2 * Copyright (c) 1989, 1993
3 * The Regents of the University of California. All rights reserved.
5 * This code is derived from software contributed to Berkeley by
6 * Rick Macklem at The University of Guelph.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95
37 * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $
38 * $DragonFly: src/sys/vfs/nfs/nfs_syscalls.c,v 1.31 2008/01/05 14:02:41 swildner Exp $
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/sysproto.h>
44 #include <sys/kernel.h>
45 #include <sys/sysctl.h>
47 #include <sys/filedesc.h>
48 #include <sys/vnode.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
55 #include <sys/resourcevar.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/domain.h>
59 #include <sys/protosw.h>
60 #include <sys/nlookup.h>
61 #include <vm/vm_zone.h>
63 #include <netinet/in.h>
64 #include <netinet/tcp.h>
69 #include "nfsm_subs.h"
70 #include "nfsrvcache.h"
75 #include <sys/thread2.h>
77 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
80 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
81 struct nfssvc_sock *slp,
84 extern int nfs_numasync;
86 extern struct nfsstats nfsstats;
87 extern int nfsrvw_procrastinate;
88 extern int nfsrvw_procrastinate_v3;
89 static int nuidhash_max = NFS_MAXUIDHASH;
92 static void nfsrv_zapsock (struct nfssvc_sock *slp);
94 static int nfssvc_iod (struct thread *);
99 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
101 SYSCTL_DECL(_vfs_nfs);
104 int nfsd_waiting = 0;
105 static struct nfsdrt nfsdrt;
106 static int nfs_numnfsd = 0;
107 static void nfsd_rt (int sotype, struct nfsrv_descript *nd,
109 static int nfssvc_addsock (struct file *, struct sockaddr *,
111 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
113 static int nfs_privport = 0;
114 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
115 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
116 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
117 static int nfs_soreserve = 65535;
118 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
121 * NFS server system calls
124 #endif /* NFS_NOSERVER */
126 * nfssvc_args(int flag, caddr_t argp)
128 * Nfs server psuedo system call for the nfsd's
129 * Based on the flag value it either:
130 * - adds a socket to the selection list
131 * - remains in the kernel as an nfsd
132 * - remains in the kernel as an nfsiod
135 sys_nfssvc(struct nfssvc_args *uap)
138 struct nlookupdata nd;
140 struct sockaddr *nam;
141 struct nfsd_args nfsdarg;
142 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
143 struct nfsd_cargs ncd;
145 struct nfssvc_sock *slp;
146 struct nfsuid *nuidp;
147 struct nfsmount *nmp;
149 #endif /* NFS_NOSERVER */
151 struct thread *td = curthread;
156 error = priv_check(td, PRIV_ROOT);
159 KKASSERT(td->td_proc); /* for ucred and p_fd */
160 while (nfssvc_sockhead_flag & SLP_INIT) {
161 nfssvc_sockhead_flag |= SLP_WANTINIT;
162 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
164 if (uap->flag & NFSSVC_BIOD)
165 error = nfssvc_iod(td);
169 #else /* !NFS_NOSERVER */
170 else if (uap->flag & NFSSVC_MNTD) {
171 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
175 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE,
178 error = nlookup(&nd);
180 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
185 if ((vp->v_flag & VROOT) == 0)
187 nmp = VFSTONFS(vp->v_mount);
191 if ((nmp->nm_state & NFSSTA_MNTD) &&
192 (uap->flag & NFSSVC_GOTAUTH) == 0)
194 nmp->nm_state |= NFSSTA_MNTD;
195 error = nfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag,
197 } else if (uap->flag & NFSSVC_ADDSOCK) {
198 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
201 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
205 * Get the client address for connected sockets.
207 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
208 nam = (struct sockaddr *)0;
210 error = getsockaddr(&nam, nfsdarg.name,
217 error = nfssvc_addsock(fp, nam, td);
220 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
223 if ((uap->flag & NFSSVC_AUTHIN) &&
224 ((nfsd = nsd->nsd_nfsd)) != NULL &&
225 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
226 slp = nfsd->nfsd_slp;
229 * First check to see if another nfsd has already
230 * added this credential.
232 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
233 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
234 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
235 (!nfsd->nfsd_nd->nd_nam2 ||
236 netaddr_match(NU_NETFAM(nuidp),
237 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
241 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
242 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
247 if (slp->ns_numuids < nuidhash_max) {
249 nuidp = (struct nfsuid *)
250 kmalloc(sizeof (struct nfsuid), M_NFSUID,
253 nuidp = (struct nfsuid *)0;
254 if ((slp->ns_flag & SLP_VALID) == 0) {
256 kfree((caddr_t)nuidp, M_NFSUID);
258 if (nuidp == (struct nfsuid *)0) {
259 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
260 LIST_REMOVE(nuidp, nu_hash);
261 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
263 if (nuidp->nu_flag & NU_NAM)
264 FREE(nuidp->nu_nam, M_SONAME);
267 nuidp->nu_cr = nsd->nsd_cr;
268 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
269 nuidp->nu_cr.cr_ngroups = NGROUPS;
270 nuidp->nu_cr.cr_ref = 1;
271 nuidp->nu_timestamp = nsd->nsd_timestamp;
272 nuidp->nu_expire = time_second + nsd->nsd_ttl;
274 * and save the session key in nu_key.
276 bcopy(nsd->nsd_key, nuidp->nu_key,
277 sizeof (nsd->nsd_key));
278 if (nfsd->nfsd_nd->nd_nam2) {
279 struct sockaddr_in *saddr;
281 saddr = (struct sockaddr_in *)
282 nfsd->nfsd_nd->nd_nam2;
283 switch (saddr->sin_family) {
285 nuidp->nu_flag |= NU_INETADDR;
287 saddr->sin_addr.s_addr;
291 nuidp->nu_flag |= NU_NAM;
293 dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
297 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
299 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
301 nfsrv_setcred(&nuidp->nu_cr,
302 &nfsd->nfsd_nd->nd_cr);
303 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
307 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
308 nfsd->nfsd_flag |= NFSD_AUTHFAIL;
309 error = nfssvc_nfsd(nsd, uap->argp, td);
311 #endif /* NFS_NOSERVER */
312 if (error == EINTR || error == ERESTART)
319 * Adds a socket to the list for servicing by nfsds.
322 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
325 struct nfssvc_sock *slp;
329 so = (struct socket *)fp->f_data;
331 tslp = (struct nfssvc_sock *)0;
333 * Add it to the list, as required.
335 if (so->so_proto->pr_protocol == IPPROTO_UDP) {
337 if (tslp->ns_flag & SLP_VALID) {
339 FREE(mynam, M_SONAME);
345 * Reserve buffer space in the socket. Note that due to bugs in
346 * Linux's delayed-ack code, serious performance degredation may
347 * occur with linux hosts if the minimum is used.
349 if (so->so_type == SOCK_STREAM)
350 siz = NFS_MAXPACKET + sizeof (u_long);
353 if (siz < nfs_soreserve)
355 if (siz > sb_max_adj) {
356 kprintf("Warning: vfs.nfs.soreserve (%d) "
357 "limited to adjusted sb_max (%ld)\n",
358 nfs_soreserve, sb_max_adj);
362 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
365 FREE(mynam, M_SONAME);
370 * Set protocol specific options { for now TCP only } and
371 * reserve some space. For datagram sockets, this can get called
372 * repeatedly for the same socket, but that isn't harmful.
374 if (so->so_type == SOCK_STREAM) {
378 bzero(&sopt, sizeof sopt);
379 sopt.sopt_level = SOL_SOCKET;
380 sopt.sopt_name = SO_KEEPALIVE;
381 sopt.sopt_val = &val;
382 sopt.sopt_valsize = sizeof val;
386 if (so->so_proto->pr_domain->dom_family == AF_INET &&
387 so->so_proto->pr_protocol == IPPROTO_TCP) {
391 bzero(&sopt, sizeof sopt);
392 sopt.sopt_level = IPPROTO_TCP;
393 sopt.sopt_name = TCP_NODELAY;
394 sopt.sopt_val = &val;
395 sopt.sopt_valsize = sizeof val;
399 so->so_rcv.ssb_flags &= ~SSB_NOINTR;
400 so->so_rcv.ssb_timeo = 0;
401 so->so_snd.ssb_flags &= ~SSB_NOINTR;
402 so->so_snd.ssb_timeo = 0;
404 slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
405 M_NFSSVC, M_WAITOK | M_ZERO);
406 STAILQ_INIT(&slp->ns_rec);
407 TAILQ_INIT(&slp->ns_uidlruhead);
408 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
415 so->so_upcallarg = (caddr_t)slp;
416 so->so_upcall = nfsrv_rcv;
417 so->so_rcv.ssb_flags |= SSB_UPCALL;
418 slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
419 nfsrv_wakenfsd(slp, 1);
425 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
426 * until it is killed by a signal.
429 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
432 struct nfssvc_sock *slp;
433 struct nfsd *nfsd = nsd->nsd_nfsd;
434 struct nfsrv_descript *nd = NULL;
435 struct mbuf *m, *mreq;
436 int error = 0, cacherep, sotype, writes_todo;
444 if (nfsd == (struct nfsd *)0) {
445 nsd->nsd_nfsd = nfsd = (struct nfsd *)
446 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
449 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
455 * Loop getting rpc requests until SIGKILL.
458 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
459 while (nfsd->nfsd_slp == NULL &&
460 (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
461 nfsd->nfsd_flag |= NFSD_WAITING;
463 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
468 if (nfsd->nfsd_slp == NULL &&
469 (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
470 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
471 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
472 == (SLP_VALID | SLP_DOREC)) {
473 slp->ns_flag &= ~SLP_DOREC;
475 nfsd->nfsd_slp = slp;
480 nfsd_head_flag &= ~NFSD_CHECKSLP;
482 if ((slp = nfsd->nfsd_slp) == (struct nfssvc_sock *)0)
484 if (slp->ns_flag & SLP_VALID) {
485 if (slp->ns_flag & SLP_DISCONN)
487 else if (slp->ns_flag & SLP_NEEDQ) {
488 slp->ns_flag &= ~SLP_NEEDQ;
489 (void) nfs_slplock(slp, 1);
490 nfsrv_rcv(slp->ns_so, (caddr_t)slp,
494 error = nfsrv_dorec(slp, nfsd, &nd);
495 cur_usec = nfs_curusec();
496 if (error && slp->ns_tq.lh_first &&
497 slp->ns_tq.lh_first->nd_time <= cur_usec) {
503 nfsd->nfsd_flag |= NFSD_REQINPROG;
507 slp = nfsd->nfsd_slp;
509 if (error || (slp->ns_flag & SLP_VALID) == 0) {
511 kfree((caddr_t)nd, M_NFSRVDESC);
514 nfsd->nfsd_slp = (struct nfssvc_sock *)0;
515 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
520 sotype = slp->ns_so->so_type;
522 getmicrotime(&nd->nd_starttime);
524 nd->nd_nam = nd->nd_nam2;
526 nd->nd_nam = slp->ns_nam;
529 * Check to see if authorization is needed.
531 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
532 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
534 ((struct sockaddr_in *)
535 nd->nd_nam)->sin_addr.s_addr;
536 nsd->nsd_authlen = nfsd->nfsd_authlen;
537 nsd->nsd_verflen = nfsd->nfsd_verflen;
538 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
539 nfsd->nfsd_authlen) &&
540 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
541 nfsd->nfsd_verflen) &&
542 !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
544 cacherep = RC_DROPIT;
546 cacherep = nfsrv_getcache(nd, slp, &mreq);
549 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
550 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
551 nd->nd_procnum = NFSPROC_NOOP;
552 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
554 } else if (nfs_privport) {
555 /* Check if source port is privileged */
557 struct sockaddr *nam = nd->nd_nam;
558 struct sockaddr_in *sin;
560 sin = (struct sockaddr_in *)nam;
561 port = ntohs(sin->sin_port);
562 if (port >= IPPORT_RESERVED &&
563 nd->nd_procnum != NFSPROC_NULL) {
564 nd->nd_procnum = NFSPROC_NOOP;
565 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
567 kprintf("NFS request from unprivileged port (%s:%d)\n",
568 inet_ntoa(sin->sin_addr), port);
575 * Loop to get all the write rpc relies that have been
581 if (nd && (nd->nd_flag & ND_NFSV3))
582 procrastinate = nfsrvw_procrastinate_v3;
584 procrastinate = nfsrvw_procrastinate;
585 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
588 error = nfsrv_writegather(&nd, slp,
589 nfsd->nfsd_td, &mreq);
591 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
592 slp, nfsd->nfsd_td, &mreq);
596 if (error != 0 && error != NFSERR_RETVOID) {
597 if (nd->nd_procnum != NQNFSPROC_VACATED)
599 nfsrv_updatecache(nd, FALSE, mreq);
601 FREE(nd->nd_nam2, M_SONAME);
604 nfsstats.srvrpccnt[nd->nd_procnum]++;
605 nfsrv_updatecache(nd, TRUE, mreq);
606 nd->nd_mrep = (struct mbuf *)0;
614 if (siz <= 0 || siz > NFS_MAXPACKET) {
615 kprintf("mbuf siz=%d\n",siz);
616 panic("Bad nfs svc reply");
619 m->m_pkthdr.len = siz;
620 m->m_pkthdr.rcvif = (struct ifnet *)0;
622 * For stream protocols, prepend a Sun RPC
625 if (sotype == SOCK_STREAM) {
626 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
629 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
631 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
632 (void) nfs_slplock(slp, 1);
633 if (slp->ns_flag & SLP_VALID)
634 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
640 nfsd_rt(sotype, nd, cacherep);
642 FREE(nd->nd_nam2, M_SONAME);
644 m_freem(nd->nd_mrep);
647 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
649 if (error == EINTR || error == ERESTART) {
650 kfree((caddr_t)nd, M_NFSRVDESC);
658 nfsd_rt(sotype, nd, cacherep);
659 m_freem(nd->nd_mrep);
661 FREE(nd->nd_nam2, M_SONAME);
665 FREE((caddr_t)nd, M_NFSRVDESC);
670 * Check to see if there are outstanding writes that
671 * need to be serviced.
673 cur_usec = nfs_curusec();
675 if (slp->ns_tq.lh_first &&
676 slp->ns_tq.lh_first->nd_time <= cur_usec) {
682 } while (writes_todo);
684 if (nfsrv_dorec(slp, nfsd, &nd)) {
685 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
686 nfsd->nfsd_slp = NULL;
691 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
693 kfree((caddr_t)nfsd, M_NFSD);
694 nsd->nsd_nfsd = (struct nfsd *)0;
695 if (--nfs_numnfsd == 0)
696 nfsrv_init(TRUE); /* Reinitialize everything */
701 * Shut down a socket associated with an nfssvc_sock structure.
702 * Should be called with the send lock set, if required.
703 * The trick here is to increment the sref at the start, so that the nfsds
704 * will stop using it and clear ns_flag at the end so that it will not be
705 * reassigned during cleanup.
708 nfsrv_zapsock(struct nfssvc_sock *slp)
710 struct nfsuid *nuidp, *nnuidp;
711 struct nfsrv_descript *nwp, *nnwp;
714 struct nfsrv_rec *rec;
716 slp->ns_flag &= ~SLP_ALLFLAGS;
719 slp->ns_fp = (struct file *)0;
721 so->so_rcv.ssb_flags &= ~SSB_UPCALL;
722 so->so_upcall = NULL;
723 so->so_upcallarg = NULL;
724 soshutdown(so, SHUT_RDWR);
727 FREE(slp->ns_nam, M_SONAME);
728 m_freem(slp->ns_raw);
729 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
731 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
733 FREE(rec->nr_address, M_SONAME);
734 m_freem(rec->nr_packet);
735 kfree(rec, M_NFSRVDESC);
737 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
739 LIST_REMOVE(nuidp, nu_hash);
740 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
741 if (nuidp->nu_flag & NU_NAM)
742 FREE(nuidp->nu_nam, M_SONAME);
743 kfree((caddr_t)nuidp, M_NFSUID);
746 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
747 nnwp = nwp->nd_tq.le_next;
748 LIST_REMOVE(nwp, nd_tq);
749 kfree((caddr_t)nwp, M_NFSRVDESC);
751 LIST_INIT(&slp->ns_tq);
757 * Derefence a server socket structure. If it has no more references and
758 * is no longer valid, you can throw it away.
761 nfsrv_slpderef(struct nfssvc_sock *slp)
763 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
764 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
765 kfree((caddr_t)slp, M_NFSSVC);
770 * Lock a socket against others.
773 nfs_slplock(struct nfssvc_sock *slp, int wait)
775 int *statep = &slp->ns_solock;
777 if (!wait && (*statep & NFSSTA_SNDLOCK))
778 return(0); /* already locked, fail */
779 while (*statep & NFSSTA_SNDLOCK) {
780 *statep |= NFSSTA_WANTSND;
781 (void) tsleep((caddr_t)statep, 0, "nfsslplck", 0);
783 *statep |= NFSSTA_SNDLOCK;
788 * Unlock the stream socket for others.
791 nfs_slpunlock(struct nfssvc_sock *slp)
793 int *statep = &slp->ns_solock;
795 if ((*statep & NFSSTA_SNDLOCK) == 0)
796 panic("nfs slpunlock");
797 *statep &= ~NFSSTA_SNDLOCK;
798 if (*statep & NFSSTA_WANTSND) {
799 *statep &= ~NFSSTA_WANTSND;
800 wakeup((caddr_t)statep);
805 * Initialize the data structures for the server.
806 * Handshake with any new nfsds starting up to avoid any chance of
810 nfsrv_init(int terminating)
812 struct nfssvc_sock *slp, *nslp;
814 if (nfssvc_sockhead_flag & SLP_INIT)
816 nfssvc_sockhead_flag |= SLP_INIT;
818 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
819 if (slp->ns_flag & SLP_VALID)
821 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
822 kfree((caddr_t)slp, M_NFSSVC);
824 nfsrv_cleancache(); /* And clear out server cache */
826 nfs_pub.np_valid = 0;
828 TAILQ_INIT(&nfssvc_sockhead);
829 nfssvc_sockhead_flag &= ~SLP_INIT;
830 if (nfssvc_sockhead_flag & SLP_WANTINIT) {
831 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
832 wakeup((caddr_t)&nfssvc_sockhead);
835 TAILQ_INIT(&nfsd_head);
836 nfsd_head_flag &= ~NFSD_CHECKSLP;
839 nfs_udpsock = (struct nfssvc_sock *)
840 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
841 STAILQ_INIT(&nfs_udpsock->ns_rec);
842 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
843 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
845 nfs_cltpsock = (struct nfssvc_sock *)
846 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
847 STAILQ_INIT(&nfs_cltpsock->ns_rec);
848 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
849 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
854 * Add entries to the server monitor log.
857 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
861 rt = &nfsdrt.drt[nfsdrt.pos];
862 if (cacherep == RC_DOIT)
864 else if (cacherep == RC_REPLY)
865 rt->flag = DRT_CACHEREPLY;
867 rt->flag = DRT_CACHEDROP;
868 if (sotype == SOCK_STREAM)
870 if (nd->nd_flag & ND_NFSV3)
871 rt->flag |= DRT_NFSV3;
872 rt->proc = nd->nd_procnum;
873 if (nd->nd_nam->sa_family == AF_INET)
874 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
876 rt->ipadr = INADDR_ANY;
877 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
878 getmicrotime(&rt->tstamp);
879 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
881 #endif /* NFS_NOSERVER */
883 static int nfs_defect = 0;
884 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
887 * Asynchronous I/O daemons for client nfs.
888 * They do read-ahead and write-behind operations on the block I/O cache.
889 * Never returns unless it fails or gets killed.
892 nfssvc_iod(struct thread *td)
896 struct nfsmount *nmp;
900 * Assign my position or return error if too many already running
903 for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
904 if (nfs_asyncdaemon[i] == 0) {
905 nfs_asyncdaemon[i]++;
913 * Just loop around doin our stuff until SIGKILL
916 while (((nmp = nfs_iodmount[myiod]) == NULL
917 || TAILQ_EMPTY(&nmp->nm_bioq))
921 nfs_iodwant[myiod] = td;
922 nfs_iodmount[myiod] = NULL;
923 error = tsleep((caddr_t)&nfs_iodwant[myiod],
924 PCATCH, "nfsidl", 0);
927 nfs_asyncdaemon[myiod] = 0;
930 nfs_iodwant[myiod] = NULL;
931 nfs_iodmount[myiod] = NULL;
935 while ((bio = TAILQ_FIRST(&nmp->nm_bioq)) != NULL) {
937 * Take one off the front of the list. The BIO's
938 * block number is normalized for DEV_BSIZE.
940 TAILQ_REMOVE(&nmp->nm_bioq, bio, bio_act);
942 if (nmp->nm_bioqwant && nmp->nm_bioqlen <= nfs_numasync) {
943 nmp->nm_bioqwant = FALSE;
944 wakeup(&nmp->nm_bioq);
946 nfs_doio((struct vnode *)bio->bio_driver_info, bio, NULL);
948 * If there are more than one iod on this mount, then defect
949 * so that the iods can be shared out fairly between the mounts
951 if (nfs_defect && nmp->nm_bioqiods > 1) {
953 ("nfssvc_iod: iod %d defecting from mount %p\n",
955 nfs_iodmount[myiod] = NULL;
965 * Get an authorization string for the uid by having the mount_nfs sitting
966 * on this mount point porpous out of the kernel and do it.
969 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
970 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
971 int *verf_len, NFSKERBKEY_T key /* return session key */)
975 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
976 nmp->nm_state |= NFSSTA_WANTAUTH;
977 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
979 error = nfs_sigintr(nmp, rep, rep->r_td);
981 nmp->nm_state &= ~NFSSTA_WANTAUTH;
985 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
986 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
987 nmp->nm_authlen = RPCAUTH_MAXSIZ;
988 nmp->nm_verfstr = verf_str;
989 nmp->nm_verflen = *verf_len;
990 nmp->nm_authuid = cred->cr_uid;
991 wakeup((caddr_t)&nmp->nm_authstr);
994 * And wait for mount_nfs to do its stuff.
996 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
997 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
999 error = nfs_sigintr(nmp, rep, rep->r_td);
1001 if (nmp->nm_state & NFSSTA_AUTHERR) {
1002 nmp->nm_state &= ~NFSSTA_AUTHERR;
1006 kfree((caddr_t)*auth_str, M_TEMP);
1008 *auth_len = nmp->nm_authlen;
1009 *verf_len = nmp->nm_verflen;
1010 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
1012 nmp->nm_state &= ~NFSSTA_HASAUTH;
1013 nmp->nm_state |= NFSSTA_WAITAUTH;
1014 if (nmp->nm_state & NFSSTA_WANTAUTH) {
1015 nmp->nm_state &= ~NFSSTA_WANTAUTH;
1016 wakeup((caddr_t)&nmp->nm_authtype);
1022 * Get a nickname authenticator and verifier.
1025 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
1026 int *auth_len, char *verf_str, int verf_len)
1028 struct nfsuid *nuidp;
1029 u_int32_t *nickp, *verfp;
1030 struct timeval ktvin, ktvout;
1033 if (verf_len < (4 * NFSX_UNSIGNED))
1034 panic("nfs_getnickauth verf too small");
1036 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
1037 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1038 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
1041 if (!nuidp || nuidp->nu_expire < time_second)
1045 * Move to the end of the lru list (end of lru == most recently used).
1047 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1048 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1050 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1051 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1052 *nickp = txdr_unsigned(nuidp->nu_nickname);
1053 *auth_str = (char *)nickp;
1054 *auth_len = 2 * NFSX_UNSIGNED;
1057 * Now we must encrypt the verifier and package it up.
1059 verfp = (u_int32_t *)verf_str;
1060 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1061 if (time_second > nuidp->nu_timestamp.tv_sec ||
1062 (time_second == nuidp->nu_timestamp.tv_sec &&
1063 time_second > nuidp->nu_timestamp.tv_usec))
1064 getmicrotime(&nuidp->nu_timestamp);
1066 nuidp->nu_timestamp.tv_usec++;
1067 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
1068 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1071 * Now encrypt the timestamp verifier in ecb mode using the session
1078 *verfp++ = ktvout.tv_sec;
1079 *verfp++ = ktvout.tv_usec;
1085 * Save the current nickname in a hash list entry on the mount point.
1088 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1089 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1092 struct nfsuid *nuidp;
1095 struct mbuf *md = *mdp;
1096 struct timeval ktvin, ktvout;
1098 char *dpos = *dposp, *cp2;
1099 int deltasec, error = 0;
1101 if (len == (3 * NFSX_UNSIGNED)) {
1102 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1103 ktvin.tv_sec = *tl++;
1104 ktvin.tv_usec = *tl++;
1105 nick = fxdr_unsigned(u_int32_t, *tl);
1108 * Decrypt the timestamp in ecb mode.
1113 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1114 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1115 deltasec = time_second - ktvout.tv_sec;
1117 deltasec = -deltasec;
1119 * If ok, add it to the hash list for the mount point.
1121 if (deltasec <= NFS_KERBCLOCKSKEW) {
1122 if (nmp->nm_numuids < nuidhash_max) {
1124 nuidp = (struct nfsuid *)
1125 kmalloc(sizeof (struct nfsuid), M_NFSUID,
1128 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1129 LIST_REMOVE(nuidp, nu_hash);
1130 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1134 nuidp->nu_cr.cr_uid = cred->cr_uid;
1135 nuidp->nu_expire = time_second + NFS_KERBTTL;
1136 nuidp->nu_timestamp = ktvout;
1137 nuidp->nu_nickname = nick;
1138 bcopy(key, nuidp->nu_key, sizeof (key));
1139 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1141 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1145 nfsm_adv(nfsm_rndup(len));