diff options
Diffstat (limited to 'rpc/rpc-lib/src')
| -rw-r--r-- | rpc/rpc-lib/src/Makefile.am | 14 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/auth-glusterfs.c | 93 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/auth-null.c | 28 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/auth-unix.c | 26 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/protocol-common.h | 122 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-clnt.c | 585 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-clnt.h | 49 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-drc.c | 898 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-drc.h | 106 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-transport.c | 150 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpc-transport.h | 38 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpcsvc-auth.c | 198 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpcsvc-common.h | 86 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpcsvc.c | 1200 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/rpcsvc.h | 122 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/xdr-common.h | 63 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/xdr-rpc.c | 21 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/xdr-rpc.h | 39 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/xdr-rpcclnt.c | 19 | ||||
| -rw-r--r-- | rpc/rpc-lib/src/xdr-rpcclnt.h | 19 |
20 files changed, 2752 insertions, 1124 deletions
diff --git a/rpc/rpc-lib/src/Makefile.am b/rpc/rpc-lib/src/Makefile.am index 8b087301c..ee2f27b25 100644 --- a/rpc/rpc-lib/src/Makefile.am +++ b/rpc/rpc-lib/src/Makefile.am @@ -1,16 +1,20 @@ 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 $(CONTRIBDIR)/sunrpc/xdr_sizeof.c libgfrpc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la +libgfrpc_la_LDFLAGS = -version-info $(LIBGFRPC_LT_VERSION) $(GF_LDFLAGS) noinst_HEADERS = rpcsvc.h rpc-transport.h xdr-common.h xdr-rpc.h xdr-rpcclnt.h \ - rpc-clnt.h rpcsvc-common.h protocol-common.h + rpc-clnt.h rpcsvc-common.h protocol-common.h rpc-drc.h -AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\ - -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS) \ +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) CLEANFILES = *~ diff --git a/rpc/rpc-lib/src/auth-glusterfs.c b/rpc/rpc-lib/src/auth-glusterfs.c index 9996bfad4..c3fc166b7 100644 --- a/rpc/rpc-lib/src/auth-glusterfs.c +++ b/rpc/rpc-lib/src/auth-glusterfs.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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. */ @@ -59,12 +50,6 @@ ret: int auth_glusterfs_request_init (rpcsvc_request_t *req, void *priv) { - if (!req) - return -1; - memset (req->verf.authdata, 0, GF_MAX_AUTH_BYTES); - req->verf.datalen = 0; - req->verf.flavour = AUTH_NULL; - return 0; } @@ -73,9 +58,9 @@ int auth_glusterfs_authenticate (rpcsvc_request_t *req, void *priv) struct auth_glusterfs_parms au = {0,}; int ret = RPCSVC_AUTH_REJECT; - int gidcount = 0; int j = 0; int i = 0; + int gidcount = 0; if (!req) return ret; @@ -105,9 +90,26 @@ 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]; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Auth Info: pid: %u, uid: %d" ", gid: %d, owner: %s", req->pid, req->uid, req->gid, lkowner_utoa (&req->lk_owner)); @@ -163,20 +165,16 @@ ret: int auth_glusterfs_v2_request_init (rpcsvc_request_t *req, void *priv) { - if (!req) - return -1; - memset (req->verf.authdata, 0, GF_MAX_AUTH_BYTES); - req->verf.datalen = 0; - req->verf.flavour = AUTH_NULL; - return 0; } int auth_glusterfs_v2_authenticate (rpcsvc_request_t *req, void *priv) { struct auth_glusterfs_parms_v2 au = {0,}; - int ret = RPCSVC_AUTH_REJECT; - int i = 0; + int ret = RPCSVC_AUTH_REJECT; + int i = 0; + int max_groups = 0; + int max_lk_owner_len = 0; if (!req) return ret; @@ -195,32 +193,59 @@ int auth_glusterfs_v2_authenticate (rpcsvc_request_t *req, void *priv) req->lk_owner.len = au.lk_owner.lk_owner_len; req->auxgidcount = au.groups.groups_len; - if (req->auxgidcount > GF_MAX_AUX_GROUPS) { + /* the number of groups and size of lk_owner depend on each other */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS (req->lk_owner.len); + max_lk_owner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER (req->auxgidcount); + + if (req->auxgidcount > max_groups) { gf_log ("", GF_LOG_WARNING, "more than max aux gids found (%d) , truncating it " "to %d and continuing", au.groups.groups_len, - GF_MAX_AUX_GROUPS); - req->auxgidcount = GF_MAX_AUX_GROUPS; + max_groups); + req->auxgidcount = max_groups; } - if (req->lk_owner.len > GF_MAX_LOCK_OWNER_LEN) { + if (req->lk_owner.len > max_lk_owner_len) { gf_log ("", GF_LOG_WARNING, - "lkowner field > 1k, failing authentication"); + "lkowner field to big (%d), depends on the number of " + "groups (%d), failing authentication", + req->lk_owner.len, req->auxgidcount); ret = RPCSVC_AUTH_REJECT; 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]; for (i = 0; i < au.lk_owner.lk_owner_len; ++i) req->lk_owner.data[i] = au.lk_owner.lk_owner_val[i]; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Auth Info: pid: %u, uid: %d" ", gid: %d, owner: %s", req->pid, req->uid, req->gid, lkowner_utoa (&req->lk_owner)); ret = RPCSVC_AUTH_ACCEPT; err: + /* TODO: instead use alloca() for these variables */ + free (au.groups.groups_val); + free (au.lk_owner.lk_owner_val); + return ret; } diff --git a/rpc/rpc-lib/src/auth-null.c b/rpc/rpc-lib/src/auth-null.c index aa6b4c1bd..b030341ab 100644 --- a/rpc/rpc-lib/src/auth-null.c +++ b/rpc/rpc-lib/src/auth-null.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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. */ @@ -31,15 +22,6 @@ int auth_null_request_init (rpcsvc_request_t *req, void *priv) { - if (!req) - return -1; - - memset (req->cred.authdata, 0, GF_MAX_AUTH_BYTES); - req->cred.datalen = 0; - - memset (req->verf.authdata, 0, GF_MAX_AUTH_BYTES); - req->verf.datalen = 0; - return 0; } diff --git a/rpc/rpc-lib/src/auth-unix.c b/rpc/rpc-lib/src/auth-unix.c index c3b58945a..27351f669 100644 --- a/rpc/rpc-lib/src/auth-unix.c +++ b/rpc/rpc-lib/src/auth-unix.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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. */ @@ -33,12 +24,6 @@ int auth_unix_request_init (rpcsvc_request_t *req, void *priv) { - if (!req) - return -1; - memset (req->verf.authdata, 0, GF_MAX_AUTH_BYTES); - req->verf.datalen = 0; - req->verf.flavour = AUTH_NULL; - return 0; } @@ -51,6 +36,7 @@ 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 971996091..99d5a6259 100644 --- a/rpc/rpc-lib/src/protocol-common.h +++ b/rpc/rpc-lib/src/protocol-common.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2007-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _PROTOCOL_COMMON_H @@ -65,6 +56,10 @@ enum gf_fop_procnum { GFS3_OP_RELEASE, GFS3_OP_RELEASEDIR, GFS3_OP_FREMOVEXATTR, + GFS3_OP_FALLOCATE, + GFS3_OP_DISCARD, + GFS3_OP_ZEROFILL, + GFS3_OP_IPC, GFS3_OP_MAXVALUE, } ; @@ -73,6 +68,8 @@ enum gf_handshake_procnum { GF_HNDSK_SETVOLUME, GF_HNDSK_GETSPEC, GF_HNDSK_PING, + GF_HNDSK_SET_LK_VER, + GF_HNDSK_EVENT_NOTIFY, GF_HNDSK_MAXVALUE, }; @@ -86,6 +83,13 @@ enum gf_pmap_procnum { GF_PMAP_MAXVALUE, }; +enum gf_aggregator_procnum { + GF_AGGREGATOR_NULL = 0, + GF_AGGREGATOR_LOOKUP, + GF_AGGREGATOR_GETLIMIT, + GF_AGGREGATOR_MAXVALUE, +}; + enum gf_pmap_port_type { GF_PMAP_PORT_FREE = 0, GF_PMAP_PORT_FOREIGN, @@ -101,8 +105,12 @@ enum gf_probe_resp { GF_PROBE_FRIEND, GF_PROBE_ANOTHER_CLUSTER, GF_PROBE_VOLUME_CONFLICT, + GF_PROBE_SAME_UUID, GF_PROBE_UNKNOWN_PEER, - GF_PROBE_ADD_FAILED + GF_PROBE_ADD_FAILED, + GF_PROBE_QUORUM_NOT_MET, + GF_PROBE_MISSED_SNAP_CONFLICT, + GF_PROBE_SNAP_CONFLICT, }; enum gf_deprobe_resp { @@ -110,13 +118,15 @@ enum gf_deprobe_resp { GF_DEPROBE_LOCALHOST, GF_DEPROBE_NOT_FRIEND, GF_DEPROBE_BRICK_EXIST, - GF_DEPROBE_FRIEND_DOWN + GF_DEPROBE_FRIEND_DOWN, + GF_DEPROBE_QUORUM_NOT_MET, }; enum gf_cbk_procnum { GF_CBK_NULL = 0, GF_CBK_FETCHSPEC, GF_CBK_INO_FLUSH, + GF_CBK_EVENT_NOTIFY, GF_CBK_MAXVALUE, }; @@ -154,6 +164,13 @@ enum gluster_cli_procnum { GLUSTER_CLI_UMOUNT, GLUSTER_CLI_HEAL_VOLUME, GLUSTER_CLI_STATEDUMP_VOLUME, + GLUSTER_CLI_LIST_VOLUME, + GLUSTER_CLI_CLRLOCKS_VOLUME, + GLUSTER_CLI_UUID_RESET, + GLUSTER_CLI_UUID_GET, + GLUSTER_CLI_COPY_FILE, + GLUSTER_CLI_SYS_EXEC, + GLUSTER_CLI_SNAP, GLUSTER_CLI_MAXVALUE, }; @@ -179,14 +196,66 @@ enum glusterd_brick_procnum { GLUSTERD_BRICK_NULL, /* 0 */ GLUSTERD_BRICK_TERMINATE, GLUSTERD_BRICK_XLATOR_INFO, - GLUSTERD_BRICK_XLATOR_HEAL, + GLUSTERD_BRICK_XLATOR_OP, GLUSTERD_BRICK_STATUS, GLUSTERD_BRICK_OP, + GLUSTERD_BRICK_XLATOR_DEFRAG, + GLUSTERD_NODE_PROFILE, + GLUSTERD_NODE_STATUS, + GLUSTERD_VOLUME_BARRIER_OP, GLUSTERD_BRICK_MAXVALUE, }; +enum glusterd_mgmt_hndsk_procnum { + GD_MGMT_HNDSK_NULL, + GD_MGMT_HNDSK_VERSIONS, + GD_MGMT_HNDSK_VERSIONS_ACK, + GD_MGMT_HNDSK_MAXVALUE, +}; + +typedef enum { + GF_AFR_OP_INVALID, + GF_AFR_OP_HEAL_INDEX, + GF_AFR_OP_HEAL_FULL, + 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_STATISTICS, + GF_AFR_OP_STATISTICS_HEAL_COUNT, + GF_AFR_OP_STATISTICS_HEAL_COUNT_PER_REPLICA, +} gf_xl_afr_op_t ; + +struct gf_gsync_detailed_status_ { + char node[NAME_MAX]; + char master[NAME_MAX]; + char brick[NAME_MAX]; + char slave_node[NAME_MAX]; + char worker_status[NAME_MAX]; + char checkpoint_status[NAME_MAX]; + char crawl_status[NAME_MAX]; + char files_syncd[NAME_MAX]; + char files_remaining[NAME_MAX]; + char bytes_remaining[NAME_MAX]; + char purges_remaining[NAME_MAX]; + char total_files_skipped[NAME_MAX]; +}; + +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, +}; + +typedef struct gf_gsync_detailed_status_ gf_gsync_status_t; + #define GLUSTER_HNDSK_PROGRAM 14398633 /* Completely random */ -#define GLUSTER_HNDSK_VERSION 1 /* 0.0.1 */ +#define GLUSTER_HNDSK_VERSION 2 /* 0.0.2 */ #define GLUSTER_PMAP_PROGRAM 34123456 #define GLUSTER_PMAP_VERSION 1 @@ -194,9 +263,13 @@ enum glusterd_brick_procnum { #define GLUSTER_CBK_PROGRAM 52743234 /* Completely random */ #define GLUSTER_CBK_VERSION 1 /* 0.0.1 */ -#define GLUSTER3_1_FOP_PROGRAM 1298437 /* Completely random */ -#define GLUSTER3_1_FOP_VERSION 330 /* 3.3.0 */ -#define GLUSTER3_1_FOP_PROCCNT GFS3_OP_MAXVALUE +#define GLUSTER_FOP_PROGRAM 1298437 /* Completely random */ +#define GLUSTER_FOP_VERSION 330 /* 3.3.0 */ +#define GLUSTER_FOP_PROCCNT GFS3_OP_MAXVALUE + +/* Aggregator */ +#define GLUSTER_AGGREGATOR_PROGRAM 29852134 /* Completely random */ +#define GLUSTER_AGGREGATOR_VERSION 1 /* Second version */ #define GD_MGMT_PROGRAM 1238433 /* Completely random */ @@ -211,4 +284,11 @@ enum glusterd_brick_procnum { #define GD_BRICK_PROGRAM 4867634 /*Completely random*/ #define GD_BRICK_VERSION 2 +/* Third version */ +#define GD_MGMT_V3_VERSION 3 + +/* OP-VERSION handshake */ +#define GD_MGMT_HNDSK_PROGRAM 1239873 /* Completely random */ +#define GD_MGMT_HNDSK_VERSION 1 + #endif /* !_PROTOCOL_COMMON_H */ diff --git a/rpc/rpc-lib/src/rpc-clnt.c b/rpc/rpc-lib/src/rpc-clnt.c index e21aac755..3106342e9 100644 --- a/rpc/rpc-lib/src/rpc-clnt.c +++ b/rpc/rpc-lib/src/rpc-clnt.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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. */ @@ -23,7 +14,7 @@ #include "config.h" #endif -#define RPC_CLNT_DEFAULT_REQUEST_COUNT 4096 +#define RPC_CLNT_DEFAULT_REQUEST_COUNT 512 #include "rpc-clnt.h" #include "byte-order.h" @@ -75,8 +66,8 @@ _is_lock_fop (struct saved_frame *sframe) { int fop = 0; - if (SFRAME_GET_PROGNUM (sframe) == GLUSTER3_1_FOP_PROGRAM && - SFRAME_GET_PROGVER (sframe) == GLUSTER3_1_FOP_VERSION) + if (SFRAME_GET_PROGNUM (sframe) == GLUSTER_FOP_PROGRAM && + SFRAME_GET_PROGVER (sframe) == GLUSTER_FOP_VERSION) fop = SFRAME_GET_PROCNUM (sframe); return ((fop == GFS3_OP_LK) || @@ -146,6 +137,7 @@ out: static void call_bail (void *data) { + rpc_transport_t *trans = NULL; struct rpc_clnt *clnt = NULL; rpc_clnt_connection_t *conn = NULL; struct timeval current; @@ -153,16 +145,30 @@ call_bail (void *data) struct saved_frame *saved_frame = NULL; struct saved_frame *trav = NULL; struct saved_frame *tmp = NULL; - struct tm frame_sent_tm; char frame_sent[256] = {0,}; - struct timeval timeout = {0,}; + struct timespec timeout = {0,}; struct iovec iov = {0,}; + char peerid[UNIX_PATH_MAX] = {0}; GF_VALIDATE_OR_GOTO ("client", data, out); clnt = data; conn = &clnt->conn; + pthread_mutex_lock (&conn->lock); + { + trans = conn->trans; + if (trans) { + strncpy (peerid, conn->trans->peerinfo.identifier, + sizeof (peerid)-1); + + } + } + pthread_mutex_unlock (&conn->lock); + /*rpc_clnt_connection_cleanup will be unwinding all saved frames, + * bailed or otherwise*/ + if (!trans) + goto out; gettimeofday (¤t, NULL); INIT_LIST_HEAD (&list); @@ -173,7 +179,7 @@ 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, @@ -182,8 +188,9 @@ call_bail (void *data) (void *) clnt); if (conn->timer == NULL) { - gf_log (conn->trans->name, GF_LOG_WARNING, - "Cannot create bailout timer"); + gf_log (conn->name, GF_LOG_WARNING, + "Cannot create bailout timer for %s", + peerid); } } @@ -200,21 +207,21 @@ call_bail (void *data) pthread_mutex_unlock (&conn->lock); list_for_each_entry_safe (trav, tmp, &list, list) { - localtime_r (&trav->saved_at.tv_sec, &frame_sent_tm); - strftime (frame_sent, 32, "%Y-%m-%d %H:%M:%S", &frame_sent_tm); + gf_time_fmt (frame_sent, sizeof frame_sent, + trav->saved_at.tv_sec, gf_timefmt_FT); snprintf (frame_sent + strlen (frame_sent), 256 - strlen (frame_sent), ".%"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", + gf_log (conn->name, GF_LOG_ERROR, + "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, peerid); clnt = rpc_clnt_ref (clnt); trav->rpcreq->rpc_status = -1; @@ -236,7 +243,7 @@ __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; @@ -250,7 +257,7 @@ __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, @@ -349,20 +356,16 @@ out: void saved_frames_unwind (struct saved_frames *saved_frames) { - struct rpc_clnt *clnt = NULL; struct saved_frame *trav = NULL; struct saved_frame *tmp = NULL; - struct tm *frame_sent_tm = NULL; - char timestr[256] = {0,}; - + char timestr[1024] = {0,}; struct iovec iov = {0,}; list_splice_init (&saved_frames->lk_sf.list, &saved_frames->sf.list); list_for_each_entry_safe (trav, tmp, &saved_frames->sf.list, list) { - frame_sent_tm = localtime (&trav->saved_at.tv_sec); - strftime (timestr, sizeof(timestr), "%Y-%m-%d %H:%M:%S", - frame_sent_tm); + gf_time_fmt (timestr, sizeof timestr, + trav->saved_at.tv_sec, gf_timefmt_FT); snprintf (timestr + strlen (timestr), sizeof(timestr) - strlen (timestr), ".%"GF_PRI_SUSECONDS, trav->saved_at.tv_usec); @@ -370,25 +373,24 @@ saved_frames_unwind (struct saved_frames *saved_frames) if (!trav->rpcreq || !trav->rpcreq->prog) continue; - gf_log_callingfn (trav->rpcreq->conn->trans->name, + gf_log_callingfn (trav->rpcreq->conn->name, GF_LOG_ERROR, "forced unwinding frame type(%s) op(%s(%d)) " - "called at %s", + "called at %s (xid=0x%x)", trav->rpcreq->prog->progname, ((trav->rpcreq->prog->procnames) ? trav->rpcreq->prog->procnames[trav->rpcreq->procnum] : "--"), - trav->rpcreq->procnum, timestr); + trav->rpcreq->procnum, timestr, + trav->rpcreq->xid); saved_frames->count--; - clnt = rpc_clnt_ref (trav->rpcreq->conn->rpc_clnt); trav->rpcreq->rpc_status = -1; trav->rpcreq->cbkfn (trav->rpcreq, &iov, 1, trav->frame); rpc_clnt_reply_deinit (trav->rpcreq, trav->rpcreq->conn->rpc_clnt->reqpool); - clnt = rpc_clnt_unref (clnt); list_del_init (&trav->list); mem_put (trav); } @@ -408,50 +410,43 @@ saved_frames_destroy (struct saved_frames *frames) void -rpc_clnt_reconnect (void *trans_ptr) +rpc_clnt_reconnect (void *conn_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; - trans = trans_ptr; - if (!trans || !trans->mydata) - return; - - conn = trans->mydata; + conn = conn_ptr; clnt = conn->rpc_clnt; pthread_mutex_lock (&conn->lock); { + trans = conn->trans; + if (!trans) { + pthread_mutex_unlock (&conn->lock); + return; + } if (conn->reconnect) gf_timer_call_cancel (clnt->ctx, conn->reconnect); conn->reconnect = 0; - if (conn->connected == 0) { - tv.tv_sec = 3; + if ((conn->connected == 0) && !clnt->disabled) { + ts.tv_sec = 3; + ts.tv_nsec = 0; - gf_log (trans->name, GF_LOG_TRACE, + gf_log (conn->name, GF_LOG_TRACE, "attempting reconnect"); ret = rpc_transport_connect (trans, conn->config.remote_port); - /* Every time there is a disconnection, processes - should try to connect to 'glusterd' (ie, default - port) or whichever port given as 'option remote-port' - in volume file. */ - /* Below code makes sure the (re-)configured port lasts - for just one successful attempt */ - if (!ret) - conn->config.remote_port = 0; - conn->reconnect = - gf_timer_call_after (clnt->ctx, tv, + gf_timer_call_after (clnt->ctx, ts, rpc_clnt_reconnect, - trans); + conn); } else { - gf_log (trans->name, GF_LOG_TRACE, + gf_log (conn->name, GF_LOG_TRACE, "breaking reconnect chain"); } } @@ -468,7 +463,7 @@ rpc_clnt_reconnect (void *trans_ptr) int rpc_clnt_fill_request_info (struct rpc_clnt *clnt, rpc_request_info_t *info) { - struct saved_frame saved_frame = {{}, 0}; + struct saved_frame saved_frame; int ret = -1; pthread_mutex_lock (&clnt->conn.lock); @@ -479,7 +474,7 @@ rpc_clnt_fill_request_info (struct rpc_clnt *clnt, rpc_request_info_t *info) pthread_mutex_unlock (&clnt->conn.lock); if (ret == -1) { - gf_log (clnt->conn.trans->name, GF_LOG_CRITICAL, + gf_log (clnt->conn.name, GF_LOG_CRITICAL, "cannot lookup the saved " "frame corresponding to xid (%d)", info->xid); goto out; @@ -539,11 +534,9 @@ rpc_clnt_connection_cleanup (rpc_clnt_connection_t *conn) clnt = conn->rpc_clnt; - gf_log (conn->trans->name, GF_LOG_TRACE, - "cleaning up state in transport object %p", conn->trans); - pthread_mutex_lock (&conn->lock); { + saved_frames = conn->saved_frames; conn->saved_frames = saved_frames_new (); @@ -554,6 +547,12 @@ rpc_clnt_connection_cleanup (rpc_clnt_connection_t *conn) } conn->connected = 0; + + if (conn->ping_timer) { + gf_timer_call_cancel (clnt->ctx, conn->ping_timer); + conn->ping_timer = NULL; + conn->ping_started = 0; + } } pthread_mutex_unlock (&conn->lock); @@ -667,7 +666,7 @@ rpc_clnt_reply_init (rpc_clnt_connection_t *conn, rpc_transport_pollin_t *msg, ret = xdr_to_rpc_reply (msgbuf, msglen, &rpcmsg, &progmsg, req->verf.authdata); if (ret != 0) { - gf_log (conn->trans->name, GF_LOG_WARNING, + gf_log (conn->name, GF_LOG_WARNING, "RPC reply decoding failed"); goto out; } @@ -678,15 +677,13 @@ rpc_clnt_reply_init (rpc_clnt_connection_t *conn, rpc_transport_pollin_t *msg, goto out; } - gf_log (conn->trans->name, GF_LOG_TRACE, - "received rpc message (RPC XID: 0x%ux" + gf_log (conn->name, GF_LOG_TRACE, + "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, saved_frame->rpcreq->prog->progver, - saved_frame->rpcreq->procnum, conn->trans->name); - - req->rpc_status = 0; + saved_frame->rpcreq->procnum, conn->name); out: if (ret != 0) { @@ -714,18 +711,19 @@ rpc_clnt_handle_cbk (struct rpc_clnt *clnt, rpc_transport_pollin_t *msg) clnt = rpc_clnt_ref (clnt); ret = xdr_to_rpc_call (msgbuf, msglen, &rpcmsg, &progmsg, NULL,NULL); if (ret == -1) { - gf_log (clnt->conn.trans->name, GF_LOG_WARNING, + gf_log (clnt->conn.name, GF_LOG_WARNING, "RPC call decoding failed"); goto out; } - gf_log (clnt->conn.trans->name, GF_LOG_TRACE, - "received rpc message (XID: 0x%lx, " - "Ver: %ld, Program: %ld, ProgVers: %ld, Proc: %ld) " + gf_log (clnt->conn.name, GF_LOG_TRACE, + "receivd rpc message (XID: 0x%" GF_PRI_RPC_XID ", " + "Ver: %" GF_PRI_RPC_VERSION ", Program: %" GF_PRI_RPC_PROG_ID ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC ") " "from rpc-transport (%s)", rpc_call_xid (&rpcmsg), rpc_call_rpcvers (&rpcmsg), rpc_call_program (&rpcmsg), rpc_call_progver (&rpcmsg), rpc_call_progproc (&rpcmsg), - clnt->conn.trans->name); + clnt->conn.name); procnum = rpc_call_progproc (&rpcmsg); @@ -744,7 +742,8 @@ rpc_clnt_handle_cbk (struct rpc_clnt *clnt, rpc_transport_pollin_t *msg) if (found && (procnum < program->numactors) && (program->actors[procnum].actor)) { - program->actors[procnum].actor (&progmsg); + program->actors[procnum].actor (clnt, program->mydata, + &progmsg); } out: @@ -767,7 +766,7 @@ rpc_clnt_handle_reply (struct rpc_clnt *clnt, rpc_transport_pollin_t *pollin) xid = ntoh32 (*((uint32_t *)pollin->vector[0].iov_base)); saved_frame = lookup_frame (conn, xid); if (saved_frame == NULL) { - gf_log (conn->trans->name, GF_LOG_ERROR, + gf_log (conn->name, GF_LOG_ERROR, "cannot lookup the saved frame for reply with xid (%u)", xid); goto out; @@ -775,7 +774,7 @@ rpc_clnt_handle_reply (struct rpc_clnt *clnt, rpc_transport_pollin_t *pollin) req = saved_frame->rpcreq; if (req == NULL) { - gf_log (conn->trans->name, GF_LOG_ERROR, + gf_log (conn->name, GF_LOG_ERROR, "no request with frame for xid (%u)", xid); goto out; } @@ -783,7 +782,7 @@ rpc_clnt_handle_reply (struct rpc_clnt *clnt, rpc_transport_pollin_t *pollin) ret = rpc_clnt_reply_init (conn, pollin, req, saved_frame); if (ret != 0) { req->rpc_status = -1; - gf_log (conn->trans->name, GF_LOG_WARNING, + gf_log (conn->name, GF_LOG_WARNING, "initialising rpc reply failed"); } @@ -838,17 +837,20 @@ 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, ...) { - rpc_clnt_connection_t *conn = NULL; - struct rpc_clnt *clnt = NULL; - int ret = -1; - rpc_request_info_t *req_info = NULL; - rpc_transport_pollin_t *pollin = NULL; - struct timeval tv = {0, }; + rpc_clnt_connection_t *conn = NULL; + struct rpc_clnt *clnt = NULL; + int ret = -1; + rpc_request_info_t *req_info = NULL; + rpc_transport_pollin_t *pollin = NULL; + struct timespec ts = {0, }; + void *clnt_mydata = NULL; conn = mydata; if (conn == NULL) { @@ -865,13 +867,15 @@ rpc_clnt_notify (rpc_transport_t *trans, void *mydata, pthread_mutex_lock (&conn->lock); { - if (conn->reconnect == NULL) { - tv.tv_sec = 10; + if (!conn->rpc_clnt->disabled + && (conn->reconnect == NULL)) { + 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); + conn); } } pthread_mutex_unlock (&conn->lock); @@ -883,9 +887,13 @@ 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. - */ + if (clnt->notifyfn) { + clnt_mydata = clnt->mydata; + clnt->mydata = NULL; + ret = clnt->notifyfn (clnt, clnt_mydata, + RPC_CLNT_DESTROY, NULL); + } + rpc_clnt_destroy (clnt); ret = 0; break; @@ -929,6 +937,14 @@ rpc_clnt_notify (rpc_transport_t *trans, void *mydata, case RPC_TRANSPORT_CONNECT: { + /* Every time there is a disconnection, processes + should try to connect to 'glusterd' (ie, default + port) or whichever port given as 'option remote-port' + in volume file. */ + /* Below code makes sure the (re-)configured port lasts + for just one successful attempt */ + conn->config.remote_port = 0; + if (clnt->notifyfn) ret = clnt->notifyfn (clnt, clnt->mydata, RPC_CLNT_CONNECT, NULL); @@ -955,16 +971,23 @@ rpc_clnt_connection_deinit (rpc_clnt_connection_t *conn) } -inline int +static inline int rpc_clnt_connection_init (struct rpc_clnt *clnt, glusterfs_ctx_t *ctx, dict_t *options, char *name) { int ret = -1; rpc_clnt_connection_t *conn = NULL; + rpc_transport_t *trans = NULL; conn = &clnt->conn; pthread_mutex_init (&clnt->conn.lock, NULL); + conn->name = gf_strdup (name); + if (!conn->name) { + ret = -1; + goto out; + } + ret = dict_get_int32 (options, "frame-timeout", &conn->frame_timeout); if (ret >= 0) { @@ -975,25 +998,28 @@ rpc_clnt_connection_init (struct rpc_clnt *clnt, glusterfs_ctx_t *ctx, "defaulting frame-timeout to 30mins"); conn->frame_timeout = 1800; } + conn->rpc_clnt = clnt; - conn->trans = rpc_transport_load (ctx, options, name); - if (!conn->trans) { + trans = rpc_transport_load (ctx, options, name); + if (!trans) { gf_log (name, GF_LOG_WARNING, "loading of new rpc-transport" " failed"); ret = -1; goto out; } + rpc_transport_ref (trans); - rpc_transport_ref (conn->trans); - - conn->rpc_clnt = clnt; + pthread_mutex_lock (&conn->lock); + { + conn->trans = trans; + trans = NULL; + } + pthread_mutex_unlock (&conn->lock); ret = rpc_transport_register_notify (conn->trans, rpc_clnt_notify, conn); if (ret == -1) { gf_log (name, GF_LOG_WARNING, "registering notify failed"); - rpc_clnt_connection_cleanup (conn); - conn = NULL; goto out; } @@ -1001,19 +1027,32 @@ rpc_clnt_connection_init (struct rpc_clnt *clnt, glusterfs_ctx_t *ctx, if (!conn->saved_frames) { gf_log (name, GF_LOG_WARNING, "creation of saved_frames " "failed"); - rpc_clnt_connection_cleanup (conn); + ret = -1; goto out; } ret = 0; out: + if (ret) { + pthread_mutex_lock (&conn->lock); + { + trans = conn->trans; + conn->trans = NULL; + } + pthread_mutex_unlock (&conn->lock); + if (trans) + rpc_transport_unref (trans); + //conn cleanup needs to be done since we might have failed to + // register notification. + rpc_clnt_connection_cleanup (conn); + } return ret; } struct rpc_clnt * -rpc_clnt_new (dict_t *options, - glusterfs_ctx_t *ctx, char *name) +rpc_clnt_new (dict_t *options, glusterfs_ctx_t *ctx, char *name, + uint32_t reqpool_size) { int ret = -1; struct rpc_clnt *rpc = NULL; @@ -1026,8 +1065,10 @@ rpc_clnt_new (dict_t *options, pthread_mutex_init (&rpc->lock, NULL); rpc->ctx = ctx; - rpc->reqpool = mem_pool_new (struct rpc_req, - RPC_CLNT_DEFAULT_REQUEST_COUNT); + if (!reqpool_size) + reqpool_size = RPC_CLNT_DEFAULT_REQUEST_COUNT; + + rpc->reqpool = mem_pool_new (struct rpc_req, reqpool_size); if (rpc->reqpool == NULL) { pthread_mutex_destroy (&rpc->lock); GF_FREE (rpc); @@ -1036,7 +1077,7 @@ rpc_clnt_new (dict_t *options, } rpc->saved_frames_pool = mem_pool_new (struct saved_frame, - RPC_CLNT_DEFAULT_REQUEST_COUNT); + reqpool_size); if (rpc->saved_frames_pool == NULL) { pthread_mutex_destroy (&rpc->lock); mem_pool_destroy (rpc->reqpool); @@ -1057,6 +1098,8 @@ rpc_clnt_new (dict_t *options, goto out; } + rpc->auth_null = dict_get_str_boolean (options, "auth-null", 0); + rpc = rpc_clnt_ref (rpc); INIT_LIST_HEAD (&rpc->programs); @@ -1075,7 +1118,7 @@ rpc_clnt_start (struct rpc_clnt *rpc) conn = &rpc->conn; - rpc_clnt_reconnect (conn->trans); + rpc_clnt_reconnect (conn); return 0; } @@ -1091,17 +1134,34 @@ rpc_clnt_register_notify (struct rpc_clnt *rpc, rpc_clnt_notify_t fn, return 0; } +/* used for GF_LOG_OCCASIONALLY() */ +static int gf_auth_max_groups_log = 0; + ssize_t xdr_serialize_glusterfs_auth (char *dest, struct auth_glusterfs_parms_v2 *au) { ssize_t ret = -1; XDR xdr; + u_long ngroups = 0; + int max_groups = 0; if ((!dest) || (!au)) return -1; + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS (au->lk_owner.lk_owner_len); + xdrmem_create (&xdr, dest, GF_MAX_AUTH_BYTES, XDR_ENCODE); + if (au->groups.groups_len > max_groups) { + ngroups = au->groups.groups_len; + au->groups.groups_len = max_groups; + + GF_LOG_OCCASIONALLY (gf_auth_max_groups_log, + THIS->name, GF_LOG_WARNING, + "too many groups, reducing %ld -> %d", + ngroups, max_groups); + } + if (!xdr_auth_glusterfs_parms_v2 (&xdr, au)) { gf_log (THIS->name, GF_LOG_WARNING, "failed to encode auth glusterfs elements"); @@ -1112,12 +1172,15 @@ xdr_serialize_glusterfs_auth (char *dest, struct auth_glusterfs_parms_v2 *au) ret = (((size_t)(&xdr)->x_private) - ((size_t)(&xdr)->x_base)); ret: + if (ngroups) + au->groups.groups_len = ngroups; + return ret; } int -rpc_clnt_fill_request (int prognum, int progver, int procnum, int payload, +rpc_clnt_fill_request (int prognum, int progver, int procnum, uint64_t xid, struct auth_glusterfs_parms_v2 *au, struct rpc_msg *request, char *auth_data) { @@ -1137,19 +1200,26 @@ rpc_clnt_fill_request (int prognum, int progver, int procnum, int payload, request->rm_call.cb_vers = progver; request->rm_call.cb_proc = procnum; - /* TODO: Using AUTH_GLUSTERFS for time-being. Make it modular in - * future so it is easy to plug-in new authentication schemes. + /* TODO: Using AUTH_(GLUSTERFS/NULL) in a kludgy way for time-being. + * Make it modular in future so it is easy to plug-in new + * authentication schemes. */ - ret = xdr_serialize_glusterfs_auth (auth_data, au); - if (ret == -1) { - gf_log ("rpc-clnt", GF_LOG_DEBUG, "cannot encode credentials"); - goto out; - } - - request->rm_call.cb_cred.oa_flavor = AUTH_GLUSTERFS_v2; - request->rm_call.cb_cred.oa_base = auth_data; - request->rm_call.cb_cred.oa_length = ret; + if (auth_data) { + ret = xdr_serialize_glusterfs_auth (auth_data, au); + if (ret == -1) { + gf_log ("rpc-clnt", GF_LOG_DEBUG, + "cannot encode credentials"); + goto out; + } + request->rm_call.cb_cred.oa_flavor = AUTH_GLUSTERFS_v2; + request->rm_call.cb_cred.oa_base = auth_data; + request->rm_call.cb_cred.oa_length = ret; + } else { + request->rm_call.cb_cred.oa_flavor = AUTH_NULL; + request->rm_call.cb_cred.oa_base = NULL; + request->rm_call.cb_cred.oa_length = 0; + } request->rm_call.cb_verf.oa_flavor = AUTH_NONE; request->rm_call.cb_verf.oa_base = NULL; request->rm_call.cb_verf.oa_length = 0; @@ -1197,7 +1267,7 @@ out: struct iobuf * rpc_clnt_record_build_record (struct rpc_clnt *clnt, int prognum, int progver, - int procnum, size_t payload, uint64_t xid, + int procnum, size_t hdrsize, uint64_t xid, struct auth_glusterfs_parms_v2 *au, struct iovec *recbuf) { @@ -1207,16 +1277,33 @@ rpc_clnt_record_build_record (struct rpc_clnt *clnt, int prognum, int progver, struct iovec recordhdr = {0, }; size_t pagesize = 0; int ret = -1; + size_t xdr_size = 0; char auth_data[GF_MAX_AUTH_BYTES] = {0, }; if ((!clnt) || (!recbuf) || (!au)) { goto out; } + /* Fill the rpc structure and XDR it into the buffer got above. */ + if (clnt->auth_null) + ret = rpc_clnt_fill_request (prognum, progver, procnum, + xid, NULL, &request, NULL); + else + ret = rpc_clnt_fill_request (prognum, progver, procnum, + xid, au, &request, auth_data); + + if (ret == -1) { + gf_log (clnt->conn.name, GF_LOG_WARNING, + "cannot build a rpc-request xid (%"PRIu64")", xid); + goto out; + } + + xdr_size = xdr_sizeof ((xdrproc_t)xdr_callmsg, &request); + /* First, try to get a pointer into the buffer which the RPC * layer can use. */ - request_iob = iobuf_get (clnt->ctx->iobuf_pool); + request_iob = iobuf_get2 (clnt->ctx->iobuf_pool, (xdr_size + hdrsize)); if (!request_iob) { goto out; } @@ -1225,20 +1312,11 @@ rpc_clnt_record_build_record (struct rpc_clnt *clnt, int prognum, int progver, record = iobuf_ptr (request_iob); /* Now we have it. */ - /* Fill the rpc structure and XDR it into the buffer got above. */ - ret = rpc_clnt_fill_request (prognum, progver, procnum, payload, xid, - au, &request, auth_data); - if (ret == -1) { - gf_log (clnt->conn.trans->name, GF_LOG_WARNING, - "cannot build a rpc-request xid (%"PRIu64")", xid); - goto out; - } - recordhdr = rpc_clnt_record_build_header (record, pagesize, &request, - payload); + hdrsize); if (!recordhdr.iov_base) { - gf_log (clnt->conn.trans->name, GF_LOG_ERROR, + gf_log (clnt->conn.name, GF_LOG_ERROR, "Failed to build record header"); iobuf_unref (request_iob); request_iob = NULL; @@ -1256,12 +1334,14 @@ out: struct iobuf * rpc_clnt_record (struct rpc_clnt *clnt, call_frame_t *call_frame, - rpc_clnt_prog_t *prog,int procnum, size_t payload_len, + rpc_clnt_prog_t *prog, int procnum, size_t hdrlen, struct iovec *rpchdr, uint64_t callid) { struct auth_glusterfs_parms_v2 au = {0, }; struct iobuf *request_iob = NULL; char owner[4] = {0,}; + int max_groups = 0; + int max_lkowner_len = 0; if (!prog || !rpchdr || !call_frame) { goto out; @@ -1288,20 +1368,38 @@ rpc_clnt_record (struct rpc_clnt *clnt, call_frame_t *call_frame, au.lk_owner.lk_owner_len = 4; } - gf_log (clnt->conn.trans->name, GF_LOG_TRACE, "Auth Info: pid: %u, uid: %d" + /* The number of groups and the size of lk_owner depend on oneother. + * We can truncate the groups, but should not touch the lk_owner. */ + max_groups = GF_AUTH_GLUSTERFS_MAX_GROUPS (au.lk_owner.lk_owner_len); + if (au.groups.groups_len > max_groups) { + GF_LOG_OCCASIONALLY (gf_auth_max_groups_log, clnt->conn.name, + GF_LOG_WARNING, "truncating grouplist " + "from %d to %d", au.groups.groups_len, + max_groups); + + au.groups.groups_len = max_groups; + } + + max_lkowner_len = GF_AUTH_GLUSTERFS_MAX_LKOWNER (au.groups.groups_len); + if (au.lk_owner.lk_owner_len > max_lkowner_len) { + gf_log (clnt->conn.name, GF_LOG_ERROR, "lkowner field is too " + "big (%d), it does not fit in the rpc-header", + au.lk_owner.lk_owner_len); + errno = E2BIG; + goto out; + } + + gf_log (clnt->conn.name, GF_LOG_TRACE, "Auth Info: pid: %u, uid: %d" ", gid: %d, owner: %s", au.pid, au.uid, au.gid, lkowner_utoa (&call_frame->root->lk_owner)); - /* Assuming the client program would like to speak to the same version of - * program on server. - */ request_iob = rpc_clnt_record_build_record (clnt, prog->prognum, prog->progver, - procnum, payload_len, + procnum, hdrlen, callid, &au, rpchdr); if (!request_iob) { - gf_log (clnt->conn.trans->name, GF_LOG_WARNING, + gf_log (clnt->conn.name, GF_LOG_WARNING, "cannot build rpc-record"); goto out; } @@ -1312,7 +1410,7 @@ out: int rpcclnt_cbk_program_register (struct rpc_clnt *clnt, - rpcclnt_cb_program_t *program) + rpcclnt_cb_program_t *program, void *mydata) { int ret = -1; char already_registered = 0; @@ -1337,7 +1435,7 @@ rpcclnt_cbk_program_register (struct rpc_clnt *clnt, pthread_mutex_unlock (&clnt->lock); if (already_registered) { - gf_log_callingfn (clnt->conn.trans->name, GF_LOG_DEBUG, + gf_log_callingfn (clnt->conn.name, GF_LOG_DEBUG, "already registered"); ret = 0; goto out; @@ -1352,6 +1450,8 @@ rpcclnt_cbk_program_register (struct rpc_clnt *clnt, memcpy (tmp, program, sizeof (*tmp)); INIT_LIST_HEAD (&tmp->program); + tmp->mydata = mydata; + pthread_mutex_lock (&clnt->lock); { list_add_tail (&tmp->program, &clnt->programs); @@ -1359,14 +1459,14 @@ rpcclnt_cbk_program_register (struct rpc_clnt *clnt, pthread_mutex_unlock (&clnt->lock); ret = 0; - gf_log (clnt->conn.trans->name, GF_LOG_DEBUG, + gf_log (clnt->conn.name, GF_LOG_DEBUG, "New program registered: %s, Num: %d, Ver: %d", program->progname, program->prognum, program->progver); out: if (ret == -1) { - gf_log (clnt->conn.trans->name, GF_LOG_ERROR, + gf_log (clnt->conn.name, GF_LOG_ERROR, "Program registration failed:" " %s, Num: %d, Ver: %d", program->progname, program->prognum, program->progver); @@ -1399,6 +1499,8 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, goto out; } + conn = &rpc->conn; + rpcreq = mem_get (rpc->reqpool); if (rpcreq == NULL) { goto out; @@ -1418,8 +1520,6 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, callid = rpc_clnt_new_callid (rpc); - conn = &rpc->conn; - rpcreq->prog = prog; rpcreq->procnum = procnum; rpcreq->conn = conn; @@ -1432,16 +1532,11 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, proglen += iov_length (proghdr, proghdrcount); } - if (progpayload) { - proglen += iov_length (progpayload, - progpayloadcount); - } - request_iob = rpc_clnt_record (rpc, frame, prog, procnum, proglen, &rpchdr, callid); if (!request_iob) { - gf_log (conn->trans->name, GF_LOG_WARNING, + gf_log (conn->name, GF_LOG_WARNING, "cannot build rpc-record"); goto out; } @@ -1468,33 +1563,27 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, if (conn->connected == 0) { ret = rpc_transport_connect (conn->trans, conn->config.remote_port); - /* Below code makes sure the (re-)configured port lasts - for just one successful connect attempt */ - if (!ret) - conn->config.remote_port = 0; } - ret = rpc_transport_submit_request (rpc->conn.trans, - &req); + ret = rpc_transport_submit_request (conn->trans, &req); if (ret == -1) { - gf_log (conn->trans->name, GF_LOG_WARNING, + gf_log (conn->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); + rpcreq->procnum, conn->name); } if ((ret >= 0) && frame) { - gettimeofday (&conn->last_sent, NULL); /* Save the frame in queue */ __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); + rpcreq->procnum, conn->name); } } pthread_mutex_unlock (&conn->lock); @@ -1506,7 +1595,9 @@ rpc_clnt_submit (struct rpc_clnt *rpc, rpc_clnt_prog_t *prog, ret = 0; out: - iobuf_unref (request_iob); + if (request_iob) { + iobuf_unref (request_iob); + } if (new_iobref && iobref) { iobref_unref (iobref); @@ -1538,19 +1629,24 @@ rpc_clnt_ref (struct rpc_clnt *rpc) static void -rpc_clnt_destroy (struct rpc_clnt *rpc) +rpc_clnt_trigger_destroy (struct rpc_clnt *rpc) { + rpc_clnt_connection_t *conn = NULL; + if (!rpc) return; - if (rpc->conn.trans) { - rpc->conn.trans->mydata = NULL; - rpc_transport_unref (rpc->conn.trans); - //rpc_transport_destroy (rpc->conn.trans); - } + conn = &rpc->conn; + rpc_clnt_disable (rpc); + rpc_transport_unref (conn->trans); +} + +static void +rpc_clnt_destroy (struct rpc_clnt *rpc) +{ + if (!rpc) + return; - rpc_clnt_connection_cleanup (&rpc->conn); - rpc_clnt_reconnect_cleanup (&rpc->conn); saved_frames_destroy (rpc->conn.saved_frames); pthread_mutex_destroy (&rpc->lock); pthread_mutex_destroy (&rpc->conn.lock); @@ -1577,19 +1673,89 @@ 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) +{ + rpc_clnt_connection_t *conn = NULL; + rpc_transport_t *trans = NULL; + + if (!rpc) { + goto out; + } + + conn = &rpc->conn; + + pthread_mutex_lock (&conn->lock); + { + rpc->disabled = 1; + + if (conn->timer) { + gf_timer_call_cancel (rpc->ctx, conn->timer); + conn->timer = NULL; + } + + if (conn->reconnect) { + gf_timer_call_cancel (rpc->ctx, conn->reconnect); + conn->reconnect = NULL; + } + conn->connected = 0; + + if (conn->ping_timer) { + gf_timer_call_cancel (rpc->ctx, conn->ping_timer); + conn->ping_timer = NULL; + conn->ping_started = 0; + } + trans = conn->trans; + conn->trans = NULL; + + } + pthread_mutex_unlock (&conn->lock); + + if (trans) { + rpc_transport_disconnect (trans); + } + +out: + return; +} + + void rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config) { if (config->rpc_timeout) { if (config->rpc_timeout != rpc->conn.config.rpc_timeout) - gf_log (rpc->conn.trans->name, GF_LOG_INFO, + gf_log (rpc->conn.name, GF_LOG_INFO, "changing timeout to %d (from %d)", config->rpc_timeout, rpc->conn.config.rpc_timeout); @@ -1598,7 +1764,7 @@ rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config) if (config->remote_port) { if (config->remote_port != rpc->conn.config.remote_port) - gf_log (rpc->conn.trans->name, GF_LOG_INFO, + gf_log (rpc->conn.name, GF_LOG_INFO, "changing port to %d (from %d)", config->remote_port, rpc->conn.config.remote_port); @@ -1610,13 +1776,13 @@ rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config) if (rpc->conn.config.remote_host) { if (strcmp (rpc->conn.config.remote_host, config->remote_host)) - gf_log (rpc->conn.trans->name, GF_LOG_INFO, - "changing port to %s (from %s)", + gf_log (rpc->conn.name, GF_LOG_INFO, + "changing hostname to %s (from %s)", config->remote_host, rpc->conn.config.remote_host); - FREE (rpc->conn.config.remote_host); + GF_FREE (rpc->conn.config.remote_host); } else { - gf_log (rpc->conn.trans->name, GF_LOG_INFO, + gf_log (rpc->conn.name, GF_LOG_INFO, "setting hostname to %s", config->remote_host); } @@ -1624,54 +1790,3 @@ 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) -{ - 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; - - *options = dict; -out: - if (ret) { - if (fpath) - 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 7034da705..bd4820605 100644 --- a/rpc/rpc-lib/src/rpc-clnt.h +++ b/rpc/rpc-lib/src/rpc-clnt.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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_CLNT_H @@ -28,7 +19,8 @@ typedef enum { RPC_CLNT_CONNECT, RPC_CLNT_DISCONNECT, - RPC_CLNT_MSG + RPC_CLNT_MSG, + RPC_CLNT_DESTROY } rpc_clnt_event_t; @@ -87,7 +79,7 @@ typedef struct rpc_clnt_program { int numproc; } rpc_clnt_prog_t; -typedef int (*rpcclnt_cb_fn) (void *data); +typedef int (*rpcclnt_cb_fn) (struct rpc_clnt *rpc, void *mydata, void *data); /* The descriptor for each procedure/actor that runs * over the RPC service. @@ -115,6 +107,9 @@ typedef struct rpcclnt_cb_program { /* list member to link to list of registered services with rpc_clnt */ struct list_head program; + + /* Needed for passing back in cb_actor */ + void *mydata; } rpcclnt_cb_program_t; @@ -149,6 +144,7 @@ struct rpc_clnt_connection { struct timeval last_sent; struct timeval last_received; int32_t ping_started; + char *name; }; typedef struct rpc_clnt_connection rpc_clnt_connection_t; @@ -169,7 +165,7 @@ struct rpc_req { void *conn_private; }; -struct rpc_clnt { +typedef struct rpc_clnt { pthread_mutex_t lock; rpc_clnt_notify_t notifyfn; rpc_clnt_connection_t conn; @@ -186,11 +182,13 @@ struct rpc_clnt { glusterfs_ctx_t *ctx; int refcount; -}; + int auth_null; + char disabled; +} rpc_clnt_t; struct rpc_clnt *rpc_clnt_new (dict_t *options, glusterfs_ctx_t *ctx, - char *name); + char *name, uint32_t reqpool_size); int rpc_clnt_start (struct rpc_clnt *rpc); @@ -226,10 +224,11 @@ rpc_clnt_ref (struct rpc_clnt *rpc); struct rpc_clnt * rpc_clnt_unref (struct rpc_clnt *rpc); +int rpc_clnt_connection_cleanup (rpc_clnt_connection_t *conn); + void rpc_clnt_set_connected (rpc_clnt_connection_t *conn); void rpc_clnt_unset_connected (rpc_clnt_connection_t *conn); - void rpc_clnt_reconnect (void *trans_ptr); void rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config); @@ -238,8 +237,12 @@ void rpc_clnt_reconfig (struct rpc_clnt *rpc, struct rpc_clnt_config *config); * procedure handlers. */ int rpcclnt_cbk_program_register (struct rpc_clnt *svc, - rpcclnt_cb_program_t *program); + rpcclnt_cb_program_t *program, void *mydata); + +void +rpc_clnt_disable (struct rpc_clnt *rpc); + +char +rpc_clnt_is_disabled (struct rpc_clnt *rpc); -int -rpc_clnt_transport_unix_options_build (dict_t **options, char *filepath); #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..bc37324d3 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.c @@ -0,0 +1,898 @@ +/* + 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; + drc_cached_op_t *req = NULL; + drc_cached_op_t *reply = NULL; + + GF_ASSERT (item); + GF_ASSERT (rb_node_data); + GF_ASSERT (param); + + req = (drc_cached_op_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->progversion == 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; + drc_cached_op_t new = { + .xid = req->xid, + .prognum = req->prognum, + .progversion = req->progver, + .procnum = req->procnum, + }; + + 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, &new); + + 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; + + 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; + + GF_ASSERT (svc); + GF_ASSERT (options); + + /* 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; + } + + gf_log (GF_RPCSVC, GF_LOG_INFO, "DRC is turned %s", (ret?"ON":"OFF")); + + /*DRC off, nothing to do */ + if (ret == _gf_false) + return (0); + + drc = GF_CALLOC (1, sizeof (rpcsvc_drc_globals_t), + gf_common_mt_drc_globals_t); + if (!drc) + return (-1); + + LOCK_INIT (&drc->lock); + svc->drc = drc; + + LOCK (&drc->lock); + + /* 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; + 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_deinit (rpcsvc_t *svc) +{ + rpcsvc_drc_globals_t *drc = NULL; + + if (!svc) + return (-1); + + drc = svc->drc; + if (!drc) + 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); +} + +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; + + /* Input sanitization */ + if ((!svc) || (!options)) + return (-1); + + /* If DRC was not enabled before, Let rpcsvc_drc_init() to + * take care of DRC initialization part. + */ + drc = svc->drc; + if (!drc) { + return rpcsvc_drc_init(svc, options); + } + + /* DRC was already enabled before. Going to be reconfigured. Check + * if reconfigured options contain "nfs.drc" and "nfs.drc-size". + * + * NB: If DRC is "OFF", "drc-size" has no role to play. + * So, "drc-size" gets evaluated IFF DRC is "ON". + * + * If DRC is reconfigured, + * case 1: DRC is "ON" + * sub-case 1: drc-size remains same + * ACTION: Nothing to do. + * sub-case 2: drc-size just changed + * ACTION: rpcsvc_drc_deinit() followed by + * rpcsvc_drc_init(). + * + * case 2: DRC is "OFF" + * ACTION: rpcsvc_drc_deinit() + */ + ret = dict_get_str_boolean (options, "nfs.drc", _gf_true); + if (ret < 0) { + enable_drc = _gf_true; + } else { + enable_drc = ret; + } + + /* case 1: DRC is "ON"*/ + if (enable_drc) { + /* Fetch drc-size if reconfigured */ + if (dict_get_uint32 (options, "nfs.drc-size", &drc_size)) + drc_size = DRC_DEFAULT_CACHE_SIZE; + + /* case 1: sub-case 1*/ + if (drc->global_cache_size == drc_size) + return (0); + + /* case 1: sub-case 2*/ + (void) rpcsvc_drc_deinit (svc); + return rpcsvc_drc_init (svc, options); + } + + /* case 2: DRC is "OFF" */ + gf_log (GF_RPCSVC, GF_LOG_INFO, "DRC is manually turned OFF"); + return rpcsvc_drc_deinit (svc); +} diff --git a/rpc/rpc-lib/src/rpc-drc.h b/rpc/rpc-lib/src/rpc-drc.h new file mode 100644 index 000000000..c42c2a2c2 --- /dev/null +++ b/rpc/rpc-lib/src/rpc-drc.h @@ -0,0 +1,106 @@ +/* + 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; +}; + +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_deinit (rpcsvc_t *svc); + +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 b97ba61bf..c24d41084 100644 --- a/rpc/rpc-lib/src/rpc-transport.c +++ b/rpc/rpc-lib/src/rpc-transport.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. */ #include <dlfcn.h> @@ -78,6 +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) @@ -154,6 +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); @@ -178,7 +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 { @@ -259,13 +264,15 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) handle = dlopen (name, RTLD_NOW|RTLD_GLOBAL); if (handle == NULL) { gf_log ("rpc-transport", GF_LOG_ERROR, "%s", dlerror ()); - gf_log ("rpc-transport", GF_LOG_ERROR, + gf_log ("rpc-transport", GF_LOG_WARNING, "volume '%s': transport-type '%s' is not valid or " "not found on this machine", trans_name, type); goto fail; } + trans->dl_handle = handle; + trans->ops = dlsym (handle, "tops"); if (trans->ops == NULL) { gf_log ("rpc-transport", GF_LOG_ERROR, @@ -273,22 +280,22 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) goto fail; } - trans->init = dlsym (handle, "init"); + *VOID(&(trans->init)) = dlsym (handle, "init"); if (trans->init == NULL) { gf_log ("rpc-transport", GF_LOG_ERROR, "dlsym (gf_rpc_transport_init) on %s", dlerror ()); goto fail; } - trans->fini = dlsym (handle, "fini"); + *VOID(&(trans->fini)) = dlsym (handle, "fini"); if (trans->fini == NULL) { gf_log ("rpc-transport", GF_LOG_ERROR, "dlsym (gf_rpc_transport_fini) on %s", dlerror ()); goto fail; } - trans->reconfigure = dlsym (handle, "reconfigure"); - if (trans->fini == NULL) { + *VOID(&(trans->reconfigure)) = dlsym (handle, "reconfigure"); + if (trans->reconfigure == NULL) { gf_log ("rpc-transport", GF_LOG_DEBUG, "dlsym (gf_rpc_transport_reconfigure) on %s", dlerror()); } @@ -299,14 +306,15 @@ 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"); @@ -317,7 +325,7 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) trans->options = options; pthread_mutex_init (&trans->lock, NULL); - trans->xl = THIS; + trans->xl = this; ret = trans->init (trans); if (ret != 0) { @@ -326,25 +334,27 @@ rpc_transport_load (glusterfs_ctx_t *ctx, dict_t *options, char *trans_name) goto fail; } - return_trans = trans; + return_trans = trans; - if (name) { - GF_FREE (name); - } + GF_FREE (name); return return_trans; fail: if (trans) { - if (trans->name) { - GF_FREE (trans->name); - } + GF_FREE (trans->name); + + if (trans->dl_handle) + dlclose (trans->dl_handle); GF_FREE (trans); } - if (name) { - GF_FREE (name); + GF_FREE (name); + + if (vol_opt && !list_empty (&vol_opt->list)) { + list_del_init (&vol_opt->list); + GF_FREE (vol_opt); } return NULL; @@ -432,8 +442,10 @@ rpc_transport_destroy (rpc_transport_t *this) pthread_mutex_destroy (&this->lock); - if (this->name) - GF_FREE (this->name); + GF_FREE (this->name); + + if (this->dl_handle) + dlclose (this->dl_handle); GF_FREE (this); fail: @@ -470,7 +482,7 @@ rpc_transport_unref (rpc_transport_t *this) pthread_mutex_lock (&this->lock); { - refcount = --this->refcount; + refcount = --this->refcount; } pthread_mutex_unlock (&this->lock); @@ -478,7 +490,9 @@ rpc_transport_unref (rpc_transport_t *this) if (this->mydata) this->notify (this, this->mydata, RPC_TRANSPORT_CLEANUP, NULL); - rpc_transport_destroy (this); + this->mydata = NULL; + this->notify = NULL; + rpc_transport_destroy (this); } ret = 0; @@ -520,6 +534,8 @@ out: return ret; } + + //give negative values to skip setting that value //this function asserts if both the values are negative. //why call it if you dont set it. @@ -546,6 +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) { @@ -580,7 +653,7 @@ rpc_transport_inet_options_build (dict_t **options, const char *hostname, "failed to set remote-port with %d", port); goto out; } - ret = dict_set_str (dict, "transport.address-family", "inet/inet6"); + ret = dict_set_str (dict, "transport.address-family", "inet"); if (ret) { gf_log (THIS->name, GF_LOG_WARNING, "failed to set addr-family with inet"); @@ -597,8 +670,7 @@ rpc_transport_inet_options_build (dict_t **options, const char *hostname, *options = dict; out: if (ret) { - if (host) - GF_FREE (host); + GF_FREE (host); if (dict) dict_unref (dict); } diff --git a/rpc/rpc-lib/src/rpc-transport.h b/rpc/rpc-lib/src/rpc-transport.h index f838d4baf..2db9072ae 100644 --- a/rpc/rpc-lib/src/rpc-transport.h +++ b/rpc/rpc-lib/src/rpc-transport.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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_TRANSPORT_H__ @@ -80,6 +71,11 @@ struct peer_info { struct sockaddr_storage sockaddr; socklen_t sockaddr_len; char identifier[UNIX_PATH_MAX]; + // OP-VERSION of clients + uint32_t max_op_version; + uint32_t min_op_version; + //Volume mounted by client + char volname[1024]; }; typedef struct peer_info peer_info_t; @@ -190,16 +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); @@ -214,6 +213,7 @@ struct rpc_transport { struct list_head list; int bind_insecure; + void *dl_handle; /* handle of dlopen() */ }; struct rpc_transport_ops { @@ -237,6 +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); }; @@ -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 930eabdda..384e4a75d 100644 --- a/rpc/rpc-lib/src/rpcsvc-auth.c +++ b/rpc/rpc-lib/src/rpcsvc-auth.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. */ #include "rpcsvc.h" @@ -187,6 +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; @@ -213,6 +227,41 @@ rpcsvc_set_allow_insecure (rpcsvc_t *svc, dict_t *options) } int +rpcsvc_set_root_squash (rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + uid_t anonuid = -1; + gid_t anongid = -1; + + GF_ASSERT (svc); + GF_ASSERT (options); + + ret = dict_get_str_boolean (options, "root-squash", 0); + if (ret != -1) + svc->root_squash = ret; + else + svc->root_squash = _gf_false; + + ret = dict_get_uint32 (options, "anonuid", &anonuid); + if (!ret) + svc->anonuid = anonuid; + else + svc->anonuid = RPC_NOBODY_UID; + + ret = dict_get_uint32 (options, "anongid", &anongid); + if (!ret) + svc->anongid = anongid; + else + svc->anongid = RPC_NOBODY_GID; + + if (svc->root_squash) + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "root squashing enabled " + "(uid=%d, gid=%d)", svc->anonuid, svc->anongid); + + return 0; +} + +int rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options) { int ret = -1; @@ -221,6 +270,8 @@ rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options) return -1; (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"); @@ -237,6 +288,25 @@ 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) @@ -299,22 +369,36 @@ ret: int -rpcsvc_auth_request_init (rpcsvc_request_t *req) +rpcsvc_auth_request_init (rpcsvc_request_t *req, struct rpc_msg *callmsg) { - int ret = -1; + int32_t ret = 0; rpcsvc_auth_t *auth = NULL; - if (!req) - return -1; + if (!req || !callmsg) { + ret = -1; + goto err; + } + + req->cred.flavour = rpc_call_cred_flavour (callmsg); + req->cred.datalen = rpc_call_cred_len (callmsg); + req->verf.flavour = rpc_call_verf_flavour (callmsg); + req->verf.datalen = rpc_call_verf_len (callmsg); auth = rpcsvc_auth_get_handler (req); - if (!auth) + if (!auth) { + ret = -1; goto err; - ret = 0; + } + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Auth handler: %s", auth->authname); - if (!auth->authops->request_init) - ret = auth->authops->request_init (req, auth->authprivate); + if (auth->authops->request_init) + ret = auth->authops->request_init (req, auth->authprivate); + + /* reset to auxgidlarge during + unsersialize if necessary */ + req->auxgids = req->auxgidsmall; + req->auxgidlarge = NULL; err: return ret; } @@ -354,14 +438,10 @@ 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; @@ -379,59 +459,27 @@ 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 b03776dee..3c16abeb7 100644 --- a/rpc/rpc-lib/src/rpcsvc-common.h +++ b/rpc/rpc-lib/src/rpcsvc-common.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _RPCSVC_COMMON_H @@ -39,6 +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. */ @@ -59,24 +52,77 @@ 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; + uid_t anonuid; + gid_t anongid; 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 e0fc29184..1c13048f2 100644 --- a/rpc/rpc-lib/src/rpcsvc.c +++ b/rpc/rpc-lib/src/rpcsvc.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 @@ -36,6 +27,9 @@ #include "xdr-common.h" #include "xdr-generic.h" #include "rpc-common-xdr.h" +#include "syncop.h" +#include "rpc-drc.h" +#include "protocol-common.h" #include <errno.h> #include <pthread.h> @@ -49,6 +43,7 @@ #include <stdio.h> #include "xdr-rpcclnt.h" +#include "glusterfs-acl.h" struct rpcsvc_program gluster_dump_prog; @@ -65,6 +60,9 @@ int rpcsvc_notify (rpc_transport_t *trans, void *mydata, rpc_transport_event_t event, void *data, ...); +static int +match_subnet_v4 (const char *addrtok, const char *ipaddr); + rpcsvc_notify_wrapper_t * rpcsvc_notify_wrapper_alloc (void) { @@ -135,6 +133,72 @@ rpcsvc_get_program_vector_sizer (rpcsvc_t *svc, uint32_t prognum, return NULL; } +gf_boolean_t +rpcsvc_can_outstanding_req_be_ignored (rpcsvc_request_t *req) +{ + /* + * If outstanding_rpc_limit is reached because of blocked locks and + * throttling is attempted then no unlock requests will be received. So + * the outstanding request count will never change i.e. it will always + * be equal to the limit. This also leads to ping timer expiry on + * client. + */ + + /* + * This is a hack and a necessity until grantedlock == fop completion. + * Ideally if we get a blocking lock request which cannot be granted + * right now, we should unwind the fop saying “request registered, will + * notify you when granted”, which is very hard to implement at the + * moment. Until we bring in such mechanism, we will need to live with + * not rate-limiting INODELK/ENTRYLK/LK fops + */ + + if ((req->prognum == GLUSTER_FOP_PROGRAM) && + (req->progver == GLUSTER_FOP_VERSION)) { + if ((req->procnum == GFS3_OP_INODELK) || + (req->procnum == GFS3_OP_FINODELK) || + (req->procnum == GFS3_OP_ENTRYLK) || + (req->procnum == GFS3_OP_FENTRYLK) || + (req->procnum == GFS3_OP_LK)) + return _gf_true; + } + return _gf_false; +} + +int +rpcsvc_request_outstanding (rpcsvc_request_t *req, int delta) +{ + int ret = 0; + int old_count = 0; + int new_count = 0; + int limit = 0; + + if (rpcsvc_can_outstanding_req_be_ignored (req)) + return 0; + + pthread_mutex_lock (&req->trans->lock); + { + limit = req->svc->outstanding_rpc_limit; + if (!limit) + goto unlock; + + old_count = req->trans->outstanding_rpc_count; + req->trans->outstanding_rpc_count += delta; + new_count = req->trans->outstanding_rpc_count; + + if (old_count <= limit && new_count > limit) + ret = rpc_transport_throttle (req->trans, _gf_true); + + if (old_count > limit && new_count <= limit) + ret = rpc_transport_throttle (req->trans, _gf_false); + } +unlock: + pthread_mutex_unlock (&req->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. @@ -170,7 +234,11 @@ rpcsvc_program_actor (rpcsvc_request_t *req) if (!found) { if (err != PROG_MISMATCH) { - gf_log (GF_RPCSVC, GF_LOG_WARNING, + /* log in DEBUG when nfs clients try to see if + * ACL requests are accepted by nfs server + */ + gf_log (GF_RPCSVC, (req->prognum == ACL_PROGRAM) ? + GF_LOG_DEBUG : GF_LOG_WARNING, "RPC program not available (req %u %u)", req->prognum, req->progver); err = PROG_UNAVAIL; @@ -209,6 +277,8 @@ rpcsvc_program_actor (rpcsvc_request_t *req) goto err; } + req->synctask = program->synctask; + err = SUCCESS; gf_log (GF_RPCSVC, GF_LOG_TRACE, "Actor found: %s - %s", program->progname, actor->procname); @@ -223,7 +293,7 @@ err: /* this procedure can only pass 4 arguments to registered notifyfn. To send more * arguments call wrapper->notify directly. */ -inline void +static inline void rpcsvc_program_notify (rpcsvc_listener_t *listener, rpcsvc_event_t event, void *data) { @@ -246,7 +316,7 @@ out: } -inline int +static inline int rpcsvc_accept (rpcsvc_t *svc, rpc_transport_t *listen_trans, rpc_transport_t *new_trans) { @@ -276,8 +346,21 @@ rpcsvc_request_destroy (rpcsvc_request_t *req) iobref_unref (req->iobref); } + 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. + */ + if (req->prognum) //Only for initialized requests + rpcsvc_request_outstanding (req, -1); + rpc_transport_unref (req->trans); + GF_FREE (req->auxgidlarge); + mem_put (req); out: @@ -307,7 +390,9 @@ rpcsvc_request_init (rpcsvc_t *svc, rpc_transport_t *trans, req->msg[0] = progmsg; req->iobref = iobref_ref (msg->iobref); if (msg->vectored) { - for (i = 1; i < msg->count; i++) { + /* msg->vector[2] is defined in structure. prevent a + out of bound access */ + for (i = 1; i < min (msg->count, 2); i++) { req->msg[i] = msg->vector[i]; } } @@ -322,13 +407,7 @@ rpcsvc_request_init (rpcsvc_t *svc, rpc_transport_t *trans, * been copied into the required sections of the req structure, * we just need to fill in the meta-data about it now. */ - req->cred.flavour = rpc_call_cred_flavour (callmsg); - req->cred.datalen = rpc_call_cred_len (callmsg); - req->verf.flavour = rpc_call_verf_flavour (callmsg); - req->verf.datalen = rpc_call_verf_len (callmsg); - - /* AUTH */ - rpcsvc_auth_request_init (req); + rpcsvc_auth_request_init (req, callmsg); return req; } @@ -375,19 +454,28 @@ rpcsvc_request_create (rpcsvc_t *svc, rpc_transport_t *trans, ret = -1; rpcsvc_request_init (svc, trans, &rpcmsg, progmsg, msg, req); - gf_log (GF_RPCSVC, GF_LOG_TRACE, "received rpc-message (XID: 0x%lx, " - "Ver: %ld, Program: %ld, ProgVers: %ld, Proc: %ld) from" - " rpc-transport (%s)", rpc_call_xid (&rpcmsg), + gf_log (GF_RPCSVC, GF_LOG_TRACE, "received rpc-message " + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION ", Program: %" GF_PRI_RPC_PROG_ID ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC ") " + "from rpc-transport (%s)", rpc_call_xid (&rpcmsg), rpc_call_rpcvers (&rpcmsg), rpc_call_program (&rpcmsg), rpc_call_progver (&rpcmsg), rpc_call_progproc (&rpcmsg), trans->name); + /* 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. + */ + if (req->prognum) //Only for initialized requests + ret = rpcsvc_request_outstanding (req, +1); + if (rpc_call_rpcvers (&rpcmsg) != 2) { /* LOG- TODO: print rpc version, also print the peerinfo from transport */ gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC version not supported " - "(XID: 0x%lx, Ver: %ld, Prog: %ld, ProgVers: %ld, " - "Proc: %ld) from trans (%s)", rpc_call_xid (&rpcmsg), + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION ", Program: %" GF_PRI_RPC_PROG_ID ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC ") " + "from trans (%s)", rpc_call_xid (&rpcmsg), rpc_call_rpcvers (&rpcmsg), rpc_call_program (&rpcmsg), rpc_call_progver (&rpcmsg), rpc_call_progproc (&rpcmsg), trans->name); @@ -403,8 +491,9 @@ rpcsvc_request_create (rpcsvc_t *svc, rpc_transport_t *trans, */ rpcsvc_request_seterr (req, AUTH_ERROR); gf_log (GF_RPCSVC, GF_LOG_ERROR, "auth failed on request. " - "(XID: 0x%lx, Ver: %ld, Prog: %ld, ProgVers: %ld, " - "Proc: %ld) from trans (%s)", rpc_call_xid (&rpcmsg), + "(XID: 0x%" GF_PRI_RPC_XID ", Ver: %" GF_PRI_RPC_VERSION ", Program: %" GF_PRI_RPC_PROG_ID ", " + "ProgVers: %" GF_PRI_RPC_PROG_VERS ", Proc: %" GF_PRI_RPC_PROC ") " + "from trans (%s)", rpc_call_xid (&rpcmsg), rpc_call_rpcvers (&rpcmsg), rpc_call_program (&rpcmsg), rpc_call_progver (&rpcmsg), rpc_call_progproc (&rpcmsg), trans->name); @@ -417,6 +506,7 @@ 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) { @@ -432,15 +522,39 @@ err: int +rpcsvc_check_and_reply_error (int ret, call_frame_t *frame, void *opaque) +{ + rpcsvc_request_t *req = NULL; + + req = opaque; + + if (ret) + gf_log ("rpcsvc", GF_LOG_ERROR, + "rpc actor failed to complete successfully"); + + if (ret == RPCSVC_ACTOR_ERROR) { + ret = rpcsvc_error_reply (req); + if (ret) + gf_log ("rpcsvc", GF_LOG_WARNING, + "failed to queue error reply"); + } + + 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_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; @@ -476,7 +590,7 @@ 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; @@ -494,40 +608,76 @@ 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; - if (req->count == 2) { - if (actor->vector_actor) { - ret = actor->vector_actor (req, &req->msg[1], 1, - req->iobref); - } else { - 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; - } - } else if (actor->actor) { - ret = actor->actor (req); + 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; } - } -err_reply: - if (ret == RPCSVC_ACTOR_ERROR) { - ret = rpcsvc_error_reply (req); + 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, + req); + } else { + ret = actor_fn (req); + } } - if (ret) - gf_log ("rpcsvc", GF_LOG_WARNING, "failed to queue error reply"); +err_reply: + ret = rpcsvc_check_and_reply_error (ret, NULL, req); /* No need to propagate error beyond this function since the reply * has now been queued. */ ret = 0; -err: +out: return ret; } @@ -545,6 +695,9 @@ rpcsvc_handle_disconnect (rpcsvc_t *svc, rpc_transport_t *trans) pthread_mutex_lock (&svc->rpclock); { + if (!svc->notify_count) + goto unlock; + wrappers = GF_CALLOC (svc->notify_count, sizeof (*wrapper), gf_common_mt_rpcsvc_wrapper_t); if (!wrappers) { @@ -685,15 +838,9 @@ err: return txrecord; } -inline int -rpcsvc_get_callid (rpcsvc_t *rpc) -{ - return GF_UNIVERSAL_ANSWER; -} - int rpcsvc_fill_callback (int prognum, int progver, int procnum, int payload, - uint64_t xid, struct rpc_msg *request) + uint32_t xid, struct rpc_msg *request) { int ret = -1; @@ -758,9 +905,9 @@ out: return txrecord; } -struct iobuf * +static struct iobuf * rpcsvc_callback_build_record (rpcsvc_t *rpc, int prognum, int progver, - int procnum, size_t payload, uint64_t xid, + int procnum, size_t payload, u_long xid, struct iovec *recbuf) { struct rpc_msg request = {0, }; @@ -768,16 +915,28 @@ rpcsvc_callback_build_record (rpcsvc_t *rpc, int prognum, int progver, char *record = NULL; struct iovec recordhdr = {0, }; size_t pagesize = 0; + size_t xdr_size = 0; int ret = -1; if ((!rpc) || (!recbuf)) { goto out; } + /* Fill the rpc structure and XDR it into the buffer got above. */ + ret = rpcsvc_fill_callback (prognum, progver, procnum, payload, xid, + &request); + if (ret == -1) { + gf_log ("rpcsvc", GF_LOG_WARNING, "cannot build a rpc-request " + "xid (%" GF_PRI_RPC_XID ")", xid); + goto out; + } + /* First, try to get a pointer into the buffer which the RPC * layer can use. */ - request_iob = iobuf_get (rpc->ctx->iobuf_pool); + xdr_size = xdr_sizeof ((xdrproc_t)xdr_callmsg, &request); + + request_iob = iobuf_get2 (rpc->ctx->iobuf_pool, (xdr_size + payload)); if (!request_iob) { goto out; } @@ -786,15 +945,6 @@ rpcsvc_callback_build_record (rpcsvc_t *rpc, int prognum, int progver, record = iobuf_ptr (request_iob); /* Now we have it. */ - /* Fill the rpc structure and XDR it into the buffer got above. */ - ret = rpcsvc_fill_callback (prognum, progver, procnum, payload, xid, - &request); - if (ret == -1) { - gf_log ("rpcsvc", GF_LOG_WARNING, "cannot build a rpc-request " - "xid (%"PRIu64")", xid); - goto out; - } - recordhdr = rpcsvc_callback_build_header (record, pagesize, &request, payload); @@ -824,7 +974,6 @@ rpcsvc_callback_submit (rpcsvc_t *rpc, rpc_transport_t *trans, rpc_transport_req_t req; int ret = -1; int proglen = 0; - uint64_t callid = 0; if (!rpc) { goto out; @@ -832,15 +981,14 @@ rpcsvc_callback_submit (rpcsvc_t *rpc, rpc_transport_t *trans, memset (&req, 0, sizeof (req)); - callid = rpcsvc_get_callid (rpc); - if (proghdr) { proglen += iov_length (proghdr, proghdrcount); } request_iob = rpcsvc_callback_build_record (rpc, prog->prognum, prog->progver, procnum, - proglen, callid, + proglen, + GF_UNIVERSAL_ANSWER, &rpchdr); if (!request_iob) { gf_log ("rpcsvc", GF_LOG_WARNING, @@ -868,21 +1016,22 @@ out: return ret; } -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; @@ -938,13 +1087,14 @@ out: */ struct iobuf * rpcsvc_record_build_record (rpcsvc_request_t *req, size_t payload, - struct iovec *recbuf) + size_t hdrlen, struct iovec *recbuf) { struct rpc_msg reply; struct iobuf *replyiob = NULL; char *record = NULL; struct iovec recordhdr = {0, }; size_t pagesize = 0; + size_t xdr_size = 0; rpcsvc_t *svc = NULL; int ret = -1; @@ -952,19 +1102,25 @@ rpcsvc_record_build_record (rpcsvc_request_t *req, size_t payload, return NULL; svc = req->svc; - replyiob = iobuf_get (svc->ctx->iobuf_pool); - pagesize = iobuf_pagesize (replyiob); - if (!replyiob) { - goto err_exit; - } - - record = iobuf_ptr (replyiob); /* Now we have it. */ /* Fill the rpc structure and XDR it into the buffer got above. */ ret = rpcsvc_fill_reply (req, &reply); if (ret) goto err_exit; + xdr_size = xdr_sizeof ((xdrproc_t)xdr_replymsg, &reply); + + /* Payload would include 'readv' size etc too, where as + that comes as another payload iobuf */ + replyiob = iobuf_get2 (svc->ctx->iobuf_pool, (xdr_size + hdrlen)); + if (!replyiob) { + goto err_exit; + } + + pagesize = iobuf_pagesize (replyiob); + + record = iobuf_ptr (replyiob); /* Now we have it. */ + recordhdr = rpcsvc_record_build_header (record, pagesize, reply, payload); if (!recordhdr.iov_base) { @@ -1019,7 +1175,9 @@ rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec *proghdr, struct iovec recordhdr = {0, }; rpc_transport_t *trans = NULL; size_t msglen = 0; + size_t hdrlen = 0; char new_iobref = 0; + rpcsvc_drc_globals_t *drc = NULL; if ((!req) || (!req->trans)) return -1; @@ -1037,7 +1195,7 @@ rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec *proghdr, gf_log (GF_RPCSVC, GF_LOG_TRACE, "Tx message: %zu", msglen); /* Build the buffer containing the encoded RPC reply. */ - replyiob = rpcsvc_record_build_record (req, msglen, &recordhdr); + replyiob = rpcsvc_record_build_record (req, msglen, hdrlen, &recordhdr); if (!replyiob) { gf_log (GF_RPCSVC, GF_LOG_ERROR,"Reply record creation failed"); goto disconnect_exit; @@ -1054,24 +1212,35 @@ 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) && (rpcsvc_need_drc (req))) { + 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); + req->procnum, trans ? 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, - req->procnum, trans->name); + req->procnum, trans ? trans->name : ""); } disconnect_exit: @@ -1097,7 +1266,7 @@ rpcsvc_error_reply (rpcsvc_request_t *req) if (!req) return -1; - gf_log_callingfn ("", GF_LOG_WARNING, "sending a RPC error reply"); + gf_log_callingfn ("", GF_LOG_DEBUG, "sending a RPC error reply"); /* At this point the req should already have been filled with the * appropriate RPC error numbers. @@ -1110,20 +1279,21 @@ 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" - " portmap"); + " portmap %d %d %u", newprog->prognum, newprog->progver, port); goto out; } - ret = 0; + ret = 0; /* SUCCESS */ out: return ret; } @@ -1132,7 +1302,7 @@ out: inline int rpcsvc_program_unregister_portmap (rpcsvc_program_t *prog) { - int ret = 0; + int ret = -1; if (!prog) goto out; @@ -1254,28 +1424,44 @@ rpcsvc_submit_message (rpcsvc_request_t *req, struct iovec *proghdr, int -rpcsvc_program_unregister (rpcsvc_t *svc, rpcsvc_program_t *prog) +rpcsvc_program_unregister (rpcsvc_t *svc, rpcsvc_program_t *program) { int ret = -1; - - if (!svc || !prog) { + rpcsvc_program_t *prog = NULL; + if (!svc || !program) { goto out; } - ret = rpcsvc_program_unregister_portmap (prog); + ret = rpcsvc_program_unregister_portmap (program); if (ret == -1) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "portmap unregistration of" " program failed"); goto out; } + pthread_mutex_lock (&svc->rpclock); + { + list_for_each_entry (prog, &svc->programs, program) { + if ((prog->prognum == program->prognum) + && (prog->progver == program->progver)) { + break; + } + } + } + pthread_mutex_unlock (&svc->rpclock); + + if (prog == NULL) { + ret = -1; + goto out; + } + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Program unregistered: %s, Num: %d," " Ver: %d, Port: %d", prog->progname, prog->prognum, prog->progver, prog->progport); pthread_mutex_lock (&svc->rpclock); { - list_del (&prog->program); + list_del_init (&prog->program); } pthread_mutex_unlock (&svc->rpclock); @@ -1283,8 +1469,8 @@ rpcsvc_program_unregister (rpcsvc_t *svc, rpcsvc_program_t *prog) out: if (ret == -1) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "Program unregistration failed" - ": %s, Num: %d, Ver: %d, Port: %d", prog->progname, - prog->prognum, prog->progver, prog->progport); + ": %s, Num: %d, Ver: %d, Port: %d", program->progname, + program->prognum, program->progver, program->progport); } return ret; @@ -1473,6 +1659,7 @@ rpcsvc_create_listeners (rpcsvc_t *svc, dict_t *options, char *name) } GF_FREE (transport_name); + transport_name = NULL; count++; } @@ -1484,17 +1671,13 @@ rpcsvc_create_listeners (rpcsvc_t *svc, dict_t *options, char *name) transport_type = NULL; out: - if (str != NULL) { - GF_FREE (str); - } + GF_FREE (str); - if (transport_type != NULL) { - GF_FREE (transport_type); - } + GF_FREE (transport_type); - if (tmp != NULL) { - GF_FREE (tmp); - } + GF_FREE (tmp); + + GF_FREE (transport_name); return count; } @@ -1671,15 +1854,17 @@ rpcsvc_dump (rpcsvc_request_t *req) uint32_t dump_rsp_len = 0; if (!req) - goto fail; + goto sendrsp; ret = build_prog_details (req, &rsp); if (ret < 0) { op_errno = -ret; - goto fail; + goto sendrsp; } -fail: + op_errno = 0; + +sendrsp: rsp.op_errno = gf_errno_to_error (op_errno); rsp.op_ret = ret; @@ -1691,15 +1876,12 @@ fail: ret = xdr_serialize_generic (iov, &rsp, (xdrproc_t)xdr_gf_dump_rsp); if (ret < 0) { - if (req) - req->rpc_err = GARBAGE_ARGS; - op_errno = EINVAL; - goto fail; + ret = RPCSVC_ACTOR_ERROR; + } else { + rpcsvc_submit_generic (req, &iov, 1, NULL, 0, NULL); + ret = 0; } - ret = rpcsvc_submit_generic (req, &iov, 1, NULL, 0, - NULL); - free_prog_details (&rsp); return ret; @@ -1737,13 +1919,92 @@ rpcsvc_init_options (rpcsvc_t *svc, dict_t *options) if (!svc->register_portmap) gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Portmap registration " "disabled"); - ret = 0; 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; @@ -1782,23 +2043,70 @@ rpcsvc_transport_unix_options_build (dict_t **options, char *filepath) *options = dict; out: if (ret) { - if (fpath) - GF_FREE (fpath); + GF_FREE (fpath); if (dict) dict_unref (dict); } return ret; } +/* + * Configure() the rpc.outstanding-rpc-limit param. + * If dict_get_int32() for dict-key "rpc.outstanding-rpc-limit" FAILS, + * it would set the value as "defvalue". Otherwise it would fetch the + * value and round up to multiple-of-8. defvalue must be +ve. + * + * NB: defval or set-value "0" is special which means unlimited/65536. + */ +int +rpcsvc_set_outstanding_rpc_limit (rpcsvc_t *svc, dict_t *options, int defvalue) +{ + int ret = -1; /* FAILURE */ + int rpclim = 0; + static char *rpclimkey = "rpc.outstanding-rpc-limit"; + + if ((!svc) || (!options)) + return (-1); + + if ((defvalue < RPCSVC_MIN_OUTSTANDING_RPC_LIMIT) || + (defvalue > RPCSVC_MAX_OUTSTANDING_RPC_LIMIT)) { + return (-1); + } + + /* Fetch the rpc.outstanding-rpc-limit from dict. */ + ret = dict_get_int32 (options, rpclimkey, &rpclim); + if (ret < 0) { + /* Fall back to default for FAILURE */ + rpclim = defvalue; + } + + /* Round up to multiple-of-8. It must not exceed + * RPCSVC_MAX_OUTSTANDING_RPC_LIMIT. + */ + rpclim = ((rpclim + 8 - 1) >> 3) * 8; + 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 * -rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options) +rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options, + uint32_t poolcount) { rpcsvc_t *svc = NULL; - int ret = -1, poolcount = 0; + int ret = -1; - if ((!ctx) || (!options)) + if ((!xl) || (!ctx) || (!options)) return NULL; svc = GF_CALLOC (1, sizeof (*svc), gf_common_mt_rpcsvc_t); @@ -1817,7 +2125,8 @@ rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options) goto free_svc; } - poolcount = RPCSVC_POOLCOUNT_MULT * svc->memfactor; + if (!poolcount) + poolcount = RPCSVC_POOLCOUNT_MULT * svc->memfactor; gf_log (GF_RPCSVC, GF_LOG_TRACE, "rx pool: %d", poolcount); svc->rxpool = mem_pool_new (rpcsvc_request_t, poolcount); @@ -1848,6 +2157,7 @@ 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) { @@ -1860,17 +2170,16 @@ 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 *svptr = NULL; + int ret = -1; + char *addrtok = NULL; + char *addrstr = NULL; + char *dup_addrstr = NULL; + char *svptr = NULL; - if ((!options) || (!clstr)) - return -1; - - if (!dict_get (options, pattern)) + if ((!options) || (!ip)) return -1; ret = dict_get_str (options, pattern, &addrstr); @@ -1884,88 +2193,98 @@ rpcsvc_transport_peer_check_search (dict_t *options, char *pattern, char *clstr) goto err; } - addrtok = strtok_r (addrstr, ",", &svptr); + dup_addrstr = gf_strdup (addrstr); + addrtok = strtok_r (dup_addrstr, ",", &svptr); while (addrtok) { /* 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 (hostname) { +#ifdef FNM_CASEFOLD + ret = fnmatch (addrtok, hostname, FNM_CASEFOLD); +#else + ret = fnmatch (addrtok, hostname, 0); +#endif + if (ret == 0) + goto err; + } + + /* Compare IPv4 subnetwork */ + if (strchr (addrtok, '/')) { + ret = match_subnet_v4 (addrtok, ip); + if (ret == 0) + goto err; + } + addrtok = strtok_r (NULL, ",", &svptr); } ret = -1; err: + GF_FREE (dup_addrstr); return ret; } -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; @@ -1976,313 +2295,144 @@ 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]; - - 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; - } - - 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]; - struct sockaddr_storage sastorage = {0,}; - - 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, - &sastorage, sizeof (sastorage)); + 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; - } - - 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); + ret = dict_get_str (options, srchstr, &allow_str); + GF_FREE (srchstr); + if (ret < 0) + return RPCSVC_AUTH_REJECT; - if (namelookup) - ret = rpcsvc_combine_gen_spec_addr_checks (addrchk, - namechk); - else - ret = addrchk; + 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; + } - return ret; -} + ret = dict_get_str (options, srchstr, &reject_str); + GF_FREE (srchstr); -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; + /* + * If "reject_str" is being set as '*' (anonymous), then NFS-server + * would reject everything. If the "reject_str" is not set and + * "allow_str" is set as '*' (anonymous), then NFS-server would + * accept mount requests from all clients. + */ + if (reject_str != NULL) { + if (!strcmp ("*", reject_str)) + return RPCSVC_AUTH_REJECT; + } else { + if (!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) { - struct sockaddr_storage sastorage = {0,}; - struct sockaddr_in *sa = NULL; + union gf_sock_union sock_union; int ret = RPCSVC_AUTH_REJECT; - socklen_t sasize = sizeof (sa); + 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; + memset (&sock_union, 0, sizeof (sock_union)); + if ((!svc) || (!volname) || (!trans)) return ret; - sa = (struct sockaddr_in*) &sastorage; - ret = rpcsvc_transport_peeraddr (trans, NULL, 0, &sastorage, - sasize); + ret = rpcsvc_transport_peeraddr (trans, NULL, 0, &sock_union.storage, + sinsize); if (ret != 0) { gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get peer addr: %s", gai_strerror (ret)); @@ -2290,7 +2440,7 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, goto err; } - port = ntohs (sa->sin_port); + port = ntohs (sock_union.sin.sin_port); gf_log (GF_RPCSVC, GF_LOG_TRACE, "Client port: %d", (int)port); /* If the port is already a privileged one, dont bother with checking * options. @@ -2301,23 +2451,6 @@ 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"); @@ -2325,25 +2458,22 @@ 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 @@ -2351,6 +2481,9 @@ rpcsvc_transport_privport_check (rpcsvc_t *svc, char *volname, " allowed"); err: + if (srchstr) + GF_FREE (srchstr); + return ret; } @@ -2372,22 +2505,87 @@ rpcsvc_volume_allowed (dict_t *options, char *volname) goto out; } - if (!dict_get (options, srchstr)) { - GF_FREE (srchstr); - srchstr = globalrule; - ret = dict_get_str (options, srchstr, &addrstr); - } else + if (!dict_get (options, srchstr)) + ret = dict_get_str (options, globalrule, &addrstr); + else ret = dict_get_str (options, srchstr, &addrstr); out: + GF_FREE (srchstr); + return addrstr; } +/* + * match_subnet_v4() takes subnetwork address pattern and checks + * if the target IPv4 address has the same network address with + * the help of network mask. + * + * Returns 0 for SUCCESS and -1 otherwise. + * + * NB: Validation of subnetwork address pattern is not required + * as it's already being done at the time of CLI SET. + */ +static int +match_subnet_v4 (const char *addrtok, const char *ipaddr) +{ + char *slash = NULL; + char *netaddr = NULL; + long prefixlen = -1; + int ret = -1; + uint32_t shift = 0; + struct sockaddr_in sin1 = {0, }; + struct sockaddr_in sin2 = {0, }; + struct sockaddr_in mask = {0, }; + + /* Copy the input */ + netaddr = gf_strdup (addrtok); + if (netaddr == NULL) /* ENOMEM */ + goto out; + + /* Find the network socket addr of target */ + if (inet_pton (AF_INET, ipaddr, &sin1.sin_addr) == 0) + goto out; + + /* Find the network socket addr of subnet pattern */ + slash = strchr (netaddr, '/'); + *slash = '\0'; + if (inet_pton (AF_INET, netaddr, &sin2.sin_addr) == 0) + goto out; + + /* + * Find the network mask in network byte order. + * NB: 32 : Max len of IPv4 address. + */ + prefixlen = atoi (slash + 1); + shift = 32 - (uint32_t)prefixlen; + mask.sin_addr.s_addr = htonl ((uint32_t)~0 << shift); + + /* + * Check if both have same network address. + * Extract the network address from the IP addr by applying the + * network mask. If they match, return SUCCESS. i.e. + * + * (x == y) <=> (x ^ y == 0) + * (x & y) ^ (x & z) <=> x & (y ^ z) + * + * ((ip1 & mask) == (ip2 & mask)) <=> ((mask & (ip1 ^ ip2)) == 0) + */ + if (((mask.sin_addr.s_addr) & + (sin1.sin_addr.s_addr ^ sin2.sin_addr.s_addr)) != 0) + goto out; + + ret = 0; /* SUCCESS */ +out: + GF_FREE (netaddr); + return ret; +} + rpcsvc_actor_t gluster_dump_actors[] = { - [GF_DUMP_NULL] = {"NULL", GF_DUMP_NULL, NULL, NULL, NULL, 0}, - [GF_DUMP_DUMP] = {"DUMP", GF_DUMP_DUMP, rpcsvc_dump, NULL, NULL, 0}, - [GF_DUMP_MAXVALUE] = {"MAXVALUE", GF_DUMP_MAXVALUE, NULL, 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 b15476704..5a991d499 100644 --- a/rpc/rpc-lib/src/rpcsvc.h +++ b/rpc/rpc-lib/src/rpcsvc.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _RPCSVC_H @@ -47,16 +38,25 @@ #define MAX_IOVEC 16 #endif +#define RPCSVC_DEFAULT_OUTSTANDING_RPC_LIMIT 64 /* Default for protocol/server */ +#define RPCSVC_DEF_NFS_OUTSTANDING_RPC_LIMIT 16 /* Default for nfs/server */ +#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)) #define RPCSVC_FRAGHDR_SIZE 4 /* 4-byte RPC fragment header size */ #define RPCSVC_DEFAULT_LISTEN_PORT GF_DEFAULT_BASE_PORT -#define RPCSVC_DEFAULT_MEMFACTOR 15 +#define RPCSVC_DEFAULT_MEMFACTOR 8 #define RPCSVC_EVENTPOOL_SIZE_MULT 1024 -#define RPCSVC_POOLCOUNT_MULT 35 +#define RPCSVC_POOLCOUNT_MULT 64 #define RPCSVC_CONN_READ (128 * GF_UNIT_KB) #define RPCSVC_PAGE_SIZE (128 * GF_UNIT_KB) +#define RPC_ROOT_UID 0 +#define RPC_ROOT_GID 0 +#define RPC_NOBODY_UID 65534 +#define RPC_NOBODY_GID 65534 /* RPC Record States */ #define RPCSVC_READ_FRAGHDR 1 @@ -145,6 +145,9 @@ 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. @@ -183,7 +186,9 @@ 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; @@ -233,6 +238,9 @@ struct rpcsvc_request { */ rpcsvc_auth_data_t verf; + /* Execute this request's actor function as a synctask? */ + gf_boolean_t synctask; + /* Container for a RPC program wanting to store a temp * request-specific item. */ @@ -240,6 +248,12 @@ struct rpcsvc_request { /* Container for transport to store request-specific item */ void *trans_private; + + /* 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)) @@ -247,8 +261,6 @@ struct rpcsvc_request { #define rpcsvc_request_program_private(req) (((rpcsvc_program_t *)((req)->prog))->private) #define rpcsvc_request_accepted(req) ((req)->rpc_status == MSG_ACCEPTED) #define rpcsvc_request_accepted_success(req) ((req)->rpc_err == SUCCESS) -#define rpcsvc_request_uid(req) ((req)->uid) -#define rpcsvc_request_gid(req) ((req)->gid) #define rpcsvc_request_prog_minauth(req) (rpcsvc_request_program(req)->min_auth) #define rpcsvc_request_cred_flavour(req) (rpcsvc_auth_flavour(req->cred)) #define rpcsvc_request_verf_flavour(req) (rpcsvc_auth_flavour(req->verf)) @@ -265,7 +277,24 @@ struct rpcsvc_request { #define rpcsvc_request_set_vecstate(req, state) ((req)->vecstate = state) #define rpcsvc_request_vecstate(req) ((req)->vecstate) #define rpcsvc_request_transport(req) ((req)->trans) - +#define rpcsvc_request_transport_ref(req) (rpc_transport_ref((req)->trans)) +#define RPC_AUTH_ROOT_SQUASH(req) \ + do { \ + int gidcount = 0; \ + if (req->svc->root_squash) { \ + if (req->uid == RPC_ROOT_UID) \ + req->uid = req->svc->anonuid; \ + if (req->gid == RPC_ROOT_GID) \ + req->gid = req->svc->anongid; \ + \ + for (gidcount = 0; gidcount < req->auxgidcount; \ + ++gidcount) { \ + if (!req->auxgids[gidcount]) \ + req->auxgids[gidcount] = \ + req->svc->anongid; \ + } \ + } \ + } while (0); #define RPCSVC_ACTOR_SUCCESS 0 #define RPCSVC_ACTOR_ERROR (-1) @@ -284,9 +313,8 @@ struct rpcsvc_request { * */ typedef int (*rpcsvc_actor) (rpcsvc_request_t *req); -typedef int (*rpcsvc_vector_actor) (rpcsvc_request_t *req, struct iovec *vec, - int count, struct iobref *iobref); -typedef int (*rpcsvc_vector_sizer) (int state, ssize_t *readsize, char *addr); +typedef int (*rpcsvc_vector_sizer) (int state, ssize_t *readsize, + char *base_addr, char *curr_addr); /* Every protocol actor will also need to specify the function the RPC layer * will use to serialize or encode the message into XDR format just before @@ -300,7 +328,6 @@ 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. @@ -318,11 +345,11 @@ typedef struct rpcsvc_actor_desc { * handler for letting the RPC program read the data from the network * directly into its aligned buffers. */ - rpcsvc_vector_actor vector_actor; rpcsvc_vector_sizer vector_sizer; /* 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 @@ -377,6 +404,9 @@ struct rpcsvc_program { */ int min_auth; + /* Execute actor function as a synctask? */ + gf_boolean_t synctask; + /* list member to link to list of registered services with rpcsvc */ struct list_head program; }; @@ -413,13 +443,20 @@ 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. * Called in main. */ extern rpcsvc_t * -rpcsvc_init (xlator_t *xl, glusterfs_ctx_t *ctx, dict_t *options); +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); @@ -431,6 +468,13 @@ 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); @@ -451,17 +495,17 @@ rpcsvc_error_reply (rpcsvc_request_t *req); extern int rpcsvc_transport_peername (rpc_transport_t *trans, char *hostname, int hostlen); -extern inline int +extern int 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 @@ -511,12 +555,15 @@ struct rpcsvc_auth_list { }; extern int -rpcsvc_auth_request_init (rpcsvc_request_t *req); +rpcsvc_auth_request_init (rpcsvc_request_t *req, struct rpc_msg *callmsg); 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 @@ -533,9 +580,6 @@ 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); @@ -543,16 +587,22 @@ 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 defvalue); +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-common.h b/rpc/rpc-lib/src/xdr-common.h index d5198d20f..631f4b284 100644 --- a/rpc/rpc-lib/src/xdr-common.h +++ b/rpc/rpc-lib/src/xdr-common.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _XDR_COMMON_H_ @@ -28,6 +19,7 @@ #include <rpc/types.h> #include <sys/types.h> #include <rpc/xdr.h> +#include <rpc/auth.h> #include <sys/uio.h> #ifdef __NetBSD__ @@ -43,28 +35,61 @@ enum gf_dump_procnum { #define GLUSTER_DUMP_PROGRAM 123451501 /* Completely random */ #define GLUSTER_DUMP_VERSION 1 -#define GF_MAX_AUTH_BYTES 2048 +/* MAX_AUTH_BYTES is restricted to 400 bytes, see + * http://tools.ietf.org/html/rfc5531#section-8.2 */ +#define GF_MAX_AUTH_BYTES MAX_AUTH_BYTES + +/* The size of an AUTH_GLUSTERFS_V2 structure: + * + * 1 | pid + * 1 | uid + * 1 | gid + * 1 | groups_len + * XX | groups_val (GF_MAX_AUX_GROUPS=65535) + * 1 | lk_owner_len + * YY | lk_owner_val (GF_MAX_LOCK_OWNER_LEN=1024) + * ----+------------------------------------------- + * 5 | total xdr-units + * + * one XDR-unit is defined as BYTES_PER_XDR_UNIT = 4 bytes + * MAX_AUTH_BYTES = 400 is the maximum, this is 100 xdr-units. + * XX + YY can be 95 to fill the 100 xdr-units. + * + * Note that the on-wire protocol has tighter requirements than the internal + * structures. It is possible for xlators to use more groups and a bigger + * lk_owner than that can be sent by a GlusterFS-client. + */ +#define GF_AUTH_GLUSTERFS_MAX_GROUPS(lk_owner_len) \ + (95 - lk_owner_len) +#define GF_AUTH_GLUSTERFS_MAX_LKOWNER(groups_len) \ + (95 - groups_len) + +#ifdef GF_LINUX_HOST_OS +#define xdr_u_int32_t xdr_uint32_t +#define xdr_u_int64_t xdr_uint64_t +#endif -#if GF_DARWIN_HOST_OS +#ifdef GF_DARWIN_HOST_OS #define xdr_u_quad_t xdr_u_int64_t #define xdr_quad_t xdr_int64_t #define xdr_uint32_t xdr_u_int32_t +#define xdr_uint64_t xdr_u_int64_t #define uint64_t u_int64_t +unsigned long xdr_sizeof (xdrproc_t func, void *data); #endif #if defined(__NetBSD__) #define xdr_u_quad_t xdr_u_int64_t #define xdr_quad_t xdr_int64_t #define xdr_uint32_t xdr_u_int32_t +#define xdr_uint64_t xdr_u_int64_t #endif - -#if GF_SOLARIS_HOST_OS +#ifdef GF_SOLARIS_HOST_OS #define u_quad_t uint64_t #define quad_t int64_t #define xdr_u_quad_t xdr_uint64_t #define xdr_quad_t xdr_int64_t -#define xdr_uint32_t xdr_uint32_t #endif /* Returns the address of the byte that follows the diff --git a/rpc/rpc-lib/src/xdr-rpc.c b/rpc/rpc-lib/src/xdr-rpc.c index e4e5b1087..adb48a531 100644 --- a/rpc/rpc-lib/src/xdr-rpc.c +++ b/rpc/rpc-lib/src/xdr-rpc.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 @@ -43,7 +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-lib/src/xdr-rpc.h b/rpc/rpc-lib/src/xdr-rpc.h index 226f8e8b3..4663ef796 100644 --- a/rpc/rpc-lib/src/xdr-rpc.h +++ b/rpc/rpc-lib/src/xdr-rpc.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _XDR_RPC_H_ @@ -43,7 +34,8 @@ typedef enum { AUTH_GLUSTERFS = 5, - AUTH_GLUSTERFS_v2 = 6, + AUTH_GLUSTERFS_v2 = 390039, /* using a number from 'unused' range, + from the list available in RFC5531 */ } gf_rpc_authtype_t; /* Converts a given network buffer from its XDR format to a structure @@ -86,4 +78,21 @@ xdr_to_auth_unix_cred (char *msgbuf, int msglen, struct authunix_parms *au, #define rpc_call_verf_flavour(call) (rpc_opaque_auth_flavour ((&(call)->ru.RM_cmb.cb_verf))) #define rpc_call_verf_len(call) (rpc_opaque_auth_len ((&(call)->ru.RM_cmb.cb_verf))) + +#ifdef GF_DARWIN_HOST_OS +#define GF_PRI_RPC_XID PRIu32 +#define GF_PRI_RPC_VERSION PRIu32 +#define GF_PRI_RPC_PROG_ID PRIu32 +#define GF_PRI_RPC_PROG_VERS PRIu32 +#define GF_PRI_RPC_PROC PRIu32 +#define GF_PRI_RPC_PROC_VERSION PRIu32 +#else +#define GF_PRI_RPC_XID PRIu64 +#define GF_PRI_RPC_VERSION PRIu64 +#define GF_PRI_RPC_PROG_ID PRIu64 +#define GF_PRI_RPC_PROG_VERS PRIu64 +#define GF_PRI_RPC_PROC PRIu64 +#define GF_PRI_RPC_PROC_VERSION PRIu64 +#endif + #endif diff --git a/rpc/rpc-lib/src/xdr-rpcclnt.c b/rpc/rpc-lib/src/xdr-rpcclnt.c index 69daa98cb..810d1961b 100644 --- a/rpc/rpc-lib/src/xdr-rpcclnt.c +++ b/rpc/rpc-lib/src/xdr-rpcclnt.c @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 diff --git a/rpc/rpc-lib/src/xdr-rpcclnt.h b/rpc/rpc-lib/src/xdr-rpcclnt.h index 3bcf5b54b..c08d872f8 100644 --- a/rpc/rpc-lib/src/xdr-rpcclnt.h +++ b/rpc/rpc-lib/src/xdr-rpcclnt.h @@ -1,20 +1,11 @@ /* - Copyright (c) 2010-2011 Gluster, Inc. <http://www.gluster.com> + Copyright (c) 2008-2012 Red Hat, Inc. <http://www.redhat.com> This file is part of GlusterFS. - GlusterFS is free software; you can redistribute it and/or modify - it under the terms of the GNU General Public License as published - by the Free Software Foundation; either version 3 of the License, - or (at your option) any later version. - - GlusterFS is distributed in the hope that it will be useful, but - WITHOUT ANY WARRANTY; without even the implied warranty of - MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - General Public License for more details. - - You should have received a copy of the GNU General Public License - along with this program. If not, see - <http://www.gnu.org/licenses/>. + 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 _XDR_RPCCLNT_H |
