nlm_prot_impl.c revision 214048
162587Sitojun/*- 295023Ssuz * Copyright (c) 2008 Isilon Inc http://www.isilon.com/ 362587Sitojun * Authors: Doug Rabson <dfr@rabson.org> 4139826Simp * Developed with Red Inc: Alfred Perlstein <alfred@freebsd.org> 553541Sshin * 653541Sshin * Redistribution and use in source and binary forms, with or without 753541Sshin * modification, are permitted provided that the following conditions 853541Sshin * are met: 953541Sshin * 1. Redistributions of source code must retain the above copyright 1053541Sshin * notice, this list of conditions and the following disclaimer. 1153541Sshin * 2. Redistributions in binary form must reproduce the above copyright 1253541Sshin * notice, this list of conditions and the following disclaimer in the 1353541Sshin * documentation and/or other materials provided with the distribution. 1453541Sshin * 1553541Sshin * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 1653541Sshin * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 1753541Sshin * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 1853541Sshin * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 1953541Sshin * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 2053541Sshin * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 2153541Sshin * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 2253541Sshin * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 2353541Sshin * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 2453541Sshin * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 2553541Sshin * SUCH DAMAGE. 2653541Sshin */ 2753541Sshin 2853541Sshin#include "opt_inet6.h" 2953541Sshin 3053541Sshin#include <sys/cdefs.h> 3153541Sshin__FBSDID("$FreeBSD: head/sys/nlm/nlm_prot_impl.c 214048 2010-10-19 00:20:00Z rmacklem $"); 3253541Sshin 3353541Sshin#include <sys/param.h> 3453541Sshin#include <sys/fail.h> 3553541Sshin#include <sys/fcntl.h> 3653541Sshin#include <sys/kernel.h> 3753541Sshin#include <sys/kthread.h> 3853541Sshin#include <sys/lockf.h> 3953541Sshin#include <sys/malloc.h> 4053541Sshin#include <sys/mount.h> 4153541Sshin#if __FreeBSD_version >= 700000 4253541Sshin#include <sys/priv.h> 4353541Sshin#endif 4453541Sshin#include <sys/proc.h> 4553541Sshin#include <sys/socket.h> 4653541Sshin#include <sys/socketvar.h> 4753541Sshin#include <sys/syscall.h> 4853541Sshin#include <sys/sysctl.h> 4953541Sshin#include <sys/sysent.h> 5062587Sitojun#include <sys/syslog.h> 5153541Sshin#include <sys/sysproto.h> 5262587Sitojun#include <sys/systm.h> 53121684Sume#include <sys/taskqueue.h> 54121684Sume#include <sys/unistd.h> 5553541Sshin#include <sys/vnode.h> 5653541Sshin 5753541Sshin#include <nfs/nfsproto.h> 5853541Sshin#include <nfs/nfs_lock.h> 5953541Sshin 6053541Sshin#include <nlm/nlm_prot.h> 6153541Sshin#include <nlm/sm_inter.h> 6253541Sshin#include <nlm/nlm.h> 6362587Sitojun#include <rpc/rpc_com.h> 6453541Sshin#include <rpc/rpcb_prot.h> 6562587Sitojun 6662587SitojunMALLOC_DEFINE(M_NLM, "NLM", "Network Lock Manager"); 6762587Sitojun 6862587Sitojun/* 6962587Sitojun * If a host is inactive (and holds no locks) for this amount of 7053541Sshin * seconds, we consider it idle and stop tracking it. 71121346Sume */ 72121346Sume#define NLM_IDLE_TIMEOUT 30 73121346Sume 74121346Sume/* 75121346Sume * We check the host list for idle every few seconds. 76121346Sume */ 77121346Sume#define NLM_IDLE_PERIOD 5 7853541Sshin 79121346Sume/* 80121346Sume * We only look for GRANTED_RES messages for a little while. 81121346Sume */ 82121355Sume#define NLM_EXPIRE_TIMEOUT 10 83121346Sume 84121345Sume/* 8569774Sphk * Support for sysctl vfs.nlm.sysid 8662587Sitojun */ 8753541SshinSYSCTL_NODE(_vfs, OID_AUTO, nlm, CTLFLAG_RW, NULL, "Network Lock Manager"); 8853541SshinSYSCTL_NODE(_vfs_nlm, OID_AUTO, sysid, CTLFLAG_RW, NULL, ""); 8953541Sshin 90157927Sps/* 91157927Sps * Syscall hooks 92157927Sps */ 93157927Spsstatic int nlm_syscall_offset = SYS_nlm_syscall; 94157927Spsstatic struct sysent nlm_syscall_prev_sysent; 95157927Sps#if __FreeBSD_version < 700000 96157927Spsstatic struct sysent nlm_syscall_sysent = { 97157927Sps (sizeof(struct nlm_syscall_args) / sizeof(register_t)) | SYF_MPSAFE, 9853541Sshin (sy_call_t *) nlm_syscall 9953541Sshin}; 10053541Sshin#else 10153541SshinMAKE_SYSENT(nlm_syscall); 10277969Sjesper#endif 103121345Sumestatic bool_t nlm_syscall_registered = FALSE; 104157927Sps 105157927Sps/* 10677969Sjesper * Debug level passed in from userland. We also support a sysctl hook 107121346Sume * so that it can be changed on a live system. 108121346Sume */ 10953541Sshinstatic int nlm_debug_level; 11053541SshinSYSCTL_INT(_debug, OID_AUTO, nlm_debug, CTLFLAG_RW, &nlm_debug_level, 0, ""); 11153541Sshin 11253541Sshin#define NLM_DEBUG(_level, args...) \ 11362587Sitojun do { \ 11462587Sitojun if (nlm_debug_level >= (_level)) \ 11562587Sitojun log(LOG_DEBUG, args); \ 11662587Sitojun } while(0) 11762587Sitojun#define NLM_ERR(args...) \ 11862587Sitojun do { \ 11962587Sitojun log(LOG_ERR, args); \ 12062587Sitojun } while(0) 12162587Sitojun 12262587Sitojun/* 12362587Sitojun * Grace period handling. The value of nlm_grace_threshold is the 12462587Sitojun * value of time_uptime after which we are serving requests normally. 12562587Sitojun */ 12662587Sitojunstatic time_t nlm_grace_threshold; 12762587Sitojun 12862587Sitojun/* 12962587Sitojun * We check for idle hosts if time_uptime is greater than 13062587Sitojun * nlm_next_idle_check, 13162587Sitojun */ 13262587Sitojunstatic time_t nlm_next_idle_check; 13362587Sitojun 13462587Sitojun/* 13562587Sitojun * A socket to use for RPC - shared by all IPv4 RPC clients. 13662587Sitojun */ 13762587Sitojunstatic struct socket *nlm_socket; 13862587Sitojun 13962587Sitojun#ifdef INET6 14062587Sitojun 14162587Sitojun/* 14253541Sshin * A socket to use for RPC - shared by all IPv6 RPC clients. 14353541Sshin */ 14453541Sshinstatic struct socket *nlm_socket6; 14553541Sshin 14653541Sshin#endif 14753541Sshin 14853541Sshin/* 14953541Sshin * An RPC client handle that can be used to communicate with the local 15053541Sshin * NSM. 15153541Sshin */ 15253541Sshinstatic CLIENT *nlm_nsm; 15362587Sitojun 154121630Sume/* 155121630Sume * An AUTH handle for the server's creds. 156121630Sume */ 15753541Sshinstatic AUTH *nlm_auth; 15853541Sshin 15962587Sitojun/* 16053541Sshin * A zero timeval for sending async RPC messages. 161121684Sume */ 16253541Sshinstruct timeval nlm_zero_tv = { 0, 0 }; 16362587Sitojun 16462587Sitojun/* 16553541Sshin * The local NSM state number 16653541Sshin */ 16762587Sitojunint nlm_nsm_state; 16862587Sitojun 16962587Sitojun 170120856Sume/* 17162587Sitojun * A lock to protect the host list and waiting lock list. 17253541Sshin */ 17353541Sshinstatic struct mtx nlm_global_lock; 17453541Sshin 17553541Sshin/* 176121630Sume * Locks: 177121630Sume * (l) locked by nh_lock 17853541Sshin * (s) only accessed via server RPC which is single threaded 17953541Sshin * (g) locked by nlm_global_lock 18053541Sshin * (c) const until freeing 18153541Sshin * (a) modified using atomic ops 18253541Sshin */ 18353541Sshin 18453541Sshin/* 18553541Sshin * A pending client-side lock request, stored on the nlm_waiting_locks 18653541Sshin * list. 18753541Sshin */ 18853541Sshinstruct nlm_waiting_lock { 18953541Sshin TAILQ_ENTRY(nlm_waiting_lock) nw_link; /* (g) */ 19053541Sshin bool_t nw_waiting; /* (g) */ 19153541Sshin nlm4_lock nw_lock; /* (c) */ 19253541Sshin union nfsfh nw_fh; /* (c) */ 19353541Sshin struct vnode *nw_vp; /* (c) */ 19453541Sshin}; 19553541SshinTAILQ_HEAD(nlm_waiting_lock_list, nlm_waiting_lock); 19653541Sshin 19753541Sshinstruct nlm_waiting_lock_list nlm_waiting_locks; /* (g) */ 19853541Sshin 199120891Sume/* 200120891Sume * A pending server-side asynchronous lock request, stored on the 20153541Sshin * nh_pending list of the NLM host. 20253541Sshin */ 20353541Sshinstruct nlm_async_lock { 20453541Sshin TAILQ_ENTRY(nlm_async_lock) af_link; /* (l) host's list of locks */ 20553541Sshin struct task af_task; /* (c) async callback details */ 20653541Sshin void *af_cookie; /* (l) lock manager cancel token */ 207120891Sume struct vnode *af_vp; /* (l) vnode to lock */ 20862587Sitojun struct flock af_fl; /* (c) lock details */ 20953541Sshin struct nlm_host *af_host; /* (c) host which is locking */ 21053541Sshin CLIENT *af_rpc; /* (c) rpc client to send message */ 211121345Sume nlm4_testargs af_granted; /* (c) notification details */ 21278064Sume time_t af_expiretime; /* (c) notification time */ 213121345Sume}; 214121345SumeTAILQ_HEAD(nlm_async_lock_list, nlm_async_lock); 215121345Sume 216121345Sume/* 217121345Sume * NLM host. 218121345Sume */ 219121345Sumeenum nlm_host_state { 220121345Sume NLM_UNMONITORED, 221121345Sume NLM_MONITORED, 222121345Sume NLM_MONITOR_FAILED, 22353541Sshin NLM_RECOVERING 22453541Sshin}; 22553541Sshin 22653541Sshinstruct nlm_rpc { 22753541Sshin CLIENT *nr_client; /* (l) RPC client handle */ 22853541Sshin time_t nr_create_time; /* (l) when client was created */ 22953541Sshin}; 23053541Sshin 23153541Sshinstruct nlm_host { 23253541Sshin struct mtx nh_lock; 23353541Sshin volatile u_int nh_refs; /* (a) reference count */ 23453541Sshin TAILQ_ENTRY(nlm_host) nh_link; /* (g) global list of hosts */ 23553541Sshin char nh_caller_name[MAXNAMELEN]; /* (c) printable name of host */ 23653541Sshin uint32_t nh_sysid; /* (c) our allocaed system ID */ 23753541Sshin char nh_sysid_string[10]; /* (c) string rep. of sysid */ 238121345Sume struct sockaddr_storage nh_addr; /* (s) remote address of host */ 239121345Sume struct nlm_rpc nh_srvrpc; /* (l) RPC for server replies */ 240121345Sume struct nlm_rpc nh_clntrpc; /* (l) RPC for client requests */ 24153541Sshin rpcvers_t nh_vers; /* (s) NLM version of host */ 24278064Sume int nh_state; /* (s) last seen NSM state of host */ 24378064Sume enum nlm_host_state nh_monstate; /* (l) local NSM monitoring state */ 24478064Sume time_t nh_idle_timeout; /* (s) Time at which host is idle */ 24578064Sume struct sysctl_ctx_list nh_sysctl; /* (c) vfs.nlm.sysid nodes */ 24678064Sume uint32_t nh_grantcookie; /* (l) grant cookie counter */ 24753541Sshin struct nlm_async_lock_list nh_pending; /* (l) pending async locks */ 248121607Sume struct nlm_async_lock_list nh_granted; /* (l) granted locks */ 24953541Sshin struct nlm_async_lock_list nh_finished; /* (l) finished async locks */ 25053541Sshin}; 25162587SitojunTAILQ_HEAD(nlm_host_list, nlm_host); 25253541Sshin 25353541Sshinstatic struct nlm_host_list nlm_hosts; /* (g) */ 25453541Sshinstatic uint32_t nlm_next_sysid = 1; /* (g) */ 25562587Sitojun 25653541Sshinstatic void nlm_host_unmonitor(struct nlm_host *); 25762587Sitojun 25862587Sitojunstruct nlm_grantcookie { 25962587Sitojun uint32_t ng_sysid; 26053541Sshin uint32_t ng_cookie; 26153541Sshin}; 26253541Sshin 26353541Sshinstatic inline uint32_t 26453541Sshinng_sysid(struct netobj *src) 26553541Sshin{ 266121345Sume 267121345Sume return ((struct nlm_grantcookie *)src->n_bytes)->ng_sysid; 26853541Sshin} 26953541Sshin 27053541Sshinstatic inline uint32_t 27153541Sshinng_cookie(struct netobj *src) 27253541Sshin{ 27353541Sshin 27453541Sshin return ((struct nlm_grantcookie *)src->n_bytes)->ng_cookie; 27553541Sshin} 276120891Sume 277120891Sume/**********************************************************************/ 27853541Sshin 27953541Sshin/* 28053541Sshin * Initialise NLM globals. 28153541Sshin */ 28253541Sshinstatic void 28353541Sshinnlm_init(void *dummy) 28453541Sshin{ 28553541Sshin int error; 28662587Sitojun 28753541Sshin mtx_init(&nlm_global_lock, "nlm_global_lock", NULL, MTX_DEF); 28853541Sshin TAILQ_INIT(&nlm_waiting_locks); 28953541Sshin TAILQ_INIT(&nlm_hosts); 29053541Sshin 291120891Sume error = syscall_register(&nlm_syscall_offset, &nlm_syscall_sysent, 292120891Sume &nlm_syscall_prev_sysent); 293121345Sume if (error) 294120856Sume NLM_ERR("Can't register NLM syscall\n"); 29553541Sshin else 296120891Sume nlm_syscall_registered = TRUE; 29753541Sshin} 298120891SumeSYSINIT(nlm_init, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_init, NULL); 299120891Sume 300121345Sumestatic void 301120856Sumenlm_uninit(void *dummy) 30253541Sshin{ 30353541Sshin 30453541Sshin if (nlm_syscall_registered) 30553541Sshin syscall_deregister(&nlm_syscall_offset, 30653541Sshin &nlm_syscall_prev_sysent); 30753541Sshin} 30853541SshinSYSUNINIT(nlm_uninit, SI_SUB_LOCK, SI_ORDER_FIRST, nlm_uninit, NULL); 30953541Sshin 31053541Sshin/* 31153541Sshin * Create a netobj from an arbitrary source. 31253541Sshin */ 31353541Sshinvoid 31453541Sshinnlm_make_netobj(struct netobj *dst, caddr_t src, size_t srcsize, 31553541Sshin struct malloc_type *type) 31653541Sshin{ 31753541Sshin 31853541Sshin dst->n_len = srcsize; 31953541Sshin dst->n_bytes = malloc(srcsize, type, M_WAITOK); 32062587Sitojun memcpy(dst->n_bytes, src, srcsize); 32153541Sshin} 32253541Sshin 32353541Sshin/* 32453541Sshin * Copy a struct netobj. 32553541Sshin */ 32653541Sshinvoid 32753541Sshinnlm_copy_netobj(struct netobj *dst, struct netobj *src, 32853541Sshin struct malloc_type *type) 32953541Sshin{ 33053541Sshin 33153541Sshin nlm_make_netobj(dst, src->n_bytes, src->n_len, type); 33253541Sshin} 333120891Sume 334120891Sume 335120891Sume/* 33653541Sshin * Create an RPC client handle for the given (address,prog,vers) 33753541Sshin * triple using UDP. 33853541Sshin */ 33953541Sshinstatic CLIENT * 34062587Sitojunnlm_get_rpc(struct sockaddr *sa, rpcprog_t prog, rpcvers_t vers) 341121607Sume{ 34262587Sitojun char *wchan = "nlmrcv"; 34362587Sitojun const char* protofmly; 34462587Sitojun struct sockaddr_storage ss; 34562587Sitojun struct socket *so; 34662587Sitojun CLIENT *rpcb; 34762587Sitojun struct timeval timo; 34862587Sitojun RPCB parms; 34953541Sshin char *uaddr; 35053541Sshin enum clnt_stat stat = RPC_SUCCESS; 35153541Sshin int rpcvers = RPCBVERS4; 35253541Sshin bool_t do_tcp = FALSE; 35353541Sshin bool_t tryagain = FALSE; 35453541Sshin struct portmap mapping; 35553541Sshin u_short port = 0; 35653541Sshin 35753541Sshin /* 35853541Sshin * First we need to contact the remote RPCBIND service to find 35953541Sshin * the right port. 36053541Sshin */ 361121684Sume memcpy(&ss, sa, sa->sa_len); 362121684Sume switch (ss.ss_family) { 363121684Sume case AF_INET: 364121684Sume ((struct sockaddr_in *)&ss)->sin_port = htons(111); 365121684Sume protofmly = "inet"; 366121684Sume so = nlm_socket; 367121684Sume break; 368121684Sume 369121684Sume#ifdef INET6 370121684Sume case AF_INET6: 371121684Sume ((struct sockaddr_in6 *)&ss)->sin6_port = htons(111); 372121684Sume protofmly = "inet6"; 373121684Sume so = nlm_socket6; 374121684Sume break; 375121684Sume#endif 376121684Sume 377121684Sume default: 378121684Sume /* 379121684Sume * Unsupported address family - fail. 380121684Sume */ 38153541Sshin return (NULL); 38253541Sshin } 38353541Sshin 38453541Sshin rpcb = clnt_dg_create(so, (struct sockaddr *)&ss, 38553541Sshin RPCBPROG, rpcvers, 0, 0); 38653541Sshin if (!rpcb) 38753541Sshin return (NULL); 38862587Sitojun 38953541Sshintry_tcp: 39062587Sitojun parms.r_prog = prog; 39162587Sitojun parms.r_vers = vers; 39262587Sitojun if (do_tcp) 39362587Sitojun parms.r_netid = "tcp"; 39462587Sitojun else 39562587Sitojun parms.r_netid = "udp"; 39662587Sitojun parms.r_addr = ""; 39762587Sitojun parms.r_owner = ""; 39862587Sitojun 39962587Sitojun /* 40062587Sitojun * Use the default timeout. 40162587Sitojun */ 40262587Sitojun timo.tv_sec = 25; 40362587Sitojun timo.tv_usec = 0; 40462587Sitojunagain: 40562587Sitojun switch (rpcvers) { 40662587Sitojun case RPCBVERS4: 40762587Sitojun case RPCBVERS: 40862587Sitojun /* 40962587Sitojun * Try RPCBIND 4 then 3. 41062587Sitojun */ 41162587Sitojun uaddr = NULL; 41262587Sitojun stat = CLNT_CALL(rpcb, (rpcprog_t) RPCBPROC_GETADDR, 41362587Sitojun (xdrproc_t) xdr_rpcb, &parms, 41462587Sitojun (xdrproc_t) xdr_wrapstring, &uaddr, timo); 41562587Sitojun if (stat == RPC_SUCCESS) { 41662587Sitojun /* 41762587Sitojun * We have a reply from the remote RPCBIND - turn it 41862587Sitojun * into an appropriate address and make a new client 41962587Sitojun * that can talk to the remote NLM. 42062587Sitojun * 42162587Sitojun * XXX fixup IPv6 scope ID. 42262587Sitojun */ 42362587Sitojun struct netbuf *a; 42462587Sitojun a = __rpc_uaddr2taddr_af(ss.ss_family, uaddr); 42553541Sshin if (!a) { 42653541Sshin tryagain = TRUE; 42753541Sshin } else { 428121345Sume tryagain = FALSE; 429121345Sume memcpy(&ss, a->buf, a->len); 43053541Sshin free(a->buf, M_RPC); 43153541Sshin free(a, M_RPC); 43253541Sshin xdr_free((xdrproc_t) xdr_wrapstring, &uaddr); 43353541Sshin } 43453541Sshin } 43576899Ssumikawa if (tryagain || stat == RPC_PROGVERSMISMATCH) { 43653541Sshin if (rpcvers == RPCBVERS4) 43753541Sshin rpcvers = RPCBVERS; 43853541Sshin else if (rpcvers == RPCBVERS) 43976899Ssumikawa rpcvers = PMAPVERS; 44076899Ssumikawa CLNT_CONTROL(rpcb, CLSET_VERS, &rpcvers); 44153541Sshin goto again; 44253541Sshin } 44353541Sshin break; 44453541Sshin case PMAPVERS: 44553541Sshin /* 44653541Sshin * Try portmap. 44776899Ssumikawa */ 44853541Sshin mapping.pm_prog = parms.r_prog; 44953541Sshin mapping.pm_vers = parms.r_vers; 45053541Sshin mapping.pm_prot = do_tcp ? IPPROTO_TCP : IPPROTO_UDP; 45176899Ssumikawa mapping.pm_port = 0; 45276899Ssumikawa 45353541Sshin stat = CLNT_CALL(rpcb, (rpcprog_t) PMAPPROC_GETPORT, 45453541Sshin (xdrproc_t) xdr_portmap, &mapping, 45553541Sshin (xdrproc_t) xdr_u_short, &port, timo); 45662587Sitojun 45753541Sshin if (stat == RPC_SUCCESS) { 45853541Sshin switch (ss.ss_family) { 45953541Sshin case AF_INET: 46053541Sshin ((struct sockaddr_in *)&ss)->sin_port = 46153541Sshin htons(port); 46253541Sshin break; 46353541Sshin 46453541Sshin#ifdef INET6 46553541Sshin case AF_INET6: 46653541Sshin ((struct sockaddr_in6 *)&ss)->sin6_port = 467121345Sume htons(port); 468121345Sume break; 46962587Sitojun#endif 47062587Sitojun } 47162587Sitojun } 47262587Sitojun break; 47362587Sitojun default: 47462587Sitojun panic("invalid rpcvers %d", rpcvers); 47553541Sshin } 47653541Sshin /* 47753541Sshin * We may have a positive response from the portmapper, but the NLM 47853541Sshin * service was not found. Make sure we received a valid port. 479121345Sume */ 48053541Sshin switch (ss.ss_family) { 48153541Sshin case AF_INET: 48253541Sshin port = ((struct sockaddr_in *)&ss)->sin_port; 48353541Sshin break; 48453541Sshin#ifdef INET6 485121345Sume case AF_INET6: 48653541Sshin port = ((struct sockaddr_in6 *)&ss)->sin6_port; 48753541Sshin break; 48853541Sshin#endif 48953541Sshin } 49053541Sshin if (stat != RPC_SUCCESS || !port) { 49153541Sshin /* 49253541Sshin * If we were able to talk to rpcbind or portmap, but the udp 49353541Sshin * variant wasn't available, ask about tcp. 49453541Sshin * 49562587Sitojun * XXX - We could also check for a TCP portmapper, but 49653541Sshin * if the host is running a portmapper at all, we should be able 49762587Sitojun * to hail it over UDP. 49862587Sitojun */ 49953541Sshin if (stat == RPC_SUCCESS && !do_tcp) { 50053541Sshin do_tcp = TRUE; 50153541Sshin goto try_tcp; 50262587Sitojun } 50362587Sitojun 50462587Sitojun /* Otherwise, bad news. */ 50553541Sshin NLM_ERR("NLM: failed to contact remote rpcbind, " 50653541Sshin "stat = %d, port = %d\n", (int) stat, port); 50753541Sshin CLNT_DESTROY(rpcb); 50853541Sshin return (NULL); 50962587Sitojun } 51062587Sitojun 51153541Sshin if (do_tcp) { 51253541Sshin /* 51353541Sshin * Destroy the UDP client we used to speak to rpcbind and 51453541Sshin * recreate as a TCP client. 51562587Sitojun */ 51662587Sitojun struct netconfig *nconf = NULL; 51762587Sitojun 51853541Sshin CLNT_DESTROY(rpcb); 51953541Sshin 52053541Sshin switch (ss.ss_family) { 52153541Sshin case AF_INET: 52262587Sitojun nconf = getnetconfigent("tcp"); 52353541Sshin break; 52453541Sshin#ifdef INET6 52562587Sitojun case AF_INET6: 52662587Sitojun nconf = getnetconfigent("tcp6"); 52762587Sitojun break; 52862587Sitojun#endif 529111119Simp } 53062587Sitojun 531121345Sume rpcb = clnt_reconnect_create(nconf, (struct sockaddr *)&ss, 53262587Sitojun prog, vers, 0, 0); 53362587Sitojun CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 53462587Sitojun rpcb->cl_auth = nlm_auth; 53562587Sitojun 53662587Sitojun } else { 53762587Sitojun /* 53853541Sshin * Re-use the client we used to speak to rpcbind. 53953541Sshin */ 54053541Sshin CLNT_CONTROL(rpcb, CLSET_SVC_ADDR, &ss); 54153541Sshin CLNT_CONTROL(rpcb, CLSET_PROG, &prog); 54253541Sshin CLNT_CONTROL(rpcb, CLSET_VERS, &vers); 54353541Sshin CLNT_CONTROL(rpcb, CLSET_WAITCHAN, wchan); 54453541Sshin rpcb->cl_auth = nlm_auth; 54553541Sshin } 54653541Sshin 54753541Sshin return (rpcb); 54853541Sshin} 549121345Sume 55053541Sshin/* 55153541Sshin * This async callback after when an async lock request has been 55253541Sshin * granted. We notify the host which initiated the request. 55353541Sshin */ 55453541Sshinstatic void 55553541Sshinnlm_lock_callback(void *arg, int pending) 55653541Sshin{ 55753541Sshin struct nlm_async_lock *af = (struct nlm_async_lock *) arg; 55853541Sshin struct rpc_callextra ext; 559120891Sume 56053541Sshin NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) granted," 56153541Sshin " cookie %d:%d\n", af, af->af_host->nh_caller_name, 56253541Sshin af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie), 56353541Sshin ng_cookie(&af->af_granted.cookie)); 56453541Sshin 56553541Sshin /* 56653541Sshin * Send the results back to the host. 56753541Sshin * 56853541Sshin * Note: there is a possible race here with nlm_host_notify 56953541Sshin * destroying the RPC client. To avoid problems, the first 570121345Sume * thing nlm_host_notify does is to cancel pending async lock 57153541Sshin * requests. 57253541Sshin */ 57353541Sshin memset(&ext, 0, sizeof(ext)); 574121346Sume ext.rc_auth = nlm_auth; 57553541Sshin if (af->af_host->nh_vers == NLM_VERS4) { 57653541Sshin nlm4_granted_msg_4(&af->af_granted, 57753541Sshin NULL, af->af_rpc, &ext, nlm_zero_tv); 57853541Sshin } else { 57953541Sshin /* 58053541Sshin * Back-convert to legacy protocol 58153541Sshin */ 58253541Sshin nlm_testargs granted; 58353541Sshin granted.cookie = af->af_granted.cookie; 58453541Sshin granted.exclusive = af->af_granted.exclusive; 58553541Sshin granted.alock.caller_name = 58653541Sshin af->af_granted.alock.caller_name; 58753541Sshin granted.alock.fh = af->af_granted.alock.fh; 58853541Sshin granted.alock.oh = af->af_granted.alock.oh; 58953541Sshin granted.alock.svid = af->af_granted.alock.svid; 59053541Sshin granted.alock.l_offset = 591121355Sume af->af_granted.alock.l_offset; 592121345Sume granted.alock.l_len = 59353541Sshin af->af_granted.alock.l_len; 59453541Sshin 59553541Sshin nlm_granted_msg_1(&granted, 59653541Sshin NULL, af->af_rpc, &ext, nlm_zero_tv); 59753541Sshin } 59853541Sshin 59953541Sshin /* 60053541Sshin * Move this entry to the nh_granted list. 60153541Sshin */ 60253541Sshin af->af_expiretime = time_uptime + NLM_EXPIRE_TIMEOUT; 60353541Sshin mtx_lock(&af->af_host->nh_lock); 60453541Sshin TAILQ_REMOVE(&af->af_host->nh_pending, af, af_link); 60553541Sshin TAILQ_INSERT_TAIL(&af->af_host->nh_granted, af, af_link); 60653541Sshin mtx_unlock(&af->af_host->nh_lock); 60753541Sshin} 60853541Sshin 60953541Sshin/* 610120891Sume * Free an async lock request. The request must have been removed from 61153541Sshin * any list. 61253541Sshin */ 61353541Sshinstatic void 61453541Sshinnlm_free_async_lock(struct nlm_async_lock *af) 61553541Sshin{ 61662587Sitojun /* 61753541Sshin * Free an async lock. 61862587Sitojun */ 61953541Sshin if (af->af_rpc) 62053541Sshin CLNT_RELEASE(af->af_rpc); 621121345Sume xdr_free((xdrproc_t) xdr_nlm4_testargs, &af->af_granted); 62253541Sshin if (af->af_vp) 62353541Sshin vrele(af->af_vp); 62453541Sshin free(af, M_NLM); 62553541Sshin} 62653541Sshin 62753541Sshin/* 62853541Sshin * Cancel our async request - this must be called with 62953541Sshin * af->nh_host->nh_lock held. This is slightly complicated by a 63053541Sshin * potential race with our own callback. If we fail to cancel the 63153541Sshin * lock, it must already have been granted - we make sure our async 63253541Sshin * task has completed by calling taskqueue_drain in this case. 63353541Sshin */ 634121345Sumestatic int 635121355Sumenlm_cancel_async_lock(struct nlm_async_lock *af) 636121345Sume{ 63753541Sshin struct nlm_host *host = af->af_host; 63853541Sshin int error; 63953541Sshin 64053541Sshin mtx_assert(&host->nh_lock, MA_OWNED); 64153541Sshin 64253541Sshin mtx_unlock(&host->nh_lock); 64353541Sshin 64453541Sshin error = VOP_ADVLOCKASYNC(af->af_vp, NULL, F_CANCEL, &af->af_fl, 64553541Sshin F_REMOTE, NULL, &af->af_cookie); 64653541Sshin 64753541Sshin if (error) { 64853541Sshin /* 64953541Sshin * We failed to cancel - make sure our callback has 650121345Sume * completed before we continue. 651121355Sume */ 652121345Sume taskqueue_drain(taskqueue_thread, &af->af_task); 65353541Sshin } 65453541Sshin 65553541Sshin mtx_lock(&host->nh_lock); 65653541Sshin 65753541Sshin if (!error) { 65853541Sshin NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) " 65953541Sshin "cancelled\n", af, host->nh_caller_name, host->nh_sysid); 66053541Sshin 661121345Sume /* 662121355Sume * Remove from the nh_pending list and free now that 663121345Sume * we are safe from the callback. 66453541Sshin */ 66553541Sshin TAILQ_REMOVE(&host->nh_pending, af, af_link); 66653541Sshin mtx_unlock(&host->nh_lock); 66753541Sshin nlm_free_async_lock(af); 66853541Sshin mtx_lock(&host->nh_lock); 66953541Sshin } 67053541Sshin 67153541Sshin return (error); 67253541Sshin} 67353541Sshin 674121345Sumestatic void 675121355Sumenlm_check_expired_locks(struct nlm_host *host) 676121345Sume{ 67753541Sshin struct nlm_async_lock *af; 67853541Sshin time_t uptime = time_uptime; 67953541Sshin 68053541Sshin mtx_lock(&host->nh_lock); 68153541Sshin while ((af = TAILQ_FIRST(&host->nh_granted)) != NULL 68278064Sume && uptime >= af->af_expiretime) { 68353541Sshin NLM_DEBUG(2, "NLM: async lock %p for %s (sysid %d) expired," 68453541Sshin " cookie %d:%d\n", af, af->af_host->nh_caller_name, 68553541Sshin af->af_host->nh_sysid, ng_sysid(&af->af_granted.cookie), 68653541Sshin ng_cookie(&af->af_granted.cookie)); 68753541Sshin TAILQ_REMOVE(&host->nh_granted, af, af_link); 68853541Sshin mtx_unlock(&host->nh_lock); 68953541Sshin nlm_free_async_lock(af); 69053541Sshin mtx_lock(&host->nh_lock); 691121345Sume } 69253541Sshin while ((af = TAILQ_FIRST(&host->nh_finished)) != NULL) { 69353541Sshin TAILQ_REMOVE(&host->nh_finished, af, af_link); 69453541Sshin mtx_unlock(&host->nh_lock); 69553541Sshin nlm_free_async_lock(af); 69653541Sshin mtx_lock(&host->nh_lock); 69753541Sshin } 69853541Sshin mtx_unlock(&host->nh_lock); 69953541Sshin} 70053541Sshin 70153541Sshin/* 70253541Sshin * Free resources used by a host. This is called after the reference 70353541Sshin * count has reached zero so it doesn't need to worry about locks. 70453541Sshin */ 70553541Sshinstatic void 70653541Sshinnlm_host_destroy(struct nlm_host *host) 70753541Sshin{ 70878064Sume 70978064Sume mtx_lock(&nlm_global_lock); 71053541Sshin TAILQ_REMOVE(&nlm_hosts, host, nh_link); 71153541Sshin mtx_unlock(&nlm_global_lock); 71253541Sshin 71353541Sshin if (host->nh_srvrpc.nr_client) 714121345Sume CLNT_RELEASE(host->nh_srvrpc.nr_client); 71562587Sitojun if (host->nh_clntrpc.nr_client) 71662587Sitojun CLNT_RELEASE(host->nh_clntrpc.nr_client); 71762587Sitojun mtx_destroy(&host->nh_lock); 71862587Sitojun sysctl_ctx_free(&host->nh_sysctl); 71962587Sitojun free(host, M_NLM); 72062587Sitojun} 72162587Sitojun 72262587Sitojun/* 72362587Sitojun * Thread start callback for client lock recovery 72462587Sitojun */ 72562587Sitojunstatic void 72662587Sitojunnlm_client_recovery_start(void *arg) 72762587Sitojun{ 72862587Sitojun struct nlm_host *host = (struct nlm_host *) arg; 72962587Sitojun 73062587Sitojun NLM_DEBUG(1, "NLM: client lock recovery for %s started\n", 73153541Sshin host->nh_caller_name); 73253541Sshin 73353541Sshin nlm_client_recovery(host); 73453541Sshin 73553541Sshin NLM_DEBUG(1, "NLM: client lock recovery for %s completed\n", 73653541Sshin host->nh_caller_name); 73753541Sshin 73853541Sshin host->nh_monstate = NLM_MONITORED; 739121345Sume nlm_host_release(host); 740121346Sume 74153541Sshin kthread_exit(); 74253541Sshin} 74353541Sshin 74453541Sshin/* 74553541Sshin * This is called when we receive a host state change notification. We 74653541Sshin * unlock any active locks owned by the host. When rpc.lockd is 747121345Sume * shutting down, this function is called with newstate set to zero 74853541Sshin * which allows us to cancel any pending async locks and clear the 749 * locking state. 750 */ 751static void 752nlm_host_notify(struct nlm_host *host, int newstate) 753{ 754 struct nlm_async_lock *af; 755 756 if (newstate) { 757 NLM_DEBUG(1, "NLM: host %s (sysid %d) rebooted, new " 758 "state is %d\n", host->nh_caller_name, 759 host->nh_sysid, newstate); 760 } 761 762 /* 763 * Cancel any pending async locks for this host. 764 */ 765 mtx_lock(&host->nh_lock); 766 while ((af = TAILQ_FIRST(&host->nh_pending)) != NULL) { 767 /* 768 * nlm_cancel_async_lock will remove the entry from 769 * nh_pending and free it. 770 */ 771 nlm_cancel_async_lock(af); 772 } 773 mtx_unlock(&host->nh_lock); 774 nlm_check_expired_locks(host); 775 776 /* 777 * The host just rebooted - trash its locks. 778 */ 779 lf_clearremotesys(host->nh_sysid); 780 host->nh_state = newstate; 781 782 /* 783 * If we have any remote locks for this host (i.e. it 784 * represents a remote NFS server that our local NFS client 785 * has locks for), start a recovery thread. 786 */ 787 if (newstate != 0 788 && host->nh_monstate != NLM_RECOVERING 789 && lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid) > 0) { 790 struct thread *td; 791 host->nh_monstate = NLM_RECOVERING; 792 refcount_acquire(&host->nh_refs); 793 kthread_add(nlm_client_recovery_start, host, curproc, &td, 0, 0, 794 "NFS lock recovery for %s", host->nh_caller_name); 795 } 796} 797 798/* 799 * Sysctl handler to count the number of locks for a sysid. 800 */ 801static int 802nlm_host_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 803{ 804 struct nlm_host *host; 805 int count; 806 807 host = oidp->oid_arg1; 808 count = lf_countlocks(host->nh_sysid); 809 return sysctl_handle_int(oidp, &count, 0, req); 810} 811 812/* 813 * Sysctl handler to count the number of client locks for a sysid. 814 */ 815static int 816nlm_host_client_lock_count_sysctl(SYSCTL_HANDLER_ARGS) 817{ 818 struct nlm_host *host; 819 int count; 820 821 host = oidp->oid_arg1; 822 count = lf_countlocks(NLM_SYSID_CLIENT | host->nh_sysid); 823 return sysctl_handle_int(oidp, &count, 0, req); 824} 825 826/* 827 * Create a new NLM host. 828 */ 829static struct nlm_host * 830nlm_create_host(const char* caller_name) 831{ 832 struct nlm_host *host; 833 struct sysctl_oid *oid; 834 835 mtx_assert(&nlm_global_lock, MA_OWNED); 836 837 NLM_DEBUG(1, "NLM: new host %s (sysid %d)\n", 838 caller_name, nlm_next_sysid); 839 host = malloc(sizeof(struct nlm_host), M_NLM, M_NOWAIT|M_ZERO); 840 if (!host) 841 return (NULL); 842 mtx_init(&host->nh_lock, "nh_lock", NULL, MTX_DEF); 843 host->nh_refs = 1; 844 strlcpy(host->nh_caller_name, caller_name, MAXNAMELEN); 845 host->nh_sysid = nlm_next_sysid++; 846 snprintf(host->nh_sysid_string, sizeof(host->nh_sysid_string), 847 "%d", host->nh_sysid); 848 host->nh_vers = 0; 849 host->nh_state = 0; 850 host->nh_monstate = NLM_UNMONITORED; 851 host->nh_grantcookie = 1; 852 TAILQ_INIT(&host->nh_pending); 853 TAILQ_INIT(&host->nh_granted); 854 TAILQ_INIT(&host->nh_finished); 855 TAILQ_INSERT_TAIL(&nlm_hosts, host, nh_link); 856 857 mtx_unlock(&nlm_global_lock); 858 859 sysctl_ctx_init(&host->nh_sysctl); 860 oid = SYSCTL_ADD_NODE(&host->nh_sysctl, 861 SYSCTL_STATIC_CHILDREN(_vfs_nlm_sysid), 862 OID_AUTO, host->nh_sysid_string, CTLFLAG_RD, NULL, ""); 863 SYSCTL_ADD_STRING(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 864 "hostname", CTLFLAG_RD, host->nh_caller_name, 0, ""); 865 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 866 "version", CTLFLAG_RD, &host->nh_vers, 0, ""); 867 SYSCTL_ADD_INT(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 868 "monitored", CTLFLAG_RD, &host->nh_monstate, 0, ""); 869 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 870 "lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 871 nlm_host_lock_count_sysctl, "I", ""); 872 SYSCTL_ADD_PROC(&host->nh_sysctl, SYSCTL_CHILDREN(oid), OID_AUTO, 873 "client_lock_count", CTLTYPE_INT | CTLFLAG_RD, host, 0, 874 nlm_host_client_lock_count_sysctl, "I", ""); 875 876 mtx_lock(&nlm_global_lock); 877 878 return (host); 879} 880 881/* 882 * Acquire the next sysid for remote locks not handled by the NLM. 883 */ 884uint32_t 885nlm_acquire_next_sysid(void) 886{ 887 uint32_t next_sysid; 888 889 mtx_lock(&nlm_global_lock); 890 next_sysid = nlm_next_sysid++; 891 mtx_unlock(&nlm_global_lock); 892 return (next_sysid); 893} 894 895/* 896 * Return non-zero if the address parts of the two sockaddrs are the 897 * same. 898 */ 899static int 900nlm_compare_addr(const struct sockaddr *a, const struct sockaddr *b) 901{ 902 const struct sockaddr_in *a4, *b4; 903#ifdef INET6 904 const struct sockaddr_in6 *a6, *b6; 905#endif 906 907 if (a->sa_family != b->sa_family) 908 return (FALSE); 909 910 switch (a->sa_family) { 911 case AF_INET: 912 a4 = (const struct sockaddr_in *) a; 913 b4 = (const struct sockaddr_in *) b; 914 return !memcmp(&a4->sin_addr, &b4->sin_addr, 915 sizeof(a4->sin_addr)); 916#ifdef INET6 917 case AF_INET6: 918 a6 = (const struct sockaddr_in6 *) a; 919 b6 = (const struct sockaddr_in6 *) b; 920 return !memcmp(&a6->sin6_addr, &b6->sin6_addr, 921 sizeof(a6->sin6_addr)); 922#endif 923 } 924 925 return (0); 926} 927 928/* 929 * Check for idle hosts and stop monitoring them. We could also free 930 * the host structure here, possibly after a larger timeout but that 931 * would require some care to avoid races with 932 * e.g. nlm_host_lock_count_sysctl. 933 */ 934static void 935nlm_check_idle(void) 936{ 937 struct nlm_host *host; 938 939 mtx_assert(&nlm_global_lock, MA_OWNED); 940 941 if (time_uptime <= nlm_next_idle_check) 942 return; 943 944 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 945 946 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 947 if (host->nh_monstate == NLM_MONITORED 948 && time_uptime > host->nh_idle_timeout) { 949 mtx_unlock(&nlm_global_lock); 950 if (lf_countlocks(host->nh_sysid) > 0 951 || lf_countlocks(NLM_SYSID_CLIENT 952 + host->nh_sysid)) { 953 host->nh_idle_timeout = 954 time_uptime + NLM_IDLE_TIMEOUT; 955 mtx_lock(&nlm_global_lock); 956 continue; 957 } 958 nlm_host_unmonitor(host); 959 mtx_lock(&nlm_global_lock); 960 } 961 } 962} 963 964/* 965 * Search for an existing NLM host that matches the given name 966 * (typically the caller_name element of an nlm4_lock). If none is 967 * found, create a new host. If 'addr' is non-NULL, record the remote 968 * address of the host so that we can call it back for async 969 * responses. If 'vers' is greater than zero then record the NLM 970 * program version to use to communicate with this client. 971 */ 972struct nlm_host * 973nlm_find_host_by_name(const char *name, const struct sockaddr *addr, 974 rpcvers_t vers) 975{ 976 struct nlm_host *host; 977 978 mtx_lock(&nlm_global_lock); 979 980 /* 981 * The remote host is determined by caller_name. 982 */ 983 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 984 if (!strcmp(host->nh_caller_name, name)) 985 break; 986 } 987 988 if (!host) { 989 host = nlm_create_host(name); 990 if (!host) { 991 mtx_unlock(&nlm_global_lock); 992 return (NULL); 993 } 994 } 995 refcount_acquire(&host->nh_refs); 996 997 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 998 999 /* 1000 * If we have an address for the host, record it so that we 1001 * can send async replies etc. 1002 */ 1003 if (addr) { 1004 1005 KASSERT(addr->sa_len < sizeof(struct sockaddr_storage), 1006 ("Strange remote transport address length")); 1007 1008 /* 1009 * If we have seen an address before and we currently 1010 * have an RPC client handle, make sure the address is 1011 * the same, otherwise discard the client handle. 1012 */ 1013 if (host->nh_addr.ss_len && host->nh_srvrpc.nr_client) { 1014 if (!nlm_compare_addr( 1015 (struct sockaddr *) &host->nh_addr, 1016 addr) 1017 || host->nh_vers != vers) { 1018 CLIENT *client; 1019 mtx_lock(&host->nh_lock); 1020 client = host->nh_srvrpc.nr_client; 1021 host->nh_srvrpc.nr_client = NULL; 1022 mtx_unlock(&host->nh_lock); 1023 if (client) { 1024 CLNT_RELEASE(client); 1025 } 1026 } 1027 } 1028 memcpy(&host->nh_addr, addr, addr->sa_len); 1029 host->nh_vers = vers; 1030 } 1031 1032 nlm_check_idle(); 1033 1034 mtx_unlock(&nlm_global_lock); 1035 1036 return (host); 1037} 1038 1039/* 1040 * Search for an existing NLM host that matches the given remote 1041 * address. If none is found, create a new host with the requested 1042 * address and remember 'vers' as the NLM protocol version to use for 1043 * that host. 1044 */ 1045struct nlm_host * 1046nlm_find_host_by_addr(const struct sockaddr *addr, int vers) 1047{ 1048 /* 1049 * Fake up a name using inet_ntop. This buffer is 1050 * large enough for an IPv6 address. 1051 */ 1052 char tmp[sizeof "ffff:ffff:ffff:ffff:ffff:ffff:255.255.255.255"]; 1053 struct nlm_host *host; 1054 1055 switch (addr->sa_family) { 1056 case AF_INET: 1057 inet_ntop(AF_INET, 1058 &((const struct sockaddr_in *) addr)->sin_addr, 1059 tmp, sizeof tmp); 1060 break; 1061#ifdef INET6 1062 case AF_INET6: 1063 inet_ntop(AF_INET6, 1064 &((const struct sockaddr_in6 *) addr)->sin6_addr, 1065 tmp, sizeof tmp); 1066 break; 1067#endif 1068 default: 1069 strcmp(tmp, "<unknown>"); 1070 } 1071 1072 1073 mtx_lock(&nlm_global_lock); 1074 1075 /* 1076 * The remote host is determined by caller_name. 1077 */ 1078 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1079 if (nlm_compare_addr(addr, 1080 (const struct sockaddr *) &host->nh_addr)) 1081 break; 1082 } 1083 1084 if (!host) { 1085 host = nlm_create_host(tmp); 1086 if (!host) { 1087 mtx_unlock(&nlm_global_lock); 1088 return (NULL); 1089 } 1090 memcpy(&host->nh_addr, addr, addr->sa_len); 1091 host->nh_vers = vers; 1092 } 1093 refcount_acquire(&host->nh_refs); 1094 1095 host->nh_idle_timeout = time_uptime + NLM_IDLE_TIMEOUT; 1096 1097 nlm_check_idle(); 1098 1099 mtx_unlock(&nlm_global_lock); 1100 1101 return (host); 1102} 1103 1104/* 1105 * Find the NLM host that matches the value of 'sysid'. If none 1106 * exists, return NULL. 1107 */ 1108static struct nlm_host * 1109nlm_find_host_by_sysid(int sysid) 1110{ 1111 struct nlm_host *host; 1112 1113 TAILQ_FOREACH(host, &nlm_hosts, nh_link) { 1114 if (host->nh_sysid == sysid) { 1115 refcount_acquire(&host->nh_refs); 1116 return (host); 1117 } 1118 } 1119 1120 return (NULL); 1121} 1122 1123void nlm_host_release(struct nlm_host *host) 1124{ 1125 if (refcount_release(&host->nh_refs)) { 1126 /* 1127 * Free the host 1128 */ 1129 nlm_host_destroy(host); 1130 } 1131} 1132 1133/* 1134 * Unregister this NLM host with the local NSM due to idleness. 1135 */ 1136static void 1137nlm_host_unmonitor(struct nlm_host *host) 1138{ 1139 mon_id smmonid; 1140 sm_stat_res smstat; 1141 struct timeval timo; 1142 enum clnt_stat stat; 1143 1144 NLM_DEBUG(1, "NLM: unmonitoring %s (sysid %d)\n", 1145 host->nh_caller_name, host->nh_sysid); 1146 1147 /* 1148 * We put our assigned system ID value in the priv field to 1149 * make it simpler to find the host if we are notified of a 1150 * host restart. 1151 */ 1152 smmonid.mon_name = host->nh_caller_name; 1153 smmonid.my_id.my_name = "localhost"; 1154 smmonid.my_id.my_prog = NLM_PROG; 1155 smmonid.my_id.my_vers = NLM_SM; 1156 smmonid.my_id.my_proc = NLM_SM_NOTIFY; 1157 1158 timo.tv_sec = 25; 1159 timo.tv_usec = 0; 1160 stat = CLNT_CALL(nlm_nsm, SM_UNMON, 1161 (xdrproc_t) xdr_mon, &smmonid, 1162 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1163 1164 if (stat != RPC_SUCCESS) { 1165 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1166 return; 1167 } 1168 if (smstat.res_stat == stat_fail) { 1169 NLM_ERR("Local NSM refuses to unmonitor %s\n", 1170 host->nh_caller_name); 1171 return; 1172 } 1173 1174 host->nh_monstate = NLM_UNMONITORED; 1175} 1176 1177/* 1178 * Register this NLM host with the local NSM so that we can be 1179 * notified if it reboots. 1180 */ 1181void 1182nlm_host_monitor(struct nlm_host *host, int state) 1183{ 1184 mon smmon; 1185 sm_stat_res smstat; 1186 struct timeval timo; 1187 enum clnt_stat stat; 1188 1189 if (state && !host->nh_state) { 1190 /* 1191 * This is the first time we have seen an NSM state 1192 * value for this host. We record it here to help 1193 * detect host reboots. 1194 */ 1195 host->nh_state = state; 1196 NLM_DEBUG(1, "NLM: host %s (sysid %d) has NSM state %d\n", 1197 host->nh_caller_name, host->nh_sysid, state); 1198 } 1199 1200 mtx_lock(&host->nh_lock); 1201 if (host->nh_monstate != NLM_UNMONITORED) { 1202 mtx_unlock(&host->nh_lock); 1203 return; 1204 } 1205 host->nh_monstate = NLM_MONITORED; 1206 mtx_unlock(&host->nh_lock); 1207 1208 NLM_DEBUG(1, "NLM: monitoring %s (sysid %d)\n", 1209 host->nh_caller_name, host->nh_sysid); 1210 1211 /* 1212 * We put our assigned system ID value in the priv field to 1213 * make it simpler to find the host if we are notified of a 1214 * host restart. 1215 */ 1216 smmon.mon_id.mon_name = host->nh_caller_name; 1217 smmon.mon_id.my_id.my_name = "localhost"; 1218 smmon.mon_id.my_id.my_prog = NLM_PROG; 1219 smmon.mon_id.my_id.my_vers = NLM_SM; 1220 smmon.mon_id.my_id.my_proc = NLM_SM_NOTIFY; 1221 memcpy(smmon.priv, &host->nh_sysid, sizeof(host->nh_sysid)); 1222 1223 timo.tv_sec = 25; 1224 timo.tv_usec = 0; 1225 stat = CLNT_CALL(nlm_nsm, SM_MON, 1226 (xdrproc_t) xdr_mon, &smmon, 1227 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1228 1229 if (stat != RPC_SUCCESS) { 1230 NLM_ERR("Failed to contact local NSM - rpc error %d\n", stat); 1231 return; 1232 } 1233 if (smstat.res_stat == stat_fail) { 1234 NLM_ERR("Local NSM refuses to monitor %s\n", 1235 host->nh_caller_name); 1236 mtx_lock(&host->nh_lock); 1237 host->nh_monstate = NLM_MONITOR_FAILED; 1238 mtx_unlock(&host->nh_lock); 1239 return; 1240 } 1241 1242 host->nh_monstate = NLM_MONITORED; 1243} 1244 1245/* 1246 * Return an RPC client handle that can be used to talk to the NLM 1247 * running on the given host. 1248 */ 1249CLIENT * 1250nlm_host_get_rpc(struct nlm_host *host, bool_t isserver) 1251{ 1252 struct nlm_rpc *rpc; 1253 CLIENT *client; 1254 1255 mtx_lock(&host->nh_lock); 1256 1257 if (isserver) 1258 rpc = &host->nh_srvrpc; 1259 else 1260 rpc = &host->nh_clntrpc; 1261 1262 /* 1263 * We can't hold onto RPC handles for too long - the async 1264 * call/reply protocol used by some NLM clients makes it hard 1265 * to tell when they change port numbers (e.g. after a 1266 * reboot). Note that if a client reboots while it isn't 1267 * holding any locks, it won't bother to notify us. We 1268 * expire the RPC handles after two minutes. 1269 */ 1270 if (rpc->nr_client && time_uptime > rpc->nr_create_time + 2*60) { 1271 client = rpc->nr_client; 1272 rpc->nr_client = NULL; 1273 mtx_unlock(&host->nh_lock); 1274 CLNT_RELEASE(client); 1275 mtx_lock(&host->nh_lock); 1276 } 1277 1278 if (!rpc->nr_client) { 1279 mtx_unlock(&host->nh_lock); 1280 client = nlm_get_rpc((struct sockaddr *)&host->nh_addr, 1281 NLM_PROG, host->nh_vers); 1282 mtx_lock(&host->nh_lock); 1283 1284 if (client) { 1285 if (rpc->nr_client) { 1286 mtx_unlock(&host->nh_lock); 1287 CLNT_DESTROY(client); 1288 mtx_lock(&host->nh_lock); 1289 } else { 1290 rpc->nr_client = client; 1291 rpc->nr_create_time = time_uptime; 1292 } 1293 } 1294 } 1295 1296 client = rpc->nr_client; 1297 if (client) 1298 CLNT_ACQUIRE(client); 1299 mtx_unlock(&host->nh_lock); 1300 1301 return (client); 1302 1303} 1304 1305int nlm_host_get_sysid(struct nlm_host *host) 1306{ 1307 1308 return (host->nh_sysid); 1309} 1310 1311int 1312nlm_host_get_state(struct nlm_host *host) 1313{ 1314 1315 return (host->nh_state); 1316} 1317 1318void * 1319nlm_register_wait_lock(struct nlm4_lock *lock, struct vnode *vp) 1320{ 1321 struct nlm_waiting_lock *nw; 1322 1323 nw = malloc(sizeof(struct nlm_waiting_lock), M_NLM, M_WAITOK); 1324 nw->nw_lock = *lock; 1325 memcpy(&nw->nw_fh.fh_bytes, nw->nw_lock.fh.n_bytes, 1326 nw->nw_lock.fh.n_len); 1327 nw->nw_lock.fh.n_bytes = nw->nw_fh.fh_bytes; 1328 nw->nw_waiting = TRUE; 1329 nw->nw_vp = vp; 1330 mtx_lock(&nlm_global_lock); 1331 TAILQ_INSERT_TAIL(&nlm_waiting_locks, nw, nw_link); 1332 mtx_unlock(&nlm_global_lock); 1333 1334 return nw; 1335} 1336 1337void 1338nlm_deregister_wait_lock(void *handle) 1339{ 1340 struct nlm_waiting_lock *nw = handle; 1341 1342 mtx_lock(&nlm_global_lock); 1343 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1344 mtx_unlock(&nlm_global_lock); 1345 1346 free(nw, M_NLM); 1347} 1348 1349int 1350nlm_wait_lock(void *handle, int timo) 1351{ 1352 struct nlm_waiting_lock *nw = handle; 1353 int error; 1354 1355 /* 1356 * If the granted message arrived before we got here, 1357 * nw->nw_waiting will be FALSE - in that case, don't sleep. 1358 */ 1359 mtx_lock(&nlm_global_lock); 1360 error = 0; 1361 if (nw->nw_waiting) 1362 error = msleep(nw, &nlm_global_lock, PCATCH, "nlmlock", timo); 1363 TAILQ_REMOVE(&nlm_waiting_locks, nw, nw_link); 1364 if (error) { 1365 /* 1366 * The granted message may arrive after the 1367 * interrupt/timeout but before we manage to lock the 1368 * mutex. Detect this by examining nw_lock. 1369 */ 1370 if (!nw->nw_waiting) 1371 error = 0; 1372 } else { 1373 /* 1374 * If nlm_cancel_wait is called, then error will be 1375 * zero but nw_waiting will still be TRUE. We 1376 * translate this into EINTR. 1377 */ 1378 if (nw->nw_waiting) 1379 error = EINTR; 1380 } 1381 mtx_unlock(&nlm_global_lock); 1382 1383 free(nw, M_NLM); 1384 1385 return (error); 1386} 1387 1388void 1389nlm_cancel_wait(struct vnode *vp) 1390{ 1391 struct nlm_waiting_lock *nw; 1392 1393 mtx_lock(&nlm_global_lock); 1394 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1395 if (nw->nw_vp == vp) { 1396 wakeup(nw); 1397 } 1398 } 1399 mtx_unlock(&nlm_global_lock); 1400} 1401 1402 1403/**********************************************************************/ 1404 1405/* 1406 * Syscall interface with userland. 1407 */ 1408 1409extern void nlm_prog_0(struct svc_req *rqstp, SVCXPRT *transp); 1410extern void nlm_prog_1(struct svc_req *rqstp, SVCXPRT *transp); 1411extern void nlm_prog_3(struct svc_req *rqstp, SVCXPRT *transp); 1412extern void nlm_prog_4(struct svc_req *rqstp, SVCXPRT *transp); 1413 1414static int 1415nlm_register_services(SVCPOOL *pool, int addr_count, char **addrs) 1416{ 1417 static rpcvers_t versions[] = { 1418 NLM_SM, NLM_VERS, NLM_VERSX, NLM_VERS4 1419 }; 1420 static void (*dispatchers[])(struct svc_req *, SVCXPRT *) = { 1421 nlm_prog_0, nlm_prog_1, nlm_prog_3, nlm_prog_4 1422 }; 1423 static const int version_count = sizeof(versions) / sizeof(versions[0]); 1424 1425 SVCXPRT **xprts; 1426 char netid[16]; 1427 char uaddr[128]; 1428 struct netconfig *nconf; 1429 int i, j, error; 1430 1431 if (!addr_count) { 1432 NLM_ERR("NLM: no service addresses given - can't start server"); 1433 return (EINVAL); 1434 } 1435 1436 xprts = malloc(addr_count * sizeof(SVCXPRT *), M_NLM, M_WAITOK|M_ZERO); 1437 for (i = 0; i < version_count; i++) { 1438 for (j = 0; j < addr_count; j++) { 1439 /* 1440 * Create transports for the first version and 1441 * then just register everything else to the 1442 * same transports. 1443 */ 1444 if (i == 0) { 1445 char *up; 1446 1447 error = copyin(&addrs[2*j], &up, 1448 sizeof(char*)); 1449 if (error) 1450 goto out; 1451 error = copyinstr(up, netid, sizeof(netid), 1452 NULL); 1453 if (error) 1454 goto out; 1455 error = copyin(&addrs[2*j+1], &up, 1456 sizeof(char*)); 1457 if (error) 1458 goto out; 1459 error = copyinstr(up, uaddr, sizeof(uaddr), 1460 NULL); 1461 if (error) 1462 goto out; 1463 nconf = getnetconfigent(netid); 1464 if (!nconf) { 1465 NLM_ERR("Can't lookup netid %s\n", 1466 netid); 1467 error = EINVAL; 1468 goto out; 1469 } 1470 xprts[j] = svc_tp_create(pool, dispatchers[i], 1471 NLM_PROG, versions[i], uaddr, nconf); 1472 if (!xprts[j]) { 1473 NLM_ERR("NLM: unable to create " 1474 "(NLM_PROG, %d).\n", versions[i]); 1475 error = EINVAL; 1476 goto out; 1477 } 1478 freenetconfigent(nconf); 1479 } else { 1480 nconf = getnetconfigent(xprts[j]->xp_netid); 1481 rpcb_unset(NLM_PROG, versions[i], nconf); 1482 if (!svc_reg(xprts[j], NLM_PROG, versions[i], 1483 dispatchers[i], nconf)) { 1484 NLM_ERR("NLM: can't register " 1485 "(NLM_PROG, %d)\n", versions[i]); 1486 error = EINVAL; 1487 goto out; 1488 } 1489 } 1490 } 1491 } 1492 error = 0; 1493out: 1494 for (j = 0; j < addr_count; j++) { 1495 if (xprts[j]) 1496 SVC_RELEASE(xprts[j]); 1497 } 1498 free(xprts, M_NLM); 1499 return (error); 1500} 1501 1502/* 1503 * Main server entry point. Contacts the local NSM to get its current 1504 * state and send SM_UNMON_ALL. Registers the NLM services and then 1505 * services requests. Does not return until the server is interrupted 1506 * by a signal. 1507 */ 1508static int 1509nlm_server_main(int addr_count, char **addrs) 1510{ 1511 struct thread *td = curthread; 1512 int error; 1513 SVCPOOL *pool = NULL; 1514 struct sockopt opt; 1515 int portlow; 1516#ifdef INET6 1517 struct sockaddr_in6 sin6; 1518#endif 1519 struct sockaddr_in sin; 1520 my_id id; 1521 sm_stat smstat; 1522 struct timeval timo; 1523 enum clnt_stat stat; 1524 struct nlm_host *host, *nhost; 1525 struct nlm_waiting_lock *nw; 1526 vop_advlock_t *old_nfs_advlock; 1527 vop_reclaim_t *old_nfs_reclaim; 1528 int v4_used; 1529#ifdef INET6 1530 int v6_used; 1531#endif 1532 1533 if (nlm_socket) { 1534 NLM_ERR("NLM: can't start server - " 1535 "it appears to be running already\n"); 1536 return (EPERM); 1537 } 1538 1539 memset(&opt, 0, sizeof(opt)); 1540 1541 nlm_socket = NULL; 1542 error = socreate(AF_INET, &nlm_socket, SOCK_DGRAM, 0, 1543 td->td_ucred, td); 1544 if (error) { 1545 NLM_ERR("NLM: can't create IPv4 socket - error %d\n", error); 1546 return (error); 1547 } 1548 opt.sopt_dir = SOPT_SET; 1549 opt.sopt_level = IPPROTO_IP; 1550 opt.sopt_name = IP_PORTRANGE; 1551 portlow = IP_PORTRANGE_LOW; 1552 opt.sopt_val = &portlow; 1553 opt.sopt_valsize = sizeof(portlow); 1554 sosetopt(nlm_socket, &opt); 1555 1556#ifdef INET6 1557 nlm_socket6 = NULL; 1558 error = socreate(AF_INET6, &nlm_socket6, SOCK_DGRAM, 0, 1559 td->td_ucred, td); 1560 if (error) { 1561 NLM_ERR("NLM: can't create IPv6 socket - error %d\n", error); 1562 goto out; 1563 return (error); 1564 } 1565 opt.sopt_dir = SOPT_SET; 1566 opt.sopt_level = IPPROTO_IPV6; 1567 opt.sopt_name = IPV6_PORTRANGE; 1568 portlow = IPV6_PORTRANGE_LOW; 1569 opt.sopt_val = &portlow; 1570 opt.sopt_valsize = sizeof(portlow); 1571 sosetopt(nlm_socket6, &opt); 1572#endif 1573 1574 nlm_auth = authunix_create(curthread->td_ucred); 1575 1576#ifdef INET6 1577 memset(&sin6, 0, sizeof(sin6)); 1578 sin6.sin6_len = sizeof(sin6); 1579 sin6.sin6_family = AF_INET6; 1580 sin6.sin6_addr = in6addr_loopback; 1581 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin6, SM_PROG, SM_VERS); 1582 if (!nlm_nsm) { 1583#endif 1584 memset(&sin, 0, sizeof(sin)); 1585 sin.sin_len = sizeof(sin); 1586 sin.sin_family = AF_INET; 1587 sin.sin_addr.s_addr = htonl(INADDR_LOOPBACK); 1588 nlm_nsm = nlm_get_rpc((struct sockaddr *) &sin, SM_PROG, 1589 SM_VERS); 1590#ifdef INET6 1591 } 1592#endif 1593 1594 if (!nlm_nsm) { 1595 NLM_ERR("Can't start NLM - unable to contact NSM\n"); 1596 error = EINVAL; 1597 goto out; 1598 } 1599 1600 pool = svcpool_create("NLM", NULL); 1601 1602 error = nlm_register_services(pool, addr_count, addrs); 1603 if (error) 1604 goto out; 1605 1606 memset(&id, 0, sizeof(id)); 1607 id.my_name = "NFS NLM"; 1608 1609 timo.tv_sec = 25; 1610 timo.tv_usec = 0; 1611 stat = CLNT_CALL(nlm_nsm, SM_UNMON_ALL, 1612 (xdrproc_t) xdr_my_id, &id, 1613 (xdrproc_t) xdr_sm_stat, &smstat, timo); 1614 1615 if (stat != RPC_SUCCESS) { 1616 struct rpc_err err; 1617 1618 CLNT_GETERR(nlm_nsm, &err); 1619 NLM_ERR("NLM: unexpected error contacting NSM, " 1620 "stat=%d, errno=%d\n", stat, err.re_errno); 1621 error = EINVAL; 1622 goto out; 1623 } 1624 1625 NLM_DEBUG(1, "NLM: local NSM state is %d\n", smstat.state); 1626 nlm_nsm_state = smstat.state; 1627 1628 old_nfs_advlock = nfs_advlock_p; 1629 nfs_advlock_p = nlm_advlock; 1630 old_nfs_reclaim = nfs_reclaim_p; 1631 nfs_reclaim_p = nlm_reclaim; 1632 1633 svc_run(pool); 1634 error = 0; 1635 1636 nfs_advlock_p = old_nfs_advlock; 1637 nfs_reclaim_p = old_nfs_reclaim; 1638 1639out: 1640 if (pool) 1641 svcpool_destroy(pool); 1642 1643 /* 1644 * We are finished communicating with the NSM. 1645 */ 1646 if (nlm_nsm) { 1647 CLNT_RELEASE(nlm_nsm); 1648 nlm_nsm = NULL; 1649 } 1650 1651 /* 1652 * Trash all the existing state so that if the server 1653 * restarts, it gets a clean slate. This is complicated by the 1654 * possibility that there may be other threads trying to make 1655 * client locking requests. 1656 * 1657 * First we fake a client reboot notification which will 1658 * cancel any pending async locks and purge remote lock state 1659 * from the local lock manager. We release the reference from 1660 * nlm_hosts to the host (which may remove it from the list 1661 * and free it). After this phase, the only entries in the 1662 * nlm_host list should be from other threads performing 1663 * client lock requests. We arrange to defer closing the 1664 * sockets until the last RPC client handle is released. 1665 */ 1666 v4_used = 0; 1667#ifdef INET6 1668 v6_used = 0; 1669#endif 1670 mtx_lock(&nlm_global_lock); 1671 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 1672 wakeup(nw); 1673 } 1674 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1675 mtx_unlock(&nlm_global_lock); 1676 nlm_host_notify(host, 0); 1677 nlm_host_release(host); 1678 mtx_lock(&nlm_global_lock); 1679 } 1680 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, nhost) { 1681 mtx_lock(&host->nh_lock); 1682 if (host->nh_srvrpc.nr_client 1683 || host->nh_clntrpc.nr_client) { 1684 if (host->nh_addr.ss_family == AF_INET) 1685 v4_used++; 1686#ifdef INET6 1687 if (host->nh_addr.ss_family == AF_INET6) 1688 v6_used++; 1689#endif 1690 /* 1691 * Note that the rpc over udp code copes 1692 * correctly with the fact that a socket may 1693 * be used by many rpc handles. 1694 */ 1695 if (host->nh_srvrpc.nr_client) 1696 CLNT_CONTROL(host->nh_srvrpc.nr_client, 1697 CLSET_FD_CLOSE, 0); 1698 if (host->nh_clntrpc.nr_client) 1699 CLNT_CONTROL(host->nh_clntrpc.nr_client, 1700 CLSET_FD_CLOSE, 0); 1701 } 1702 mtx_unlock(&host->nh_lock); 1703 } 1704 mtx_unlock(&nlm_global_lock); 1705 1706 AUTH_DESTROY(nlm_auth); 1707 1708 if (!v4_used) 1709 soclose(nlm_socket); 1710 nlm_socket = NULL; 1711#ifdef INET6 1712 if (!v6_used) 1713 soclose(nlm_socket6); 1714 nlm_socket6 = NULL; 1715#endif 1716 1717 return (error); 1718} 1719 1720int 1721nlm_syscall(struct thread *td, struct nlm_syscall_args *uap) 1722{ 1723 int error; 1724 1725#if __FreeBSD_version >= 700000 1726 error = priv_check(td, PRIV_NFS_LOCKD); 1727#else 1728 error = suser(td); 1729#endif 1730 if (error) 1731 return (error); 1732 1733 nlm_debug_level = uap->debug_level; 1734 nlm_grace_threshold = time_uptime + uap->grace_period; 1735 nlm_next_idle_check = time_uptime + NLM_IDLE_PERIOD; 1736 1737 return nlm_server_main(uap->addr_count, uap->addrs); 1738} 1739 1740/**********************************************************************/ 1741 1742/* 1743 * NLM implementation details, called from the RPC stubs. 1744 */ 1745 1746 1747void 1748nlm_sm_notify(struct nlm_sm_status *argp) 1749{ 1750 uint32_t sysid; 1751 struct nlm_host *host; 1752 1753 NLM_DEBUG(3, "nlm_sm_notify(): mon_name = %s\n", argp->mon_name); 1754 memcpy(&sysid, &argp->priv, sizeof(sysid)); 1755 host = nlm_find_host_by_sysid(sysid); 1756 if (host) { 1757 nlm_host_notify(host, argp->state); 1758 nlm_host_release(host); 1759 } 1760} 1761 1762static void 1763nlm_convert_to_fhandle_t(fhandle_t *fhp, struct netobj *p) 1764{ 1765 memcpy(fhp, p->n_bytes, sizeof(fhandle_t)); 1766} 1767 1768struct vfs_state { 1769 struct mount *vs_mp; 1770 struct vnode *vs_vp; 1771 int vs_vfslocked; 1772 int vs_vnlocked; 1773}; 1774 1775static int 1776nlm_get_vfs_state(struct nlm_host *host, struct svc_req *rqstp, 1777 fhandle_t *fhp, struct vfs_state *vs) 1778{ 1779 int error, exflags; 1780 struct ucred *cred = NULL, *credanon; 1781 1782 memset(vs, 0, sizeof(*vs)); 1783 1784 vs->vs_mp = vfs_getvfs(&fhp->fh_fsid); 1785 if (!vs->vs_mp) { 1786 return (ESTALE); 1787 } 1788 vs->vs_vfslocked = VFS_LOCK_GIANT(vs->vs_mp); 1789 1790 error = VFS_CHECKEXP(vs->vs_mp, (struct sockaddr *)&host->nh_addr, 1791 &exflags, &credanon, NULL, NULL); 1792 if (error) 1793 goto out; 1794 1795 if (exflags & MNT_EXRDONLY || (vs->vs_mp->mnt_flag & MNT_RDONLY)) { 1796 error = EROFS; 1797 goto out; 1798 } 1799 1800 error = VFS_FHTOVP(vs->vs_mp, &fhp->fh_fid, &vs->vs_vp); 1801 if (error) 1802 goto out; 1803 vs->vs_vnlocked = TRUE; 1804 1805 if (!svc_getcred(rqstp, &cred, NULL)) { 1806 error = EINVAL; 1807 goto out; 1808 } 1809 if (cred->cr_uid == 0 || (exflags & MNT_EXPORTANON)) { 1810 crfree(cred); 1811 cred = credanon; 1812 credanon = NULL; 1813 } 1814 1815 /* 1816 * Check cred. 1817 */ 1818 error = VOP_ACCESS(vs->vs_vp, VWRITE, cred, curthread); 1819 if (error) 1820 goto out; 1821 1822#if __FreeBSD_version < 800011 1823 VOP_UNLOCK(vs->vs_vp, 0, curthread); 1824#else 1825 VOP_UNLOCK(vs->vs_vp, 0); 1826#endif 1827 vs->vs_vnlocked = FALSE; 1828 1829out: 1830 if (cred) 1831 crfree(cred); 1832 if (credanon) 1833 crfree(credanon); 1834 1835 return (error); 1836} 1837 1838static void 1839nlm_release_vfs_state(struct vfs_state *vs) 1840{ 1841 1842 if (vs->vs_vp) { 1843 if (vs->vs_vnlocked) 1844 vput(vs->vs_vp); 1845 else 1846 vrele(vs->vs_vp); 1847 } 1848 if (vs->vs_mp) 1849 vfs_rel(vs->vs_mp); 1850 VFS_UNLOCK_GIANT(vs->vs_vfslocked); 1851} 1852 1853static nlm4_stats 1854nlm_convert_error(int error) 1855{ 1856 1857 if (error == ESTALE) 1858 return nlm4_stale_fh; 1859 else if (error == EROFS) 1860 return nlm4_rofs; 1861 else 1862 return nlm4_failed; 1863} 1864 1865int 1866nlm_do_test(nlm4_testargs *argp, nlm4_testres *result, struct svc_req *rqstp, 1867 CLIENT **rpcp) 1868{ 1869 fhandle_t fh; 1870 struct vfs_state vs; 1871 struct nlm_host *host, *bhost; 1872 int error, sysid; 1873 struct flock fl; 1874 1875 memset(result, 0, sizeof(*result)); 1876 memset(&vs, 0, sizeof(vs)); 1877 1878 host = nlm_find_host_by_name(argp->alock.caller_name, 1879 svc_getrpccaller(rqstp), rqstp->rq_vers); 1880 if (!host) { 1881 result->stat.stat = nlm4_denied_nolocks; 1882 return (ENOMEM); 1883 } 1884 1885 NLM_DEBUG(3, "nlm_do_test(): caller_name = %s (sysid = %d)\n", 1886 host->nh_caller_name, host->nh_sysid); 1887 1888 nlm_check_expired_locks(host); 1889 sysid = host->nh_sysid; 1890 1891 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1892 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1893 1894 if (time_uptime < nlm_grace_threshold) { 1895 result->stat.stat = nlm4_denied_grace_period; 1896 goto out; 1897 } 1898 1899 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 1900 if (error) { 1901 result->stat.stat = nlm_convert_error(error); 1902 goto out; 1903 } 1904 1905 fl.l_start = argp->alock.l_offset; 1906 fl.l_len = argp->alock.l_len; 1907 fl.l_pid = argp->alock.svid; 1908 fl.l_sysid = sysid; 1909 fl.l_whence = SEEK_SET; 1910 if (argp->exclusive) 1911 fl.l_type = F_WRLCK; 1912 else 1913 fl.l_type = F_RDLCK; 1914 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_GETLK, &fl, F_REMOTE); 1915 if (error) { 1916 result->stat.stat = nlm4_failed; 1917 goto out; 1918 } 1919 1920 if (fl.l_type == F_UNLCK) { 1921 result->stat.stat = nlm4_granted; 1922 } else { 1923 result->stat.stat = nlm4_denied; 1924 result->stat.nlm4_testrply_u.holder.exclusive = 1925 (fl.l_type == F_WRLCK); 1926 result->stat.nlm4_testrply_u.holder.svid = fl.l_pid; 1927 bhost = nlm_find_host_by_sysid(fl.l_sysid); 1928 if (bhost) { 1929 /* 1930 * We don't have any useful way of recording 1931 * the value of oh used in the original lock 1932 * request. Ideally, the test reply would have 1933 * a space for the owning host's name allowing 1934 * our caller's NLM to keep track. 1935 * 1936 * As far as I can see, Solaris uses an eight 1937 * byte structure for oh which contains a four 1938 * byte pid encoded in local byte order and 1939 * the first four bytes of the host 1940 * name. Linux uses a variable length string 1941 * 'pid@hostname' in ascii but doesn't even 1942 * return that in test replies. 1943 * 1944 * For the moment, return nothing in oh 1945 * (already zero'ed above). 1946 */ 1947 nlm_host_release(bhost); 1948 } 1949 result->stat.nlm4_testrply_u.holder.l_offset = fl.l_start; 1950 result->stat.nlm4_testrply_u.holder.l_len = fl.l_len; 1951 } 1952 1953out: 1954 nlm_release_vfs_state(&vs); 1955 if (rpcp) 1956 *rpcp = nlm_host_get_rpc(host, TRUE); 1957 nlm_host_release(host); 1958 return (0); 1959} 1960 1961int 1962nlm_do_lock(nlm4_lockargs *argp, nlm4_res *result, struct svc_req *rqstp, 1963 bool_t monitor, CLIENT **rpcp) 1964{ 1965 fhandle_t fh; 1966 struct vfs_state vs; 1967 struct nlm_host *host; 1968 int error, sysid; 1969 struct flock fl; 1970 1971 memset(result, 0, sizeof(*result)); 1972 memset(&vs, 0, sizeof(vs)); 1973 1974 host = nlm_find_host_by_name(argp->alock.caller_name, 1975 svc_getrpccaller(rqstp), rqstp->rq_vers); 1976 if (!host) { 1977 result->stat.stat = nlm4_denied_nolocks; 1978 return (ENOMEM); 1979 } 1980 1981 NLM_DEBUG(3, "nlm_do_lock(): caller_name = %s (sysid = %d)\n", 1982 host->nh_caller_name, host->nh_sysid); 1983 1984 if (monitor && host->nh_state && argp->state 1985 && host->nh_state != argp->state) { 1986 /* 1987 * The host rebooted without telling us. Trash its 1988 * locks. 1989 */ 1990 nlm_host_notify(host, argp->state); 1991 } 1992 1993 nlm_check_expired_locks(host); 1994 sysid = host->nh_sysid; 1995 1996 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 1997 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 1998 1999 if (time_uptime < nlm_grace_threshold && !argp->reclaim) { 2000 result->stat.stat = nlm4_denied_grace_period; 2001 goto out; 2002 } 2003 2004 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2005 if (error) { 2006 result->stat.stat = nlm_convert_error(error); 2007 goto out; 2008 } 2009 2010 fl.l_start = argp->alock.l_offset; 2011 fl.l_len = argp->alock.l_len; 2012 fl.l_pid = argp->alock.svid; 2013 fl.l_sysid = sysid; 2014 fl.l_whence = SEEK_SET; 2015 if (argp->exclusive) 2016 fl.l_type = F_WRLCK; 2017 else 2018 fl.l_type = F_RDLCK; 2019 if (argp->block) { 2020 struct nlm_async_lock *af; 2021 CLIENT *client; 2022 struct nlm_grantcookie cookie; 2023 2024 /* 2025 * First, make sure we can contact the host's NLM. 2026 */ 2027 client = nlm_host_get_rpc(host, TRUE); 2028 if (!client) { 2029 result->stat.stat = nlm4_failed; 2030 goto out; 2031 } 2032 2033 /* 2034 * First we need to check and see if there is an 2035 * existing blocked lock that matches. This could be a 2036 * badly behaved client or an RPC re-send. If we find 2037 * one, just return nlm4_blocked. 2038 */ 2039 mtx_lock(&host->nh_lock); 2040 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 2041 if (af->af_fl.l_start == fl.l_start 2042 && af->af_fl.l_len == fl.l_len 2043 && af->af_fl.l_pid == fl.l_pid 2044 && af->af_fl.l_type == fl.l_type) { 2045 break; 2046 } 2047 } 2048 if (!af) { 2049 cookie.ng_sysid = host->nh_sysid; 2050 cookie.ng_cookie = host->nh_grantcookie++; 2051 } 2052 mtx_unlock(&host->nh_lock); 2053 if (af) { 2054 CLNT_RELEASE(client); 2055 result->stat.stat = nlm4_blocked; 2056 goto out; 2057 } 2058 2059 af = malloc(sizeof(struct nlm_async_lock), M_NLM, 2060 M_WAITOK|M_ZERO); 2061 TASK_INIT(&af->af_task, 0, nlm_lock_callback, af); 2062 af->af_vp = vs.vs_vp; 2063 af->af_fl = fl; 2064 af->af_host = host; 2065 af->af_rpc = client; 2066 /* 2067 * We use M_RPC here so that we can xdr_free the thing 2068 * later. 2069 */ 2070 nlm_make_netobj(&af->af_granted.cookie, 2071 (caddr_t)&cookie, sizeof(cookie), M_RPC); 2072 af->af_granted.exclusive = argp->exclusive; 2073 af->af_granted.alock.caller_name = 2074 strdup(argp->alock.caller_name, M_RPC); 2075 nlm_copy_netobj(&af->af_granted.alock.fh, 2076 &argp->alock.fh, M_RPC); 2077 nlm_copy_netobj(&af->af_granted.alock.oh, 2078 &argp->alock.oh, M_RPC); 2079 af->af_granted.alock.svid = argp->alock.svid; 2080 af->af_granted.alock.l_offset = argp->alock.l_offset; 2081 af->af_granted.alock.l_len = argp->alock.l_len; 2082 2083 /* 2084 * Put the entry on the pending list before calling 2085 * VOP_ADVLOCKASYNC. We do this in case the lock 2086 * request was blocked (returning EINPROGRESS) but 2087 * then granted before we manage to run again. The 2088 * client may receive the granted message before we 2089 * send our blocked reply but thats their problem. 2090 */ 2091 mtx_lock(&host->nh_lock); 2092 TAILQ_INSERT_TAIL(&host->nh_pending, af, af_link); 2093 mtx_unlock(&host->nh_lock); 2094 2095 error = VOP_ADVLOCKASYNC(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE, 2096 &af->af_task, &af->af_cookie); 2097 2098 /* 2099 * If the lock completed synchronously, just free the 2100 * tracking structure now. 2101 */ 2102 if (error != EINPROGRESS) { 2103 CLNT_RELEASE(af->af_rpc); 2104 mtx_lock(&host->nh_lock); 2105 TAILQ_REMOVE(&host->nh_pending, af, af_link); 2106 mtx_unlock(&host->nh_lock); 2107 xdr_free((xdrproc_t) xdr_nlm4_testargs, 2108 &af->af_granted); 2109 free(af, M_NLM); 2110 } else { 2111 NLM_DEBUG(2, "NLM: pending async lock %p for %s " 2112 "(sysid %d)\n", af, host->nh_caller_name, sysid); 2113 /* 2114 * Don't vrele the vnode just yet - this must 2115 * wait until either the async callback 2116 * happens or the lock is cancelled. 2117 */ 2118 vs.vs_vp = NULL; 2119 } 2120 } else { 2121 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_SETLK, &fl, F_REMOTE); 2122 } 2123 2124 if (error) { 2125 if (error == EINPROGRESS) { 2126 result->stat.stat = nlm4_blocked; 2127 } else if (error == EDEADLK) { 2128 result->stat.stat = nlm4_deadlck; 2129 } else if (error == EAGAIN) { 2130 result->stat.stat = nlm4_denied; 2131 } else { 2132 result->stat.stat = nlm4_failed; 2133 } 2134 } else { 2135 if (monitor) 2136 nlm_host_monitor(host, argp->state); 2137 result->stat.stat = nlm4_granted; 2138 } 2139 2140out: 2141 nlm_release_vfs_state(&vs); 2142 if (rpcp) 2143 *rpcp = nlm_host_get_rpc(host, TRUE); 2144 nlm_host_release(host); 2145 return (0); 2146} 2147 2148int 2149nlm_do_cancel(nlm4_cancargs *argp, nlm4_res *result, struct svc_req *rqstp, 2150 CLIENT **rpcp) 2151{ 2152 fhandle_t fh; 2153 struct vfs_state vs; 2154 struct nlm_host *host; 2155 int error, sysid; 2156 struct flock fl; 2157 struct nlm_async_lock *af; 2158 2159 memset(result, 0, sizeof(*result)); 2160 memset(&vs, 0, sizeof(vs)); 2161 2162 host = nlm_find_host_by_name(argp->alock.caller_name, 2163 svc_getrpccaller(rqstp), rqstp->rq_vers); 2164 if (!host) { 2165 result->stat.stat = nlm4_denied_nolocks; 2166 return (ENOMEM); 2167 } 2168 2169 NLM_DEBUG(3, "nlm_do_cancel(): caller_name = %s (sysid = %d)\n", 2170 host->nh_caller_name, host->nh_sysid); 2171 2172 nlm_check_expired_locks(host); 2173 sysid = host->nh_sysid; 2174 2175 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2176 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2177 2178 if (time_uptime < nlm_grace_threshold) { 2179 result->stat.stat = nlm4_denied_grace_period; 2180 goto out; 2181 } 2182 2183 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2184 if (error) { 2185 result->stat.stat = nlm_convert_error(error); 2186 goto out; 2187 } 2188 2189 fl.l_start = argp->alock.l_offset; 2190 fl.l_len = argp->alock.l_len; 2191 fl.l_pid = argp->alock.svid; 2192 fl.l_sysid = sysid; 2193 fl.l_whence = SEEK_SET; 2194 if (argp->exclusive) 2195 fl.l_type = F_WRLCK; 2196 else 2197 fl.l_type = F_RDLCK; 2198 2199 /* 2200 * First we need to try and find the async lock request - if 2201 * there isn't one, we give up and return nlm4_denied. 2202 */ 2203 mtx_lock(&host->nh_lock); 2204 2205 TAILQ_FOREACH(af, &host->nh_pending, af_link) { 2206 if (af->af_fl.l_start == fl.l_start 2207 && af->af_fl.l_len == fl.l_len 2208 && af->af_fl.l_pid == fl.l_pid 2209 && af->af_fl.l_type == fl.l_type) { 2210 break; 2211 } 2212 } 2213 2214 if (!af) { 2215 mtx_unlock(&host->nh_lock); 2216 result->stat.stat = nlm4_denied; 2217 goto out; 2218 } 2219 2220 error = nlm_cancel_async_lock(af); 2221 2222 if (error) { 2223 result->stat.stat = nlm4_denied; 2224 } else { 2225 result->stat.stat = nlm4_granted; 2226 } 2227 2228 mtx_unlock(&host->nh_lock); 2229 2230out: 2231 nlm_release_vfs_state(&vs); 2232 if (rpcp) 2233 *rpcp = nlm_host_get_rpc(host, TRUE); 2234 nlm_host_release(host); 2235 return (0); 2236} 2237 2238int 2239nlm_do_unlock(nlm4_unlockargs *argp, nlm4_res *result, struct svc_req *rqstp, 2240 CLIENT **rpcp) 2241{ 2242 fhandle_t fh; 2243 struct vfs_state vs; 2244 struct nlm_host *host; 2245 int error, sysid; 2246 struct flock fl; 2247 2248 memset(result, 0, sizeof(*result)); 2249 memset(&vs, 0, sizeof(vs)); 2250 2251 host = nlm_find_host_by_name(argp->alock.caller_name, 2252 svc_getrpccaller(rqstp), rqstp->rq_vers); 2253 if (!host) { 2254 result->stat.stat = nlm4_denied_nolocks; 2255 return (ENOMEM); 2256 } 2257 2258 NLM_DEBUG(3, "nlm_do_unlock(): caller_name = %s (sysid = %d)\n", 2259 host->nh_caller_name, host->nh_sysid); 2260 2261 nlm_check_expired_locks(host); 2262 sysid = host->nh_sysid; 2263 2264 nlm_convert_to_fhandle_t(&fh, &argp->alock.fh); 2265 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2266 2267 if (time_uptime < nlm_grace_threshold) { 2268 result->stat.stat = nlm4_denied_grace_period; 2269 goto out; 2270 } 2271 2272 error = nlm_get_vfs_state(host, rqstp, &fh, &vs); 2273 if (error) { 2274 result->stat.stat = nlm_convert_error(error); 2275 goto out; 2276 } 2277 2278 fl.l_start = argp->alock.l_offset; 2279 fl.l_len = argp->alock.l_len; 2280 fl.l_pid = argp->alock.svid; 2281 fl.l_sysid = sysid; 2282 fl.l_whence = SEEK_SET; 2283 fl.l_type = F_UNLCK; 2284 error = VOP_ADVLOCK(vs.vs_vp, NULL, F_UNLCK, &fl, F_REMOTE); 2285 2286 /* 2287 * Ignore the error - there is no result code for failure, 2288 * only for grace period. 2289 */ 2290 result->stat.stat = nlm4_granted; 2291 2292out: 2293 nlm_release_vfs_state(&vs); 2294 if (rpcp) 2295 *rpcp = nlm_host_get_rpc(host, TRUE); 2296 nlm_host_release(host); 2297 return (0); 2298} 2299 2300int 2301nlm_do_granted(nlm4_testargs *argp, nlm4_res *result, struct svc_req *rqstp, 2302 2303 CLIENT **rpcp) 2304{ 2305 struct nlm_host *host; 2306 struct nlm_waiting_lock *nw; 2307 2308 memset(result, 0, sizeof(*result)); 2309 2310 host = nlm_find_host_by_addr(svc_getrpccaller(rqstp), rqstp->rq_vers); 2311 if (!host) { 2312 result->stat.stat = nlm4_denied_nolocks; 2313 return (ENOMEM); 2314 } 2315 2316 nlm_copy_netobj(&result->cookie, &argp->cookie, M_RPC); 2317 result->stat.stat = nlm4_denied; 2318 KFAIL_POINT_CODE(DEBUG_FP, nlm_deny_grant, goto out); 2319 2320 mtx_lock(&nlm_global_lock); 2321 TAILQ_FOREACH(nw, &nlm_waiting_locks, nw_link) { 2322 if (!nw->nw_waiting) 2323 continue; 2324 if (argp->alock.svid == nw->nw_lock.svid 2325 && argp->alock.l_offset == nw->nw_lock.l_offset 2326 && argp->alock.l_len == nw->nw_lock.l_len 2327 && argp->alock.fh.n_len == nw->nw_lock.fh.n_len 2328 && !memcmp(argp->alock.fh.n_bytes, nw->nw_lock.fh.n_bytes, 2329 nw->nw_lock.fh.n_len)) { 2330 nw->nw_waiting = FALSE; 2331 wakeup(nw); 2332 result->stat.stat = nlm4_granted; 2333 break; 2334 } 2335 } 2336 mtx_unlock(&nlm_global_lock); 2337 2338out: 2339 if (rpcp) 2340 *rpcp = nlm_host_get_rpc(host, TRUE); 2341 nlm_host_release(host); 2342 return (0); 2343} 2344 2345void 2346nlm_do_granted_res(nlm4_res *argp, struct svc_req *rqstp) 2347{ 2348 struct nlm_host *host = NULL; 2349 struct nlm_async_lock *af = NULL; 2350 int error; 2351 2352 if (argp->cookie.n_len != sizeof(struct nlm_grantcookie)) { 2353 NLM_DEBUG(1, "NLM: bogus grant cookie"); 2354 goto out; 2355 } 2356 2357 host = nlm_find_host_by_sysid(ng_sysid(&argp->cookie)); 2358 if (!host) { 2359 NLM_DEBUG(1, "NLM: Unknown host rejected our grant"); 2360 goto out; 2361 } 2362 2363 mtx_lock(&host->nh_lock); 2364 TAILQ_FOREACH(af, &host->nh_granted, af_link) 2365 if (ng_cookie(&argp->cookie) == 2366 ng_cookie(&af->af_granted.cookie)) 2367 break; 2368 if (af) 2369 TAILQ_REMOVE(&host->nh_granted, af, af_link); 2370 mtx_unlock(&host->nh_lock); 2371 2372 if (!af) { 2373 NLM_DEBUG(1, "NLM: host %s (sysid %d) replied to our grant " 2374 "with unrecognized cookie %d:%d", host->nh_caller_name, 2375 host->nh_sysid, ng_sysid(&argp->cookie), 2376 ng_cookie(&argp->cookie)); 2377 goto out; 2378 } 2379 2380 if (argp->stat.stat != nlm4_granted) { 2381 af->af_fl.l_type = F_UNLCK; 2382 error = VOP_ADVLOCK(af->af_vp, NULL, F_UNLCK, &af->af_fl, F_REMOTE); 2383 if (error) { 2384 NLM_DEBUG(1, "NLM: host %s (sysid %d) rejected our grant " 2385 "and we failed to unlock (%d)", host->nh_caller_name, 2386 host->nh_sysid, error); 2387 goto out; 2388 } 2389 2390 NLM_DEBUG(5, "NLM: async lock %p rejected by host %s (sysid %d)", 2391 af, host->nh_caller_name, host->nh_sysid); 2392 } else { 2393 NLM_DEBUG(5, "NLM: async lock %p accepted by host %s (sysid %d)", 2394 af, host->nh_caller_name, host->nh_sysid); 2395 } 2396 2397 out: 2398 if (af) 2399 nlm_free_async_lock(af); 2400 if (host) 2401 nlm_host_release(host); 2402} 2403 2404void 2405nlm_do_free_all(nlm4_notify *argp) 2406{ 2407 struct nlm_host *host, *thost; 2408 2409 TAILQ_FOREACH_SAFE(host, &nlm_hosts, nh_link, thost) { 2410 if (!strcmp(host->nh_caller_name, argp->name)) 2411 nlm_host_notify(host, argp->state); 2412 } 2413} 2414 2415/* 2416 * Kernel module glue 2417 */ 2418static int 2419nfslockd_modevent(module_t mod, int type, void *data) 2420{ 2421 2422 return (0); 2423} 2424static moduledata_t nfslockd_mod = { 2425 "nfslockd", 2426 nfslockd_modevent, 2427 NULL, 2428}; 2429DECLARE_MODULE(nfslockd, nfslockd_mod, SI_SUB_VFS, SI_ORDER_ANY); 2430 2431/* So that loader and kldload(2) can find us, wherever we are.. */ 2432MODULE_DEPEND(nfslockd, krpc, 1, 1, 1); 2433MODULE_DEPEND(nfslockd, nfs, 1, 1, 1); 2434MODULE_DEPEND(nfslockd, nfslock, 1, 1, 1); 2435MODULE_VERSION(nfslockd, 1); 2436