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 <sys/mutex.h>
62 #include <vm/vm_zone.h>
64 #include <sys/mutex2.h>
66 #include <netinet/in.h>
67 #include <netinet/tcp.h>
72 #include "nfsm_subs.h"
73 #include "nfsrvcache.h"
78 #include <sys/thread2.h>
80 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
83 extern int32_t (*nfsrv3_procs[NFS_NPROCS]) (struct nfsrv_descript *nd,
84 struct nfssvc_sock *slp,
87 extern int nfs_numasync;
89 extern struct nfsstats nfsstats;
90 extern int nfsrvw_procrastinate;
91 extern int nfsrvw_procrastinate_v3;
92 static int nuidhash_max = NFS_MAXUIDHASH;
95 static void nfsrv_zapsock (struct nfssvc_sock *slp);
97 static int nfssvc_iod (struct thread *);
102 static int nfs_asyncdaemon[NFS_MAXASYNCDAEMON];
104 SYSCTL_DECL(_vfs_nfs);
107 int nfsd_waiting = 0;
108 static struct nfsdrt nfsdrt;
109 static int nfs_numnfsd = 0;
110 static void nfsd_rt (int sotype, struct nfsrv_descript *nd,
112 static int nfssvc_addsock (struct file *, struct sockaddr *,
114 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
116 static int nfs_privport = 0;
117 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
118 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
119 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
120 static int nfs_soreserve = 65535;
121 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
124 * NFS server system calls
127 #endif /* NFS_NOSERVER */
129 * nfssvc_args(int flag, caddr_t argp)
131 * Nfs server psuedo system call for the nfsd's
132 * Based on the flag value it either:
133 * - adds a socket to the selection list
134 * - remains in the kernel as an nfsd
135 * - remains in the kernel as an nfsiod
138 sys_nfssvc(struct nfssvc_args *uap)
141 struct nlookupdata nd;
143 struct sockaddr *nam;
144 struct nfsd_args nfsdarg;
145 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
146 struct nfsd_cargs ncd;
148 struct nfssvc_sock *slp;
149 struct nfsuid *nuidp;
150 struct nfsmount *nmp;
152 #endif /* NFS_NOSERVER */
154 struct thread *td = curthread;
159 error = priv_check(td, PRIV_ROOT);
162 KKASSERT(td->td_proc); /* for ucred and p_fd */
163 while (nfssvc_sockhead_flag & SLP_INIT) {
164 nfssvc_sockhead_flag |= SLP_WANTINIT;
165 (void) tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
167 if (uap->flag & NFSSVC_BIOD)
168 error = nfssvc_iod(td);
172 #else /* !NFS_NOSERVER */
173 else if (uap->flag & NFSSVC_MNTD) {
174 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
178 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE,
181 error = nlookup(&nd);
183 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
188 if ((vp->v_flag & VROOT) == 0)
190 nmp = VFSTONFS(vp->v_mount);
194 if ((nmp->nm_state & NFSSTA_MNTD) &&
195 (uap->flag & NFSSVC_GOTAUTH) == 0)
197 nmp->nm_state |= NFSSTA_MNTD;
198 error = nfs_clientd(nmp, td->td_proc->p_ucred, &ncd, uap->flag,
200 } else if (uap->flag & NFSSVC_ADDSOCK) {
201 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
204 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
208 * Get the client address for connected sockets.
210 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
213 error = getsockaddr(&nam, nfsdarg.name,
220 error = nfssvc_addsock(fp, nam, td);
223 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
226 if ((uap->flag & NFSSVC_AUTHIN) &&
227 ((nfsd = nsd->nsd_nfsd)) != NULL &&
228 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
229 slp = nfsd->nfsd_slp;
232 * First check to see if another nfsd has already
233 * added this credential.
235 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
236 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
237 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
238 (!nfsd->nfsd_nd->nd_nam2 ||
239 netaddr_match(NU_NETFAM(nuidp),
240 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
244 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
245 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
250 if (slp->ns_numuids < nuidhash_max) {
252 nuidp = (struct nfsuid *)
253 kmalloc(sizeof (struct nfsuid), M_NFSUID,
257 if ((slp->ns_flag & SLP_VALID) == 0) {
259 kfree((caddr_t)nuidp, M_NFSUID);
262 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
263 LIST_REMOVE(nuidp, nu_hash);
264 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
266 if (nuidp->nu_flag & NU_NAM)
267 FREE(nuidp->nu_nam, M_SONAME);
270 nuidp->nu_cr = nsd->nsd_cr;
271 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
272 nuidp->nu_cr.cr_ngroups = NGROUPS;
273 nuidp->nu_cr.cr_ref = 1;
274 nuidp->nu_timestamp = nsd->nsd_timestamp;
275 nuidp->nu_expire = time_second + nsd->nsd_ttl;
277 * and save the session key in nu_key.
279 bcopy(nsd->nsd_key, nuidp->nu_key,
280 sizeof (nsd->nsd_key));
281 if (nfsd->nfsd_nd->nd_nam2) {
282 struct sockaddr_in *saddr;
284 saddr = (struct sockaddr_in *)
285 nfsd->nfsd_nd->nd_nam2;
286 switch (saddr->sin_family) {
288 nuidp->nu_flag |= NU_INETADDR;
290 saddr->sin_addr.s_addr;
294 nuidp->nu_flag |= NU_NAM;
296 dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
300 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
302 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
304 nfsrv_setcred(&nuidp->nu_cr,
305 &nfsd->nfsd_nd->nd_cr);
306 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
310 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
311 nfsd->nfsd_flag |= NFSD_AUTHFAIL;
312 error = nfssvc_nfsd(nsd, uap->argp, td);
314 #endif /* NFS_NOSERVER */
315 if (error == EINTR || error == ERESTART)
322 * Adds a socket to the list for servicing by nfsds.
325 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
328 struct nfssvc_sock *slp;
332 so = (struct socket *)fp->f_data;
336 * Add it to the list, as required.
338 if (so->so_proto->pr_protocol == IPPROTO_UDP) {
340 if (tslp->ns_flag & SLP_VALID) {
342 FREE(mynam, M_SONAME);
348 * Reserve buffer space in the socket. Note that due to bugs in
349 * Linux's delayed-ack code, serious performance degredation may
350 * occur with linux hosts if the minimum is used.
352 if (so->so_type == SOCK_STREAM)
353 siz = NFS_MAXPACKET + sizeof (u_long);
356 if (siz < nfs_soreserve)
358 if (siz > sb_max_adj) {
359 kprintf("Warning: vfs.nfs.soreserve (%d) "
360 "limited to adjusted sb_max (%ld)\n",
361 nfs_soreserve, sb_max_adj);
365 error = soreserve(so, siz, siz, &td->td_proc->p_rlimit[RLIMIT_SBSIZE]);
368 FREE(mynam, M_SONAME);
373 * Set protocol specific options { for now TCP only } and
374 * reserve some space. For datagram sockets, this can get called
375 * repeatedly for the same socket, but that isn't harmful.
377 if (so->so_type == SOCK_STREAM) {
381 bzero(&sopt, sizeof sopt);
382 sopt.sopt_level = SOL_SOCKET;
383 sopt.sopt_name = SO_KEEPALIVE;
384 sopt.sopt_val = &val;
385 sopt.sopt_valsize = sizeof val;
389 if (so->so_proto->pr_domain->dom_family == AF_INET &&
390 so->so_proto->pr_protocol == IPPROTO_TCP) {
394 bzero(&sopt, sizeof sopt);
395 sopt.sopt_level = IPPROTO_TCP;
396 sopt.sopt_name = TCP_NODELAY;
397 sopt.sopt_val = &val;
398 sopt.sopt_valsize = sizeof val;
402 so->so_rcv.ssb_flags &= ~SSB_NOINTR;
403 so->so_rcv.ssb_timeo = 0;
404 so->so_snd.ssb_flags &= ~SSB_NOINTR;
405 so->so_snd.ssb_timeo = 0;
407 slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
408 M_NFSSVC, M_WAITOK | M_ZERO);
409 mtx_init(&slp->ns_solock);
410 STAILQ_INIT(&slp->ns_rec);
411 TAILQ_INIT(&slp->ns_uidlruhead);
412 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
419 so->so_upcallarg = (caddr_t)slp;
420 so->so_upcall = nfsrv_rcv;
421 so->so_rcv.ssb_flags |= SSB_UPCALL;
422 slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
423 nfsrv_wakenfsd(slp, 1);
429 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
430 * until it is killed by a signal.
433 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
436 struct nfssvc_sock *slp;
437 struct nfsd *nfsd = nsd->nsd_nfsd;
438 struct nfsrv_descript *nd = NULL;
439 struct mbuf *m, *mreq;
440 int error = 0, cacherep, sotype, writes_todo;
449 nsd->nsd_nfsd = nfsd = (struct nfsd *)
450 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
453 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
459 * Loop getting rpc requests until SIGKILL.
462 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
463 while (nfsd->nfsd_slp == NULL &&
464 (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
465 nfsd->nfsd_flag |= NFSD_WAITING;
467 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
472 if (nfsd->nfsd_slp == NULL &&
473 (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
474 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
475 if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
476 == (SLP_VALID | SLP_DOREC)) {
477 slp->ns_flag &= ~SLP_DOREC;
479 nfsd->nfsd_slp = slp;
484 nfsd_head_flag &= ~NFSD_CHECKSLP;
486 if ((slp = nfsd->nfsd_slp) == NULL)
488 if (slp->ns_flag & SLP_VALID) {
489 if (slp->ns_flag & SLP_DISCONN)
491 else if (slp->ns_flag & SLP_NEEDQ) {
492 slp->ns_flag &= ~SLP_NEEDQ;
493 (void) nfs_slplock(slp, 1);
494 nfsrv_rcv(slp->ns_so, (caddr_t)slp,
498 error = nfsrv_dorec(slp, nfsd, &nd);
499 cur_usec = nfs_curusec();
500 if (error && slp->ns_tq.lh_first &&
501 slp->ns_tq.lh_first->nd_time <= cur_usec) {
507 nfsd->nfsd_flag |= NFSD_REQINPROG;
511 slp = nfsd->nfsd_slp;
513 if (error || (slp->ns_flag & SLP_VALID) == 0) {
515 kfree((caddr_t)nd, M_NFSRVDESC);
518 nfsd->nfsd_slp = NULL;
519 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
524 sotype = slp->ns_so->so_type;
526 getmicrotime(&nd->nd_starttime);
528 nd->nd_nam = nd->nd_nam2;
530 nd->nd_nam = slp->ns_nam;
533 * Check to see if authorization is needed.
535 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
536 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
538 ((struct sockaddr_in *)
539 nd->nd_nam)->sin_addr.s_addr;
540 nsd->nsd_authlen = nfsd->nfsd_authlen;
541 nsd->nsd_verflen = nfsd->nfsd_verflen;
542 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
543 nfsd->nfsd_authlen) &&
544 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
545 nfsd->nfsd_verflen) &&
546 !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
548 cacherep = RC_DROPIT;
550 cacherep = nfsrv_getcache(nd, slp, &mreq);
553 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
554 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
555 nd->nd_procnum = NFSPROC_NOOP;
556 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
558 } else if (nfs_privport) {
559 /* Check if source port is privileged */
561 struct sockaddr *nam = nd->nd_nam;
562 struct sockaddr_in *sin;
564 sin = (struct sockaddr_in *)nam;
565 port = ntohs(sin->sin_port);
566 if (port >= IPPORT_RESERVED &&
567 nd->nd_procnum != NFSPROC_NULL) {
568 nd->nd_procnum = NFSPROC_NOOP;
569 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
571 kprintf("NFS request from unprivileged port (%s:%d)\n",
572 inet_ntoa(sin->sin_addr), port);
579 * Loop to get all the write rpc relies that have been
585 if (nd && (nd->nd_flag & ND_NFSV3))
586 procrastinate = nfsrvw_procrastinate_v3;
588 procrastinate = nfsrvw_procrastinate;
589 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
592 error = nfsrv_writegather(&nd, slp,
593 nfsd->nfsd_td, &mreq);
595 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
596 slp, nfsd->nfsd_td, &mreq);
600 if (error != 0 && error != NFSERR_RETVOID) {
601 if (nd->nd_procnum != NQNFSPROC_VACATED)
603 nfsrv_updatecache(nd, FALSE, mreq);
605 FREE(nd->nd_nam2, M_SONAME);
608 nfsstats.srvrpccnt[nd->nd_procnum]++;
609 nfsrv_updatecache(nd, TRUE, mreq);
618 if (siz <= 0 || siz > NFS_MAXPACKET) {
619 kprintf("mbuf siz=%d\n",siz);
620 panic("Bad nfs svc reply");
623 m->m_pkthdr.len = siz;
624 m->m_pkthdr.rcvif = NULL;
626 * For stream protocols, prepend a Sun RPC
629 if (sotype == SOCK_STREAM) {
630 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
633 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
635 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
636 (void) nfs_slplock(slp, 1);
637 if (slp->ns_flag & SLP_VALID)
638 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
644 nfsd_rt(sotype, nd, cacherep);
646 FREE(nd->nd_nam2, M_SONAME);
648 m_freem(nd->nd_mrep);
651 if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
653 if (error == EINTR || error == ERESTART) {
654 kfree((caddr_t)nd, M_NFSRVDESC);
662 nfsd_rt(sotype, nd, cacherep);
663 m_freem(nd->nd_mrep);
665 FREE(nd->nd_nam2, M_SONAME);
669 FREE((caddr_t)nd, M_NFSRVDESC);
674 * Check to see if there are outstanding writes that
675 * need to be serviced.
677 cur_usec = nfs_curusec();
679 if (slp->ns_tq.lh_first &&
680 slp->ns_tq.lh_first->nd_time <= cur_usec) {
686 } while (writes_todo);
688 if (nfsrv_dorec(slp, nfsd, &nd)) {
689 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
690 nfsd->nfsd_slp = NULL;
695 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
697 kfree((caddr_t)nfsd, M_NFSD);
698 nsd->nsd_nfsd = NULL;
699 if (--nfs_numnfsd == 0)
700 nfsrv_init(TRUE); /* Reinitialize everything */
705 * Shut down a socket associated with an nfssvc_sock structure.
706 * Should be called with the send lock set, if required.
707 * The trick here is to increment the sref at the start, so that the nfsds
708 * will stop using it and clear ns_flag at the end so that it will not be
709 * reassigned during cleanup.
712 nfsrv_zapsock(struct nfssvc_sock *slp)
714 struct nfsuid *nuidp, *nnuidp;
715 struct nfsrv_descript *nwp, *nnwp;
718 struct nfsrv_rec *rec;
720 slp->ns_flag &= ~SLP_ALLFLAGS;
725 so->so_rcv.ssb_flags &= ~SSB_UPCALL;
726 so->so_upcall = NULL;
727 so->so_upcallarg = NULL;
728 soshutdown(so, SHUT_RDWR);
731 FREE(slp->ns_nam, M_SONAME);
732 m_freem(slp->ns_raw);
733 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
735 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
737 FREE(rec->nr_address, M_SONAME);
738 m_freem(rec->nr_packet);
739 kfree(rec, M_NFSRVDESC);
741 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
743 LIST_REMOVE(nuidp, nu_hash);
744 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
745 if (nuidp->nu_flag & NU_NAM)
746 FREE(nuidp->nu_nam, M_SONAME);
747 kfree((caddr_t)nuidp, M_NFSUID);
750 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
751 nnwp = nwp->nd_tq.le_next;
752 LIST_REMOVE(nwp, nd_tq);
753 kfree((caddr_t)nwp, M_NFSRVDESC);
755 LIST_INIT(&slp->ns_tq);
761 * Derefence a server socket structure. If it has no more references and
762 * is no longer valid, you can throw it away.
765 nfsrv_slpderef(struct nfssvc_sock *slp)
767 if (--(slp->ns_sref) == 0 && (slp->ns_flag & SLP_VALID) == 0) {
768 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
769 kfree((caddr_t)slp, M_NFSSVC);
774 * Lock a socket against others.
776 * Returns 0 on failure, 1 on success.
779 nfs_slplock(struct nfssvc_sock *slp, int wait)
781 mtx_t mtx = &slp->ns_solock;
784 mtx_lock_ex(mtx, "nfsslplck", 0, 0);
786 } else if (mtx_lock_ex_try(mtx) == 0) {
794 * Unlock the stream socket for others.
797 nfs_slpunlock(struct nfssvc_sock *slp)
799 mtx_t mtx = &slp->ns_solock;
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 mtx_init(&nfs_udpsock->ns_solock);
842 STAILQ_INIT(&nfs_udpsock->ns_rec);
843 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
844 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
846 nfs_cltpsock = (struct nfssvc_sock *)
847 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
848 mtx_init(&nfs_cltpsock->ns_solock);
849 STAILQ_INIT(&nfs_cltpsock->ns_rec);
850 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
851 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
856 * Add entries to the server monitor log.
859 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
863 rt = &nfsdrt.drt[nfsdrt.pos];
864 if (cacherep == RC_DOIT)
866 else if (cacherep == RC_REPLY)
867 rt->flag = DRT_CACHEREPLY;
869 rt->flag = DRT_CACHEDROP;
870 if (sotype == SOCK_STREAM)
872 if (nd->nd_flag & ND_NFSV3)
873 rt->flag |= DRT_NFSV3;
874 rt->proc = nd->nd_procnum;
875 if (nd->nd_nam->sa_family == AF_INET)
876 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
878 rt->ipadr = INADDR_ANY;
879 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
880 getmicrotime(&rt->tstamp);
881 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
883 #endif /* NFS_NOSERVER */
885 static int nfs_defect = 0;
886 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
889 * Asynchronous I/O daemons for client nfs.
890 * They do read-ahead and write-behind operations on the block I/O cache.
891 * Never returns unless it fails or gets killed.
894 nfssvc_iod(struct thread *td)
898 struct nfsmount *nmp;
902 * Assign my position or return error if too many already running
905 for (i = 0; i < NFS_MAXASYNCDAEMON; i++)
906 if (nfs_asyncdaemon[i] == 0) {
907 nfs_asyncdaemon[i]++;
915 * Just loop around doin our stuff until SIGKILL
918 while (((nmp = nfs_iodmount[myiod]) == NULL
919 || TAILQ_EMPTY(&nmp->nm_bioq))
923 nfs_iodwant[myiod] = td;
924 nfs_iodmount[myiod] = NULL;
925 error = tsleep((caddr_t)&nfs_iodwant[myiod],
926 PCATCH, "nfsidl", 0);
929 nfs_asyncdaemon[myiod] = 0;
932 nfs_iodwant[myiod] = NULL;
933 nfs_iodmount[myiod] = NULL;
937 while ((bio = TAILQ_FIRST(&nmp->nm_bioq)) != NULL) {
939 * Take one off the front of the list. The BIO's
940 * block number is normalized for DEV_BSIZE.
942 TAILQ_REMOVE(&nmp->nm_bioq, bio, bio_act);
944 if (nmp->nm_bioqwant && nmp->nm_bioqlen <= nfs_numasync) {
945 nmp->nm_bioqwant = FALSE;
946 wakeup(&nmp->nm_bioq);
948 nfs_doio((struct vnode *)bio->bio_driver_info, bio, NULL);
951 * If there are more than one iod on this mount, then defect
952 * so that the iods can be shared out fairly between the mounts
954 if (nfs_defect && nmp->nm_bioqiods > 1) {
956 ("nfssvc_iod: iod %d defecting from mount %p\n",
958 nfs_iodmount[myiod] = NULL;
968 * Get an authorization string for the uid by having the mount_nfs sitting
969 * on this mount point porpous out of the kernel and do it.
972 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
973 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
974 int *verf_len, NFSKERBKEY_T key /* return session key */)
978 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
979 nmp->nm_state |= NFSSTA_WANTAUTH;
980 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
982 error = nfs_sigintr(nmp, rep, rep->r_td);
984 nmp->nm_state &= ~NFSSTA_WANTAUTH;
988 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
989 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
990 nmp->nm_authlen = RPCAUTH_MAXSIZ;
991 nmp->nm_verfstr = verf_str;
992 nmp->nm_verflen = *verf_len;
993 nmp->nm_authuid = cred->cr_uid;
994 wakeup((caddr_t)&nmp->nm_authstr);
997 * And wait for mount_nfs to do its stuff.
999 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
1000 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
1001 "nfsauth2", 2 * hz);
1002 error = nfs_sigintr(nmp, rep, rep->r_td);
1004 if (nmp->nm_state & NFSSTA_AUTHERR) {
1005 nmp->nm_state &= ~NFSSTA_AUTHERR;
1009 kfree((caddr_t)*auth_str, M_TEMP);
1011 *auth_len = nmp->nm_authlen;
1012 *verf_len = nmp->nm_verflen;
1013 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
1015 nmp->nm_state &= ~NFSSTA_HASAUTH;
1016 nmp->nm_state |= NFSSTA_WAITAUTH;
1017 if (nmp->nm_state & NFSSTA_WANTAUTH) {
1018 nmp->nm_state &= ~NFSSTA_WANTAUTH;
1019 wakeup((caddr_t)&nmp->nm_authtype);
1025 * Get a nickname authenticator and verifier.
1028 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
1029 int *auth_len, char *verf_str, int verf_len)
1031 struct nfsuid *nuidp;
1032 u_int32_t *nickp, *verfp;
1033 struct timeval ktvin, ktvout;
1036 if (verf_len < (4 * NFSX_UNSIGNED))
1037 panic("nfs_getnickauth verf too small");
1039 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
1040 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1041 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
1044 if (!nuidp || nuidp->nu_expire < time_second)
1048 * Move to the end of the lru list (end of lru == most recently used).
1050 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1051 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1053 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1054 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1055 *nickp = txdr_unsigned(nuidp->nu_nickname);
1056 *auth_str = (char *)nickp;
1057 *auth_len = 2 * NFSX_UNSIGNED;
1060 * Now we must encrypt the verifier and package it up.
1062 verfp = (u_int32_t *)verf_str;
1063 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1064 if (time_second > nuidp->nu_timestamp.tv_sec ||
1065 (time_second == nuidp->nu_timestamp.tv_sec &&
1066 time_second > nuidp->nu_timestamp.tv_usec))
1067 getmicrotime(&nuidp->nu_timestamp);
1069 nuidp->nu_timestamp.tv_usec++;
1070 ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
1071 ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1074 * Now encrypt the timestamp verifier in ecb mode using the session
1081 *verfp++ = ktvout.tv_sec;
1082 *verfp++ = ktvout.tv_usec;
1088 * Save the current nickname in a hash list entry on the mount point.
1091 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1092 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1095 struct nfsuid *nuidp;
1098 struct mbuf *md = *mdp;
1099 struct timeval ktvin, ktvout;
1101 char *dpos = *dposp, *cp2;
1102 int deltasec, error = 0;
1104 if (len == (3 * NFSX_UNSIGNED)) {
1105 nfsm_dissect(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
1106 ktvin.tv_sec = *tl++;
1107 ktvin.tv_usec = *tl++;
1108 nick = fxdr_unsigned(u_int32_t, *tl);
1111 * Decrypt the timestamp in ecb mode.
1116 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1117 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1118 deltasec = time_second - ktvout.tv_sec;
1120 deltasec = -deltasec;
1122 * If ok, add it to the hash list for the mount point.
1124 if (deltasec <= NFS_KERBCLOCKSKEW) {
1125 if (nmp->nm_numuids < nuidhash_max) {
1127 nuidp = (struct nfsuid *)
1128 kmalloc(sizeof (struct nfsuid), M_NFSUID,
1131 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1132 LIST_REMOVE(nuidp, nu_hash);
1133 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1137 nuidp->nu_cr.cr_uid = cred->cr_uid;
1138 nuidp->nu_expire = time_second + NFS_KERBTTL;
1139 nuidp->nu_timestamp = ktvout;
1140 nuidp->nu_nickname = nick;
1141 bcopy(key, nuidp->nu_key, sizeof (key));
1142 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1144 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1148 nfsm_adv(nfsm_rndup(len));