kernel - Fix races created by a comedy of circumstansces (3)
[dragonfly.git] / sys / vfs / nfs / nfs_syscalls.c
blobac4936aaded54c1ea6a339abc0303e57a3224782
1 /*
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
10 * are met:
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. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30 * SUCH DAMAGE.
32 * @(#)nfs_syscalls.c 8.5 (Berkeley) 3/30/95
33 * $FreeBSD: src/sys/nfs/nfs_syscalls.c,v 1.58.2.1 2000/11/26 02:30:06 dillon Exp $
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/sysproto.h>
39 #include <sys/kernel.h>
40 #include <sys/sysctl.h>
41 #include <sys/file.h>
42 #include <sys/filedesc.h>
43 #include <sys/vnode.h>
44 #include <sys/malloc.h>
45 #include <sys/mount.h>
46 #include <sys/proc.h>
47 #include <sys/priv.h>
48 #include <sys/buf.h>
49 #include <sys/mbuf.h>
50 #include <sys/resourcevar.h>
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/domain.h>
54 #include <sys/protosw.h>
55 #include <sys/nlookup.h>
57 #include <sys/mutex2.h>
58 #include <sys/thread2.h>
60 #include <netinet/in.h>
61 #include <netinet/tcp.h>
62 #include "xdr_subs.h"
63 #include "rpcv2.h"
64 #include "nfsproto.h"
65 #include "nfs.h"
66 #include "nfsm_subs.h"
67 #include "nfsrvcache.h"
68 #include "nfsmount.h"
69 #include "nfsnode.h"
70 #include "nfsrtt.h"
72 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
74 static int nuidhash_max = NFS_MAXUIDHASH;
76 #ifndef NFS_NOSERVER
77 static void nfsrv_zapsock (struct nfssvc_sock *slp);
78 #endif
80 #define TRUE 1
81 #define FALSE 0
83 SYSCTL_DECL(_vfs_nfs);
85 #ifndef NFS_NOSERVER
86 int nfsd_waiting = 0;
87 static struct nfsdrt nfsdrt;
88 static int nfs_numnfsd = 0;
89 static void nfsd_rt (int sotype, struct nfsrv_descript *nd,
90 int cacherep);
91 static int nfssvc_addsock (struct file *, struct sockaddr *,
92 struct thread *);
93 static int nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
95 static int nfs_privport = 0;
96 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport,
97 0, "Enable privileged source port checks");
98 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0,
99 "Enable NFS request procrastination");
100 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0,
101 "Enable NFSv3 request procrastination");
102 int nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO;
103 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0,
104 "Minimum NFS socket buffer size reservation");
107 * NFS server system calls
110 #endif /* NFS_NOSERVER */
112 * nfssvc_args(int flag, caddr_t argp)
114 * Nfs server psuedo system call for the nfsd's
115 * Based on the flag value it either:
116 * - adds a socket to the selection list
117 * - remains in the kernel as an nfsd
118 * - remains in the kernel as an nfsiod
120 * MPALMOSTSAFE
123 sys_nfssvc(struct nfssvc_args *uap)
125 #ifndef NFS_NOSERVER
126 struct nlookupdata nd;
127 struct file *fp;
128 struct sockaddr *nam;
129 struct nfsd_args nfsdarg;
130 struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
131 struct nfsd_cargs ncd;
132 struct nfsd *nfsd;
133 struct nfssvc_sock *slp;
134 struct nfsuid *nuidp;
135 struct nfsmount *nmp;
136 struct vnode *vp;
137 #endif /* NFS_NOSERVER */
138 int error;
139 struct thread *td = curthread;
142 * Must be super user
144 error = priv_check(td, PRIV_ROOT);
145 if (error)
146 return (error);
148 lwkt_gettoken(&nfs_token);
150 while (nfssvc_sockhead_flag & SLP_INIT) {
151 nfssvc_sockhead_flag |= SLP_WANTINIT;
152 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
154 if (uap->flag & NFSSVC_BIOD)
155 error = ENXIO; /* no longer need nfsiod's */
156 #ifdef NFS_NOSERVER
157 else
158 error = ENXIO;
159 #else /* !NFS_NOSERVER */
160 else if (uap->flag & NFSSVC_MNTD) {
161 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
162 if (error)
163 goto done;
164 vp = NULL;
165 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE,
166 NLC_FOLLOW);
167 if (error == 0)
168 error = nlookup(&nd);
169 if (error == 0)
170 error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);
171 nlookup_done(&nd);
172 if (error)
173 goto done;
175 if ((vp->v_flag & VROOT) == 0)
176 error = EINVAL;
177 nmp = VFSTONFS(vp->v_mount);
178 vput(vp);
179 if (error)
180 goto done;
181 if ((nmp->nm_state & NFSSTA_MNTD) &&
182 (uap->flag & NFSSVC_GOTAUTH) == 0) {
183 error = 0;
184 goto done;
186 nmp->nm_state |= NFSSTA_MNTD;
187 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag,
188 uap->argp, td);
189 } else if (uap->flag & NFSSVC_ADDSOCK) {
190 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
191 if (error)
192 goto done;
193 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
194 if (error)
195 goto done;
197 * Get the client address for connected sockets.
199 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
200 nam = NULL;
201 else {
202 error = getsockaddr(&nam, nfsdarg.name,
203 nfsdarg.namelen);
204 if (error) {
205 fdrop(fp);
206 goto done;
209 error = nfssvc_addsock(fp, nam, td);
210 fdrop(fp);
211 } else {
212 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
213 if (error)
214 goto done;
215 if ((uap->flag & NFSSVC_AUTHIN) &&
216 ((nfsd = nsd->nsd_nfsd)) != NULL &&
217 (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
218 slp = nfsd->nfsd_slp;
221 * First check to see if another nfsd has already
222 * added this credential.
224 for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
225 nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
226 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
227 (!nfsd->nfsd_nd->nd_nam2 ||
228 netaddr_match(AF_INET,
229 &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
230 break;
232 if (nuidp) {
233 nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
234 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
235 } else {
237 * Nope, so we will.
239 if (slp->ns_numuids < nuidhash_max) {
240 slp->ns_numuids++;
241 nuidp = (struct nfsuid *)
242 kmalloc(sizeof (struct nfsuid), M_NFSUID,
243 M_WAITOK);
244 } else
245 nuidp = NULL;
246 if ((slp->ns_flag & SLP_VALID) == 0) {
247 if (nuidp)
248 kfree((caddr_t)nuidp, M_NFSUID);
249 } else {
250 if (nuidp == NULL) {
251 nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
252 LIST_REMOVE(nuidp, nu_hash);
253 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
254 nu_lru);
255 if (nuidp->nu_flag & NU_NAM)
256 kfree(nuidp->nu_nam, M_SONAME);
258 nuidp->nu_flag = 0;
259 nuidp->nu_cr = nsd->nsd_cr;
260 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
261 nuidp->nu_cr.cr_ngroups = NGROUPS;
262 nuidp->nu_cr.cr_ref = 1;
263 nuidp->nu_timestamp = nsd->nsd_timestamp;
264 nuidp->nu_expire = time_uptime + nsd->nsd_ttl;
266 * and save the session key in nu_key.
268 bcopy(nsd->nsd_key, nuidp->nu_key,
269 sizeof (nsd->nsd_key));
270 if (nfsd->nfsd_nd->nd_nam2) {
271 struct sockaddr_in *saddr;
273 saddr = (struct sockaddr_in *)
274 nfsd->nfsd_nd->nd_nam2;
275 switch (saddr->sin_family) {
276 case AF_INET:
277 nuidp->nu_flag |= NU_INETADDR;
278 nuidp->nu_inetaddr =
279 saddr->sin_addr.s_addr;
280 break;
281 default:
282 nuidp->nu_flag |= NU_NAM;
283 nuidp->nu_nam =
284 dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
285 break;
288 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
289 nu_lru);
290 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
291 nuidp, nu_hash);
292 nfsrv_setcred(&nuidp->nu_cr,
293 &nfsd->nfsd_nd->nd_cr);
294 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
298 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
299 nfsd->nfsd_flag |= NFSD_AUTHFAIL;
300 error = nfssvc_nfsd(nsd, uap->argp, td);
302 #endif /* NFS_NOSERVER */
303 if (error == EINTR || error == ERESTART)
304 error = 0;
305 done:
306 lwkt_reltoken(&nfs_token);
307 return (error);
310 #ifndef NFS_NOSERVER
312 * Adds a socket to the list for servicing by nfsds.
314 static int
315 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
317 int siz;
318 struct nfssvc_sock *slp;
319 struct socket *so;
320 int error;
322 so = (struct socket *)fp->f_data;
323 #if 0
324 tslp = NULL;
326 * Add it to the list, as required.
328 if (so->so_proto->pr_protocol == IPPROTO_UDP) {
329 tslp = nfs_udpsock;
330 if (tslp->ns_flag & SLP_VALID) {
331 if (mynam != NULL)
332 kfree(mynam, M_SONAME);
333 return (EPERM);
336 #endif
338 * Reserve buffer space in the socket. Note that due to bugs in
339 * Linux's delayed-ack code, serious performance degredation may
340 * occur with linux hosts if the minimum is used.
342 * NFS sockets are not limited to the standard sb_max or by
343 * resource limits.
345 if (so->so_type == SOCK_STREAM)
346 siz = NFS_MAXPACKET + sizeof (u_long);
347 else
348 siz = NFS_MAXPACKET;
349 if (siz < nfs_soreserve)
350 siz = nfs_soreserve;
352 error = soreserve(so, siz, siz, NULL);
353 if (error) {
354 if (mynam != NULL)
355 kfree(mynam, M_SONAME);
356 return (error);
360 * Set protocol specific options { for now TCP only } and
361 * reserve some space. For datagram sockets, this can get called
362 * repeatedly for the same socket, but that isn't harmful.
364 if (so->so_type == SOCK_STREAM) {
365 struct sockopt sopt;
366 int val;
368 bzero(&sopt, sizeof sopt);
369 sopt.sopt_level = SOL_SOCKET;
370 sopt.sopt_name = SO_KEEPALIVE;
371 sopt.sopt_val = &val;
372 sopt.sopt_valsize = sizeof val;
373 val = 1;
374 sosetopt(so, &sopt);
376 if (so->so_proto->pr_domain->dom_family == AF_INET &&
377 so->so_proto->pr_protocol == IPPROTO_TCP) {
378 struct sockopt sopt;
379 int val;
381 bzero(&sopt, sizeof sopt);
382 sopt.sopt_level = IPPROTO_TCP;
383 sopt.sopt_name = TCP_NODELAY;
384 sopt.sopt_val = &val;
385 sopt.sopt_valsize = sizeof val;
386 val = 1;
387 sosetopt(so, &sopt);
389 bzero(&sopt, sizeof sopt);
390 sopt.sopt_level = IPPROTO_TCP;
391 sopt.sopt_name = TCP_FASTKEEP;
392 sopt.sopt_val = &val;
393 sopt.sopt_valsize = sizeof val;
394 val = 1;
395 sosetopt(so, &sopt);
397 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_NOINTR);
398 so->so_rcv.ssb_timeo = 0;
399 atomic_clear_int(&so->so_snd.ssb_flags, SSB_NOINTR);
400 so->so_snd.ssb_timeo = 0;
403 * Clear AUTOSIZE, otherwise the socket buffer could be reduced
404 * to the point where rpc's cannot be queued using the mbuf
405 * interface.
407 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
408 atomic_clear_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
410 slp = kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
411 mtx_init(&slp->ns_solock, "nfsvc");
412 STAILQ_INIT(&slp->ns_rec);
413 TAILQ_INIT(&slp->ns_uidlruhead);
414 lwkt_token_init(&slp->ns_token, "nfssrv_token");
416 lwkt_gettoken(&nfs_token);
417 nfsrv_slpref(slp);
418 TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
419 lwkt_gettoken(&slp->ns_token);
421 slp->ns_so = so;
422 slp->ns_nam = mynam;
423 fhold(fp);
424 slp->ns_fp = fp;
426 so->so_upcallarg = (caddr_t)slp;
427 so->so_upcall = nfsrv_rcv_upcall;
428 atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
429 slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
430 nfsrv_wakenfsd(slp, 1);
432 lwkt_reltoken(&slp->ns_token);
433 lwkt_reltoken(&nfs_token);
435 return (0);
439 * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
440 * until it is killed by a signal.
442 static int
443 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
445 int siz;
446 struct nfssvc_sock *slp;
447 struct nfsd *nfsd = nsd->nsd_nfsd;
448 struct nfsrv_descript *nd = NULL;
449 struct mbuf *m, *mreq;
450 int error, cacherep, sotype, writes_todo;
451 int procrastinate;
452 int slplocked;
453 u_quad_t cur_usec;
455 #ifndef nolint
456 cacherep = RC_DOIT;
457 writes_todo = 0;
458 #endif
459 lwkt_gettoken(&nfs_token);
461 if (nfsd == NULL) {
462 nsd->nsd_nfsd = nfsd = (struct nfsd *)
463 kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
464 nfsd->nfsd_td = td;
465 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
466 nfs_numnfsd++;
470 * Loop getting rpc requests until SIGKILL.
472 for (;;) {
473 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
474 while (nfsd->nfsd_slp == NULL &&
475 (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
476 nfsd->nfsd_flag |= NFSD_WAITING;
477 nfsd_waiting++;
478 error = tsleep(nfsd, PCATCH, "nfsd", 0);
479 nfsd_waiting--;
480 if (error && nfsd->nfsd_slp == NULL)
481 goto done;
483 if (nfsd->nfsd_slp == NULL &&
484 (nfsd_head_flag & NFSD_CHECKSLP)) {
485 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
486 if ((slp->ns_flag & SLP_ACTION_MASK) ||
487 slp->ns_needq_upcall) {
488 nfsrv_slpref(slp);
489 nfsd->nfsd_slp = slp;
490 break;
493 if (slp == NULL)
494 nfsd_head_flag &= ~NFSD_CHECKSLP;
496 if ((slp = nfsd->nfsd_slp) == NULL)
497 continue;
499 lwkt_reltoken(&nfs_token);
500 lwkt_gettoken(&slp->ns_token);
502 if (slp->ns_needq_upcall) {
503 slp->ns_needq_upcall = 0;
504 slp->ns_flag |= SLP_NEEDQ;
507 if (slp->ns_flag & SLP_VALID) {
509 * We can both process additional received
510 * data into new records and process existing
511 * records. This keeps the pipeline hot by
512 * allowing the tcp socket to continue to
513 * drain while we are processing records.
515 while (slp->ns_flag & (SLP_DISCONN|SLP_NEEDQ)) {
516 if (slp->ns_flag & SLP_DISCONN) {
517 nfsrv_zapsock(slp);
518 break;
520 if (slp->ns_flag & SLP_NEEDQ) {
521 nfs_slplock(slp, 1);
522 if ((slp->ns_flag &
523 SLP_NEEDQ) == 0) {
524 nfs_slpunlock(slp);
525 continue;
527 nfsrv_rcv(slp->ns_so,
528 (caddr_t)slp,
529 M_WAITOK);
530 nfs_slpunlock(slp);
531 break;
534 error = nfsrv_dorec(slp, nfsd, &nd);
535 cur_usec = nfs_curusec();
536 if (error && slp->ns_tq.lh_first &&
537 slp->ns_tq.lh_first->nd_time <= cur_usec) {
538 error = 0;
539 cacherep = RC_DOIT;
540 writes_todo = 1;
541 } else {
542 writes_todo = 0;
544 nfsd->nfsd_flag |= NFSD_REQINPROG;
545 } else {
546 slp->ns_flag &= ~SLP_ACTION_MASK;
547 error = 0;
549 } else {
550 error = 0;
551 slp = nfsd->nfsd_slp;
553 lwkt_reltoken(&nfs_token);
554 lwkt_gettoken(&slp->ns_token);
556 if (slp->ns_needq_upcall) {
557 slp->ns_needq_upcall = 0;
558 slp->ns_flag |= SLP_NEEDQ;
560 if (NFSRV_RECLIMIT(slp) == 0 &&
561 (slp->ns_flag & SLP_NEEDQ)) {
562 nfs_slplock(slp, 1);
563 if (NFSRV_RECLIMIT(slp) == 0 &&
564 (slp->ns_flag & SLP_NEEDQ)) {
565 nfsrv_rcv(slp->ns_so, (caddr_t)slp,
566 M_WAITOK);
568 nfs_slpunlock(slp);
573 * nfs_token not held here. slp token is held.
575 if (error || (slp->ns_flag & SLP_VALID) == 0) {
576 if (nd) {
577 kfree((caddr_t)nd, M_NFSRVDESC);
578 nd = NULL;
580 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
581 if (slp->ns_flag & SLP_ACTION_MASK) {
582 lwkt_reltoken(&slp->ns_token);
583 lwkt_gettoken(&nfs_token);
584 } else {
585 nfsd->nfsd_slp = NULL;
586 lwkt_reltoken(&slp->ns_token);
587 lwkt_gettoken(&nfs_token);
588 nfsrv_slpderef(slp);
590 continue;
592 sotype = slp->ns_so->so_type;
595 * Execute the NFS request - handle the server side cache
597 * nfs_token not held here. slp token is held.
599 if (nd) {
600 getmicrotime(&nd->nd_starttime);
601 if (nd->nd_nam2)
602 nd->nd_nam = nd->nd_nam2;
603 else
604 nd->nd_nam = slp->ns_nam;
607 * Check to see if authorization is needed.
609 if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
610 nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
611 nsd->nsd_haddr =
612 ((struct sockaddr_in *)
613 nd->nd_nam)->sin_addr.s_addr;
614 nsd->nsd_authlen = nfsd->nfsd_authlen;
615 nsd->nsd_verflen = nfsd->nfsd_verflen;
616 if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
617 nfsd->nfsd_authlen) &&
618 !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
619 nfsd->nfsd_verflen) &&
620 !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
622 lwkt_reltoken(&slp->ns_token);
623 return (ENEEDAUTH);
625 cacherep = RC_DROPIT;
626 } else {
627 cacherep = nfsrv_getcache(nd, slp, &mreq);
630 if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
631 nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
632 nd->nd_procnum = NFSPROC_NOOP;
633 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
634 cacherep = RC_DOIT;
635 } else if (nfs_privport) {
636 /* Check if source port is privileged */
637 u_short port;
638 struct sockaddr *nam = nd->nd_nam;
639 struct sockaddr_in *sin;
641 sin = (struct sockaddr_in *)nam;
642 port = ntohs(sin->sin_port);
643 if (port >= IPPORT_RESERVED &&
644 nd->nd_procnum != NFSPROC_NULL) {
645 nd->nd_procnum = NFSPROC_NOOP;
646 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
647 cacherep = RC_DOIT;
648 kprintf("NFS request from unprivileged port (%s:%d)\n",
649 inet_ntoa(sin->sin_addr), port);
655 * Execute the NFS request - direct execution
657 * Loop to get all the write rpc replies that have been
658 * gathered together.
660 * nfs_token not held here. slp token is held.
662 do {
663 switch (cacherep) {
664 case RC_DOIT:
665 if (nd && (nd->nd_flag & ND_NFSV3))
666 procrastinate = nfsrvw_procrastinate_v3;
667 else
668 procrastinate = nfsrvw_procrastinate;
669 if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
670 procrastinate > 0)
672 error = nfsrv_writegather(&nd, slp,
673 nfsd->nfsd_td, &mreq);
674 } else {
675 /* NOT YET lwkt_reltoken(&slp->ns_token); */
676 error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
677 slp, nfsd->nfsd_td, &mreq);
678 /* NOT YET lwkt_gettoken(&slp->ns_token); */
679 lwpkthreaddeferred(); /* vnlru issues */
681 if (mreq == NULL)
682 break;
683 if (error != 0 && error != NFSERR_RETVOID) {
684 if (nd->nd_procnum != NQNFSPROC_VACATED)
685 nfsstats.srv_errs++;
686 nfsrv_updatecache(nd, FALSE, mreq);
687 if (nd->nd_nam2)
688 kfree(nd->nd_nam2, M_SONAME);
689 break;
691 nfsstats.srvrpccnt[nd->nd_procnum]++;
692 nfsrv_updatecache(nd, TRUE, mreq);
693 nd->nd_mrep = NULL;
694 /* FALL THROUGH */
695 case RC_REPLY:
696 m = mreq;
697 siz = 0;
698 while (m) {
699 siz += m->m_len;
700 m = m->m_next;
702 if (siz <= 0 || siz > NFS_MAXPACKET) {
703 kprintf("mbuf siz=%d\n",siz);
704 panic("Bad nfs svc reply");
706 m = mreq;
707 m->m_pkthdr.len = siz;
708 m->m_pkthdr.rcvif = NULL;
710 * For stream protocols, prepend a Sun RPC
711 * Record Mark.
713 if (sotype == SOCK_STREAM) {
714 M_PREPEND(m, NFSX_UNSIGNED, M_WAITOK);
715 if (m == NULL) {
716 error = ENOBUFS;
717 slplocked = 0;
718 goto skip;
720 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
722 if ((slp->ns_flag & SLP_VALID) &&
723 (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)){
724 nfs_slplock(slp, 1);
725 slplocked = 1;
726 } else {
727 slplocked = 0;
729 if (slp->ns_flag & SLP_VALID) {
730 error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
731 } else {
732 error = EPIPE;
733 m_freem(m);
735 skip:
736 if (nfsrtton)
737 nfsd_rt(sotype, nd, cacherep);
738 if (nd->nd_nam2)
739 kfree(nd->nd_nam2, M_SONAME);
740 if (nd->nd_mrep)
741 m_freem(nd->nd_mrep);
742 if (error == EPIPE || error == ENOBUFS)
743 nfsrv_zapsock(slp);
744 if (slplocked)
745 nfs_slpunlock(slp);
746 if (error == EINTR || error == ERESTART) {
747 kfree((caddr_t)nd, M_NFSRVDESC);
748 lwkt_reltoken(&slp->ns_token);
749 lwkt_gettoken(&nfs_token);
750 nfsd->nfsd_slp = NULL;
751 nfsrv_slpderef(slp);
752 goto done;
754 break;
755 case RC_DROPIT:
756 if (nfsrtton)
757 nfsd_rt(sotype, nd, cacherep);
758 m_freem(nd->nd_mrep);
759 if (nd->nd_nam2)
760 kfree(nd->nd_nam2, M_SONAME);
761 break;
763 if (nd) {
764 kfree((caddr_t)nd, M_NFSRVDESC);
765 nd = NULL;
769 * Check to see if there are outstanding writes that
770 * need to be serviced.
772 cur_usec = nfs_curusec();
773 if (slp->ns_tq.lh_first &&
774 slp->ns_tq.lh_first->nd_time <= cur_usec) {
775 cacherep = RC_DOIT;
776 writes_todo = 1;
777 } else {
778 writes_todo = 0;
780 } while (writes_todo);
783 * nfs_token not held here. slp token is held.
785 if (nfsrv_dorec(slp, nfsd, &nd)) {
786 nfsd->nfsd_flag &= ~NFSD_REQINPROG;
787 if (slp->ns_flag & SLP_ACTION_MASK) {
788 lwkt_reltoken(&slp->ns_token);
789 lwkt_gettoken(&nfs_token);
790 } else {
791 nfsd->nfsd_slp = NULL;
792 lwkt_reltoken(&slp->ns_token);
793 lwkt_gettoken(&nfs_token);
794 nfsrv_slpderef(slp);
796 } else {
797 lwkt_reltoken(&slp->ns_token);
798 lwkt_gettoken(&nfs_token);
801 done:
802 TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
803 kfree((caddr_t)nfsd, M_NFSD);
804 nsd->nsd_nfsd = NULL;
805 if (--nfs_numnfsd == 0)
806 nfsrv_init(TRUE); /* Reinitialize everything */
808 lwkt_reltoken(&nfs_token);
809 return (error);
813 * Shut down a socket associated with an nfssvc_sock structure.
814 * Should be called with the send lock set, if required.
816 * The trick here is to increment the sref at the start, so that the nfsds
817 * will stop using it and clear ns_flag at the end so that it will not be
818 * reassigned during cleanup.
820 * That said, while we shutdown() the socket here, we don't actually destroy
821 * it until the final deref as there might be other code in the middle of
822 * using it.
824 static void
825 nfsrv_zapsock(struct nfssvc_sock *slp)
827 struct nfsuid *nuidp, *nnuidp;
828 struct nfsrv_descript *nwp, *nnwp;
829 struct socket *so;
830 struct nfsrv_rec *rec;
831 int wasvalid;
833 wasvalid = slp->ns_flag & SLP_VALID;
834 slp->ns_flag &= ~SLP_ALLFLAGS;
835 if (wasvalid && slp->ns_fp) {
836 so = slp->ns_so;
837 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
838 so->so_upcall = NULL;
839 so->so_upcallarg = NULL;
840 soshutdown(so, SHUT_RDWR);
841 if (slp->ns_nam) {
842 kfree(slp->ns_nam, M_SONAME);
843 slp->ns_nam = NULL;
846 m_freem(slp->ns_raw);
847 slp->ns_raw = NULL;
849 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
850 --slp->ns_numrec;
851 STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
852 if (rec->nr_address)
853 kfree(rec->nr_address, M_SONAME);
854 m_freem(rec->nr_packet);
855 kfree(rec, M_NFSRVDESC);
857 KKASSERT(slp->ns_numrec == 0);
859 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
860 nnuidp) {
861 LIST_REMOVE(nuidp, nu_hash);
862 TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
863 if (nuidp->nu_flag & NU_NAM)
864 kfree(nuidp->nu_nam, M_SONAME);
865 kfree((caddr_t)nuidp, M_NFSUID);
867 crit_enter(); /* XXX doesn't do anything any more */
868 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
869 nnwp = nwp->nd_tq.le_next;
870 LIST_REMOVE(nwp, nd_tq);
871 kfree((caddr_t)nwp, M_NFSRVDESC);
873 LIST_INIT(&slp->ns_tq);
874 crit_exit();
876 nfsrv_slpderef(slp);
881 * Derefence a server socket structure. If it has no more references and
882 * is no longer valid, you can throw it away.
884 * Must be holding nfs_token!
886 void
887 nfsrv_slpderef(struct nfssvc_sock *slp)
889 struct file *fp;
891 ASSERT_LWKT_TOKEN_HELD(&nfs_token);
892 if (slp->ns_sref == 1) {
893 KKASSERT((slp->ns_flag & SLP_VALID) == 0);
894 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
895 slp->ns_sref = 0;
896 fp = slp->ns_fp;
897 slp->ns_fp = NULL;
898 slp->ns_so = NULL;
899 if (fp)
900 closef(fp, NULL);
901 kfree((caddr_t)slp, M_NFSSVC);
902 } else {
903 --slp->ns_sref;
907 void
908 nfsrv_slpref(struct nfssvc_sock *slp)
910 ASSERT_LWKT_TOKEN_HELD(&nfs_token);
911 ++slp->ns_sref;
915 * Lock a socket against others.
917 * Returns 0 on failure, 1 on success.
920 nfs_slplock(struct nfssvc_sock *slp, int wait)
922 mtx_t *mtx = &slp->ns_solock;
924 if (wait) {
925 mtx_lock_ex(mtx, 0, 0);
926 return(1);
927 } else if (mtx_lock_ex_try(mtx) == 0) {
928 return(1);
929 } else {
930 return(0);
935 * Unlock the stream socket for others.
937 void
938 nfs_slpunlock(struct nfssvc_sock *slp)
940 mtx_t *mtx = &slp->ns_solock;
942 mtx_unlock(mtx);
946 * Initialize the data structures for the server.
947 * Handshake with any new nfsds starting up to avoid any chance of
948 * corruption.
950 void
951 nfsrv_init(int terminating)
953 struct nfssvc_sock *slp, *nslp;
955 lwkt_gettoken(&nfs_token);
956 if (nfssvc_sockhead_flag & SLP_INIT)
957 panic("nfsd init");
958 nfssvc_sockhead_flag |= SLP_INIT;
960 if (terminating) {
961 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
962 nfsrv_slpref(slp);
963 lwkt_gettoken(&slp->ns_token);
964 if (slp->ns_flag & SLP_VALID)
965 nfsrv_zapsock(slp);
966 lwkt_reltoken(&slp->ns_token);
967 nfsrv_slpderef(slp);
969 nfsrv_cleancache(); /* And clear out server cache */
970 } else {
971 nfs_pub.np_valid = 0;
974 TAILQ_INIT(&nfssvc_sockhead);
975 nfssvc_sockhead_flag &= ~SLP_INIT;
976 if (nfssvc_sockhead_flag & SLP_WANTINIT) {
977 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
978 wakeup((caddr_t)&nfssvc_sockhead);
981 TAILQ_INIT(&nfsd_head);
982 nfsd_head_flag &= ~NFSD_CHECKSLP;
984 lwkt_reltoken(&nfs_token);
986 #if 0
987 nfs_udpsock = (struct nfssvc_sock *)
988 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
989 mtx_init(&nfs_udpsock->ns_solock);
990 STAILQ_INIT(&nfs_udpsock->ns_rec);
991 TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
992 TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
994 nfs_cltpsock = (struct nfssvc_sock *)
995 kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
996 mtx_init(&nfs_cltpsock->ns_solock);
997 STAILQ_INIT(&nfs_cltpsock->ns_rec);
998 TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
999 TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
1000 #endif
1004 * Add entries to the server monitor log.
1006 static void
1007 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
1009 struct drt *rt;
1011 rt = &nfsdrt.drt[nfsdrt.pos];
1012 if (cacherep == RC_DOIT)
1013 rt->flag = 0;
1014 else if (cacherep == RC_REPLY)
1015 rt->flag = DRT_CACHEREPLY;
1016 else
1017 rt->flag = DRT_CACHEDROP;
1018 if (sotype == SOCK_STREAM)
1019 rt->flag |= DRT_TCP;
1020 if (nd->nd_flag & ND_NFSV3)
1021 rt->flag |= DRT_NFSV3;
1022 rt->proc = nd->nd_procnum;
1023 if (nd->nd_nam->sa_family == AF_INET)
1024 rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
1025 else
1026 rt->ipadr = INADDR_ANY;
1027 rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
1028 getmicrotime(&rt->tstamp);
1029 nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
1031 #endif /* NFS_NOSERVER */
1034 * Get an authorization string for the uid by having the mount_nfs sitting
1035 * on this mount point porpous out of the kernel and do it.
1038 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
1039 struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
1040 int *verf_len, NFSKERBKEY_T key /* return session key */)
1042 int error = 0;
1044 while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
1045 nmp->nm_state |= NFSSTA_WANTAUTH;
1046 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
1047 "nfsauth1", 2 * hz);
1048 error = nfs_sigintr(nmp, rep, rep->r_td);
1049 if (error) {
1050 nmp->nm_state &= ~NFSSTA_WANTAUTH;
1051 return (error);
1054 nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
1055 nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
1056 nmp->nm_authlen = RPCAUTH_MAXSIZ;
1057 nmp->nm_verfstr = verf_str;
1058 nmp->nm_verflen = *verf_len;
1059 nmp->nm_authuid = cred->cr_uid;
1060 wakeup((caddr_t)&nmp->nm_authstr);
1063 * And wait for mount_nfs to do its stuff.
1065 while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
1066 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
1067 "nfsauth2", 2 * hz);
1068 error = nfs_sigintr(nmp, rep, rep->r_td);
1070 if (nmp->nm_state & NFSSTA_AUTHERR) {
1071 nmp->nm_state &= ~NFSSTA_AUTHERR;
1072 error = EAUTH;
1074 if (error)
1075 kfree((caddr_t)*auth_str, M_TEMP);
1076 else {
1077 *auth_len = nmp->nm_authlen;
1078 *verf_len = nmp->nm_verflen;
1079 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (NFSKERBKEY_T));
1081 nmp->nm_state &= ~NFSSTA_HASAUTH;
1082 nmp->nm_state |= NFSSTA_WAITAUTH;
1083 if (nmp->nm_state & NFSSTA_WANTAUTH) {
1084 nmp->nm_state &= ~NFSSTA_WANTAUTH;
1085 wakeup((caddr_t)&nmp->nm_authtype);
1087 return (error);
1091 * Get a nickname authenticator and verifier.
1094 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
1095 int *auth_len, char *verf_str, int verf_len)
1097 struct nfsuid *nuidp;
1098 u_int32_t *nickp, *verfp;
1099 struct timeval ktvout;
1101 #ifdef DIAGNOSTIC
1102 if (verf_len < (4 * NFSX_UNSIGNED))
1103 panic("nfs_getnickauth verf too small");
1104 #endif
1105 for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
1106 nuidp != NULL; nuidp = nuidp->nu_hash.le_next) {
1107 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
1108 break;
1110 if (!nuidp || nuidp->nu_expire < time_uptime)
1111 return (EACCES);
1114 * Move to the end of the lru list (end of lru == most recently used).
1116 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1117 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1119 nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1120 *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1121 *nickp = txdr_unsigned(nuidp->nu_nickname);
1122 *auth_str = (char *)nickp;
1123 *auth_len = 2 * NFSX_UNSIGNED;
1126 * Now we must encrypt the verifier and package it up.
1128 verfp = (u_int32_t *)verf_str;
1129 *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1130 if (time_second != nuidp->nu_timestamp.tv_sec ||
1131 (time_second == nuidp->nu_timestamp.tv_sec &&
1132 time_second > nuidp->nu_timestamp.tv_usec)) /* XXX */
1133 getmicrotime(&nuidp->nu_timestamp);
1134 else
1135 nuidp->nu_timestamp.tv_usec++;
1138 * Now encrypt the timestamp verifier in ecb mode using the session
1139 * key.
1141 #ifdef NFSKERB
1143 #else
1144 ktvout.tv_sec = 0;
1145 ktvout.tv_usec = 0;
1146 #endif
1148 *verfp++ = ktvout.tv_sec;
1149 *verfp++ = ktvout.tv_usec;
1150 *verfp = 0;
1151 return (0);
1155 * Save the current nickname in a hash list entry on the mount point.
1158 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1159 NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1160 struct mbuf *mrep)
1162 struct nfsuid *nuidp;
1163 u_int32_t *tl;
1164 struct timeval ktvin, ktvout;
1165 u_int32_t nick;
1166 int deltasec, error = 0;
1167 struct nfsm_info info;
1169 info.md = *mdp;
1170 info.dpos = *dposp;
1171 info.mrep = mrep;
1173 if (len == (3 * NFSX_UNSIGNED)) {
1174 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
1175 ktvin.tv_sec = *tl++;
1176 ktvin.tv_usec = *tl++;
1177 nick = fxdr_unsigned(u_int32_t, *tl);
1180 * Decrypt the timestamp in ecb mode.
1182 #ifdef NFSKERB
1184 #else
1185 ktvout.tv_sec = 0;
1186 ktvout.tv_usec = 0;
1187 #endif
1188 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1189 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1190 deltasec = time_second - ktvout.tv_sec;
1191 if (deltasec < 0)
1192 deltasec = -deltasec;
1194 * If ok, add it to the hash list for the mount point.
1196 if (deltasec <= NFS_KERBCLOCKSKEW) {
1197 if (nmp->nm_numuids < nuidhash_max) {
1198 nmp->nm_numuids++;
1199 nuidp = (struct nfsuid *)
1200 kmalloc(sizeof (struct nfsuid), M_NFSUID,
1201 M_WAITOK);
1202 } else {
1203 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1204 LIST_REMOVE(nuidp, nu_hash);
1205 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1206 nu_lru);
1208 nuidp->nu_flag = 0;
1209 nuidp->nu_cr.cr_uid = cred->cr_uid;
1210 nuidp->nu_expire = time_uptime + NFS_KERBTTL;
1211 nuidp->nu_timestamp = ktvout;
1212 nuidp->nu_nickname = nick;
1213 bcopy(key, nuidp->nu_key, sizeof (NFSKERBKEY_T));
1214 TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1215 nu_lru);
1216 LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1217 nuidp, nu_hash);
1219 } else {
1220 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
1222 nfsmout:
1223 *mdp = info.md;
1224 *dposp = info.dpos;
1225 return (error);