From 8b2949db0d56bdf5842abcb72437cc7dccd884df Mon Sep 17 00:00:00 2001 From: Shehjar Tikoo Date: Wed, 31 Mar 2010 07:27:02 +0000 Subject: nfs: Add RPCv2 service Signed-off-by: Shehjar Tikoo Signed-off-by: Anand V. Avati BUG: 399 (NFS translator with Mount v3 and NFS v3 support) URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=399 --- xlators/nfs/lib/src/Makefile.am | 4 +- xlators/nfs/lib/src/auth-null.c | 71 + xlators/nfs/lib/src/auth-unix.c | 91 ++ xlators/nfs/lib/src/rpc-socket.c | 358 +++++ xlators/nfs/lib/src/rpc-socket.h | 65 + xlators/nfs/lib/src/rpcsvc-auth.c | 391 ++++++ xlators/nfs/lib/src/rpcsvc.c | 2743 +++++++++++++++++++++++++++++++++++++ xlators/nfs/lib/src/rpcsvc.h | 715 ++++++++++ 8 files changed, 4436 insertions(+), 2 deletions(-) create mode 100644 xlators/nfs/lib/src/auth-null.c create mode 100644 xlators/nfs/lib/src/auth-unix.c create mode 100644 xlators/nfs/lib/src/rpc-socket.c create mode 100644 xlators/nfs/lib/src/rpc-socket.h create mode 100644 xlators/nfs/lib/src/rpcsvc-auth.c create mode 100644 xlators/nfs/lib/src/rpcsvc.c create mode 100644 xlators/nfs/lib/src/rpcsvc.h diff --git a/xlators/nfs/lib/src/Makefile.am b/xlators/nfs/lib/src/Makefile.am index d0221f4940b..a6090bb2896 100644 --- a/xlators/nfs/lib/src/Makefile.am +++ b/xlators/nfs/lib/src/Makefile.am @@ -1,10 +1,10 @@ lib_LTLIBRARIES = librpcsvc.la librpcsvc_la_LDFLAGS = -module -avoidversion -librpcsvc_la_SOURCES = msg-nfs3.c xdr-nfs3.c xdr-rpc.c +librpcsvc_la_SOURCES = msg-nfs3.c xdr-nfs3.c xdr-rpc.c auth-unix.c rpcsvc-auth.c rpcsvc.c auth-null.c rpc-socket.c librpcsvc_la_LIBADD = $(top_builddir)/libglusterfs/src/libglusterfs.la -noinst_HEADERS = xdr-rpc.h msg-nfs3.h xdr-common.h xdr-nfs3.h +noinst_HEADERS = xdr-rpc.h msg-nfs3.h xdr-common.h xdr-nfs3.h rpc-socket.h rpcsvc.h AM_CFLAGS = -fPIC -D_FILE_OFFSET_BITS=64 -D_GNU_SOURCE -Wall -D$(GF_HOST_OS)\ -I$(top_srcdir)/libglusterfs/src -shared -nostartfiles $(GF_CFLAGS) diff --git a/xlators/nfs/lib/src/auth-null.c b/xlators/nfs/lib/src/auth-null.c new file mode 100644 index 00000000000..b162db11247 --- /dev/null +++ b/xlators/nfs/lib/src/auth-null.c @@ -0,0 +1,71 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc.h" +#include "list.h" +#include "dict.h" + + +int +auth_null_request_init (rpcsvc_request_t *req, void *priv) +{ + if (!req) + return -1; + + memset (req->cred.authdata, 0, RPCSVC_MAX_AUTH_BYTES); + req->cred.datalen = 0; + + memset (req->verf.authdata, 0, RPCSVC_MAX_AUTH_BYTES); + req->verf.datalen = 0; + + return 0; +} + +int auth_null_authenticate (rpcsvc_request_t *req, void *priv) +{ + /* Always succeed. */ + return RPCSVC_AUTH_ACCEPT; +} + +rpcsvc_auth_ops_t auth_null_ops = { + .conn_init = NULL, + .request_init = auth_null_request_init, + .authenticate = auth_null_authenticate +}; + +rpcsvc_auth_t rpcsvc_auth_null = { + .authname = "AUTH_NULL", + .authnum = AUTH_NULL, + .authops = &auth_null_ops, + .authprivate = NULL +}; + + +rpcsvc_auth_t * +rpcsvc_auth_null_init (rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_null; +} + diff --git a/xlators/nfs/lib/src/auth-unix.c b/xlators/nfs/lib/src/auth-unix.c new file mode 100644 index 00000000000..0eaf0686654 --- /dev/null +++ b/xlators/nfs/lib/src/auth-unix.c @@ -0,0 +1,91 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + + + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc.h" +#include "list.h" +#include "dict.h" +#include "xdr-rpc.h" + + +int +auth_unix_request_init (rpcsvc_request_t *req, void *priv) +{ + if (!req) + return -1; + memset (req->verf.authdata, 0, RPCSVC_MAX_AUTH_BYTES); + req->verf.datalen = 0; + req->verf.flavour = AUTH_NULL; + + return 0; +} + +int auth_unix_authenticate (rpcsvc_request_t *req, void *priv) +{ + int ret = RPCSVC_AUTH_REJECT; + struct authunix_parms aup; + char machname[MAX_MACHINE_NAME]; + + if (!req) + return ret; + + ret = xdr_to_auth_unix_cred (req->cred.authdata, req->cred.datalen, + &aup, machname, req->auxgids); + if (ret == -1) { + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + req->uid = aup.aup_uid; + req->gid = aup.aup_gid; + req->auxgidcount = aup.aup_len; + + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Auth Info: machine name: %s, uid: %d" + ", gid: %d", machname, req->uid, req->gid); + ret = RPCSVC_AUTH_ACCEPT; +err: + return ret; +} + +rpcsvc_auth_ops_t auth_unix_ops = { + .conn_init = NULL, + .request_init = auth_unix_request_init, + .authenticate = auth_unix_authenticate +}; + +rpcsvc_auth_t rpcsvc_auth_unix = { + .authname = "AUTH_UNIX", + .authnum = AUTH_UNIX, + .authops = &auth_unix_ops, + .authprivate = NULL +}; + + +rpcsvc_auth_t * +rpcsvc_auth_unix_init (rpcsvc_t *svc, dict_t *options) +{ + return &rpcsvc_auth_unix; +} + diff --git a/xlators/nfs/lib/src/rpc-socket.c b/xlators/nfs/lib/src/rpc-socket.c new file mode 100644 index 00000000000..01f114a8530 --- /dev/null +++ b/xlators/nfs/lib/src/rpc-socket.c @@ -0,0 +1,358 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpc-socket.h" +#include "rpcsvc.h" +#include "dict.h" +#include "logging.h" +#include "byte-order.h" +#include "common-utils.h" +#include "compat-errno.h" + +#include +#include +#include +#include +#include +#include + +static int +rpcsvc_socket_server_get_local_socket (int addrfam, char *listenhost, + uint16_t listenport, + struct sockaddr *addr, + socklen_t *addr_len) +{ + struct addrinfo hints, *res = 0; + char service[NI_MAXSERV]; + int ret = -1; + + memset (service, 0, sizeof (service)); + sprintf (service, "%d", listenport); + + memset (&hints, 0, sizeof (hints)); + addr->sa_family = hints.ai_family = addrfam; + hints.ai_socktype = SOCK_STREAM; + hints.ai_flags = AI_ADDRCONFIG | AI_PASSIVE; + + ret = getaddrinfo(listenhost, service, &hints, &res); + if (ret != 0) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, + "getaddrinfo failed for host %s, service %s (%s)", + listenhost, service, gai_strerror (ret)); + ret = -1; + goto err; + } + + memcpy (addr, res->ai_addr, res->ai_addrlen); + *addr_len = res->ai_addrlen; + + freeaddrinfo (res); + ret = 0; + +err: + return ret; +} + + +int +rpcsvc_socket_listen (int addrfam, char *listenhost, uint16_t listenport) +{ + int sock = -1; + struct sockaddr_storage sockaddr; + socklen_t sockaddr_len; + int flags = 0; + int ret = -1; + int opt = 1; + + ret = rpcsvc_socket_server_get_local_socket (addrfam, listenhost, + listenport,SA (&sockaddr), + &sockaddr_len); + + if (ret == -1) + return ret; + + sock = socket (SA (&sockaddr)->sa_family, SOCK_STREAM, 0); + if (sock == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "socket creation failed" + " (%s)", strerror (errno)); + goto err; + } + + flags = fcntl (sock, F_GETFL); + if (flags == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "cannot get socket flags" + " (%s)", strerror(errno)); + goto close_err; + } + + ret = fcntl (sock, F_SETFL, flags | O_NONBLOCK); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "cannot set socket " + "non-blocking (%s)", strerror (errno)); + goto close_err; + } + + ret = setsockopt (sock, SOL_SOCKET, SO_REUSEADDR, &opt, sizeof (opt)); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "setsockopt() for " + "SO_REUSEADDR failed (%s)", strerror (errno)); + goto close_err; + } + + ret = bind (sock, (struct sockaddr *)&sockaddr, sockaddr_len); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "binding socket failed:" + " %s", strerror (errno)); + if (errno == EADDRINUSE) + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "Port is already" + " in use"); + goto close_err; + } + + ret = listen (sock, 10); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "could not listen on" + " socket (%s)", strerror (errno)); + goto close_err; + } + + return sock; + +close_err: + close (sock); + sock = -1; + +err: + return sock; +} + + +int +rpcsvc_socket_accept (int listenfd) +{ + int new_sock = -1; + struct sockaddr_storage new_sockaddr = {0, }; + socklen_t addrlen = sizeof (new_sockaddr); + int flags = 0; + int ret = -1; + int on = 1; + + new_sock = accept (listenfd, SA (&new_sockaddr), &addrlen); + if (new_sock == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR,"accept on socket failed"); + goto err; + } + + flags = fcntl (new_sock, F_GETFL); + if (flags == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "cannot get socket flags" + " (%s)", strerror(errno)); + goto close_err; + } + + ret = fcntl (new_sock, F_SETFL, flags | O_NONBLOCK); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "cannot set socket " + "non-blocking (%s)", strerror (errno)); + goto close_err; + } + +#ifdef TCP_NODELAY + ret = setsockopt(new_sock, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "cannot set no-delay " + " socket option"); + } +#endif + + return new_sock; + +close_err: + close (new_sock); + new_sock = -1; + +err: + return new_sock; +} + +ssize_t +rpcsvc_socket_read (int sockfd, char *readaddr, size_t readsize) +{ + ssize_t dataread = 0; + ssize_t readlen = -1; + + if (!readaddr) + return -1; + + while (readsize > 0) { + readlen = read (sockfd, readaddr, readsize); + if (readlen == -1) { + if (errno != EAGAIN) { + dataread = -1; + break; + } else + break; + } else if (readlen == 0) + break; + + dataread += readlen; + readaddr += readlen; + readsize -= readlen; + } + + return dataread; +} + + +ssize_t +rpcsvc_socket_write (int sockfd, char *buffer, size_t size) +{ + size_t writelen = -1; + ssize_t written = 0; + + if (!buffer) + return -1; + + while (size > 0) { + writelen = write (sockfd, buffer, size); + if (writelen == -1) { + if (errno != EAGAIN) { + written = -1; + break; + } else + break; + } else if (writelen == 0) + break; + + written += writelen; + size -= writelen; + buffer += writelen; + } + + return written; +} + + +int +rpcsvc_socket_peername (int sockfd, char *hostname, int hostlen) +{ + struct sockaddr sa; + socklen_t sl = sizeof (sa); + int ret = EAI_FAIL; + + if (!hostname) + return ret; + + ret = getpeername (sockfd, &sa, &sl); + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "Failed to get peer name:" + " %s", strerror (errno)); + ret = EAI_FAIL; + goto err; + } + + ret = getnameinfo (&sa, sl, hostname, hostlen, NULL, 0, 0); + if (ret != 0) + goto err; + +err: + return ret; +} + + +int +rpcsvc_socket_peeraddr (int sockfd, char *addrstr, int addrlen, + struct sockaddr *returnsa, socklen_t sasize) +{ + struct sockaddr sa; + int ret = EAI_FAIL; + + if (returnsa) + ret = getpeername (sockfd, returnsa, &sasize); + else { + sasize = sizeof (sa); + ret = getpeername (sockfd, &sa, &sasize); + } + + if (ret == -1) { + gf_log (GF_RPCSVC_SOCK, GF_LOG_ERROR, "Failed to get peer addr:" + " %s", strerror (errno)); + ret = EAI_FAIL; + goto err; + } + + /* If caller did not specify a string into which the address can be + * stored, dont bother getting it. + */ + if (!addrstr) { + ret = 0; + goto err; + } + + if (returnsa) + ret = getnameinfo (returnsa, sasize, addrstr, addrlen, NULL, 0, + NI_NUMERICHOST); + else + ret = getnameinfo (&sa, sasize, addrstr, addrlen, NULL, 0, + NI_NUMERICHOST); + +err: + return ret; +} + + +int +rpcsvc_socket_block_tx (int sockfd) +{ + int ret = -1; + int on = 1; + +#ifdef TCP_CORK + ret = setsockopt(sockfd, IPPROTO_TCP, TCP_CORK, &on, sizeof(on)); +#endif + +#ifdef TCP_NOPUSH + ret = setsockopt(sockfd, IPPROTO_TCP, TCP_NOPUSH, &on, sizeof(on)); +#endif + + return ret; +} + + +int +rpcsvc_socket_unblock_tx (int sockfd) +{ + int ret = -1; + int off = 0; + +#ifdef TCP_CORK + ret = setsockopt(sockfd, IPPROTO_TCP, TCP_CORK, &off, sizeof(off)); +#endif + +#ifdef TCP_NOPUSH + ret = setsockopt(sockfd, IPPROTO_TCP, TCP_NOPUSH, &off, sizeof(off)); +#endif + return ret; +} + diff --git a/xlators/nfs/lib/src/rpc-socket.h b/xlators/nfs/lib/src/rpc-socket.h new file mode 100644 index 00000000000..3a50c97a98d --- /dev/null +++ b/xlators/nfs/lib/src/rpc-socket.h @@ -0,0 +1,65 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + +#ifndef _RPCSVC_SOCKET_H_ +#define _RPCSVC_SOCKET_H_ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc.h" +#include "dict.h" +#include "logging.h" +#include "byte-order.h" +#include "common-utils.h" +#include "compat-errno.h" + +#include +#include +#include +#include + +#define SA(ptr) ((struct sockaddr *)ptr) +#define GF_RPCSVC_SOCK "rpc-socket" +extern int +rpcsvc_socket_listen (int addrfam, char *listenhost, uint16_t listenport); + +extern int +rpcsvc_socket_accept (int listenfd); + +extern ssize_t +rpcsvc_socket_read (int sockfd, char *readaddr, size_t readsize); + +extern ssize_t +rpcsvc_socket_write (int sockfd, char *buffer, size_t size); + +extern int +rpcsvc_socket_peername (int sockfd, char *hostname, int hostlen); + +extern int +rpcsvc_socket_peeraddr (int sockfd, char *addrstr, int addrlen, + struct sockaddr *returnsa, socklen_t sasize); +extern int +rpcsvc_socket_block_tx (int sockfd); + +extern int +rpcsvc_socket_unblock_tx (int sockfd); +#endif diff --git a/xlators/nfs/lib/src/rpcsvc-auth.c b/xlators/nfs/lib/src/rpcsvc-auth.c new file mode 100644 index 00000000000..38697965bbd --- /dev/null +++ b/xlators/nfs/lib/src/rpcsvc-auth.c @@ -0,0 +1,391 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + +#include "rpcsvc.h" +#include "logging.h" +#include "dict.h" + +extern rpcsvc_auth_t * +rpcsvc_auth_null_init (rpcsvc_t *svc, dict_t *options); + +extern rpcsvc_auth_t * +rpcsvc_auth_unix_init (rpcsvc_t *svc, dict_t *options); + +int +rpcsvc_auth_add_initer (struct list_head *list, char *idfier, + rpcsvc_auth_initer_t init) +{ + struct rpcsvc_auth_list *new = NULL; + + if ((!list) || (!init) || (!idfier)) + return -1; + + new = CALLOC (1, sizeof (*new)); + if (!new) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Memory allocation failed"); + return -1; + } + + new->init = init; + strcpy (new->name, idfier); + INIT_LIST_HEAD (&new->authlist); + list_add_tail (&new->authlist, list); + return 0; +} + + + +int +rpcsvc_auth_add_initers (rpcsvc_t *svc) +{ + int ret = -1; + + ret = rpcsvc_auth_add_initer (&svc->authschemes, "auth-unix", + (rpcsvc_auth_initer_t) + rpcsvc_auth_unix_init); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_UNIX"); + goto err; + } + + ret = rpcsvc_auth_add_initer (&svc->authschemes, "auth-null", + (rpcsvc_auth_initer_t) + rpcsvc_auth_null_init); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to add AUTH_NULL"); + goto err; + } + + ret = 0; +err: + return 0; +} + + +int +rpcsvc_auth_init_auth (rpcsvc_t *svc, dict_t *options, + struct rpcsvc_auth_list *authitem) +{ + int ret = -1; + + if ((!svc) || (!options) || (!authitem)) + return -1; + + if (!authitem->init) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "No init function defined"); + ret = -1; + goto err; + } + + authitem->auth = authitem->init (svc, options); + if (!authitem->auth) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Registration of auth failed:" + " %s", authitem->name); + ret = -1; + goto err; + } + + authitem->enable = 1; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Authentication enabled: %s", + authitem->auth->authname); + + ret = 0; +err: + return ret; +} + + +int +rpcsvc_auth_init_auths (rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + + if (!svc) + return -1; + + if (list_empty (&svc->authschemes)) { + gf_log (GF_RPCSVC, GF_LOG_WARNING, "No authentication!"); + ret = 0; + goto err; + } + + /* If auth null and sys are not disabled by the user, we must enable + * it by default. This is a globally default rule, the user is still + * allowed to disable the two for particular subvolumes. + */ + if (!dict_get (options, "rpc-auth.auth-null")) + ret = dict_set_dynstr (options, "rpc-auth.auth-null", "on"); + + if (!dict_get (options, "rpc-auth.auth-unix")) + ret = dict_set_dynstr (options, "rpc-auth.auth-unix", "on"); + + list_for_each_entry_safe (auth, tmp, &svc->authschemes, authlist) { + ret = rpcsvc_auth_init_auth (svc, options, auth); + if (ret == -1) + goto err; + } + + ret = 0; +err: + return ret; + +} + +int +rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options) +{ + int ret = -1; + + if ((!svc) || (!options)) + return -1; + + ret = rpcsvc_auth_add_initers (svc); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to add initers"); + goto out; + } + + ret = rpcsvc_auth_init_auths (svc, options); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to init auth schemes"); + goto out; + } + +out: + return ret; +} + + +rpcsvc_auth_t * +__rpcsvc_auth_get_handler (rpcsvc_request_t *req) +{ + int ret = -1; + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + rpcsvc_t *svc = NULL; + + if (!req) + return NULL; + + svc = rpcsvc_request_service (req); + if (list_empty (&svc->authschemes)) { + gf_log (GF_RPCSVC, GF_LOG_WARNING, "No authentication!"); + ret = 0; + goto err; + } + + list_for_each_entry_safe (auth, tmp, &svc->authschemes, authlist) { + if (!auth->enable) + continue; + if (auth->auth->authnum == req->cred.flavour) + goto err; + + } + + auth = NULL; +err: + if (auth) + return auth->auth; + else + return NULL; +} + +rpcsvc_auth_t * +rpcsvc_auth_get_handler (rpcsvc_request_t *req) +{ + rpcsvc_auth_t *auth = NULL; + + auth = __rpcsvc_auth_get_handler (req); + if (auth) + goto ret; + + gf_log (GF_RPCSVC, GF_LOG_TRACE, "No auth handler: %d", + req->cred.flavour); + + /* The requested scheme was not available so fall back the to one + * scheme that will always be present. + */ + req->cred.flavour = AUTH_NULL; + req->verf.flavour = AUTH_NULL; + auth = __rpcsvc_auth_get_handler (req); +ret: + return auth; +} + + +int +rpcsvc_auth_request_init (rpcsvc_request_t *req) +{ + int ret = -1; + rpcsvc_auth_t *auth = NULL; + + if (!req) + return -1; + + auth = rpcsvc_auth_get_handler (req); + if (!auth) + 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); + +err: + return ret; +} + + +int +rpcsvc_authenticate (rpcsvc_request_t *req) +{ + int ret = RPCSVC_AUTH_REJECT; + rpcsvc_auth_t *auth = NULL; + int minauth = 0; + + if (!req) + return ret; + + minauth = rpcsvc_request_prog_minauth (req); + if (minauth > rpcsvc_request_cred_flavour (req)) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Auth too weak"); + rpcsvc_request_set_autherr (req, AUTH_TOOWEAK); + goto err; + } + + auth = rpcsvc_auth_get_handler (req); + if (!auth) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "No auth handler found"); + goto err; + } + + if (auth->authops->authenticate) + ret = auth->authops->authenticate (req, auth->authprivate); + +err: + return ret; +} + + +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; + + struct rpcsvc_auth_list *auth = NULL; + struct rpcsvc_auth_list *tmp = NULL; + + if ((!svc) || (!autharr) || (!volname)) + return -1; + + memset (autharr, 0, arrlen * sizeof(int)); + if (list_empty (&svc->authschemes)) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "No authentication!"); + goto err; + } + + list_for_each_entry_safe (auth, tmp, &svc->authschemes, authlist) { + if (count >= arrlen) + break; + + gen = asprintf (&srchstr, "rpc-auth.%s", auth->name); + if (gen == -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"); + } + + FREE (srchstr); + spec = 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"); + } + + FREE (srchstr); + final = rpcsvc_combine_gen_spec_volume_checks (gen, spec); + if (final == RPCSVC_AUTH_ACCEPT) { + autharr[count] = auth->auth->authnum; + ++count; + } + } + +err: + return count; +} + + +gid_t * +rpcsvc_auth_unix_auxgids (rpcsvc_request_t *req, int *arrlen) +{ + if ((!req) || (!arrlen)) + return NULL; + + if (req->cred.flavour != AUTH_UNIX) + return NULL; + + *arrlen = req->auxgidcount; + if (*arrlen == 0) + return NULL; + + return &req->auxgids[0]; +} + diff --git a/xlators/nfs/lib/src/rpcsvc.c b/xlators/nfs/lib/src/rpcsvc.c new file mode 100644 index 00000000000..e76ee16087a --- /dev/null +++ b/xlators/nfs/lib/src/rpcsvc.c @@ -0,0 +1,2743 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "rpcsvc.h" +#include "rpc-socket.h" +#include "dict.h" +#include "logging.h" +#include "byte-order.h" +#include "common-utils.h" +#include "compat-errno.h" +#include "list.h" +#include "xdr-rpc.h" +#include "iobuf.h" +#include "globals.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + + +#define rpcsvc_alloc_request(con, request) \ + do { \ + request = (rpcsvc_request_t *) mem_get ((con)->rxpool); \ + memset (request, 0, sizeof (rpcsvc_request_t)); \ + } while (0) \ + +/* The generic event handler for every stage */ +void * +rpcsvc_stage_proc (void *arg) +{ + rpcsvc_stage_t *stg = (rpcsvc_stage_t *)arg; + + if (!stg) + return NULL; + + event_dispatch (stg->eventpool); + return NULL; +} + + +rpcsvc_stage_t * +rpcsvc_stage_init (rpcsvc_t *svc) +{ + rpcsvc_stage_t *stg = NULL; + int ret = -1; + size_t stacksize = RPCSVC_THREAD_STACK_SIZE; + pthread_attr_t stgattr; + unsigned int eventpoolsize = 0; + + if (!svc) + return NULL; + + stg = CALLOC (1, sizeof(*stg)); + if (!stg) + return NULL; + + eventpoolsize = svc->memfactor * RPCSVC_EVENTPOOL_SIZE_MULT; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "event pool size: %d", eventpoolsize); + stg->eventpool = event_pool_new (eventpoolsize); + if (!stg->eventpool) + goto free_stg; + + pthread_attr_init (&stgattr); + ret = pthread_attr_setstacksize (&stgattr, stacksize); + if (ret == EINVAL) + gf_log (GF_RPCSVC, GF_LOG_WARNING, + "Using default thread stack size"); + + ret = pthread_create (&stg->tid, &stgattr, rpcsvc_stage_proc, + (void *)stg); + if (ret != 0) { + ret = -1; + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Stage creation failed"); + goto free_stg; + } + + stg->svc = svc; + ret = 0; +free_stg: + if (ret == -1) { + FREE (stg); + stg = NULL; + } + + return stg; +} + + +int +rpcsvc_init_options (rpcsvc_t *svc, dict_t *options) +{ + svc->memfactor = RPCSVC_DEFAULT_MEMFACTOR; + return 0; +} + + +/* The global RPC service initializer. + * Starts up the stages and then waits for RPC program registrations + * to come in. + */ +rpcsvc_t * +rpcsvc_init (glusterfs_ctx_t *ctx, dict_t *options) +{ + rpcsvc_t *svc = NULL; + int ret = -1; + + if ((!ctx) || (!options)) + return NULL; + + svc = CALLOC (1, sizeof (*svc)); + if (!svc) + return NULL; + + pthread_mutex_init (&svc->rpclock, NULL); + INIT_LIST_HEAD (&svc->stages); + INIT_LIST_HEAD (&svc->authschemes); + + ret = rpcsvc_init_options (svc, options); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to init options"); + goto free_svc; + } + + ret = rpcsvc_auth_init (svc, options); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to init " + "authentication"); + goto free_svc; + } + + ret = -1; + svc->defaultstage = rpcsvc_stage_init (svc); + if (!svc->defaultstage) { + gf_log (GF_RPCSVC, GF_LOG_ERROR,"RPC service init failed."); + goto free_svc; + } + svc->options = options; + svc->ctx = ctx; + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "RPC service inited."); + + ret = 0; +free_svc: + if (ret == -1) { + FREE (svc); + svc = NULL; + } + + return svc; +} + + +/* Once multi-threaded support is complete, we'll be able to round-robin + * the various incoming connections over the many available stages. This + * function selects one from among all the stages. + */ +rpcsvc_stage_t * +rpcsvc_select_stage (rpcsvc_t *rpcservice) +{ + if (!rpcservice) + return NULL; + + return rpcservice->defaultstage; +} + + +int +rpcsvc_conn_peer_check_search (dict_t *options, char *pattern, char *clstr) +{ + int ret = -1; + char *addrtok = NULL; + char *addrstr = NULL; + char *svptr = NULL; + + if ((!options) || (!clstr)) + return -1; + + if (!dict_get (options, pattern)) + return -1; + + ret = dict_get_str (options, pattern, &addrstr); + if (ret < 0) { + ret = -1; + goto err; + } + + if (!addrstr) { + ret = -1; + goto err; + } + + addrtok = strtok_r (addrstr, ",", &svptr); + while (addrtok) { + + ret = fnmatch (addrtok, clstr, FNM_CASEFOLD); + if (ret == 0) + goto err; + + addrtok = strtok_r (NULL, ",", &svptr); + } + + ret = -1; +err: + + return ret; +} + + +int +rpcsvc_conn_peer_check_allow (dict_t *options, char *volname, char *clstr) +{ + int ret = RPCSVC_AUTH_DONTCARE; + char *srchstr = NULL; + char globalrule[] = "rpc-auth.addr.allow"; + + if ((!options) || (!clstr)) + 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 = 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 = rpcsvc_conn_peer_check_search (options, srchstr, clstr); + if (volname) + FREE (srchstr); + + if (ret == 0) + ret = RPCSVC_AUTH_ACCEPT; + else + ret = RPCSVC_AUTH_DONTCARE; +out: + return ret; +} + +int +rpcsvc_conn_peer_check_reject (dict_t *options, char *volname, char *clstr) +{ + int ret = RPCSVC_AUTH_DONTCARE; + char *srchstr = NULL; + char generalrule[] = "rpc-auth.addr.reject"; + + if ((!options) || (!clstr)) + return ret; + + if (volname) { + ret = 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 = rpcsvc_conn_peer_check_search (options, srchstr, clstr); + if (volname) + FREE (srchstr); + + if (ret == 0) + ret = RPCSVC_AUTH_REJECT; + else + ret = RPCSVC_AUTH_DONTCARE; +out: + return ret; +} + + +/* 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 | + */ +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; +} + + + +/* 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; +} + + +int +rpcsvc_conn_peer_check_name (dict_t *options, char *volname, + rpcsvc_conn_t *conn) +{ + int ret = RPCSVC_AUTH_REJECT; + int aret = RPCSVC_AUTH_REJECT; + int rjret = RPCSVC_AUTH_REJECT; + char clstr[RPCSVC_PEER_STRLEN]; + + if (!conn) + return ret; + + ret = rpcsvc_conn_peername (conn, 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_conn_peer_check_allow (options, volname, clstr); + rjret = rpcsvc_conn_peer_check_reject (options, volname, clstr); + + ret = rpcsvc_combine_allow_reject_volume_check (aret, rjret); + +err: + return ret; +} + + +int +rpcsvc_conn_peer_check_addr (dict_t *options, char *volname,rpcsvc_conn_t *conn) +{ + int ret = RPCSVC_AUTH_REJECT; + int aret = RPCSVC_AUTH_DONTCARE; + int rjret = RPCSVC_AUTH_REJECT; + char clstr[RPCSVC_PEER_STRLEN]; + + if (!conn) + return ret; + + ret = rpcsvc_conn_peeraddr (conn, clstr, RPCSVC_PEER_STRLEN, NULL, 0); + 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_conn_peer_check_allow (options, volname, clstr); + rjret = rpcsvc_conn_peer_check_reject (options, volname, clstr); + + ret = rpcsvc_combine_allow_reject_volume_check (aret, rjret); +err: + return ret; +} + + +int +rpcsvc_conn_check_volume_specific (dict_t *options, char *volname, + rpcsvc_conn_t *conn) +{ + int namechk = RPCSVC_AUTH_REJECT; + int addrchk = RPCSVC_AUTH_REJECT; + gf_boolean_t namelookup = _gf_true; + char *namestr = NULL; + int ret = 0; + + if ((!options) || (!volname) || (!conn)) + return RPCSVC_AUTH_REJECT; + + /* Enabled 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_conn_peer_check_name (options, volname, conn); + addrchk = rpcsvc_conn_peer_check_addr (options, volname, conn); + + if (namelookup) + ret = rpcsvc_combine_gen_spec_addr_checks (addrchk, namechk); + else + ret = addrchk; + + return ret; +} + + +int +rpcsvc_conn_check_volume_general (dict_t *options, rpcsvc_conn_t *conn) +{ + int addrchk = RPCSVC_AUTH_REJECT; + int namechk = RPCSVC_AUTH_REJECT; + gf_boolean_t namelookup = _gf_true; + char *namestr = NULL; + int ret = 0; + + if ((!options) || (!conn)) + return RPCSVC_AUTH_REJECT; + + /* Enabled 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_conn_peer_check_name (options, NULL, conn); + addrchk = rpcsvc_conn_peer_check_addr (options, NULL, conn); + + if (namelookup) + ret = rpcsvc_combine_gen_spec_addr_checks (addrchk, namechk); + else + ret = addrchk; + + return ret; +} + +int +rpcsvc_conn_peer_check (dict_t *options, char *volname, rpcsvc_conn_t *conn) +{ + int general_chk = RPCSVC_AUTH_REJECT; + int specific_chk = RPCSVC_AUTH_REJECT; + + if ((!options) || (!volname) || (!conn)) + return RPCSVC_AUTH_REJECT; + + general_chk = rpcsvc_conn_check_volume_general (options, conn); + specific_chk = rpcsvc_conn_check_volume_specific (options, volname, + conn); + + return rpcsvc_combine_gen_spec_volume_checks (general_chk,specific_chk); +} + + +char * +rpcsvc_volume_allowed (dict_t *options, char *volname) +{ + char globalrule[] = "rpc-auth.addr.allow"; + char *srchstr = NULL; + char *addrstr = NULL; + int ret = -1; + + if ((!options) || (!volname)) + return NULL; + + ret = asprintf (&srchstr, "rpc-auth.addr.%s.allow", volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + goto out; + } + + if (!dict_get (options, srchstr)) { + FREE (srchstr); + srchstr = globalrule; + ret = dict_get_str (options, srchstr, &addrstr); + } else + ret = dict_get_str (options, srchstr, &addrstr); + +out: + return addrstr; +} + + +/* Initialize the core of a connection */ +rpcsvc_conn_t * +rpcsvc_conn_init (rpcsvc_t *svc, rpcsvc_program_t *prog, int sockfd) +{ + rpcsvc_conn_t *conn = NULL; + int ret = -1; + unsigned int poolcount = 0; + + conn = CALLOC (1, sizeof(*conn)); + if (!conn) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "memory allocation failed"); + return NULL; + } + + conn->sockfd = sockfd; + conn->program = (void *)prog; + INIT_LIST_HEAD (&conn->txbufs); + poolcount = RPCSVC_POOLCOUNT_MULT * svc->memfactor; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "tx pool: %d", poolcount); + conn->txpool = mem_pool_new (rpcsvc_txbuf_t, poolcount); + if (!conn->txpool) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "mem pool allocation failed"); + goto free_conn; + } + + gf_log (GF_RPCSVC, GF_LOG_TRACE, "rx pool: %d", poolcount); + conn->rxpool = mem_pool_new (rpcsvc_request_t, poolcount); + if (!conn->rxpool) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "mem pool allocation failed"); + goto free_txp; + } + + /* Cannot consider a connection connected unless the user of this + * connection decides it is ready to use. It is possible that we have + * to free this connection soon after. That free will not happpen + * unless the state is disconnected. + */ + conn->connstate = RPCSVC_CONNSTATE_DISCONNECTED; + pthread_mutex_init (&conn->connlock, NULL); + conn->connref = 0; + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "New connection inited: sockfd: %d", + sockfd); + + ret = 0; +free_txp: + if (ret == -1) + mem_pool_destroy (conn->txpool); + +free_conn: + if (ret == -1) { + FREE (conn); + conn = NULL; + } + + return conn; +} + + +void +rpcsvc_conn_destroy (rpcsvc_conn_t *conn) +{ + mem_pool_destroy (conn->txpool); + mem_pool_destroy (conn->rxpool); + + if (conn->program->conn_destroy) + conn->program->conn_destroy (conn->program->private, conn); + + /* Need to destory record state, txlists etc. */ + FREE (conn); + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Connection destroyed"); +} + + +int +__rpcsvc_conn_unref (rpcsvc_conn_t *conn) +{ + --conn->connref; + return conn->connref; +} + + +void +__rpcsvc_conn_deinit (rpcsvc_conn_t *conn) +{ + if (!conn) + return; + + if ((conn->stage) && (conn->stage->eventpool)) { + event_unregister (conn->stage->eventpool, conn->sockfd, + conn->eventidx); + } + + if (rpcsvc_conn_check_active (conn)) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Connection de-activated:" + " sockfd: %d", conn->sockfd); + conn->connstate = RPCSVC_CONNSTATE_DISCONNECTED; + } + + if (conn->sockfd != -1) { + close (conn->sockfd); + conn->sockfd = -1; + } +} + + +void +rpcsvc_conn_deinit (rpcsvc_conn_t *conn) +{ + int ref = 0; + + if (!conn) + return; + + pthread_mutex_lock (&conn->connlock); + { + __rpcsvc_conn_deinit (conn); + ref = __rpcsvc_conn_unref (conn); + } + pthread_mutex_unlock (&conn->connlock); + + if (ref == 0) + rpcsvc_conn_destroy (conn); + + return; +} + + +void +rpcsvc_conn_unref (rpcsvc_conn_t *conn) +{ + int ref = 0; + if (!conn) + return; + + pthread_mutex_lock (&conn->connlock); + { + ref = __rpcsvc_conn_unref (conn); + } + pthread_mutex_unlock (&conn->connlock); + + if (ref == 0) + rpcsvc_conn_destroy (conn); +} + + +int +rpcsvc_conn_active (rpcsvc_conn_t *conn) +{ + int status = 0; + + if (!conn) + return 0; + + pthread_mutex_lock (&conn->connlock); + { + status = rpcsvc_conn_check_active (conn); + } + pthread_mutex_unlock (&conn->connlock); + + return status; +} + + + +void +rpcsvc_conn_ref (rpcsvc_conn_t *conn) +{ + if (!conn) + return; + + pthread_mutex_lock (&conn->connlock); + { + ++conn->connref; + } + pthread_mutex_unlock (&conn->connlock); + + return; +} + + +void +rpcsvc_conn_state_init (rpcsvc_conn_t *conn) +{ + if (!conn) + return; + + ++conn->connref; + conn->connstate = RPCSVC_CONNSTATE_CONNECTED; +} + +/* Builds a rpcsvc_conn_t with the aim of listening on it. + */ +rpcsvc_conn_t * +rpcsvc_conn_listen_init (rpcsvc_t *svc, rpcsvc_program_t *newprog) +{ + rpcsvc_conn_t *conn = NULL; + int sock = -1; + + if (!newprog) + return NULL; + + sock = rpcsvc_socket_listen (newprog->progaddrfamily, newprog->proghost, + newprog->progport); + if (sock == -1) + goto err; + + conn = rpcsvc_conn_init (svc, newprog, sock); + if (!conn) + goto sock_close_err; + + rpcsvc_conn_state_init (conn); +sock_close_err: + if (!conn) + close (sock); + +err: + return conn; +} + +void +rpcsvc_record_init (rpcsvc_record_state_t *rs, struct iobuf_pool *pool) +{ + if (!rs) + return; + + rs->state = RPCSVC_READ_FRAGHDR; + rs->vecstate = 0; + rs->remainingfraghdr = RPCSVC_FRAGHDR_SIZE; + rs->remainingfrag = 0; + rs->fragsize = 0; + rs->recordsize = 0; + rs->islastfrag = 0; + + /* If the rs preserves a ref to the iob used by the previous request, + * we must unref it here to prevent memory leak. + * If program actor wanted to keep that memory around, it should've + * refd it on entry into the actor. + */ + if (rs->activeiob) + iobuf_unref (rs->activeiob); + + if (rs->vectoriob) { + iobuf_unref (rs->vectoriob); + rs->vectoriob = NULL; + } + + rs->activeiob = iobuf_get (pool); + rs->fragcurrent = iobuf_ptr (rs->activeiob); + + memset (rs->fragheader, 0, RPCSVC_FRAGHDR_SIZE); + rs->hdrcurrent = &rs->fragheader[0]; + +} + + +int +rpcsvc_conn_privport_check (rpcsvc_t *svc, char *volname, rpcsvc_conn_t *conn) +{ + struct sockaddr_in sa; + int ret = RPCSVC_AUTH_REJECT; + socklen_t sasize = sizeof (sa); + 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; + + if ((!svc) || (!volname) || (!conn)) + return ret; + + ret = rpcsvc_conn_peeraddr (conn, NULL, 0, (struct sockaddr *)&sa, + sasize); + if (ret != 0) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get peer addr: %s", + gai_strerror (ret)); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + port = ntohs (sa.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. + */ + if (port <= 1024) { + ret = RPCSVC_AUTH_ACCEPT; + goto err; + } + + /* 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 = asprintf (&srchstr, "rpc-auth.ports.%s.insecure", volname); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "asprintf failed"); + ret = RPCSVC_AUTH_REJECT; + goto err; + } + + if (dict_get (svc->options, srchstr)) { + ret = dict_get_str (svc->options, srchstr, &valstr); + if (ret == 0) { + ret = gf_string2boolean (srchstr, &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); + if (ret == RPCSVC_AUTH_ACCEPT) + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Unprivileged port allowed"); + else + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Unprivileged port not" + " allowed"); + +err: + return ret; +} + +/* Inits a rpcsvc_conn_t after accepting the connection. + */ +rpcsvc_conn_t * +rpcsvc_conn_accept_init (rpcsvc_t *svc, int listenfd, + rpcsvc_program_t *destprog) +{ + rpcsvc_conn_t *newconn = NULL; + int sock = -1; + int ret = -1; + + sock = rpcsvc_socket_accept (listenfd); + if (sock == -1) + goto err; + + newconn = rpcsvc_conn_init (svc, destprog, sock); + if (!newconn) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to init conn object"); + ret = -1; + goto err; + } + + rpcsvc_record_init (&newconn->rstate, svc->ctx->iobuf_pool); + rpcsvc_conn_state_init (newconn); + if (destprog->conn_init) + destprog->conn_init (destprog->private, newconn); + ret = 0; + +err: + if (ret == -1) + close (sock); + + return newconn; +} + + +/* Once the connection has been created, we need to associate it with + * a stage so that the selected stage will handle the event on this connection. + * This function also allows the caller to decide which handler should + * be executed in the context of the stage, and also which specific events + * should be handed to the handler when running in this particular stage. + */ +int +rpcsvc_stage_conn_associate (rpcsvc_stage_t *stg, rpcsvc_conn_t *conn, + event_handler_t handler, void *data) +{ + int ret = -1; + + if ((!stg) || (!conn)) + return -1; + + conn->stage = stg; + conn->eventidx = event_register (stg->eventpool, conn->sockfd, handler, + data, 1, 0); + if (conn->eventidx == -1) + goto err; + + ret = 0; +err: + return ret; +} + + +/* Depending on the state we're in, return the size of the next read request. */ +size_t +rpcsvc_record_read_size (rpcsvc_record_state_t *rs) +{ + size_t toread = -1; + + if (!rs) + return -1; + + if (rpcsvc_record_readfraghdr (rs)) + toread = rs->remainingfraghdr; + else if (rpcsvc_record_readfrag (rs)) + toread = rs->remainingfrag; + else + toread = RPCSVC_CONN_READ; + + return toread; +} + + +uint32_t +rpcsvc_record_extract_fraghdr (char *fraghdr) +{ + uint32_t hdr = 0; + if (!fraghdr) + return 0; + + memcpy ((void *)&hdr, fraghdr, sizeof (hdr)); + + hdr = ntohl (hdr); + return hdr; +} + + +ssize_t +rpcsvc_record_read_complete_fraghdr (rpcsvc_record_state_t *rs,ssize_t dataread) +{ + uint32_t remhdr = 0; + char *fraghdrstart = NULL; + uint32_t fraghdr = 0; + + fraghdrstart = &rs->fragheader[0]; + remhdr = rs->remainingfraghdr; + fraghdr = rpcsvc_record_extract_fraghdr (fraghdrstart); + rs->fragsize = RPCSVC_FRAGSIZE (fraghdr); + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Received fragment size: %d", + rs->fragsize); + if (rpcsvc_record_vectored (rs)) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC header," + " remaining: %d", RPCSVC_BARERPC_MSGSZ); + rs->remainingfrag = RPCSVC_BARERPC_MSGSZ; + } else { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Regular RPC header," + " remaining: %d", rs->fragsize); + rs->remainingfrag = rs->fragsize; + } + + rs->state = RPCSVC_READ_FRAG; + dataread -= remhdr; + rs->remainingfraghdr -= remhdr; + rs->islastfrag = RPCSVC_LASTFRAG (fraghdr); + + return dataread; +} + + +ssize_t +rpcsvc_record_read_partial_fraghdr (rpcsvc_record_state_t *rs, ssize_t dataread) +{ + + /* In case we got less than even the remaining header size, + * we need to consume it all and wait for remaining frag hdr + * bytes to come in. + */ + rs->remainingfraghdr -= dataread; + rpcsvc_record_update_currenthdr (rs, dataread); + dataread = 0; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Fragment header remaining: %d", + rs->remainingfraghdr); + + return dataread; +} + + +ssize_t +rpcsvc_record_update_fraghdr (rpcsvc_record_state_t *rs, ssize_t dataread) +{ + if ((!rs) || (dataread <= 0)) + return -1; + + /* Why are we even here, we're not supposed to be in the fragment + * header processing state. + */ + if (!rpcsvc_record_readfraghdr(rs)) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "record state inconsistent" + ": request to update frag header when state is not" + "RPCSVC_READ_FRAGHDR"); + return -1; + } + + /* Again, if header has been read then the state member above should've + * been different, this is crazy. We should not be here. + */ + if (rs->remainingfraghdr == 0) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "record state inconsistent" + ": request to update frag header when frag header" + "remaining is 0."); + return -1; + } + + /* We've definitely got the full header now and may be even more. */ + if (dataread >= rs->remainingfraghdr) + dataread = rpcsvc_record_read_complete_fraghdr (rs, dataread); + else + dataread = rpcsvc_record_read_partial_fraghdr (rs, dataread); + + return dataread; +} + +ssize_t +rpcsvc_record_read_complete_frag (rpcsvc_record_state_t *rs, ssize_t dataread) +{ + uint32_t remfrag; + + /* Since the frag is now complete, change the state to the next + * one, i.e. to read the header of the next fragment. + */ + remfrag = rs->remainingfrag; + rs->state = RPCSVC_READ_FRAGHDR; + dataread -= remfrag; + + /* This will be 0 now. */ + rs->remainingfrag -= remfrag; + + /* Now that the fragment is complete, we must update the + * record size. Recall that fragsize was got from the frag + * header. + */ + rs->recordsize += rs->fragsize; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Fragment remaining: %d", + rs->remainingfrag); + + return dataread; +} + + +ssize_t +rpcsvc_record_read_partial_frag (rpcsvc_record_state_t *rs, ssize_t dataread) +{ + /* Just take whatever has come through the current network buffer. */ + rs->remainingfrag -= dataread; + + rpcsvc_record_update_currentfrag (rs, dataread); + /* Since we know we're consuming the whole buffer from dataread + * simply setting to 0 zero is fine. + */ + dataread = 0; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Fragment remaining: %d", + rs->remainingfrag); + return dataread; +} + + +ssize_t +rpcsvc_record_update_frag (rpcsvc_record_state_t *rs, ssize_t dataread) +{ + if ((!rs) || (dataread <= 0)) + return -1; + + if (!rpcsvc_record_readfrag (rs)) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "record state inconsistent" + ": request to update fragment when record state is not" + "RPCSVC_READ_FRAG."); + return -1; + } + + if (rs->remainingfrag == 0) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "record state inconsistent" + ": request to update fragment when there is no fragment" + " data remaining to be read."); + return -1; + } + + /* We've read in more data than the current fragment requires. */ + if (dataread >= rs->remainingfrag) + dataread = rpcsvc_record_read_complete_frag (rs, dataread); + else + dataread = rpcsvc_record_read_partial_frag (rs, dataread); + + return dataread; +} + + +/* This needs to change to returning errors, since + * we need to return RPC specific error messages when some + * of the pointers below are NULL. + */ +rpcsvc_actor_t * +rpcsvc_program_actor (rpcsvc_conn_t *conn, rpcsvc_request_t *req) +{ + rpcsvc_program_t *program = NULL; + int err = SYSTEM_ERR; + rpcsvc_actor_t *actor = NULL; + + if ((!conn) || (!req)) + goto err; + + program = (rpcsvc_program_t *)conn->program; + if (!program) + goto err; + + if (req->prognum != program->prognum) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC program not available"); + err = PROG_UNAVAIL; + goto err; + } + + if (!program->actors) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC System error"); + err = SYSTEM_ERR; + goto err; + } + + if (req->progver != program->progver) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC program version not" + " available"); + err = PROG_MISMATCH; + goto err; + } + + if ((req->procnum < 0) || (req->procnum >= program->numactors)) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC Program procedure not" + " available"); + err = PROC_UNAVAIL; + goto err; + } + + actor = &program->actors[req->procnum]; + if (!actor->actor) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC Program procedure not" + " available"); + err = PROC_UNAVAIL; + actor = NULL; + goto err; + } + + err = SUCCESS; + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Actor found: %s - %s", + program->progname, actor->procname); +err: + if (req) + req->rpc_err = err; + + return actor; +} + + +rpcsvc_txbuf_t * +rpcsvc_init_txbuf (rpcsvc_conn_t *conn, struct iovec msg, struct iobuf *iob, + struct iobref *iobref, int txflags) +{ + rpcsvc_txbuf_t *txbuf = NULL; + + txbuf = (rpcsvc_txbuf_t *) mem_get(conn->txpool); + if (!txbuf) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get txbuf"); + return NULL; + } + + memset (txbuf, 0, sizeof (*txbuf)); + INIT_LIST_HEAD (&txbuf->txlist); + txbuf->buf = msg; + + /* If it was required, this iob must've been ref'd already + * so I dont have to bother here. + */ + txbuf->iob = iob; + txbuf->iobref = iobref; + txbuf->offset = 0; + txbuf->txbehave = txflags; + + return txbuf; +} + + +int +rpcsvc_conn_append_txlist (rpcsvc_conn_t *conn, struct iovec msg, + struct iobuf *iob, int txflags) +{ + rpcsvc_txbuf_t *txbuf = NULL; + + if ((!conn) || (!msg.iov_base) || (!iob)) + return -1; + + txbuf = rpcsvc_init_txbuf (conn, msg, iob, NULL, txflags); + if (!txbuf) + return -1; + + list_add_tail (&txbuf->txlist, &conn->txbufs); + return 0; +} + + +void +rpcsvc_set_lastfrag (uint32_t *fragsize) { + (*fragsize) |= 0x80000000U; +} + +void +rpcsvc_set_frag_header_size (uint32_t size, char *haddr) +{ + size = htonl (size); + memcpy (haddr, &size, sizeof (size)); +} + +void +rpcsvc_set_last_frag_header_size (uint32_t size, char *haddr) +{ + rpcsvc_set_lastfrag (&size); + rpcsvc_set_frag_header_size (size, haddr); +} + + +/* Given the RPC reply structure and the payload handed by the RPC program, + * encode the RPC record header into the buffer pointed by recordstart. + */ +struct iovec +rpcsvc_record_build_header (char *recordstart, size_t rlen, + struct rpc_msg reply, size_t payload) +{ + struct iovec replyhdr; + struct iovec txrecord = {0, 0}; + size_t fraglen = 0; + int ret = -1; + + /* After leaving aside the 4 bytes for the fragment header, lets + * encode the RPC reply structure into the buffer given to us. + */ + ret = rpc_reply_to_xdr (&reply,(recordstart + RPCSVC_FRAGHDR_SIZE), + rlen, &replyhdr); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to create RPC reply"); + goto err; + } + + fraglen = payload + replyhdr.iov_len; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Reply fraglen %zu, payload: %zu, " + "rpc hdr: %ld", fraglen, payload, replyhdr.iov_len); + + /* Since we're not spreading RPC records over mutiple fragments + * we just set this fragment as the first and last fragment for this + * record. + */ + rpcsvc_set_last_frag_header_size (fraglen, recordstart); + + /* Even though the RPC record starts at recordstart+RPCSVC_FRAGHDR_SIZE + * we need to transmit the record with the fragment header, which starts + * at recordstart. + */ + txrecord.iov_base = recordstart; + + /* Remember, this is only the vec for the RPC header and does not + * include the payload above. We needed the payload only to calculate + * the size of the full fragment. This size is sent in the fragment + * header. + */ + txrecord.iov_len = RPCSVC_FRAGHDR_SIZE + replyhdr.iov_len; + +err: + return txrecord; +} + + +int +rpcsvc_conn_submit (rpcsvc_conn_t *conn, struct iovec hdr, + struct iobuf *hdriob, struct iovec msgvec, + struct iobuf *msgiob) +{ + int ret = -1; + + if ((!conn) || (!hdr.iov_base) || (!hdriob)) + return -1; + + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Tx Header: %ld, payload: %ld", + hdr.iov_len, msgvec.iov_len); + /* Now that we have both the RPC and Program buffers in xdr format + * lets hand it to the transmission layer. + */ + pthread_mutex_lock (&conn->connlock); + { + if (!rpcsvc_conn_check_active (conn)) { + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Connection inactive"); + goto unlock_err; + } + + ret = rpcsvc_conn_append_txlist (conn, hdr, hdriob, + RPCSVC_TXB_FIRST); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to append " + "header to transmission list"); + goto unlock_err; + } + + /* It is possible that this RPC reply is an error reply. In that + * case we might not have been handed a payload. + */ + ret = 0; + if (msgiob) + ret = rpcsvc_conn_append_txlist (conn, msgvec, msgiob, + RPCSVC_TXB_LAST); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to append" + " payload to transmission list"); + goto unlock_err; + } + } +unlock_err: + pthread_mutex_unlock (&conn->connlock); + + if (ret == -1) + goto err; + + /* Tell event pool, we're interested in poll_out to trigger flush + * of our tx buffers. + */ + conn->eventidx = event_select_on (conn->stage->eventpool, conn->sockfd, + conn->eventidx, -1, 1); + ret = 0; +err: + + return ret; +} + + +int +rpcsvc_fill_reply (rpcsvc_request_t *req, struct rpc_msg *reply) +{ + rpcsvc_program_t *prog = NULL; + if ((!req) || (!reply)) + return -1; + + prog = rpcsvc_request_program (req); + rpc_fill_empty_reply (reply, req->xid); + + if (req->rpc_stat == MSG_DENIED) + rpc_fill_denied_reply (reply, req->rpc_err, req->auth_err); + else if (req->rpc_stat == MSG_ACCEPTED) + rpc_fill_accepted_reply (reply, req->rpc_err, prog->proglowvers, + prog->proghighvers, req->verf.flavour, + req->verf.datalen, + req->verf.authdata); + else + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Invalid rpc_stat value"); + + return 0; +} + + +/* Given a request and the reply payload, build a reply and encodes the reply + * into a record header. This record header is encoded into the vector pointed + * to be recbuf. + * msgvec is the buffer that points to the payload of the RPC program. + * This buffer can be NULL, if an RPC error reply is being constructed. + * The only reason it is needed here is that in case the buffer is provided, + * we should account for the length of that buffer in the RPC fragment header. + */ +struct iobuf * +rpcsvc_record_build_record (rpcsvc_request_t *req, size_t payload, + struct iovec *recbuf) +{ + struct rpc_msg reply; + struct iobuf *replyiob = NULL; + char *record = NULL; + struct iovec recordhdr = {0, }; + size_t pagesize = 0; + rpcsvc_conn_t *conn = NULL; + rpcsvc_t *svc = NULL; + + if ((!req) || (!req->conn) || (!recbuf)) + return NULL; + + /* First, try to get a pointer into the buffer which the RPC + * layer can use. + */ + conn = req->conn; + svc = rpcsvc_conn_rpcsvc (conn); + replyiob = iobuf_get (svc->ctx->iobuf_pool); + pagesize = iobpool_pagesize ((struct iobuf_pool *)svc->ctx->iobuf_pool); + if (!replyiob) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to get iobuf"); + goto err_exit; + } + + record = iobuf_ptr (replyiob); /* Now we have it. */ + + /* Fill the rpc structure and XDR it into the buffer got above. */ + rpcsvc_fill_reply (req, &reply); + recordhdr = rpcsvc_record_build_header (record, pagesize, reply, + payload); + if (!recordhdr.iov_base) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to build record " + " header"); + iobuf_unref (replyiob); + replyiob = NULL; + recbuf->iov_base = NULL; + goto err_exit; + } + + recbuf->iov_base = recordhdr.iov_base; + recbuf->iov_len = recordhdr.iov_len; +err_exit: + return replyiob; +} + + +/* + * The function to submit a program message to the RPC service. + * This message is added to the transmission queue of the + * conn. + * + * Program callers are not expected to use the msgvec->iov_base + * address for anything else. + * Nor are they expected to free it once this function returns. + * Once the transmission of the buffer is completed by the RPC service, + * the memory area as referenced through @msg will be unrefed. + * If a higher layer does not want anything to do with this iobuf + * after this function returns, it should call unref on it. For keeping + * it around till the transmission is actually complete, rpcsvc also refs it. + * * + * If this function returns an error by returning -1, the + * higher layer programs should assume that a disconnection happened + * and should know that the conn memory area as well as the req structure + * has been freed internally. + * + * For now, this function assumes that a submit is always called + * to send a new record. Later, if there is a situation where different + * buffers for the same record come from different sources, then we'll + * need to change this code to account for multiple submit calls adding + * the buffers into a single record. + */ + +int +rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec msgvec, + struct iobuf *msg) +{ + int ret = -1; + struct iobuf *replyiob = NULL; + struct iovec recordhdr = {0, }; + rpcsvc_conn_t *conn = NULL; + + if ((!req) || (!req->conn)) + return -1; + + conn = req->conn; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Tx message: %ld", msgvec.iov_len); + /* Build the buffer containing the encoded RPC reply. */ + replyiob = rpcsvc_record_build_record (req, msgvec.iov_len, &recordhdr); + if (!replyiob) { + gf_log (GF_RPCSVC, GF_LOG_ERROR,"Reply record creation failed"); + goto disconnect_exit; + } + + /* Must ref the iobuf got from higher layer so that the higher layer + * can rest assured that it can unref it and leave the final freeing + * of the buffer to us. Note msg can be NULL if an RPC-only message + * was being sent. Happens when an RPC error reply is being sent. + */ + if (msg) + iobuf_ref (msg); + ret = rpcsvc_conn_submit (conn, recordhdr, replyiob, msgvec, msg); + mem_put (conn->rxpool, req); + + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to submit message"); + iobuf_unref (replyiob); + } + +disconnect_exit: + /* Note that a unref is called everytime a reply is sent. This is in + * response to the ref that is performed on the conn when a request is + * handed to the RPC program. + * + * The catch, however, is that if the reply is an rpc error, we must + * not unref. This is because the ref only contains + * references for the actors to which the request was handed plus one + * reference maintained by the RPC layer. By unrefing for a case where + * no actor was called, we will be losing the ref held for the RPC + * layer. + */ + if ((rpcsvc_request_accepted (req)) && + (rpcsvc_request_accepted_success (req))) + rpcsvc_conn_unref (conn); + + return ret; +} + + +int +rpcsvc_request_attach_vector (rpcsvc_request_t *req, struct iovec msgvec, + struct iobuf *iob, struct iobref *iobref, + int finalvector) +{ + rpcsvc_txbuf_t *txb = NULL; + int txflags = 0; + + if ((!req) || (!msgvec.iov_base)) + return -1; + + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Tx Vector: %ld", msgvec.iov_len); + if (finalvector) + txflags |= RPCSVC_TXB_LAST; + /* We only let the user decide whether this is the last vector for the + * record, since the first vector is always the RPC header. + */ + txb = rpcsvc_init_txbuf (req->conn, msgvec, iob, iobref, txflags); + if (!txb) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Could not init tx buf"); + return -1; + } + + req->payloadsize += msgvec.iov_len; + if (iob) + iobuf_ref (iob); + if (iobref) + iobref_ref (iobref); + list_add_tail (&txb->txlist, &req->txlist); + + return 0; +} + + +int +rpcsvc_submit_vectors (rpcsvc_request_t *req) +{ + int ret = -1; + struct iobuf *replyiob = NULL; + struct iovec recordhdr = {0, }; + rpcsvc_txbuf_t *rpctxb = NULL; + + if ((!req) || (!req->conn)) + return -1; + + /* Build the buffer containing the encoded RPC reply. */ + replyiob = rpcsvc_record_build_record (req, req->payloadsize, + &recordhdr); + if (!replyiob) { + gf_log (GF_RPCSVC, GF_LOG_ERROR,"Reply record creation failed"); + goto disconnect_exit; + } + + rpctxb = rpcsvc_init_txbuf (req->conn, recordhdr, replyiob, NULL, + RPCSVC_TXB_FIRST); + if (!rpctxb) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to create tx buf"); + goto disconnect_exit; + } + + pthread_mutex_lock (&req->conn->connlock); + { + list_splice_init (&req->txlist, &req->conn->txbufs); + list_add (&rpctxb->txlist, &req->conn->txbufs); + } + pthread_mutex_unlock (&req->conn->connlock); + + ret = 0; + req->conn->eventidx = event_select_on (req->conn->stage->eventpool, + req->conn->sockfd, + req->conn->eventidx, -1, 1); +disconnect_exit: + /* Note that a unref is called everytime a reply is sent. This is in + * response to the ref that is performed on the conn when a request is + * handed to the RPC program. + */ + rpcsvc_conn_unref (req->conn); + if (ret == -1) + iobuf_unref (replyiob); + + mem_put (req->conn->rxpool, req); + return ret; +} + + +int +rpcsvc_error_reply (rpcsvc_request_t *req) +{ + struct iovec dummyvec = {0, }; + + if (!req) + return -1; + + /* At this point the req should already have been filled with the + * appropriate RPC error numbers. + */ + return rpcsvc_submit_generic (req, dummyvec, NULL); +} + + +rpcsvc_request_t * +rpcsvc_request_init (rpcsvc_conn_t *conn, struct rpc_msg *callmsg, + struct iovec progmsg, rpcsvc_request_t *req) +{ + if ((!conn) || (!callmsg)|| (!req)) + return NULL; + + + /* We start a RPC request as always denied. */ + req->rpc_stat = MSG_DENIED; + req->xid = rpc_call_xid (callmsg); + req->prognum = rpc_call_program (callmsg); + req->progver = rpc_call_progver (callmsg); + req->procnum = rpc_call_progproc (callmsg); + req->conn = conn; + req->msg = progmsg; + req->recordiob = conn->rstate.activeiob; + INIT_LIST_HEAD (&req->txlist); + req->payloadsize = 0; + + /* By this time, the data bytes for the auth scheme would have already + * 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); + return req; +} + + +rpcsvc_request_t * +rpcsvc_request_create (rpcsvc_conn_t *conn) +{ + char *msgbuf = NULL; + struct rpc_msg rpcmsg; + struct iovec progmsg; /* RPC Program payload */ + rpcsvc_request_t *req = NULL; + int ret = -1; + + if (!conn) + return NULL; + + /* We need to allocate the request before actually calling + * rpcsvc_request_init on the request so that we, can fill the auth + * data directly into the request structure from the message iobuf. + * This avoids a need to keep a temp buffer into which the auth data + * would've been copied otherwise. + */ + rpcsvc_alloc_request (conn, req); + if (!req) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed to alloc request"); + goto err; + } + + msgbuf = iobuf_ptr (conn->rstate.activeiob); + ret = xdr_to_rpc_call (msgbuf, conn->rstate.recordsize, &rpcmsg, + &progmsg, req->cred.authdata,req->verf.authdata); + + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC call decoding failed"); + rpcsvc_request_seterr (req, GARBAGE_ARGS); + goto err; + } + + ret = -1; + rpcsvc_request_init (conn, &rpcmsg, progmsg, req); + + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "RPC XID: %lx, Ver: %ld, Program: %ld," + " ProgVers: %ld, Proc: %ld", rpc_call_xid (&rpcmsg), + rpc_call_rpcvers (&rpcmsg), rpc_call_program (&rpcmsg), + rpc_call_progver (&rpcmsg), rpc_call_progproc (&rpcmsg)); + + if (rpc_call_rpcvers (&rpcmsg) != 2) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "RPC version not supported"); + rpcsvc_request_seterr (req, RPC_MISMATCH); + goto err; + } + + ret = rpcsvc_authenticate (req); + if (ret == RPCSVC_AUTH_REJECT) { + /* No need to set auth_err, that is the responsibility of + * the authentication handler since only that know what exact + * error happened. + */ + rpcsvc_request_seterr (req, AUTH_ERROR); + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Failed authentication"); + ret = -1; + goto err; + } + + + /* If the error is not RPC_MISMATCH, we consider the call as accepted + * since we are not handling authentication failures for now. + */ + req->rpc_stat = MSG_ACCEPTED; + ret = 0; +err: + if (ret == -1) { + ret = rpcsvc_error_reply (req); + req = NULL; + } + + return req; +} + + +int +rpcsvc_handle_rpc_call (rpcsvc_conn_t *conn) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_request_t *req = NULL; + int ret = -1; + + if (!conn) + return -1; + + req = rpcsvc_request_create (conn); + if (!req) + goto err; + + if (!rpcsvc_request_accepted (req)) + goto err_reply; + + actor = rpcsvc_program_actor (conn, req); + if (!actor) + goto err_reply; + + if ((actor) && (actor->actor)) { + rpcsvc_conn_ref (conn); + ret = actor->actor (req); + } + +err_reply: + if (ret == RPCSVC_ACTOR_ERROR) + ret = rpcsvc_error_reply (req); + + /* No need to propagate error beyond this function since the reply + * has now been queued. */ + ret = 0; +err: + return ret; +} + +#define rpc_call_cred_addr(rs) (iobuf_ptr ((rs)->activeiob) + RPCSVC_BARERPC_MSGSZ - 4) + +uint32_t +rpcsvc_call_credlen (rpcsvc_record_state_t *rs) +{ + char *credaddr = NULL; + uint32_t credlen_nw = 0; + uint32_t credlen_host = 0; + + /* Position to the start of the credential length field. */ + credaddr = rpc_call_cred_addr (rs); + credlen_nw = *(uint32_t *)credaddr; + credlen_host = ntohl (credlen_nw); + + return credlen_host; +} + +uint32_t +rpcsvc_call_verflen (rpcsvc_record_state_t *rs) +{ + char *verfaddr = NULL; + uint32_t verflen_nw = 0; + uint32_t verflen_host = 0; + uint32_t credlen = 0; + + /* Position to the start of the verifier length field. */ + credlen = rpcsvc_call_credlen (rs); + verfaddr = (rpc_call_cred_addr (rs) + 4 + credlen); + verflen_nw = *(uint32_t *)verfaddr; + verflen_host = ntohl (verflen_nw); + + return verflen_host; +} + + +void +rpcsvc_update_vectored_verf (rpcsvc_record_state_t *rs) +{ + if (!rs) + return; + + rs->recordsize += rpcsvc_call_verflen (rs); + return; +} + + +void +rpcsvc_handle_vectored_prep_rpc_call (rpcsvc_conn_t *conn) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_request_t *req = NULL; + rpcsvc_record_state_t *rs = NULL; + rpcsvc_t *svc = NULL; + int ret = -1; + ssize_t remfrag = RPCSVC_ACTOR_ERROR; + int newbuf = 0; + + if (!conn) + return; + + rs = &conn->rstate; + + /* In case one of the steps below fails, we need to make sure that the + * remaining frag in the kernel's buffers are read-out so that the + * requests that follow can be served. + */ + rs->remainingfrag = rs->fragsize - rs->recordsize; + rs->vecstate = RPCSVC_VECTOR_IGNORE; + req = rpcsvc_request_create (conn); + svc = rpcsvc_conn_rpcsvc (conn); + if (!req) + goto err; + + if (!rpcsvc_request_accepted (req)) + goto err_reply; + + actor = rpcsvc_program_actor (conn, req); + if (!actor) + goto err_reply; + + if (!actor->vector_sizer) { + ret = -1; + rpcsvc_request_seterr (req, PROC_UNAVAIL); + goto err_reply; + } + + rpcsvc_conn_ref (conn); + ret = actor->vector_sizer (req, &remfrag, &newbuf); + rpcsvc_conn_unref (conn); + + if (ret == RPCSVC_ACTOR_ERROR) { + ret = -1; + rpcsvc_request_seterr (req, SYSTEM_ERR); + goto err_reply; + } + + rs->remainingfrag = remfrag; + rs->vecstate = RPCSVC_VECTOR_READPROCHDR; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC proc header remaining:" + " %d", rs->remainingfrag); + conn->vectoredreq = req; + + /* Store the reference to the current frag pointer. This is where the + * proc header will be read into. + */ + req->msg.iov_base = rs->fragcurrent; + req->msg.iov_len = rs->remainingfrag; + ret = 0; + +err_reply: + if (ret == -1) + ret = rpcsvc_error_reply (req); + + /* No need to propagate error beyond this function since the reply + * has now been queued. */ + ret = 0; +err: + return; +} + + +void +rpcsvc_update_vectored_verfsz (rpcsvc_conn_t *conn) +{ + rpcsvc_record_state_t *rs = NULL; + uint32_t verflen = 0; + + if (!conn) + return; + + rs = &conn->rstate; + + verflen = rpcsvc_call_verflen (rs); + rs->recordsize += 8; + if (verflen > 0) { + rs->remainingfrag = verflen; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC verf remaining: " + " %d", rs->remainingfrag); + rs->vecstate = RPCSVC_VECTOR_READVERF; + } else { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC preparing call"); + rpcsvc_handle_vectored_prep_rpc_call (conn); + } + + return; +} + + +void +rpcsvc_update_vectored_cred (rpcsvc_record_state_t *rs) +{ + uint32_t credlen = 0; + + if (!rs) + return; + + credlen = rpcsvc_call_credlen (rs); + /* Update remainingfrag to read the 8 bytes needed for + * reading verf flavour and verf len. + */ + rs->remainingfrag = (2 * sizeof (uint32_t)); + rs->vecstate = RPCSVC_VECTOR_READVERFSZ; + rs->recordsize += credlen; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC verfsz remaining: %d", + rs->remainingfrag); + + return; +} + +void +rpcsvc_update_vectored_barerpc (rpcsvc_record_state_t *rs) +{ + uint32_t credlen = 0; + + if (!rs) + return; + + credlen = rpcsvc_call_credlen (rs); + rs->recordsize = RPCSVC_BARERPC_MSGSZ; + if (credlen == 0) { + rs->remainingfrag = 8; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC verfsz remaining" + ": %d", rs->remainingfrag); + rs->vecstate = RPCSVC_VECTOR_READVERFSZ; + } else { + rs->remainingfrag = credlen; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC cred remaining: " + "%d", rs->remainingfrag); + rs->vecstate = RPCSVC_VECTOR_READCRED; + } + + return; +} + + +void +rpcsvc_handle_vectored_rpc_call (rpcsvc_conn_t *conn) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_request_t *req = NULL; + rpcsvc_record_state_t *rs = NULL; + rpcsvc_t *svc = NULL; + int ret = -1; + ssize_t remfrag = -1; + int newbuf = 0; + + if (!conn) + return; + + rs = &conn->rstate; + + req = conn->vectoredreq; + svc = rpcsvc_conn_rpcsvc (conn); + + if (!req) + goto err; + + actor = rpcsvc_program_actor (conn, req); + if (!actor) + goto err_reply; + + if (!actor->vector_sizer) { + ret = -1; + rpcsvc_request_seterr (req, PROC_UNAVAIL); + goto err_reply; + } + + req->msg.iov_len = (unsigned long)((long)rs->fragcurrent - (long)req->msg.iov_base); + rpcsvc_conn_ref (conn); + ret = actor->vector_sizer (req, &remfrag, &newbuf); + rpcsvc_conn_unref (conn); + if (ret == RPCSVC_ACTOR_ERROR) { + ret = -1; + rpcsvc_request_seterr (req, SYSTEM_ERR); + goto err_reply; + } + + if (newbuf) { + rs->vectoriob = iobuf_get (svc->ctx->iobuf_pool); + rs->fragcurrent = iobuf_ptr (rs->vectoriob); + rs->vecstate = RPCSVC_VECTOR_READVEC; + rs->remainingfrag = remfrag; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC buf remaining:" + " %d", rs->remainingfrag); + } else { + rs->remainingfrag = remfrag; + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC proc remaining:" + " %d", rs->remainingfrag); + } + + ret = 0; +err_reply: + if (ret == -1) + ret = rpcsvc_error_reply (req); + + /* No need to propagate error beyond this function since the reply + * has now been queued. */ + ret = 0; +err: + return; +} + + + +void +rpcsvc_record_vectored_call_actor (rpcsvc_conn_t *conn) +{ + rpcsvc_actor_t *actor = NULL; + rpcsvc_request_t *req = NULL; + rpcsvc_record_state_t *rs = NULL; + rpcsvc_t *svc = NULL; + int ret = -1; + + if (!conn) + return; + + rs = &conn->rstate; + req = conn->vectoredreq; + svc = rpcsvc_conn_rpcsvc (conn); + + if (!req) + goto err; + + actor = rpcsvc_program_actor (conn, req); + if (!actor) + goto err_reply; + + if (actor->vector_actor) { + rpcsvc_conn_ref (conn); + ret = actor->vector_actor (req, rs->vectoriob); + } else { + rpcsvc_request_seterr (req, PROC_UNAVAIL); + gf_log (GF_RPCSVC, GF_LOG_ERROR, "No vectored handler present"); + ret = RPCSVC_ACTOR_ERROR; + } + +err_reply: + if (ret == RPCSVC_ACTOR_ERROR) + ret = rpcsvc_error_reply (req); + + /* No need to propagate error beyond this function since the reply + * has now been queued. */ + ret = 0; +err: + return; +} + + + +ssize_t +rpcsvc_update_vectored_state (rpcsvc_conn_t *conn) +{ + rpcsvc_record_state_t *rs = NULL; + rpcsvc_t *svc = NULL; + + if (!conn) + return 0; + + /* At this point, we can be confident that the activeiob contains + * exactly the first RPCSVC_BARERPC_MSGSZ bytes needed in order to + * determine the program and actor. So the next state will be + * to read the credentials. + * + * But first, try to determine how many more bytes do we need from the + * network to complete the RPC message including the credentials. + */ + + rs = &conn->rstate; + if (rpcsvc_record_vectored_baremsg (rs)) + rpcsvc_update_vectored_barerpc (rs); + else if (rpcsvc_record_vectored_cred (rs)) + rpcsvc_update_vectored_cred (rs); + else if (rpcsvc_record_vectored_verfsz (rs)) + rpcsvc_update_vectored_verfsz (conn); + else if (rpcsvc_record_vectored_verfread (rs)) { + rpcsvc_update_vectored_verf (rs); + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC preparing call"); + rpcsvc_handle_vectored_prep_rpc_call (conn); + } else if (rpcsvc_record_vectored_readprochdr (rs)) + rpcsvc_handle_vectored_rpc_call (conn); + else if (rpcsvc_record_vectored_ignore (rs)) { + svc = rpcsvc_conn_rpcsvc (conn); + rpcsvc_record_init (rs, svc->ctx->iobuf_pool); + } else if (rpcsvc_record_vectored_readvec (rs)) { + svc = rpcsvc_conn_rpcsvc (conn); + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored RPC vector read"); + rpcsvc_record_vectored_call_actor (conn); + rpcsvc_record_init (rs, svc->ctx->iobuf_pool); + } + + return 0; +} + + +ssize_t +rpcsvc_record_read_partial_frag (rpcsvc_record_state_t *rs, ssize_t dataread); + +ssize_t +rpcsvc_update_vectored_msg (rpcsvc_conn_t *conn, ssize_t dataread) +{ + + if (!conn) + return dataread; + + /* find out how much of the bare msg is pending and set that up to be + * read into the updated fragcurrent along with the updated size into + * remainingfrag. + */ + + + /* Incidently, the logic needed here is similar to a regular partial + * fragment read since we've already set the remainingfrag member in + * rstate to be RPCSVC_BARERPC_MSGSZ for the purpose of a vectored + * fragment. + */ + return rpcsvc_record_read_partial_frag (&conn->rstate, dataread); +} + +/* FIX: As a first version of vectored reading, I am assuming dataread will + * always be equal to RPCSVC_BARERPC_MSGSZ for the sake of simplicity on the + * belief that we're never actually reading more bytes than needed in each + * poll_in. + */ +ssize_t +rpcsvc_handle_vectored_frag (rpcsvc_conn_t *conn, ssize_t dataread) +{ + if (!conn) + return dataread; + + /* At this point we can be confident that only the frag size has been + * read from the network. Now it is up to us to have the remaining RPC + * fields given to us here. + */ + + /* Since the poll_in handler uses the remainingfrag field to determine + * how much to read from the network, we'll hack this scheme to tell + * the poll_in to read at most RPCSVC_BARERPC_MSGSZ bytes. This is done + * to, as a first step, identify which (program, actor) we need to call. + */ + + dataread = rpcsvc_update_vectored_msg (conn, dataread); + + if (conn->rstate.remainingfrag == 0) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored frag complete"); + dataread = rpcsvc_update_vectored_state (conn); + } + + return dataread; +} + + +int +rpcsvc_record_update_state (rpcsvc_conn_t *conn, ssize_t dataread) +{ + rpcsvc_record_state_t *rs = NULL; + rpcsvc_t *svc = NULL; + + if (!conn) + return -1; + + rs = &conn->rstate; + /* At entry into this function, fragcurrent will be pointing to the\ + * start of the area into which dataread number of bytes were read. + */ + + if (rpcsvc_record_readfraghdr(rs)) + dataread = rpcsvc_record_update_fraghdr (rs, dataread); + + if (rpcsvc_record_readfrag(rs)) { + /* The minimum needed for triggering the vectored handler is + * the frag size field. The fragsize member remains set to this + * size till this request is completely extracted from the + * network. Once all the data has been read from the network, + * the request structure would've been created. The point being + * that even if it takes multiple calls to network IO for + * getting the vectored fragment, we can continue to use this + * condition as the flag to tell us that this is a vectored + * fragment. + */ + if ((dataread > 0) && (rpcsvc_record_vectored (rs))) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Vectored frag"); + dataread = rpcsvc_handle_vectored_frag (conn, dataread); + } else if (dataread > 0) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Regular frag"); + dataread = rpcsvc_record_update_frag (rs, dataread); + } + } + + /* This should not happen. We are never reading more than the current + * fragment needs. Something is seriously wrong. + */ + if (dataread > 0) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Data Left: %ld", dataread); + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Unwanted data read from " + " connection."); + } + + /* If we're now supposed to wait for a new fragment header and if the + * fragment that we just completed in the previous call to + * rpcsvc_record_update_frag was the last fragment for the current + * RPC record, then, it is time to perform the translation from + * XDR formatted buffer in activeiob followed by the upcall to the + * protocol actor. + */ + if ((rpcsvc_record_readfraghdr(rs)) && (rs->islastfrag)) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Full Record Received."); + rpcsvc_handle_rpc_call (conn); + svc = rpcsvc_conn_rpcsvc (conn); + rpcsvc_record_init (rs, svc->ctx->iobuf_pool); + } + + return 0; +} + + +char * +rpcsvc_record_read_addr (rpcsvc_record_state_t *rs) +{ + + if (rpcsvc_record_readfraghdr (rs)) + return rpcsvc_record_currenthdr_addr (rs); + else if (rpcsvc_record_readfrag (rs)) + return rpcsvc_record_currentfrag_addr (rs); + + return NULL; +} + + +int +rpcsvc_conn_data_poll_in (rpcsvc_conn_t *conn) +{ + ssize_t dataread = -1; + size_t readsize = 0; + char *readaddr = NULL; + int ret = -1; + + readaddr = rpcsvc_record_read_addr (&conn->rstate); + if (!readaddr) + goto err; + + readsize = rpcsvc_record_read_size (&conn->rstate); + if (readsize == -1) + goto err; + + dataread = rpcsvc_socket_read (conn->sockfd, readaddr, readsize); + gf_log (GF_RPCSVC, GF_LOG_TRACE, "conn: 0x%lx, readsize: %ld, dataread: %ld", + (long)conn, readsize, dataread); + + if (dataread > 0) + ret = rpcsvc_record_update_state (conn, dataread); + +err: + return ret; +} + + +int +rpcsvc_conn_data_poll_err (rpcsvc_conn_t *conn) +{ + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Received error event"); + rpcsvc_conn_deinit (conn); + return 0; +} + + +int +__rpcsvc_conn_data_poll_out (rpcsvc_conn_t *conn) +{ + rpcsvc_txbuf_t *txbuf = NULL; + rpcsvc_txbuf_t *tmp = NULL; + ssize_t written = -1; + char *writeaddr = NULL; + size_t writesize = -1; + + if (!conn) + return -1; + + /* Attempt transmission of each of the pending buffers */ + list_for_each_entry_safe (txbuf, tmp, &conn->txbufs, txlist) { +tx_remaining: + writeaddr = (char *)(txbuf->buf.iov_base + txbuf->offset); + writesize = (txbuf->buf.iov_len - txbuf->offset); + + if (txbuf->txbehave & RPCSVC_TXB_FIRST) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "First Tx Buf"); + rpcsvc_socket_block_tx (conn->sockfd); + } + + written = rpcsvc_socket_write (conn->sockfd, writeaddr, + writesize); + if (txbuf->txbehave & RPCSVC_TXB_LAST) { + gf_log (GF_RPCSVC, GF_LOG_TRACE, "Last Tx Buf"); + rpcsvc_socket_unblock_tx (conn->sockfd); + } + gf_log (GF_RPCSVC, GF_LOG_TRACE, "conn: 0x%lx, Tx request: %ld," + " Tx sent: %ld", (long)conn, writesize, written); + + /* There was an error transmitting this buffer */ + if (written == -1) + break; + + if (written >= 0) + txbuf->offset += written; + + /* If the current buffer has been completely transmitted, + * delete it from the list and move on to the next buffer. + */ + if (txbuf->offset == txbuf->buf.iov_len) { + /* It doesnt matter who ref'ed this iobuf, rpcsvc for + * its own header or a RPC program. + */ + if (txbuf->iob) + iobuf_unref (txbuf->iob); + if (txbuf->iobref) + iobref_unref (txbuf->iobref); + + list_del (&txbuf->txlist); + mem_put (conn->txpool, txbuf); + } else + /* If the current buffer is incompletely tx'd, do not + * go to the head of the loop, since that moves us to + * the next buffer. + */ + goto tx_remaining; + } + + /* If we've broken out of the loop above then we must unblock + * the transmission now. + */ + rpcsvc_socket_unblock_tx (conn->sockfd); + if (list_empty (&conn->txbufs)) + conn->eventidx = event_select_on (conn->stage->eventpool, + conn->sockfd, conn->eventidx, + -1, 0); + + return 0; +} + + +int +rpcsvc_conn_data_poll_out (rpcsvc_conn_t *conn) +{ + if (!conn) + return -1; + + + pthread_mutex_lock (&conn->connlock); + { + __rpcsvc_conn_data_poll_out (conn); + } + pthread_mutex_unlock (&conn->connlock); + + return 0; +} + + +int +rpcsvc_conn_data_handler (int fd, int idx, void *data, int poll_in, int poll_out + , int poll_err) +{ + rpcsvc_conn_t *conn = NULL; + int ret = 0; + + if (!data) + return 0; + + conn = (rpcsvc_conn_t *)data; + + if (poll_out) + ret = rpcsvc_conn_data_poll_out (conn); + + if (poll_err) { + ret = rpcsvc_conn_data_poll_err (conn); + return 0; + } + + if (poll_in) { + ret = 0; + ret = rpcsvc_conn_data_poll_in (conn); + } + + if (ret == -1) + rpcsvc_conn_data_poll_err (conn); + + return 0; +} + + +int +rpcsvc_conn_listening_handler (int fd, int idx, void *data, int poll_in, + int poll_out, int poll_err) +{ + rpcsvc_conn_t *newconn = NULL; + rpcsvc_stage_t *selectedstage = NULL; + int ret = -1; + rpcsvc_conn_t *conn = NULL; + rpcsvc_program_t *prog = NULL; + rpcsvc_t *svc = NULL; + + if (!poll_in) + return 0; + + conn = (rpcsvc_conn_t *)data; + prog = (rpcsvc_program_t *)conn->program; + svc = rpcsvc_conn_rpcsvc (conn); + newconn = rpcsvc_conn_accept_init (svc, fd, prog); + if (!newconn) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "failed to accept connection"); + goto err; + } + + selectedstage = rpcsvc_select_stage (svc); + if (!selectedstage) + goto close_err; + + /* Now that we've accepted the connection, we need to associate + * its events to a stage. + */ + ret = rpcsvc_stage_conn_associate (selectedstage, newconn, + rpcsvc_conn_data_handler, newconn); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "could not associated stage " + " with new connection"); + goto close_err; + } + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "New Connection: Program %s, Num: %d," + " Ver: %d, Port: %d", prog->progname, prog->prognum, + prog->progver, prog->progport); + ret = 0; +close_err: + if (ret == -1) + rpcsvc_conn_unref (newconn); + +err: + return ret; +} + + +/* Register the program with the local portmapper service. */ +int +rpcsvc_program_register_portmap (rpcsvc_program_t *newprog) +{ + if (!newprog) + return -1; + + if (!(pmap_set(newprog->prognum, newprog->progver, IPPROTO_TCP, + newprog->progport))) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Could not register with" + " portmap"); + return -1; + } + + return 0; +} + + +int +rpcsvc_program_unregister_portmap (rpcsvc_program_t *prog) +{ + if (!prog) + return -1; + + if (!(pmap_unset(prog->prognum, prog->progver))) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Could not unregister with" + " portmap"); + return -1; + } + + return 0; +} + + +int +rpcsvc_stage_program_register (rpcsvc_stage_t *stg, rpcsvc_program_t *newprog) +{ + rpcsvc_conn_t *newconn = NULL; + rpcsvc_t *svc = NULL; + + if ((!stg) || (!newprog)) + return -1; + + svc = rpcsvc_stage_service (stg); + /* Create a listening socket */ + newconn = rpcsvc_conn_listen_init (svc, newprog); + if (!newconn) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "could not create listening" + " connection"); + return -1; + } + + if ((rpcsvc_stage_conn_associate (stg, newconn, + rpcsvc_conn_listening_handler, + newconn)) == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR,"could not associate stage with" + " listening connection"); + return -1; + } + + return 0; +} + + +int +rpcsvc_program_register (rpcsvc_t *svc, rpcsvc_program_t program) +{ + rpcsvc_program_t *newprog = NULL; + rpcsvc_stage_t *selectedstage = NULL; + int ret = -1; + + if (!svc) + return -1; + + newprog = CALLOC (1, sizeof(*newprog)); + if (!newprog) + return -1; + + if (!program.actors) + goto free_prog; + + memcpy (newprog, &program, sizeof (program)); + selectedstage = rpcsvc_select_stage (svc); + + ret = rpcsvc_stage_program_register (selectedstage, newprog); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "stage registration of program" + " failed"); + goto free_prog; + } + + ret = rpcsvc_program_register_portmap (newprog); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "portmap registration of" + " program failed"); + goto free_prog; + } + + ret = 0; + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "New program registered: %s, Num: %d," + " Ver: %d, Port: %d", newprog->progname, newprog->prognum, + newprog->progver, newprog->progport); + +free_prog: + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "Program registration failed:" + " %s, Num: %d, Ver: %d, Port: %d", newprog->progname, + newprog->prognum, newprog->progver, newprog->progport); + FREE (newprog); + } + + return ret; +} + +/* The only difference between the generic submit and this one is that the + * generic submit is also used for submitting RPC error replies in where there + * are no payloads so the msgvec and msgbuf can be NULL. + * Since RPC programs should be using this function along with their payloads + * we must perform NULL checks before calling the generic submit. + */ +int +rpcsvc_submit_message (rpcsvc_request_t *req, struct iovec msgvec, + struct iobuf *msg) +{ + if ((!req) || (!req->conn) || (!msg) || (!msgvec.iov_base)) + return -1; + + return rpcsvc_submit_generic (req, msgvec, msg); +} + + +int +rpcsvc_program_unregister (rpcsvc_t *svc, rpcsvc_program_t prog) +{ + int ret = -1; + + if (!svc) + return -1; + + /* TODO: De-init the listening connection for this program. */ + ret = rpcsvc_program_unregister_portmap (&prog); + if (ret == -1) { + gf_log (GF_RPCSVC, GF_LOG_ERROR, "portmap unregistration of" + " program failed"); + goto err; + } + + ret = 0; + gf_log (GF_RPCSVC, GF_LOG_DEBUG, "Program unregistered: %s, Num: %d," + " Ver: %d, Port: %d", prog.progname, prog.prognum, + prog.progver, prog.progport); + +err: + 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); + + return ret; +} + + +int +rpcsvc_conn_peername (rpcsvc_conn_t *conn, char *hostname, int hostlen) +{ + if (!conn) + return -1; + + return rpcsvc_socket_peername (conn->sockfd, hostname, hostlen); +} + + +int +rpcsvc_conn_peeraddr (rpcsvc_conn_t *conn, char *addrstr, int addrlen, + struct sockaddr *sa, socklen_t sasize) +{ + if (!conn) + return -1; + + return rpcsvc_socket_peeraddr (conn->sockfd, addrstr, addrlen, sa, + sasize); +} + diff --git a/xlators/nfs/lib/src/rpcsvc.h b/xlators/nfs/lib/src/rpcsvc.h new file mode 100644 index 00000000000..2746288f82b --- /dev/null +++ b/xlators/nfs/lib/src/rpcsvc.h @@ -0,0 +1,715 @@ +/* + Copyright (c) 2010 Gluster, Inc. + 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 + . +*/ + +#ifndef _RPCSVC_H +#define _RPCSVC_H + + +#ifndef _CONFIG_H +#define _CONFIG_H +#include "config.h" +#endif + +#include "event.h" +#include "transport.h" +#include "logging.h" +#include "dict.h" +#include "mem-pool.h" +#include "list.h" +#include "iobuf.h" +#include "xdr-rpc.h" +#include "glusterfs.h" + +#include +#include + +#define GF_RPCSVC "rpc-service" +#define RPCSVC_THREAD_STACK_SIZE ((size_t)(1024 * GF_UNIT_KB)) + +#define RPCSVC_DEFAULT_MEMFACTOR 15 +#define RPCSVC_EVENTPOOL_SIZE_MULT 1024 +#define RPCSVC_POOLCOUNT_MULT 35 +#define RPCSVC_CONN_READ (128 * GF_UNIT_KB) +#define RPCSVC_PAGE_SIZE (128 * GF_UNIT_KB) + +/* Defines for RPC record and fragment assembly */ + +#define RPCSVC_FRAGHDR_SIZE 4 /* 4-byte RPC fragment header size */ + +/* Given the 4-byte fragment header, returns non-zero if this fragment + * is the last fragment for the RPC record being assemebled. + * RPC Record marking standard defines a 32 bit value as the fragment + * header with the MSB signifying whether the fragment is the last + * fragment for the record being asembled. + */ +#define RPCSVC_LASTFRAG(fraghdr) ((uint32_t)(fraghdr & 0x80000000U)) + +/* Given the 4-byte fragment header, extracts the bits that contain + * the fragment size. + */ +#define RPCSVC_FRAGSIZE(fraghdr) ((uint32_t)(fraghdr & 0x7fffffffU)) + +/* RPC Record States */ +#define RPCSVC_READ_FRAGHDR 1 +#define RPCSVC_READ_FRAG 2 +/* The size in bytes, if crossed by a fragment will be handed over to the + * vectored actor so that it can allocate its buffers the way it wants. + * In our RPC layer, we assume that vectored RPC requests/records are never + * spread over multiple RPC fragments since that prevents us from determining + * whether the record should be handled in RPC layer completely or handed to + * the vectored handler. + */ +#define RPCSVC_VECTORED_FRAGSZ 4096 +#define RPCSVC_VECTOR_READCRED 1003 +#define RPCSVC_VECTOR_READVERFSZ 1004 +#define RPCSVC_VECTOR_READVERF 1005 +#define RPCSVC_VECTOR_IGNORE 1006 +#define RPCSVC_VECTOR_READVEC 1007 +#define RPCSVC_VECTOR_READPROCHDR 1008 + +#define rpcsvc_record_vectored_baremsg(rs) (((rs)->state == RPCSVC_READ_FRAG) && (rs)->vecstate == 0) +#define rpcsvc_record_vectored_cred(rs) ((rs)->vecstate == RPCSVC_VECTOR_READCRED) +#define rpcsvc_record_vectored_verfsz(rs) ((rs)->vecstate == RPCSVC_VECTOR_READVERFSZ) +#define rpcsvc_record_vectored_verfread(rs) ((rs)->vecstate == RPCSVC_VECTOR_READVERF) +#define rpcsvc_record_vectored_ignore(rs) ((rs)->vecstate == RPCSVC_VECTOR_IGNORE) +#define rpcsvc_record_vectored_readvec(rs) ((rs)->vecstate == RPCSVC_VECTOR_READVEC) +#define rpcsvc_record_vectored_readprochdr(rs) ((rs)->vecstate == RPCSVC_VECTOR_READPROCHDR) +#define rpcsvc_record_vectored(rs) ((rs)->fragsize > RPCSVC_VECTORED_FRAGSZ) +/* Includes bytes up to and including the credential length field. The credlen + * will be followed by @credlen bytes of credential data which will have to be + * read separately by the vectored reader. After the credentials comes the + * verifier which will also have to be read separately including the 8 bytes of + * verf flavour and verflen. + */ +#define RPCSVC_BARERPC_MSGSZ 32 +#define rpcsvc_record_readfraghdr(rs) ((rs)->state == RPCSVC_READ_FRAGHDR) +#define rpcsvc_record_readfrag(rs) ((rs)->state == RPCSVC_READ_FRAG) + +#define rpcsvc_conn_rpcsvc(conn) ((conn)->stage->svc) +#define RPCSVC_LOWVERS 2 +#define RPCSVC_HIGHVERS 2 + +typedef struct rpc_svc_program rpcsvc_program_t; +/* A Stage is the event handler thread together with + * the connections being served by this thread. + * It is called a stage because all the actors, i.e, protocol actors, + * defined by higher level users of the RPC layer, are executed here. + */ +typedef struct rpc_svc_stage_context { + pthread_t tid; + struct event_pool *eventpool; /* Per-stage event-pool */ + void *svc; /* Ref to the rpcsvc_t */ +} rpcsvc_stage_t; + + +/* RPC Records and Fragments assembly state. + * This is per-connection state that is used to determine + * how much data has come in, how much more needs to be read + * and where it needs to be read. + * + * All this state is then used to re-assemble network buffers into + * RPC fragments, which are then re-assembled into RPC records. + * + * See RFC 1831: "RPC: Remote Procedure Call Protocol Specification Version 2", + * particularly the section on Record Marking Standard. + */ +typedef struct rpcsvc_record_state { + + /* Pending messages storage + * This memory area is currently being used to assemble + * the latest RPC record. + * + * Note that this buffer contains the data other than the + * fragment headers received from the network. This is so that we can + * directly pass this buffer to higher layers without requiring to + * perform memory copies and marshalling of data. + */ + struct iobuf *activeiob; + + struct iobuf *vectoriob; + /* The pointer into activeiob memory, into which will go the + * contents from the next read from the network. + */ + char *fragcurrent; + + /* Size of the currently incomplete RPC fragment. + * This is filled in when the fragment header comes in. + * Even though only the 31 least significant bits are used from the + * fragment header, we use a 32 bit variable to store the size. + */ + uint32_t fragsize; + + /* The fragment header is always read in here so that + * the RPC messages contained in a RPC records can be processed + * separately without copying them out of the activeiob above. + */ + char fragheader[RPCSVC_FRAGHDR_SIZE]; + char *hdrcurrent; + + /* Bytes remaining to come in for the current fragment. */ + uint32_t remainingfrag; + + /* It is possible for the frag header to be split over separate + * read calls, so we need to keep track of how much is left. + */ + uint32_t remainingfraghdr; + + /* Record size, the total size of the RPC record, i.e. the total + * of all fragment sizes received till now. Does not include the size + * of a partial fragment which is continuing to be assembled right now. + */ + int recordsize; + + /* Current state of the record */ + int state; + + /* Current state of the vectored reading process. */ + int vecstate; + + /* Set to non-zero when the currently partial or complete fragment is + * the last fragment being received for the current RPC record. + */ + uint32_t islastfrag; + +} rpcsvc_record_state_t; + + +#define RPCSVC_CONNSTATE_CONNECTED 1 +#define RPCSVC_CONNSTATE_DISCONNECTED 2 + +#define rpcsvc_conn_check_active(conn) ((conn)->connstate==RPCSVC_CONNSTATE_CONNECTED) + +typedef struct rpcsvc_request rpcsvc_request_t; +/* Contains the state for each connection that is used for transmitting and + * receiving RPC messages. + * + * There is also an eventidx because each connection's fd is added to the event + * pool of the stage to which a connection belongs. + * Anything that can be accessed by a RPC program must be synced through + * connlock. + */ +typedef struct rpc_conn_state { + + /* Transport or connection state */ + + /* Once we start working on RDMA support, this TCP specific state will + * have to be abstracted away. + */ + int sockfd; + int eventidx; + int windowsize; + + /* Reference to the stage which is handling this + * connection. + */ + rpcsvc_stage_t *stage; + + /* RPC Records and Fragments assembly state. + * All incoming data is staged here before being + * called a full RPC message. + */ + rpcsvc_record_state_t rstate; + + /* It is possible that a client disconnects while + * the higher layer RPC service is busy in a call. + * In this case, we cannot just free the conn + * structure, since the higher layer service could + * still have a reference to it. + * The refcount avoids freeing until all references + * have been given up, although the connection is clos()ed at the first + * call to unref. + */ + int connref; + pthread_mutex_t connlock; + int connstate; + + /* The program that is listening for requests on this connection. */ + rpcsvc_program_t *program; + + /* List of buffers awaiting transmission */ + /* Accesses to txbufs between multiple threads calling + * rpcsvc_submit is synced through connlock. Prefer spinlock over + * mutex because this is a low overhead op that needs simple + * appending to the tx list. + */ + struct list_head txbufs; + + /* Mem pool for the txbufs above. */ + struct mem_pool *txpool; + + /* Memory pool for rpcsvc_request_t */ + struct mem_pool *rxpool; + + /* The request which hasnt yet been handed to the RPC program because + * this request is being treated as a vector request and so needs some + * more data to be got from the network. + */ + rpcsvc_request_t *vectoredreq; +} rpcsvc_conn_t; + + +#define RPCSVC_MAX_AUTH_BYTES 400 +typedef struct rpcsvc_auth_data { + int flavour; + int datalen; + char authdata[RPCSVC_MAX_AUTH_BYTES]; +} rpcsvc_auth_data_t; + +#define rpcsvc_auth_flavour(au) ((au).flavour) + +/* The container for the RPC call handed up to an actor. + * Dynamically allocated. Lives till the call reply is completely + * transmitted. + * */ +struct rpcsvc_request { + /* Connection over which this request came. */ + rpcsvc_conn_t *conn; + + /* The identifier for the call from client. + * Needed to pair the reply with the call. + */ + uint32_t xid; + + int prognum; + + int progver; + + int procnum; + /* Uid and gid filled by the rpc-auth module during the authentication + * phase. + */ + uid_t uid; + gid_t gid; + + /* Might want to move this to AUTH_UNIX specifix state since this array + * is not available for every authenticatino scheme. + */ + gid_t auxgids[NGRPS]; + int auxgidcount; + + + /* The RPC message payload, contains the data required + * by the program actors. This is the buffer that will need to + * be de-xdred by the actor. + */ + struct iovec msg; + + /* The full message buffer allocated to store the RPC headers. + * This buffer is ref'd when allocated why RPC svc and unref'd after + * the buffer is handed to the actor. That means if the actor or any + * higher layer wants to keep this buffer around, they too must ref it + * right after entering the program actor. + */ + struct iobuf *recordiob; + + /* Status of the RPC call, whether it was accepted or denied. */ + int rpc_stat; + + /* In case, the call was denied, the RPC error is stored here + * till the reply is sent. + */ + int rpc_err; + + /* In case the failure happened because of an authentication problem + * , this value needs to be assigned the correct auth error number. + */ + int auth_err; + + /* There can be cases of RPC requests where the reply needs to + * be built from multiple sources. For eg. where even the NFS reply can + * contain a payload, as in the NFSv3 read reply. Here the RPC header + * ,NFS header and the read data are brought together separately from + * different buffers, so we need to stage the buffers temporarily here + * before all of them get added to the connection's transmission list. + */ + struct list_head txlist; + + /* While the reply record is being built, this variable keeps track + * of how many bytes have been added to the record. + */ + size_t payloadsize; + + /* The credentials extracted from the rpc request */ + rpcsvc_auth_data_t cred; + + /* The verified extracted from the rpc request. In request side + * processing this contains the verifier sent by the client, on reply + * side processing, it is filled with the verified that will be + * sent to the client. + */ + rpcsvc_auth_data_t verf; + + /* Container for a RPC program wanting to store a temp + * request-specific item. + */ + void *private; + +}; + +#define rpcsvc_request_program(req) ((rpcsvc_program_t *)((req)->conn->program)) +#define rpcsvc_request_program_private(req) (((rpcsvc_program_t *)((req)->conn->program))->private) +#define rpcsvc_request_conn(req) (req)->conn +#define rpcsvc_request_accepted(req) ((req)->rpc_stat == 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_stage_service(stg) ((rpcsvc_t *)((stg)->svc)) +#define rpcsvc_conn_stage(conn) ((conn)->stage) +#define rpcsvc_request_service(req) (rpcsvc_stage_service(rpcsvc_conn_stage(rpcsvc_request_conn(req)))) +#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)) + +#define rpcsvc_request_uid(req) ((req)->uid) +#define rpcsvc_request_gid(req) ((req)->gid) +#define rpcsvc_request_private(req) ((req)->private) +#define rpcsvc_request_xid(req) ((req)->xid) +#define rpcsvc_request_set_private(req,prv) (req)->private = (void *)(prv) +#define rpcsvc_request_record_iob(rq) ((rq)->recordiob) +#define rpcsvc_request_record_ref(req) (iobuf_ref ((req)->recordiob)) +#define rpcsvc_request_record_unref(req) (iobuf_unref ((req)->recordiob)) + + +#define RPCSVC_ACTOR_SUCCESS 0 +#define RPCSVC_ACTOR_ERROR (-1) + +/* Functor for every type of protocol actor + * must be defined like this. + * + * See the request structure for info on how to handle the request + * in the program actor. + * + * On successful santify checks inside the actor, it should return + * RPCSVC_ACTOR_SUCCESS. + * On an error, on which the RPC layer is expected to return a reply, the actor + * should return RPCSVC_ACTOR_ERROR. + * + */ +typedef int (*rpcsvc_actor) (rpcsvc_request_t *req); +typedef int (*rpcsvc_vector_actor) (rpcsvc_request_t *req, struct iobuf *iob); +typedef int (*rpcsvc_vector_sizer) (rpcsvc_request_t *req, ssize_t *readsize, + int *newiob); + +/* 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 + * transmitting on the connection. + */ +typedef void *(*rpcsvc_encode_reply) (void *msg); + +/* Once the reply has been transmitted, the message will have to be de-allocated + * , so every actor will need to provide a function that deallocates the message + * it had allocated as a response. + */ +typedef void (*rpcsvc_deallocate_reply) (void *msg); + + +#define RPCSVC_NAME_MAX 32 +/* The descriptor for each procedure/actor that runs + * over the RPC service. + */ +typedef struct rpc_svc_actor_desc { + char procname[RPCSVC_NAME_MAX]; + int procnum; + rpcsvc_actor actor; + + /* Handler for cases where the RPC requests fragments are large enough + * to benefit from being decoded into aligned memory addresses. While + * decoding the request in a non-vectored manner, due to the nature of + * the XDR scheme, RPC cannot guarantee memory aligned addresses for + * the resulting message-specific structures. Allowing a specialized + * handler for letting the RPC program read the data from the network + * directly into its alligned buffers. + */ + rpcsvc_vector_actor vector_actor; + rpcsvc_vector_sizer vector_sizer; + +} rpcsvc_actor_t; + +typedef int (*rpcsvc_conn_notify_fn) (void *progpriv, rpcsvc_conn_t *conn); + +/* Describes a program and its version along with the function pointers + * required to handle the procedures/actors of each program/version. + * Never changed ever by any thread so no need for a lock. + */ +struct rpc_svc_program { + char progname[RPCSVC_NAME_MAX]; + int prognum; + int progver; + uint16_t progport; /* Registered with portmap */ + int progaddrfamily; /* AF_INET or AF_INET6 */ + char *proghost; /* Bind host, can be NULL */ + rpcsvc_actor_t *actors; /* All procedure handlers */ + int numactors; /* Num actors in actor array */ + int proghighvers; /* Highest ver for program + supported by the system. */ + int proglowvers; /* Lowest ver */ + + /* Program specific state handed to actors */ + void *private; + + /* This upcall is made when a connection's refcount reaches 0 and the + * connection is about to be destroyed. We want to let the RPC program + * know that it should also now free any state it is maintaining + * for this connection. + */ + rpcsvc_conn_notify_fn conn_destroy; + + /* Used to tell RPC program to init the state it needs to associate + * with the new connection. + */ + rpcsvc_conn_notify_fn conn_init; + + /* An integer that identifies the min auth strength that is required + * by this protocol, for eg. MOUNT3 needs AUTH_UNIX at least. + * See RFC 1813, Section 5.2.1. + */ + int min_auth; +}; + + +/* Contains global state required for all the RPC services. + */ +typedef struct rpc_svc_state { + + /* Contains the list of rpcsvc_stage_t + * list of (program, version) handlers. + * other options. + */ + + /* At this point, lock is not used to protect anything. Later, it'll + * be used for protecting stages. + */ + pthread_mutex_t rpclock; + + /* This is the first stage that is inited, so that any RPC based + * services that do not need multi-threaded support can just use the + * service right away. This is not added to the stages list + * declared later. + * This is also the stage over which all service listeners are run. + */ + rpcsvc_stage_t *defaultstage; + + /* When we have multi-threaded RPC support, we'll use this to link + * to the multiple Stages. + */ + struct list_head stages; /* All stages */ + + unsigned int memfactor; + + /* List of the authentication schemes available. */ + struct list_head authschemes; + + /* Reference to the options */ + dict_t *options; + + /* Allow insecure ports. */ + int allow_insecure; + + glusterfs_ctx_t *ctx; +} rpcsvc_t; + + +/* All users of RPC services should use this API to register their + * procedure handlers. + */ +extern int +rpcsvc_program_register (rpcsvc_t *svc, rpcsvc_program_t program); + +extern int +rpcsvc_program_unregister (rpcsvc_t *svc, rpcsvc_program_t program); + +/* Inits the global RPC service data structures. + * Called in main. + */ +extern rpcsvc_t * +rpcsvc_init (glusterfs_ctx_t *ctx, dict_t *options); + + +extern int +rpcsvc_submit_message (rpcsvc_request_t * req, struct iovec msg, + struct iobuf *iob); + +int +rpcsvc_submit_generic (rpcsvc_request_t *req, struct iovec msgvec, + struct iobuf *msg); +#define rpcsvc_record_currentfrag_addr(rs) ((rs)->fragcurrent) +#define rpcsvc_record_currenthdr_addr(rs) ((rs)->hdrcurrent) + +#define rpcsvc_record_update_currentfrag(rs, size) \ + do { \ + (rs)->fragcurrent += size; \ + } while (0) \ + +#define rpcsvc_record_update_currenthdr(rs, size) \ + do { \ + (rs)->hdrcurrent += size; \ + } while (0) \ + + +/* These are used to differentiate between multiple txbufs which form + * a single RPC record. For eg, one purpose we use these for is to + * prevent dividing a RPC record over multiple TCP segments. Multiple + * TCP segments are possible for a single RPC record because we generally do not + * have control over how the kernel's TCP segments the buffers when putting + * them on the wire. So, on Linux, we use these to set TCP_CORK to create + * a single TCP segment from multiple txbufs that are part of the same RPC + * record. This improves network performance by reducing tiny message + * transmissions. + */ +#define RPCSVC_TXB_FIRST 0x1 +#define RPCSVC_TXB_LAST 0x2 + +/* The list of buffers appended to a connection's pending + * transmission list. + */ +typedef struct rpcsvc_txbuf { + struct list_head txlist; + /* The iobuf which contains the full message to be transmitted */ + struct iobuf *iob; + + /* For vectored messages from an RPC program, we need to be able + * maintain a ref to an iobuf which we do not have access to directly + * except through the iobref which in turn could've been passed to + * the RPC program by a higher layer. + * + * So either the iob is defined or iobref is defined for a reply, + * never both. + */ + struct iobref *iobref; + /* In order to handle non-blocking writes, we'll need to keep track of + * how much data from an iobuf has been written and where the next + * transmission needs to start from. This iov.base points to the base of + * the iobuf, iov.len is the size of iobuf being used for the message + * from the total size in the iobuf. + */ + struct iovec buf; + /* offset is the point from where the next transmission for this buffer + * should start. + */ + size_t offset; + + /* This is a special field that tells us what kind of transmission + * behaviour to provide to a particular buffer. + * See the RPCSVC_TXB_* defines for more info. + */ + int txbehave; +} rpcsvc_txbuf_t; + +extern int +rpcsvc_error_reply (rpcsvc_request_t *req); + +#define RPCSVC_PEER_STRLEN 1024 +#define RPCSVC_AUTH_ACCEPT 1 +#define RPCSVC_AUTH_REJECT 2 +#define RPCSVC_AUTH_DONTCARE 3 + +extern int +rpcsvc_conn_peername (rpcsvc_conn_t *conn, char *hostname, int hostlen); + +extern int +rpcsvc_conn_peeraddr (rpcsvc_conn_t *conn, char *addrstr, int addrlen, + struct sockaddr *returnsa, socklen_t sasize); + +extern int +rpcsvc_conn_peer_check (dict_t *options, char *volname, rpcsvc_conn_t *conn); + +extern int +rpcsvc_conn_privport_check (rpcsvc_t *svc, char *volname, rpcsvc_conn_t *conn); +#define rpcsvc_request_seterr(req, err) (req)->rpc_err = err +#define rpcsvc_request_set_autherr(req, err) (req)->auth_err = err + +extern void +rpcsvc_conn_deinit (rpcsvc_conn_t *conn); +extern void rpcsvc_conn_ref (rpcsvc_conn_t *conn); +extern void rpcsvc_conn_unref (rpcsvc_conn_t *conn); + +extern int rpcsvc_submit_vectors (rpcsvc_request_t *req); + +extern int rpcsvc_request_attach_vector (rpcsvc_request_t *req, + struct iovec msgvec, struct iobuf *iob, + struct iobref *ioref, int finalvector); + + +typedef int (*auth_init_conn) (rpcsvc_conn_t *conn, void *priv); +typedef int (*auth_init_request) (rpcsvc_request_t *req, void *priv); +typedef int (*auth_request_authenticate) (rpcsvc_request_t *req, void *priv); + +/* This structure needs to be registered by every authentication scheme. + * Our authentication schemes are stored per connection because + * each connection will end up using a different authentication scheme. + */ +typedef struct rpcsvc_auth_ops { + auth_init_conn conn_init; + auth_init_request request_init; + auth_request_authenticate authenticate; +} rpcsvc_auth_ops_t; + +typedef struct rpcsvc_auth_flavour_desc { + char authname[RPCSVC_NAME_MAX]; + int authnum; + rpcsvc_auth_ops_t *authops; + void *authprivate; +} rpcsvc_auth_t; + +typedef void * (*rpcsvc_auth_initer_t) (rpcsvc_t *svc, dict_t *options); + +struct rpcsvc_auth_list { + struct list_head authlist; + rpcsvc_auth_initer_t init; + /* Should be the name with which we identify the auth scheme given + * in the volfile options. + * This should be different from the authname in rpc_auth_t + * in way that makes it easier to specify this scheme in the volfile. + * This is because the technical names of the schemes can be a bit + * arcane. + */ + char name[RPCSVC_NAME_MAX]; + rpcsvc_auth_t *auth; + int enable; +}; + +extern int +rpcsvc_auth_request_init (rpcsvc_request_t *req); + +extern int +rpcsvc_auth_init (rpcsvc_t *svc, dict_t *options); + +extern int +rpcsvc_auth_conn_init (rpcsvc_conn_t *conn); + +extern int +rpcsvc_authenticate (rpcsvc_request_t *req); + +extern int +rpcsvc_auth_array (rpcsvc_t *svc, char *volname, int *autharr, int arrlen); + +/* If the request has been sent using AUTH_UNIX, this function returns the + * auxiliary gids as an array, otherwise, it returns NULL. + * Move to auth-unix specific source file when we need to modularize the + * authentication code even further to support mode auth schemes. + */ +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); +#endif -- cgit