = strlen (buf) + 1; - nbio->count = 1; - return; -} - -static inline void -gf_rdma_fill_handshake_ack (char *buf, struct gf_rdma_nbio *nbio) -{ - sprintf (buf, "DONE\n"); - nbio->vector.iov_base = buf; - nbio->vector.iov_len = strlen (buf) + 1; - nbio->count = 1; - return; -} - -static int -gf_rdma_handshake_pollin (rpc_transport_t *this) -{ - int ret = 0; - gf_rdma_private_t *priv = NULL; - char *buf = NULL; - int32_t recv_buf_size = 0, send_buf_size; - socklen_t sock_len = 0; - - priv = this->private; - buf = priv->handshake.incoming.buf; - - if (priv->handshake.incoming.state == GF_RDMA_HANDSHAKE_COMPLETE) { - return -1; - } - - pthread_mutex_lock (&priv->write_mutex); - { - while (priv->handshake.incoming.state != GF_RDMA_HANDSHAKE_COMPLETE) - { - switch (priv->handshake.incoming.state) - { - case GF_RDMA_HANDSHAKE_START: - buf = priv->handshake.incoming.buf = GF_CALLOC (1, 256, gf_common_mt_char); - gf_rdma_fill_handshake_data (buf, &priv->handshake.incoming, priv); - buf[0] = 0; - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_RECEIVING_DATA; - break; - - case GF_RDMA_HANDSHAKE_RECEIVING_DATA: - ret = __tcp_readv (this, - &priv->handshake.incoming.vector, - priv->handshake.incoming.count, - &priv->handshake.incoming.pending_vector, - &priv->handshake.incoming.pending_count); - if (ret == -1) { - goto unlock; - } - - if (ret > 0) { - gf_log (this->name, GF_LOG_TRACE, - "partial header read on NB socket. continue later"); - goto unlock; - } - - if (!ret) { - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_RECEIVED_DATA; - } - break; - - case GF_RDMA_HANDSHAKE_RECEIVED_DATA: - ret = sscanf (buf, - "QP1:RECV_BLKSIZE=%08x:SEND_BLKSIZE=%08x\n" - "QP1:LID=%04x:QPN=%06x:PSN=%06x\n", - &recv_buf_size, - &send_buf_size, - &priv->peer.remote_lid, - &priv->peer.remote_qpn, - &priv->peer.remote_psn); - - if ((ret != 5) && (strncmp (buf, "QP1:", 4))) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_CRITICAL, - "%s: remote-host(%s)'s " - "transport type is different", - this->name, - this->peerinfo.identifier); - ret = -1; - goto unlock; - } - - if (recv_buf_size < priv->peer.recv_size) - priv->peer.recv_size = recv_buf_size; - if (send_buf_size < priv->peer.send_size) - priv->peer.send_size = send_buf_size; - - gf_log (GF_RDMA_LOG_NAME, GF_LOG_TRACE, - "%s: transacted recv_size=%d " - "send_size=%d", - this->name, priv->peer.recv_size, - priv->peer.send_size); - - priv->peer.quota = priv->peer.send_count; - - if (gf_rdma_connect_qp (this)) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_ERROR, - "%s: failed to connect with " - "remote QP", this->name); - ret = -1; - goto unlock; - } - gf_rdma_fill_handshake_ack (buf, &priv->handshake.incoming); - buf[0] = 0; - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_RECEIVING_ACK; - break; - - case GF_RDMA_HANDSHAKE_RECEIVING_ACK: - ret = __tcp_readv (this, - &priv->handshake.incoming.vector, - priv->handshake.incoming.count, - &priv->handshake.incoming.pending_vector, - &priv->handshake.incoming.pending_count); - if (ret == -1) { - goto unlock; - } - - if (ret > 0) { - gf_log (this->name, GF_LOG_TRACE, - "partial header read on NB " - "socket. continue later"); - goto unlock; - } - - if (!ret) { - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_RECEIVED_ACK; - } - break; - - case GF_RDMA_HANDSHAKE_RECEIVED_ACK: - if (strncmp (buf, "DONE", 4)) { - gf_log (GF_RDMA_LOG_NAME, - GF_LOG_DEBUG, - "%s: handshake-3 did not " - "return 'DONE' (%s)", - this->name, buf); - ret = -1; - goto unlock; - } - ret = 0; - priv->connected = 1; - sock_len = sizeof (struct sockaddr_storage); - getpeername (priv->sock, - (struct sockaddr *) &this->peerinfo.sockaddr, - &sock_len); - - GF_FREE (priv->handshake.incoming.buf); - priv->handshake.incoming.buf = NULL; - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_COMPLETE; - } - } - } -unlock: - pthread_mutex_unlock (&priv->write_mutex); - - if (ret == -1) { - rpc_transport_disconnect (this); - } else { - ret = 0; - } - - - if (!ret && priv->connected) { - if (priv->is_server) { - ret = rpc_transport_notify (priv->listener, - RPC_TRANSPORT_ACCEPT, - this); - } else { - ret = rpc_transport_notify (this, RPC_TRANSPORT_CONNECT, - this); - } - } - - return ret; -} - -static int -gf_rdma_handshake_pollout (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = NULL; - char *buf = NULL; - int32_t ret = 0; - - priv = this->private; - buf = priv->handshake.outgoing.buf; - - if (priv->handshake.outgoing.state == GF_RDMA_HANDSHAKE_COMPLETE) { - return 0; - } - - pthread_mutex_unlock (&priv->write_mutex); - { - while (priv->handshake.outgoing.state - != GF_RDMA_HANDSHAKE_COMPLETE) - { - switch (priv->handshake.outgoing.state) - { - case GF_RDMA_HANDSHAKE_START: - buf = priv->handshake.outgoing.buf - = GF_CALLOC (1, 256, gf_common_mt_char); - gf_rdma_fill_handshake_data (buf, - &priv->handshake.outgoing, priv); - priv->handshake.outgoing.state - = GF_RDMA_HANDSHAKE_SENDING_DATA; - break; - - case GF_RDMA_HANDSHAKE_SENDING_DATA: - ret = __tcp_writev (this, - &priv->handshake.outgoing.vector, - priv->handshake.outgoing.count, - &priv->handshake.outgoing.pending_vector, - &priv->handshake.outgoing.pending_count); - if (ret == -1) { - goto unlock; - } - - if (ret > 0) { - gf_log (this->name, GF_LOG_TRACE, - "partial header read on NB " - "socket. continue later"); - goto unlock; - } - - if (!ret) { - priv->handshake.outgoing.state - = GF_RDMA_HANDSHAKE_SENT_DATA; - } - break; - - case GF_RDMA_HANDSHAKE_SENT_DATA: - gf_rdma_fill_handshake_ack (buf, - &priv->handshake.outgoing); - priv->handshake.outgoing.state - = GF_RDMA_HANDSHAKE_SENDING_ACK; - break; - - case GF_RDMA_HANDSHAKE_SENDING_ACK: - ret = __tcp_writev (this, - &priv->handshake.outgoing.vector, - priv->handshake.outgoing.count, - &priv->handshake.outgoing.pending_vector, - &priv->handshake.outgoing.pending_count); - - if (ret == -1) { - goto unlock; - } - - if (ret > 0) { - gf_log (this->name, GF_LOG_TRACE, - "partial header read on NB " - "socket. continue later"); - goto unlock; - } - - if (!ret) { - GF_FREE (priv->handshake.outgoing.buf); - priv->handshake.outgoing.buf = NULL; - priv->handshake.outgoing.state - = GF_RDMA_HANDSHAKE_COMPLETE; - } - break; - } - } - } -unlock: - pthread_mutex_unlock (&priv->write_mutex); - - if (ret == -1) { - rpc_transport_disconnect (this); - } else { - ret = 0; - } - - return ret; -} - -static int -gf_rdma_handshake_pollerr (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = this->private; - char need_unref = 0, connected = 0; - - gf_log_callingfn (GF_RDMA_LOG_NAME, GF_LOG_WARNING, - "%s: peer (%s) disconnected, cleaning up", - this->name, this->peerinfo.identifier); - - pthread_mutex_lock (&priv->write_mutex); - { - __gf_rdma_teardown (this); - - connected = priv->connected; - if (priv->sock != -1) { - event_unregister (this->ctx->event_pool, - priv->sock, priv->idx); - need_unref = 1; - - if (close (priv->sock) != 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "close () - error: %s", - strerror (errno)); - } - priv->tcp_connected = priv->connected = 0; - priv->sock = -1; - } - - if (priv->handshake.incoming.buf) { - GF_FREE (priv->handshake.incoming.buf); - priv->handshake.incoming.buf = NULL; - } - - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_START; - - if (priv->handshake.outgoing.buf) { - GF_FREE (priv->handshake.outgoing.buf); - priv->handshake.outgoing.buf = NULL; - } - - priv->handshake.outgoing.state = GF_RDMA_HANDSHAKE_START; - } - pthread_mutex_unlock (&priv->write_mutex); - - if (connected) { - rpc_transport_notify (this, RPC_TRANSPORT_DISCONNECT, this); - } - - if (need_unref) - rpc_transport_unref (this); - - return 0; -} - - -static int -tcp_connect_finish (rpc_transport_t *this) -{ - gf_rdma_private_t *priv = NULL; - int error = 0, ret = 0; - - priv = this->private; - pthread_mutex_lock (&priv->write_mutex); - { - ret = __tcp_connect_finish (priv->sock); - - if (!ret) { - this->myinfo.sockaddr_len = - sizeof (this->myinfo.sockaddr); - ret = getsockname (priv->sock, - (struct sockaddr *)&this->myinfo.sockaddr, - &this->myinfo.sockaddr_len); - if (ret == -1) - { - gf_log (this->name, GF_LOG_ERROR, - "getsockname on new client-socket %d " - "failed (%s)", - priv->sock, strerror (errno)); - close (priv->sock); - error = 1; - goto unlock; - } - - gf_rdma_get_transport_identifiers (this); - priv->tcp_connected = 1; - } - - if (ret == -1 && errno != EINPROGRESS) { - gf_log (this->name, GF_LOG_ERROR, - "tcp connect to %s failed (%s)", - this->peerinfo.identifier, strerror (errno)); - error = 1; - } - } -unlock: - pthread_mutex_unlock (&priv->write_mutex); - - if (error) { - rpc_transport_disconnect (this); - } - - return ret; -} - -static int -gf_rdma_event_handler (int fd, int idx, void *data, - int poll_in, int poll_out, int poll_err) -{ - rpc_transport_t *this = NULL; - gf_rdma_private_t *priv = NULL; - gf_rdma_options_t *options = NULL; - int ret = 0; - - this = data; - priv = this->private; - if (!priv->tcp_connected) { - ret = tcp_connect_finish (this); - if (priv->tcp_connected) { - options = &priv->options; - - priv->peer.send_count = options->send_count; - priv->peer.recv_count = options->recv_count; - priv->peer.send_size = options->send_size; - priv->peer.recv_size = options->recv_size; - - if ((ret = gf_rdma_create_qp (this)) < 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not create QP", - this->name); - rpc_transport_disconnect (this); - } - } - } - - if (!ret && poll_out && priv->tcp_connected) { - ret = gf_rdma_handshake_pollout (this); - } - - if (!ret && !poll_err && poll_in && priv->tcp_connected) { - if (priv->handshake.incoming.state - == GF_RDMA_HANDSHAKE_COMPLETE) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: pollin received on tcp socket (peer: %s) " - "after handshake is complete", - this->name, this->peerinfo.identifier); - gf_rdma_handshake_pollerr (this); - return 0; - } - - ret = gf_rdma_handshake_pollin (this); - if (ret < 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_WARNING, - "handshake pollin failed"); - } - } - - if (ret < 0 || poll_err) { - ret = gf_rdma_handshake_pollerr (this); - } - - return 0; -} - -static int -__tcp_nonblock (int fd) -{ - int flags = 0; - int ret = -1; - - flags = fcntl (fd, F_GETFL); - - if (flags != -1) - ret = fcntl (fd, F_SETFL, flags | O_NONBLOCK); - - return ret; -} - -static int32_t gf_rdma_connect (struct rpc_transport *this, int port) { gf_rdma_private_t *priv = NULL; int32_t ret = 0; - gf_boolean_t non_blocking = 1; union gf_sock_union sock_union = {{0, }, }; socklen_t sockaddr_len = 0; + gf_rdma_peer_t *peer = NULL; + gf_rdma_ctx_t *rdma_ctx = NULL; + gf_boolean_t connected = _gf_false; priv = this->private; + peer = &priv->peer; + + rpc_transport_ref (this); + ret = gf_rdma_client_get_remote_sockaddr (this, &sock_union.sa, &sockaddr_len, port); if (ret != 0) { gf_log (this->name, GF_LOG_DEBUG, "cannot get remote address to connect"); - return ret; + goto out; } + rdma_ctx = this->ctx->ib; pthread_mutex_lock (&priv->write_mutex); { - if (priv->sock != -1) { - ret = 0; + if (peer->cm_id != NULL) { + ret = -1; + errno = EINPROGRESS; + connected = _gf_true; goto unlock; } - priv->sock = socket (sock_union.sa.sa_family, SOCK_STREAM, 0); + priv->entity = GF_RDMA_CLIENT; - if (priv->sock == -1) { + ret = rdma_create_id (rdma_ctx->rdma_cm_event_channel, + &peer->cm_id, this, RDMA_PS_TCP); + if (ret != 0) { gf_log (this->name, GF_LOG_ERROR, - "socket () - error: %s", strerror (errno)); + "creation of rdma_cm_id failed (%s)", + strerror (errno)); ret = -errno; goto unlock; } - gf_log (this->name, GF_LOG_TRACE, - "socket fd = %d", priv->sock); - memcpy (&this->peerinfo.sockaddr, &sock_union.storage, sockaddr_len); this->peerinfo.sockaddr_len = sockaddr_len; @@ -4735,201 +4341,84 @@ gf_rdma_connect (struct rpc_transport *this, int port) ((struct sockaddr *) &this->myinfo.sockaddr)->sa_family = ((struct sockaddr *)&this->peerinfo.sockaddr)->sa_family; - if (non_blocking) - { - ret = __tcp_nonblock (priv->sock); - - if (ret == -1) - { - gf_log (this->name, GF_LOG_ERROR, - "could not set socket %d to non " - "blocking mode (%s)", - priv->sock, strerror (errno)); - close (priv->sock); - priv->sock = -1; - goto unlock; - } - } - ret = gf_rdma_client_bind (this, (struct sockaddr *)&this->myinfo.sockaddr, &this->myinfo.sockaddr_len, - priv->sock); - if (ret == -1) - { + peer->cm_id); + if (ret != 0) { gf_log (this->name, GF_LOG_WARNING, "client bind failed: %s", strerror (errno)); - close (priv->sock); - priv->sock = -1; goto unlock; } - ret = connect (priv->sock, - (struct sockaddr *)&this->peerinfo.sockaddr, - this->peerinfo.sockaddr_len); - if (ret == -1 && errno != EINPROGRESS) - { - gf_log (this->name, GF_LOG_ERROR, - "connection attempt failed (%s)", + ret = rdma_resolve_addr (peer->cm_id, NULL, &sock_union.sa, + 2000); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "rdma_resolve_addr failed (%s)", strerror (errno)); - close (priv->sock); - priv->sock = -1; goto unlock; } - priv->tcp_connected = priv->connected = 0; - - rpc_transport_ref (this); - - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_START; - priv->handshake.outgoing.state = GF_RDMA_HANDSHAKE_START; - - priv->idx = event_register (this->ctx->event_pool, - priv->sock, gf_rdma_event_handler, - this, 1, 1); + priv->connected = 0; } unlock: pthread_mutex_unlock (&priv->write_mutex); - return ret; -} - -static int -gf_rdma_server_event_handler (int fd, int idx, void *data, - int poll_in, int poll_out, int poll_err) -{ - int32_t main_sock = -1; - rpc_transport_t *this = NULL, *trans = NULL; - gf_rdma_private_t *priv = NULL; - gf_rdma_private_t *trans_priv = NULL; - gf_rdma_options_t *options = NULL; - - if (!poll_in) { - return 0; - } - - trans = data; - trans_priv = (gf_rdma_private_t *) trans->private; - - this = GF_CALLOC (1, sizeof (rpc_transport_t), - gf_common_mt_rpc_transport_t); - if (this == NULL) { - return -1; - } - - this->listener = trans; - - priv = GF_CALLOC (1, sizeof (gf_rdma_private_t), - gf_common_mt_rdma_private_t); - if (priv == NULL) { - GF_FREE (priv); - return -1; - } - this->private = priv; - /* Copy all the rdma related values in priv, from trans_priv - as other than QP, all the values remain same */ - priv->device = trans_priv->device; - priv->options = trans_priv->options; - priv->is_server = 1; - priv->listener = trans; - - options = &priv->options; - - this->ops = trans->ops; - this->init = trans->init; - this->fini = trans->fini; - this->ctx = trans->ctx; - this->name = gf_strdup (trans->name); - this->notify = trans->notify; - this->mydata = trans->mydata; - - memcpy (&this->myinfo.sockaddr, &trans->myinfo.sockaddr, - trans->myinfo.sockaddr_len); - this->myinfo.sockaddr_len = trans->myinfo.sockaddr_len; - - main_sock = (trans_priv)->sock; - this->peerinfo.sockaddr_len = sizeof (this->peerinfo.sockaddr); - priv->sock = accept (main_sock, - (struct sockaddr *)&this->peerinfo.sockaddr, - &this->peerinfo.sockaddr_len); - if (priv->sock == -1) { - gf_log ("rdma/server", GF_LOG_ERROR, - "accept() failed: %s", - strerror (errno)); - GF_FREE (this->private); - GF_FREE (this); - return -1; - } - - priv->peer.trans = this; - rpc_transport_ref (this); - - gf_rdma_get_transport_identifiers (this); - - priv->tcp_connected = 1; - priv->handshake.incoming.state = GF_RDMA_HANDSHAKE_START; - priv->handshake.outgoing.state = GF_RDMA_HANDSHAKE_START; - - priv->peer.send_count = options->send_count; - priv->peer.recv_count = options->recv_count; - priv->peer.send_size = options->send_size; - priv->peer.recv_size = options->recv_size; - INIT_LIST_HEAD (&priv->peer.ioq); +out: + if (ret != 0) { + if (!connected) { + gf_rdma_teardown (this); + } - if (gf_rdma_create_qp (this) < 0) { - gf_log (GF_RDMA_LOG_NAME, GF_LOG_ERROR, - "%s: could not create QP", - this->name); - rpc_transport_disconnect (this); - return -1; + rpc_transport_unref (this); } - priv->idx = event_register (this->ctx->event_pool, priv->sock, - gf_rdma_event_handler, this, 1, 1); - - pthread_mutex_init (&priv->read_mutex, NULL); - pthread_mutex_init (&priv->write_mutex, NULL); - pthread_mutex_init (&priv->recv_mutex, NULL); - /* pthread_cond_init (&priv->recv_cond, NULL); */ - return 0; + return ret; } + static int32_t gf_rdma_listen (rpc_transport_t *this) { union gf_sock_union sock_union = {{0, }, }; socklen_t sockaddr_len = 0; gf_rdma_private_t *priv = NULL; - int opt = 1, ret = 0; + gf_rdma_peer_t *peer = NULL; + int ret = 0; + gf_rdma_ctx_t *rdma_ctx = NULL; char service[NI_MAXSERV], host[NI_MAXHOST]; priv = this->private; - memset (&sock_union, 0, sizeof (sock_union)); - ret = gf_rdma_server_get_local_sockaddr (this, - &sock_union.sa, + peer = &priv->peer; + + priv->entity = GF_RDMA_SERVER_LISTENER; + + rdma_ctx = this->ctx->ib; + + ret = gf_rdma_server_get_local_sockaddr (this, &sock_union.sa, &sockaddr_len); if (ret != 0) { - gf_log (this->name, GF_LOG_DEBUG, + gf_log (this->name, GF_LOG_WARNING, "cannot find network address of server to bind to"); goto err; } - priv->sock = socket (sock_union.sa.sa_family, SOCK_STREAM, 0); - if (priv->sock == -1) { - gf_log ("rdma/server", GF_LOG_CRITICAL, - "init: failed to create socket, error: %s", + ret = rdma_create_id (rdma_ctx->rdma_cm_event_channel, + &peer->cm_id, this, RDMA_PS_TCP); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "creation of rdma_cm_id failed (%s)", strerror (errno)); - GF_FREE (this->private); - ret = -1; goto err; } - memcpy (&this->myinfo.sockaddr, &sock_union.storage, sockaddr_len); + memcpy (&this->myinfo.sockaddr, &sock_union.storage, + sockaddr_len); this->myinfo.sockaddr_len = sockaddr_len; ret = getnameinfo ((struct sockaddr *)&this->myinfo.sockaddr, - this->myinfo.sockaddr_len, - host, sizeof (host), + this->myinfo.sockaddr_len, host, sizeof (host), service, sizeof (service), NI_NUMERICHOST); if (ret != 0) { @@ -4937,34 +4426,38 @@ gf_rdma_listen (rpc_transport_t *this) "getnameinfo failed (%s)", gai_strerror (ret)); goto err; } + sprintf (this->myinfo.identifier, "%s:%s", host, service); - setsockopt (priv->sock, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof (opt)); - if (bind (priv->sock, &sock_union.sa, sockaddr_len) != 0) { - ret = -1; - gf_log ("rdma/server", GF_LOG_ERROR, - "init: failed to bind to socket for %s (%s)", - this->myinfo.identifier, strerror (errno)); + ret = rdma_bind_addr (peer->cm_id, &sock_union.sa); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "rdma_bind_addr failed (%s)", strerror (errno)); goto err; } - if (listen (priv->sock, 10) != 0) { - gf_log ("rdma/server", GF_LOG_ERROR, - "init: listen () failed on socket for %s (%s)", - this->myinfo.identifier, strerror (errno)); - ret = -1; + ret = rdma_listen (peer->cm_id, 10); + if (ret != 0) { + gf_log (this->name, GF_LOG_WARNING, + "rdma_listen failed (%s)", strerror (errno)); goto err; } - /* Register the main socket */ - priv->idx = event_register (this->ctx->event_pool, priv->sock, - gf_rdma_server_event_handler, - rpc_transport_ref (this), 1, 0); + rpc_transport_ref (this); + ret = 0; err: + if (ret < 0) { + if (peer->cm_id != NULL) { + rdma_destroy_id (peer->cm_id); + peer->cm_id = NULL; + } + } + return ret; } + struct rpc_transport_ops tops = { .submit_request = gf_rdma_submit_request, .submit_reply = gf_rdma_submit_reply, @@ -4983,7 +4476,6 @@ init (rpc_transport_t *this) return -1; this->private = priv; - priv->sock = -1; if (gf_rdma_init (this)) { gf_log (this->name, GF_LOG_ERROR, @@ -5007,13 +4499,6 @@ fini (struct rpc_transport *this) if (priv) { pthread_mutex_destroy (&priv->recv_mutex); pthread_mutex_destroy (&priv->write_mutex); - pthread_mutex_destroy (&priv->read_mutex); - - /* pthread_cond_destroy (&priv->recv_cond); */ - if (priv->sock != -1) { - event_unregister (this->ctx->event_pool, - priv->sock, priv->idx); - } gf_log (this->name, GF_LOG_TRACE, "called fini on transport: %p", this); diff --git a/rpc/rpc-transport/rdma/src/rdma.h b/rpc/rpc-transport/rdma/src/rdma.h index 687d6005f..7f76244f0 100644 --- a/rpc/rpc-transport/rdma/src/rdma.h +++ b/rpc/rpc-transport/rdma/src/rdma.h @@ -29,6 +29,7 @@ #include <list.h> #include <arpa/inet.h> #include <infiniband/verbs.h> +#include <rdma/rdma_cma.h> /* FIXME: give appropriate values to these macros */ #define GF_DEFAULT_RDMA_LISTEN_PORT (GF_DEFAULT_BASE_PORT + 1) @@ -230,30 +231,33 @@ typedef enum __gf_rdma_send_post_type { /* represents one communication peer, two per transport_t */ struct __gf_rdma_peer { - rpc_transport_t *trans; - struct ibv_qp *qp; + rpc_transport_t *trans; + struct rdma_cm_id *cm_id; + struct ibv_qp *qp; + pthread_t rdma_event_thread; + char quota_set; int32_t recv_count; int32_t send_count; int32_t recv_size; int32_t send_size; - int32_t quota; + int32_t quota; union { - struct list_head ioq; + struct list_head ioq; struct { - gf_rdma_ioq_t *ioq_next; - gf_rdma_ioq_t *ioq_prev; + gf_rdma_ioq_t *ioq_next; + gf_rdma_ioq_t *ioq_prev; }; }; /* QP attributes, needed to connect with remote QP */ - int32_t local_lid; - int32_t local_psn; - int32_t local_qpn; - int32_t remote_lid; - int32_t remote_psn; - int32_t remote_qpn; + int32_t local_lid; + int32_t local_psn; + int32_t local_qpn; + int32_t remote_lid; + int32_t remote_psn; + int32_t remote_qpn; }; typedef struct __gf_rdma_peer gf_rdma_peer_t; @@ -320,33 +324,19 @@ struct __gf_rdma_device { struct ibv_comp_channel *send_chan, *recv_chan; struct ibv_cq *send_cq, *recv_cq; gf_rdma_queue_t sendq, recvq; - pthread_t send_thread, recv_thread; + pthread_t send_thread, recv_thread, async_event_thread; struct mem_pool *request_ctx_pool; struct mem_pool *ioq_pool; struct mem_pool *reply_info_pool; }; typedef struct __gf_rdma_device gf_rdma_device_t; -typedef enum { - GF_RDMA_HANDSHAKE_START = 0, - GF_RDMA_HANDSHAKE_SENDING_DATA, - GF_RDMA_HANDSHAKE_RECEIVING_DATA, - GF_RDMA_HANDSHAKE_SENT_DATA, - GF_RDMA_HANDSHAKE_RECEIVED_DATA, - GF_RDMA_HANDSHAKE_SENDING_ACK, - GF_RDMA_HANDSHAKE_RECEIVING_ACK, - GF_RDMA_HANDSHAKE_RECEIVED_ACK, - GF_RDMA_HANDSHAKE_COMPLETE, -} gf_rdma_handshake_state_t; - -struct gf_rdma_nbio { - int state; - char *buf; - int count; - struct iovec vector; - struct iovec *pending_vector; - int pending_count; +struct __gf_rdma_ctx { + gf_rdma_device_t *device; + struct rdma_event_channel *rdma_cm_event_channel; + pthread_t rdma_cm_thread; }; +typedef struct __gf_rdma_ctx gf_rdma_ctx_t; struct __gf_rdma_request_context { struct ibv_mr *mr[GF_RDMA_MAX_SEGMENTS]; @@ -358,46 +348,35 @@ struct __gf_rdma_request_context { }; typedef struct __gf_rdma_request_context gf_rdma_request_context_t; +typedef enum { + GF_RDMA_SERVER_LISTENER, + GF_RDMA_SERVER, + GF_RDMA_CLIENT, +} gf_rdma_transport_entity_t; + struct __gf_rdma_private { - int32_t sock; - int32_t idx; - unsigned char connected; - unsigned char tcp_connected; - unsigned char ib_connected; - in_addr_t addr; + int32_t idx; + unsigned char connected; + in_addr_t addr; unsigned short port; /* IB Verbs Driver specific variables, pointers */ - gf_rdma_peer_t peer; + gf_rdma_peer_t peer; struct __gf_rdma_device *device; - gf_rdma_options_t options; + gf_rdma_options_t options; /* Used by trans->op->receive */ - char *data_ptr; - int32_t data_offset; - int32_t data_len; + char *data_ptr; + int32_t data_offset; + int32_t data_len; /* Mutex */ - pthread_mutex_t read_mutex; - pthread_mutex_t write_mutex; - pthread_barrier_t handshake_barrier; - char handshake_ret; - char is_server; - rpc_transport_t *listener; - - pthread_mutex_t recv_mutex; - pthread_cond_t recv_cond; - - /* used during gf_rdma_handshake */ - struct { - struct gf_rdma_nbio incoming; - struct gf_rdma_nbio outgoing; - int state; - gf_rdma_header_t header; - char *buf; - size_t size; - } handshake; + pthread_mutex_t write_mutex; + rpc_transport_t *listener; + pthread_mutex_t recv_mutex; + pthread_cond_t recv_cond; + gf_rdma_transport_entity_t entity; }; -typedef struct __gf_rdma_private gf_rdma_private_t; +typedef struct __gf_rdma_private gf_rdma_private_t; #endif /* _XPORT_GF_RDMA_H */ diff --git a/rpc/rpc-transport/socket/src/socket.c b/rpc/rpc-transport/socket/src/socket.c index fffc137f6..93da3f296 100644 --- a/rpc/rpc-transport/socket/src/socket.c +++ b/rpc/rpc-transport/socket/src/socket.c @@ -147,21 +147,14 @@ typedef int SSL_trinary_func (SSL *, void *, int); &in->pending_vector, \ &in->pending_count, \ &bytes_read); \ - if (ret == -1) { \ - if (priv->read_fail_log) \ - gf_log (this->name, GF_LOG_WARNING, \ - "reading from socket failed." \ - "Error (%s), peer (%s)", \ - strerror (errno), \ - this->peerinfo.identifier); \ + if (ret == -1) \ break; \ - } \ __socket_proto_update_priv_after_read (priv, ret, bytes_read); \ } -int socket_init (rpc_transport_t *this); +static int socket_init (rpc_transport_t *this); -void +static void ssl_dump_error_stack (const char *caller) { unsigned long errnum = 0; @@ -175,7 +168,7 @@ ssl_dump_error_stack (const char *caller) } } -int +static int ssl_do (rpc_transport_t *this, void *buf, size_t len, SSL_trinary_func *func) { int r = (-1); @@ -246,7 +239,7 @@ out: #define ssl_read_one(t,b,l) ssl_do((t),(b),(l),(SSL_trinary_func *)SSL_read) #define ssl_write_one(t,b,l) ssl_do((t),(b),(l),(SSL_trinary_func *)SSL_write) -int +static int ssl_setup_connection (rpc_transport_t *this, int server) { X509 *peer = NULL; @@ -311,7 +304,7 @@ done: } -void +static void ssl_teardown_connection (socket_private_t *priv) { SSL_shutdown(priv->ssl_ssl); @@ -321,7 +314,7 @@ ssl_teardown_connection (socket_private_t *priv) } -ssize_t +static ssize_t __socket_ssl_readv (rpc_transport_t *this, struct iovec *opvector, int opcount) { socket_private_t *priv = NULL; @@ -341,7 +334,7 @@ __socket_ssl_readv (rpc_transport_t *this, struct iovec *opvector, int opcount) } -ssize_t +static ssize_t __socket_ssl_read (rpc_transport_t *this, void *buf, size_t count) { struct iovec iov = {0, }; @@ -356,7 +349,7 @@ __socket_ssl_read (rpc_transport_t *this, void *buf, size_t count) } -int +static int __socket_cached_read (rpc_transport_t *this, struct iovec *opvector, int opcount) { socket_private_t *priv = NULL; @@ -424,6 +417,19 @@ out: return ret; } +static gf_boolean_t +__does_socket_rwv_error_need_logging (socket_private_t *priv, int write) +{ + int read = !write; + + if (priv->connected == -1) /* Didn't even connect, of course it fails */ + return _gf_false; + + if (read && (priv->read_fail_log == _gf_false)) + return _gf_false; + + return _gf_true; +} /* * return value: @@ -432,7 +438,7 @@ out: * > 0 = incomplete */ -int +static int __socket_rwv (rpc_transport_t *this, struct iovec *vector, int count, struct iovec **pending_vector, int *pending_count, size_t *bytes, int write) @@ -507,11 +513,15 @@ __socket_rwv (rpc_transport_t *this, struct iovec *vector, int count, if (errno == EINTR) continue; - if (write || (!write && priv->read_fail_log)) + if (__does_socket_rwv_error_need_logging (priv, + write)) { gf_log (this->name, GF_LOG_WARNING, - "%s failed (%s)", + "%s on %s failed (%s)", write ? "writev":"readv", + this->peerinfo.identifier, strerror (errno)); + } + if (priv->use_ssl) { ssl_dump_error_stack(this->name); } @@ -562,7 +572,7 @@ out: } -int +static int __socket_readv (rpc_transport_t *this, struct iovec *vector, int count, struct iovec **pending_vector, int *pending_count, size_t *bytes) @@ -576,7 +586,7 @@ __socket_readv (rpc_transport_t *this, struct iovec *vector, int count, } -int +static int __socket_writev (rpc_transport_t *this, struct iovec *vector, int count, struct iovec **pending_vector, int *pending_count) { @@ -589,7 +599,7 @@ __socket_writev (rpc_transport_t *this, struct iovec *vector, int count, } -int +static int __socket_shutdown (rpc_transport_t *this) { int ret = -1; @@ -608,7 +618,7 @@ __socket_shutdown (rpc_transport_t *this) return ret; } -int +static int __socket_disconnect (rpc_transport_t *this) { int ret = -1; @@ -619,25 +629,23 @@ __socket_disconnect (rpc_transport_t *this) priv = this->private; + gf_log (this->name, GF_LOG_TRACE, + "disconnecting %p, state=%u gen=%u sock=%d", this, + priv->ot_state, priv->ot_gen, priv->sock); + if (priv->sock != -1) { ret = __socket_shutdown(this); if (priv->own_thread) { - /* - * Without this, reconnect (= disconnect + connect) - * won't work except by accident. - */ - close(priv->sock); - priv->sock = -1; /* - * Closing the socket forces an error that will wake - * up the polling thread. Wait for it to notice and - * respond. + * Without this, reconnect (= disconnect + connect) + * won't work except by accident. */ - if (priv->ot_state == OT_ALIVE) { - priv->ot_state = OT_DYING; - pthread_cond_wait(&priv->ot_event,&priv->lock); - } - } + close(priv->sock); + priv->sock = -1; + gf_log (this->name, GF_LOG_TRACE, + "OT_PLEASE_DIE on %p", this); + priv->ot_state = OT_PLEASE_DIE; + } else if (priv->use_ssl) { ssl_teardown_connection(priv); } @@ -648,7 +656,7 @@ out: } -int +static int __socket_server_bind (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -704,7 +712,7 @@ out: } -int +static int __socket_nonblock (int fd) { int flags = 0; @@ -718,7 +726,7 @@ __socket_nonblock (int fd) return ret; } -int +static int __socket_nodelay (int fd) { int on = 1; @@ -794,7 +802,7 @@ err: } -int +static int __socket_connect_finish (int fd) { int ret = -1; @@ -812,7 +820,7 @@ __socket_connect_finish (int fd) } -void +static void __socket_reset (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -849,13 +857,13 @@ out: } -void +static void socket_set_lastfrag (uint32_t *fragsize) { (*fragsize) |= 0x80000000U; } -void +static void socket_set_frag_header_size (uint32_t size, char *haddr) { size = htonl (size); @@ -863,14 +871,14 @@ socket_set_frag_header_size (uint32_t size, char *haddr) } -void +static void socket_set_last_frag_header_size (uint32_t size, char *haddr) { socket_set_lastfrag (&size); socket_set_frag_header_size (size, haddr); } -struct ioq * +static struct ioq * __socket_ioq_new (rpc_transport_t *this, rpc_transport_msg_t *msg) { struct ioq *entry = NULL; @@ -937,7 +945,7 @@ out: } -void +static void __socket_ioq_entry_free (struct ioq *entry) { GF_VALIDATE_OR_GOTO ("socket", entry, out); @@ -954,7 +962,7 @@ out: } -void +static void __socket_ioq_flush (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -975,7 +983,7 @@ out: } -int +static int __socket_ioq_churn_entry (rpc_transport_t *this, struct ioq *entry, int direct) { int ret = -1; @@ -1009,7 +1017,7 @@ __socket_ioq_churn_entry (rpc_transport_t *this, struct ioq *entry, int direct) } -int +static int __socket_ioq_churn (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -1042,7 +1050,7 @@ out: } -int +static int socket_event_poll_err (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -1067,7 +1075,7 @@ out: } -int +static int socket_event_poll_out (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -1405,7 +1413,7 @@ __socket_read_request (rpc_transport_t *this) buf = rpc_procnum_addr (iobuf_ptr (in->iobuf)); procnum = ntoh32 (*((uint32_t *)buf)); - if (this->listener) { + if (priv->is_server) { /* this check is needed as rpcsvc and rpc-clnt * actor structures are not same */ vector_sizer = @@ -1911,7 +1919,7 @@ void __socket_reset_priv (socket_private_t *priv) } -int +static int __socket_proto_state_machine (rpc_transport_t *this, rpc_transport_pollin_t **pollin) { @@ -1952,17 +1960,8 @@ __socket_proto_state_machine (rpc_transport_t *this, &in->pending_vector, &in->pending_count, NULL); - if (ret == -1) { - if (priv->read_fail_log == 1) { - gf_log (this->name, - ((priv->connected == 1) ? - GF_LOG_WARNING : GF_LOG_DEBUG), - "reading from socket failed. Error (%s)" - ", peer (%s)", strerror (errno), - this->peerinfo.identifier); - } + if (ret == -1) goto out; - } if (ret > 0) { gf_log (this->name, GF_LOG_TRACE, "partial " @@ -2083,7 +2082,7 @@ out: } -int +static int socket_proto_state_machine (rpc_transport_t *this, rpc_transport_pollin_t **pollin) { @@ -2106,17 +2105,22 @@ out: } -int +static int socket_event_poll_in (rpc_transport_t *this) { int ret = -1; rpc_transport_pollin_t *pollin = NULL; + socket_private_t *priv = this->private; ret = socket_proto_state_machine (this, &pollin); if (pollin != NULL) { + priv->ot_state = OT_CALLBACK; ret = rpc_transport_notify (this, RPC_TRANSPORT_MSG_RECEIVED, pollin); + if (priv->ot_state == OT_CALLBACK) { + priv->ot_state = OT_RUNNING; + } rpc_transport_pollin_destroy (pollin); } @@ -2124,7 +2128,7 @@ socket_event_poll_in (rpc_transport_t *this) } -int +static int socket_connect_finish (rpc_transport_t *this) { int ret = -1; @@ -2158,8 +2162,6 @@ socket_connect_finish (rpc_transport_t *this) priv->connect_finish_log = 1; } __socket_disconnect (this); - notify_rpc = 1; - event = RPC_TRANSPORT_DISCONNECT; goto unlock; } @@ -2198,7 +2200,7 @@ out: /* reads rpc_requests during pollin */ -int +static int socket_event_handler (int fd, int idx, void *data, int poll_in, int poll_out, int poll_err) { @@ -2243,7 +2245,7 @@ out: } -void * +static void * socket_poller (void *ctx) { rpc_transport_t *this = ctx; @@ -2251,6 +2253,9 @@ socket_poller (void *ctx) struct pollfd pfd[2] = {{0,},}; gf_boolean_t to_write = _gf_false; int ret = 0; + uint32_t gen = 0; + + priv->ot_state = OT_RUNNING; if (priv->use_ssl) { if (ssl_setup_connection(this,priv->connected) < 0) { @@ -2286,6 +2291,7 @@ socket_poller (void *ctx) "asynchronous rpc_transport_notify failed"); } + gen = priv->ot_gen; for (;;) { pthread_mutex_lock(&priv->lock); to_write = !list_empty(&priv->ioq); @@ -2322,6 +2328,13 @@ socket_poller (void *ctx) else if (errno == ENOTCONN) { ret = 0; } + if (priv->ot_state == OT_PLEASE_DIE) { + gf_log (this->name, GF_LOG_TRACE, + "OT_IDLE on %p (input request)", + this); + priv->ot_state = OT_IDLE; + break; + } } else if (pfd[1].revents & POLL_MASK_OUTPUT) { ret = socket_event_poll_out(this); @@ -2332,6 +2345,13 @@ socket_poller (void *ctx) else if (errno == ENOTCONN) { ret = 0; } + if (priv->ot_state == OT_PLEASE_DIE) { + gf_log (this->name, GF_LOG_TRACE, + "OT_IDLE on %p (output request)", + this); + priv->ot_state = OT_IDLE; + break; + } } else { /* @@ -2352,21 +2372,17 @@ socket_poller (void *ctx) "error in polling loop"); break; } + if (priv->ot_gen != gen) { + gf_log (this->name, GF_LOG_TRACE, + "generation mismatch, my %u != %u", + gen, priv->ot_gen); + return NULL; + } } err: /* All (and only) I/O errors should come here. */ pthread_mutex_lock(&priv->lock); - if (priv->ot_state == OT_ALIVE) { - /* - * We have to do this if we're here because of an error we - * detected ourselves, but need to avoid a recursive call - * if our death is the result of an external disconnect. - */ - __socket_shutdown(this); - close(priv->sock); - priv->sock = -1; - } if (priv->ssl_ssl) { /* * We're always responsible for this part, but only actually @@ -2375,40 +2391,45 @@ err: */ ssl_teardown_connection(priv); } + __socket_shutdown(this); + close(priv->sock); + priv->sock = -1; priv->ot_state = OT_IDLE; - /* - * We expect there to be only one waiter, but if there do happen to - * be multiple it's probably better to unblock them than to let them - * hang. If there are none, this is a harmless no-op. - */ - pthread_cond_broadcast(&priv->ot_event); pthread_mutex_unlock(&priv->lock); - rpc_transport_notify (this->listener, RPC_TRANSPORT_DISCONNECT, this); - rpc_transport_unref (this); + rpc_transport_notify (this->listener, RPC_TRANSPORT_DISCONNECT, + this); + rpc_transport_unref (this); return NULL; } -void +static void socket_spawn (rpc_transport_t *this) { socket_private_t *priv = this->private; - if (priv->ot_state == OT_ALIVE) { + switch (priv->ot_state) { + case OT_IDLE: + case OT_PLEASE_DIE: + break; + default: gf_log (this->name, GF_LOG_WARNING, "refusing to start redundant poller"); return; } - priv->ot_state = OT_ALIVE; + priv->ot_gen += 7; + priv->ot_state = OT_SPAWNING; + gf_log (this->name, GF_LOG_TRACE, + "spawning %p with gen %u", this, priv->ot_gen); - if (pthread_create(&priv->thread,NULL,socket_poller,this) != 0) { + if (gf_thread_create(&priv->thread,NULL,socket_poller,this) != 0) { gf_log (this->name, GF_LOG_ERROR, "could not create poll thread"); } } -int +static int socket_server_event_handler (int fd, int idx, void *data, int poll_in, int poll_out, int poll_err) { @@ -2446,7 +2467,7 @@ socket_server_event_handler (int fd, int idx, void *data, goto unlock; } - if (priv->nodelay) { + if (priv->nodelay && (new_sockaddr.ss_family != AF_UNIX)) { ret = __socket_nodelay (new_sock); if (ret == -1) { gf_log (this->name, GF_LOG_WARNING, @@ -2456,7 +2477,8 @@ socket_server_event_handler (int fd, int idx, void *data, } } - if (priv->keepalive) { + if (priv->keepalive && + new_sockaddr.ss_family != AF_UNIX) { ret = __socket_keepalive (new_sock, new_sockaddr.ss_family, priv->keepaliveintvl, @@ -2472,6 +2494,15 @@ socket_server_event_handler (int fd, int idx, void *data, if (!new_trans) goto unlock; + ret = pthread_mutex_init(&new_trans->lock, NULL); + if (ret == -1) { + gf_log (this->name, GF_LOG_WARNING, + "pthread_mutex_init() failed: %s", + strerror (errno)); + close (new_sock); + goto unlock; + } + new_trans->name = gf_strdup (this->name); memcpy (&new_trans->peerinfo.sockaddr, &new_sockaddr, @@ -2543,6 +2574,7 @@ socket_server_event_handler (int fd, int idx, void *data, * connection. */ new_priv->connected = 1; + new_priv->is_server = _gf_true; rpc_transport_ref (new_trans); if (new_priv->own_thread) { @@ -2585,7 +2617,7 @@ out: } -int +static int socket_disconnect (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -2607,7 +2639,7 @@ out: } -int +static int socket_connect (rpc_transport_t *this, int port) { int ret = -1; @@ -2646,6 +2678,10 @@ socket_connect (rpc_transport_t *this, int port) goto err; } + gf_log (this->name, GF_LOG_TRACE, + "connecting %p, state=%u gen=%u sock=%d", this, + priv->ot_state, priv->ot_gen, priv->sock); + ret = socket_client_get_remote_sockaddr (this, &sock_union.sa, &sockaddr_len, &sa_family); if (ret == -1) { @@ -2715,7 +2751,7 @@ socket_connect (rpc_transport_t *this, int port) } } - if (priv->nodelay) { + if (priv->nodelay && (sa_family != AF_UNIX)) { ret = __socket_nodelay (priv->sock); if (ret == -1) { @@ -2725,7 +2761,7 @@ socket_connect (rpc_transport_t *this, int port) } } - if (priv->keepalive) { + if (priv->keepalive && sa_family != AF_UNIX) { ret = __socket_keepalive (priv->sock, sa_family, priv->keepaliveintvl, @@ -2758,13 +2794,30 @@ socket_connect (rpc_transport_t *this, int port) goto unlock; } + if (!priv->use_ssl && !priv->bio && !priv->own_thread) { + ret = __socket_nonblock (priv->sock); + if (ret == -1) { + gf_log (this->name, GF_LOG_ERROR, + "NBIO on %d failed (%s)", + priv->sock, strerror (errno)); + close (priv->sock); + priv->sock = -1; + goto unlock; + } + } + ret = connect (priv->sock, SA (&this->peerinfo.sockaddr), this->peerinfo.sockaddr_len); if (ret == -1 && ((errno != EINPROGRESS) && (errno != ENOENT))) { - gf_log (this->name, GF_LOG_ERROR, - "connection attempt failed (%s)", - strerror (errno)); + /* For unix path based sockets, the socket path is + * cryptic (md5sum of path) and may not be useful for + * the user in debugging so log it in DEBUG + */ + gf_log (this->name, ((sa_family == AF_UNIX) ? + GF_LOG_DEBUG : GF_LOG_ERROR), + "connection attempt on %s failed, (%s)", + this->peerinfo.identifier, strerror (errno)); close (priv->sock); priv->sock = -1; goto unlock; @@ -2799,6 +2852,7 @@ socket_connect (rpc_transport_t *this, int port) * initializing a client connection. */ priv->connected = 0; + priv->is_server = _gf_false; rpc_transport_ref (this); if (priv->own_thread) { @@ -2829,7 +2883,7 @@ err: } -int +static int socket_listen (rpc_transport_t *this) { socket_private_t * priv = NULL; @@ -2911,7 +2965,7 @@ socket_listen (rpc_transport_t *this) } } - if (priv->nodelay) { + if (priv->nodelay && (sa_family != AF_UNIX)) { ret = __socket_nodelay (priv->sock); if (ret == -1) { gf_log (this->name, GF_LOG_ERROR, @@ -2980,7 +3034,7 @@ out: } -int32_t +static int32_t socket_submit_request (rpc_transport_t *this, rpc_transport_req_t *req) { socket_private_t *priv = NULL; @@ -3054,7 +3108,7 @@ out: } -int32_t +static int32_t socket_submit_reply (rpc_transport_t *this, rpc_transport_reply_t *reply) { socket_private_t *priv = NULL; @@ -3128,7 +3182,7 @@ out: } -int32_t +static int32_t socket_getpeername (rpc_transport_t *this, char *hostname, int hostlen) { int32_t ret = -1; @@ -3147,7 +3201,7 @@ out: } -int32_t +static int32_t socket_getpeeraddr (rpc_transport_t *this, char *peeraddr, int addrlen, struct sockaddr_storage *sa, socklen_t salen) { @@ -3168,7 +3222,7 @@ out: } -int32_t +static int32_t socket_getmyname (rpc_transport_t *this, char *hostname, int hostlen) { int32_t ret = -1; @@ -3187,7 +3241,7 @@ out: } -int32_t +static int32_t socket_getmyaddr (rpc_transport_t *this, char *myaddr, int addrlen, struct sockaddr_storage *sa, socklen_t salen) { @@ -3207,6 +3261,25 @@ out: } +static int +socket_throttle (rpc_transport_t *this, gf_boolean_t onoff) +{ + socket_private_t *priv = NULL; + + priv = this->private; + + /* The way we implement throttling is by taking off + POLLIN event from the polled flags. This way we + never get called with the POLLIN event and therefore + will never read() any more data until throttling + is turned off. + */ + priv->idx = event_select_on (this->ctx->event_pool, priv->sock, + priv->idx, (int) !onoff, -1); + return 0; +} + + struct rpc_transport_ops tops = { .listen = socket_listen, .connect = socket_connect, @@ -3217,6 +3290,7 @@ struct rpc_transport_ops tops = { .get_peeraddr = socket_getpeeraddr, .get_myname = socket_getmyname, .get_myaddr = socket_getmyaddr, + .throttle = socket_throttle, }; int @@ -3273,7 +3347,7 @@ out: } -int +static int socket_init (rpc_transport_t *this) { socket_private_t *priv = NULL; @@ -3522,7 +3596,6 @@ socket_init (rpc_transport_t *this) if (priv->own_thread) { priv->ot_state = OT_IDLE; - pthread_cond_init (&priv->ot_event, NULL); } out: diff --git a/rpc/rpc-transport/socket/src/socket.h b/rpc/rpc-transport/socket/src/socket.h index bb342d998..e0b412fcc 100644 --- a/rpc/rpc-transport/socket/src/socket.h +++ b/rpc/rpc-transport/socket/src/socket.h @@ -186,8 +186,10 @@ struct gf_sock_incoming { typedef enum { OT_IDLE, /* Uninitialized or termination complete. */ - OT_ALIVE, /* Past pthread_create, no error/disconnect. */ - OT_DYING, /* Disconnect in progress. */ + OT_SPAWNING, /* Past pthread_create but not in thread yet. */ + OT_RUNNING, /* Poller thread running normally. */ + OT_CALLBACK, /* Poller thread in the middle of a callback. */ + OT_PLEASE_DIE, /* Poller termination requested. */ } ot_state_t; typedef struct { @@ -229,7 +231,8 @@ typedef struct { int pipe[2]; gf_boolean_t own_thread; ot_state_t ot_state; - pthread_cond_t ot_event; + uint32_t ot_gen; + gf_boolean_t is_server; } socket_private_t; diff --git a/rpc/xdr/src/cli1-xdr.c b/rpc/xdr/src/cli1-xdr.c index b8780af91..97b210e14 100644 --- a/rpc/xdr/src/cli1-xdr.c +++ b/rpc/xdr/src/cli1-xdr.c @@ -179,136 +179,51 @@ xdr_gf_cli_status_type (XDR *xdrs, gf_cli_status_type *objp) } bool_t -xdr_gf_cli_req (XDR *xdrs, gf_cli_req *objp) +xdr_gf1_cli_snapshot (XDR *xdrs, gf1_cli_snapshot *objp) { register int32_t *buf; buf = NULL; - if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + if (!xdr_enum (xdrs, (enum_t *) objp)) return FALSE; return TRUE; } bool_t -xdr_gf_cli_rsp (XDR *xdrs, gf_cli_rsp *objp) +xdr_gf1_cli_snapshot_config (XDR *xdrs, gf1_cli_snapshot_config *objp) { register int32_t *buf; buf = NULL; - if (!xdr_int (xdrs, &objp->op_ret)) - return FALSE; - if (!xdr_int (xdrs, &objp->op_errno)) - return FALSE; - if (!xdr_string (xdrs, &objp->op_errstr, ~0)) - return FALSE; - if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + if (!xdr_enum (xdrs, (enum_t *) objp)) return FALSE; return TRUE; } bool_t -xdr_gf1_cli_probe_req (XDR *xdrs, gf1_cli_probe_req *objp) +xdr_gf_cli_req (XDR *xdrs, gf_cli_req *objp) { register int32_t *buf; buf = NULL; - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; - if (!xdr_int (xdrs, &objp->port)) + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) return FALSE; return TRUE; } bool_t -xdr_gf1_cli_probe_rsp (XDR *xdrs, gf1_cli_probe_rsp *objp) +xdr_gf_cli_rsp (XDR *xdrs, gf_cli_rsp *objp) { register int32_t *buf; buf = NULL; - - if (xdrs->x_op == XDR_ENCODE) { - buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); - if (buf == NULL) { - if (!xdr_int (xdrs, &objp->op_ret)) - return FALSE; - if (!xdr_int (xdrs, &objp->op_errno)) - return FALSE; - if (!xdr_int (xdrs, &objp->port)) - return FALSE; - - } else { - IXDR_PUT_LONG(buf, objp->op_ret); - IXDR_PUT_LONG(buf, objp->op_errno); - IXDR_PUT_LONG(buf, objp->port); - } - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; - if (!xdr_string (xdrs, &objp->op_errstr, ~0)) - return FALSE; - return TRUE; - } else if (xdrs->x_op == XDR_DECODE) { - buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); - if (buf == NULL) { - if (!xdr_int (xdrs, &objp->op_ret)) - return FALSE; - if (!xdr_int (xdrs, &objp->op_errno)) - return FALSE; - if (!xdr_int (xdrs, &objp->port)) - return FALSE; - - } else { - objp->op_ret = IXDR_GET_LONG(buf); - objp->op_errno = IXDR_GET_LONG(buf); - objp->port = IXDR_GET_LONG(buf); - } - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; - if (!xdr_string (xdrs, &objp->op_errstr, ~0)) - return FALSE; - return TRUE; - } - if (!xdr_int (xdrs, &objp->op_ret)) return FALSE; if (!xdr_int (xdrs, &objp->op_errno)) return FALSE; - if (!xdr_int (xdrs, &objp->port)) - return FALSE; - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; if (!xdr_string (xdrs, &objp->op_errstr, ~0)) return FALSE; - return TRUE; -} - -bool_t -xdr_gf1_cli_deprobe_req (XDR *xdrs, gf1_cli_deprobe_req *objp) -{ - register int32_t *buf; - buf = NULL; - - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; - if (!xdr_int (xdrs, &objp->port)) - return FALSE; - if (!xdr_int (xdrs, &objp->flags)) - return FALSE; - return TRUE; -} - -bool_t -xdr_gf1_cli_deprobe_rsp (XDR *xdrs, gf1_cli_deprobe_rsp *objp) -{ - register int32_t *buf; - buf = NULL; - - if (!xdr_int (xdrs, &objp->op_ret)) - return FALSE; - if (!xdr_int (xdrs, &objp->op_errno)) - return FALSE; - if (!xdr_string (xdrs, &objp->hostname, ~0)) - return FALSE; - if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) return FALSE; return TRUE; } diff --git a/rpc/xdr/src/cli1-xdr.h b/rpc/xdr/src/cli1-xdr.h index 58ce1101e..5e8c29fbb 100644 --- a/rpc/xdr/src/cli1-xdr.h +++ b/rpc/xdr/src/cli1-xdr.h @@ -48,6 +48,10 @@ enum gf_defrag_status_t { GF_DEFRAG_STATUS_STOPPED = 2, GF_DEFRAG_STATUS_COMPLETE = 3, GF_DEFRAG_STATUS_FAILED = 4, + GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED = 5, + GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED = 6, + GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE = 7, + GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED = 8, }; typedef enum gf_defrag_status_t gf_defrag_status_t; @@ -92,7 +96,8 @@ enum gf_quota_type { typedef enum gf_quota_type gf_quota_type; enum gf1_cli_friends_list { - GF_CLI_LIST_ALL = 1, + GF_CLI_LIST_PEERS = 1, + GF_CLI_LIST_POOL_NODES = 2, }; typedef enum gf1_cli_friends_list gf1_cli_friends_list; @@ -120,6 +125,8 @@ enum gf1_cli_gsync_set { GF_GSYNC_OPTION_TYPE_CONFIG = 3, GF_GSYNC_OPTION_TYPE_STATUS = 4, GF_GSYNC_OPTION_TYPE_ROTATE = 5, + GF_GSYNC_OPTION_TYPE_CREATE = 6, + GF_GSYNC_OPTION_TYPE_DELETE = 7, }; typedef enum gf1_cli_gsync_set gf1_cli_gsync_set; @@ -152,6 +159,7 @@ enum gf_cli_status_type { GF_CLI_STATUS_FD = 0x0008, GF_CLI_STATUS_CALLPOOL = 0x0010, GF_CLI_STATUS_DETAIL = 0x0020, + GF_CLI_STATUS_TASKS = 0x0040, GF_CLI_STATUS_MASK = 0x00FF, GF_CLI_STATUS_VOL = 0x0100, GF_CLI_STATUS_ALL = 0x0200, @@ -161,6 +169,41 @@ enum gf_cli_status_type { }; typedef enum gf_cli_status_type gf_cli_status_type; +enum gf1_cli_snapshot { + GF_SNAP_OPTION_TYPE_NONE = 0, + GF_SNAP_OPTION_TYPE_CREATE = 1, + GF_SNAP_OPTION_TYPE_DELETE = 2, + GF_SNAP_OPTION_TYPE_RESTORE = 3, + GF_SNAP_OPTION_TYPE_START = 4, + GF_SNAP_OPTION_TYPE_STOP = 5, + GF_SNAP_OPTION_TYPE_LIST = 6, + GF_SNAP_OPTION_TYPE_STATUS = 7, + GF_SNAP_OPTION_TYPE_CONFIG = 8, + GF_SNAP_OPTION_TYPE_INFO = 9, +}; +typedef enum gf1_cli_snapshot gf1_cli_snapshot; + +enum gf1_cli_snapshot_info { + GF_SNAP_INFO_TYPE_ALL = 0, + GF_SNAP_INFO_TYPE_SNAP = 1, + GF_SNAP_INFO_TYPE_VOL = 2, +}; +typedef enum gf1_cli_snapshot_info gf1_cli_snapshot_info; + +enum gf1_cli_snapshot_config { + GF_SNAP_CONFIG_TYPE_NONE = 0, + GF_SNAP_CONFIG_TYPE_SET = 1, + GF_SNAP_CONFIG_DISPLAY = 2, +}; +typedef enum gf1_cli_snapshot_config gf1_cli_snapshot_config; + +enum gf1_cli_snapshot_status { + GF_SNAP_STATUS_TYPE_ALL = 0, + GF_SNAP_STATUS_TYPE_SNAP = 1, + GF_SNAP_STATUS_TYPE_VOL = 2, +}; +typedef enum gf1_cli_snapshot_status gf1_cli_snapshot_status; + struct gf_cli_req { struct { u_int dict_len; @@ -180,36 +223,6 @@ struct gf_cli_rsp { }; typedef struct gf_cli_rsp gf_cli_rsp; -struct gf1_cli_probe_req { - char *hostname; - int port; -}; -typedef struct gf1_cli_probe_req gf1_cli_probe_req; - -struct gf1_cli_probe_rsp { - int op_ret; - int op_errno; - int port; - char *hostname; - char *op_errstr; -}; -typedef struct gf1_cli_probe_rsp gf1_cli_probe_rsp; - -struct gf1_cli_deprobe_req { - char *hostname; - int port; - int flags; -}; -typedef struct gf1_cli_deprobe_req gf1_cli_deprobe_req; - -struct gf1_cli_deprobe_rsp { - int op_ret; - int op_errno; - char *hostname; - char *op_errstr; -}; -typedef struct gf1_cli_deprobe_rsp gf1_cli_deprobe_rsp; - struct gf1_cli_peer_list_req { int flags; struct { @@ -302,12 +315,10 @@ extern bool_t xdr_gf1_cli_gsync_set (XDR *, gf1_cli_gsync_set*); extern bool_t xdr_gf1_cli_stats_op (XDR *, gf1_cli_stats_op*); extern bool_t xdr_gf1_cli_top_op (XDR *, gf1_cli_top_op*); extern bool_t xdr_gf_cli_status_type (XDR *, gf_cli_status_type*); +extern bool_t xdr_gf1_cli_snapshot (XDR *, gf1_cli_snapshot*); +extern bool_t xdr_gf1_cli_snapshot_config (XDR *, gf1_cli_snapshot_config*); extern bool_t xdr_gf_cli_req (XDR *, gf_cli_req*); extern bool_t xdr_gf_cli_rsp (XDR *, gf_cli_rsp*); -extern bool_t xdr_gf1_cli_probe_req (XDR *, gf1_cli_probe_req*); -extern bool_t xdr_gf1_cli_probe_rsp (XDR *, gf1_cli_probe_rsp*); -extern bool_t xdr_gf1_cli_deprobe_req (XDR *, gf1_cli_deprobe_req*); -extern bool_t xdr_gf1_cli_deprobe_rsp (XDR *, gf1_cli_deprobe_rsp*); extern bool_t xdr_gf1_cli_peer_list_req (XDR *, gf1_cli_peer_list_req*); extern bool_t xdr_gf1_cli_peer_list_rsp (XDR *, gf1_cli_peer_list_rsp*); extern bool_t xdr_gf1_cli_fsm_log_req (XDR *, gf1_cli_fsm_log_req*); @@ -334,12 +345,10 @@ extern bool_t xdr_gf1_cli_gsync_set (); extern bool_t xdr_gf1_cli_stats_op (); extern bool_t xdr_gf1_cli_top_op (); extern bool_t xdr_gf_cli_status_type (); +extern bool_t xdr_gf1_cli_snapshot (); +extern bool_t xdr_gf1_cli_snapshot_config (); extern bool_t xdr_gf_cli_req (); extern bool_t xdr_gf_cli_rsp (); -extern bool_t xdr_gf1_cli_probe_req (); -extern bool_t xdr_gf1_cli_probe_rsp (); -extern bool_t xdr_gf1_cli_deprobe_req (); -extern bool_t xdr_gf1_cli_deprobe_rsp (); extern bool_t xdr_gf1_cli_peer_list_req (); extern bool_t xdr_gf1_cli_peer_list_rsp (); extern bool_t xdr_gf1_cli_fsm_log_req (); diff --git a/rpc/xdr/src/cli1-xdr.x b/rpc/xdr/src/cli1-xdr.x index cb22080cc..f9d29b7e1 100644 --- a/rpc/xdr/src/cli1-xdr.x +++ b/rpc/xdr/src/cli1-xdr.x @@ -11,7 +11,11 @@ GF_DEFRAG_STATUS_STARTED, GF_DEFRAG_STATUS_STOPPED, GF_DEFRAG_STATUS_COMPLETE, - GF_DEFRAG_STATUS_FAILED + GF_DEFRAG_STATUS_FAILED, + GF_DEFRAG_STATUS_LAYOUT_FIX_STARTED, + GF_DEFRAG_STATUS_LAYOUT_FIX_STOPPED, + GF_DEFRAG_STATUS_LAYOUT_FIX_COMPLETE, + GF_DEFRAG_STATUS_LAYOUT_FIX_FAILED } ; enum gf1_cluster_type { @@ -51,7 +55,8 @@ enum gf_quota_type { }; enum gf1_cli_friends_list { - GF_CLI_LIST_ALL = 1 + GF_CLI_LIST_PEERS = 1, + GF_CLI_LIST_POOL_NODES = 2 } ; enum gf1_cli_get_volume { @@ -74,7 +79,9 @@ enum gf1_cli_gsync_set { GF_GSYNC_OPTION_TYPE_STOP, GF_GSYNC_OPTION_TYPE_CONFIG, GF_GSYNC_OPTION_TYPE_STATUS, - GF_GSYNC_OPTION_TYPE_ROTATE + GF_GSYNC_OPTION_TYPE_ROTATE, + GF_GSYNC_OPTION_TYPE_CREATE, + GF_GSYNC_OPTION_TYPE_DELETE }; enum gf1_cli_stats_op { @@ -106,6 +113,7 @@ enum gf_cli_status_type { GF_CLI_STATUS_FD = 0x0008, /*0000000001000*/ GF_CLI_STATUS_CALLPOOL = 0x0010, /*0000000010000*/ GF_CLI_STATUS_DETAIL = 0x0020, /*0000000100000*/ + GF_CLI_STATUS_TASKS = 0x0040, /*0000001000000*/ GF_CLI_STATUS_MASK = 0x00FF, /*0000011111111 Used to get the op*/ GF_CLI_STATUS_VOL = 0x0100, /*0000100000000*/ GF_CLI_STATUS_ALL = 0x0200, /*0001000000000*/ @@ -114,6 +122,26 @@ enum gf_cli_status_type { GF_CLI_STATUS_SHD = 0x1000 /*1000000000000*/ }; +/* Identifiers for snapshot clis */ +enum gf1_cli_snapshot { + GF_SNAP_OPTION_TYPE_NONE = 0, + GF_SNAP_OPTION_TYPE_CREATE, + GF_SNAP_OPTION_TYPE_DELETE, + GF_SNAP_OPTION_TYPE_RESTORE, + GF_SNAP_OPTION_TYPE_START, + GF_SNAP_OPTION_TYPE_STOP, + GF_SNAP_OPTION_TYPE_LIST, + GF_SNAP_OPTION_TYPE_STATUS, + GF_SNAP_OPTION_TYPE_CONFIG +}; + +enum gf1_cli_snapshot_config { + GF_SNAP_CONFIG_TYPE_NONE = 0, + GF_SNAP_CONFIG_TYPE_SET, + GF_SNAP_CONFIG_DISPLAY, + +}; + struct gf_cli_req { opaque dict<>; } ; @@ -125,32 +153,6 @@ enum gf_cli_status_type { opaque dict<>; } ; - struct gf1_cli_probe_req { - string hostname<>; - int port; -} ; - - struct gf1_cli_probe_rsp { - int op_ret; - int op_errno; - int port; - string hostname<>; - string op_errstr<>; -} ; - - struct gf1_cli_deprobe_req { - string hostname<>; - int port; - int flags; -} ; - - struct gf1_cli_deprobe_rsp { - int op_ret; - int op_errno; - string hostname<>; - string op_errstr<>; -} ; - struct gf1_cli_peer_list_req { int flags; opaque dict<>; diff --git a/rpc/xdr/src/glusterd1-xdr.c b/rpc/xdr/src/glusterd1-xdr.c index 213b48bc6..7fa98aaeb 100644 --- a/rpc/xdr/src/glusterd1-xdr.c +++ b/rpc/xdr/src/glusterd1-xdr.c @@ -491,3 +491,433 @@ xdr_gd1_mgmt_brick_op_rsp (XDR *xdrs, gd1_mgmt_brick_op_rsp *objp) return FALSE; return TRUE; } + +bool_t +xdr_gd1_mgmt_v3_lock_req (XDR *xdrs, gd1_mgmt_v3_lock_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_vector (xdrs, (char *)objp->txn_id, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_lock_rsp (XDR *xdrs, gd1_mgmt_v3_lock_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_vector (xdrs, (char *)objp->txn_id, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_pre_val_req (XDR *xdrs, gd1_mgmt_v3_pre_val_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_pre_val_rsp (XDR *xdrs, gd1_mgmt_v3_pre_val_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + + if (xdrs->x_op == XDR_ENCODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + IXDR_PUT_LONG(buf, objp->op); + IXDR_PUT_LONG(buf, objp->op_ret); + IXDR_PUT_LONG(buf, objp->op_errno); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } else if (xdrs->x_op == XDR_DECODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + objp->op = IXDR_GET_LONG(buf); + objp->op_ret = IXDR_GET_LONG(buf); + objp->op_errno = IXDR_GET_LONG(buf); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_brick_op_req (XDR *xdrs, gd1_mgmt_v3_brick_op_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_brick_op_rsp (XDR *xdrs, gd1_mgmt_v3_brick_op_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + + if (xdrs->x_op == XDR_ENCODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + IXDR_PUT_LONG(buf, objp->op); + IXDR_PUT_LONG(buf, objp->op_ret); + IXDR_PUT_LONG(buf, objp->op_errno); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } else if (xdrs->x_op == XDR_DECODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + objp->op = IXDR_GET_LONG(buf); + objp->op_ret = IXDR_GET_LONG(buf); + objp->op_errno = IXDR_GET_LONG(buf); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_commit_req (XDR *xdrs, gd1_mgmt_v3_commit_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_commit_rsp (XDR *xdrs, gd1_mgmt_v3_commit_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + + if (xdrs->x_op == XDR_ENCODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + IXDR_PUT_LONG(buf, objp->op); + IXDR_PUT_LONG(buf, objp->op_ret); + IXDR_PUT_LONG(buf, objp->op_errno); + } + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + return TRUE; + } else if (xdrs->x_op == XDR_DECODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + objp->op = IXDR_GET_LONG(buf); + objp->op_ret = IXDR_GET_LONG(buf); + objp->op_errno = IXDR_GET_LONG(buf); + } + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + return TRUE; + } + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_post_val_req (XDR *xdrs, gd1_mgmt_v3_post_val_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_post_val_rsp (XDR *xdrs, gd1_mgmt_v3_post_val_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + + if (xdrs->x_op == XDR_ENCODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + IXDR_PUT_LONG(buf, objp->op); + IXDR_PUT_LONG(buf, objp->op_ret); + IXDR_PUT_LONG(buf, objp->op_errno); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } else if (xdrs->x_op == XDR_DECODE) { + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + buf = XDR_INLINE (xdrs, 3 * BYTES_PER_XDR_UNIT); + if (buf == NULL) { + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + + } else { + objp->op = IXDR_GET_LONG(buf); + objp->op_ret = IXDR_GET_LONG(buf); + objp->op_errno = IXDR_GET_LONG(buf); + } + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; + } + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_string (xdrs, &objp->op_errstr, ~0)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_unlock_req (XDR *xdrs, gd1_mgmt_v3_unlock_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_vector (xdrs, (char *)objp->txn_id, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_int (xdrs, &objp->op)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gd1_mgmt_v3_unlock_rsp (XDR *xdrs, gd1_mgmt_v3_unlock_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_vector (xdrs, (char *)objp->uuid, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_vector (xdrs, (char *)objp->txn_id, 16, + sizeof (u_char), (xdrproc_t) xdr_u_char)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->dict.dict_val, (u_int *) &objp->dict.dict_len, ~0)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + return TRUE; +} diff --git a/rpc/xdr/src/glusterd1-xdr.h b/rpc/xdr/src/glusterd1-xdr.h index c35930cad..b6be23d06 100644 --- a/rpc/xdr/src/glusterd1-xdr.h +++ b/rpc/xdr/src/glusterd1-xdr.h @@ -202,6 +202,145 @@ struct gd1_mgmt_brick_op_rsp { }; typedef struct gd1_mgmt_brick_op_rsp gd1_mgmt_brick_op_rsp; +struct gd1_mgmt_v3_lock_req { + u_char uuid[16]; + u_char txn_id[16]; + int op; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_lock_req gd1_mgmt_v3_lock_req; + +struct gd1_mgmt_v3_lock_rsp { + u_char uuid[16]; + u_char txn_id[16]; + struct { + u_int dict_len; + char *dict_val; + } dict; + int op_ret; + int op_errno; +}; +typedef struct gd1_mgmt_v3_lock_rsp gd1_mgmt_v3_lock_rsp; + +struct gd1_mgmt_v3_pre_val_req { + u_char uuid[16]; + int op; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_pre_val_req gd1_mgmt_v3_pre_val_req; + +struct gd1_mgmt_v3_pre_val_rsp { + u_char uuid[16]; + int op; + int op_ret; + int op_errno; + char *op_errstr; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_pre_val_rsp gd1_mgmt_v3_pre_val_rsp; + +struct gd1_mgmt_v3_brick_op_req { + u_char uuid[16]; + int op; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_brick_op_req gd1_mgmt_v3_brick_op_req; + +struct gd1_mgmt_v3_brick_op_rsp { + u_char uuid[16]; + int op; + int op_ret; + int op_errno; + char *op_errstr; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_brick_op_rsp gd1_mgmt_v3_brick_op_rsp; + +struct gd1_mgmt_v3_commit_req { + u_char uuid[16]; + int op; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_commit_req gd1_mgmt_v3_commit_req; + +struct gd1_mgmt_v3_commit_rsp { + u_char uuid[16]; + int op; + int op_ret; + int op_errno; + struct { + u_int dict_len; + char *dict_val; + } dict; + char *op_errstr; +}; +typedef struct gd1_mgmt_v3_commit_rsp gd1_mgmt_v3_commit_rsp; + +struct gd1_mgmt_v3_post_val_req { + u_char uuid[16]; + int op; + int op_ret; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_post_val_req gd1_mgmt_v3_post_val_req; + +struct gd1_mgmt_v3_post_val_rsp { + u_char uuid[16]; + int op; + int op_ret; + int op_errno; + char *op_errstr; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_post_val_rsp gd1_mgmt_v3_post_val_rsp; + +struct gd1_mgmt_v3_unlock_req { + u_char uuid[16]; + u_char txn_id[16]; + int op; + struct { + u_int dict_len; + char *dict_val; + } dict; +}; +typedef struct gd1_mgmt_v3_unlock_req gd1_mgmt_v3_unlock_req; + +struct gd1_mgmt_v3_unlock_rsp { + u_char uuid[16]; + u_char txn_id[16]; + struct { + u_int dict_len; + char *dict_val; + } dict; + int op_ret; + int op_errno; +}; +typedef struct gd1_mgmt_v3_unlock_rsp gd1_mgmt_v3_unlock_rsp; + /* the xdr functions */ #if defined(__STDC__) || defined(__cplusplus) @@ -224,6 +363,18 @@ extern bool_t xdr_gd1_mgmt_friend_update (XDR *, gd1_mgmt_friend_update*); extern bool_t xdr_gd1_mgmt_friend_update_rsp (XDR *, gd1_mgmt_friend_update_rsp*); extern bool_t xdr_gd1_mgmt_brick_op_req (XDR *, gd1_mgmt_brick_op_req*); extern bool_t xdr_gd1_mgmt_brick_op_rsp (XDR *, gd1_mgmt_brick_op_rsp*); +extern bool_t xdr_gd1_mgmt_v3_lock_req (XDR *, gd1_mgmt_v3_lock_req*); +extern bool_t xdr_gd1_mgmt_v3_lock_rsp (XDR *, gd1_mgmt_v3_lock_rsp*); +extern bool_t xdr_gd1_mgmt_v3_pre_val_req (XDR *, gd1_mgmt_v3_pre_val_req*); +extern bool_t xdr_gd1_mgmt_v3_pre_val_rsp (XDR *, gd1_mgmt_v3_pre_val_rsp*); +extern bool_t xdr_gd1_mgmt_v3_brick_op_req (XDR *, gd1_mgmt_v3_brick_op_req*); +extern bool_t xdr_gd1_mgmt_v3_brick_op_rsp (XDR *, gd1_mgmt_v3_brick_op_rsp*); +extern bool_t xdr_gd1_mgmt_v3_commit_req (XDR *, gd1_mgmt_v3_commit_req*); +extern bool_t xdr_gd1_mgmt_v3_commit_rsp (XDR *, gd1_mgmt_v3_commit_rsp*); +extern bool_t xdr_gd1_mgmt_v3_post_val_req (XDR *, gd1_mgmt_v3_post_val_req*); +extern bool_t xdr_gd1_mgmt_v3_post_val_rsp (XDR *, gd1_mgmt_v3_post_val_rsp*); +extern bool_t xdr_gd1_mgmt_v3_unlock_req (XDR *, gd1_mgmt_v3_unlock_req*); +extern bool_t xdr_gd1_mgmt_v3_unlock_rsp (XDR *, gd1_mgmt_v3_unlock_rsp*); #else /* K&R C */ extern bool_t xdr_glusterd_volume_status (); @@ -245,6 +396,18 @@ extern bool_t xdr_gd1_mgmt_friend_update (); extern bool_t xdr_gd1_mgmt_friend_update_rsp (); extern bool_t xdr_gd1_mgmt_brick_op_req (); extern bool_t xdr_gd1_mgmt_brick_op_rsp (); +extern bool_t xdr_gd1_mgmt_v3_lock_req (); +extern bool_t xdr_gd1_mgmt_v3_lock_rsp (); +extern bool_t xdr_gd1_mgmt_v3_pre_val_req (); +extern bool_t xdr_gd1_mgmt_v3_pre_val_rsp (); +extern bool_t xdr_gd1_mgmt_v3_brick_op_req (); +extern bool_t xdr_gd1_mgmt_v3_brick_op_rsp (); +extern bool_t xdr_gd1_mgmt_v3_commit_req (); +extern bool_t xdr_gd1_mgmt_v3_commit_rsp (); +extern bool_t xdr_gd1_mgmt_v3_post_val_req (); +extern bool_t xdr_gd1_mgmt_v3_post_val_rsp (); +extern bool_t xdr_gd1_mgmt_v3_unlock_req (); +extern bool_t xdr_gd1_mgmt_v3_unlock_rsp (); #endif /* K&R C */ diff --git a/rpc/xdr/src/glusterd1-xdr.x b/rpc/xdr/src/glusterd1-xdr.x index fc1bb58b4..f5c45c9e4 100644 --- a/rpc/xdr/src/glusterd1-xdr.x +++ b/rpc/xdr/src/glusterd1-xdr.x @@ -125,3 +125,94 @@ struct gd1_mgmt_brick_op_rsp { opaque output<>; string op_errstr<>; } ; + +struct gd1_mgmt_v3_lock_req { + unsigned char uuid[16]; + unsigned char txn_id[16]; + int op; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_lock_rsp { + unsigned char uuid[16]; + unsigned char txn_id[16]; + opaque dict<>; + int op_ret; + int op_errno; +} ; + +struct gd1_mgmt_v3_pre_val_req { + unsigned char uuid[16]; + int op; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_pre_val_rsp { + unsigned char uuid[16]; + int op; + int op_ret; + int op_errno; + string op_errstr<>; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_brick_op_req { + unsigned char uuid[16]; + int op; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_brick_op_rsp { + unsigned char uuid[16]; + int op; + int op_ret; + int op_errno; + string op_errstr<>; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_commit_req { + unsigned char uuid[16]; + int op; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_commit_rsp { + unsigned char uuid[16]; + int op; + int op_ret; + int op_errno; + opaque dict<>; + string op_errstr<>; +} ; + +struct gd1_mgmt_v3_post_val_req { + unsigned char uuid[16]; + int op; + int op_ret; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_post_val_rsp { + unsigned char uuid[16]; + int op; + int op_ret; + int op_errno; + string op_errstr<>; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_unlock_req { + unsigned char uuid[16]; + unsigned char txn_id[16]; + int op; + opaque dict<>; +} ; + +struct gd1_mgmt_v3_unlock_rsp { + unsigned char uuid[16]; + unsigned char txn_id[16]; + opaque dict<>; + int op_ret; + int op_errno; +} ; diff --git a/rpc/xdr/src/glusterfs3-xdr.c b/rpc/xdr/src/glusterfs3-xdr.c index a502b2ea1..3205c551e 100644 --- a/rpc/xdr/src/glusterfs3-xdr.c +++ b/rpc/xdr/src/glusterfs3-xdr.c @@ -1507,6 +1507,125 @@ xdr_gfs3_fsetattr_rsp (XDR *xdrs, gfs3_fsetattr_rsp *objp) } bool_t +xdr_gfs3_fallocate_req (XDR *xdrs, gfs3_fallocate_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_opaque (xdrs, objp->gfid, 16)) + return FALSE; + if (!xdr_quad_t (xdrs, &objp->fd)) + return FALSE; + if (!xdr_u_int (xdrs, &objp->flags)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->offset)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->size)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gfs3_fallocate_rsp (XDR *xdrs, gfs3_fallocate_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpre)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpost)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gfs3_discard_req (XDR *xdrs, gfs3_discard_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_opaque (xdrs, objp->gfid, 16)) + return FALSE; + if (!xdr_quad_t (xdrs, &objp->fd)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->offset)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->size)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gfs3_discard_rsp (XDR *xdrs, gfs3_discard_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpre)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpost)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gfs3_zerofill_req (XDR *xdrs, gfs3_zerofill_req *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_opaque (xdrs, objp->gfid, 16)) + return FALSE; + if (!xdr_quad_t (xdrs, &objp->fd)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->offset)) + return FALSE; + if (!xdr_u_quad_t (xdrs, &objp->size)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, + (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + +bool_t +xdr_gfs3_zerofill_rsp (XDR *xdrs, gfs3_zerofill_rsp *objp) +{ + register int32_t *buf; + buf = NULL; + + if (!xdr_int (xdrs, &objp->op_ret)) + return FALSE; + if (!xdr_int (xdrs, &objp->op_errno)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpre)) + return FALSE; + if (!xdr_gf_iatt (xdrs, &objp->statpost)) + return FALSE; + if (!xdr_bytes (xdrs, (char **)&objp->xdata.xdata_val, + (u_int *) &objp->xdata.xdata_len, ~0)) + return FALSE; + return TRUE; +} + + +bool_t xdr_gfs3_rchecksum_req (XDR *xdrs, gfs3_rchecksum_req *objp) { register int32_t *buf; diff --git a/rpc/xdr/src/glusterfs3-xdr.h b/rpc/xdr/src/glusterfs3-xdr.h index 0268d1a14..13566e694 100644 --- a/rpc/xdr/src/glusterfs3-xdr.h +++ b/rpc/xdr/src/glusterfs3-xdr.h @@ -887,6 +887,80 @@ struct gfs3_fsetattr_rsp { }; typedef struct gfs3_fsetattr_rsp gfs3_fsetattr_rsp; +struct gfs3_fallocate_req { + char gfid[16]; + quad_t fd; + u_int flags; + u_quad_t offset; + u_quad_t size; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_fallocate_req gfs3_fallocate_req; + +struct gfs3_fallocate_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_fallocate_rsp gfs3_fallocate_rsp; + +struct gfs3_discard_req { + char gfid[16]; + quad_t fd; + u_quad_t offset; + u_quad_t size; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_discard_req gfs3_discard_req; + +struct gfs3_discard_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_discard_rsp gfs3_discard_rsp; + +struct gfs3_zerofill_req { + char gfid[16]; + quad_t fd; + u_quad_t offset; + u_quad_t size; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_zerofill_req gfs3_zerofill_req; + +struct gfs3_zerofill_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + struct { + u_int xdata_len; + char *xdata_val; + } xdata; +}; +typedef struct gfs3_zerofill_rsp gfs3_zerofill_rsp; + + struct gfs3_rchecksum_req { quad_t fd; u_quad_t offset; @@ -1182,6 +1256,12 @@ extern bool_t xdr_gfs3_setattr_req (XDR *, gfs3_setattr_req*); extern bool_t xdr_gfs3_setattr_rsp (XDR *, gfs3_setattr_rsp*); extern bool_t xdr_gfs3_fsetattr_req (XDR *, gfs3_fsetattr_req*); extern bool_t xdr_gfs3_fsetattr_rsp (XDR *, gfs3_fsetattr_rsp*); +extern bool_t xdr_gfs3_fallocate_req (XDR *, gfs3_fallocate_req*); +extern bool_t xdr_gfs3_fallocate_rsp (XDR *, gfs3_fallocate_rsp*); +extern bool_t xdr_gfs3_discard_req (XDR *, gfs3_discard_req*); +extern bool_t xdr_gfs3_discard_rsp (XDR *, gfs3_discard_rsp*); +extern bool_t xdr_gfs3_zerofill_req (XDR *, gfs3_zerofill_req*); +extern bool_t xdr_gfs3_zerofill_rsp (XDR *, gfs3_zerofill_rsp*); extern bool_t xdr_gfs3_rchecksum_req (XDR *, gfs3_rchecksum_req*); extern bool_t xdr_gfs3_rchecksum_rsp (XDR *, gfs3_rchecksum_rsp*); extern bool_t xdr_gf_setvolume_req (XDR *, gf_setvolume_req*); @@ -1276,6 +1356,12 @@ extern bool_t xdr_gfs3_setattr_req (); extern bool_t xdr_gfs3_setattr_rsp (); extern bool_t xdr_gfs3_fsetattr_req (); extern bool_t xdr_gfs3_fsetattr_rsp (); +extern bool_t xdr_gfs3_fallocate_req (); +extern bool_t xdr_gfs3_fallocate_rsp (); +extern bool_t xdr_gfs3_discard_req (); +extern bool_t xdr_gfs3_discard_rsp (); +extern bool_t xdr_gfs3_zerofill_req (); +extern bool_t xdr_gfs3_zerofill_rsp (); extern bool_t xdr_gfs3_rchecksum_req (); extern bool_t xdr_gfs3_rchecksum_rsp (); extern bool_t xdr_gf_setvolume_req (); diff --git a/rpc/xdr/src/glusterfs3-xdr.x b/rpc/xdr/src/glusterfs3-xdr.x index 063f302d9..1edbda3ad 100644 --- a/rpc/xdr/src/glusterfs3-xdr.x +++ b/rpc/xdr/src/glusterfs3-xdr.x @@ -566,6 +566,56 @@ struct gfs3_fstat_req { opaque xdata<>; /* Extra data */ } ; + struct gfs3_fallocate_req { + opaque gfid[16]; + hyper fd; + unsigned int flags; + unsigned hyper offset; + unsigned hyper size; + opaque xdata<>; /* Extra data */ +} ; + + struct gfs3_fallocate_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + opaque xdata<>; /* Extra data */ +} ; + + struct gfs3_discard_req { + opaque gfid[16]; + hyper fd; + unsigned hyper offset; + unsigned hyper size; + opaque xdata<>; /* Extra data */ +} ; + + struct gfs3_discard_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + opaque xdata<>; /* Extra data */ +} ; + + struct gfs3_zerofill_req { + opaque gfid[16]; + hyper fd; + unsigned hyper offset; + unsigned hyper size; + opaque xdata<>; +} ; + + struct gfs3_zerofill_rsp { + int op_ret; + int op_errno; + struct gf_iatt statpre; + struct gf_iatt statpost; + opaque xdata<>; +} ; + + struct gfs3_rchecksum_req { hyper fd; unsigned hyper offset; diff --git a/rpc/xdr/src/xdr-nfs3.h b/rpc/xdr/src/xdr-nfs3.h index 964632be1..6f6b0e1f9 100644 --- a/rpc/xdr/src/xdr-nfs3.h +++ b/rpc/xdr/src/xdr-nfs3.h @@ -1039,8 +1039,10 @@ typedef struct exportnode exportnode; #define MOUNT3_PROC_COUNT 6 #define MOUNT1_NULL 0 +#define MOUNT1_MNT 1 #define MOUNT1_DUMP 2 #define MOUNT1_UMNT 3 +#define MOUNT1_UMNTALL 4 #define MOUNT1_EXPORT 5 #define MOUNT1_PROC_COUNT 6 /* the xdr functions */ |