X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/743b15655a24ee3fe9f458f383003e011db0558f..15129b1c8dbb3650c63b70adb1cad9af601c6c17:/bsd/nfs/nfs_socket.c diff --git a/bsd/nfs/nfs_socket.c b/bsd/nfs/nfs_socket.c index 2da38e94c..c163bfa04 100644 --- a/bsd/nfs/nfs_socket.c +++ b/bsd/nfs/nfs_socket.c @@ -1,23 +1,29 @@ /* - * Copyright (c) 2000-2005 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2011 Apple Inc. All rights reserved. * - * @APPLE_LICENSE_HEADER_START@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * - * The contents of this file constitute Original Code as defined in and - * are subject to the Apple Public Source License Version 1.1 (the - * "License"). You may not use this file except in compliance with the - * License. Please obtain a copy of the License at - * http://www.apple.com/publicsource and read it before using this file. + * This file contains Original Code and/or Modifications of Original Code + * as defined in and that are subject to the Apple Public Source License + * Version 2.0 (the 'License'). You may not use this file except in + * compliance with the License. The rights granted to you under the License + * may not be used to create, or enable the creation or redistribution of, + * unlawful or unlicensed copies of an Apple operating system, or to + * circumvent, violate, or enable the circumvention or violation of, any + * terms of an Apple operating system software license agreement. * - * This Original Code and all software distributed under the License are - * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER + * Please obtain a copy of the License at + * http://www.opensource.apple.com/apsl/ and read it before using this file. + * + * The Original Code and all software distributed under the License are + * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, - * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the - * License for the specific language governing rights and limitations - * under the License. + * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. + * Please see the License for the specific language governing rights and + * limitations under the License. * - * @APPLE_LICENSE_HEADER_END@ + * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */ /* @@ -66,6 +72,7 @@ #include #include #include +#include #include #include #include @@ -77,38 +84,90 @@ #include #include #include -#include #include #include #include #include #include +#include #include +#include #include #include #include +#include #include #include #include #include +#include #include #include -#include -#include +#define NFS_SOCK_DBG(...) NFS_DBG(NFS_FAC_SOCK, 7, ## __VA_ARGS__) + +/* XXX */ +boolean_t current_thread_aborted(void); +kern_return_t thread_terminate(thread_t); + + +#if NFSSERVER +int nfsrv_sock_max_rec_queue_length = 128; /* max # RPC records queued on (UDP) socket */ + +int nfsrv_getstream(struct nfsrv_sock *,int); +int nfsrv_getreq(struct nfsrv_descript *); +extern int nfsv3_procid[NFS_NPROCS]; +#endif /* NFSSERVER */ + +/* + * compare two sockaddr structures + */ +int +nfs_sockaddr_cmp(struct sockaddr *sa1, struct sockaddr *sa2) +{ + if (!sa1) + return (-1); + if (!sa2) + return (1); + if (sa1->sa_family != sa2->sa_family) + return ((sa1->sa_family < sa2->sa_family) ? -1 : 1); + if (sa1->sa_len != sa2->sa_len) + return ((sa1->sa_len < sa2->sa_len) ? -1 : 1); + if (sa1->sa_family == AF_INET) + return (bcmp(&((struct sockaddr_in*)sa1)->sin_addr, + &((struct sockaddr_in*)sa2)->sin_addr, sizeof(((struct sockaddr_in*)sa1)->sin_addr))); + if (sa1->sa_family == AF_INET6) + return (bcmp(&((struct sockaddr_in6*)sa1)->sin6_addr, + &((struct sockaddr_in6*)sa2)->sin6_addr, sizeof(((struct sockaddr_in6*)sa1)->sin6_addr))); + return (-1); +} -#define FSDBG(A, B, C, D, E) \ - KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_NONE, \ - (int)(B), (int)(C), (int)(D), (int)(E), 0) -#define FSDBG_TOP(A, B, C, D, E) \ - KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_START, \ - (int)(B), (int)(C), (int)(D), (int)(E), 0) -#define FSDBG_BOT(A, B, C, D, E) \ - KERNEL_DEBUG((FSDBG_CODE(DBG_FSRW, (A))) | DBG_FUNC_END, \ - (int)(B), (int)(C), (int)(D), (int)(E), 0) +#if NFSCLIENT + +int nfs_connect_search_new_socket(struct nfsmount *, struct nfs_socket_search *, struct timeval *); +int nfs_connect_search_socket_connect(struct nfsmount *, struct nfs_socket *, int); +int nfs_connect_search_ping(struct nfsmount *, struct nfs_socket *, struct timeval *); +void nfs_connect_search_socket_found(struct nfsmount *, struct nfs_socket_search *, struct nfs_socket *); +void nfs_connect_search_socket_reap(struct nfsmount *, struct nfs_socket_search *, struct timeval *); +int nfs_connect_search_check(struct nfsmount *, struct nfs_socket_search *, struct timeval *); +int nfs_reconnect(struct nfsmount *); +int nfs_connect_setup(struct nfsmount *); +void nfs_mount_sock_thread(void *, wait_result_t); +void nfs_udp_rcv(socket_t, void*, int); +void nfs_tcp_rcv(socket_t, void*, int); +void nfs_sock_poke(struct nfsmount *); +void nfs_request_match_reply(struct nfsmount *, mbuf_t); +void nfs_reqdequeue(struct nfsreq *); +void nfs_reqbusy(struct nfsreq *); +struct nfsreq *nfs_reqnext(struct nfsreq *); +int nfs_wait_reply(struct nfsreq *); +void nfs_softterm(struct nfsreq *); +int nfs_can_squish(struct nfsmount *); +int nfs_is_squishy(struct nfsmount *); +int nfs_is_dead(int, struct nfsmount *); /* * Estimate rto for an nfs rpc sent via. an unreliable datagram. @@ -129,17 +188,6 @@ ((((n)->nm_srtt[t-1] + 7) >> 3) + (n)->nm_sdrtt[t-1] + 1))) #define NFS_SRTT(r) (r)->r_nmp->nm_srtt[proct[(r)->r_procnum] - 1] #define NFS_SDRTT(r) (r)->r_nmp->nm_sdrtt[proct[(r)->r_procnum] - 1] -/* - * External data, mostly RPC constants in XDR form - */ -extern u_long rpc_reply, rpc_msgdenied, rpc_mismatch, rpc_vers, rpc_auth_unix, - rpc_msgaccepted, rpc_call, rpc_autherr, - rpc_auth_kerb; -extern u_long nfs_prog; -extern struct nfsstats nfsstats; -extern int nfsv3_procid[NFS_NPROCS]; -extern int nfs_ticks; -extern u_long nfs_xidwrap; /* * Defines which timer to use for the procnum. @@ -169,1978 +217,5484 @@ static int proct[NFS_NPROCS] = { #define NFS_CWNDSCALE 256 #define NFS_MAXCWND (NFS_CWNDSCALE * 32) static int nfs_backoff[8] = { 2, 4, 8, 16, 32, 64, 128, 256, }; -int nfsrtton = 0; -struct nfsrtt nfsrtt; -static int nfs_rcvlock(struct nfsreq *); -static void nfs_rcvunlock(struct nfsreq *); -static int nfs_receive(struct nfsreq *rep, mbuf_t *mp); -static int nfs_reconnect(struct nfsreq *rep); -static void nfs_repdequeue(struct nfsreq *rep); +/* + * Increment location index to next address/server/location. + */ +void +nfs_location_next(struct nfs_fs_locations *nlp, struct nfs_location_index *nlip) +{ + uint8_t loc = nlip->nli_loc; + uint8_t serv = nlip->nli_serv; + uint8_t addr = nlip->nli_addr; + + /* move to next address */ + addr++; + if (addr >= nlp->nl_locations[loc]->nl_servers[serv]->ns_addrcount) { + /* no more addresses on current server, go to first address of next server */ +next_server: + addr = 0; + serv++; + if (serv >= nlp->nl_locations[loc]->nl_servcount) { + /* no more servers on current location, go to first server of next location */ + serv = 0; + loc++; + if (loc >= nlp->nl_numlocs) + loc = 0; /* after last location, wrap back around to first location */ + } + } + /* + * It's possible for this next server to not have any addresses. + * Check for that here and go to the next server. + * But bail out if we've managed to come back around to the original + * location that was passed in. (That would mean no servers had any + * addresses. And we don't want to spin here forever.) + */ + if ((loc == nlip->nli_loc) && (serv == nlip->nli_serv) && (addr == nlip->nli_addr)) + return; + if (addr >= nlp->nl_locations[loc]->nl_servers[serv]->ns_addrcount) + goto next_server; -/* XXX */ -boolean_t current_thread_aborted(void); -kern_return_t thread_terminate(thread_t); + nlip->nli_loc = loc; + nlip->nli_serv = serv; + nlip->nli_addr = addr; +} -#ifndef NFS_NOSERVER -static int nfsrv_getstream(struct nfssvc_sock *,int); - -int (*nfsrv3_procs[NFS_NPROCS])(struct nfsrv_descript *nd, - struct nfssvc_sock *slp, - proc_t procp, - mbuf_t *mreqp) = { - nfsrv_null, - nfsrv_getattr, - nfsrv_setattr, - nfsrv_lookup, - nfsrv3_access, - nfsrv_readlink, - nfsrv_read, - nfsrv_write, - nfsrv_create, - nfsrv_mkdir, - nfsrv_symlink, - nfsrv_mknod, - nfsrv_remove, - nfsrv_rmdir, - nfsrv_rename, - nfsrv_link, - nfsrv_readdir, - nfsrv_readdirplus, - nfsrv_statfs, - nfsrv_fsinfo, - nfsrv_pathconf, - nfsrv_commit, - nfsrv_noop -}; -#endif /* NFS_NOSERVER */ +/* + * Compare two location indices. + */ +int +nfs_location_index_cmp(struct nfs_location_index *nlip1, struct nfs_location_index *nlip2) +{ + if (nlip1->nli_loc != nlip2->nli_loc) + return (nlip1->nli_loc - nlip2->nli_loc); + if (nlip1->nli_serv != nlip2->nli_serv) + return (nlip1->nli_serv - nlip2->nli_serv); + return (nlip1->nli_addr - nlip2->nli_addr); +} +/* + * Get the mntfromname (or path portion only) for a given location. + */ +void +nfs_location_mntfromname(struct nfs_fs_locations *locs, struct nfs_location_index idx, char *s, int size, int pathonly) +{ + struct nfs_fs_location *fsl = locs->nl_locations[idx.nli_loc]; + char *p; + int cnt, i; + + p = s; + if (!pathonly) { + cnt = snprintf(p, size, "%s:", fsl->nl_servers[idx.nli_serv]->ns_name); + p += cnt; + size -= cnt; + } + if (fsl->nl_path.np_compcount == 0) { + /* mounting root export on server */ + if (size > 0) { + *p++ = '/'; + *p++ = '\0'; + } + return; + } + /* append each server path component */ + for (i=0; (size > 0) && (i < (int)fsl->nl_path.np_compcount); i++) { + cnt = snprintf(p, size, "/%s", fsl->nl_path.np_components[i]); + p += cnt; + size -= cnt; + } +} /* - * attempt to bind a socket to a reserved port + * NFS client connect socket upcall. + * (Used only during socket connect/search.) */ -static int -nfs_bind_resv(struct nfsmount *nmp) +void +nfs_connect_upcall(socket_t so, void *arg, __unused int waitflag) { - socket_t so = nmp->nm_so; - struct sockaddr_in sin; - int error; - u_short tport; + struct nfs_socket *nso = arg; + size_t rcvlen; + mbuf_t m; + int error = 0, recv = 1; - if (!so) - return (EINVAL); + if (nso->nso_flags & NSO_CONNECTING) { + NFS_SOCK_DBG("nfs connect - socket %p upcall - connecting\n", nso); + wakeup(nso->nso_wake); + return; + } - sin.sin_len = sizeof (struct sockaddr_in); - sin.sin_family = AF_INET; - sin.sin_addr.s_addr = INADDR_ANY; - tport = IPPORT_RESERVED - 1; - sin.sin_port = htons(tport); + lck_mtx_lock(&nso->nso_lock); + if ((nso->nso_flags & (NSO_UPCALL|NSO_DISCONNECTING|NSO_DEAD)) || !(nso->nso_flags & NSO_PINGING)) { + NFS_SOCK_DBG("nfs connect - socket %p upcall - nevermind\n", nso); + lck_mtx_unlock(&nso->nso_lock); + return; + } + NFS_SOCK_DBG("nfs connect - socket %p upcall\n", nso); + nso->nso_flags |= NSO_UPCALL; - while (((error = sock_bind(so, (struct sockaddr *) &sin)) == EADDRINUSE) && - (--tport > IPPORT_RESERVED / 2)) - sin.sin_port = htons(tport); - return (error); -} + /* loop while we make error-free progress */ + while (!error && recv) { + /* make sure we're still interested in this socket */ + if (nso->nso_flags & (NSO_DISCONNECTING|NSO_DEAD)) + break; + lck_mtx_unlock(&nso->nso_lock); + m = NULL; + if (nso->nso_sotype == SOCK_STREAM) { + error = nfs_rpc_record_read(so, &nso->nso_rrs, MSG_DONTWAIT, &recv, &m); + } else { + rcvlen = 1000000; + error = sock_receivembuf(so, NULL, &m, MSG_DONTWAIT, &rcvlen); + recv = m ? 1 : 0; + } + lck_mtx_lock(&nso->nso_lock); + if (m) { + /* match response with request */ + struct nfsm_chain nmrep; + uint32_t reply = 0, rxid = 0, verf_type, verf_len; + uint32_t reply_status, rejected_status, accepted_status; + + nfsm_chain_dissect_init(error, &nmrep, m); + nfsm_chain_get_32(error, &nmrep, rxid); + nfsm_chain_get_32(error, &nmrep, reply); + if (!error && ((reply != RPC_REPLY) || (rxid != nso->nso_pingxid))) + error = EBADRPC; + nfsm_chain_get_32(error, &nmrep, reply_status); + if (!error && (reply_status == RPC_MSGDENIED)) { + nfsm_chain_get_32(error, &nmrep, rejected_status); + if (!error) + error = (rejected_status == RPC_MISMATCH) ? ERPCMISMATCH : EACCES; + } + nfsm_chain_get_32(error, &nmrep, verf_type); /* verifier flavor */ + nfsm_chain_get_32(error, &nmrep, verf_len); /* verifier length */ + nfsmout_if(error); + if (verf_len) + nfsm_chain_adv(error, &nmrep, nfsm_rndup(verf_len)); + nfsm_chain_get_32(error, &nmrep, accepted_status); + nfsmout_if(error); + if ((accepted_status == RPC_PROGMISMATCH) && !nso->nso_version) { + uint32_t minvers, maxvers; + nfsm_chain_get_32(error, &nmrep, minvers); + nfsm_chain_get_32(error, &nmrep, maxvers); + nfsmout_if(error); + if (nso->nso_protocol == PMAPPROG) { + if ((minvers > RPCBVERS4) || (maxvers < PMAPVERS)) + error = EPROGMISMATCH; + else if ((nso->nso_saddr->sa_family == AF_INET) && + (PMAPVERS >= minvers) && (PMAPVERS <= maxvers)) + nso->nso_version = PMAPVERS; + else if (nso->nso_saddr->sa_family == AF_INET6) { + if ((RPCBVERS4 >= minvers) && (RPCBVERS4 <= maxvers)) + nso->nso_version = RPCBVERS4; + else if ((RPCBVERS3 >= minvers) && (RPCBVERS3 <= maxvers)) + nso->nso_version = RPCBVERS3; + } + } else if (nso->nso_protocol == NFS_PROG) { + if ((minvers > NFS_VER4) || (maxvers < NFS_VER2)) + error = EPROGMISMATCH; + else if ((NFS_VER3 >= minvers) && (NFS_VER3 <= maxvers)) + nso->nso_version = NFS_VER3; + else if ((NFS_VER2 >= minvers) && (NFS_VER2 <= maxvers)) + nso->nso_version = NFS_VER2; + else if ((NFS_VER4 >= minvers) && (NFS_VER4 <= maxvers)) + nso->nso_version = NFS_VER4; + } + if (!error && nso->nso_version) + accepted_status = RPC_SUCCESS; + } + if (!error) { + switch (accepted_status) { + case RPC_SUCCESS: + error = 0; + break; + case RPC_PROGUNAVAIL: + error = EPROGUNAVAIL; + break; + case RPC_PROGMISMATCH: + error = EPROGMISMATCH; + break; + case RPC_PROCUNAVAIL: + error = EPROCUNAVAIL; + break; + case RPC_GARBAGE: + error = EBADRPC; + break; + case RPC_SYSTEM_ERR: + default: + error = EIO; + break; + } + } +nfsmout: + nso->nso_flags &= ~NSO_PINGING; + if (error) { + nso->nso_error = error; + nso->nso_flags |= NSO_DEAD; + } else { + nso->nso_flags |= NSO_VERIFIED; + } + mbuf_freem(m); + /* wake up search thread */ + wakeup(nso->nso_wake); + break; + } + } -/* - * variables for managing the nfs_bind_resv_thread - */ -int nfs_resv_mounts = 0; -static int nfs_bind_resv_thread_state = 0; -#define NFS_BIND_RESV_THREAD_STATE_INITTED 1 -#define NFS_BIND_RESV_THREAD_STATE_RUNNING 2 -lck_grp_t *nfs_bind_resv_lck_grp; -lck_grp_attr_t *nfs_bind_resv_lck_grp_attr; -lck_attr_t *nfs_bind_resv_lck_attr; -lck_mtx_t *nfs_bind_resv_mutex; -struct nfs_bind_resv_request { - TAILQ_ENTRY(nfs_bind_resv_request) brr_chain; - struct nfsmount *brr_nmp; - int brr_error; -}; -static TAILQ_HEAD(, nfs_bind_resv_request) nfs_bind_resv_request_queue; + nso->nso_flags &= ~NSO_UPCALL; + if ((error != EWOULDBLOCK) && (error || !recv)) { + /* problems with the socket... */ + nso->nso_error = error ? error : EPIPE; + nso->nso_flags |= NSO_DEAD; + wakeup(nso->nso_wake); + } + if (nso->nso_flags & NSO_DISCONNECTING) + wakeup(&nso->nso_flags); + lck_mtx_unlock(&nso->nso_lock); +} /* - * thread to handle any reserved port bind requests + * Create/initialize an nfs_socket structure. */ -static void -nfs_bind_resv_thread(void) +int +nfs_socket_create( + __unused struct nfsmount *nmp, + struct sockaddr *sa, + int sotype, + in_port_t port, + uint32_t protocol, + uint32_t vers, + int resvport, + struct nfs_socket **nsop) { - struct nfs_bind_resv_request *brreq; + struct nfs_socket *nso; + struct timeval now; + int error; +#ifdef NFS_SOCKET_DEBUGGING + char naddr[MAX_IPv6_STR_LEN]; + void *sinaddr; + + if (sa->sa_family == AF_INET) + sinaddr = &((struct sockaddr_in*)sa)->sin_addr; + else + sinaddr = &((struct sockaddr_in6*)sa)->sin6_addr; + if (inet_ntop(sa->sa_family, sinaddr, naddr, sizeof(naddr)) != naddr) + strlcpy(naddr, "", sizeof(naddr)); +#else + char naddr[1] = { 0 }; +#endif - nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_RUNNING; + *nsop = NULL; - while (nfs_resv_mounts > 0) { - lck_mtx_lock(nfs_bind_resv_mutex); - while ((brreq = TAILQ_FIRST(&nfs_bind_resv_request_queue))) { - TAILQ_REMOVE(&nfs_bind_resv_request_queue, brreq, brr_chain); - lck_mtx_unlock(nfs_bind_resv_mutex); - brreq->brr_error = nfs_bind_resv(brreq->brr_nmp); - wakeup(brreq); - lck_mtx_lock(nfs_bind_resv_mutex); + /* Create the socket. */ + MALLOC(nso, struct nfs_socket *, sizeof(struct nfs_socket), M_TEMP, M_WAITOK|M_ZERO); + if (nso) + MALLOC(nso->nso_saddr, struct sockaddr *, sa->sa_len, M_SONAME, M_WAITOK|M_ZERO); + if (!nso || !nso->nso_saddr) { + if (nso) + FREE(nso, M_TEMP); + return (ENOMEM); + } + lck_mtx_init(&nso->nso_lock, nfs_request_grp, LCK_ATTR_NULL); + nso->nso_sotype = sotype; + if (nso->nso_sotype == SOCK_STREAM) + nfs_rpc_record_state_init(&nso->nso_rrs); + microuptime(&now); + nso->nso_timestamp = now.tv_sec; + bcopy(sa, nso->nso_saddr, sa->sa_len); + if (sa->sa_family == AF_INET) + ((struct sockaddr_in*)nso->nso_saddr)->sin_port = htons(port); + else if (sa->sa_family == AF_INET6) + ((struct sockaddr_in6*)nso->nso_saddr)->sin6_port = htons(port); + nso->nso_protocol = protocol; + nso->nso_version = vers; + + error = sock_socket(sa->sa_family, nso->nso_sotype, 0, NULL, NULL, &nso->nso_so); + + /* Some servers require that the client port be a reserved port number. */ + if (!error && resvport && ((sa->sa_family == AF_INET) || (sa->sa_family == AF_INET6))) { + struct sockaddr_storage ss; + int level = (sa->sa_family == AF_INET) ? IPPROTO_IP : IPPROTO_IPV6; + int optname = (sa->sa_family == AF_INET) ? IP_PORTRANGE : IPV6_PORTRANGE; + int portrange = IP_PORTRANGE_LOW; + + error = sock_setsockopt(nso->nso_so, level, optname, &portrange, sizeof(portrange)); + if (!error) { /* bind now to check for failure */ + ss.ss_len = sa->sa_len; + ss.ss_family = sa->sa_family; + if (ss.ss_family == AF_INET) { + ((struct sockaddr_in*)&ss)->sin_addr.s_addr = INADDR_ANY; + ((struct sockaddr_in*)&ss)->sin_port = htons(0); + } else if (ss.ss_family == AF_INET6) { + ((struct sockaddr_in6*)&ss)->sin6_addr = in6addr_any; + ((struct sockaddr_in6*)&ss)->sin6_port = htons(0); + } else { + error = EINVAL; + } + if (!error) + error = sock_bind(nso->nso_so, (struct sockaddr*)&ss); } - msleep((caddr_t)&nfs_bind_resv_request_queue, - nfs_bind_resv_mutex, PSOCK | PDROP, - "nfs_bind_resv_request_queue", 0); } - nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_INITTED; - (void) thread_terminate(current_thread()); + if (error) { + NFS_SOCK_DBG("nfs connect %s error %d creating socket %p %s type %d%s port %d prot %d %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, error, nso, naddr, sotype, + resvport ? "r" : "", port, protocol, vers); + nfs_socket_destroy(nso); + } else { + NFS_SOCK_DBG("nfs connect %s created socket %p %s type %d%s port %d prot %d %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, naddr, + sotype, resvport ? "r" : "", port, protocol, vers); + *nsop = nso; + } + return (error); } -int -nfs_bind_resv_thread_wake(void) +/* + * Destroy an nfs_socket structure. + */ +void +nfs_socket_destroy(struct nfs_socket *nso) { - if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_RUNNING) - return (EIO); - wakeup(&nfs_bind_resv_request_queue); - return (0); + struct timespec ts = { 4, 0 }; + + lck_mtx_lock(&nso->nso_lock); + nso->nso_flags |= NSO_DISCONNECTING; + if (nso->nso_flags & NSO_UPCALL) /* give upcall a chance to complete */ + msleep(&nso->nso_flags, &nso->nso_lock, PZERO-1, "nfswaitupcall", &ts); + lck_mtx_unlock(&nso->nso_lock); + sock_shutdown(nso->nso_so, SHUT_RDWR); + sock_close(nso->nso_so); + if (nso->nso_sotype == SOCK_STREAM) + nfs_rpc_record_state_cleanup(&nso->nso_rrs); + lck_mtx_destroy(&nso->nso_lock, nfs_request_grp); + if (nso->nso_saddr) + FREE(nso->nso_saddr, M_SONAME); + if (nso->nso_saddr2) + FREE(nso->nso_saddr2, M_SONAME); + NFS_SOCK_DBG("nfs connect - socket %p destroyed\n", nso); + FREE(nso, M_TEMP); } /* - * underprivileged procs call this to request nfs_bind_resv_thread - * to perform the reserved port binding for them. + * Set common socket options on an nfs_socket. */ -static int -nfs_bind_resv_nopriv(struct nfsmount *nmp) +void +nfs_socket_options(struct nfsmount *nmp, struct nfs_socket *nso) { - struct nfs_bind_resv_request brreq; - int error; + /* + * Set socket send/receive timeouts + * - Receive timeout shouldn't matter because most receives are performed + * in the socket upcall non-blocking. + * - Send timeout should allow us to react to a blocked socket. + * Soft mounts will want to abort sooner. + */ + struct timeval timeo; + int on = 1, proto; - if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_RUNNING) { - if (nfs_bind_resv_thread_state < NFS_BIND_RESV_THREAD_STATE_INITTED) { - nfs_bind_resv_lck_grp_attr = lck_grp_attr_alloc_init(); - lck_grp_attr_setstat(nfs_bind_resv_lck_grp_attr); - nfs_bind_resv_lck_grp = lck_grp_alloc_init("nfs_bind_resv", nfs_bind_resv_lck_grp_attr); - nfs_bind_resv_lck_attr = lck_attr_alloc_init(); - nfs_bind_resv_mutex = lck_mtx_alloc_init(nfs_bind_resv_lck_grp, nfs_bind_resv_lck_attr); - TAILQ_INIT(&nfs_bind_resv_request_queue); - nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_INITTED; - } - kernel_thread(kernel_task, nfs_bind_resv_thread); - nfs_bind_resv_thread_state = NFS_BIND_RESV_THREAD_STATE_RUNNING; + timeo.tv_usec = 0; + timeo.tv_sec = (NMFLAG(nmp, SOFT) || nfs_can_squish(nmp)) ? 5 : 60; + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo)); + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo)); + if (nso->nso_sotype == SOCK_STREAM) { + /* Assume that SOCK_STREAM always requires a connection */ + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on)); + /* set nodelay for TCP */ + sock_gettype(nso->nso_so, NULL, NULL, &proto); + if (proto == IPPROTO_TCP) + sock_setsockopt(nso->nso_so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); } + if (nso->nso_sotype == SOCK_DGRAM) { /* set socket buffer sizes for UDP */ + int reserve = NFS_UDPSOCKBUF; + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_SNDBUF, &reserve, sizeof(reserve)); + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_RCVBUF, &reserve, sizeof(reserve)); + } + /* set SO_NOADDRERR to detect network changes ASAP */ + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on)); + /* just playin' it safe with upcalls */ + sock_setsockopt(nso->nso_so, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on)); + /* socket should be interruptible if the mount is */ + if (!NMFLAG(nmp, INTR)) + sock_nointerrupt(nso->nso_so, 1); +} - brreq.brr_nmp = nmp; - brreq.brr_error = 0; - - lck_mtx_lock(nfs_bind_resv_mutex); - TAILQ_INSERT_TAIL(&nfs_bind_resv_request_queue, &brreq, brr_chain); - lck_mtx_unlock(nfs_bind_resv_mutex); +/* + * Release resources held in an nfs_socket_search. + */ +void +nfs_socket_search_cleanup(struct nfs_socket_search *nss) +{ + struct nfs_socket *nso, *nsonext; - error = nfs_bind_resv_thread_wake(); - if (error) { - TAILQ_REMOVE(&nfs_bind_resv_request_queue, &brreq, brr_chain); - /* Note: we might be able to simply restart the thread */ - return (error); + TAILQ_FOREACH_SAFE(nso, &nss->nss_socklist, nso_link, nsonext) { + TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link); + nss->nss_sockcnt--; + nfs_socket_destroy(nso); } + if (nss->nss_sock) { + nfs_socket_destroy(nss->nss_sock); + nss->nss_sock = NULL; + } +} - tsleep((caddr_t)&brreq, PSOCK, "nfsbindresv", 0); - - return (brreq.brr_error); +/* + * Prefer returning certain errors over others. + * This function returns a ranking of the given error. + */ +int +nfs_connect_error_class(int error) +{ + switch (error) { + case 0: + return (0); + case ETIMEDOUT: + case EAGAIN: + return (1); + case EPIPE: + case EADDRNOTAVAIL: + case ENETDOWN: + case ENETUNREACH: + case ENETRESET: + case ECONNABORTED: + case ECONNRESET: + case EISCONN: + case ENOTCONN: + case ESHUTDOWN: + case ECONNREFUSED: + case EHOSTDOWN: + case EHOSTUNREACH: + return (2); + case ERPCMISMATCH: + case EPROCUNAVAIL: + case EPROGMISMATCH: + case EPROGUNAVAIL: + return (3); + case EBADRPC: + return (4); + default: + return (5); + } } /* - * Initialize sockets and congestion for a new NFS connection. - * We do not free the sockaddr if error. + * Make sure a socket search returns the best error. + */ +void +nfs_socket_search_update_error(struct nfs_socket_search *nss, int error) +{ + if (nfs_connect_error_class(error) >= nfs_connect_error_class(nss->nss_error)) + nss->nss_error = error; +} + +/* nfs_connect_search_new_socket: + * Given a socket search structure for an nfs mount try to find a new socket from the set of addresses specified + * by nss. + * + * nss_last is set to -1 at initialization to indicate the first time. Its set to -2 if address was found but + * could not be used or if a socket timed out. */ int -nfs_connect( - struct nfsmount *nmp, - __unused struct nfsreq *rep) +nfs_connect_search_new_socket(struct nfsmount *nmp, struct nfs_socket_search *nss, struct timeval *now) { - socket_t so; - int error, rcvreserve, sndreserve; - struct sockaddr *saddr; - struct timeval timeo; + struct nfs_fs_location *fsl; + struct nfs_fs_server *fss; + struct sockaddr_storage ss; + struct nfs_socket *nso; + char *addrstr; + int error = 0; + - nmp->nm_so = 0; - saddr = mbuf_data(nmp->nm_nam); - error = sock_socket(saddr->sa_family, nmp->nm_sotype, - nmp->nm_soproto, 0, 0, &nmp->nm_so); - if (error) { - goto bad; - } - so = nmp->nm_so; + NFS_SOCK_DBG("nfs connect %s nss_addrcnt = %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss->nss_addrcnt); /* - * Some servers require that the client port be a reserved port number. + * while there are addresses and: + * we have no sockets or + * the last address failed and did not produce a socket (nss_last < 0) or + * Its been a while (2 seconds) and we have less than the max number of concurrent sockets to search (4) + * then attempt to create a socket with the current address. */ - if (saddr->sa_family == AF_INET && (nmp->nm_flag & NFSMNT_RESVPORT)) { - proc_t p; - /* - * sobind() requires current_proc() to have superuser privs. - * If this bind is part of a reconnect, and the current proc - * doesn't have superuser privs, we hand the sobind() off to - * a kernel thread to process. - */ - if ((nmp->nm_state & NFSSTA_MOUNTED) && - (p = current_proc()) && suser(kauth_cred_get(), 0)) { - /* request nfs_bind_resv_thread() to do bind */ - error = nfs_bind_resv_nopriv(nmp); - } else { - error = nfs_bind_resv(nmp); + while (nss->nss_addrcnt > 0 && ((nss->nss_last < 0) || (nss->nss_sockcnt == 0) || + ((nss->nss_sockcnt < 4) && (now->tv_sec >= (nss->nss_last + 2))))) { + if (nmp->nm_sockflags & NMSOCK_UNMOUNT) + return (EINTR); + /* Can we convert the address to a sockaddr? */ + fsl = nmp->nm_locations.nl_locations[nss->nss_nextloc.nli_loc]; + fss = fsl->nl_servers[nss->nss_nextloc.nli_serv]; + addrstr = fss->ns_addresses[nss->nss_nextloc.nli_addr]; + if (!nfs_uaddr2sockaddr(addrstr, (struct sockaddr*)&ss)) { + nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc); + nss->nss_addrcnt -= 1; + nss->nss_last = -2; + continue; + } + /* Check that socket family is acceptable. */ + if (nmp->nm_sofamily && (ss.ss_family != nmp->nm_sofamily)) { + nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc); + nss->nss_addrcnt -= 1; + nss->nss_last = -2; + continue; } + + /* Create the socket. */ + error = nfs_socket_create(nmp, (struct sockaddr*)&ss, nss->nss_sotype, + nss->nss_port, nss->nss_protocol, nss->nss_version, + ((nss->nss_protocol == NFS_PROG) && NMFLAG(nmp, RESVPORT)), &nso); if (error) - goto bad; - } + return (error); - /* - * Protocols that do not require connections may be optionally left - * unconnected for servers that reply from a port other than NFS_PORT. - */ - if (nmp->nm_flag & NFSMNT_NOCONN) { - if (nmp->nm_sotype == SOCK_STREAM) { - error = ENOTCONN; - goto bad; - } - } else { - struct timeval tv; - tv.tv_sec = 2; - tv.tv_usec = 0; - error = sock_connect(so, mbuf_data(nmp->nm_nam), MSG_DONTWAIT); - if (error && error != EINPROGRESS) { - goto bad; + nso->nso_location = nss->nss_nextloc; + nso->nso_wake = nss; + error = sock_setupcall(nso->nso_so, nfs_connect_upcall, nso); + if (error) { + lck_mtx_lock(&nso->nso_lock); + nso->nso_error = error; + nso->nso_flags |= NSO_DEAD; + lck_mtx_unlock(&nso->nso_lock); } + + TAILQ_INSERT_TAIL(&nss->nss_socklist, nso, nso_link); + nss->nss_sockcnt++; + nfs_location_next(&nmp->nm_locations, &nss->nss_nextloc); + nss->nss_addrcnt -= 1; - while ((error = sock_connectwait(so, &tv)) == EINPROGRESS) { - if (rep && (error = nfs_sigintr(nmp, rep, rep->r_procp))) { - goto bad; - } - } + nss->nss_last = now->tv_sec; } + + if (nss->nss_addrcnt == 0 && nss->nss_last < 0) + nss->nss_last = now->tv_sec; - /* - * Always time out on recieve, this allows us to reconnect the - * socket to deal with network changes. - */ - timeo.tv_usec = 0; - timeo.tv_sec = 2; - error = sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo)); - if (nmp->nm_flag & (NFSMNT_SOFT | NFSMNT_INT)) { - timeo.tv_sec = 5; - } else { - timeo.tv_sec = 0; - } - error = sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo)); - - if (nmp->nm_sotype == SOCK_DGRAM) { - sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3; - rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * - (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); - } else if (nmp->nm_sotype == SOCK_SEQPACKET) { - sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR) * 3; - rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR) * - (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); - } else { - int proto; - int on = 1; - - sock_gettype(so, NULL, NULL, &proto); - if (nmp->nm_sotype != SOCK_STREAM) - panic("nfscon sotype"); + return (error); +} - // Assume that SOCK_STREAM always requires a connection - sock_setsockopt(so, SOL_SOCKET, SO_KEEPALIVE, &on, sizeof(on)); - - if (proto == IPPROTO_TCP) { - sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); +/* + * nfs_connect_search_socket_connect: Connect an nfs socket nso for nfsmount nmp. + * If successful set the socket options for the socket as require from the mount. + * + * Assumes: nso->nso_lock is held on entry and return. + */ +int +nfs_connect_search_socket_connect(struct nfsmount *nmp, struct nfs_socket *nso, int verbose) +{ + int error; + + if ((nso->nso_sotype != SOCK_STREAM) && NMFLAG(nmp, NOCONNECT)) { + /* no connection needed, just say it's already connected */ + NFS_SOCK_DBG("nfs connect %s UDP socket %p noconnect\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + nso->nso_flags |= NSO_CONNECTED; + nfs_socket_options(nmp, nso); + return (1); /* Socket is connected and setup */ + } else if (!(nso->nso_flags & NSO_CONNECTING)) { + /* initiate the connection */ + nso->nso_flags |= NSO_CONNECTING; + lck_mtx_unlock(&nso->nso_lock); + NFS_SOCK_DBG("nfs connect %s connecting socket %p\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + error = sock_connect(nso->nso_so, nso->nso_saddr, MSG_DONTWAIT); + lck_mtx_lock(&nso->nso_lock); + if (error && (error != EINPROGRESS)) { + nso->nso_error = error; + nso->nso_flags |= NSO_DEAD; + return (0); } - - sndreserve = (nmp->nm_wsize + NFS_MAXPKTHDR + sizeof (u_long)) * 3; - rcvreserve = (nmp->nm_rsize + NFS_MAXPKTHDR + sizeof (u_long)) * - (nmp->nm_readahead > 0 ? nmp->nm_readahead + 1 : 2); - } - - if (sndreserve > NFS_MAXSOCKBUF) - sndreserve = NFS_MAXSOCKBUF; - if (rcvreserve > NFS_MAXSOCKBUF) - rcvreserve = NFS_MAXSOCKBUF; - error = sock_setsockopt(so, SOL_SOCKET, SO_SNDBUF, &sndreserve, sizeof(sndreserve)); - if (error) { - goto bad; } - error = sock_setsockopt(so, SOL_SOCKET, SO_RCVBUF, &rcvreserve, sizeof(rcvreserve)); - if (error) { - goto bad; + if (nso->nso_flags & NSO_CONNECTING) { + /* check the connection */ + if (sock_isconnected(nso->nso_so)) { + NFS_SOCK_DBG("nfs connect %s socket %p is connected\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + nso->nso_flags &= ~NSO_CONNECTING; + nso->nso_flags |= NSO_CONNECTED; + nfs_socket_options(nmp, nso); + return (1); /* Socket is connected and setup */ + } else { + int optlen = sizeof(error); + error = 0; + sock_getsockopt(nso->nso_so, SOL_SOCKET, SO_ERROR, &error, &optlen); + if (error) { /* we got an error on the socket */ + NFS_SOCK_DBG("nfs connect %s socket %p connection error %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error); + if (verbose) + printf("nfs connect socket error %d for %s\n", + error, vfs_statfs(nmp->nm_mountp)->f_mntfromname); + nso->nso_error = error; + nso->nso_flags |= NSO_DEAD; + return (0); + } + } } - - sock_nointerrupt(so, 1); - - /* Initialize other non-zero congestion variables */ - nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = - nmp->nm_srtt[3] = (NFS_TIMEO << 3); - nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = - nmp->nm_sdrtt[3] = 0; - nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ - nmp->nm_sent = 0; - FSDBG(529, nmp, nmp->nm_state, nmp->nm_soflags, nmp->nm_cwnd); - nmp->nm_timeouts = 0; - return (0); - -bad: - nfs_disconnect(nmp); - return (error); + + return (0); /* Waiting to be connected */ } /* - * Reconnect routine: - * Called when a connection is broken on a reliable protocol. - * - clean up the old socket - * - nfs_connect() again - * - set R_MUSTRESEND for all outstanding requests on mount point - * If this fails the mount point is DEAD! - * nb: Must be called with the nfs_sndlock() set on the mount point. + * nfs_connect_search_ping: Send a null proc on the nso socket. */ -static int -nfs_reconnect(struct nfsreq *rep) +int +nfs_connect_search_ping(struct nfsmount *nmp, struct nfs_socket *nso, struct timeval *now) { - struct nfsreq *rp; - struct nfsmount *nmp = rep->r_nmp; + /* initiate a NULL RPC request */ + uint64_t xid = nso->nso_pingxid; + mbuf_t m, mreq = NULL; + struct msghdr msg; + size_t reqlen, sentlen; + uint32_t vers = nso->nso_version; int error; - nfs_disconnect(nmp); - while ((error = nfs_connect(nmp, rep))) { - if (error == EINTR || error == ERESTART) - return (EINTR); - if (error == EIO) - return (EIO); - nfs_down(rep->r_nmp, rep->r_procp, error, NFSSTA_TIMEO, - "can not connect"); - rep->r_flags |= R_TPRINTFMSG; - if (!(nmp->nm_state & NFSSTA_MOUNTED)) { - /* we're not yet completely mounted and */ - /* we can't reconnect, so we fail */ - return (error); + if (!vers) { + if (nso->nso_protocol == PMAPPROG) + vers = (nso->nso_saddr->sa_family == AF_INET) ? PMAPVERS : RPCBVERS4; + else if (nso->nso_protocol == NFS_PROG) + vers = NFS_VER3; + } + lck_mtx_unlock(&nso->nso_lock); + error = nfsm_rpchead2(nmp, nso->nso_sotype, nso->nso_protocol, vers, 0, RPCAUTH_SYS, + vfs_context_ucred(vfs_context_kernel()), NULL, NULL, &xid, &mreq); + lck_mtx_lock(&nso->nso_lock); + if (!error) { + nso->nso_flags |= NSO_PINGING; + nso->nso_pingxid = R_XID32(xid); + nso->nso_reqtimestamp = now->tv_sec; + bzero(&msg, sizeof(msg)); + if ((nso->nso_sotype != SOCK_STREAM) && !sock_isconnected(nso->nso_so)) { + msg.msg_name = nso->nso_saddr; + msg.msg_namelen = nso->nso_saddr->sa_len; } - if ((error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp))) - return (error); - tsleep((caddr_t)&lbolt, PSOCK, "nfscon", 0); + for (reqlen=0, m=mreq; m; m = mbuf_next(m)) + reqlen += mbuf_len(m); + lck_mtx_unlock(&nso->nso_lock); + error = sock_sendmbuf(nso->nso_so, &msg, mreq, 0, &sentlen); + NFS_SOCK_DBG("nfs connect %s verifying socket %p send rv %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error); + lck_mtx_lock(&nso->nso_lock); + if (!error && (sentlen != reqlen)) + error = ETIMEDOUT; } - - /* - * Loop through outstanding request list and fix up all requests - * on old socket. - */ - TAILQ_FOREACH(rp, &nfs_reqq, r_chain) { - if (rp->r_nmp == nmp) - rp->r_flags |= R_MUSTRESEND; + if (error) { + nso->nso_error = error; + nso->nso_flags |= NSO_DEAD; + return (0); } - return (0); + + return (1); } /* - * NFS disconnect. Clean up and unlink. + * nfs_connect_search_socket_found: Take the found socket of the socket search list and assign it to the searched socket. + * Set the nfs socket protocol and version if needed. */ void -nfs_disconnect(struct nfsmount *nmp) +nfs_connect_search_socket_found(struct nfsmount *nmp __unused, struct nfs_socket_search *nss, struct nfs_socket *nso) { - socket_t so; - - if (nmp->nm_so) { - so = nmp->nm_so; - nmp->nm_so = 0; - sock_shutdown(so, 2); - sock_close(so); + NFS_SOCK_DBG("nfs connect %s socket %p verified\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + if (!nso->nso_version) { + /* If the version isn't set, the default must have worked. */ + if (nso->nso_protocol == PMAPPROG) + nso->nso_version = (nso->nso_saddr->sa_family == AF_INET) ? PMAPVERS : RPCBVERS4; + if (nso->nso_protocol == NFS_PROG) + nso->nso_version = NFS_VER3; } + TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link); + nss->nss_sockcnt--; + nss->nss_sock = nso; } /* - * This is the nfs send routine. For connection based socket types, it - * must be called with an nfs_sndlock() on the socket. - * "rep == NULL" indicates that it has been called from a server. - * For the client side: - * - return EINTR if the RPC is terminated, 0 otherwise - * - set R_MUSTRESEND if the send fails for any reason - * - do any cleanup required by recoverable socket errors (???) - * For the server side: - * - return EINTR or ERESTART if interrupted by a signal - * - return EPIPE if a connection is lost for connection based sockets (TCP...) - * - do any cleanup required by recoverable socket errors (???) + * nfs_connect_search_socket_reap: For each socket in the search list mark any timed out socket as dead and remove from + * the list. Dead socket are then destroyed. */ -int -nfs_send(so, nam, top, rep) - socket_t so; - mbuf_t nam; - mbuf_t top; - struct nfsreq *rep; +void +nfs_connect_search_socket_reap(struct nfsmount *nmp __unused, struct nfs_socket_search *nss, struct timeval *now) { - struct sockaddr *sendnam; - int error, error2, sotype, flags; - u_long xidqueued = 0; - struct nfsreq *rp; - char savenametolog[MAXPATHLEN]; - struct msghdr msg; + struct nfs_socket *nso, *nsonext; - if (rep) { - error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp); - if (error) { - mbuf_freem(top); - return (error); + TAILQ_FOREACH_SAFE(nso, &nss->nss_socklist, nso_link, nsonext) { + lck_mtx_lock(&nso->nso_lock); + if (now->tv_sec >= (nso->nso_timestamp + nss->nss_timeo)) { + /* took too long */ + NFS_SOCK_DBG("nfs connect %s socket %p timed out\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + nso->nso_error = ETIMEDOUT; + nso->nso_flags |= NSO_DEAD; } - if ((so = rep->r_nmp->nm_so) == NULL) { - rep->r_flags |= R_MUSTRESEND; - mbuf_freem(top); - return (0); + if (!(nso->nso_flags & NSO_DEAD)) { + lck_mtx_unlock(&nso->nso_lock); + continue; } - rep->r_flags &= ~R_MUSTRESEND; - TAILQ_FOREACH(rp, &nfs_reqq, r_chain) - if (rp == rep) - break; - if (rp) - xidqueued = rp->r_xid; + lck_mtx_unlock(&nso->nso_lock); + NFS_SOCK_DBG("nfs connect %s reaping socket %p %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, nso->nso_error); + nfs_socket_search_update_error(nss, nso->nso_error); + TAILQ_REMOVE(&nss->nss_socklist, nso, nso_link); + nss->nss_sockcnt--; + nfs_socket_destroy(nso); + /* If there are more sockets to try, force the starting of another socket */ + if (nss->nss_addrcnt > 0) + nss->nss_last = -2; } - sock_gettype(so, NULL, &sotype, NULL); - if ((sotype == SOCK_STREAM) || (sock_isconnected(so)) || - (nam == 0)) - sendnam = (struct sockaddr *)0; - else - sendnam = mbuf_data(nam); +} - if (sotype == SOCK_SEQPACKET) - flags = MSG_EOR; - else - flags = 0; +/* + * nfs_connect_search_check: Check on the status of search and wait for replies if needed. + */ +int +nfs_connect_search_check(struct nfsmount *nmp, struct nfs_socket_search *nss, struct timeval *now) +{ + int error; - /* - * Save the name here in case mount point goes away if we block. - * The name is using local stack and is large, but don't - * want to block if we malloc. - */ - if (rep) - strncpy(savenametolog, - vfs_statfs(rep->r_nmp->nm_mountp)->f_mntfromname, - MAXPATHLEN - 1); - bzero(&msg, sizeof(msg)); - msg.msg_name = (caddr_t)sendnam; - msg.msg_namelen = sendnam == 0 ? 0 : sendnam->sa_len; - error = sock_sendmbuf(so, &msg, top, flags, NULL); + /* log a warning if connect is taking a while */ + if (((now->tv_sec - nss->nss_timestamp) >= 8) && ((nss->nss_flags & (NSS_VERBOSE|NSS_WARNED)) == NSS_VERBOSE)) { + printf("nfs_connect: socket connect taking a while for %s\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname); + nss->nss_flags |= NSS_WARNED; + } + if (nmp->nm_sockflags & NMSOCK_UNMOUNT) + return (EINTR); + if ((error = nfs_sigintr(nmp, NULL, current_thread(), 0))) + return (error); + + /* If we were succesfull at sending a ping, wait up to a second for a reply */ + if (nss->nss_last >= 0) + tsleep(nss, PSOCK, "nfs_connect_search_wait", hz); + + return (0); +} + + +/* + * Continue the socket search until we have something to report. + */ +int +nfs_connect_search_loop(struct nfsmount *nmp, struct nfs_socket_search *nss) +{ + struct nfs_socket *nso; + struct timeval now; + int error; + int verbose = (nss->nss_flags & NSS_VERBOSE); + +loop: + microuptime(&now); + NFS_SOCK_DBG("nfs connect %s search %ld\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, now.tv_sec); + /* add a new socket to the socket list if needed and available */ + error = nfs_connect_search_new_socket(nmp, nss, &now); if (error) { - if (rep) { - if (xidqueued) { - TAILQ_FOREACH(rp, &nfs_reqq, r_chain) - if (rp == rep && rp->r_xid == xidqueued) - break; - if (!rp) - panic("nfs_send: error %d xid %x gone", - error, xidqueued); + NFS_SOCK_DBG("nfs connect returned %d\n", error); + return (error); + } + + /* check each active socket on the list and try to push it along */ + TAILQ_FOREACH(nso, &nss->nss_socklist, nso_link) { + lck_mtx_lock(&nso->nso_lock); + + /* If not connected connect it */ + if (!(nso->nso_flags & NSO_CONNECTED)) { + if (!nfs_connect_search_socket_connect(nmp, nso, verbose)) { + lck_mtx_unlock(&nso->nso_lock); + continue; } - log(LOG_INFO, "nfs send error %d for server %s\n", - error, savenametolog); - /* - * Deal with errors for the client side. - */ - error2 = nfs_sigintr(rep->r_nmp, rep, rep->r_procp); - if (error2) { - error = error2; - } else { - rep->r_flags |= R_MUSTRESEND; + } + + /* If the socket hasn't been verified or in a ping, ping it. We also handle UDP retransmits */ + if (!(nso->nso_flags & (NSO_PINGING|NSO_VERIFIED)) || + ((nso->nso_sotype == SOCK_DGRAM) && (now.tv_sec >= nso->nso_reqtimestamp+2))) { + if (!nfs_connect_search_ping(nmp, nso, &now)) { + lck_mtx_unlock(&nso->nso_lock); + continue; } - } else - log(LOG_INFO, "nfsd send error %d\n", error); + } - /* - * Handle any recoverable (soft) socket errors here. (???) - */ - if (error != EINTR && error != ERESTART && error != EIO && - error != EWOULDBLOCK && error != EPIPE) { - error = 0; + /* Has the socket been verified by the up call routine? */ + if (nso->nso_flags & NSO_VERIFIED) { + /* WOOHOO!! This socket looks good! */ + nfs_connect_search_socket_found(nmp, nss, nso); + lck_mtx_unlock(&nso->nso_lock); + break; } + lck_mtx_unlock(&nso->nso_lock); } + + /* Check for timed out sockets and mark as dead and then remove all dead sockets. */ + nfs_connect_search_socket_reap(nmp, nss, &now); + + /* + * Keep looping if we haven't found a socket yet and we have more + * sockets to (continue to) try. + */ + error = 0; + if (!nss->nss_sock && (!TAILQ_EMPTY(&nss->nss_socklist) || nss->nss_addrcnt)) { + error = nfs_connect_search_check(nmp, nss, &now); + if (!error) + goto loop; + } + + NFS_SOCK_DBG("nfs connect %s returning %d\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, error); return (error); } /* - * Receive a Sun RPC Request/Reply. For SOCK_DGRAM, the work is all - * done by soreceive(), but for SOCK_STREAM we must deal with the Record - * Mark and consolidate the data into a new mbuf list. - * nb: Sometimes TCP passes the data up to soreceive() in long lists of - * small mbufs. - * For SOCK_STREAM we must be very careful to read an entire record once - * we have read any of it, even if the system call has been interrupted. + * Initialize a new NFS connection. + * + * Search for a location to connect a socket to and initialize the connection. + * + * An NFS mount may have multiple locations/servers/addresses available. + * We attempt to connect to each one asynchronously and will start + * several sockets in parallel if other locations are slow to answer. + * We'll use the first NFS socket we can successfully set up. + * + * The search may involve contacting the portmapper service first. + * + * A mount's initial connection may require negotiating some parameters such + * as socket type and NFS version. */ -static int -nfs_receive(struct nfsreq *rep, mbuf_t *mp) -{ - socket_t so; - struct iovec_32 aio; - mbuf_t m, mlast; - u_long len, fraglen; - int error, error2, sotype; - proc_t p = current_proc(); /* XXX */ - struct msghdr msg; - size_t rcvlen; - int lastfragment; +int +nfs_connect(struct nfsmount *nmp, int verbose, int timeo) +{ + struct nfs_socket_search nss; + struct nfs_socket *nso, *nsonfs; + struct sockaddr_storage ss; + struct sockaddr *saddr, *oldsaddr; + sock_upcall upcall; + struct timeval now, start; + int error, savederror, nfsvers; + uint8_t sotype = nmp->nm_sotype ? nmp->nm_sotype : SOCK_STREAM; + fhandle_t *fh = NULL; + char *path = NULL; + in_port_t port; + int addrtotal = 0; + + /* paranoia... check that we have at least one address in the locations */ + uint32_t loc, serv; + for (loc=0; loc < nmp->nm_locations.nl_numlocs; loc++) { + for (serv=0; serv < nmp->nm_locations.nl_locations[loc]->nl_servcount; serv++) { + addrtotal += nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_addrcount; + if (nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_addrcount == 0) + NFS_SOCK_DBG("nfs connect %s search, server %s has no addresses\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, + nmp->nm_locations.nl_locations[loc]->nl_servers[serv]->ns_name); + } + } - /* - * Set up arguments for soreceive() - */ - *mp = NULL; - sotype = rep->r_nmp->nm_sotype; + if (addrtotal == 0) { + NFS_SOCK_DBG("nfs connect %s search failed, no addresses\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname); + return (EINVAL); + } else + NFS_SOCK_DBG("nfs connect %s has %d addresses\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, addrtotal); + + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags |= NMSOCK_CONNECTING; + nmp->nm_nss = &nss; + lck_mtx_unlock(&nmp->nm_lock); + microuptime(&start); + savederror = error = 0; - /* - * For reliable protocols, lock against other senders/receivers - * in case a reconnect is necessary. - * For SOCK_STREAM, first get the Record Mark to find out how much - * more there is to get. - * We must lock the socket against other receivers - * until we have an entire rpc request/reply. - */ - if (sotype != SOCK_DGRAM) { - error = nfs_sndlock(rep); - if (error) - return (error); tryagain: - /* - * Check for fatal errors and resending request. - */ - /* - * Ugh: If a reconnect attempt just happened, nm_so - * would have changed. NULL indicates a failed - * attempt that has essentially shut down this - * mount point. - */ - if ((error = nfs_sigintr(rep->r_nmp, rep, p)) || rep->r_mrep) { - nfs_sndunlock(rep); - if (error) - return (error); - return (EINTR); - } - so = rep->r_nmp->nm_so; - if (!so) { - error = nfs_reconnect(rep); - if (error) { - nfs_sndunlock(rep); - return (error); + /* initialize socket search state */ + bzero(&nss, sizeof(nss)); + nss.nss_addrcnt = addrtotal; + nss.nss_error = savederror; + TAILQ_INIT(&nss.nss_socklist); + nss.nss_sotype = sotype; + nss.nss_startloc = nmp->nm_locations.nl_current; + nss.nss_timestamp = start.tv_sec; + nss.nss_timeo = timeo; + if (verbose) + nss.nss_flags |= NSS_VERBOSE; + + /* First time connecting, we may need to negotiate some things */ + if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) { + if (!nmp->nm_vers) { + /* No NFS version specified... */ + if (!nmp->nm_nfsport || (!NM_OMATTR_GIVEN(nmp, FH) && !nmp->nm_mountport)) { + /* ...connect to portmapper first if we (may) need any ports. */ + nss.nss_port = PMAPPORT; + nss.nss_protocol = PMAPPROG; + nss.nss_version = 0; + } else { + /* ...connect to NFS port first. */ + nss.nss_port = nmp->nm_nfsport; + nss.nss_protocol = NFS_PROG; + nss.nss_version = 0; } + } else if (nmp->nm_vers >= NFS_VER4) { + /* For NFSv4, we use the given (or default) port. */ + nss.nss_port = nmp->nm_nfsport ? nmp->nm_nfsport : NFS_PORT; + nss.nss_protocol = NFS_PROG; + nss.nss_version = 4; + } else { + /* For NFSv3/v2... */ + if (!nmp->nm_nfsport || (!NM_OMATTR_GIVEN(nmp, FH) && !nmp->nm_mountport)) { + /* ...connect to portmapper first if we need any ports. */ + nss.nss_port = PMAPPORT; + nss.nss_protocol = PMAPPROG; + nss.nss_version = 0; + } else { + /* ...connect to NFS port first. */ + nss.nss_port = nmp->nm_nfsport; + nss.nss_protocol = NFS_PROG; + nss.nss_version = nmp->nm_vers; + } + } + NFS_SOCK_DBG("nfs connect first %s, so type %d port %d prot %d %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss.nss_sotype, nss.nss_port, + nss.nss_protocol, nss.nss_version); + } else { + /* we've connected before, just connect to NFS port */ + if (!nmp->nm_nfsport) { + /* need to ask portmapper which port that would be */ + nss.nss_port = PMAPPORT; + nss.nss_protocol = PMAPPROG; + nss.nss_version = 0; + } else { + nss.nss_port = nmp->nm_nfsport; + nss.nss_protocol = NFS_PROG; + nss.nss_version = nmp->nm_vers; + } + NFS_SOCK_DBG("nfs connect %s, so type %d port %d prot %d %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nss.nss_sotype, nss.nss_port, + nss.nss_protocol, nss.nss_version); + } + + /* Set next location to first valid location. */ + /* If start location is invalid, find next location. */ + nss.nss_nextloc = nss.nss_startloc; + if ((nss.nss_nextloc.nli_serv >= nmp->nm_locations.nl_locations[nss.nss_nextloc.nli_loc]->nl_servcount) || + (nss.nss_nextloc.nli_addr >= nmp->nm_locations.nl_locations[nss.nss_nextloc.nli_loc]->nl_servers[nss.nss_nextloc.nli_serv]->ns_addrcount)) { + nfs_location_next(&nmp->nm_locations, &nss.nss_nextloc); + if (!nfs_location_index_cmp(&nss.nss_nextloc, &nss.nss_startloc)) { + NFS_SOCK_DBG("nfs connect %s search failed, couldn't find a valid location index\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname); + return (ENOENT); + } + } + nss.nss_last = -1; + +keepsearching: + + error = nfs_connect_search_loop(nmp, &nss); + if (error || !nss.nss_sock) { + /* search failed */ + nfs_socket_search_cleanup(&nss); + if (!error && (nss.nss_sotype == SOCK_STREAM) && !nmp->nm_sotype && (nmp->nm_vers < NFS_VER4)) { + /* Try using UDP */ + sotype = SOCK_DGRAM; + savederror = nss.nss_error; + NFS_SOCK_DBG("nfs connect %s TCP failed %d %d, trying UDP\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, error, nss.nss_error); goto tryagain; } - while (rep->r_flags & R_MUSTRESEND) { - error = mbuf_copym(rep->r_mreq, 0, MBUF_COPYALL, MBUF_WAITOK, &m); + if (!error) + error = nss.nss_error ? nss.nss_error : ETIMEDOUT; + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags &= ~NMSOCK_CONNECTING; + nmp->nm_nss = NULL; + lck_mtx_unlock(&nmp->nm_lock); + if (nss.nss_flags & NSS_WARNED) + log(LOG_INFO, "nfs_connect: socket connect aborted for %s\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname); + if (fh) + FREE(fh, M_TEMP); + if (path) + FREE_ZONE(path, MAXPATHLEN, M_NAMEI); + NFS_SOCK_DBG("nfs connect %s search failed, returning %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, error); + return (error); + } + + /* try to use nss_sock */ + nso = nss.nss_sock; + nss.nss_sock = NULL; + + /* We may be speaking to portmap first... to determine port(s). */ + if (nso->nso_saddr->sa_family == AF_INET) + port = ntohs(((struct sockaddr_in*)nso->nso_saddr)->sin_port); + else + port = ntohs(((struct sockaddr_in6*)nso->nso_saddr)->sin6_port); + if (port == PMAPPORT) { + /* Use this portmapper port to get the port #s we need. */ + NFS_SOCK_DBG("nfs connect %s got portmapper socket %p\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + + /* remove the connect upcall so nfs_portmap_lookup() can use this socket */ + sock_setupcall(nso->nso_so, NULL, NULL); + + /* Set up socket address and port for NFS socket. */ + bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len); + + /* If NFS version not set, try NFSv3 then NFSv2. */ + nfsvers = nmp->nm_vers ? nmp->nm_vers : NFS_VER3; + + if (!(port = nmp->nm_nfsport)) { + if (ss.ss_family == AF_INET) + ((struct sockaddr_in*)&ss)->sin_port = htons(0); + else if (ss.ss_family == AF_INET6) + ((struct sockaddr_in6*)&ss)->sin6_port = htons(0); + error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss, + nso->nso_so, NFS_PROG, nfsvers, + (nso->nso_sotype == SOCK_DGRAM) ? IPPROTO_UDP : IPPROTO_TCP, timeo); if (!error) { - OSAddAtomic(1, (SInt32*)&nfsstats.rpcretries); - error = nfs_send(so, rep->r_nmp->nm_nam, m, rep); + if (ss.ss_family == AF_INET) + port = ntohs(((struct sockaddr_in*)&ss)->sin_port); + else if (ss.ss_family == AF_INET6) + port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port); + if (!port) + error = EPROGUNAVAIL; } - /* - * we also hold rcv lock so rep is still - * legit this point - */ - if (error) { - if (error == EINTR || error == ERESTART || - (error = nfs_reconnect(rep))) { - nfs_sndunlock(rep); - return (error); + if (error && !nmp->nm_vers) { + nfsvers = NFS_VER2; + error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss, + nso->nso_so, NFS_PROG, nfsvers, + (nso->nso_sotype == SOCK_DGRAM) ? IPPROTO_UDP : IPPROTO_TCP, timeo); + if (!error) { + if (ss.ss_family == AF_INET) + port = ntohs(((struct sockaddr_in*)&ss)->sin_port); + else if (ss.ss_family == AF_INET6) + port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port); + if (!port) + error = EPROGUNAVAIL; } - goto tryagain; + } + if (error) { + nfs_socket_search_update_error(&nss, error); + nfs_socket_destroy(nso); + goto keepsearching; } } - nfs_sndunlock(rep); - if (sotype == SOCK_STREAM) { + /* Create NFS protocol socket and add it to the list of sockets. */ + error = nfs_socket_create(nmp, (struct sockaddr*)&ss, nso->nso_sotype, port, + NFS_PROG, nfsvers, NMFLAG(nmp, RESVPORT), &nsonfs); + if (error) { + nfs_socket_search_update_error(&nss, error); + nfs_socket_destroy(nso); + goto keepsearching; + } + nsonfs->nso_location = nso->nso_location; + nsonfs->nso_wake = &nss; + error = sock_setupcall(nsonfs->nso_so, nfs_connect_upcall, nsonfs); + if (error) { + nfs_socket_search_update_error(&nss, error); + nfs_socket_destroy(nsonfs); + nfs_socket_destroy(nso); + goto keepsearching; + } + TAILQ_INSERT_TAIL(&nss.nss_socklist, nsonfs, nso_link); + nss.nss_sockcnt++; + if ((nfsvers < NFS_VER4) && !(nmp->nm_sockflags & NMSOCK_HASCONNECTED) && !NM_OMATTR_GIVEN(nmp, FH)) { + /* Set up socket address and port for MOUNT socket. */ error = 0; - len = 0; - lastfragment = 0; - mlast = NULL; - while (!error && !lastfragment) { - aio.iov_base = (uintptr_t) &fraglen; - aio.iov_len = sizeof(u_long); - bzero(&msg, sizeof(msg)); - msg.msg_iov = (struct iovec *) &aio; - msg.msg_iovlen = 1; - do { - error = sock_receive(so, &msg, MSG_WAITALL, &rcvlen); - if (!rep->r_nmp) /* if unmounted then bailout */ - goto shutout; - if (error == EWOULDBLOCK && rep) { - error2 = nfs_sigintr(rep->r_nmp, rep, p); - if (error2) - error = error2; - } - } while (error == EWOULDBLOCK); - if (!error && rcvlen < aio.iov_len) { - /* only log a message if we got a partial word */ - if (rcvlen != 0) - log(LOG_INFO, - "short receive (%d/%d) from nfs server %s\n", - rcvlen, sizeof(u_long), - vfs_statfs(rep->r_nmp->nm_mountp)->f_mntfromname); - error = EPIPE; - } - if (error) - goto errout; - lastfragment = ntohl(fraglen) & 0x80000000; - fraglen = ntohl(fraglen) & ~0x80000000; - len += fraglen; - /* - * This is SERIOUS! We are out of sync with the sender - * and forcing a disconnect/reconnect is all I can do. - */ - if (len > NFS_MAXPACKET) { - log(LOG_ERR, "%s (%d) from nfs server %s\n", - "impossible RPC record length", len, - vfs_statfs(rep->r_nmp->nm_mountp)->f_mntfromname); - error = EFBIG; - goto errout; - } - - m = NULL; - do { - rcvlen = fraglen; - error = sock_receivembuf(so, NULL, &m, MSG_WAITALL, &rcvlen); - if (!rep->r_nmp) /* if unmounted then bailout */ { - goto shutout; - } - } while (error == EWOULDBLOCK || error == EINTR || - error == ERESTART); - - if (!error && fraglen > rcvlen) { - log(LOG_INFO, - "short receive (%d/%d) from nfs server %s\n", - rcvlen, fraglen, - vfs_statfs(rep->r_nmp->nm_mountp)->f_mntfromname); - error = EPIPE; - mbuf_freem(m); - } - if (!error) { - if (!*mp) { - *mp = m; - mlast = m; - } else { - error = mbuf_setnext(mlast, m); - if (error) { - printf("nfs_receive: mbuf_setnext failed %d\n", error); - mbuf_freem(m); - } - } - while (mbuf_next(mlast)) - mlast = mbuf_next(mlast); - } + bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len); + port = nmp->nm_mountport; + if (ss.ss_family == AF_INET) + ((struct sockaddr_in*)&ss)->sin_port = htons(port); + else if (ss.ss_family == AF_INET6) + ((struct sockaddr_in6*)&ss)->sin6_port = htons(port); + if (!port) { + /* Get port/sockaddr for MOUNT version corresponding to NFS version. */ + /* If NFS version is unknown, optimistically choose for NFSv3. */ + int mntvers = (nfsvers == NFS_VER2) ? RPCMNT_VER1 : RPCMNT_VER3; + int mntproto = (NM_OMFLAG(nmp, MNTUDP) || (nso->nso_sotype == SOCK_DGRAM)) ? IPPROTO_UDP : IPPROTO_TCP; + error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss, + nso->nso_so, RPCPROG_MNT, mntvers, mntproto, timeo); } - } else { - bzero(&msg, sizeof(msg)); - do { - rcvlen = 100000000; - error = sock_receivembuf(so, &msg, mp, 0, &rcvlen); - if (!rep->r_nmp) /* if unmounted then bailout */ { - goto shutout; - } - if (error == EWOULDBLOCK && rep) { - error2 = nfs_sigintr(rep->r_nmp, rep, p); - if (error2) { - return (error2); - } - } - } while (error == EWOULDBLOCK); - - if ((msg.msg_flags & MSG_EOR) == 0) - printf("Egad!!\n"); - if (!error && *mp == NULL) - error = EPIPE; - len = rcvlen; - } -errout: - if (error && error != EINTR && error != ERESTART) { - mbuf_freem(*mp); - *mp = NULL; - if (error != EPIPE) - log(LOG_INFO, - "receive error %d from nfs server %s\n", error, - vfs_statfs(rep->r_nmp->nm_mountp)->f_mntfromname); - error = nfs_sndlock(rep); if (!error) { - error = nfs_reconnect(rep); - if (!error) - goto tryagain; - nfs_sndunlock(rep); + if (ss.ss_family == AF_INET) + port = ntohs(((struct sockaddr_in*)&ss)->sin_port); + else if (ss.ss_family == AF_INET6) + port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port); + if (!port) + error = EPROGUNAVAIL; + } + /* create sockaddr for MOUNT */ + if (!error) + MALLOC(nsonfs->nso_saddr2, struct sockaddr *, ss.ss_len, M_SONAME, M_WAITOK|M_ZERO); + if (!error && !nsonfs->nso_saddr2) + error = ENOMEM; + if (!error) + bcopy(&ss, nsonfs->nso_saddr2, ss.ss_len); + if (error) { + lck_mtx_lock(&nsonfs->nso_lock); + nsonfs->nso_error = error; + nsonfs->nso_flags |= NSO_DEAD; + lck_mtx_unlock(&nsonfs->nso_lock); } } - } else { - /* - * We could have failed while rebinding the datagram socket - * so we need to attempt to rebind here. - */ - if ((so = rep->r_nmp->nm_so) == NULL) { - error = nfs_sndlock(rep); + nfs_socket_destroy(nso); + goto keepsearching; + } + + /* nso is an NFS socket */ + NFS_SOCK_DBG("nfs connect %s got NFS socket %p\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso); + + /* If NFS version wasn't specified, it was determined during the connect. */ + nfsvers = nmp->nm_vers ? nmp->nm_vers : (int)nso->nso_version; + + /* Perform MOUNT call for initial NFSv2/v3 connection/mount. */ + if ((nfsvers < NFS_VER4) && !(nmp->nm_sockflags & NMSOCK_HASCONNECTED) && !NM_OMATTR_GIVEN(nmp, FH)) { + error = 0; + saddr = nso->nso_saddr2; + if (!saddr) { + /* Need sockaddr for MOUNT port */ + bcopy(nso->nso_saddr, &ss, nso->nso_saddr->sa_len); + port = nmp->nm_mountport; + if (ss.ss_family == AF_INET) + ((struct sockaddr_in*)&ss)->sin_port = htons(port); + else if (ss.ss_family == AF_INET6) + ((struct sockaddr_in6*)&ss)->sin6_port = htons(port); + if (!port) { + /* Get port/sockaddr for MOUNT version corresponding to NFS version. */ + int mntvers = (nfsvers == NFS_VER2) ? RPCMNT_VER1 : RPCMNT_VER3; + int mntproto = (NM_OMFLAG(nmp, MNTUDP) || (nso->nso_sotype == SOCK_DGRAM)) ? IPPROTO_UDP : IPPROTO_TCP; + error = nfs_portmap_lookup(nmp, vfs_context_current(), (struct sockaddr*)&ss, + NULL, RPCPROG_MNT, mntvers, mntproto, timeo); + if (ss.ss_family == AF_INET) + port = ntohs(((struct sockaddr_in*)&ss)->sin_port); + else if (ss.ss_family == AF_INET6) + port = ntohs(((struct sockaddr_in6*)&ss)->sin6_port); + } if (!error) { - error = nfs_reconnect(rep); - nfs_sndunlock(rep); + if (port) + saddr = (struct sockaddr*)&ss; + else + error = EPROGUNAVAIL; } - if (error) - return (error); - if (!rep->r_nmp) /* if unmounted then bailout */ - return (ENXIO); - so = rep->r_nmp->nm_so; } - bzero(&msg, sizeof(msg)); - len = 0; - do { - rcvlen = 1000000; - error = sock_receivembuf(so, &msg, mp, 0, &rcvlen); - if (!rep->r_nmp) /* if unmounted then bailout */ - goto shutout; - if (error) { - error2 = nfs_sigintr(rep->r_nmp, rep, p); - if (error2) { - error = error2; - goto shutout; - } + if (saddr) + MALLOC(fh, fhandle_t *, sizeof(fhandle_t), M_TEMP, M_WAITOK|M_ZERO); + if (saddr && fh) + MALLOC_ZONE(path, char *, MAXPATHLEN, M_NAMEI, M_WAITOK); + if (!saddr || !fh || !path) { + if (!error) + error = ENOMEM; + if (fh) + FREE(fh, M_TEMP); + if (path) + FREE_ZONE(path, MAXPATHLEN, M_NAMEI); + fh = NULL; + path = NULL; + nfs_socket_search_update_error(&nss, error); + nfs_socket_destroy(nso); + goto keepsearching; + } + nfs_location_mntfromname(&nmp->nm_locations, nso->nso_location, path, MAXPATHLEN, 1); + error = nfs3_mount_rpc(nmp, saddr, nso->nso_sotype, nfsvers, + path, vfs_context_current(), timeo, fh, &nmp->nm_servsec); + NFS_SOCK_DBG("nfs connect %s socket %p mount %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error); + if (!error) { + /* Make sure we can agree on a security flavor. */ + int o, s; /* indices into mount option and server security flavor lists */ + int found = 0; + + if ((nfsvers == NFS_VER3) && !nmp->nm_servsec.count) { + /* Some servers return an empty list to indicate RPCAUTH_SYS? */ + nmp->nm_servsec.count = 1; + nmp->nm_servsec.flavors[0] = RPCAUTH_SYS; } - /* Reconnect for all errors. We may be receiving - * soft/hard/blocking errors because of a network - * change. - * XXX: we should rate limit or delay this - * to once every N attempts or something. - * although TCP doesn't seem to. - */ - if (error) { - error2 = nfs_sndlock(rep); - if (!error2) { - error2 = nfs_reconnect(rep); - if (error2) - error = error2; - else if (!rep->r_nmp) /* if unmounted then bailout */ - error = ENXIO; - else - so = rep->r_nmp->nm_so; - nfs_sndunlock(rep); - } else { - error = error2; + if (nmp->nm_sec.count) { + /* Choose the first flavor in our list that the server supports. */ + if (!nmp->nm_servsec.count) { + /* we don't know what the server supports, just use our first choice */ + nmp->nm_auth = nmp->nm_sec.flavors[0]; + found = 1; + } + for (o=0; !found && (o < nmp->nm_sec.count); o++) + for (s=0; !found && (s < nmp->nm_servsec.count); s++) + if (nmp->nm_sec.flavors[o] == nmp->nm_servsec.flavors[s]) { + nmp->nm_auth = nmp->nm_sec.flavors[o]; + found = 1; + } + } else { + /* Choose the first one we support from the server's list. */ + if (!nmp->nm_servsec.count) { + nmp->nm_auth = RPCAUTH_SYS; + found = 1; } + for (s=0; s < nmp->nm_servsec.count; s++) + switch (nmp->nm_servsec.flavors[s]) { + case RPCAUTH_SYS: + /* prefer RPCAUTH_SYS to RPCAUTH_NONE */ + if (found && (nmp->nm_auth == RPCAUTH_NONE)) + found = 0; + case RPCAUTH_NONE: + case RPCAUTH_KRB5: + case RPCAUTH_KRB5I: + case RPCAUTH_KRB5P: + if (!found) { + nmp->nm_auth = nmp->nm_servsec.flavors[s]; + found = 1; + } + break; + } } - } while (error == EWOULDBLOCK); + error = !found ? EAUTH : 0; + } + FREE_ZONE(path, MAXPATHLEN, M_NAMEI); + path = NULL; + if (error) { + nfs_socket_search_update_error(&nss, error); + FREE(fh, M_TEMP); + fh = NULL; + nfs_socket_destroy(nso); + goto keepsearching; + } + if (nmp->nm_fh) + FREE(nmp->nm_fh, M_TEMP); + nmp->nm_fh = fh; + fh = NULL; + NFS_BITMAP_SET(nmp->nm_flags, NFS_MFLAG_CALLUMNT); } -shutout: + + /* put the real upcall in place */ + upcall = (nso->nso_sotype == SOCK_STREAM) ? nfs_tcp_rcv : nfs_udp_rcv; + error = sock_setupcall(nso->nso_so, upcall, nmp); if (error) { - mbuf_freem(*mp); - *mp = NULL; + nfs_socket_search_update_error(&nss, error); + nfs_socket_destroy(nso); + goto keepsearching; } - return (error); -} - -/* - * Implement receipt of reply on a socket. - * We must search through the list of received datagrams matching them - * with outstanding requests using the xid, until ours is found. - */ -/* ARGSUSED */ -int -nfs_reply(myrep) - struct nfsreq *myrep; -{ - struct nfsreq *rep; - struct nfsmount *nmp = myrep->r_nmp; - long t1; - mbuf_t mrep, md; - u_long rxid, *tl; - caddr_t dpos, cp2; - int error; - /* - * Loop around until we get our own reply - */ - for (;;) { - /* - * Lock against other receivers so that I don't get stuck in - * sbwait() after someone else has received my reply for me. - * Also necessary for connection based protocols to avoid - * race conditions during a reconnect. - * If nfs_rcvlock() returns EALREADY, that means that - * the reply has already been recieved by another - * process and we can return immediately. In this - * case, the lock is not taken to avoid races with - * other processes. - */ - error = nfs_rcvlock(myrep); - if (error == EALREADY) - return (0); - if (error) - return (error); - - /* - * If we slept after putting bits otw, then reply may have - * arrived. In which case returning is required, or we - * would hang trying to nfs_receive an already received reply. - */ - if (myrep->r_mrep != NULL) { - nfs_rcvunlock(myrep); - FSDBG(530, myrep->r_xid, myrep, myrep->r_nmp, -1); - return (0); + if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) { + /* set mntfromname to this location */ + if (!NM_OMATTR_GIVEN(nmp, MNTFROM)) + nfs_location_mntfromname(&nmp->nm_locations, nso->nso_location, + vfs_statfs(nmp->nm_mountp)->f_mntfromname, + sizeof(vfs_statfs(nmp->nm_mountp)->f_mntfromname), 0); + /* some negotiated values need to remain unchanged for the life of the mount */ + if (!nmp->nm_sotype) + nmp->nm_sotype = nso->nso_sotype; + if (!nmp->nm_vers) { + nmp->nm_vers = nfsvers; + /* If we negotiated NFSv4, set nm_nfsport if we ended up on the standard NFS port */ + if ((nfsvers >= NFS_VER4) && !NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_PORT)) { + if (nso->nso_saddr->sa_family == AF_INET) + port = ((struct sockaddr_in*)nso->nso_saddr)->sin_port = htons(port); + else if (nso->nso_saddr->sa_family == AF_INET6) + port = ((struct sockaddr_in6*)nso->nso_saddr)->sin6_port = htons(port); + else + port = 0; + if (port == NFS_PORT) + nmp->nm_nfsport = NFS_PORT; + } } - /* - * Get the next Rpc reply off the socket. Assume myrep->r_nmp - * is still intact by checks done in nfs_rcvlock. - */ - error = nfs_receive(myrep, &mrep); - /* - * Bailout asap if nfsmount struct gone (unmounted). - */ - if (!myrep->r_nmp) { - FSDBG(530, myrep->r_xid, myrep, nmp, -2); - if (mrep) - mbuf_freem(mrep); - return (ENXIO); + /* do some version-specific pre-mount set up */ + if (nmp->nm_vers >= NFS_VER4) { + microtime(&now); + nmp->nm_mounttime = ((uint64_t)now.tv_sec << 32) | now.tv_usec; + if (!NMFLAG(nmp, NOCALLBACK)) + nfs4_mount_callback_setup(nmp); } - if (error) { - FSDBG(530, myrep->r_xid, myrep, nmp, error); - nfs_rcvunlock(myrep); - - /* Bailout asap if nfsmount struct gone (unmounted). */ - if (!myrep->r_nmp) { - if (mrep) - mbuf_freem(mrep); - return (ENXIO); - } + } - /* - * Ignore routing errors on connectionless protocols?? - */ - if (NFSIGNORE_SOERROR(nmp->nm_sotype, error)) { - if (nmp->nm_so) { - int clearerror; - int optlen = sizeof(clearerror); - sock_getsockopt(nmp->nm_so, SOL_SOCKET, SO_ERROR, &clearerror, &optlen); + /* Initialize NFS socket state variables */ + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_srtt[0] = nmp->nm_srtt[1] = nmp->nm_srtt[2] = + nmp->nm_srtt[3] = (NFS_TIMEO << 3); + nmp->nm_sdrtt[0] = nmp->nm_sdrtt[1] = nmp->nm_sdrtt[2] = + nmp->nm_sdrtt[3] = 0; + if (nso->nso_sotype == SOCK_DGRAM) { + nmp->nm_cwnd = NFS_MAXCWND / 2; /* Initial send window */ + nmp->nm_sent = 0; + } else if (nso->nso_sotype == SOCK_STREAM) { + nmp->nm_timeouts = 0; + } + nmp->nm_sockflags &= ~NMSOCK_CONNECTING; + nmp->nm_sockflags |= NMSOCK_SETUP; + /* move the socket to the mount structure */ + nmp->nm_nso = nso; + oldsaddr = nmp->nm_saddr; + nmp->nm_saddr = nso->nso_saddr; + lck_mtx_unlock(&nmp->nm_lock); + error = nfs_connect_setup(nmp); + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags &= ~NMSOCK_SETUP; + if (!error) { + nmp->nm_sockflags |= NMSOCK_READY; + wakeup(&nmp->nm_sockflags); + } + if (error) { + NFS_SOCK_DBG("nfs connect %s socket %p setup failed %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, nso, error); + nfs_socket_search_update_error(&nss, error); + nmp->nm_saddr = oldsaddr; + if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) { + /* undo settings made prior to setup */ + if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_SOCKET_TYPE)) + nmp->nm_sotype = 0; + if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_VERSION)) { + if (nmp->nm_vers >= NFS_VER4) { + if (!NFS_BITMAP_ISSET(nmp->nm_mattrs, NFS_MATTR_NFS_PORT)) + nmp->nm_nfsport = 0; + if (nmp->nm_cbid) + nfs4_mount_callback_shutdown(nmp); + if (IS_VALID_CRED(nmp->nm_mcred)) + kauth_cred_unref(&nmp->nm_mcred); + bzero(&nmp->nm_un, sizeof(nmp->nm_un)); } - continue; + nmp->nm_vers = 0; } - if (mrep) - mbuf_freem(mrep); - return (error); } + lck_mtx_unlock(&nmp->nm_lock); + nmp->nm_nso = NULL; + nfs_socket_destroy(nso); + goto keepsearching; + } - /* - * We assume all is fine, but if we did not have an error - * and mrep is 0, better not dereference it. nfs_receive - * calls soreceive which carefully sets error=0 when it got - * errors on sbwait (tsleep). In most cases, I assume that's - * so we could go back again. In tcp case, EPIPE is returned. - * In udp, case nfs_receive gets back here with no error and no - * mrep. Is the right fix to have soreceive check for process - * aborted after sbwait and return something non-zero? Should - * nfs_receive give an EPIPE? Too risky to play with those - * two this late in game for a shutdown problem. Instead, - * just check here and get out. (ekn) - */ - if (!mrep) { - nfs_rcvunlock(myrep); - FSDBG(530, myrep->r_xid, myrep, nmp, -3); - return (ENXIO); /* sounds good */ - } - - /* - * Get the xid and check that it is an rpc reply - */ - md = mrep; - dpos = mbuf_data(md); - nfsm_dissect(tl, u_long *, 2*NFSX_UNSIGNED); - rxid = *tl++; - if (*tl != rpc_reply) { - OSAddAtomic(1, (SInt32*)&nfsstats.rpcinvalid); - mbuf_freem(mrep); -nfsmout: - if (nmp->nm_state & NFSSTA_RCVLOCK) - nfs_rcvunlock(myrep); - continue; - } + /* update current location */ + if ((nmp->nm_locations.nl_current.nli_flags & NLI_VALID) && + (nmp->nm_locations.nl_current.nli_serv != nso->nso_location.nli_serv)) { + /* server has changed, we should initiate failover/recovery */ + // XXX + } + nmp->nm_locations.nl_current = nso->nso_location; + nmp->nm_locations.nl_current.nli_flags |= NLI_VALID; - /* - * Loop through the request list to match up the reply - * Iff no match, just drop the datagram - */ - TAILQ_FOREACH(rep, &nfs_reqq, r_chain) { - if (rep->r_mrep == NULL && rxid == rep->r_xid) { - /* Found it.. */ - rep->r_mrep = mrep; - rep->r_md = md; - rep->r_dpos = dpos; - /* - * If we're tracking the round trip time - * then we update the circular log here - * with the stats from our current request. - */ - if (nfsrtton) { - struct rttl *rt; - - rt = &nfsrtt.rttl[nfsrtt.pos]; - rt->proc = rep->r_procnum; - rt->rto = NFS_RTO(nmp, proct[rep->r_procnum]); - rt->sent = nmp->nm_sent; - rt->cwnd = nmp->nm_cwnd; - if (proct[rep->r_procnum] == 0) - panic("nfs_reply: proct[%d] is zero", rep->r_procnum); - rt->srtt = nmp->nm_srtt[proct[rep->r_procnum] - 1]; - rt->sdrtt = nmp->nm_sdrtt[proct[rep->r_procnum] - 1]; - rt->fsid = vfs_statfs(nmp->nm_mountp)->f_fsid; - microtime(&rt->tstamp); // XXX unused - if (rep->r_flags & R_TIMING) - rt->rtt = rep->r_rtt; - else - rt->rtt = 1000000; - nfsrtt.pos = (nfsrtt.pos + 1) % NFSRTTLOGSIZ; - } - /* - * Update congestion window. - * Do the additive increase of - * one rpc/rtt. - */ - FSDBG(530, rep->r_xid, rep, nmp->nm_sent, - nmp->nm_cwnd); - if (nmp->nm_cwnd <= nmp->nm_sent) { - nmp->nm_cwnd += - (NFS_CWNDSCALE * NFS_CWNDSCALE + - (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; - if (nmp->nm_cwnd > NFS_MAXCWND) - nmp->nm_cwnd = NFS_MAXCWND; - } - if (rep->r_flags & R_SENT) { - rep->r_flags &= ~R_SENT; - nmp->nm_sent -= NFS_CWNDSCALE; - } - /* - * Update rtt using a gain of 0.125 on the mean - * and a gain of 0.25 on the deviation. - */ - if (rep->r_flags & R_TIMING) { - /* - * Since the timer resolution of - * NFS_HZ is so course, it can often - * result in r_rtt == 0. Since - * r_rtt == N means that the actual - * rtt is between N+dt and N+2-dt ticks, - * add 1. - */ - if (proct[rep->r_procnum] == 0) - panic("nfs_reply: proct[%d] is zero", rep->r_procnum); - t1 = rep->r_rtt + 1; - t1 -= (NFS_SRTT(rep) >> 3); - NFS_SRTT(rep) += t1; - if (t1 < 0) - t1 = -t1; - t1 -= (NFS_SDRTT(rep) >> 2); - NFS_SDRTT(rep) += t1; - } - nmp->nm_timeouts = 0; - break; + if (!(nmp->nm_sockflags & NMSOCK_HASCONNECTED)) { + /* We have now successfully connected... make a note of it. */ + nmp->nm_sockflags |= NMSOCK_HASCONNECTED; + } + + lck_mtx_unlock(&nmp->nm_lock); + if (oldsaddr) + FREE(oldsaddr, M_SONAME); + + if (nss.nss_flags & NSS_WARNED) + log(LOG_INFO, "nfs_connect: socket connect completed for %s\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname); + + nmp->nm_nss = NULL; + nfs_socket_search_cleanup(&nss); + if (fh) + FREE(fh, M_TEMP); + if (path) + FREE_ZONE(path, MAXPATHLEN, M_NAMEI); + NFS_SOCK_DBG("nfs connect %s success\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname); + return (0); +} + + +/* setup & confirm socket connection is functional */ +int +nfs_connect_setup(struct nfsmount *nmp) +{ + int error = 0; + + if (nmp->nm_vers >= NFS_VER4) { + if (nmp->nm_state & NFSSTA_CLIENTID) { + /* first, try to renew our current state */ + error = nfs4_renew(nmp, R_SETUP); + if ((error == NFSERR_ADMIN_REVOKED) || + (error == NFSERR_CB_PATH_DOWN) || + (error == NFSERR_EXPIRED) || + (error == NFSERR_LEASE_MOVED) || + (error == NFSERR_STALE_CLIENTID)) { + lck_mtx_lock(&nmp->nm_lock); + nfs_need_recover(nmp, error); + lck_mtx_unlock(&nmp->nm_lock); } } - nfs_rcvunlock(myrep); - /* - * If not matched to a request, drop it. - * If it's mine, get out. - */ - if (rep == 0) { - OSAddAtomic(1, (SInt32*)&nfsstats.rpcunexpected); - mbuf_freem(mrep); - } else if (rep == myrep) { - if (rep->r_mrep == NULL) - panic("nfs_reply: nil r_mrep"); - return (0); - } - FSDBG(530, myrep->r_xid, myrep, rep, - rep ? rep->r_xid : myrep->r_flags); + error = nfs4_setclientid(nmp); } + return (error); } /* - * nfs_request - goes something like this - * - fill in request struct - * - links it into list - * - calls nfs_send() for first transmit - * - calls nfs_receive() to get reply - * - break down rpc header and return with nfs reply pointed to - * by mrep or error - * nb: always frees up mreq mbuf list + * NFS socket reconnect routine: + * Called when a connection is broken. + * - disconnect the old socket + * - nfs_connect() again + * - set R_MUSTRESEND for all outstanding requests on mount point + * If this fails the mount point is DEAD! */ int -nfs_request(vp, mp, mrest, procnum, procp, cred, mrp, mdp, dposp, xidp) - vnode_t vp; - mount_t mp; - mbuf_t mrest; - int procnum; - proc_t procp; - kauth_cred_t cred; - mbuf_t *mrp; - mbuf_t *mdp; - caddr_t *dposp; - u_int64_t *xidp; -{ - mbuf_t m, mrep, m2; - struct nfsreq re, *rep; - u_long *tl; - int i; - struct nfsmount *nmp; - mbuf_t md, mheadend; - char nickv[RPCX_NICKVERF]; - time_t waituntil; - caddr_t dpos, cp2; - int t1, error = 0, mrest_len, auth_len, auth_type; - int trylater_delay = NFS_TRYLATERDEL, failed_auth = 0; - int verf_len, verf_type; - u_long xid; - char *auth_str, *verf_str; - NFSKERBKEY_T key; /* save session key */ - int nmsotype; +nfs_reconnect(struct nfsmount *nmp) +{ + struct nfsreq *rq; struct timeval now; + thread_t thd = current_thread(); + int error, wentdown = 0, verbose = 1; + time_t lastmsg; + int timeo; - if (mrp) - *mrp = NULL; - if (xidp) - *xidp = 0; - nmp = VFSTONFS(mp); - - rep = &re; + microuptime(&now); + lastmsg = now.tv_sec - (nmp->nm_tprintf_delay - nmp->nm_tprintf_initial_delay); - if (vp) - nmp = VFSTONFS(vnode_mount(vp)); - if (nmp == NULL || - (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_TIMEO)) == - (NFSSTA_FORCE|NFSSTA_TIMEO)) { - mbuf_freem(mrest); - return (ENXIO); - } - nmsotype = nmp->nm_sotype; + nfs_disconnect(nmp); - FSDBG_TOP(531, vp, procnum, nmp, rep); - rep->r_nmp = nmp; - rep->r_vp = vp; - rep->r_procp = procp; - rep->r_procnum = procnum; - microuptime(&now); - rep->r_lastmsg = now.tv_sec - - ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay)); - i = 0; - m = mrest; - while (m) { - i += mbuf_len(m); - m = mbuf_next(m); - } - mrest_len = i; + lck_mtx_lock(&nmp->nm_lock); + timeo = nfs_is_squishy(nmp) ? 8 : 30; + lck_mtx_unlock(&nmp->nm_lock); - /* - * Get the RPC header with authorization. - */ -kerbauth: - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; - if (!nmp) { - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); + while ((error = nfs_connect(nmp, verbose, timeo))) { + verbose = 0; + nfs_disconnect(nmp); + if ((error == EINTR) || (error == ERESTART)) + return (EINTR); + if (error == EIO) + return (EIO); + microuptime(&now); + if ((lastmsg + nmp->nm_tprintf_delay) < now.tv_sec) { + lastmsg = now.tv_sec; + nfs_down(nmp, thd, error, NFSSTA_TIMEO, "can not connect"); + wentdown = 1; + } + lck_mtx_lock(&nmp->nm_lock); + if (!(nmp->nm_state & NFSSTA_MOUNTED)) { + /* we're not yet completely mounted and */ + /* we can't reconnect, so we fail */ + lck_mtx_unlock(&nmp->nm_lock); + NFS_SOCK_DBG("Not mounted returning %d\n", error); + return (error); + } + nfs_mount_check_dead_timeout(nmp); + if ((error = nfs_sigintr(nmp, NULL, thd, 1))) { + lck_mtx_unlock(&nmp->nm_lock); + return (error); + } + lck_mtx_unlock(&nmp->nm_lock); + tsleep(nfs_reconnect, PSOCK, "nfs_reconnect_delay", 2*hz); + if ((error = nfs_sigintr(nmp, NULL, thd, 0))) + return (error); + } + + if (wentdown) + nfs_up(nmp, thd, NFSSTA_TIMEO, "connected"); + + /* + * Loop through outstanding request list and mark all requests + * as needing a resend. (Though nfs_need_reconnect() probably + * marked them all already.) + */ + lck_mtx_lock(nfs_request_mutex); + TAILQ_FOREACH(rq, &nfs_reqq, r_chain) { + if (rq->r_nmp == nmp) { + lck_mtx_lock(&rq->r_mtx); + if (!rq->r_error && !rq->r_nmrep.nmc_mhead && !(rq->r_flags & R_MUSTRESEND)) { + rq->r_flags |= R_MUSTRESEND; + rq->r_rtt = -1; + wakeup(rq); + if ((rq->r_flags & (R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC) + nfs_asyncio_resend(rq); + } + lck_mtx_unlock(&rq->r_mtx); + } + } + lck_mtx_unlock(nfs_request_mutex); + return (0); +} + +/* + * NFS disconnect. Clean up and unlink. + */ +void +nfs_disconnect(struct nfsmount *nmp) +{ + struct nfs_socket *nso; + + lck_mtx_lock(&nmp->nm_lock); +tryagain: + if (nmp->nm_nso) { + struct timespec ts = { 1, 0 }; + if (nmp->nm_state & NFSSTA_SENDING) { /* wait for sending to complete */ + nmp->nm_state |= NFSSTA_WANTSND; + msleep(&nmp->nm_state, &nmp->nm_lock, PZERO-1, "nfswaitsending", &ts); + goto tryagain; + } + if (nmp->nm_sockflags & NMSOCK_POKE) { /* wait for poking to complete */ + msleep(&nmp->nm_sockflags, &nmp->nm_lock, PZERO-1, "nfswaitpoke", &ts); + goto tryagain; + } + nmp->nm_sockflags |= NMSOCK_DISCONNECTING; + nmp->nm_sockflags &= ~NMSOCK_READY; + nso = nmp->nm_nso; + nmp->nm_nso = NULL; + if (nso->nso_saddr == nmp->nm_saddr) + nso->nso_saddr = NULL; + lck_mtx_unlock(&nmp->nm_lock); + nfs_socket_destroy(nso); + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags &= ~NMSOCK_DISCONNECTING; + lck_mtx_unlock(&nmp->nm_lock); + } else { + lck_mtx_unlock(&nmp->nm_lock); + } +} + +/* + * mark an NFS mount as needing a reconnect/resends. + */ +void +nfs_need_reconnect(struct nfsmount *nmp) +{ + struct nfsreq *rq; + + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags &= ~(NMSOCK_READY|NMSOCK_SETUP); + lck_mtx_unlock(&nmp->nm_lock); + + /* + * Loop through outstanding request list and + * mark all requests as needing a resend. + */ + lck_mtx_lock(nfs_request_mutex); + TAILQ_FOREACH(rq, &nfs_reqq, r_chain) { + if (rq->r_nmp == nmp) { + lck_mtx_lock(&rq->r_mtx); + if (!rq->r_error && !rq->r_nmrep.nmc_mhead && !(rq->r_flags & R_MUSTRESEND)) { + rq->r_flags |= R_MUSTRESEND; + rq->r_rtt = -1; + wakeup(rq); + if ((rq->r_flags & (R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC) + nfs_asyncio_resend(rq); + } + lck_mtx_unlock(&rq->r_mtx); + } + } + lck_mtx_unlock(nfs_request_mutex); +} + + +/* + * thread to handle miscellaneous async NFS socket work (reconnects/resends) + */ +void +nfs_mount_sock_thread(void *arg, __unused wait_result_t wr) +{ + struct nfsmount *nmp = arg; + struct timespec ts = { 30, 0 }; + thread_t thd = current_thread(); + struct nfsreq *req; + struct timeval now; + int error, dofinish; + nfsnode_t np; + int do_reconnect_sleep = 0; + + lck_mtx_lock(&nmp->nm_lock); + + while (!(nmp->nm_sockflags & NMSOCK_READY) || + !TAILQ_EMPTY(&nmp->nm_resendq) || + !LIST_EMPTY(&nmp->nm_monlist) || + nmp->nm_deadto_start || + (nmp->nm_state & NFSSTA_RECOVER) || + ((nmp->nm_vers >= NFS_VER4) && !TAILQ_EMPTY(&nmp->nm_dreturnq))) + { + if (nmp->nm_sockflags & NMSOCK_UNMOUNT) + break; + /* do reconnect, if necessary */ + if (!(nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & NFSSTA_FORCE)) { + if (nmp->nm_reconnect_start <= 0) { + microuptime(&now); + nmp->nm_reconnect_start = now.tv_sec; + } + lck_mtx_unlock(&nmp->nm_lock); + NFS_SOCK_DBG("nfs reconnect %s\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname); + /* + * XXX We don't want to call reconnect again right away if returned errors + * before that may not have blocked. This has caused spamming null procs + * from machines in the pass. + */ + if (do_reconnect_sleep) + tsleep(nfs_mount_sock_thread, PSOCK, "nfs_reconnect_sock_thread_delay", hz); + error = nfs_reconnect(nmp); + if (error) { + int lvl = 7; + if (error == EIO || error == EINTR) { + lvl = (do_reconnect_sleep++ % 600) ? 7 : 0; + } + nfs_printf(NFS_FAC_SOCK, lvl, "nfs reconnect %s: returned %d\n", + vfs_statfs(nmp->nm_mountp)->f_mntfromname, error); + } else { + nmp->nm_reconnect_start = 0; + do_reconnect_sleep = 0; + } + lck_mtx_lock(&nmp->nm_lock); + } + if ((nmp->nm_sockflags & NMSOCK_READY) && + (nmp->nm_state & NFSSTA_RECOVER) && + !(nmp->nm_sockflags & NMSOCK_UNMOUNT) && + !(nmp->nm_state & NFSSTA_FORCE)) { + /* perform state recovery */ + lck_mtx_unlock(&nmp->nm_lock); + nfs_recover(nmp); + lck_mtx_lock(&nmp->nm_lock); + } + /* handle NFSv4 delegation returns */ + while ((nmp->nm_vers >= NFS_VER4) && !(nmp->nm_state & NFSSTA_FORCE) && + (nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & NFSSTA_RECOVER) && + ((np = TAILQ_FIRST(&nmp->nm_dreturnq)))) { + lck_mtx_unlock(&nmp->nm_lock); + nfs4_delegation_return(np, R_RECOVER, thd, nmp->nm_mcred); + lck_mtx_lock(&nmp->nm_lock); + } + /* do resends, if necessary/possible */ + while ((((nmp->nm_sockflags & NMSOCK_READY) && !(nmp->nm_state & NFSSTA_RECOVER)) || (nmp->nm_state & NFSSTA_FORCE)) && + ((req = TAILQ_FIRST(&nmp->nm_resendq)))) { + if (req->r_resendtime) + microuptime(&now); + while (req && !(nmp->nm_state & NFSSTA_FORCE) && req->r_resendtime && (now.tv_sec < req->r_resendtime)) + req = TAILQ_NEXT(req, r_rchain); + if (!req) + break; + TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain); + req->r_rchain.tqe_next = NFSREQNOLIST; + lck_mtx_unlock(&nmp->nm_lock); + lck_mtx_lock(&req->r_mtx); + if (req->r_error || req->r_nmrep.nmc_mhead) { + dofinish = req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT); + req->r_flags &= ~R_RESENDQ; + wakeup(req); + lck_mtx_unlock(&req->r_mtx); + if (dofinish) + nfs_asyncio_finish(req); + lck_mtx_lock(&nmp->nm_lock); + continue; + } + if ((req->r_flags & R_RESTART) || nfs_request_using_gss(req)) { + req->r_flags &= ~R_RESTART; + req->r_resendtime = 0; + lck_mtx_unlock(&req->r_mtx); + /* async RPCs on GSS mounts need to be rebuilt and resent. */ + nfs_reqdequeue(req); + if (nfs_request_using_gss(req)) { + nfs_gss_clnt_rpcdone(req); + error = nfs_gss_clnt_args_restore(req); + if (error == ENEEDAUTH) + req->r_xid = 0; + } + NFS_SOCK_DBG("nfs async%s restart: p %d x 0x%llx f 0x%x rtt %d\n", + nfs_request_using_gss(req) ? " gss" : "", req->r_procnum, req->r_xid, + req->r_flags, req->r_rtt); + error = !req->r_nmp ? ENXIO : 0; /* unmounted? */ + if (!error) + error = nfs_sigintr(nmp, req, req->r_thread, 0); + if (!error) + error = nfs_request_add_header(req); + if (!error) + error = nfs_request_send(req, 0); + lck_mtx_lock(&req->r_mtx); + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + if (error) + req->r_error = error; + wakeup(req); + dofinish = error && req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT); + lck_mtx_unlock(&req->r_mtx); + if (dofinish) + nfs_asyncio_finish(req); + lck_mtx_lock(&nmp->nm_lock); + error = 0; + continue; + } + NFS_SOCK_DBG("nfs async resend: p %d x 0x%llx f 0x%x rtt %d\n", + req->r_procnum, req->r_xid, req->r_flags, req->r_rtt); + error = !req->r_nmp ? ENXIO : 0; /* unmounted? */ + if (!error) + error = nfs_sigintr(nmp, req, req->r_thread, 0); + if (!error) { + req->r_flags |= R_SENDING; + lck_mtx_unlock(&req->r_mtx); + error = nfs_send(req, 0); + lck_mtx_lock(&req->r_mtx); + if (!error) { + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + wakeup(req); + lck_mtx_unlock(&req->r_mtx); + lck_mtx_lock(&nmp->nm_lock); + continue; + } + } + req->r_error = error; + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + wakeup(req); + dofinish = req->r_callback.rcb_func && !(req->r_flags & R_WAITSENT); + lck_mtx_unlock(&req->r_mtx); + if (dofinish) + nfs_asyncio_finish(req); + lck_mtx_lock(&nmp->nm_lock); + } + if (nmp->nm_deadto_start) + nfs_mount_check_dead_timeout(nmp); + if (nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD)) + break; + /* check monitored nodes, if necessary/possible */ + if (!LIST_EMPTY(&nmp->nm_monlist)) { + nmp->nm_state |= NFSSTA_MONITOR_SCAN; + LIST_FOREACH(np, &nmp->nm_monlist, n_monlink) { + if (!(nmp->nm_sockflags & NMSOCK_READY) || (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING|NFSSTA_FORCE))) + break; + np->n_mflag |= NMMONSCANINPROG; + lck_mtx_unlock(&nmp->nm_lock); + error = nfs_getattr(np, NULL, vfs_context_kernel(), (NGA_UNCACHED|NGA_MONITOR)); + if (!error && ISSET(np->n_flag, NUPDATESIZE)) /* update quickly to avoid multiple events */ + nfs_data_update_size(np, 0); + lck_mtx_lock(&nmp->nm_lock); + np->n_mflag &= ~NMMONSCANINPROG; + if (np->n_mflag & NMMONSCANWANT) { + np->n_mflag &= ~NMMONSCANWANT; + wakeup(&np->n_mflag); + } + if (error || !(nmp->nm_sockflags & NMSOCK_READY) || (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING|NFSSTA_FORCE))) + break; + } + nmp->nm_state &= ~NFSSTA_MONITOR_SCAN; + if (nmp->nm_state & NFSSTA_UNMOUNTING) + wakeup(&nmp->nm_state); /* let unmounting thread know scan is done */ + } + if ((nmp->nm_sockflags & NMSOCK_READY) || (nmp->nm_state & (NFSSTA_RECOVER|NFSSTA_UNMOUNTING))) { + if (nmp->nm_deadto_start || !TAILQ_EMPTY(&nmp->nm_resendq) || + (nmp->nm_state & NFSSTA_RECOVER)) + ts.tv_sec = 1; + else + ts.tv_sec = 5; + msleep(&nmp->nm_sockthd, &nmp->nm_lock, PSOCK, "nfssockthread", &ts); + } + } + + /* If we're unmounting, send the unmount RPC, if requested/appropriate. */ + if ((nmp->nm_sockflags & NMSOCK_UNMOUNT) && + (nmp->nm_state & NFSSTA_MOUNTED) && NMFLAG(nmp, CALLUMNT) && + (nmp->nm_vers < NFS_VER4) && !(nmp->nm_state & (NFSSTA_FORCE|NFSSTA_DEAD))) { + lck_mtx_unlock(&nmp->nm_lock); + nfs3_umount_rpc(nmp, vfs_context_kernel(), + (nmp->nm_sockflags & NMSOCK_READY) ? 6 : 2); + lck_mtx_lock(&nmp->nm_lock); + } + + if (nmp->nm_sockthd == thd) + nmp->nm_sockthd = NULL; + lck_mtx_unlock(&nmp->nm_lock); + wakeup(&nmp->nm_sockthd); + thread_terminate(thd); +} + +/* start or wake a mount's socket thread */ +void +nfs_mount_sock_thread_wake(struct nfsmount *nmp) +{ + if (nmp->nm_sockthd) + wakeup(&nmp->nm_sockthd); + else if (kernel_thread_start(nfs_mount_sock_thread, nmp, &nmp->nm_sockthd) == KERN_SUCCESS) + thread_deallocate(nmp->nm_sockthd); +} + +/* + * Check if we should mark the mount dead because the + * unresponsive mount has reached the dead timeout. + * (must be called with nmp locked) + */ +void +nfs_mount_check_dead_timeout(struct nfsmount *nmp) +{ + struct timeval now; + + if (nmp->nm_deadto_start == 0) + return; + if (nmp->nm_state & NFSSTA_DEAD) + return; + nfs_is_squishy(nmp); + if (nmp->nm_curdeadtimeout <= 0) + return; + microuptime(&now); + if ((now.tv_sec - nmp->nm_deadto_start) < nmp->nm_curdeadtimeout) + return; + printf("nfs server %s: %sdead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, + (nmp->nm_curdeadtimeout != nmp->nm_deadtimeout) ? "squished " : ""); + nmp->nm_state |= NFSSTA_DEAD; + vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_DEAD, 0); +} + +/* + * NFS callback channel socket state + */ +struct nfs_callback_socket +{ + TAILQ_ENTRY(nfs_callback_socket) ncbs_link; + socket_t ncbs_so; /* the socket */ + struct sockaddr_storage ncbs_saddr; /* socket address */ + struct nfs_rpc_record_state ncbs_rrs; /* RPC record parsing state */ + time_t ncbs_stamp; /* last accessed at */ + uint32_t ncbs_flags; /* see below */ +}; +#define NCBSOCK_UPCALL 0x0001 +#define NCBSOCK_UPCALLWANT 0x0002 +#define NCBSOCK_DEAD 0x0004 + +/* + * NFS callback channel state + * + * One listening socket for accepting socket connections from servers and + * a list of connected sockets to handle callback requests on. + * Mounts registered with the callback channel are assigned IDs and + * put on a list so that the callback request handling code can match + * the requests up with mounts. + */ +socket_t nfs4_cb_so = NULL; +socket_t nfs4_cb_so6 = NULL; +in_port_t nfs4_cb_port = 0; +in_port_t nfs4_cb_port6 = 0; +uint32_t nfs4_cb_id = 0; +uint32_t nfs4_cb_so_usecount = 0; +TAILQ_HEAD(nfs4_cb_sock_list,nfs_callback_socket) nfs4_cb_socks; +TAILQ_HEAD(nfs4_cb_mount_list,nfsmount) nfs4_cb_mounts; + +int nfs4_cb_handler(struct nfs_callback_socket *, mbuf_t); + +/* + * Set up the callback channel for the NFS mount. + * + * Initializes the callback channel socket state and + * assigns a callback ID to the mount. + */ +void +nfs4_mount_callback_setup(struct nfsmount *nmp) +{ + struct sockaddr_in sin; + struct sockaddr_in6 sin6; + socket_t so = NULL; + socket_t so6 = NULL; + struct timeval timeo; + int error, on = 1; + in_port_t port; + + lck_mtx_lock(nfs_global_mutex); + if (nfs4_cb_id == 0) { + TAILQ_INIT(&nfs4_cb_mounts); + TAILQ_INIT(&nfs4_cb_socks); + nfs4_cb_id++; + } + nmp->nm_cbid = nfs4_cb_id++; + if (nmp->nm_cbid == 0) + nmp->nm_cbid = nfs4_cb_id++; + nfs4_cb_so_usecount++; + TAILQ_INSERT_HEAD(&nfs4_cb_mounts, nmp, nm_cblink); + + if (nfs4_cb_so) { + lck_mtx_unlock(nfs_global_mutex); + return; + } + + /* IPv4 */ + error = sock_socket(AF_INET, SOCK_STREAM, IPPROTO_TCP, nfs4_cb_accept, NULL, &nfs4_cb_so); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d creating listening IPv4 socket\n", error); + goto fail; + } + so = nfs4_cb_so; + + sock_setsockopt(so, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)); + sin.sin_len = sizeof(struct sockaddr_in); + sin.sin_family = AF_INET; + sin.sin_addr.s_addr = htonl(INADDR_ANY); + sin.sin_port = htons(nfs_callback_port); /* try to use specified port */ + error = sock_bind(so, (struct sockaddr *)&sin); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d binding listening IPv4 socket\n", error); + goto fail; + } + error = sock_getsockname(so, (struct sockaddr *)&sin, sin.sin_len); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d getting listening IPv4 socket port\n", error); + goto fail; + } + nfs4_cb_port = ntohs(sin.sin_port); + + error = sock_listen(so, 32); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d on IPv4 listen\n", error); + goto fail; + } + + /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */ + timeo.tv_usec = 0; + timeo.tv_sec = 60; + error = sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback setup: error %d setting IPv4 socket rx timeout\n", error); + error = sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback setup: error %d setting IPv4 socket tx timeout\n", error); + sock_setsockopt(so, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); + sock_setsockopt(so, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on)); + sock_setsockopt(so, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on)); + error = 0; + + /* IPv6 */ + error = sock_socket(AF_INET6, SOCK_STREAM, IPPROTO_TCP, nfs4_cb_accept, NULL, &nfs4_cb_so6); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d creating listening IPv6 socket\n", error); + goto fail; + } + so6 = nfs4_cb_so6; + + sock_setsockopt(so6, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)); + sock_setsockopt(so6, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof(on)); + /* try to use specified port or same port as IPv4 */ + port = nfs_callback_port ? nfs_callback_port : nfs4_cb_port; +ipv6_bind_again: + sin6.sin6_len = sizeof(struct sockaddr_in6); + sin6.sin6_family = AF_INET6; + sin6.sin6_addr = in6addr_any; + sin6.sin6_port = htons(port); + error = sock_bind(so6, (struct sockaddr *)&sin6); + if (error) { + if (port != nfs_callback_port) { + /* if we simply tried to match the IPv4 port, then try any port */ + port = 0; + goto ipv6_bind_again; + } + log(LOG_INFO, "nfs callback setup: error %d binding listening IPv6 socket\n", error); + goto fail; + } + error = sock_getsockname(so6, (struct sockaddr *)&sin6, sin6.sin6_len); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d getting listening IPv6 socket port\n", error); + goto fail; + } + nfs4_cb_port6 = ntohs(sin6.sin6_port); + + error = sock_listen(so6, 32); + if (error) { + log(LOG_INFO, "nfs callback setup: error %d on IPv6 listen\n", error); + goto fail; + } + + /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */ + timeo.tv_usec = 0; + timeo.tv_sec = 60; + error = sock_setsockopt(so6, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback setup: error %d setting IPv6 socket rx timeout\n", error); + error = sock_setsockopt(so6, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback setup: error %d setting IPv6 socket tx timeout\n", error); + sock_setsockopt(so6, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); + sock_setsockopt(so6, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on)); + sock_setsockopt(so6, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on)); + error = 0; + +fail: + if (error) { + nfs4_cb_so = nfs4_cb_so6 = NULL; + lck_mtx_unlock(nfs_global_mutex); + if (so) { + sock_shutdown(so, SHUT_RDWR); + sock_close(so); + } + if (so6) { + sock_shutdown(so6, SHUT_RDWR); + sock_close(so6); + } + } else { + lck_mtx_unlock(nfs_global_mutex); + } +} + +/* + * Shut down the callback channel for the NFS mount. + * + * Clears the mount's callback ID and releases the mounts + * reference on the callback socket. Last reference dropped + * will also shut down the callback socket(s). + */ +void +nfs4_mount_callback_shutdown(struct nfsmount *nmp) +{ + struct nfs_callback_socket *ncbsp; + socket_t so, so6; + struct nfs4_cb_sock_list cb_socks; + struct timespec ts = {1,0}; + + lck_mtx_lock(nfs_global_mutex); + TAILQ_REMOVE(&nfs4_cb_mounts, nmp, nm_cblink); + /* wait for any callbacks in progress to complete */ + while (nmp->nm_cbrefs) + msleep(&nmp->nm_cbrefs, nfs_global_mutex, PSOCK, "cbshutwait", &ts); + nmp->nm_cbid = 0; + if (--nfs4_cb_so_usecount) { + lck_mtx_unlock(nfs_global_mutex); + return; + } + so = nfs4_cb_so; + so6 = nfs4_cb_so6; + nfs4_cb_so = nfs4_cb_so6 = NULL; + TAILQ_INIT(&cb_socks); + TAILQ_CONCAT(&cb_socks, &nfs4_cb_socks, ncbs_link); + lck_mtx_unlock(nfs_global_mutex); + if (so) { + sock_shutdown(so, SHUT_RDWR); + sock_close(so); + } + if (so6) { + sock_shutdown(so6, SHUT_RDWR); + sock_close(so6); + } + while ((ncbsp = TAILQ_FIRST(&cb_socks))) { + TAILQ_REMOVE(&cb_socks, ncbsp, ncbs_link); + sock_shutdown(ncbsp->ncbs_so, SHUT_RDWR); + sock_close(ncbsp->ncbs_so); + nfs_rpc_record_state_cleanup(&ncbsp->ncbs_rrs); + FREE(ncbsp, M_TEMP); + } +} + +/* + * Check periodically for stale/unused nfs callback sockets + */ +#define NFS4_CB_TIMER_PERIOD 30 +#define NFS4_CB_IDLE_MAX 300 +void +nfs4_callback_timer(__unused void *param0, __unused void *param1) +{ + struct nfs_callback_socket *ncbsp, *nextncbsp; + struct timeval now; + +loop: + lck_mtx_lock(nfs_global_mutex); + if (TAILQ_EMPTY(&nfs4_cb_socks)) { + nfs4_callback_timer_on = 0; + lck_mtx_unlock(nfs_global_mutex); + return; + } + microuptime(&now); + TAILQ_FOREACH_SAFE(ncbsp, &nfs4_cb_socks, ncbs_link, nextncbsp) { + if (!(ncbsp->ncbs_flags & NCBSOCK_DEAD) && + (now.tv_sec < (ncbsp->ncbs_stamp + NFS4_CB_IDLE_MAX))) + continue; + TAILQ_REMOVE(&nfs4_cb_socks, ncbsp, ncbs_link); + lck_mtx_unlock(nfs_global_mutex); + sock_shutdown(ncbsp->ncbs_so, SHUT_RDWR); + sock_close(ncbsp->ncbs_so); + nfs_rpc_record_state_cleanup(&ncbsp->ncbs_rrs); + FREE(ncbsp, M_TEMP); + goto loop; + } + nfs4_callback_timer_on = 1; + nfs_interval_timer_start(nfs4_callback_timer_call, + NFS4_CB_TIMER_PERIOD * 1000); + lck_mtx_unlock(nfs_global_mutex); +} + +/* + * Accept a new callback socket. + */ +void +nfs4_cb_accept(socket_t so, __unused void *arg, __unused int waitflag) +{ + socket_t newso = NULL; + struct nfs_callback_socket *ncbsp; + struct nfsmount *nmp; + struct timeval timeo, now; + int error, on = 1, ip; + + if (so == nfs4_cb_so) + ip = 4; + else if (so == nfs4_cb_so6) + ip = 6; + else + return; + + /* allocate/initialize a new nfs_callback_socket */ + MALLOC(ncbsp, struct nfs_callback_socket *, sizeof(struct nfs_callback_socket), M_TEMP, M_WAITOK); + if (!ncbsp) { + log(LOG_ERR, "nfs callback accept: no memory for new socket\n"); + return; + } + bzero(ncbsp, sizeof(*ncbsp)); + ncbsp->ncbs_saddr.ss_len = (ip == 4) ? sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6); + nfs_rpc_record_state_init(&ncbsp->ncbs_rrs); + + /* accept a new socket */ + error = sock_accept(so, (struct sockaddr*)&ncbsp->ncbs_saddr, + ncbsp->ncbs_saddr.ss_len, MSG_DONTWAIT, + nfs4_cb_rcv, ncbsp, &newso); + if (error) { + log(LOG_INFO, "nfs callback accept: error %d accepting IPv%d socket\n", error, ip); + FREE(ncbsp, M_TEMP); + return; + } + + /* set up the new socket */ + /* receive timeout shouldn't matter. If timeout on send, we'll want to drop the socket */ + timeo.tv_usec = 0; + timeo.tv_sec = 60; + error = sock_setsockopt(newso, SOL_SOCKET, SO_RCVTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback socket: error %d setting IPv%d socket rx timeout\n", error, ip); + error = sock_setsockopt(newso, SOL_SOCKET, SO_SNDTIMEO, &timeo, sizeof(timeo)); + if (error) + log(LOG_INFO, "nfs callback socket: error %d setting IPv%d socket tx timeout\n", error, ip); + sock_setsockopt(newso, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); + sock_setsockopt(newso, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)); + sock_setsockopt(newso, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on)); + sock_setsockopt(newso, SOL_SOCKET, SO_UPCALLCLOSEWAIT, &on, sizeof(on)); + + ncbsp->ncbs_so = newso; + microuptime(&now); + ncbsp->ncbs_stamp = now.tv_sec; + + lck_mtx_lock(nfs_global_mutex); + + /* add it to the list */ + TAILQ_INSERT_HEAD(&nfs4_cb_socks, ncbsp, ncbs_link); + + /* verify it's from a host we have mounted */ + TAILQ_FOREACH(nmp, &nfs4_cb_mounts, nm_cblink) { + /* check if socket's source address matches this mount's server address */ + if (!nmp->nm_saddr) + continue; + if (nfs_sockaddr_cmp((struct sockaddr*)&ncbsp->ncbs_saddr, nmp->nm_saddr) == 0) + break; + } + if (!nmp) /* we don't want this socket, mark it dead */ + ncbsp->ncbs_flags |= NCBSOCK_DEAD; + + /* make sure the callback socket cleanup timer is running */ + /* (shorten the timer if we've got a socket we don't want) */ + if (!nfs4_callback_timer_on) { + nfs4_callback_timer_on = 1; + nfs_interval_timer_start(nfs4_callback_timer_call, + !nmp ? 500 : (NFS4_CB_TIMER_PERIOD * 1000)); + } else if (!nmp && (nfs4_callback_timer_on < 2)) { + nfs4_callback_timer_on = 2; + thread_call_cancel(nfs4_callback_timer_call); + nfs_interval_timer_start(nfs4_callback_timer_call, 500); + } + + lck_mtx_unlock(nfs_global_mutex); +} + +/* + * Receive mbufs from callback sockets into RPC records and process each record. + * Detect connection has been closed and shut down. + */ +void +nfs4_cb_rcv(socket_t so, void *arg, __unused int waitflag) +{ + struct nfs_callback_socket *ncbsp = arg; + struct timespec ts = {1,0}; + struct timeval now; + mbuf_t m; + int error = 0, recv = 1; + + lck_mtx_lock(nfs_global_mutex); + while (ncbsp->ncbs_flags & NCBSOCK_UPCALL) { + /* wait if upcall is already in progress */ + ncbsp->ncbs_flags |= NCBSOCK_UPCALLWANT; + msleep(ncbsp, nfs_global_mutex, PSOCK, "cbupcall", &ts); + } + ncbsp->ncbs_flags |= NCBSOCK_UPCALL; + lck_mtx_unlock(nfs_global_mutex); + + /* loop while we make error-free progress */ + while (!error && recv) { + error = nfs_rpc_record_read(so, &ncbsp->ncbs_rrs, MSG_DONTWAIT, &recv, &m); + if (m) /* handle the request */ + error = nfs4_cb_handler(ncbsp, m); + } + + /* note: no error and no data indicates server closed its end */ + if ((error != EWOULDBLOCK) && (error || !recv)) { + /* + * Socket is either being closed or should be. + * We can't close the socket in the context of the upcall. + * So we mark it as dead and leave it for the cleanup timer to reap. + */ + ncbsp->ncbs_stamp = 0; + ncbsp->ncbs_flags |= NCBSOCK_DEAD; + } else { + microuptime(&now); + ncbsp->ncbs_stamp = now.tv_sec; + } + + lck_mtx_lock(nfs_global_mutex); + ncbsp->ncbs_flags &= ~NCBSOCK_UPCALL; + lck_mtx_unlock(nfs_global_mutex); + wakeup(ncbsp); +} + +/* + * Handle an NFS callback channel request. + */ +int +nfs4_cb_handler(struct nfs_callback_socket *ncbsp, mbuf_t mreq) +{ + socket_t so = ncbsp->ncbs_so; + struct nfsm_chain nmreq, nmrep; + mbuf_t mhead = NULL, mrest = NULL, m; + struct msghdr msg; + struct nfsmount *nmp; + fhandle_t fh; + nfsnode_t np; + nfs_stateid stateid; + uint32_t bitmap[NFS_ATTR_BITMAP_LEN], rbitmap[NFS_ATTR_BITMAP_LEN], bmlen, truncate, attrbytes; + uint32_t val, xid, procnum, taglen, cbid, numops, op, status; + uint32_t auth_type, auth_len; + uint32_t numres, *pnumres; + int error = 0, replen, len; + size_t sentlen = 0; + + xid = numops = op = status = procnum = taglen = cbid = 0; + + nfsm_chain_dissect_init(error, &nmreq, mreq); + nfsm_chain_get_32(error, &nmreq, xid); // RPC XID + nfsm_chain_get_32(error, &nmreq, val); // RPC Call + nfsm_assert(error, (val == RPC_CALL), EBADRPC); + nfsm_chain_get_32(error, &nmreq, val); // RPC Version + nfsm_assert(error, (val == RPC_VER2), ERPCMISMATCH); + nfsm_chain_get_32(error, &nmreq, val); // RPC Program Number + nfsm_assert(error, (val == NFS4_CALLBACK_PROG), EPROGUNAVAIL); + nfsm_chain_get_32(error, &nmreq, val); // NFS Callback Program Version Number + nfsm_assert(error, (val == NFS4_CALLBACK_PROG_VERSION), EPROGMISMATCH); + nfsm_chain_get_32(error, &nmreq, procnum); // NFS Callback Procedure Number + nfsm_assert(error, (procnum <= NFSPROC4_CB_COMPOUND), EPROCUNAVAIL); + + /* Handle authentication */ + /* XXX just ignore auth for now - handling kerberos may be tricky */ + nfsm_chain_get_32(error, &nmreq, auth_type); // RPC Auth Flavor + nfsm_chain_get_32(error, &nmreq, auth_len); // RPC Auth Length + nfsm_assert(error, (auth_len <= RPCAUTH_MAXSIZ), EBADRPC); + if (!error && (auth_len > 0)) + nfsm_chain_adv(error, &nmreq, nfsm_rndup(auth_len)); + nfsm_chain_adv(error, &nmreq, NFSX_UNSIGNED); // verifier flavor (should be AUTH_NONE) + nfsm_chain_get_32(error, &nmreq, auth_len); // verifier length + nfsm_assert(error, (auth_len <= RPCAUTH_MAXSIZ), EBADRPC); + if (!error && (auth_len > 0)) + nfsm_chain_adv(error, &nmreq, nfsm_rndup(auth_len)); + if (error) { + status = error; + error = 0; + goto nfsmout; + } + + switch (procnum) { + case NFSPROC4_CB_NULL: + status = NFSERR_RETVOID; + break; + case NFSPROC4_CB_COMPOUND: + /* tag, minorversion, cb ident, numops, op array */ + nfsm_chain_get_32(error, &nmreq, taglen); /* tag length */ + nfsm_assert(error, (val <= NFS4_OPAQUE_LIMIT), EBADRPC); + + /* start building the body of the response */ + nfsm_mbuf_get(error, &mrest, nfsm_rndup(taglen) + 5*NFSX_UNSIGNED); + nfsm_chain_init(&nmrep, mrest); + + /* copy tag from request to response */ + nfsm_chain_add_32(error, &nmrep, taglen); /* tag length */ + for (len = (int)taglen; !error && (len > 0); len -= NFSX_UNSIGNED) { + nfsm_chain_get_32(error, &nmreq, val); + nfsm_chain_add_32(error, &nmrep, val); + } + + /* insert number of results placeholder */ + numres = 0; + nfsm_chain_add_32(error, &nmrep, numres); + pnumres = (uint32_t*)(nmrep.nmc_ptr - NFSX_UNSIGNED); + + nfsm_chain_get_32(error, &nmreq, val); /* minorversion */ + nfsm_assert(error, (val == 0), NFSERR_MINOR_VERS_MISMATCH); + nfsm_chain_get_32(error, &nmreq, cbid); /* callback ID */ + nfsm_chain_get_32(error, &nmreq, numops); /* number of operations */ + if (error) { + if ((error == EBADRPC) || (error == NFSERR_MINOR_VERS_MISMATCH)) + status = error; + else if ((error == ENOBUFS) || (error == ENOMEM)) + status = NFSERR_RESOURCE; + else + status = NFSERR_SERVERFAULT; + error = 0; + nfsm_chain_null(&nmrep); + goto nfsmout; + } + /* match the callback ID to a registered mount */ + lck_mtx_lock(nfs_global_mutex); + TAILQ_FOREACH(nmp, &nfs4_cb_mounts, nm_cblink) { + if (nmp->nm_cbid != cbid) + continue; + /* verify socket's source address matches this mount's server address */ + if (!nmp->nm_saddr) + continue; + if (nfs_sockaddr_cmp((struct sockaddr*)&ncbsp->ncbs_saddr, nmp->nm_saddr) == 0) + break; + } + /* mark the NFS mount as busy */ + if (nmp) + nmp->nm_cbrefs++; + lck_mtx_unlock(nfs_global_mutex); + if (!nmp) { + /* if no mount match, just drop socket. */ + error = EPERM; + nfsm_chain_null(&nmrep); + goto out; + } + + /* process ops, adding results to mrest */ + while (numops > 0) { + numops--; + nfsm_chain_get_32(error, &nmreq, op); + if (error) + break; + switch (op) { + case NFS_OP_CB_GETATTR: + // (FH, BITMAP) -> (STATUS, BITMAP, ATTRS) + np = NULL; + nfsm_chain_get_fh(error, &nmreq, NFS_VER4, &fh); + bmlen = NFS_ATTR_BITMAP_LEN; + nfsm_chain_get_bitmap(error, &nmreq, bitmap, bmlen); + if (error) { + status = error; + error = 0; + numops = 0; /* don't process any more ops */ + } else { + /* find the node for the file handle */ + error = nfs_nget(nmp->nm_mountp, NULL, NULL, fh.fh_data, fh.fh_len, NULL, NULL, RPCAUTH_UNKNOWN, NG_NOCREATE, &np); + if (error || !np) { + status = NFSERR_BADHANDLE; + error = 0; + np = NULL; + numops = 0; /* don't process any more ops */ + } + } + nfsm_chain_add_32(error, &nmrep, op); + nfsm_chain_add_32(error, &nmrep, status); + if (!error && (status == EBADRPC)) + error = status; + if (np) { + /* only allow returning size, change, and mtime attrs */ + NFS_CLEAR_ATTRIBUTES(&rbitmap); + attrbytes = 0; + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_CHANGE)) { + NFS_BITMAP_SET(&rbitmap, NFS_FATTR_CHANGE); + attrbytes += 2 * NFSX_UNSIGNED; + } + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_SIZE)) { + NFS_BITMAP_SET(&rbitmap, NFS_FATTR_SIZE); + attrbytes += 2 * NFSX_UNSIGNED; + } + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_TIME_MODIFY)) { + NFS_BITMAP_SET(&rbitmap, NFS_FATTR_TIME_MODIFY); + attrbytes += 3 * NFSX_UNSIGNED; + } + nfsm_chain_add_bitmap(error, &nmrep, rbitmap, NFS_ATTR_BITMAP_LEN); + nfsm_chain_add_32(error, &nmrep, attrbytes); + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_CHANGE)) + nfsm_chain_add_64(error, &nmrep, + np->n_vattr.nva_change + ((np->n_flag & NMODIFIED) ? 1 : 0)); + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_SIZE)) + nfsm_chain_add_64(error, &nmrep, np->n_size); + if (NFS_BITMAP_ISSET(&bitmap, NFS_FATTR_TIME_MODIFY)) { + nfsm_chain_add_64(error, &nmrep, np->n_vattr.nva_timesec[NFSTIME_MODIFY]); + nfsm_chain_add_32(error, &nmrep, np->n_vattr.nva_timensec[NFSTIME_MODIFY]); + } + nfs_node_unlock(np); + vnode_put(NFSTOV(np)); + np = NULL; + } + /* + * If we hit an error building the reply, we can't easily back up. + * So we'll just update the status and hope the server ignores the + * extra garbage. + */ + break; + case NFS_OP_CB_RECALL: + // (STATEID, TRUNCATE, FH) -> (STATUS) + np = NULL; + nfsm_chain_get_stateid(error, &nmreq, &stateid); + nfsm_chain_get_32(error, &nmreq, truncate); + nfsm_chain_get_fh(error, &nmreq, NFS_VER4, &fh); + if (error) { + status = error; + error = 0; + numops = 0; /* don't process any more ops */ + } else { + /* find the node for the file handle */ + error = nfs_nget(nmp->nm_mountp, NULL, NULL, fh.fh_data, fh.fh_len, NULL, NULL, RPCAUTH_UNKNOWN, NG_NOCREATE, &np); + if (error || !np) { + status = NFSERR_BADHANDLE; + error = 0; + np = NULL; + numops = 0; /* don't process any more ops */ + } else if (!(np->n_openflags & N_DELEG_MASK) || + bcmp(&np->n_dstateid, &stateid, sizeof(stateid))) { + /* delegation stateid state doesn't match */ + status = NFSERR_BAD_STATEID; + numops = 0; /* don't process any more ops */ + } + if (!status) /* add node to recall queue, and wake socket thread */ + nfs4_delegation_return_enqueue(np); + if (np) { + nfs_node_unlock(np); + vnode_put(NFSTOV(np)); + } + } + nfsm_chain_add_32(error, &nmrep, op); + nfsm_chain_add_32(error, &nmrep, status); + if (!error && (status == EBADRPC)) + error = status; + break; + case NFS_OP_CB_ILLEGAL: + default: + nfsm_chain_add_32(error, &nmrep, NFS_OP_CB_ILLEGAL); + status = NFSERR_OP_ILLEGAL; + nfsm_chain_add_32(error, &nmrep, status); + numops = 0; /* don't process any more ops */ + break; + } + numres++; + } + + if (!status && error) { + if (error == EBADRPC) + status = error; + else if ((error == ENOBUFS) || (error == ENOMEM)) + status = NFSERR_RESOURCE; + else + status = NFSERR_SERVERFAULT; + error = 0; + } + + /* Now, set the numres field */ + *pnumres = txdr_unsigned(numres); + nfsm_chain_build_done(error, &nmrep); + nfsm_chain_null(&nmrep); + + /* drop the callback reference on the mount */ + lck_mtx_lock(nfs_global_mutex); + nmp->nm_cbrefs--; + if (!nmp->nm_cbid) + wakeup(&nmp->nm_cbrefs); + lck_mtx_unlock(nfs_global_mutex); + break; + } + +nfsmout: + if (status == EBADRPC) + OSAddAtomic64(1, &nfsstats.rpcinvalid); + + /* build reply header */ + error = mbuf_gethdr(MBUF_WAITOK, MBUF_TYPE_DATA, &mhead); + nfsm_chain_init(&nmrep, mhead); + nfsm_chain_add_32(error, &nmrep, 0); /* insert space for an RPC record mark */ + nfsm_chain_add_32(error, &nmrep, xid); + nfsm_chain_add_32(error, &nmrep, RPC_REPLY); + if ((status == ERPCMISMATCH) || (status & NFSERR_AUTHERR)) { + nfsm_chain_add_32(error, &nmrep, RPC_MSGDENIED); + if (status & NFSERR_AUTHERR) { + nfsm_chain_add_32(error, &nmrep, RPC_AUTHERR); + nfsm_chain_add_32(error, &nmrep, (status & ~NFSERR_AUTHERR)); + } else { + nfsm_chain_add_32(error, &nmrep, RPC_MISMATCH); + nfsm_chain_add_32(error, &nmrep, RPC_VER2); + nfsm_chain_add_32(error, &nmrep, RPC_VER2); + } + } else { + /* reply status */ + nfsm_chain_add_32(error, &nmrep, RPC_MSGACCEPTED); + /* XXX RPCAUTH_NULL verifier */ + nfsm_chain_add_32(error, &nmrep, RPCAUTH_NULL); + nfsm_chain_add_32(error, &nmrep, 0); + /* accepted status */ + switch (status) { + case EPROGUNAVAIL: + nfsm_chain_add_32(error, &nmrep, RPC_PROGUNAVAIL); + break; + case EPROGMISMATCH: + nfsm_chain_add_32(error, &nmrep, RPC_PROGMISMATCH); + nfsm_chain_add_32(error, &nmrep, NFS4_CALLBACK_PROG_VERSION); + nfsm_chain_add_32(error, &nmrep, NFS4_CALLBACK_PROG_VERSION); + break; + case EPROCUNAVAIL: + nfsm_chain_add_32(error, &nmrep, RPC_PROCUNAVAIL); + break; + case EBADRPC: + nfsm_chain_add_32(error, &nmrep, RPC_GARBAGE); + break; + default: + nfsm_chain_add_32(error, &nmrep, RPC_SUCCESS); + if (status != NFSERR_RETVOID) + nfsm_chain_add_32(error, &nmrep, status); + break; + } + } + nfsm_chain_build_done(error, &nmrep); + if (error) { + nfsm_chain_null(&nmrep); + goto out; + } + error = mbuf_setnext(nmrep.nmc_mcur, mrest); + if (error) { + printf("nfs cb: mbuf_setnext failed %d\n", error); + goto out; + } + mrest = NULL; + /* Calculate the size of the reply */ + replen = 0; + for (m = nmrep.nmc_mhead; m; m = mbuf_next(m)) + replen += mbuf_len(m); + mbuf_pkthdr_setlen(mhead, replen); + error = mbuf_pkthdr_setrcvif(mhead, NULL); + nfsm_chain_set_recmark(error, &nmrep, (replen - NFSX_UNSIGNED) | 0x80000000); + nfsm_chain_null(&nmrep); + + /* send the reply */ + bzero(&msg, sizeof(msg)); + error = sock_sendmbuf(so, &msg, mhead, 0, &sentlen); + mhead = NULL; + if (!error && ((int)sentlen != replen)) + error = EWOULDBLOCK; + if (error == EWOULDBLOCK) /* inability to send response is considered fatal */ + error = ETIMEDOUT; +out: + if (error) + nfsm_chain_cleanup(&nmrep); + if (mhead) + mbuf_freem(mhead); + if (mrest) mbuf_freem(mrest); + if (mreq) + mbuf_freem(mreq); + return (error); +} + + +/* + * Initialize an nfs_rpc_record_state structure. + */ +void +nfs_rpc_record_state_init(struct nfs_rpc_record_state *nrrsp) +{ + bzero(nrrsp, sizeof(*nrrsp)); + nrrsp->nrrs_markerleft = sizeof(nrrsp->nrrs_fragleft); +} + +/* + * Clean up an nfs_rpc_record_state structure. + */ +void +nfs_rpc_record_state_cleanup(struct nfs_rpc_record_state *nrrsp) +{ + if (nrrsp->nrrs_m) { + mbuf_freem(nrrsp->nrrs_m); + nrrsp->nrrs_m = nrrsp->nrrs_mlast = NULL; + } +} + +/* + * Read the next (marked) RPC record from the socket. + * + * *recvp returns if any data was received. + * *mp returns the next complete RPC record + */ +int +nfs_rpc_record_read(socket_t so, struct nfs_rpc_record_state *nrrsp, int flags, int *recvp, mbuf_t *mp) +{ + struct iovec aio; + struct msghdr msg; + size_t rcvlen; + int error = 0; + mbuf_t m; + + *recvp = 0; + *mp = NULL; + + /* read the TCP RPC record marker */ + while (!error && nrrsp->nrrs_markerleft) { + aio.iov_base = ((char*)&nrrsp->nrrs_fragleft + + sizeof(nrrsp->nrrs_fragleft) - nrrsp->nrrs_markerleft); + aio.iov_len = nrrsp->nrrs_markerleft; + bzero(&msg, sizeof(msg)); + msg.msg_iov = &aio; + msg.msg_iovlen = 1; + error = sock_receive(so, &msg, flags, &rcvlen); + if (error || !rcvlen) + break; + *recvp = 1; + nrrsp->nrrs_markerleft -= rcvlen; + if (nrrsp->nrrs_markerleft) + continue; + /* record marker complete */ + nrrsp->nrrs_fragleft = ntohl(nrrsp->nrrs_fragleft); + if (nrrsp->nrrs_fragleft & 0x80000000) { + nrrsp->nrrs_lastfrag = 1; + nrrsp->nrrs_fragleft &= ~0x80000000; + } + nrrsp->nrrs_reclen += nrrsp->nrrs_fragleft; + if (nrrsp->nrrs_reclen > NFS_MAXPACKET) { + /* This is SERIOUS! We are out of sync with the sender. */ + log(LOG_ERR, "impossible RPC record length (%d) on callback", nrrsp->nrrs_reclen); + error = EFBIG; + } + } + + /* read the TCP RPC record fragment */ + while (!error && !nrrsp->nrrs_markerleft && nrrsp->nrrs_fragleft) { + m = NULL; + rcvlen = nrrsp->nrrs_fragleft; + error = sock_receivembuf(so, NULL, &m, flags, &rcvlen); + if (error || !rcvlen || !m) + break; + *recvp = 1; + /* append mbufs to list */ + nrrsp->nrrs_fragleft -= rcvlen; + if (!nrrsp->nrrs_m) { + nrrsp->nrrs_m = m; + } else { + error = mbuf_setnext(nrrsp->nrrs_mlast, m); + if (error) { + printf("nfs tcp rcv: mbuf_setnext failed %d\n", error); + mbuf_freem(m); + break; + } + } + while (mbuf_next(m)) + m = mbuf_next(m); + nrrsp->nrrs_mlast = m; + } + + /* done reading fragment? */ + if (!error && !nrrsp->nrrs_markerleft && !nrrsp->nrrs_fragleft) { + /* reset socket fragment parsing state */ + nrrsp->nrrs_markerleft = sizeof(nrrsp->nrrs_fragleft); + if (nrrsp->nrrs_lastfrag) { + /* RPC record complete */ + *mp = nrrsp->nrrs_m; + /* reset socket record parsing state */ + nrrsp->nrrs_reclen = 0; + nrrsp->nrrs_m = nrrsp->nrrs_mlast = NULL; + nrrsp->nrrs_lastfrag = 0; + } + } + + return (error); +} + + + +/* + * The NFS client send routine. + * + * Send the given NFS request out the mount's socket. + * Holds nfs_sndlock() for the duration of this call. + * + * - check for request termination (sigintr) + * - wait for reconnect, if necessary + * - UDP: check the congestion window + * - make a copy of the request to send + * - UDP: update the congestion window + * - send the request + * + * If sent successfully, R_MUSTRESEND and R_RESENDERR are cleared. + * rexmit count is also updated if this isn't the first send. + * + * If the send is not successful, make sure R_MUSTRESEND is set. + * If this wasn't the first transmit, set R_RESENDERR. + * Also, undo any UDP congestion window changes made. + * + * If the error appears to indicate that the socket should + * be reconnected, mark the socket for reconnection. + * + * Only return errors when the request should be aborted. + */ +int +nfs_send(struct nfsreq *req, int wait) +{ + struct nfsmount *nmp; + struct nfs_socket *nso; + int error, error2, sotype, rexmit, slpflag = 0, needrecon; + struct msghdr msg; + struct sockaddr *sendnam; + mbuf_t mreqcopy; + size_t sentlen = 0; + struct timespec ts = { 2, 0 }; + +again: + error = nfs_sndlock(req); + if (error) { + lck_mtx_lock(&req->r_mtx); + req->r_error = error; + req->r_flags &= ~R_SENDING; + lck_mtx_unlock(&req->r_mtx); + return (error); + } + + error = nfs_sigintr(req->r_nmp, req, NULL, 0); + if (error) { + nfs_sndunlock(req); + lck_mtx_lock(&req->r_mtx); + req->r_error = error; + req->r_flags &= ~R_SENDING; + lck_mtx_unlock(&req->r_mtx); + return (error); + } + nmp = req->r_nmp; + sotype = nmp->nm_sotype; + + /* + * If it's a setup RPC but we're not in SETUP... must need reconnect. + * If it's a recovery RPC but the socket's not ready... must need reconnect. + */ + if (((req->r_flags & R_SETUP) && !(nmp->nm_sockflags & NMSOCK_SETUP)) || + ((req->r_flags & R_RECOVER) && !(nmp->nm_sockflags & NMSOCK_READY))) { + error = ETIMEDOUT; + nfs_sndunlock(req); + lck_mtx_lock(&req->r_mtx); + req->r_error = error; + req->r_flags &= ~R_SENDING; + lck_mtx_unlock(&req->r_mtx); + return (error); + } + + /* If the socket needs reconnection, do that now. */ + /* wait until socket is ready - unless this request is part of setup */ + lck_mtx_lock(&nmp->nm_lock); + if (!(nmp->nm_sockflags & NMSOCK_READY) && + !((nmp->nm_sockflags & NMSOCK_SETUP) && (req->r_flags & R_SETUP))) { + if (NMFLAG(nmp, INTR) && !(req->r_flags & R_NOINTR)) + slpflag |= PCATCH; + lck_mtx_unlock(&nmp->nm_lock); + nfs_sndunlock(req); + if (!wait) { + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_SENDING; + req->r_flags |= R_MUSTRESEND; + req->r_rtt = 0; + lck_mtx_unlock(&req->r_mtx); + return (0); + } + NFS_SOCK_DBG("nfs_send: 0x%llx wait reconnect\n", req->r_xid); + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_MUSTRESEND; + req->r_rtt = 0; + lck_mtx_unlock(&req->r_mtx); + lck_mtx_lock(&nmp->nm_lock); + while (!(nmp->nm_sockflags & NMSOCK_READY)) { + /* don't bother waiting if the socket thread won't be reconnecting it */ + if (nmp->nm_state & NFSSTA_FORCE) { + error = EIO; + break; + } + if (NMFLAG(nmp, SOFT) && (nmp->nm_reconnect_start > 0)) { + struct timeval now; + microuptime(&now); + if ((now.tv_sec - nmp->nm_reconnect_start) >= 8) { + /* soft mount in reconnect for a while... terminate ASAP */ + OSAddAtomic64(1, &nfsstats.rpctimeouts); + req->r_flags |= R_SOFTTERM; + req->r_error = error = ETIMEDOUT; + break; + } + } + /* make sure socket thread is running, then wait */ + nfs_mount_sock_thread_wake(nmp); + if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 1))) + break; + msleep(req, &nmp->nm_lock, slpflag|PSOCK, "nfsconnectwait", &ts); + slpflag = 0; + } + lck_mtx_unlock(&nmp->nm_lock); + if (error) { + lck_mtx_lock(&req->r_mtx); + req->r_error = error; + req->r_flags &= ~R_SENDING; + lck_mtx_unlock(&req->r_mtx); + return (error); + } + goto again; + } + nso = nmp->nm_nso; + /* note that we're using the mount's socket to do the send */ + nmp->nm_state |= NFSSTA_SENDING; /* will be cleared by nfs_sndunlock() */ + lck_mtx_unlock(&nmp->nm_lock); + if (!nso) { + nfs_sndunlock(req); + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_SENDING; + req->r_flags |= R_MUSTRESEND; + req->r_rtt = 0; + lck_mtx_unlock(&req->r_mtx); + return (0); + } + + lck_mtx_lock(&req->r_mtx); + rexmit = (req->r_flags & R_SENT); + + if (sotype == SOCK_DGRAM) { + lck_mtx_lock(&nmp->nm_lock); + if (!(req->r_flags & R_CWND) && (nmp->nm_sent >= nmp->nm_cwnd)) { + /* if we can't send this out yet, wait on the cwnd queue */ + slpflag = (NMFLAG(nmp, INTR) && req->r_thread) ? PCATCH : 0; + lck_mtx_unlock(&nmp->nm_lock); + nfs_sndunlock(req); + req->r_flags &= ~R_SENDING; + req->r_flags |= R_MUSTRESEND; + lck_mtx_unlock(&req->r_mtx); + if (!wait) { + req->r_rtt = 0; + return (0); + } + lck_mtx_lock(&nmp->nm_lock); + while (nmp->nm_sent >= nmp->nm_cwnd) { + if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 1))) + break; + TAILQ_INSERT_TAIL(&nmp->nm_cwndq, req, r_cchain); + msleep(req, &nmp->nm_lock, slpflag | (PZERO - 1), "nfswaitcwnd", &ts); + slpflag = 0; + if ((req->r_cchain.tqe_next != NFSREQNOLIST)) { + TAILQ_REMOVE(&nmp->nm_cwndq, req, r_cchain); + req->r_cchain.tqe_next = NFSREQNOLIST; + } + } + lck_mtx_unlock(&nmp->nm_lock); + goto again; + } + /* + * We update these *before* the send to avoid racing + * against others who may be looking to send requests. + */ + if (!rexmit) { + /* first transmit */ + req->r_flags |= R_CWND; + nmp->nm_sent += NFS_CWNDSCALE; + } else { + /* + * When retransmitting, turn timing off + * and divide congestion window by 2. + */ + req->r_flags &= ~R_TIMING; + nmp->nm_cwnd >>= 1; + if (nmp->nm_cwnd < NFS_CWNDSCALE) + nmp->nm_cwnd = NFS_CWNDSCALE; + } + lck_mtx_unlock(&nmp->nm_lock); + } + + req->r_flags &= ~R_MUSTRESEND; + lck_mtx_unlock(&req->r_mtx); + + error = mbuf_copym(req->r_mhead, 0, MBUF_COPYALL, + wait ? MBUF_WAITOK : MBUF_DONTWAIT, &mreqcopy); + if (error) { + if (wait) + log(LOG_INFO, "nfs_send: mbuf copy failed %d\n", error); + nfs_sndunlock(req); + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_SENDING; + req->r_flags |= R_MUSTRESEND; + req->r_rtt = 0; + lck_mtx_unlock(&req->r_mtx); + return (0); + } + + bzero(&msg, sizeof(msg)); + if ((sotype != SOCK_STREAM) && !sock_isconnected(nso->nso_so) && ((sendnam = nmp->nm_saddr))) { + msg.msg_name = (caddr_t)sendnam; + msg.msg_namelen = sendnam->sa_len; + } + error = sock_sendmbuf(nso->nso_so, &msg, mreqcopy, 0, &sentlen); +#ifdef NFS_SOCKET_DEBUGGING + if (error || (sentlen != req->r_mreqlen)) + NFS_SOCK_DBG("nfs_send: 0x%llx sent %d/%d error %d\n", + req->r_xid, (int)sentlen, (int)req->r_mreqlen, error); +#endif + if (!error && (sentlen != req->r_mreqlen)) + error = EWOULDBLOCK; + needrecon = ((sotype == SOCK_STREAM) && sentlen && (sentlen != req->r_mreqlen)); + + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_SENDING; + req->r_rtt = 0; + if (rexmit && (++req->r_rexmit > NFS_MAXREXMIT)) + req->r_rexmit = NFS_MAXREXMIT; + + if (!error) { + /* SUCCESS */ + req->r_flags &= ~R_RESENDERR; + if (rexmit) + OSAddAtomic64(1, &nfsstats.rpcretries); + req->r_flags |= R_SENT; + if (req->r_flags & R_WAITSENT) { + req->r_flags &= ~R_WAITSENT; + wakeup(req); + } + nfs_sndunlock(req); + lck_mtx_unlock(&req->r_mtx); + return (0); + } + + /* send failed */ + req->r_flags |= R_MUSTRESEND; + if (rexmit) + req->r_flags |= R_RESENDERR; + if ((error == EINTR) || (error == ERESTART)) + req->r_error = error; + lck_mtx_unlock(&req->r_mtx); + + if (sotype == SOCK_DGRAM) { + /* + * Note: even though a first send may fail, we consider + * the request sent for congestion window purposes. + * So we don't need to undo any of the changes made above. + */ + /* + * Socket errors ignored for connectionless sockets?? + * For now, ignore them all + */ + if ((error != EINTR) && (error != ERESTART) && + (error != EWOULDBLOCK) && (error != EIO) && (nso == nmp->nm_nso)) { + int clearerror = 0, optlen = sizeof(clearerror); + sock_getsockopt(nso->nso_so, SOL_SOCKET, SO_ERROR, &clearerror, &optlen); +#ifdef NFS_SOCKET_DEBUGGING + if (clearerror) + NFS_SOCK_DBG("nfs_send: ignoring UDP socket error %d so %d\n", + error, clearerror); +#endif + } + } + + /* check if it appears we should reconnect the socket */ + switch (error) { + case EWOULDBLOCK: + /* if send timed out, reconnect if on TCP */ + if (sotype != SOCK_STREAM) + break; + case EPIPE: + case EADDRNOTAVAIL: + case ENETDOWN: + case ENETUNREACH: + case ENETRESET: + case ECONNABORTED: + case ECONNRESET: + case ENOTCONN: + case ESHUTDOWN: + case ECONNREFUSED: + case EHOSTDOWN: + case EHOSTUNREACH: + needrecon = 1; + break; + } + if (needrecon && (nso == nmp->nm_nso)) { /* mark socket as needing reconnect */ + NFS_SOCK_DBG("nfs_send: 0x%llx need reconnect %d\n", req->r_xid, error); + nfs_need_reconnect(nmp); + } + + nfs_sndunlock(req); + + /* + * Don't log some errors: + * EPIPE errors may be common with servers that drop idle connections. + * EADDRNOTAVAIL may occur on network transitions. + * ENOTCONN may occur under some network conditions. + */ + if ((error == EPIPE) || (error == EADDRNOTAVAIL) || (error == ENOTCONN)) + error = 0; + if (error && (error != EINTR) && (error != ERESTART)) + log(LOG_INFO, "nfs send error %d for server %s\n", error, + !req->r_nmp ? "" : + vfs_statfs(req->r_nmp->nm_mountp)->f_mntfromname); + + if (nfs_is_dead(error, nmp)) + error = EIO; + + /* prefer request termination error over other errors */ + error2 = nfs_sigintr(req->r_nmp, req, req->r_thread, 0); + if (error2) + error = error2; + + /* only allow the following errors to be returned */ + if ((error != EINTR) && (error != ERESTART) && (error != EIO) && + (error != ENXIO) && (error != ETIMEDOUT)) + error = 0; + return (error); +} + +/* + * NFS client socket upcalls + * + * Pull RPC replies out of an NFS mount's socket and match them + * up with the pending request. + * + * The datagram code is simple because we always get whole + * messages out of the socket. + * + * The stream code is more involved because we have to parse + * the RPC records out of the stream. + */ + +/* NFS client UDP socket upcall */ +void +nfs_udp_rcv(socket_t so, void *arg, __unused int waitflag) +{ + struct nfsmount *nmp = arg; + struct nfs_socket *nso = nmp->nm_nso; + size_t rcvlen; + mbuf_t m; + int error = 0; + + if (nmp->nm_sockflags & NMSOCK_CONNECTING) + return; + + do { + /* make sure we're on the current socket */ + if (!nso || (nso->nso_so != so)) + return; + + m = NULL; + rcvlen = 1000000; + error = sock_receivembuf(so, NULL, &m, MSG_DONTWAIT, &rcvlen); + if (m) + nfs_request_match_reply(nmp, m); + } while (m && !error); + + if (error && (error != EWOULDBLOCK)) { + /* problems with the socket... mark for reconnection */ + NFS_SOCK_DBG("nfs_udp_rcv: need reconnect %d\n", error); + nfs_need_reconnect(nmp); + } +} + +/* NFS client TCP socket upcall */ +void +nfs_tcp_rcv(socket_t so, void *arg, __unused int waitflag) +{ + struct nfsmount *nmp = arg; + struct nfs_socket *nso = nmp->nm_nso; + struct nfs_rpc_record_state nrrs; + mbuf_t m; + int error = 0; + int recv = 1; + + if (nmp->nm_sockflags & NMSOCK_CONNECTING) + return; + + /* make sure we're on the current socket */ + lck_mtx_lock(&nmp->nm_lock); + nso = nmp->nm_nso; + if (!nso || (nso->nso_so != so) || (nmp->nm_sockflags & (NMSOCK_DISCONNECTING))) { + lck_mtx_unlock(&nmp->nm_lock); + return; + } + lck_mtx_unlock(&nmp->nm_lock); + + /* make sure this upcall should be trying to do work */ + lck_mtx_lock(&nso->nso_lock); + if (nso->nso_flags & (NSO_UPCALL|NSO_DISCONNECTING|NSO_DEAD)) { + lck_mtx_unlock(&nso->nso_lock); + return; + } + nso->nso_flags |= NSO_UPCALL; + nrrs = nso->nso_rrs; + lck_mtx_unlock(&nso->nso_lock); + + /* loop while we make error-free progress */ + while (!error && recv) { + error = nfs_rpc_record_read(so, &nrrs, MSG_DONTWAIT, &recv, &m); + if (m) /* match completed response with request */ + nfs_request_match_reply(nmp, m); + } + + lck_mtx_lock(&nmp->nm_lock); + if (nmp->nm_nso == nso) { + /* still the same socket, so update socket's RPC parsing state */ + lck_mtx_unlock(&nmp->nm_lock); + lck_mtx_lock(&nso->nso_lock); + nso->nso_rrs = nrrs; + nso->nso_flags &= ~NSO_UPCALL; + lck_mtx_unlock(&nso->nso_lock); + if (nmp->nm_sockflags & NMSOCK_DISCONNECTING) + wakeup(&nmp->nm_sockflags); + } else { + lck_mtx_unlock(&nmp->nm_lock); + } +#ifdef NFS_SOCKET_DEBUGGING + if (!recv && (error != EWOULDBLOCK)) + NFS_SOCK_DBG("nfs_tcp_rcv: got nothing, error %d, got FIN?\n", error); +#endif + /* note: no error and no data indicates server closed its end */ + if ((error != EWOULDBLOCK) && (error || !recv)) { + /* problems with the socket... mark for reconnection */ + NFS_SOCK_DBG("nfs_tcp_rcv: need reconnect %d\n", error); + nfs_need_reconnect(nmp); + } +} + +/* + * "poke" a socket to try to provoke any pending errors + */ +void +nfs_sock_poke(struct nfsmount *nmp) +{ + struct iovec aio; + struct msghdr msg; + size_t len; + int error = 0; + int dummy; + + lck_mtx_lock(&nmp->nm_lock); + if ((nmp->nm_sockflags & NMSOCK_UNMOUNT) || + !(nmp->nm_sockflags & NMSOCK_READY) || !nmp->nm_nso || !nmp->nm_nso->nso_so) { + lck_mtx_unlock(&nmp->nm_lock); + return; + } + lck_mtx_unlock(&nmp->nm_lock); + aio.iov_base = &dummy; + aio.iov_len = 0; + len = 0; + bzero(&msg, sizeof(msg)); + msg.msg_iov = &aio; + msg.msg_iovlen = 1; + error = sock_send(nmp->nm_nso->nso_so, &msg, MSG_DONTWAIT, &len); + NFS_SOCK_DBG("nfs_sock_poke: error %d\n", error); + nfs_is_dead(error, nmp); +} + +/* + * Match an RPC reply with the corresponding request + */ +void +nfs_request_match_reply(struct nfsmount *nmp, mbuf_t mrep) +{ + struct nfsreq *req; + struct nfsm_chain nmrep; + u_int32_t reply = 0, rxid = 0; + int error = 0, asyncioq, t1; + + /* Get the xid and check that it is an rpc reply */ + nfsm_chain_dissect_init(error, &nmrep, mrep); + nfsm_chain_get_32(error, &nmrep, rxid); + nfsm_chain_get_32(error, &nmrep, reply); + if (error || (reply != RPC_REPLY)) { + OSAddAtomic64(1, &nfsstats.rpcinvalid); + mbuf_freem(mrep); + return; + } + + /* + * Loop through the request list to match up the reply + * Iff no match, just drop it. + */ + lck_mtx_lock(nfs_request_mutex); + TAILQ_FOREACH(req, &nfs_reqq, r_chain) { + if (req->r_nmrep.nmc_mhead || (rxid != R_XID32(req->r_xid))) + continue; + /* looks like we have it, grab lock and double check */ + lck_mtx_lock(&req->r_mtx); + if (req->r_nmrep.nmc_mhead || (rxid != R_XID32(req->r_xid))) { + lck_mtx_unlock(&req->r_mtx); + continue; + } + /* Found it.. */ + req->r_nmrep = nmrep; + lck_mtx_lock(&nmp->nm_lock); + if (nmp->nm_sotype == SOCK_DGRAM) { + /* + * Update congestion window. + * Do the additive increase of one rpc/rtt. + */ + FSDBG(530, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd); + if (nmp->nm_cwnd <= nmp->nm_sent) { + nmp->nm_cwnd += + ((NFS_CWNDSCALE * NFS_CWNDSCALE) + + (nmp->nm_cwnd >> 1)) / nmp->nm_cwnd; + if (nmp->nm_cwnd > NFS_MAXCWND) + nmp->nm_cwnd = NFS_MAXCWND; + } + if (req->r_flags & R_CWND) { + nmp->nm_sent -= NFS_CWNDSCALE; + req->r_flags &= ~R_CWND; + } + if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) { + /* congestion window is open, poke the cwnd queue */ + struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq); + TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain); + req2->r_cchain.tqe_next = NFSREQNOLIST; + wakeup(req2); + } + } + /* + * Update rtt using a gain of 0.125 on the mean + * and a gain of 0.25 on the deviation. + */ + if (req->r_flags & R_TIMING) { + /* + * Since the timer resolution of + * NFS_HZ is so course, it can often + * result in r_rtt == 0. Since + * r_rtt == N means that the actual + * rtt is between N+dt and N+2-dt ticks, + * add 1. + */ + if (proct[req->r_procnum] == 0) + panic("nfs_request_match_reply: proct[%d] is zero", req->r_procnum); + t1 = req->r_rtt + 1; + t1 -= (NFS_SRTT(req) >> 3); + NFS_SRTT(req) += t1; + if (t1 < 0) + t1 = -t1; + t1 -= (NFS_SDRTT(req) >> 2); + NFS_SDRTT(req) += t1; + } + nmp->nm_timeouts = 0; + lck_mtx_unlock(&nmp->nm_lock); + /* signal anyone waiting on this request */ + wakeup(req); + asyncioq = (req->r_callback.rcb_func != NULL); + if (nfs_request_using_gss(req)) + nfs_gss_clnt_rpcdone(req); + lck_mtx_unlock(&req->r_mtx); + lck_mtx_unlock(nfs_request_mutex); + /* if it's an async RPC with a callback, queue it up */ + if (asyncioq) + nfs_asyncio_finish(req); + break; + } + + if (!req) { + /* not matched to a request, so drop it. */ + lck_mtx_unlock(nfs_request_mutex); + OSAddAtomic64(1, &nfsstats.rpcunexpected); + mbuf_freem(mrep); + } +} + +/* + * Wait for the reply for a given request... + * ...potentially resending the request if necessary. + */ +int +nfs_wait_reply(struct nfsreq *req) +{ + struct timespec ts = { 2, 0 }; + int error = 0, slpflag, first = 1; + + if (req->r_nmp && NMFLAG(req->r_nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR)) + slpflag = PCATCH; + else + slpflag = 0; + + lck_mtx_lock(&req->r_mtx); + while (!req->r_nmrep.nmc_mhead) { + if ((error = nfs_sigintr(req->r_nmp, req, first ? NULL : req->r_thread, 0))) + break; + if (((error = req->r_error)) || req->r_nmrep.nmc_mhead) + break; + /* check if we need to resend */ + if (req->r_flags & R_MUSTRESEND) { + NFS_SOCK_DBG("nfs wait resend: p %d x 0x%llx f 0x%x rtt %d\n", + req->r_procnum, req->r_xid, req->r_flags, req->r_rtt); + req->r_flags |= R_SENDING; + lck_mtx_unlock(&req->r_mtx); + if (nfs_request_using_gss(req)) { + /* + * It's an RPCSEC_GSS request. + * Can't just resend the original request + * without bumping the cred sequence number. + * Go back and re-build the request. + */ + lck_mtx_lock(&req->r_mtx); + req->r_flags &= ~R_SENDING; + lck_mtx_unlock(&req->r_mtx); + return (EAGAIN); + } + error = nfs_send(req, 1); + lck_mtx_lock(&req->r_mtx); + NFS_SOCK_DBG("nfs wait resend: p %d x 0x%llx f 0x%x rtt %d err %d\n", + req->r_procnum, req->r_xid, req->r_flags, req->r_rtt, error); + if (error) + break; + if (((error = req->r_error)) || req->r_nmrep.nmc_mhead) + break; + } + /* need to poll if we're P_NOREMOTEHANG */ + if (nfs_noremotehang(req->r_thread)) + ts.tv_sec = 1; + msleep(req, &req->r_mtx, slpflag | (PZERO - 1), "nfswaitreply", &ts); + first = slpflag = 0; + } + lck_mtx_unlock(&req->r_mtx); + + return (error); +} + +/* + * An NFS request goes something like this: + * (nb: always frees up mreq mbuf list) + * nfs_request_create() + * - allocates a request struct if one is not provided + * - initial fill-in of the request struct + * nfs_request_add_header() + * - add the RPC header + * nfs_request_send() + * - link it into list + * - call nfs_send() for first transmit + * nfs_request_wait() + * - call nfs_wait_reply() to wait for the reply + * nfs_request_finish() + * - break down rpc header and return with error or nfs reply + * pointed to by nmrep. + * nfs_request_rele() + * nfs_request_destroy() + * - clean up the request struct + * - free the request struct if it was allocated by nfs_request_create() + */ + +/* + * Set up an NFS request struct (allocating if no request passed in). + */ +int +nfs_request_create( + nfsnode_t np, + mount_t mp, /* used only if !np */ + struct nfsm_chain *nmrest, + int procnum, + thread_t thd, + kauth_cred_t cred, + struct nfsreq **reqp) +{ + struct nfsreq *req, *newreq = NULL; + struct nfsmount *nmp; + + req = *reqp; + if (!req) { + /* allocate a new NFS request structure */ + MALLOC_ZONE(newreq, struct nfsreq*, sizeof(*newreq), M_NFSREQ, M_WAITOK); + if (!newreq) { + mbuf_freem(nmrest->nmc_mhead); + nmrest->nmc_mhead = NULL; + return (ENOMEM); + } + req = newreq; + } + + bzero(req, sizeof(*req)); + if (req == newreq) + req->r_flags = R_ALLOCATED; + + nmp = VFSTONFS(np ? NFSTOMP(np) : mp); + if (!nmp) { + if (newreq) + FREE_ZONE(newreq, sizeof(*newreq), M_NFSREQ); + return (ENXIO); + } + lck_mtx_lock(&nmp->nm_lock); + if ((nmp->nm_state & (NFSSTA_FORCE|NFSSTA_TIMEO)) == + (NFSSTA_FORCE|NFSSTA_TIMEO)) { + lck_mtx_unlock(&nmp->nm_lock); + mbuf_freem(nmrest->nmc_mhead); + nmrest->nmc_mhead = NULL; + if (newreq) + FREE_ZONE(newreq, sizeof(*newreq), M_NFSREQ); + return (ENXIO); + } + + if ((nmp->nm_vers != NFS_VER4) && (procnum >= 0) && (procnum < NFS_NPROCS)) + OSAddAtomic64(1, &nfsstats.rpccnt[procnum]); + if ((nmp->nm_vers == NFS_VER4) && (procnum != NFSPROC4_COMPOUND) && (procnum != NFSPROC4_NULL)) + panic("nfs_request: invalid NFSv4 RPC request %d\n", procnum); + + lck_mtx_init(&req->r_mtx, nfs_request_grp, LCK_ATTR_NULL); + req->r_nmp = nmp; + req->r_np = np; + req->r_thread = thd; + if (!thd) + req->r_flags |= R_NOINTR; + if (IS_VALID_CRED(cred)) { + kauth_cred_ref(cred); + req->r_cred = cred; + } + req->r_procnum = procnum; + if (proct[procnum] > 0) + req->r_flags |= R_TIMING; + req->r_nmrep.nmc_mhead = NULL; + SLIST_INIT(&req->r_gss_seqlist); + req->r_achain.tqe_next = NFSREQNOLIST; + req->r_rchain.tqe_next = NFSREQNOLIST; + req->r_cchain.tqe_next = NFSREQNOLIST; + + /* set auth flavor to use for request */ + if (!req->r_cred) + req->r_auth = RPCAUTH_NONE; + else if (req->r_np && (req->r_np->n_auth != RPCAUTH_INVALID)) + req->r_auth = req->r_np->n_auth; + else + req->r_auth = nmp->nm_auth; + + lck_mtx_unlock(&nmp->nm_lock); + + /* move the request mbuf chain to the nfsreq */ + req->r_mrest = nmrest->nmc_mhead; + nmrest->nmc_mhead = NULL; + + req->r_flags |= R_INITTED; + req->r_refs = 1; + if (newreq) + *reqp = req; + return (0); +} + +/* + * Clean up and free an NFS request structure. + */ +void +nfs_request_destroy(struct nfsreq *req) +{ + struct nfsmount *nmp = req->r_np ? NFSTONMP(req->r_np) : req->r_nmp; + struct gss_seq *gsp, *ngsp; + struct timespec ts = { 1, 0 }; + int clearjbtimeo = 0; + + if (!req || !(req->r_flags & R_INITTED)) + return; + req->r_flags &= ~R_INITTED; + if (req->r_lflags & RL_QUEUED) + nfs_reqdequeue(req); + if (req->r_achain.tqe_next != NFSREQNOLIST) { + /* still on an async I/O queue? */ + lck_mtx_lock(nfsiod_mutex); + if (nmp && (req->r_achain.tqe_next != NFSREQNOLIST)) { + TAILQ_REMOVE(&nmp->nm_iodq, req, r_achain); + req->r_achain.tqe_next = NFSREQNOLIST; + } + lck_mtx_unlock(nfsiod_mutex); + } + lck_mtx_lock(&req->r_mtx); + if (nmp) { + lck_mtx_lock(&nmp->nm_lock); + if (req->r_flags & R_CWND) { + /* Decrement the outstanding request count. */ + req->r_flags &= ~R_CWND; + nmp->nm_sent -= NFS_CWNDSCALE; + if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) { + /* congestion window is open, poke the cwnd queue */ + struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq); + TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain); + req2->r_cchain.tqe_next = NFSREQNOLIST; + wakeup(req2); + } + } + if (req->r_rchain.tqe_next != NFSREQNOLIST) { + TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain); + req->r_rchain.tqe_next = NFSREQNOLIST; + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + } + if (req->r_cchain.tqe_next != NFSREQNOLIST) { + TAILQ_REMOVE(&nmp->nm_cwndq, req, r_cchain); + req->r_cchain.tqe_next = NFSREQNOLIST; + } + if (req->r_flags & R_JBTPRINTFMSG) { + req->r_flags &= ~R_JBTPRINTFMSG; + nmp->nm_jbreqs--; + clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0; + } + lck_mtx_unlock(&nmp->nm_lock); + } + while (req->r_flags & R_RESENDQ) + msleep(req, &req->r_mtx, (PZERO - 1), "nfsresendqwait", &ts); + lck_mtx_unlock(&req->r_mtx); + if (clearjbtimeo) + nfs_up(nmp, req->r_thread, clearjbtimeo, NULL); + if (req->r_mhead) + mbuf_freem(req->r_mhead); + else if (req->r_mrest) + mbuf_freem(req->r_mrest); + if (req->r_nmrep.nmc_mhead) + mbuf_freem(req->r_nmrep.nmc_mhead); + if (IS_VALID_CRED(req->r_cred)) + kauth_cred_unref(&req->r_cred); + if (nfs_request_using_gss(req)) + nfs_gss_clnt_rpcdone(req); + SLIST_FOREACH_SAFE(gsp, &req->r_gss_seqlist, gss_seqnext, ngsp) + FREE(gsp, M_TEMP); + if (req->r_gss_ctx) + nfs_gss_clnt_ctx_unref(req); + if (req->r_wrongsec) + FREE(req->r_wrongsec, M_TEMP); + + lck_mtx_destroy(&req->r_mtx, nfs_request_grp); + if (req->r_flags & R_ALLOCATED) + FREE_ZONE(req, sizeof(*req), M_NFSREQ); +} + +void +nfs_request_ref(struct nfsreq *req, int locked) +{ + if (!locked) + lck_mtx_lock(&req->r_mtx); + if (req->r_refs <= 0) + panic("nfsreq reference error"); + req->r_refs++; + if (!locked) + lck_mtx_unlock(&req->r_mtx); +} + +void +nfs_request_rele(struct nfsreq *req) +{ + int destroy; + + lck_mtx_lock(&req->r_mtx); + if (req->r_refs <= 0) + panic("nfsreq reference underflow"); + req->r_refs--; + destroy = (req->r_refs == 0); + lck_mtx_unlock(&req->r_mtx); + if (destroy) + nfs_request_destroy(req); +} + + +/* + * Add an (updated) RPC header with authorization to an NFS request. + */ +int +nfs_request_add_header(struct nfsreq *req) +{ + struct nfsmount *nmp; + int error = 0; + mbuf_t m; + + /* free up any previous header */ + if ((m = req->r_mhead)) { + while (m && (m != req->r_mrest)) + m = mbuf_free(m); + req->r_mhead = NULL; + } + + nmp = req->r_np ? NFSTONMP(req->r_np) : req->r_nmp; + if (!nmp) + return (ENXIO); + + error = nfsm_rpchead(req, req->r_mrest, &req->r_xid, &req->r_mhead); + if (error) + return (error); + + req->r_mreqlen = mbuf_pkthdr_len(req->r_mhead); + nmp = req->r_np ? NFSTONMP(req->r_np) : req->r_nmp; + if (!nmp) + return (ENXIO); + lck_mtx_lock(&nmp->nm_lock); + if (NMFLAG(nmp, SOFT)) + req->r_retry = nmp->nm_retry; + else + req->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ + lck_mtx_unlock(&nmp->nm_lock); + + return (error); +} + + +/* + * Queue an NFS request up and send it out. + */ +int +nfs_request_send(struct nfsreq *req, int wait) +{ + struct nfsmount *nmp; + struct timeval now; + + lck_mtx_lock(&req->r_mtx); + req->r_flags |= R_SENDING; + lck_mtx_unlock(&req->r_mtx); + + lck_mtx_lock(nfs_request_mutex); + + nmp = req->r_np ? NFSTONMP(req->r_np) : req->r_nmp; + if (!nmp) { + lck_mtx_unlock(nfs_request_mutex); return (ENXIO); } - verf_str = auth_str = (char *)0; - if (nmp->nm_flag & NFSMNT_KERB) { - verf_str = nickv; - verf_len = sizeof (nickv); - auth_type = RPCAUTH_KERB4; - bzero((caddr_t)key, sizeof (key)); - if (failed_auth || nfs_getnickauth(nmp, cred, &auth_str, - &auth_len, verf_str, verf_len)) { - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; - if (!nmp) { - FSDBG_BOT(531, 2, vp, error, rep); - mbuf_freem(mrest); - return (ENXIO); - } - error = nfs_getauth(nmp, rep, cred, &auth_str, - &auth_len, verf_str, &verf_len, key); - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; - if (!error && !nmp) - error = ENXIO; - if (error) { - FSDBG_BOT(531, 2, vp, error, rep); - mbuf_freem(mrest); - return (error); + + microuptime(&now); + if (!req->r_start) { + req->r_start = now.tv_sec; + req->r_lastmsg = now.tv_sec - + ((nmp->nm_tprintf_delay) - (nmp->nm_tprintf_initial_delay)); + } + + OSAddAtomic64(1, &nfsstats.rpcrequests); + + /* + * Chain request into list of outstanding requests. Be sure + * to put it LAST so timer finds oldest requests first. + * Make sure that the request queue timer is running + * to check for possible request timeout. + */ + TAILQ_INSERT_TAIL(&nfs_reqq, req, r_chain); + req->r_lflags |= RL_QUEUED; + if (!nfs_request_timer_on) { + nfs_request_timer_on = 1; + nfs_interval_timer_start(nfs_request_timer_call, + NFS_REQUESTDELAY); + } + lck_mtx_unlock(nfs_request_mutex); + + /* Send the request... */ + return (nfs_send(req, wait)); +} + +/* + * Call nfs_wait_reply() to wait for the reply. + */ +void +nfs_request_wait(struct nfsreq *req) +{ + req->r_error = nfs_wait_reply(req); +} + +/* + * Finish up an NFS request by dequeueing it and + * doing the initial NFS request reply processing. + */ +int +nfs_request_finish( + struct nfsreq *req, + struct nfsm_chain *nmrepp, + int *status) +{ + struct nfsmount *nmp; + mbuf_t mrep; + int verf_type = 0; + uint32_t verf_len = 0; + uint32_t reply_status = 0; + uint32_t rejected_status = 0; + uint32_t auth_status = 0; + uint32_t accepted_status = 0; + struct nfsm_chain nmrep; + int error, clearjbtimeo; + + error = req->r_error; + + if (nmrepp) + nmrepp->nmc_mhead = NULL; + + /* RPC done, unlink the request. */ + nfs_reqdequeue(req); + + mrep = req->r_nmrep.nmc_mhead; + + nmp = req->r_np ? NFSTONMP(req->r_np) : req->r_nmp; + + if ((req->r_flags & R_CWND) && nmp) { + /* + * Decrement the outstanding request count. + */ + req->r_flags &= ~R_CWND; + lck_mtx_lock(&nmp->nm_lock); + FSDBG(273, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd); + nmp->nm_sent -= NFS_CWNDSCALE; + if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) { + /* congestion window is open, poke the cwnd queue */ + struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq); + TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain); + req2->r_cchain.tqe_next = NFSREQNOLIST; + wakeup(req2); + } + lck_mtx_unlock(&nmp->nm_lock); + } + + if (nfs_request_using_gss(req)) { + /* + * If the request used an RPCSEC_GSS credential + * then reset its sequence number bit in the + * request window. + */ + nfs_gss_clnt_rpcdone(req); + + /* + * If we need to re-send, go back and re-build the + * request based on a new sequence number. + * Note that we're using the original XID. + */ + if (error == EAGAIN) { + req->r_error = 0; + if (mrep) + mbuf_freem(mrep); + error = nfs_gss_clnt_args_restore(req); // remove any trailer mbufs + req->r_nmrep.nmc_mhead = NULL; + req->r_flags |= R_RESTART; + if (error == ENEEDAUTH) { + req->r_xid = 0; // get a new XID + error = 0; + } + goto nfsmout; + } + } + + /* + * If there was a successful reply, make sure to mark the mount as up. + * If a tprintf message was given (or if this is a timed-out soft mount) + * then post a tprintf message indicating the server is alive again. + */ + if (!error) { + if ((req->r_flags & R_TPRINTFMSG) || + (nmp && NMFLAG(nmp, SOFT) && + ((nmp->nm_state & (NFSSTA_TIMEO|NFSSTA_FORCE)) == NFSSTA_TIMEO))) + nfs_up(nmp, req->r_thread, NFSSTA_TIMEO, "is alive again"); + else + nfs_up(nmp, req->r_thread, NFSSTA_TIMEO, NULL); + } + if (!error && !nmp) + error = ENXIO; + nfsmout_if(error); + + /* + * break down the RPC header and check if ok + */ + nmrep = req->r_nmrep; + nfsm_chain_get_32(error, &nmrep, reply_status); + nfsmout_if(error); + if (reply_status == RPC_MSGDENIED) { + nfsm_chain_get_32(error, &nmrep, rejected_status); + nfsmout_if(error); + if (rejected_status == RPC_MISMATCH) { + error = ENOTSUP; + goto nfsmout; + } + nfsm_chain_get_32(error, &nmrep, auth_status); + nfsmout_if(error); + switch (auth_status) { + case RPCSEC_GSS_CREDPROBLEM: + case RPCSEC_GSS_CTXPROBLEM: + /* + * An RPCSEC_GSS cred or context problem. + * We can't use it anymore. + * Restore the args, renew the context + * and set up for a resend. + */ + error = nfs_gss_clnt_args_restore(req); + if (error && error != ENEEDAUTH) + break; + + if (!error) { + error = nfs_gss_clnt_ctx_renew(req); + if (error) + break; + } + mbuf_freem(mrep); + req->r_nmrep.nmc_mhead = NULL; + req->r_xid = 0; // get a new XID + req->r_flags |= R_RESTART; + goto nfsmout; + default: + error = EACCES; + break; + } + goto nfsmout; + } + + /* Now check the verifier */ + nfsm_chain_get_32(error, &nmrep, verf_type); // verifier flavor + nfsm_chain_get_32(error, &nmrep, verf_len); // verifier length + nfsmout_if(error); + + switch (req->r_auth) { + case RPCAUTH_NONE: + case RPCAUTH_SYS: + /* Any AUTH_SYS verifier is ignored */ + if (verf_len > 0) + nfsm_chain_adv(error, &nmrep, nfsm_rndup(verf_len)); + nfsm_chain_get_32(error, &nmrep, accepted_status); + break; + case RPCAUTH_KRB5: + case RPCAUTH_KRB5I: + case RPCAUTH_KRB5P: + error = nfs_gss_clnt_verf_get(req, &nmrep, + verf_type, verf_len, &accepted_status); + break; + } + nfsmout_if(error); + + switch (accepted_status) { + case RPC_SUCCESS: + if (req->r_procnum == NFSPROC_NULL) { + /* + * The NFS null procedure is unique, + * in not returning an NFS status. + */ + *status = NFS_OK; + } else { + nfsm_chain_get_32(error, &nmrep, *status); + nfsmout_if(error); + } + + if ((nmp->nm_vers != NFS_VER2) && (*status == NFSERR_TRYLATER)) { + /* + * It's a JUKEBOX error - delay and try again + */ + int delay, slpflag = (NMFLAG(nmp, INTR) && !(req->r_flags & R_NOINTR)) ? PCATCH : 0; + + mbuf_freem(mrep); + req->r_nmrep.nmc_mhead = NULL; + if ((req->r_delay >= 30) && !(nmp->nm_state & NFSSTA_MOUNTED)) { + /* we're not yet completely mounted and */ + /* we can't complete an RPC, so we fail */ + OSAddAtomic64(1, &nfsstats.rpctimeouts); + nfs_softterm(req); + error = req->r_error; + goto nfsmout; + } + req->r_delay = !req->r_delay ? NFS_TRYLATERDEL : (req->r_delay * 2); + if (req->r_delay > 30) + req->r_delay = 30; + if (nmp->nm_tprintf_initial_delay && (req->r_delay >= nmp->nm_tprintf_initial_delay)) { + if (!(req->r_flags & R_JBTPRINTFMSG)) { + req->r_flags |= R_JBTPRINTFMSG; + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_jbreqs++; + lck_mtx_unlock(&nmp->nm_lock); + } + nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_JUKEBOXTIMEO, + "resource temporarily unavailable (jukebox)"); + } + if (NMFLAG(nmp, SOFT) && (req->r_delay == 30) && !(req->r_flags & R_NOINTR)) { + /* for soft mounts, just give up after a short while */ + OSAddAtomic64(1, &nfsstats.rpctimeouts); + nfs_softterm(req); + error = req->r_error; + goto nfsmout; + } + delay = req->r_delay; + if (req->r_callback.rcb_func) { + struct timeval now; + microuptime(&now); + req->r_resendtime = now.tv_sec + delay; + } else { + do { + if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0))) + goto nfsmout; + tsleep(nfs_request_finish, PSOCK|slpflag, "nfs_jukebox_trylater", hz); + slpflag = 0; + } while (--delay > 0); + } + req->r_xid = 0; // get a new XID + req->r_flags |= R_RESTART; + req->r_start = 0; + FSDBG(273, R_XID32(req->r_xid), nmp, req, NFSERR_TRYLATER); + return (0); + } + + if (req->r_flags & R_JBTPRINTFMSG) { + req->r_flags &= ~R_JBTPRINTFMSG; + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_jbreqs--; + clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0; + lck_mtx_unlock(&nmp->nm_lock); + nfs_up(nmp, req->r_thread, clearjbtimeo, "resource available again"); + } + + if ((nmp->nm_vers >= NFS_VER4) && (*status == NFSERR_WRONGSEC)) { + /* + * Hmmm... we need to try a different security flavor. + * The first time a request hits this, we will allocate an array + * to track flavors to try. We fill the array with the mount's + * preferred flavors or the server's preferred flavors or just the + * flavors we support. + */ + uint32_t srvflavors[NX_MAX_SEC_FLAVORS]; + int srvcount, i, j; + + /* Call SECINFO to try to get list of flavors from server. */ + srvcount = NX_MAX_SEC_FLAVORS; + nfs4_secinfo_rpc(nmp, &req->r_secinfo, req->r_cred, srvflavors, &srvcount); + + if (!req->r_wrongsec) { + /* first time... set up flavor array */ + MALLOC(req->r_wrongsec, uint32_t*, NX_MAX_SEC_FLAVORS*sizeof(uint32_t), M_TEMP, M_WAITOK); + if (!req->r_wrongsec) { + error = EACCES; + goto nfsmout; + } + i=0; + if (nmp->nm_sec.count) { /* use the mount's preferred list of flavors */ + for(; i < nmp->nm_sec.count; i++) + req->r_wrongsec[i] = nmp->nm_sec.flavors[i]; + } else if (srvcount) { /* otherwise use the server's list of flavors */ + for(; i < srvcount; i++) + req->r_wrongsec[i] = srvflavors[i]; + } else { /* otherwise, just try the flavors we support. */ + req->r_wrongsec[i++] = RPCAUTH_KRB5P; + req->r_wrongsec[i++] = RPCAUTH_KRB5I; + req->r_wrongsec[i++] = RPCAUTH_KRB5; + req->r_wrongsec[i++] = RPCAUTH_SYS; + req->r_wrongsec[i++] = RPCAUTH_NONE; + } + for(; i < NX_MAX_SEC_FLAVORS; i++) /* invalidate any remaining slots */ + req->r_wrongsec[i] = RPCAUTH_INVALID; + } + + /* clear the current flavor from the list */ + for(i=0; i < NX_MAX_SEC_FLAVORS; i++) + if (req->r_wrongsec[i] == req->r_auth) + req->r_wrongsec[i] = RPCAUTH_INVALID; + + /* find the next flavor to try */ + for(i=0; i < NX_MAX_SEC_FLAVORS; i++) + if (req->r_wrongsec[i] != RPCAUTH_INVALID) { + if (((req->r_wrongsec[i] == RPCAUTH_KRB5P) || + (req->r_wrongsec[i] == RPCAUTH_KRB5I) || + (req->r_wrongsec[i] == RPCAUTH_KRB5)) && (req->r_gss_ctx && + (req->r_gss_ctx->gss_clnt_service == RPCSEC_GSS_SVC_SYS))) { + /* don't bother trying Kerberos if we've already got a fallback context */ + req->r_wrongsec[i] = RPCAUTH_INVALID; + continue; + } + if (!srvcount) /* no server list, just try it */ + break; + /* check that it's in the server's list */ + for(j=0; j < srvcount; j++) + if (req->r_wrongsec[i] == srvflavors[j]) + break; + if (j < srvcount) /* found */ + break; + /* not found in server list */ + req->r_wrongsec[i] = RPCAUTH_INVALID; + } + if (i == NX_MAX_SEC_FLAVORS) { + /* nothing left to try! */ + error = EACCES; + goto nfsmout; + } + + /* retry with the next auth flavor */ + req->r_auth = req->r_wrongsec[i]; + req->r_xid = 0; // get a new XID + req->r_flags |= R_RESTART; + req->r_start = 0; + FSDBG(273, R_XID32(req->r_xid), nmp, req, NFSERR_WRONGSEC); + return (0); + } + if ((nmp->nm_vers >= NFS_VER4) && req->r_wrongsec) { + /* + * We renegotiated security for this request; so update the + * default security flavor for the associated node. + */ + if (req->r_np) + req->r_np->n_auth = req->r_auth; + } + + if (*status == NFS_OK) { + /* + * Successful NFS request + */ + *nmrepp = nmrep; + req->r_nmrep.nmc_mhead = NULL; + break; + } + /* Got an NFS error of some kind */ + + /* + * If the File Handle was stale, invalidate the + * lookup cache, just in case. + */ + if ((*status == ESTALE) && req->r_np) { + cache_purge(NFSTOV(req->r_np)); + /* if monitored, also send delete event */ + if (vnode_ismonitored(NFSTOV(req->r_np))) + nfs_vnode_notify(req->r_np, (VNODE_EVENT_ATTRIB|VNODE_EVENT_DELETE)); + } + if (nmp->nm_vers == NFS_VER2) + mbuf_freem(mrep); + else + *nmrepp = nmrep; + req->r_nmrep.nmc_mhead = NULL; + error = 0; + break; + case RPC_PROGUNAVAIL: + error = EPROGUNAVAIL; + break; + case RPC_PROGMISMATCH: + error = ERPCMISMATCH; + break; + case RPC_PROCUNAVAIL: + error = EPROCUNAVAIL; + break; + case RPC_GARBAGE: + error = EBADRPC; + break; + case RPC_SYSTEM_ERR: + default: + error = EIO; + break; + } +nfsmout: + if (req->r_flags & R_JBTPRINTFMSG) { + req->r_flags &= ~R_JBTPRINTFMSG; + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_jbreqs--; + clearjbtimeo = (nmp->nm_jbreqs == 0) ? NFSSTA_JUKEBOXTIMEO : 0; + lck_mtx_unlock(&nmp->nm_lock); + if (clearjbtimeo) + nfs_up(nmp, req->r_thread, clearjbtimeo, NULL); + } + FSDBG(273, R_XID32(req->r_xid), nmp, req, + (!error && (*status == NFS_OK)) ? 0xf0f0f0f0 : error); + return (error); +} + +/* + * NFS request using a GSS/Kerberos security flavor? + */ +int +nfs_request_using_gss(struct nfsreq *req) +{ + if (!req->r_gss_ctx) + return (0); + switch (req->r_auth) { + case RPCAUTH_KRB5: + case RPCAUTH_KRB5I: + case RPCAUTH_KRB5P: + return (1); + } + return (0); +} + +/* + * Perform an NFS request synchronously. + */ + +int +nfs_request( + nfsnode_t np, + mount_t mp, /* used only if !np */ + struct nfsm_chain *nmrest, + int procnum, + vfs_context_t ctx, + struct nfsreq_secinfo_args *si, + struct nfsm_chain *nmrepp, + u_int64_t *xidp, + int *status) +{ + return nfs_request2(np, mp, nmrest, procnum, + vfs_context_thread(ctx), vfs_context_ucred(ctx), + si, 0, nmrepp, xidp, status); +} + +int +nfs_request2( + nfsnode_t np, + mount_t mp, /* used only if !np */ + struct nfsm_chain *nmrest, + int procnum, + thread_t thd, + kauth_cred_t cred, + struct nfsreq_secinfo_args *si, + int flags, + struct nfsm_chain *nmrepp, + u_int64_t *xidp, + int *status) +{ + struct nfsreq rq, *req = &rq; + int error; + + if ((error = nfs_request_create(np, mp, nmrest, procnum, thd, cred, &req))) + return (error); + req->r_flags |= (flags & R_OPTMASK); + if (si) + req->r_secinfo = *si; + + FSDBG_TOP(273, R_XID32(req->r_xid), np, procnum, 0); + do { + req->r_error = 0; + req->r_flags &= ~R_RESTART; + if ((error = nfs_request_add_header(req))) + break; + if (xidp) + *xidp = req->r_xid; + if ((error = nfs_request_send(req, 1))) + break; + nfs_request_wait(req); + if ((error = nfs_request_finish(req, nmrepp, status))) + break; + } while (req->r_flags & R_RESTART); + + FSDBG_BOT(273, R_XID32(req->r_xid), np, procnum, error); + nfs_request_rele(req); + return (error); +} + + +/* + * Set up a new null proc request to exchange GSS context tokens with the + * server. Associate the context that we are setting up with the request that we + * are sending. + */ + +int +nfs_request_gss( + mount_t mp, + struct nfsm_chain *nmrest, + thread_t thd, + kauth_cred_t cred, + int flags, + struct nfs_gss_clnt_ctx *cp, /* Set to gss context to renew or setup */ + struct nfsm_chain *nmrepp, + int *status) +{ + struct nfsreq rq, *req = &rq; + int error; + + if ((error = nfs_request_create(NULL, mp, nmrest, NFSPROC_NULL, thd, cred, &req))) + return (error); + req->r_flags |= (flags & R_OPTMASK); + + if (cp == NULL) { + printf("nfs_request_gss request has no context\n"); + nfs_request_rele(req); + return (NFSERR_EAUTH); + } + nfs_gss_clnt_ctx_ref(req, cp); + + FSDBG_TOP(273, R_XID32(req->r_xid), NULL, NFSPROC_NULL, 0); + do { + req->r_error = 0; + req->r_flags &= ~R_RESTART; + if ((error = nfs_request_add_header(req))) + break; + + if ((error = nfs_request_send(req, 1))) + break; + nfs_request_wait(req); + if ((error = nfs_request_finish(req, nmrepp, status))) + break; + } while (req->r_flags & R_RESTART); + + FSDBG_BOT(273, R_XID32(req->r_xid), NULL, NFSPROC_NULL, error); + nfs_request_rele(req); + return (error); +} + +/* + * Create and start an asynchronous NFS request. + */ +int +nfs_request_async( + nfsnode_t np, + mount_t mp, /* used only if !np */ + struct nfsm_chain *nmrest, + int procnum, + thread_t thd, + kauth_cred_t cred, + struct nfsreq_secinfo_args *si, + int flags, + struct nfsreq_cbinfo *cb, + struct nfsreq **reqp) +{ + struct nfsreq *req; + struct nfsmount *nmp; + int error, sent; + + error = nfs_request_create(np, mp, nmrest, procnum, thd, cred, reqp); + req = *reqp; + FSDBG(274, (req ? R_XID32(req->r_xid) : 0), np, procnum, error); + if (error) + return (error); + req->r_flags |= (flags & R_OPTMASK); + req->r_flags |= R_ASYNC; + if (si) + req->r_secinfo = *si; + if (cb) + req->r_callback = *cb; + error = nfs_request_add_header(req); + if (!error) { + req->r_flags |= R_WAITSENT; + if (req->r_callback.rcb_func) + nfs_request_ref(req, 0); + error = nfs_request_send(req, 1); + lck_mtx_lock(&req->r_mtx); + if (!error && !(req->r_flags & R_SENT) && req->r_callback.rcb_func) { + /* make sure to wait until this async I/O request gets sent */ + int slpflag = (req->r_nmp && NMFLAG(req->r_nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR)) ? PCATCH : 0; + struct timespec ts = { 2, 0 }; + while (!(req->r_flags & R_SENT)) { + if ((req->r_flags & R_RESENDQ) && ((nmp = req->r_nmp))) { + lck_mtx_lock(&nmp->nm_lock); + if ((nmp->nm_state & NFSSTA_RECOVER) && (req->r_rchain.tqe_next != NFSREQNOLIST)) { + /* + * It's not going to get off the resend queue if we're in recovery. + * So, just take it off ourselves. We could be holding mount state + * busy and thus holding up the start of recovery. + */ + TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain); + req->r_rchain.tqe_next = NFSREQNOLIST; + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + lck_mtx_unlock(&nmp->nm_lock); + req->r_flags |= R_SENDING; + lck_mtx_unlock(&req->r_mtx); + error = nfs_send(req, 1); + lck_mtx_lock(&req->r_mtx); + if (error) + break; + continue; + } + lck_mtx_unlock(&nmp->nm_lock); + } + if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0))) + break; + msleep(req, &req->r_mtx, slpflag | (PZERO - 1), "nfswaitsent", &ts); + slpflag = 0; + } + } + sent = req->r_flags & R_SENT; + lck_mtx_unlock(&req->r_mtx); + if (error && req->r_callback.rcb_func && !sent) + nfs_request_rele(req); + } + FSDBG(274, R_XID32(req->r_xid), np, procnum, error); + if (error || req->r_callback.rcb_func) + nfs_request_rele(req); + return (error); +} + +/* + * Wait for and finish an asynchronous NFS request. + */ +int +nfs_request_async_finish( + struct nfsreq *req, + struct nfsm_chain *nmrepp, + u_int64_t *xidp, + int *status) +{ + int error = 0, asyncio = req->r_callback.rcb_func ? 1 : 0; + struct nfsmount *nmp; + + lck_mtx_lock(&req->r_mtx); + if (!asyncio) + req->r_flags |= R_ASYNCWAIT; + while (req->r_flags & R_RESENDQ) { /* wait until the request is off the resend queue */ + struct timespec ts = { 2, 0 }; + if ((nmp = req->r_nmp)) { + lck_mtx_lock(&nmp->nm_lock); + if ((nmp->nm_state & NFSSTA_RECOVER) && (req->r_rchain.tqe_next != NFSREQNOLIST)) { + /* + * It's not going to get off the resend queue if we're in recovery. + * So, just take it off ourselves. We could be holding mount state + * busy and thus holding up the start of recovery. + */ + TAILQ_REMOVE(&nmp->nm_resendq, req, r_rchain); + req->r_rchain.tqe_next = NFSREQNOLIST; + if (req->r_flags & R_RESENDQ) + req->r_flags &= ~R_RESENDQ; + lck_mtx_unlock(&nmp->nm_lock); + break; } + lck_mtx_unlock(&nmp->nm_lock); } - } else { - auth_type = RPCAUTH_UNIX; - if (cred->cr_ngroups < 1) - panic("nfsreq nogrps"); - auth_len = ((((cred->cr_ngroups - 1) > nmp->nm_numgrps) ? - nmp->nm_numgrps : (cred->cr_ngroups - 1)) << 2) + - 5 * NFSX_UNSIGNED; - } - error = nfsm_rpchead(cred, nmp->nm_flag, procnum, auth_type, auth_len, - auth_str, verf_len, verf_str, mrest, mrest_len, &mheadend, &xid, &m); - if (auth_str) - _FREE(auth_str, M_TEMP); - if (error) { - mbuf_freem(mrest); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); - return (error); + if ((error = nfs_sigintr(req->r_nmp, req, req->r_thread, 0))) + break; + msleep(req, &req->r_mtx, PZERO-1, "nfsresendqwait", &ts); } - if (xidp) - *xidp = ntohl(xid) + ((u_int64_t)nfs_xidwrap << 32); + lck_mtx_unlock(&req->r_mtx); - /* - * For stream protocols, insert a Sun RPC Record Mark. - */ - if (nmsotype == SOCK_STREAM) { - error = mbuf_prepend(&m, NFSX_UNSIGNED, MBUF_WAITOK); - if (error) { - mbuf_freem(m); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); - return (error); + if (!error) { + nfs_request_wait(req); + error = nfs_request_finish(req, nmrepp, status); + } + + while (!error && (req->r_flags & R_RESTART)) { + if (asyncio && req->r_resendtime) { /* send later */ + lck_mtx_lock(&req->r_mtx); + nfs_asyncio_resend(req); + lck_mtx_unlock(&req->r_mtx); + return (EINPROGRESS); } - *((u_long*)mbuf_data(m)) = - htonl(0x80000000 | (mbuf_pkthdr_len(m) - NFSX_UNSIGNED)); + req->r_error = 0; + req->r_flags &= ~R_RESTART; + if ((error = nfs_request_add_header(req))) + break; + if ((error = nfs_request_send(req, !asyncio))) + break; + if (asyncio) + return (EINPROGRESS); + nfs_request_wait(req); + if ((error = nfs_request_finish(req, nmrepp, status))) + break; } - rep->r_mreq = m; - rep->r_xid = xid; -tryagain: - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; - if (nmp && (nmp->nm_flag & NFSMNT_SOFT)) - rep->r_retry = nmp->nm_retry; - else - rep->r_retry = NFS_MAXREXMIT + 1; /* past clip limit */ - rep->r_rtt = rep->r_rexmit = 0; - if (proct[procnum] > 0) - rep->r_flags = R_TIMING; - else - rep->r_flags = 0; - rep->r_mrep = NULL; + if (xidp) + *xidp = req->r_xid; - /* - * Do the client side RPC. - */ - OSAddAtomic(1, (SInt32*)&nfsstats.rpcrequests); - /* - * Chain request into list of outstanding requests. Be sure - * to put it LAST so timer finds oldest requests first. - */ - TAILQ_INSERT_TAIL(&nfs_reqq, rep, r_chain); + FSDBG(275, R_XID32(req->r_xid), req->r_np, req->r_procnum, error); + nfs_request_rele(req); + return (error); +} + +/* + * Cancel a pending asynchronous NFS request. + */ +void +nfs_request_async_cancel(struct nfsreq *req) +{ + nfs_reqdequeue(req); + FSDBG(275, R_XID32(req->r_xid), req->r_np, req->r_procnum, 0xD1ED1E); + nfs_request_rele(req); +} +/* + * Flag a request as being terminated. + */ +void +nfs_softterm(struct nfsreq *req) +{ + struct nfsmount *nmp = req->r_nmp; + req->r_flags |= R_SOFTTERM; + req->r_error = ETIMEDOUT; + if (!(req->r_flags & R_CWND) || !nmp) + return; + /* update congestion window */ + req->r_flags &= ~R_CWND; + lck_mtx_lock(&nmp->nm_lock); + FSDBG(532, R_XID32(req->r_xid), req, nmp->nm_sent, nmp->nm_cwnd); + nmp->nm_sent -= NFS_CWNDSCALE; + if ((nmp->nm_sent < nmp->nm_cwnd) && !TAILQ_EMPTY(&nmp->nm_cwndq)) { + /* congestion window is open, poke the cwnd queue */ + struct nfsreq *req2 = TAILQ_FIRST(&nmp->nm_cwndq); + TAILQ_REMOVE(&nmp->nm_cwndq, req2, r_cchain); + req2->r_cchain.tqe_next = NFSREQNOLIST; + wakeup(req2); + } + lck_mtx_unlock(&nmp->nm_lock); +} + +/* + * Ensure req isn't in use by the timer, then dequeue it. + */ +void +nfs_reqdequeue(struct nfsreq *req) +{ + lck_mtx_lock(nfs_request_mutex); + while (req->r_lflags & RL_BUSY) { + req->r_lflags |= RL_WAITING; + msleep(&req->r_lflags, nfs_request_mutex, PSOCK, "reqdeq", NULL); + } + if (req->r_lflags & RL_QUEUED) { + TAILQ_REMOVE(&nfs_reqq, req, r_chain); + req->r_lflags &= ~RL_QUEUED; + } + lck_mtx_unlock(nfs_request_mutex); +} + +/* + * Busy (lock) a nfsreq, used by the nfs timer to make sure it's not + * free()'d out from under it. + */ +void +nfs_reqbusy(struct nfsreq *req) +{ + if (req->r_lflags & RL_BUSY) + panic("req locked"); + req->r_lflags |= RL_BUSY; +} + +/* + * Unbusy the nfsreq passed in, return the next nfsreq in the chain busied. + */ +struct nfsreq * +nfs_reqnext(struct nfsreq *req) +{ + struct nfsreq * nextreq; + + if (req == NULL) + return (NULL); /* - * If backing off another request or avoiding congestion, don't - * send this one now but let timer do it. If not timing a request, - * do it now. + * We need to get and busy the next req before signalling the + * current one, otherwise wakeup() may block us and we'll race to + * grab the next req. */ - if (nmp && nmp->nm_so && (nmp->nm_sotype != SOCK_DGRAM || - (nmp->nm_flag & NFSMNT_DUMBTIMR) || - nmp->nm_sent < nmp->nm_cwnd)) { - int connrequired = (nmp->nm_sotype == SOCK_STREAM); + nextreq = TAILQ_NEXT(req, r_chain); + if (nextreq != NULL) + nfs_reqbusy(nextreq); + /* unbusy and signal. */ + req->r_lflags &= ~RL_BUSY; + if (req->r_lflags & RL_WAITING) { + req->r_lflags &= ~RL_WAITING; + wakeup(&req->r_lflags); + } + return (nextreq); +} + +/* + * NFS request queue timer routine + * + * Scan the NFS request queue for any requests that have timed out. + * + * Alert the system of unresponsive servers. + * Mark expired requests on soft mounts as terminated. + * For UDP, mark/signal requests for retransmission. + */ +void +nfs_request_timer(__unused void *param0, __unused void *param1) +{ + struct nfsreq *req; + struct nfsmount *nmp; + int timeo, maxtime, finish_asyncio, error; + struct timeval now; + TAILQ_HEAD(nfs_mount_pokeq, nfsmount) nfs_mount_poke_queue; + + lck_mtx_lock(nfs_request_mutex); + req = TAILQ_FIRST(&nfs_reqq); + if (req == NULL) { /* no requests - turn timer off */ + nfs_request_timer_on = 0; + lck_mtx_unlock(nfs_request_mutex); + return; + } + + nfs_reqbusy(req); + TAILQ_INIT(&nfs_mount_poke_queue); + + microuptime(&now); + for ( ; req != NULL ; req = nfs_reqnext(req)) { + nmp = req->r_nmp; + if (!nmp) /* unmounted */ + continue; + if (req->r_error || req->r_nmrep.nmc_mhead) + continue; + if ((error = nfs_sigintr(nmp, req, req->r_thread, 0))) { + if (req->r_callback.rcb_func != NULL) { + /* async I/O RPC needs to be finished */ + lck_mtx_lock(&req->r_mtx); + req->r_error = error; + finish_asyncio = !(req->r_flags & R_WAITSENT); + wakeup(req); + lck_mtx_unlock(&req->r_mtx); + if (finish_asyncio) + nfs_asyncio_finish(req); + } + continue; + } + + lck_mtx_lock(&req->r_mtx); + + if (nmp->nm_tprintf_initial_delay && + ((req->r_rexmit > 2) || (req->r_flags & R_RESENDERR)) && + ((req->r_lastmsg + nmp->nm_tprintf_delay) < now.tv_sec)) { + req->r_lastmsg = now.tv_sec; + nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_TIMEO, + "not responding"); + req->r_flags |= R_TPRINTFMSG; + lck_mtx_lock(&nmp->nm_lock); + if (!(nmp->nm_state & NFSSTA_MOUNTED)) { + lck_mtx_unlock(&nmp->nm_lock); + /* we're not yet completely mounted and */ + /* we can't complete an RPC, so we fail */ + OSAddAtomic64(1, &nfsstats.rpctimeouts); + nfs_softterm(req); + finish_asyncio = ((req->r_callback.rcb_func != NULL) && !(req->r_flags & R_WAITSENT)); + wakeup(req); + lck_mtx_unlock(&req->r_mtx); + if (finish_asyncio) + nfs_asyncio_finish(req); + continue; + } + lck_mtx_unlock(&nmp->nm_lock); + } - if (connrequired) - error = nfs_sndlock(rep); + /* + * Put a reasonable limit on the maximum timeout, + * and reduce that limit when soft mounts get timeouts or are in reconnect. + */ + if (!NMFLAG(nmp, SOFT) && !nfs_can_squish(nmp)) + maxtime = NFS_MAXTIMEO; + else if ((req->r_flags & (R_SETUP|R_RECOVER)) || + ((nmp->nm_reconnect_start <= 0) || ((now.tv_sec - nmp->nm_reconnect_start) < 8))) + maxtime = (NFS_MAXTIMEO / (nmp->nm_timeouts+1))/2; + else + maxtime = NFS_MINTIMEO/4; /* - * Set the R_SENT before doing the send in case another thread - * processes the reply before the nfs_send returns here + * Check for request timeout. */ - if (!error) { - if ((rep->r_flags & R_MUSTRESEND) == 0) { - FSDBG(531, rep->r_xid, rep, nmp->nm_sent, - nmp->nm_cwnd); - nmp->nm_sent += NFS_CWNDSCALE; - rep->r_flags |= R_SENT; + if (req->r_rtt >= 0) { + req->r_rtt++; + lck_mtx_lock(&nmp->nm_lock); + if (req->r_flags & R_RESENDERR) { + /* with resend errors, retry every few seconds */ + timeo = 4*hz; + } else { + if (req->r_procnum == NFSPROC_NULL && req->r_gss_ctx != NULL) + timeo = NFS_MINIDEMTIMEO; // gss context setup + else if (NMFLAG(nmp, DUMBTIMER)) + timeo = nmp->nm_timeo; + else + timeo = NFS_RTO(nmp, proct[req->r_procnum]); + + /* ensure 62.5 ms floor */ + while (16 * timeo < hz) + timeo *= 2; + if (nmp->nm_timeouts > 0) + timeo *= nfs_backoff[nmp->nm_timeouts - 1]; + } + /* limit timeout to max */ + if (timeo > maxtime) + timeo = maxtime; + if (req->r_rtt <= timeo) { + lck_mtx_unlock(&nmp->nm_lock); + lck_mtx_unlock(&req->r_mtx); + continue; + } + /* The request has timed out */ + NFS_SOCK_DBG("nfs timeout: proc %d %d xid %llx rtt %d to %d # %d, t %ld/%d\n", + req->r_procnum, proct[req->r_procnum], + req->r_xid, req->r_rtt, timeo, nmp->nm_timeouts, + (now.tv_sec - req->r_start)*NFS_HZ, maxtime); + if (nmp->nm_timeouts < 8) + nmp->nm_timeouts++; + nfs_mount_check_dead_timeout(nmp); + /* if it's been a few seconds, try poking the socket */ + if ((nmp->nm_sotype == SOCK_STREAM) && + ((now.tv_sec - req->r_start) >= 3) && + !(nmp->nm_sockflags & (NMSOCK_POKE|NMSOCK_UNMOUNT)) && + (nmp->nm_sockflags & NMSOCK_READY)) { + nmp->nm_sockflags |= NMSOCK_POKE; + TAILQ_INSERT_TAIL(&nfs_mount_poke_queue, nmp, nm_pokeq); } + lck_mtx_unlock(&nmp->nm_lock); + } - error = mbuf_copym(m, 0, MBUF_COPYALL, MBUF_WAITOK, &m2); - if (!error) - error = nfs_send(nmp->nm_so, nmp->nm_nam, m2, rep); - if (connrequired) - nfs_sndunlock(rep); + /* For soft mounts (& SETUPs/RECOVERs), check for too many retransmits/timeout. */ + if ((NMFLAG(nmp, SOFT) || (req->r_flags & (R_SETUP|R_RECOVER))) && + ((req->r_rexmit >= req->r_retry) || /* too many */ + ((now.tv_sec - req->r_start)*NFS_HZ > maxtime))) { /* too long */ + OSAddAtomic64(1, &nfsstats.rpctimeouts); + lck_mtx_lock(&nmp->nm_lock); + if (!(nmp->nm_state & NFSSTA_TIMEO)) { + lck_mtx_unlock(&nmp->nm_lock); + /* make sure we note the unresponsive server */ + /* (maxtime may be less than tprintf delay) */ + nfs_down(req->r_nmp, req->r_thread, 0, NFSSTA_TIMEO, + "not responding"); + req->r_lastmsg = now.tv_sec; + req->r_flags |= R_TPRINTFMSG; + } else { + lck_mtx_unlock(&nmp->nm_lock); + } + if (req->r_flags & R_NOINTR) { + /* don't terminate nointr requests on timeout */ + lck_mtx_unlock(&req->r_mtx); + continue; + } + NFS_SOCK_DBG("nfs timer TERMINATE: p %d x 0x%llx f 0x%x rtt %d t %ld\n", + req->r_procnum, req->r_xid, req->r_flags, req->r_rtt, + now.tv_sec - req->r_start); + nfs_softterm(req); + finish_asyncio = ((req->r_callback.rcb_func != NULL) && !(req->r_flags & R_WAITSENT)); + wakeup(req); + lck_mtx_unlock(&req->r_mtx); + if (finish_asyncio) + nfs_asyncio_finish(req); + continue; } - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; - if (error) { - if (nmp) - nmp->nm_sent -= NFS_CWNDSCALE; - rep->r_flags &= ~R_SENT; + + /* for TCP, only resend if explicitly requested */ + if ((nmp->nm_sotype == SOCK_STREAM) && !(req->r_flags & R_MUSTRESEND)) { + if (++req->r_rexmit > NFS_MAXREXMIT) + req->r_rexmit = NFS_MAXREXMIT; + req->r_rtt = 0; + lck_mtx_unlock(&req->r_mtx); + continue; } - } else { - rep->r_rtt = -1; + + /* + * The request needs to be (re)sent. Kick the requester to resend it. + * (unless it's already marked as needing a resend) + */ + if ((req->r_flags & R_MUSTRESEND) && (req->r_rtt == -1)) { + lck_mtx_unlock(&req->r_mtx); + continue; + } + NFS_SOCK_DBG("nfs timer mark resend: p %d x 0x%llx f 0x%x rtt %d\n", + req->r_procnum, req->r_xid, req->r_flags, req->r_rtt); + req->r_flags |= R_MUSTRESEND; + req->r_rtt = -1; + wakeup(req); + if ((req->r_flags & (R_ASYNC|R_ASYNCWAIT|R_SENDING)) == R_ASYNC) + nfs_asyncio_resend(req); + lck_mtx_unlock(&req->r_mtx); } - /* - * Wait for the reply from our send or the timer's. - */ - if (!error || error == EPIPE) - error = nfs_reply(rep); + lck_mtx_unlock(nfs_request_mutex); - /* - * RPC done, unlink the request. - */ - nfs_repdequeue(rep); + /* poke any sockets */ + while ((nmp = TAILQ_FIRST(&nfs_mount_poke_queue))) { + TAILQ_REMOVE(&nfs_mount_poke_queue, nmp, nm_pokeq); + nfs_sock_poke(nmp); + lck_mtx_lock(&nmp->nm_lock); + nmp->nm_sockflags &= ~NMSOCK_POKE; + wakeup(&nmp->nm_sockflags); + lck_mtx_unlock(&nmp->nm_lock); + } - nmp = vp ? VFSTONFS(vnode_mount(vp)) : rep->r_nmp; + nfs_interval_timer_start(nfs_request_timer_call, NFS_REQUESTDELAY); +} - /* - * Decrement the outstanding request count. - */ - if (rep->r_flags & R_SENT) { - rep->r_flags &= ~R_SENT; /* paranoia */ - if (nmp) { - FSDBG(531, rep->r_xid, rep, nmp->nm_sent, nmp->nm_cwnd); - nmp->nm_sent -= NFS_CWNDSCALE; - } +/* + * check a thread's proc for the "noremotehang" flag. + */ +int +nfs_noremotehang(thread_t thd) +{ + proc_t p = thd ? get_bsdthreadtask_info(thd) : NULL; + return (p && proc_noremotehang(p)); +} + +/* + * Test for a termination condition pending on the process. + * This is used to determine if we need to bail on a mount. + * ETIMEDOUT is returned if there has been a soft timeout. + * EINTR is returned if there is a signal pending that is not being ignored + * and the mount is interruptable, or if we are a thread that is in the process + * of cancellation (also SIGKILL posted). + */ +extern int sigprop[NSIG+1]; +int +nfs_sigintr(struct nfsmount *nmp, struct nfsreq *req, thread_t thd, int nmplocked) +{ + proc_t p; + int error = 0; + + if (nmp == NULL) + return (ENXIO); + + if (req && (req->r_flags & R_SOFTTERM)) + return (ETIMEDOUT); /* request has been terminated. */ + if (req && (req->r_flags & R_NOINTR)) + thd = NULL; /* don't check for signal on R_NOINTR */ + + if (!nmplocked) + lck_mtx_lock(&nmp->nm_lock); + if (nmp->nm_state & NFSSTA_FORCE) { + /* If a force unmount is in progress then fail. */ + error = EIO; + } else if (nmp->nm_mountp->mnt_kern_flag & MNTK_FRCUNMOUNT) { + /* Someone is unmounting us, go soft and mark it. */ + NFS_BITMAP_SET(nmp->nm_flags, NFS_MFLAG_SOFT); + nmp->nm_state |= NFSSTA_FORCE; } - /* - * If there was a successful reply and a tprintf msg. - * tprintf a response. - */ - if (!error) - nfs_up(nmp, procp, NFSSTA_TIMEO, - (rep->r_flags & R_TPRINTFMSG) ? "is alive again" : NULL); - mrep = rep->r_mrep; - md = rep->r_md; - dpos = rep->r_dpos; - if (!error && !nmp) + /* Check if the mount is marked dead. */ + if (!error && (nmp->nm_state & NFSSTA_DEAD)) error = ENXIO; - if (error) { - mbuf_freem(rep->r_mreq); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); - return (error); - } /* - * break down the rpc header and check if ok + * If the mount is hung and we've requested not to hang + * on remote filesystems, then bail now. */ - nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); - if (*tl++ == rpc_msgdenied) { - if (*tl == rpc_mismatch) - error = EOPNOTSUPP; - else if ((nmp->nm_flag & NFSMNT_KERB) && *tl++ == rpc_autherr) { - if (!failed_auth) { - failed_auth++; - error = mbuf_setnext(mheadend, NULL); - mbuf_freem(mrep); - mbuf_freem(rep->r_mreq); - if (!error) - goto kerbauth; - printf("nfs_request: mbuf_setnext failed\n"); - } else - error = EAUTH; - } else - error = EACCES; - mbuf_freem(mrep); - mbuf_freem(rep->r_mreq); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); + if (current_proc() != kernproc && + !error && (nmp->nm_state & NFSSTA_TIMEO) && nfs_noremotehang(thd)) + error = EIO; + + if (!nmplocked) + lck_mtx_unlock(&nmp->nm_lock); + if (error) return (error); - } + + /* may not have a thread for async I/O */ + if (thd == NULL || current_proc() == kernproc) + return (0); /* - * Grab any Kerberos verifier, otherwise just throw it away. + * Check if the process is aborted, but don't interrupt if we + * were killed by a signal and this is the exiting thread which + * is attempting to dump core. */ - verf_type = fxdr_unsigned(int, *tl++); - i = fxdr_unsigned(int, *tl); - if ((nmp->nm_flag & NFSMNT_KERB) && verf_type == RPCAUTH_KERB4) { - error = nfs_savenickauth(nmp, cred, i, key, &md, &dpos, mrep); - if (error) - goto nfsmout; - } else if (i > 0) - nfsm_adv(nfsm_rndup(i)); - nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); - /* 0 == ok */ - if (*tl == 0) { - nfsm_dissect(tl, u_long *, NFSX_UNSIGNED); - if (*tl != 0) { - error = fxdr_unsigned(int, *tl); - if ((nmp->nm_flag & NFSMNT_NFSV3) && - error == NFSERR_TRYLATER) { - mbuf_freem(mrep); - error = 0; - microuptime(&now); - waituntil = now.tv_sec + trylater_delay; - while (now.tv_sec < waituntil) { - tsleep((caddr_t)&lbolt, PSOCK, "nfstrylater", 0); - microuptime(&now); - } - trylater_delay *= 2; - if (trylater_delay > 60) - trylater_delay = 60; - goto tryagain; - } + if (((p = current_proc()) != kernproc) && current_thread_aborted() && + (!(p->p_acflag & AXSIG) || (p->exit_thread != current_thread()) || + (p->p_sigacts == NULL) || + (p->p_sigacts->ps_sig < 1) || (p->p_sigacts->ps_sig > NSIG) || + !(sigprop[p->p_sigacts->ps_sig] & SA_CORE))) + return (EINTR); - /* - * If the File Handle was stale, invalidate the - * lookup cache, just in case. - */ - if ((error == ESTALE) && vp) - cache_purge(vp); - if (nmp->nm_flag & NFSMNT_NFSV3) { - *mrp = mrep; - *mdp = md; - *dposp = dpos; - error |= NFSERR_RETERR; - } else { - mbuf_freem(mrep); - error &= ~NFSERR_RETERR; - } - mbuf_freem(rep->r_mreq); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); - return (error); + /* mask off thread and process blocked signals. */ + if (NMFLAG(nmp, INTR) && ((p = get_bsdthreadtask_info(thd))) && + proc_pendingsignals(p, NFSINT_SIGMASK)) + return (EINTR); + return (0); +} + +/* + * Lock a socket against others. + * Necessary for STREAM sockets to ensure you get an entire rpc request/reply + * and also to avoid race conditions between the processes with nfs requests + * in progress when a reconnect is necessary. + */ +int +nfs_sndlock(struct nfsreq *req) +{ + struct nfsmount *nmp = req->r_nmp; + int *statep; + int error = 0, slpflag = 0; + struct timespec ts = { 0, 0 }; + + if (nmp == NULL) + return (ENXIO); + + lck_mtx_lock(&nmp->nm_lock); + statep = &nmp->nm_state; + + if (NMFLAG(nmp, INTR) && req->r_thread && !(req->r_flags & R_NOINTR)) + slpflag = PCATCH; + while (*statep & NFSSTA_SNDLOCK) { + if ((error = nfs_sigintr(nmp, req, req->r_thread, 1))) + break; + *statep |= NFSSTA_WANTSND; + if (nfs_noremotehang(req->r_thread)) + ts.tv_sec = 1; + msleep(statep, &nmp->nm_lock, slpflag | (PZERO - 1), "nfsndlck", &ts); + if (slpflag == PCATCH) { + slpflag = 0; + ts.tv_sec = 2; } + } + if (!error) + *statep |= NFSSTA_SNDLOCK; + lck_mtx_unlock(&nmp->nm_lock); + return (error); +} + +/* + * Unlock the stream socket for others. + */ +void +nfs_sndunlock(struct nfsreq *req) +{ + struct nfsmount *nmp = req->r_nmp; + int *statep, wake = 0; - *mrp = mrep; - *mdp = md; - *dposp = dpos; - mbuf_freem(rep->r_mreq); - FSDBG_BOT(531, 0xf0f0f0f0, rep->r_xid, nmp, rep); - return (0); + if (nmp == NULL) + return; + lck_mtx_lock(&nmp->nm_lock); + statep = &nmp->nm_state; + if ((*statep & NFSSTA_SNDLOCK) == 0) + panic("nfs sndunlock"); + *statep &= ~(NFSSTA_SNDLOCK|NFSSTA_SENDING); + if (*statep & NFSSTA_WANTSND) { + *statep &= ~NFSSTA_WANTSND; + wake = 1; } - mbuf_freem(mrep); - error = EPROTONOSUPPORT; -nfsmout: - mbuf_freem(rep->r_mreq); - FSDBG_BOT(531, error, rep->r_xid, nmp, rep); - return (error); + lck_mtx_unlock(&nmp->nm_lock); + if (wake) + wakeup(statep); } -#ifndef NFS_NOSERVER -/* - * Generate the rpc reply header - * siz arg. is used to decide if adding a cluster is worthwhile - */ int -nfs_rephead(siz, nd, slp, err, mrq, mbp, bposp) - int siz; - struct nfsrv_descript *nd; - struct nfssvc_sock *slp; - int err; - mbuf_t *mrq; - mbuf_t *mbp; - caddr_t *bposp; +nfs_aux_request( + struct nfsmount *nmp, + thread_t thd, + struct sockaddr *saddr, + socket_t so, + int sotype, + mbuf_t mreq, + uint32_t xid, + int bindresv, + int timeo, + struct nfsm_chain *nmrep) { - u_long *tl; - mbuf_t mreq; - caddr_t bpos; - mbuf_t mb, mb2; - int error, mlen; + int error = 0, on = 1, try, sendat = 2, soproto, recv, optlen, restoreto = 0; + socket_t newso = NULL; + struct sockaddr_storage ss; + struct timeval orig_rcvto, orig_sndto, tv = { 1, 0 }; + mbuf_t m, mrep = NULL; + struct msghdr msg; + uint32_t rxid = 0, reply = 0, reply_status, rejected_status; + uint32_t verf_type, verf_len, accepted_status; + size_t readlen, sentlen; + struct nfs_rpc_record_state nrrs; + + if (!so) { + /* create socket and set options */ + soproto = (sotype == SOCK_DGRAM) ? IPPROTO_UDP : IPPROTO_TCP; + if ((error = sock_socket(saddr->sa_family, sotype, soproto, NULL, NULL, &newso))) + goto nfsmout; - /* - * If this is a big reply, use a cluster else - * try and leave leading space for the lower level headers. - */ - siz += RPC_REPLYSIZ; - if (siz >= nfs_mbuf_minclsize) { - error = mbuf_getpacket(MBUF_WAITOK, &mreq); - } else { - error = mbuf_gethdr(MBUF_WAITOK, MBUF_TYPE_DATA, &mreq); - } - if (error) { - /* unable to allocate packet */ - /* XXX nfsstat? */ - return (error); - } - mb = mreq; - tl = mbuf_data(mreq); - mlen = 6 * NFSX_UNSIGNED; - if (siz < nfs_mbuf_minclsize) { - /* leave space for lower level headers */ - tl += 80/sizeof(*tl); /* XXX max_hdr? XXX */ - mbuf_setdata(mreq, tl, mlen); - } else { - mbuf_setlen(mreq, mlen); - } - bpos = ((caddr_t)tl) + mlen; - *tl++ = txdr_unsigned(nd->nd_retxid); - *tl++ = rpc_reply; - if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { - *tl++ = rpc_msgdenied; - if (err & NFSERR_AUTHERR) { - *tl++ = rpc_autherr; - *tl = txdr_unsigned(err & ~NFSERR_AUTHERR); - mlen -= NFSX_UNSIGNED; - mbuf_setlen(mreq, mlen); - bpos -= NFSX_UNSIGNED; - } else { - *tl++ = rpc_mismatch; - *tl++ = txdr_unsigned(RPC_VER2); - *tl = txdr_unsigned(RPC_VER2); + if (bindresv) { + int level = (saddr->sa_family == AF_INET) ? IPPROTO_IP : IPPROTO_IPV6; + int optname = (saddr->sa_family == AF_INET) ? IP_PORTRANGE : IPV6_PORTRANGE; + int portrange = IP_PORTRANGE_LOW; + error = sock_setsockopt(newso, level, optname, &portrange, sizeof(portrange)); + nfsmout_if(error); + ss.ss_len = saddr->sa_len; + ss.ss_family = saddr->sa_family; + if (ss.ss_family == AF_INET) { + ((struct sockaddr_in*)&ss)->sin_addr.s_addr = INADDR_ANY; + ((struct sockaddr_in*)&ss)->sin_port = htons(0); + } else if (ss.ss_family == AF_INET6) { + ((struct sockaddr_in6*)&ss)->sin6_addr = in6addr_any; + ((struct sockaddr_in6*)&ss)->sin6_port = htons(0); + } else { + error = EINVAL; + } + if (!error) + error = sock_bind(newso, (struct sockaddr *)&ss); + nfsmout_if(error); } - } else { - *tl++ = rpc_msgaccepted; - /* - * For Kerberos authentication, we must send the nickname - * verifier back, otherwise just RPCAUTH_NULL. - */ - if (nd->nd_flag & ND_KERBFULL) { - struct nfsuid *nuidp; - struct timeval ktvin, ktvout; - uid_t uid = kauth_cred_getuid(nd->nd_cr); - - lck_rw_lock_shared(&slp->ns_rwlock); - for (nuidp = NUIDHASH(slp, uid)->lh_first; - nuidp != 0; nuidp = nuidp->nu_hash.le_next) { - if (kauth_cred_getuid(nuidp->nu_cr) == uid && - (!nd->nd_nam2 || netaddr_match(NU_NETFAM(nuidp), - &nuidp->nu_haddr, nd->nd_nam2))) - break; - } - if (nuidp) { - ktvin.tv_sec = - txdr_unsigned(nuidp->nu_timestamp.tv_sec - 1); - ktvin.tv_usec = - txdr_unsigned(nuidp->nu_timestamp.tv_usec); + if (sotype == SOCK_STREAM) { + on = 4; /* don't wait too long for the socket to connect */ + sock_setsockopt(newso, IPPROTO_TCP, TCP_CONNECTIONTIMEOUT, &on, sizeof(on)); + error = sock_connect(newso, saddr, 0); + nfsmout_if(error); + } + if (((error = sock_setsockopt(newso, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)))) || + ((error = sock_setsockopt(newso, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)))) || + ((error = sock_setsockopt(newso, SOL_SOCKET, SO_NOADDRERR, &on, sizeof(on))))) + goto nfsmout; + so = newso; + } else { + /* make sure socket is using a one second timeout in this function */ + optlen = sizeof(orig_rcvto); + error = sock_getsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &orig_rcvto, &optlen); + if (!error) { + optlen = sizeof(orig_sndto); + error = sock_getsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &orig_sndto, &optlen); + } + if (!error) { + sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv)); + sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &tv, sizeof(tv)); + restoreto = 1; + } + } - /* - * Encrypt the timestamp in ecb mode using the - * session key. - */ -#if NFSKERB - XXX -#endif + if (sotype == SOCK_STREAM) { + sendat = 0; /* we only resend the request for UDP */ + nfs_rpc_record_state_init(&nrrs); + } - *tl++ = rpc_auth_kerb; - *tl++ = txdr_unsigned(3 * NFSX_UNSIGNED); - *tl = ktvout.tv_sec; - nfsm_build(tl, u_long *, 3 * NFSX_UNSIGNED); - *tl++ = ktvout.tv_usec; - *tl++ = txdr_unsigned(kauth_cred_getuid(nuidp->nu_cr)); - } else { - *tl++ = 0; - *tl++ = 0; - } - lck_rw_done(&slp->ns_rwlock); + for (try=0; try < timeo; try++) { + if ((error = nfs_sigintr(nmp, NULL, !try ? NULL : thd, 0))) + break; + if (!try || (try == sendat)) { + /* send the request (resending periodically for UDP) */ + if ((error = mbuf_copym(mreq, 0, MBUF_COPYALL, MBUF_WAITOK, &m))) + goto nfsmout; + bzero(&msg, sizeof(msg)); + if ((sotype == SOCK_DGRAM) && !sock_isconnected(so)) { + msg.msg_name = saddr; + msg.msg_namelen = saddr->sa_len; + } + if ((error = sock_sendmbuf(so, &msg, m, 0, &sentlen))) + goto nfsmout; + sendat *= 2; + if (sendat > 30) + sendat = 30; + } + /* wait for the response */ + if (sotype == SOCK_STREAM) { + /* try to read (more of) record */ + error = nfs_rpc_record_read(so, &nrrs, 0, &recv, &mrep); + /* if we don't have the whole record yet, we'll keep trying */ } else { - *tl++ = 0; - *tl++ = 0; + readlen = 1<<18; + bzero(&msg, sizeof(msg)); + error = sock_receivembuf(so, &msg, &mrep, 0, &readlen); } - switch (err) { - case EPROGUNAVAIL: - *tl = txdr_unsigned(RPC_PROGUNAVAIL); + if (error == EWOULDBLOCK) + continue; + nfsmout_if(error); + /* parse the response */ + nfsm_chain_dissect_init(error, nmrep, mrep); + nfsm_chain_get_32(error, nmrep, rxid); + nfsm_chain_get_32(error, nmrep, reply); + nfsmout_if(error); + if ((rxid != xid) || (reply != RPC_REPLY)) + error = EBADRPC; + nfsm_chain_get_32(error, nmrep, reply_status); + nfsmout_if(error); + if (reply_status == RPC_MSGDENIED) { + nfsm_chain_get_32(error, nmrep, rejected_status); + nfsmout_if(error); + error = (rejected_status == RPC_MISMATCH) ? ERPCMISMATCH : EACCES; + goto nfsmout; + } + nfsm_chain_get_32(error, nmrep, verf_type); /* verifier flavor */ + nfsm_chain_get_32(error, nmrep, verf_len); /* verifier length */ + nfsmout_if(error); + if (verf_len) + nfsm_chain_adv(error, nmrep, nfsm_rndup(verf_len)); + nfsm_chain_get_32(error, nmrep, accepted_status); + nfsmout_if(error); + switch (accepted_status) { + case RPC_SUCCESS: + error = 0; break; - case EPROGMISMATCH: - *tl = txdr_unsigned(RPC_PROGMISMATCH); - nfsm_build(tl, u_long *, 2 * NFSX_UNSIGNED); - // XXX hard coded versions - *tl++ = txdr_unsigned(2); - *tl = txdr_unsigned(3); + case RPC_PROGUNAVAIL: + error = EPROGUNAVAIL; break; - case EPROCUNAVAIL: - *tl = txdr_unsigned(RPC_PROCUNAVAIL); + case RPC_PROGMISMATCH: + error = EPROGMISMATCH; break; - case EBADRPC: - *tl = txdr_unsigned(RPC_GARBAGE); + case RPC_PROCUNAVAIL: + error = EPROCUNAVAIL; + break; + case RPC_GARBAGE: + error = EBADRPC; break; + case RPC_SYSTEM_ERR: default: - *tl = 0; - if (err != NFSERR_RETVOID) { - nfsm_build(tl, u_long *, NFSX_UNSIGNED); - if (err) - *tl = txdr_unsigned(nfsrv_errmap(nd, err)); - else - *tl = 0; - } + error = EIO; break; } + break; } - - if (mrq != NULL) - *mrq = mreq; - *mbp = mb; - *bposp = bpos; - if (err != 0 && err != NFSERR_RETVOID) { - OSAddAtomic(1, (SInt32*)&nfsstats.srvrpc_errs); +nfsmout: + if (restoreto) { + sock_setsockopt(so, SOL_SOCKET, SO_RCVTIMEO, &orig_rcvto, sizeof(tv)); + sock_setsockopt(so, SOL_SOCKET, SO_SNDTIMEO, &orig_sndto, sizeof(tv)); } - return (0); + if (newso) { + sock_shutdown(newso, SHUT_RDWR); + sock_close(newso); + } + mbuf_freem(mreq); + return (error); } - -#endif /* NFS_NOSERVER */ - - -/* - * From FreeBSD 1.58, a Matt Dillon fix... - * Flag a request as being about to terminate. - * The nm_sent count is decremented now to avoid deadlocks when the process - * in soreceive() hasn't yet managed to send its own request. - */ -static void -nfs_softterm(struct nfsreq *rep) +int +nfs_portmap_lookup( + struct nfsmount *nmp, + vfs_context_t ctx, + struct sockaddr *sa, + socket_t so, + uint32_t protocol, + uint32_t vers, + uint32_t ipproto, + int timeo) { + thread_t thd = vfs_context_thread(ctx); + kauth_cred_t cred = vfs_context_ucred(ctx); + struct sockaddr_storage ss; + struct sockaddr *saddr = (struct sockaddr*)&ss; + struct nfsm_chain nmreq, nmrep; + mbuf_t mreq; + int error = 0, ip, pmprog, pmvers, pmproc, ualen = 0; + uint32_t port; + uint64_t xid = 0; + char uaddr[MAX_IPv6_STR_LEN+16]; + + bcopy(sa, saddr, min(sizeof(ss), sa->sa_len)); + if (saddr->sa_family == AF_INET) { + ip = 4; + pmprog = PMAPPROG; + pmvers = PMAPVERS; + pmproc = PMAPPROC_GETPORT; + } else if (saddr->sa_family == AF_INET6) { + ip = 6; + pmprog = RPCBPROG; + pmvers = RPCBVERS4; + pmproc = RPCBPROC_GETVERSADDR; + } else { + return (EINVAL); + } + nfsm_chain_null(&nmreq); + nfsm_chain_null(&nmrep); - rep->r_flags |= R_SOFTTERM; - if (rep->r_flags & R_SENT) { - FSDBG(532, rep->r_xid, rep, rep->r_nmp->nm_sent, - rep->r_nmp->nm_cwnd); - rep->r_nmp->nm_sent -= NFS_CWNDSCALE; - rep->r_flags &= ~R_SENT; +tryagain: + /* send portmapper request to get port/uaddr */ + if (ip == 4) + ((struct sockaddr_in*)saddr)->sin_port = htons(PMAPPORT); + else + ((struct sockaddr_in6*)saddr)->sin6_port = htons(PMAPPORT); + nfsm_chain_build_alloc_init(error, &nmreq, 8*NFSX_UNSIGNED); + nfsm_chain_add_32(error, &nmreq, protocol); + nfsm_chain_add_32(error, &nmreq, vers); + if (ip == 4) { + nfsm_chain_add_32(error, &nmreq, ipproto); + nfsm_chain_add_32(error, &nmreq, 0); + } else { + if (ipproto == IPPROTO_TCP) + nfsm_chain_add_string(error, &nmreq, "tcp6", 4); + else + nfsm_chain_add_string(error, &nmreq, "udp6", 4); + nfsm_chain_add_string(error, &nmreq, "", 0); /* uaddr */ + nfsm_chain_add_string(error, &nmreq, "", 0); /* owner */ } + nfsm_chain_build_done(error, &nmreq); + nfsmout_if(error); + error = nfsm_rpchead2(nmp, (ipproto == IPPROTO_UDP) ? SOCK_DGRAM : SOCK_STREAM, + pmprog, pmvers, pmproc, RPCAUTH_SYS, cred, NULL, nmreq.nmc_mhead, + &xid, &mreq); + nfsmout_if(error); + nmreq.nmc_mhead = NULL; + error = nfs_aux_request(nmp, thd, saddr, so, (ipproto == IPPROTO_UDP) ? SOCK_DGRAM : SOCK_STREAM, + mreq, R_XID32(xid), 0, timeo, &nmrep); + + /* grab port from portmap response */ + if (ip == 4) { + nfsm_chain_get_32(error, &nmrep, port); + if (!error) + ((struct sockaddr_in*)sa)->sin_port = htons(port); + } else { + /* get uaddr string and convert to sockaddr */ + nfsm_chain_get_32(error, &nmrep, ualen); + if (!error) { + if (ualen > ((int)sizeof(uaddr)-1)) + error = EIO; + if (ualen < 1) { + /* program is not available, just return a zero port */ + bcopy(sa, saddr, min(sizeof(ss), sa->sa_len)); + ((struct sockaddr_in6*)saddr)->sin6_port = htons(0); + } else { + nfsm_chain_get_opaque(error, &nmrep, ualen, uaddr); + if (!error) { + uaddr[ualen] = '\0'; + if (!nfs_uaddr2sockaddr(uaddr, saddr)) + error = EIO; + } + } + } + if ((error == EPROGMISMATCH) || (error == EPROCUNAVAIL) || (error == EIO) || (error == EBADRPC)) { + /* remote doesn't support rpcbind version or proc (or we couldn't parse uaddr) */ + if (pmvers == RPCBVERS4) { + /* fall back to v3 and GETADDR */ + pmvers = RPCBVERS3; + pmproc = RPCBPROC_GETADDR; + nfsm_chain_cleanup(&nmreq); + nfsm_chain_cleanup(&nmrep); + bcopy(sa, saddr, min(sizeof(ss), sa->sa_len)); + xid = 0; + error = 0; + goto tryagain; + } + } + if (!error) + bcopy(saddr, sa, min(saddr->sa_len, sa->sa_len)); + } +nfsmout: + nfsm_chain_cleanup(&nmreq); + nfsm_chain_cleanup(&nmrep); + return (error); } -void -nfs_timer_funnel(void * arg) +int +nfs_msg(thread_t thd, + const char *server, + const char *msg, + int error) { - (void) thread_funnel_set(kernel_flock, TRUE); - nfs_timer(arg); - (void) thread_funnel_set(kernel_flock, FALSE); + proc_t p = thd ? get_bsdthreadtask_info(thd) : NULL; + tpr_t tpr; + if (p) + tpr = tprintf_open(p); + else + tpr = NULL; + if (error) + tprintf(tpr, "nfs server %s: %s, error %d\n", server, msg, error); + else + tprintf(tpr, "nfs server %s: %s\n", server, msg); + tprintf_close(tpr); + return (0); } -/* - * Ensure rep isn't in use by the timer, then dequeue it. - */ -static void -nfs_repdequeue(struct nfsreq *rep) -{ +#define NFS_SQUISH_MOBILE_ONLY 0x0001 /* Squish mounts only on mobile machines */ +#define NFS_SQUISH_AUTOMOUNTED_ONLY 0x0002 /* Squish mounts only if the are automounted */ +#define NFS_SQUISH_SOFT 0x0004 /* Treat all soft mounts as though they were on a mobile machine */ +#define NFS_SQUISH_QUICK 0x0008 /* Try to squish mounts more quickly. */ +#define NFS_SQUISH_SHUTDOWN 0x1000 /* Squish all mounts on shutdown. Currently not implemented */ - while ((rep->r_flags & R_BUSY)) { - rep->r_flags |= R_WAITING; - tsleep(rep, PSOCK, "repdeq", 0); - } - TAILQ_REMOVE(&nfs_reqq, rep, r_chain); -} +uint32_t nfs_squishy_flags = NFS_SQUISH_MOBILE_ONLY | NFS_SQUISH_AUTOMOUNTED_ONLY | NFS_SQUISH_QUICK; +int32_t nfs_is_mobile; + +#define NFS_SQUISHY_DEADTIMEOUT 8 /* Dead time out for squishy mounts */ +#define NFS_SQUISHY_QUICKTIMEOUT 4 /* Quicker dead time out when nfs_squish_flags NFS_SQUISH_QUICK bit is set*/ /* - * Busy (lock) a nfsreq, used by the nfs timer to make sure it's not - * free()'d out from under it. + * Could this mount be squished? */ -static void -nfs_repbusy(struct nfsreq *rep) +int +nfs_can_squish(struct nfsmount *nmp) { + uint64_t flags = vfs_flags(nmp->nm_mountp); + int softsquish = ((nfs_squishy_flags & NFS_SQUISH_SOFT) & NMFLAG(nmp, SOFT)); - if ((rep->r_flags & R_BUSY)) - panic("rep locked"); - rep->r_flags |= R_BUSY; -} + if (!softsquish && (nfs_squishy_flags & NFS_SQUISH_MOBILE_ONLY) && nfs_is_mobile == 0) + return (0); -/* - * Unbusy the nfsreq passed in, return the next nfsreq in the chain busied. - */ -static struct nfsreq * -nfs_repnext(struct nfsreq *rep) -{ - struct nfsreq * nextrep; + if ((nfs_squishy_flags & NFS_SQUISH_AUTOMOUNTED_ONLY) && (flags & MNT_AUTOMOUNTED) == 0) + return (0); - if (rep == NULL) - return (NULL); - /* - * We need to get and busy the next req before signalling the - * current one, otherwise wakeup() may block us and we'll race to - * grab the next req. - */ - nextrep = TAILQ_NEXT(rep, r_chain); - if (nextrep != NULL) - nfs_repbusy(nextrep); - /* unbusy and signal. */ - rep->r_flags &= ~R_BUSY; - if ((rep->r_flags & R_WAITING)) { - rep->r_flags &= ~R_WAITING; - wakeup(rep); - } - return (nextrep); + return (1); } /* - * Nfs timer routine - * Scan the nfsreq list and retranmit any requests that have timed out - * To avoid retransmission attempts on STREAM sockets (in the future) make - * sure to set the r_retry field to 0 (implies nm_retry == 0). + * NFS mounts default to "rw,hard" - but frequently on mobile clients + * the mount may become "not responding". It's desirable to be able + * to unmount these dead mounts, but only if there is no risk of + * losing data or crashing applications. A "squishy" NFS mount is one + * that can be force unmounted with little risk of harm. + * + * nfs_is_squishy checks if a mount is in a squishy state. A mount is + * in a squishy state iff it is allowed to be squishy and there are no + * dirty pages and there are no mmapped files and there are no files + * open for write. Mounts are allowed to be squishy is controlled by + * the settings of the nfs_squishy_flags and its mobility state. These + * flags can be set by sysctls. + * + * If nfs_is_squishy determines that we are in a squishy state we will + * update the current dead timeout to at least NFS_SQUISHY_DEADTIMEOUT + * (or NFS_SQUISHY_QUICKTIMEOUT if NFS_SQUISH_QUICK is set) (see + * above) or 1/8th of the mount's nm_deadtimeout value, otherwise we just + * update the current dead timeout with the mount's nm_deadtimeout + * value set at mount time. + * + * Assumes that nm_lock is held. + * + * Note this routine is racey, but its effects on setting the + * dead timeout only have effects when we're in trouble and are likely + * to stay that way. Since by default its only for automounted + * volumes on mobile machines; this is a reasonable trade off between + * data integrity and user experience. It can be disabled or set via + * nfs.conf file. */ -void -nfs_timer(__unused void *arg) -{ - struct nfsreq *rep; - mbuf_t m; - socket_t so; - struct nfsmount *nmp; - int timeo; - int error; -#ifndef NFS_NOSERVER - struct nfssvc_sock *slp; - u_quad_t cur_usec; -#endif /* NFS_NOSERVER */ - int flags, rexmit, cwnd, sent; - u_long xid; - struct timeval now; - rep = TAILQ_FIRST(&nfs_reqq); - if (rep != NULL) - nfs_repbusy(rep); - microuptime(&now); - for ( ; rep != NULL ; rep = nfs_repnext(rep)) { - nmp = rep->r_nmp; - if (!nmp) /* unmounted */ - continue; - if (rep->r_mrep || (rep->r_flags & R_SOFTTERM)) - continue; - if (nfs_sigintr(nmp, rep, rep->r_procp)) - continue; - if (nmp->nm_tprintf_initial_delay != 0 && - (rep->r_rexmit > 2 || (rep->r_flags & R_RESENDERR)) && - rep->r_lastmsg + nmp->nm_tprintf_delay < now.tv_sec) { - rep->r_lastmsg = now.tv_sec; - nfs_down(rep->r_nmp, rep->r_procp, 0, NFSSTA_TIMEO, - "not responding"); - rep->r_flags |= R_TPRINTFMSG; - if (!(nmp->nm_state & NFSSTA_MOUNTED)) { - /* we're not yet completely mounted and */ - /* we can't complete an RPC, so we fail */ - OSAddAtomic(1, (SInt32*)&nfsstats.rpctimeouts); - nfs_softterm(rep); - continue; - } - } - if (rep->r_rtt >= 0) { - rep->r_rtt++; - if (nmp->nm_flag & NFSMNT_DUMBTIMR) - timeo = nmp->nm_timeo; - else - timeo = NFS_RTO(nmp, proct[rep->r_procnum]); - /* ensure 62.5 ms floor */ - while (16 * timeo < hz) - timeo *= 2; - if (nmp->nm_timeouts > 0) - timeo *= nfs_backoff[nmp->nm_timeouts - 1]; - if (rep->r_rtt <= timeo) - continue; - if (nmp->nm_timeouts < 8) - nmp->nm_timeouts++; - } - /* - * Check for too many retransmits. This is never true for - * 'hard' mounts because we set r_retry to NFS_MAXREXMIT + 1 - * and never allow r_rexmit to be more than NFS_MAXREXMIT. - */ - if (rep->r_rexmit >= rep->r_retry) { /* too many */ - OSAddAtomic(1, (SInt32*)&nfsstats.rpctimeouts); - nfs_softterm(rep); - continue; - } - if (nmp->nm_sotype != SOCK_DGRAM) { - if (++rep->r_rexmit > NFS_MAXREXMIT) - rep->r_rexmit = NFS_MAXREXMIT; - continue; - } - if ((so = nmp->nm_so) == NULL) - continue; - - /* - * If there is enough space and the window allows.. - * Resend it - * Set r_rtt to -1 in case we fail to send it now. - */ - rep->r_rtt = -1; - if (((nmp->nm_flag & NFSMNT_DUMBTIMR) || - (rep->r_flags & R_SENT) || - nmp->nm_sent < nmp->nm_cwnd) && - (mbuf_copym(rep->r_mreq, 0, MBUF_COPYALL, MBUF_DONTWAIT, &m) == 0)){ - struct msghdr msg; - /* - * Iff first send, start timing - * else turn timing off, backoff timer - * and divide congestion window by 2. - * We update these *before* the send to avoid - * racing against receiving the reply. - * We save them so we can restore them on send error. - */ - flags = rep->r_flags; - rexmit = rep->r_rexmit; - cwnd = nmp->nm_cwnd; - sent = nmp->nm_sent; - xid = rep->r_xid; - if (rep->r_flags & R_SENT) { - rep->r_flags &= ~R_TIMING; - if (++rep->r_rexmit > NFS_MAXREXMIT) - rep->r_rexmit = NFS_MAXREXMIT; - nmp->nm_cwnd >>= 1; - if (nmp->nm_cwnd < NFS_CWNDSCALE) - nmp->nm_cwnd = NFS_CWNDSCALE; - OSAddAtomic(1, (SInt32*)&nfsstats.rpcretries); - } else { - rep->r_flags |= R_SENT; - nmp->nm_sent += NFS_CWNDSCALE; - } - FSDBG(535, xid, rep, nmp->nm_sent, nmp->nm_cwnd); +int +nfs_is_squishy(struct nfsmount *nmp) +{ + mount_t mp = nmp->nm_mountp; + int squishy = 0; + int timeo = (nfs_squishy_flags & NFS_SQUISH_QUICK) ? NFS_SQUISHY_QUICKTIMEOUT : NFS_SQUISHY_DEADTIMEOUT; - bzero(&msg, sizeof(msg)); - if ((nmp->nm_flag & NFSMNT_NOCONN) == NFSMNT_NOCONN) { - msg.msg_name = mbuf_data(nmp->nm_nam); - msg.msg_namelen = mbuf_len(nmp->nm_nam); - } - error = sock_sendmbuf(so, &msg, m, MSG_DONTWAIT, NULL); + NFS_SOCK_DBG("%s: nm_curdeadtiemout = %d, nfs_is_mobile = %d\n", + vfs_statfs(mp)->f_mntfromname, nmp->nm_curdeadtimeout, nfs_is_mobile); - FSDBG(535, xid, error, sent, cwnd); + if (!nfs_can_squish(nmp)) + goto out; - if (error) { - if (error == EWOULDBLOCK) { - rep->r_flags = flags; - rep->r_rexmit = rexmit; - nmp->nm_cwnd = cwnd; - nmp->nm_sent = sent; - rep->r_xid = xid; - } - else { - if (NFSIGNORE_SOERROR(nmp->nm_sotype, error)) { - int clearerror; - int optlen = sizeof(clearerror); - sock_getsockopt(nmp->nm_so, SOL_SOCKET, SO_ERROR, &clearerror, &optlen); - } - rep->r_flags = flags | R_RESENDERR; - rep->r_rexmit = rexmit; - nmp->nm_cwnd = cwnd; - nmp->nm_sent = sent; - if (flags & R_SENT) - OSAddAtomic(-1, (SInt32*)&nfsstats.rpcretries); - } - } else - rep->r_rtt = 0; - } - } - microuptime(&now); -#ifndef NFS_NOSERVER - /* - * Scan the write gathering queues for writes that need to be - * completed now. - */ - cur_usec = (u_quad_t)now.tv_sec * 1000000 + (u_quad_t)now.tv_usec; - lck_mtx_lock(nfsd_mutex); - TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) { - if (slp->ns_wgtime && (slp->ns_wgtime <= cur_usec)) - nfsrv_wakenfsd(slp); - } - while ((slp = TAILQ_FIRST(&nfssvc_deadsockhead))) { - if ((slp->ns_timestamp + 5) > now.tv_sec) - break; - TAILQ_REMOVE(&nfssvc_deadsockhead, slp, ns_chain); - nfsrv_slpfree(slp); - } - lck_mtx_unlock(nfsd_mutex); -#endif /* NFS_NOSERVER */ + timeo = (nmp->nm_deadtimeout > timeo) ? max(nmp->nm_deadtimeout/8, timeo) : timeo; + NFS_SOCK_DBG("nm_writers = %d nm_mappers = %d timeo = %d\n", nmp->nm_writers, nmp->nm_mappers, timeo); - if (nfsbuffreeuptimestamp + 30 <= now.tv_sec) { - /* - * We haven't called nfs_buf_freeup() in a little while. - * So, see if we can free up any stale/unused bufs now. + if (nmp->nm_writers == 0 && nmp->nm_mappers == 0) { + uint64_t flags = mp ? vfs_flags(mp) : 0; + squishy = 1; + + /* + * Walk the nfs nodes and check for dirty buffers it we're not + * RDONLY and we've not already been declared as squishy since + * this can be a bit expensive. */ - nfs_buf_freeup(1); + if (!(flags & MNT_RDONLY) && !(nmp->nm_state & NFSSTA_SQUISHY)) + squishy = !nfs_mount_is_dirty(mp); } - timeout(nfs_timer_funnel, (void *)0, nfs_ticks); +out: + if (squishy) + nmp->nm_state |= NFSSTA_SQUISHY; + else + nmp->nm_state &= ~NFSSTA_SQUISHY; -} + nmp->nm_curdeadtimeout = squishy ? timeo : nmp->nm_deadtimeout; + + NFS_SOCK_DBG("nm_curdeadtimeout = %d\n", nmp->nm_curdeadtimeout); + return (squishy); +} /* - * Test for a termination condition pending on the process. - * This is used to determine if we need to bail on a mount. - * EIO is returned if there has been a soft timeout. - * EINTR is returned if there is a signal pending that is not being ignored - * and the mount is interruptable, or if we are a thread that is in the process - * of cancellation (also SIGKILL posted). + * On a send operation, if we can't reach the server and we've got only one server to talk to + * and NFS_SQUISH_QUICK flag is set and we are in a squishy state then mark the mount as dead + * and ask to be forcibly unmounted. Return 1 if we're dead and 0 otherwise. */ -int -nfs_sigintr(nmp, rep, p) - struct nfsmount *nmp; - struct nfsreq *rep; - proc_t p; +static int +nfs_is_dead_lock(int error, struct nfsmount *nmp) { - sigset_t pending_sigs; - int context_good = 0; - struct nfsmount *repnmp; - extern proc_t kernproc; + if (nmp->nm_state & NFSSTA_DEAD) + return (1); - if (nmp == NULL) - return (ENXIO); - if (rep != NULL) { - repnmp = rep->r_nmp; - /* we've had a forced unmount. */ - if (repnmp == NULL) - return (ENXIO); - /* request has timed out on a 'soft' mount. */ - if (rep->r_flags & R_SOFTTERM) - return (EIO); - /* - * We're in the progress of a force unmount and there's - * been a timeout we're dead and fail IO. - */ - if ((repnmp->nm_state & (NFSSTA_FORCE|NFSSTA_TIMEO)) == - (NFSSTA_FORCE|NFSSTA_TIMEO)) - return (EIO); - /* Someone is unmounting us, go soft and mark it. */ - if (repnmp->nm_mountp->mnt_kern_flag & MNTK_FRCUNMOUNT) { - repnmp->nm_flag |= NFSMNT_SOFT; - nmp->nm_state |= NFSSTA_FORCE; - } - /* - * If the mount is hung and we've requested not to hang - * on remote filesystems, then bail now. - */ - if (p != NULL && (proc_noremotehang(p)) != 0 && - (repnmp->nm_state & NFSSTA_TIMEO) != 0) - return (EIO); - } - /* XXX: is this valid? this probably should be an assertion. */ - if (p == NULL) + if ((error != ENETUNREACH && error != EHOSTUNREACH && error != EADDRNOTAVAIL) || + !(nmp->nm_locations.nl_numlocs == 1 && nmp->nm_locations.nl_locations[0]->nl_servcount == 1)) return (0); - /* Is this thread belongs to kernel task; then abort check is not needed */ - if ((current_proc() != kernproc) && current_thread_aborted()) { - return (EINTR); + if ((nfs_squishy_flags & NFS_SQUISH_QUICK) && nfs_is_squishy(nmp)) { + printf("nfs_is_dead: nfs server %s: unreachable. Squished dead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname); + nmp->nm_state |= NFSSTA_DEAD; + vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_DEAD, 0); + return (1); } - /* mask off thread and process blocked signals. */ - - pending_sigs = proc_pendingsignals(p, NFSINT_SIGMASK); - if (pending_sigs && (nmp->nm_flag & NFSMNT_INT) != 0) - return (EINTR); return (0); } -/* - * Lock a socket against others. - * Necessary for STREAM sockets to ensure you get an entire rpc request/reply - * and also to avoid race conditions between the processes with nfs requests - * in progress when a reconnect is necessary. - */ int -nfs_sndlock(rep) - struct nfsreq *rep; +nfs_is_dead(int error, struct nfsmount *nmp) { - int *statep; - proc_t p; - int error, slpflag = 0, slptimeo = 0; + int is_dead; + + lck_mtx_lock(&nmp->nm_lock); + is_dead = nfs_is_dead_lock(error, nmp); + lck_mtx_unlock(&nmp->nm_lock); + + return (is_dead); +} + +void +nfs_down(struct nfsmount *nmp, thread_t thd, int error, int flags, const char *msg) +{ + int timeoutmask, wasunresponsive, unresponsive, softnobrowse; + uint32_t do_vfs_signal; + struct timeval now; + + if (nmp == NULL) + return; + + lck_mtx_lock(&nmp->nm_lock); + + timeoutmask = NFSSTA_TIMEO | NFSSTA_LOCKTIMEO | NFSSTA_JUKEBOXTIMEO; + if (NMFLAG(nmp, MUTEJUKEBOX)) /* jukebox timeouts don't count as unresponsive if muted */ + timeoutmask &= ~NFSSTA_JUKEBOXTIMEO; + wasunresponsive = (nmp->nm_state & timeoutmask); - if (rep->r_nmp == NULL) - return (ENXIO); - statep = &rep->r_nmp->nm_state; + /* XXX don't allow users to know about/disconnect unresponsive, soft, nobrowse mounts */ + softnobrowse = (NMFLAG(nmp, SOFT) && (vfs_flags(nmp->nm_mountp) & MNT_DONTBROWSE)); - p = rep->r_procp; - if (rep->r_nmp->nm_flag & NFSMNT_INT) - slpflag = PCATCH; - while (*statep & NFSSTA_SNDLOCK) { - error = nfs_sigintr(rep->r_nmp, rep, p); - if (error) - return (error); - *statep |= NFSSTA_WANTSND; - if (p != NULL && (proc_noremotehang(p)) != 0) - slptimeo = hz; - tsleep((caddr_t)statep, slpflag | (PZERO - 1), "nfsndlck", slptimeo); - if (slpflag == PCATCH) { - slpflag = 0; - slptimeo = 2 * hz; + if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) + nmp->nm_state |= NFSSTA_TIMEO; + if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) + nmp->nm_state |= NFSSTA_LOCKTIMEO; + if ((flags & NFSSTA_JUKEBOXTIMEO) && !(nmp->nm_state & NFSSTA_JUKEBOXTIMEO)) + nmp->nm_state |= NFSSTA_JUKEBOXTIMEO; + + unresponsive = (nmp->nm_state & timeoutmask); + + nfs_is_squishy(nmp); + + if (unresponsive && (nmp->nm_curdeadtimeout > 0)) { + microuptime(&now); + if (!wasunresponsive) { + nmp->nm_deadto_start = now.tv_sec; + nfs_mount_sock_thread_wake(nmp); + } else if ((now.tv_sec - nmp->nm_deadto_start) > nmp->nm_curdeadtimeout) { + if (!(nmp->nm_state & NFSSTA_DEAD)) + printf("nfs server %s: %sdead\n", vfs_statfs(nmp->nm_mountp)->f_mntfromname, + (nmp->nm_curdeadtimeout != nmp->nm_deadtimeout) ? "squished " : ""); + nmp->nm_state |= NFSSTA_DEAD; } - /* - * Make sure while we slept that the mountpoint didn't go away. - * nfs_sigintr and callers expect it in tact. - */ - if (!rep->r_nmp) - return (ENXIO); /* don't have lock until out of loop */ } - *statep |= NFSSTA_SNDLOCK; - return (0); + lck_mtx_unlock(&nmp->nm_lock); + + if (nmp->nm_state & NFSSTA_DEAD) + do_vfs_signal = VQ_DEAD; + else if (softnobrowse || wasunresponsive || !unresponsive) + do_vfs_signal = 0; + else + do_vfs_signal = VQ_NOTRESP; + if (do_vfs_signal) + vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, do_vfs_signal, 0); + + nfs_msg(thd, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, error); } -/* - * Unlock the stream socket for others. - */ void -nfs_sndunlock(rep) - struct nfsreq *rep; +nfs_up(struct nfsmount *nmp, thread_t thd, int flags, const char *msg) { - int *statep; + int timeoutmask, wasunresponsive, unresponsive, softnobrowse; + int do_vfs_signal; - if (rep->r_nmp == NULL) + if (nmp == NULL) return; - statep = &rep->r_nmp->nm_state; - if ((*statep & NFSSTA_SNDLOCK) == 0) - panic("nfs sndunlock"); - *statep &= ~NFSSTA_SNDLOCK; - if (*statep & NFSSTA_WANTSND) { - *statep &= ~NFSSTA_WANTSND; - wakeup((caddr_t)statep); - } + + if (msg) + nfs_msg(thd, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, 0); + + lck_mtx_lock(&nmp->nm_lock); + + timeoutmask = NFSSTA_TIMEO | NFSSTA_LOCKTIMEO | NFSSTA_JUKEBOXTIMEO; + if (NMFLAG(nmp, MUTEJUKEBOX)) /* jukebox timeouts don't count as unresponsive if muted */ + timeoutmask &= ~NFSSTA_JUKEBOXTIMEO; + wasunresponsive = (nmp->nm_state & timeoutmask); + + /* XXX don't allow users to know about/disconnect unresponsive, soft, nobrowse mounts */ + softnobrowse = (NMFLAG(nmp, SOFT) && (vfs_flags(nmp->nm_mountp) & MNT_DONTBROWSE)); + + if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) + nmp->nm_state &= ~NFSSTA_TIMEO; + if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) + nmp->nm_state &= ~NFSSTA_LOCKTIMEO; + if ((flags & NFSSTA_JUKEBOXTIMEO) && (nmp->nm_state & NFSSTA_JUKEBOXTIMEO)) + nmp->nm_state &= ~NFSSTA_JUKEBOXTIMEO; + + unresponsive = (nmp->nm_state & timeoutmask); + + nmp->nm_deadto_start = 0; + nmp->nm_curdeadtimeout = nmp->nm_deadtimeout; + nmp->nm_state &= ~NFSSTA_SQUISHY; + lck_mtx_unlock(&nmp->nm_lock); + + if (softnobrowse) + do_vfs_signal = 0; + else + do_vfs_signal = (wasunresponsive && !unresponsive); + if (do_vfs_signal) + vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESP, 1); } -static int -nfs_rcvlock(struct nfsreq *rep) + +#endif /* NFSCLIENT */ + +#if NFSSERVER + +/* + * Generate the rpc reply header + * siz arg. is used to decide if adding a cluster is worthwhile + */ +int +nfsrv_rephead( + struct nfsrv_descript *nd, + __unused struct nfsrv_sock *slp, + struct nfsm_chain *nmrepp, + size_t siz) { - int *statep; - int error, slpflag, slptimeo = 0; + mbuf_t mrep; + u_int32_t *tl; + struct nfsm_chain nmrep; + int err, error; - /* make sure we still have our mountpoint */ - if (!rep->r_nmp) { - if (rep->r_mrep != NULL) - return (EALREADY); - return (ENXIO); - } + err = nd->nd_repstat; + if (err && (nd->nd_vers == NFS_VER2)) + siz = 0; - statep = &rep->r_nmp->nm_state; - FSDBG_TOP(534, rep->r_xid, rep, rep->r_nmp, *statep); - if (rep->r_nmp->nm_flag & NFSMNT_INT) - slpflag = PCATCH; - else - slpflag = 0; - while (*statep & NFSSTA_RCVLOCK) { - if ((error = nfs_sigintr(rep->r_nmp, rep, rep->r_procp))) { - FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x100); - return (error); - } else if (rep->r_mrep != NULL) { - /* - * Don't bother sleeping if reply already arrived - */ - FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x101); - return (EALREADY); + /* + * If this is a big reply, use a cluster else + * try and leave leading space for the lower level headers. + */ + siz += RPC_REPLYSIZ; + if (siz >= nfs_mbuf_minclsize) { + error = mbuf_getpacket(MBUF_WAITOK, &mrep); + } else { + error = mbuf_gethdr(MBUF_WAITOK, MBUF_TYPE_DATA, &mrep); + } + if (error) { + /* unable to allocate packet */ + /* XXX should we keep statistics for these errors? */ + return (error); + } + if (siz < nfs_mbuf_minclsize) { + /* leave space for lower level headers */ + tl = mbuf_data(mrep); + tl += 80/sizeof(*tl); /* XXX max_hdr? XXX */ + mbuf_setdata(mrep, tl, 6 * NFSX_UNSIGNED); + } + nfsm_chain_init(&nmrep, mrep); + nfsm_chain_add_32(error, &nmrep, nd->nd_retxid); + nfsm_chain_add_32(error, &nmrep, RPC_REPLY); + if (err == ERPCMISMATCH || (err & NFSERR_AUTHERR)) { + nfsm_chain_add_32(error, &nmrep, RPC_MSGDENIED); + if (err & NFSERR_AUTHERR) { + nfsm_chain_add_32(error, &nmrep, RPC_AUTHERR); + nfsm_chain_add_32(error, &nmrep, (err & ~NFSERR_AUTHERR)); + } else { + nfsm_chain_add_32(error, &nmrep, RPC_MISMATCH); + nfsm_chain_add_32(error, &nmrep, RPC_VER2); + nfsm_chain_add_32(error, &nmrep, RPC_VER2); } - FSDBG(534, rep->r_xid, rep, rep->r_nmp, 0x102); - *statep |= NFSSTA_WANTRCV; - /* - * We need to poll if we're P_NOREMOTEHANG so that we - * call nfs_sigintr periodically above. - */ - if (rep->r_procp != NULL && - (proc_noremotehang(rep->r_procp)) != 0) - slptimeo = hz; - tsleep((caddr_t)statep, slpflag | (PZERO - 1), "nfsrcvlk", slptimeo); - if (slpflag == PCATCH) { - slpflag = 0; - slptimeo = 2 * hz; + } else { + /* reply status */ + nfsm_chain_add_32(error, &nmrep, RPC_MSGACCEPTED); + if (nd->nd_gss_context != NULL) { + /* RPCSEC_GSS verifier */ + error = nfs_gss_svc_verf_put(nd, &nmrep); + if (error) { + nfsm_chain_add_32(error, &nmrep, RPC_SYSTEM_ERR); + goto done; + } + } else { + /* RPCAUTH_NULL verifier */ + nfsm_chain_add_32(error, &nmrep, RPCAUTH_NULL); + nfsm_chain_add_32(error, &nmrep, 0); } - /* - * Make sure while we slept that the mountpoint didn't go away. - * nfs_sigintr and caller nfs_reply expect it intact. - */ - if (!rep->r_nmp) { - FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, 0x103); - return (ENXIO); /* don't have lock until out of loop */ + /* accepted status */ + switch (err) { + case EPROGUNAVAIL: + nfsm_chain_add_32(error, &nmrep, RPC_PROGUNAVAIL); + break; + case EPROGMISMATCH: + nfsm_chain_add_32(error, &nmrep, RPC_PROGMISMATCH); + /* XXX hard coded versions? */ + nfsm_chain_add_32(error, &nmrep, NFS_VER2); + nfsm_chain_add_32(error, &nmrep, NFS_VER3); + break; + case EPROCUNAVAIL: + nfsm_chain_add_32(error, &nmrep, RPC_PROCUNAVAIL); + break; + case EBADRPC: + nfsm_chain_add_32(error, &nmrep, RPC_GARBAGE); + break; + default: + nfsm_chain_add_32(error, &nmrep, RPC_SUCCESS); + if (nd->nd_gss_context != NULL) + error = nfs_gss_svc_prepare_reply(nd, &nmrep); + if (err != NFSERR_RETVOID) + nfsm_chain_add_32(error, &nmrep, + (err ? nfsrv_errmap(nd, err) : 0)); + break; } } - /* - * nfs_reply will handle it if reply already arrived. - * (We may have slept or been preempted). - */ - FSDBG_BOT(534, rep->r_xid, rep, rep->r_nmp, *statep); - *statep |= NFSSTA_RCVLOCK; + +done: + nfsm_chain_build_done(error, &nmrep); + if (error) { + /* error composing reply header */ + /* XXX should we keep statistics for these errors? */ + mbuf_freem(mrep); + return (error); + } + + *nmrepp = nmrep; + if ((err != 0) && (err != NFSERR_RETVOID)) + OSAddAtomic64(1, &nfsstats.srvrpc_errs); return (0); } /* - * Unlock the stream socket for others. + * The nfs server send routine. + * + * - return EINTR or ERESTART if interrupted by a signal + * - return EPIPE if a connection is lost for connection based sockets (TCP...) + * - do any cleanup required by recoverable socket errors (???) */ -static void -nfs_rcvunlock(struct nfsreq *rep) +int +nfsrv_send(struct nfsrv_sock *slp, mbuf_t nam, mbuf_t top) { - int *statep; - - if (rep->r_nmp == NULL) - return; - statep = &rep->r_nmp->nm_state; + int error; + socket_t so = slp->ns_so; + struct sockaddr *sendnam; + struct msghdr msg; - FSDBG(533, statep, *statep, 0, 0); - if ((*statep & NFSSTA_RCVLOCK) == 0) - panic("nfs rcvunlock"); - *statep &= ~NFSSTA_RCVLOCK; - if (*statep & NFSSTA_WANTRCV) { - *statep &= ~NFSSTA_WANTRCV; - wakeup((caddr_t)statep); + bzero(&msg, sizeof(msg)); + if (nam && !sock_isconnected(so) && (slp->ns_sotype != SOCK_STREAM)) { + if ((sendnam = mbuf_data(nam))) { + msg.msg_name = (caddr_t)sendnam; + msg.msg_namelen = sendnam->sa_len; + } } -} + error = sock_sendmbuf(so, &msg, top, 0, NULL); + if (!error) + return (0); + log(LOG_INFO, "nfsd send error %d\n", error); + + if ((error == EWOULDBLOCK) && (slp->ns_sotype == SOCK_STREAM)) + error = EPIPE; /* zap TCP sockets if they time out on send */ + + /* Handle any recoverable (soft) socket errors here. (???) */ + if (error != EINTR && error != ERESTART && error != EIO && + error != EWOULDBLOCK && error != EPIPE) + error = 0; + return (error); +} -#ifndef NFS_NOSERVER /* * Socket upcall routine for the nfsd sockets. - * The caddr_t arg is a pointer to the "struct nfssvc_sock". + * The caddr_t arg is a pointer to the "struct nfsrv_sock". * Essentially do as much as possible non-blocking, else punt and it will * be called with MBUF_WAITOK from an nfsd. */ void -nfsrv_rcv(socket_t so, caddr_t arg, int waitflag) +nfsrv_rcv(socket_t so, void *arg, int waitflag) { - struct nfssvc_sock *slp = (struct nfssvc_sock *)arg; + struct nfsrv_sock *slp = arg; - if (!nfs_numnfsd || !(slp->ns_flag & SLP_VALID)) + if (!nfsd_thread_count || !(slp->ns_flag & SLP_VALID)) return; lck_rw_lock_exclusive(&slp->ns_rwlock); @@ -2148,7 +5702,7 @@ nfsrv_rcv(socket_t so, caddr_t arg, int waitflag) /* Note: ns_rwlock gets dropped when called with MBUF_DONTWAIT */ } void -nfsrv_rcv_locked(socket_t so, struct nfssvc_sock *slp, int waitflag) +nfsrv_rcv_locked(socket_t so, struct nfsrv_sock *slp, int waitflag) { mbuf_t m, mp, mhck, m2; int ns_flag=0, error; @@ -2173,10 +5727,10 @@ nfsrv_rcv_locked(socket_t so, struct nfssvc_sock *slp, int waitflag) if (slp->ns_sotype == SOCK_STREAM) { /* * If there are already records on the queue, defer soreceive() - * to an nfsd so that there is feedback to the TCP layer that + * to an(other) nfsd so that there is feedback to the TCP layer that * the nfs servers are heavily loaded. */ - if (slp->ns_rec && waitflag == MBUF_DONTWAIT) { + if (slp->ns_rec) { ns_flag = SLP_NEEDQ; goto dorecs; } @@ -2188,7 +5742,7 @@ nfsrv_rcv_locked(socket_t so, struct nfssvc_sock *slp, int waitflag) error = sock_receivembuf(so, NULL, &mp, MSG_DONTWAIT, &bytes_read); if (error || mp == NULL) { if (error == EWOULDBLOCK) - ns_flag = SLP_NEEDQ; + ns_flag = (waitflag == MBUF_DONTWAIT) ? SLP_NEEDQ : 0; else ns_flag = SLP_DISCONN; goto dorecs; @@ -2218,11 +5772,17 @@ nfsrv_rcv_locked(socket_t so, struct nfssvc_sock *slp, int waitflag) } } else { struct sockaddr_storage nam; - + + if (slp->ns_reccnt >= nfsrv_sock_max_rec_queue_length) { + /* already have max # RPC records queued on this socket */ + ns_flag = SLP_NEEDQ; + goto dorecs; + } + bzero(&msg, sizeof(msg)); msg.msg_name = (caddr_t)&nam; msg.msg_namelen = sizeof(nam); - + do { bytes_read = 1000000000; error = sock_receivembuf(so, &msg, &mp, MSG_DONTWAIT | MSG_NEEDSA, &bytes_read); @@ -2242,29 +5802,14 @@ nfsrv_rcv_locked(socket_t so, struct nfssvc_sock *slp, int waitflag) } if (slp->ns_recend) mbuf_setnextpkt(slp->ns_recend, m); - else + else { slp->ns_rec = m; + slp->ns_flag |= SLP_DOREC; + } slp->ns_recend = m; mbuf_setnextpkt(m, NULL); + slp->ns_reccnt++; } -#if 0 - if (error) { - /* - * This may be needed in the future to support - * non-byte-stream connection-oriented protocols - * such as SCTP. - */ - /* - * This (slp->ns_sotype == SOCK_STREAM) should really - * be a check for PR_CONNREQUIRED. - */ - if ((slp->ns_sotype == SOCK_STREAM) - && error != EWOULDBLOCK) { - ns_flag = SLP_DISCONN; - goto dorecs; - } - } -#endif } while (mp); } @@ -2275,9 +5820,9 @@ dorecs: if (ns_flag) slp->ns_flag |= ns_flag; if (waitflag == MBUF_DONTWAIT) { - int wake = (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN))); + int wake = (slp->ns_flag & SLP_WORKTODO); lck_rw_done(&slp->ns_rwlock); - if (wake && nfs_numnfsd) { + if (wake && nfsd_thread_count) { lck_mtx_lock(nfsd_mutex); nfsrv_wakenfsd(slp); lck_mtx_unlock(nfsd_mutex); @@ -2290,16 +5835,14 @@ dorecs: * stream socket. The "waitflag" argument indicates whether or not it * can sleep. */ -static int -nfsrv_getstream(slp, waitflag) - struct nfssvc_sock *slp; - int waitflag; +int +nfsrv_getstream(struct nfsrv_sock *slp, int waitflag) { mbuf_t m; char *cp1, *cp2, *mdata; int len, mlen, error; mbuf_t om, m2, recm; - u_long recmark; + u_int32_t recmark; if (slp->ns_flag & SLP_GETSTREAM) panic("nfs getstream"); @@ -2339,7 +5882,7 @@ nfsrv_getstream(slp, waitflag) slp->ns_flag |= SLP_LASTFRAG; else slp->ns_flag &= ~SLP_LASTFRAG; - if (slp->ns_reclen < NFS_MINPACKET || slp->ns_reclen > NFS_MAXPACKET) { + if (slp->ns_reclen <= 0 || slp->ns_reclen > NFS_MAXPACKET) { slp->ns_flag &= ~SLP_GETSTREAM; return (EPERM); } @@ -2418,7 +5961,7 @@ nfsrv_getstream(slp, waitflag) if (slp->ns_frag == NULL) { slp->ns_frag = recm; } else { - m = slp->ns_frag; + m = slp->ns_frag; while ((m2 = mbuf_next(m))) m = m2; if ((error = mbuf_setnext(m, recm))) @@ -2427,8 +5970,10 @@ nfsrv_getstream(slp, waitflag) if (slp->ns_flag & SLP_LASTFRAG) { if (slp->ns_recend) mbuf_setnextpkt(slp->ns_recend, slp->ns_frag); - else + else { slp->ns_rec = slp->ns_frag; + slp->ns_flag |= SLP_DOREC; + } slp->ns_recend = slp->ns_frag; slp->ns_frag = NULL; } @@ -2439,18 +5984,18 @@ nfsrv_getstream(slp, waitflag) * Parse an RPC header. */ int -nfsrv_dorec(slp, nfsd, ndp) - struct nfssvc_sock *slp; - struct nfsd *nfsd; - struct nfsrv_descript **ndp; +nfsrv_dorec( + struct nfsrv_sock *slp, + struct nfsd *nfsd, + struct nfsrv_descript **ndp) { mbuf_t m; mbuf_t nam; struct nfsrv_descript *nd; - int error; + int error = 0; *ndp = NULL; - if ((slp->ns_flag & SLP_VALID) == 0 || (slp->ns_rec == NULL)) + if (!(slp->ns_flag & (SLP_VALID|SLP_DOREC)) || (slp->ns_rec == NULL)) return (ENOBUFS); MALLOC_ZONE(nd, struct nfsrv_descript *, sizeof (struct nfsrv_descript), M_NFSRVDESC, M_WAITOK); @@ -2460,8 +6005,11 @@ nfsrv_dorec(slp, nfsd, ndp) slp->ns_rec = mbuf_nextpkt(m); if (slp->ns_rec) mbuf_setnextpkt(m, NULL); - else + else { + slp->ns_flag &= ~SLP_DOREC; slp->ns_recend = NULL; + } + slp->ns_reccnt--; if (mbuf_type(m) == MBUF_TYPE_SONAME) { nam = m; m = mbuf_next(m); @@ -2469,16 +6017,19 @@ nfsrv_dorec(slp, nfsd, ndp) panic("nfsrv_dorec: mbuf_setnext failed %d\n", error); } else nam = NULL; - nd->nd_md = nd->nd_mrep = m; nd->nd_nam2 = nam; - nd->nd_dpos = mbuf_data(m); - error = nfs_getreq(nd, nfsd, TRUE); + nfsm_chain_dissect_init(error, &nd->nd_nmreq, m); + if (!error) + error = nfsrv_getreq(nd); if (error) { if (nam) mbuf_freem(nam); - FREE_ZONE((caddr_t)nd, sizeof *nd, M_NFSRVDESC); + if (nd->nd_gss_context) + nfs_gss_svc_ctx_deref(nd->nd_gss_context); + FREE_ZONE(nd, sizeof(*nd), M_NFSRVDESC); return (error); } + nd->nd_mrep = NULL; *ndp = nd; nfsd->nfsd_nd = nd; return (0); @@ -2490,269 +6041,149 @@ nfsrv_dorec(slp, nfsd, ndp) * - fill in the cred struct. */ int -nfs_getreq(nd, nfsd, has_header) - struct nfsrv_descript *nd; - struct nfsd *nfsd; - int has_header; +nfsrv_getreq(struct nfsrv_descript *nd) { + struct nfsm_chain *nmreq; int len, i; - u_long *tl; - long t1; - uio_t uiop; - caddr_t dpos, cp2, cp; - u_long nfsvers, auth_type; - uid_t nickuid; - int error = 0, ticklen; - mbuf_t mrep, md; - struct nfsuid *nuidp; + u_int32_t nfsvers, auth_type; + int error = 0; uid_t user_id; gid_t group_id; int ngroups; - struct ucred temp_cred; - struct timeval tvin, tvout, now; - char uio_buf[ UIO_SIZEOF(1) ]; -#if 0 /* until encrypted keys are implemented */ - NFSKERBKEYSCHED_T keys; /* stores key schedule */ -#endif + uint32_t val; nd->nd_cr = NULL; - - mrep = nd->nd_mrep; - md = nd->nd_md; - dpos = nd->nd_dpos; - if (has_header) { - nfsm_dissect(tl, u_long *, 10 * NFSX_UNSIGNED); - nd->nd_retxid = fxdr_unsigned(u_long, *tl++); - if (*tl++ != rpc_call) { - mbuf_freem(mrep); - return (EBADRPC); - } - } else - nfsm_dissect(tl, u_long *, 8 * NFSX_UNSIGNED); + nd->nd_gss_context = NULL; + nd->nd_gss_seqnum = 0; + nd->nd_gss_mb = NULL; + + user_id = group_id = -2; + val = auth_type = len = 0; + + nmreq = &nd->nd_nmreq; + nfsm_chain_get_32(error, nmreq, nd->nd_retxid); // XID + nfsm_chain_get_32(error, nmreq, val); // RPC Call + if (!error && (val != RPC_CALL)) + error = EBADRPC; + nfsmout_if(error); nd->nd_repstat = 0; - nd->nd_flag = 0; - if (*tl++ != rpc_vers) { + nfsm_chain_get_32(error, nmreq, val); // RPC Version + nfsmout_if(error); + if (val != RPC_VER2) { nd->nd_repstat = ERPCMISMATCH; nd->nd_procnum = NFSPROC_NOOP; return (0); } - if (*tl != nfs_prog) { + nfsm_chain_get_32(error, nmreq, val); // RPC Program Number + nfsmout_if(error); + if (val != NFS_PROG) { nd->nd_repstat = EPROGUNAVAIL; nd->nd_procnum = NFSPROC_NOOP; return (0); } - tl++; - nfsvers = fxdr_unsigned(u_long, *tl++); + nfsm_chain_get_32(error, nmreq, nfsvers);// NFS Version Number + nfsmout_if(error); if ((nfsvers < NFS_VER2) || (nfsvers > NFS_VER3)) { nd->nd_repstat = EPROGMISMATCH; nd->nd_procnum = NFSPROC_NOOP; return (0); } - else if (nfsvers == NFS_VER3) - nd->nd_flag = ND_NFSV3; - nd->nd_procnum = fxdr_unsigned(u_long, *tl++); - if (nd->nd_procnum == NFSPROC_NULL) - return (0); + nd->nd_vers = nfsvers; + nfsm_chain_get_32(error, nmreq, nd->nd_procnum);// NFS Procedure Number + nfsmout_if(error); if ((nd->nd_procnum >= NFS_NPROCS) || - (!nd->nd_flag && nd->nd_procnum > NFSV2PROC_STATFS)) { + ((nd->nd_vers == NFS_VER2) && (nd->nd_procnum > NFSV2PROC_STATFS))) { nd->nd_repstat = EPROCUNAVAIL; nd->nd_procnum = NFSPROC_NOOP; return (0); } - if ((nd->nd_flag & ND_NFSV3) == 0) + if (nfsvers != NFS_VER3) nd->nd_procnum = nfsv3_procid[nd->nd_procnum]; - auth_type = *tl++; - len = fxdr_unsigned(int, *tl++); - if (len < 0 || len > RPCAUTH_MAXSIZ) { - mbuf_freem(mrep); - return (EBADRPC); - } - - nd->nd_flag &= ~ND_KERBAUTH; - /* - * Handle auth_unix or auth_kerb. - */ - if (auth_type == rpc_auth_unix) { - len = fxdr_unsigned(int, *++tl); - if (len < 0 || len > NFS_MAXNAMLEN) { - mbuf_freem(mrep); - return (EBADRPC); - } - bzero(&temp_cred, sizeof(temp_cred)); - nfsm_adv(nfsm_rndup(len)); - nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); - user_id = fxdr_unsigned(uid_t, *tl++); - group_id = fxdr_unsigned(gid_t, *tl++); - temp_cred.cr_groups[0] = group_id; - len = fxdr_unsigned(int, *tl); - if (len < 0 || len > RPCAUTH_UNIXGIDS) { - mbuf_freem(mrep); - return (EBADRPC); - } - nfsm_dissect(tl, u_long *, (len + 2) * NFSX_UNSIGNED); + nfsm_chain_get_32(error, nmreq, auth_type); // Auth Flavor + nfsm_chain_get_32(error, nmreq, len); // Auth Length + if (!error && (len < 0 || len > RPCAUTH_MAXSIZ)) + error = EBADRPC; + nfsmout_if(error); + + /* Handle authentication */ + if (auth_type == RPCAUTH_SYS) { + struct posix_cred temp_pcred; + if (nd->nd_procnum == NFSPROC_NULL) + return (0); + nd->nd_sec = RPCAUTH_SYS; + nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED); // skip stamp + nfsm_chain_get_32(error, nmreq, len); // hostname length + if (len < 0 || len > NFS_MAXNAMLEN) + error = EBADRPC; + nfsm_chain_adv(error, nmreq, nfsm_rndup(len)); // skip hostname + nfsmout_if(error); + + /* create a temporary credential using the bits from the wire */ + bzero(&temp_pcred, sizeof(temp_pcred)); + nfsm_chain_get_32(error, nmreq, user_id); + nfsm_chain_get_32(error, nmreq, group_id); + temp_pcred.cr_groups[0] = group_id; + nfsm_chain_get_32(error, nmreq, len); // extra GID count + if ((len < 0) || (len > RPCAUTH_UNIXGIDS)) + error = EBADRPC; + nfsmout_if(error); for (i = 1; i <= len; i++) - if (i < NGROUPS) - temp_cred.cr_groups[i] = fxdr_unsigned(gid_t, *tl++); - else - tl++; + if (i < NGROUPS) + nfsm_chain_get_32(error, nmreq, temp_pcred.cr_groups[i]); + else + nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED); + nfsmout_if(error); ngroups = (len >= NGROUPS) ? NGROUPS : (len + 1); if (ngroups > 1) - nfsrvw_sort(&temp_cred.cr_groups[0], ngroups); - len = fxdr_unsigned(int, *++tl); - if (len < 0 || len > RPCAUTH_MAXSIZ) { - mbuf_freem(mrep); - return (EBADRPC); - } - temp_cred.cr_uid = user_id; - temp_cred.cr_ngroups = ngroups; - nd->nd_cr = kauth_cred_create(&temp_cred); + nfsrv_group_sort(&temp_pcred.cr_groups[0], ngroups); + nfsm_chain_adv(error, nmreq, NFSX_UNSIGNED); // verifier flavor (should be AUTH_NONE) + nfsm_chain_get_32(error, nmreq, len); // verifier length + if (len < 0 || len > RPCAUTH_MAXSIZ) + error = EBADRPC; + if (len > 0) + nfsm_chain_adv(error, nmreq, nfsm_rndup(len)); + + /* request creation of a real credential */ + temp_pcred.cr_uid = user_id; + temp_pcred.cr_ngroups = ngroups; + nd->nd_cr = posix_cred_create(&temp_pcred); if (nd->nd_cr == NULL) { nd->nd_repstat = ENOMEM; nd->nd_procnum = NFSPROC_NOOP; return (0); } - if (len > 0) - nfsm_adv(nfsm_rndup(len)); - } else if (auth_type == rpc_auth_kerb) { - switch (fxdr_unsigned(int, *tl++)) { - case RPCAKN_FULLNAME: - ticklen = fxdr_unsigned(int, *tl); - *((u_long *)nfsd->nfsd_authstr) = *tl; - uiop = uio_createwithbuffer(1, 0, UIO_SYSSPACE, UIO_READ, - &uio_buf[0], sizeof(uio_buf)); - if (!uiop) { - nd->nd_repstat = ENOMEM; - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - - // LP64todo - fix this - nfsd->nfsd_authlen = (nfsm_rndup(ticklen) + (NFSX_UNSIGNED * 2)); - if ((nfsm_rndup(ticklen) + NFSX_UNSIGNED) > (len - 2 * NFSX_UNSIGNED)) { - mbuf_freem(mrep); - return (EBADRPC); - } - uio_addiov(uiop, CAST_USER_ADDR_T(&nfsd->nfsd_authstr[4]), RPCAUTH_MAXSIZ - 4); - // LP64todo - fix this - nfsm_mtouio(uiop, uio_resid(uiop)); - nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); - if (*tl++ != rpc_auth_kerb || - fxdr_unsigned(int, *tl) != 4 * NFSX_UNSIGNED) { - printf("Bad kerb verifier\n"); - nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - nfsm_dissect(cp, caddr_t, 4 * NFSX_UNSIGNED); - tl = (u_long *)cp; - if (fxdr_unsigned(int, *tl) != RPCAKN_FULLNAME) { - printf("Not fullname kerb verifier\n"); - nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - cp += NFSX_UNSIGNED; - bcopy(cp, nfsd->nfsd_verfstr, 3 * NFSX_UNSIGNED); - nfsd->nfsd_verflen = 3 * NFSX_UNSIGNED; - nd->nd_flag |= ND_KERBFULL; - nfsd->nfsd_flag |= NFSD_NEEDAUTH; - break; - case RPCAKN_NICKNAME: - if (len != 2 * NFSX_UNSIGNED) { - printf("Kerb nickname short\n"); - nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADCRED); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - nickuid = fxdr_unsigned(uid_t, *tl); - nfsm_dissect(tl, u_long *, 2 * NFSX_UNSIGNED); - if (*tl++ != rpc_auth_kerb || - fxdr_unsigned(int, *tl) != 3 * NFSX_UNSIGNED) { - printf("Kerb nick verifier bad\n"); - nd->nd_repstat = (NFSERR_AUTHERR|AUTH_BADVERF); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - nfsm_dissect(tl, u_long *, 3 * NFSX_UNSIGNED); - tvin.tv_sec = *tl++; - tvin.tv_usec = *tl; - - for (nuidp = NUIDHASH(nfsd->nfsd_slp,nickuid)->lh_first; - nuidp != 0; nuidp = nuidp->nu_hash.le_next) { - if (kauth_cred_getuid(nuidp->nu_cr) == nickuid && - (!nd->nd_nam2 || - netaddr_match(NU_NETFAM(nuidp), - &nuidp->nu_haddr, nd->nd_nam2))) - break; - } - if (!nuidp) { - nd->nd_repstat = - (NFSERR_AUTHERR|AUTH_REJECTCRED); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - - /* - * Now, decrypt the timestamp using the session key - * and validate it. - */ -#if NFSKERB - XXX -#endif - - tvout.tv_sec = fxdr_unsigned(long, tvout.tv_sec); - tvout.tv_usec = fxdr_unsigned(long, tvout.tv_usec); - microtime(&now); - if (nuidp->nu_expire < now.tv_sec || - nuidp->nu_timestamp.tv_sec > tvout.tv_sec || - (nuidp->nu_timestamp.tv_sec == tvout.tv_sec && - nuidp->nu_timestamp.tv_usec > tvout.tv_usec)) { - nuidp->nu_expire = 0; - nd->nd_repstat = - (NFSERR_AUTHERR|AUTH_REJECTVERF); - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - bzero(&temp_cred, sizeof(temp_cred)); - ngroups = nuidp->nu_cr->cr_ngroups; - for (i = 0; i < ngroups; i++) - temp_cred.cr_groups[i] = nuidp->nu_cr->cr_groups[i]; - if (ngroups > 1) - nfsrvw_sort(&temp_cred.cr_groups[0], ngroups); - - temp_cred.cr_uid = kauth_cred_getuid(nuidp->nu_cr); - temp_cred.cr_ngroups = ngroups; - nd->nd_cr = kauth_cred_create(&temp_cred); - if (!nd->nd_cr) { - nd->nd_repstat = ENOMEM; - nd->nd_procnum = NFSPROC_NOOP; - return (0); - } - nd->nd_flag |= ND_KERBNICK; - }; + } else if (auth_type == RPCSEC_GSS) { + error = nfs_gss_svc_cred_get(nd, nmreq); + if (error) { + if (error == EINVAL) + goto nfsmout; // drop the request + nd->nd_repstat = error; + nd->nd_procnum = NFSPROC_NOOP; + return (0); + } } else { + if (nd->nd_procnum == NFSPROC_NULL) // assume it's AUTH_NONE + return (0); nd->nd_repstat = (NFSERR_AUTHERR | AUTH_REJECTCRED); nd->nd_procnum = NFSPROC_NOOP; return (0); } - - nd->nd_md = md; - nd->nd_dpos = dpos; return (0); nfsmout: - if (nd->nd_cr) - kauth_cred_rele(nd->nd_cr); + if (IS_VALID_CRED(nd->nd_cr)) + kauth_cred_unref(&nd->nd_cr); + nfsm_chain_cleanup(nmreq); return (error); } /* * Search for a sleeping nfsd and wake it up. - * SIDE EFFECT: If none found, set NFSD_CHECKSLP flag, so that one of the - * running nfsds will go look for the work in the nfssvc_sock list. + * SIDE EFFECT: If none found, make sure the socket is queued up so that one + * of the running nfsds will go look for the work in the nfsrv_sockwait list. * Note: Must be called with nfsd_mutex held. */ void -nfsrv_wakenfsd(struct nfssvc_sock *slp) +nfsrv_wakenfsd(struct nfsrv_sock *slp) { struct nfsd *nd; @@ -2760,89 +6191,21 @@ nfsrv_wakenfsd(struct nfssvc_sock *slp) return; lck_rw_lock_exclusive(&slp->ns_rwlock); - - if (nfsd_waiting) { - TAILQ_FOREACH(nd, &nfsd_head, nfsd_chain) { - if (nd->nfsd_flag & NFSD_WAITING) { - nd->nfsd_flag &= ~NFSD_WAITING; - if (nd->nfsd_slp) - panic("nfsd wakeup"); - slp->ns_sref++; - nd->nfsd_slp = slp; - lck_rw_done(&slp->ns_rwlock); - wakeup((caddr_t)nd); - return; - } - } + /* if there's work to do on this socket, make sure it's queued up */ + if ((slp->ns_flag & SLP_WORKTODO) && !(slp->ns_flag & SLP_QUEUED)) { + TAILQ_INSERT_TAIL(&nfsrv_sockwait, slp, ns_svcq); + slp->ns_flag |= SLP_WAITQ; } - - slp->ns_flag |= SLP_DOREC; - lck_rw_done(&slp->ns_rwlock); - nfsd_head_flag |= NFSD_CHECKSLP; -} -#endif /* NFS_NOSERVER */ - -static int -nfs_msg(proc_t p, - const char *server, - const char *msg, - int error) -{ - tpr_t tpr; - - if (p) - tpr = tprintf_open(p); - else - tpr = NULL; - if (error) - tprintf(tpr, "nfs server %s: %s, error %d\n", server, msg, - error); - else - tprintf(tpr, "nfs server %s: %s\n", server, msg); - tprintf_close(tpr); - return (0); -} - -void -nfs_down(nmp, proc, error, flags, msg) - struct nfsmount *nmp; - proc_t proc; - int error, flags; - const char *msg; -{ - if (nmp == NULL) + /* wake up a waiting nfsd, if possible */ + nd = TAILQ_FIRST(&nfsd_queue); + if (!nd) return; - if ((flags & NFSSTA_TIMEO) && !(nmp->nm_state & NFSSTA_TIMEO)) { - vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESP, 0); - nmp->nm_state |= NFSSTA_TIMEO; - } - if ((flags & NFSSTA_LOCKTIMEO) && !(nmp->nm_state & NFSSTA_LOCKTIMEO)) { - vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESPLOCK, 0); - nmp->nm_state |= NFSSTA_LOCKTIMEO; - } - nfs_msg(proc, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, error); -} -void -nfs_up(nmp, proc, flags, msg) - struct nfsmount *nmp; - proc_t proc; - int flags; - const char *msg; -{ - if (nmp == NULL) - return; - if (msg) - nfs_msg(proc, vfs_statfs(nmp->nm_mountp)->f_mntfromname, msg, 0); - if ((flags & NFSSTA_TIMEO) && (nmp->nm_state & NFSSTA_TIMEO)) { - nmp->nm_state &= ~NFSSTA_TIMEO; - vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESP, 1); - } - if ((flags & NFSSTA_LOCKTIMEO) && (nmp->nm_state & NFSSTA_LOCKTIMEO)) { - nmp->nm_state &= ~NFSSTA_LOCKTIMEO; - vfs_event_signal(&vfs_statfs(nmp->nm_mountp)->f_fsid, VQ_NOTRESPLOCK, 1); - } + TAILQ_REMOVE(&nfsd_queue, nd, nfsd_queue); + nd->nfsd_flag &= ~NFSD_WAITING; + wakeup(nd); } +#endif /* NFSSERVER */