Merge branch 'vendor/BZIP'
[dragonfly.git] / sys / vfs / nfs / nfs_syscalls.c
1 /*
2  * Copyright (c) 1989, 1993
3  *      The Regents of the University of California.  All rights reserved.
4  *
5  * This code is derived from software contributed to Berkeley by
6  * Rick Macklem at The University of Guelph.
7  *
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. 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.
23  *
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
34  * SUCH DAMAGE.
35  *
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 $
39  */
40
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>
46 #include <sys/file.h>
47 #include <sys/filedesc.h>
48 #include <sys/vnode.h>
49 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/proc.h>
52 #include <sys/priv.h>
53 #include <sys/buf.h>
54 #include <sys/mbuf.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>
62
63 #include <sys/mutex2.h>
64 #include <sys/thread2.h>
65
66 #include <netinet/in.h>
67 #include <netinet/tcp.h>
68 #include "xdr_subs.h"
69 #include "rpcv2.h"
70 #include "nfsproto.h"
71 #include "nfs.h"
72 #include "nfsm_subs.h"
73 #include "nfsrvcache.h"
74 #include "nfsmount.h"
75 #include "nfsnode.h"
76 #include "nfsrtt.h"
77
78 #include <sys/thread2.h>
79
80 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
81
82 static int nuidhash_max = NFS_MAXUIDHASH;
83
84 #ifndef NFS_NOSERVER
85 static void     nfsrv_zapsock (struct nfssvc_sock *slp);
86 #endif
87
88 #define TRUE    1
89 #define FALSE   0
90
91 SYSCTL_DECL(_vfs_nfs);
92
93 #ifndef NFS_NOSERVER
94 int nfsd_waiting = 0;
95 static struct nfsdrt nfsdrt;
96 static int nfs_numnfsd = 0;
97 static void     nfsd_rt (int sotype, struct nfsrv_descript *nd,
98                              int cacherep);
99 static int      nfssvc_addsock (struct file *, struct sockaddr *,
100                                     struct thread *);
101 static int      nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
102
103 static int nfs_privport = 0;
104 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
105 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
106 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
107 int     nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO;
108 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
109
110 /*
111  * NFS server system calls
112  */
113
114 #endif /* NFS_NOSERVER */
115 /*
116  * nfssvc_args(int flag, caddr_t argp)
117  *
118  * Nfs server psuedo system call for the nfsd's
119  * Based on the flag value it either:
120  * - adds a socket to the selection list
121  * - remains in the kernel as an nfsd
122  * - remains in the kernel as an nfsiod
123  *
124  * MPALMOSTSAFE
125  */
126 int
127 sys_nfssvc(struct nfssvc_args *uap)
128 {
129 #ifndef NFS_NOSERVER
130         struct nlookupdata nd;
131         struct file *fp;
132         struct sockaddr *nam;
133         struct nfsd_args nfsdarg;
134         struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
135         struct nfsd_cargs ncd;
136         struct nfsd *nfsd;
137         struct nfssvc_sock *slp;
138         struct nfsuid *nuidp;
139         struct nfsmount *nmp;
140         struct vnode *vp;
141 #endif /* NFS_NOSERVER */
142         int error;
143         struct thread *td = curthread;
144
145         /*
146          * Must be super user
147          */
148         error = priv_check(td, PRIV_ROOT);
149         if (error)
150                 return (error);
151
152         lwkt_gettoken(&nfs_token);
153
154         while (nfssvc_sockhead_flag & SLP_INIT) {
155                 nfssvc_sockhead_flag |= SLP_WANTINIT;
156                 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
157         }
158         if (uap->flag & NFSSVC_BIOD)
159                 error = ENXIO;          /* no longer need nfsiod's */
160 #ifdef NFS_NOSERVER
161         else
162                 error = ENXIO;
163 #else /* !NFS_NOSERVER */
164         else if (uap->flag & NFSSVC_MNTD) {
165                 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
166                 if (error)
167                         goto done;
168                 vp = NULL;
169                 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 
170                                         NLC_FOLLOW);
171                 if (error == 0)
172                         error = nlookup(&nd);
173                 if (error == 0)
174                         error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);   
175                 nlookup_done(&nd);
176                 if (error)
177                         goto done;
178
179                 if ((vp->v_flag & VROOT) == 0)
180                         error = EINVAL;
181                 nmp = VFSTONFS(vp->v_mount);
182                 vput(vp);
183                 if (error)
184                         goto done;
185                 if ((nmp->nm_state & NFSSTA_MNTD) &&
186                         (uap->flag & NFSSVC_GOTAUTH) == 0) {
187                         error = 0;
188                         goto done;
189                 }
190                 nmp->nm_state |= NFSSTA_MNTD;
191                 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag,
192                                     uap->argp, td);
193         } else if (uap->flag & NFSSVC_ADDSOCK) {
194                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
195                 if (error)
196                         goto done;
197                 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
198                 if (error)
199                         goto done;
200                 /*
201                  * Get the client address for connected sockets.
202                  */
203                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
204                         nam = NULL;
205                 else {
206                         error = getsockaddr(&nam, nfsdarg.name,
207                                             nfsdarg.namelen);
208                         if (error) {
209                                 fdrop(fp);
210                                 goto done;
211                         }
212                 }
213                 error = nfssvc_addsock(fp, nam, td);
214                 fdrop(fp);
215         } else {
216                 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
217                 if (error)
218                         goto done;
219                 if ((uap->flag & NFSSVC_AUTHIN) &&
220                     ((nfsd = nsd->nsd_nfsd)) != NULL &&
221                     (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
222                         slp = nfsd->nfsd_slp;
223
224                         /*
225                          * First check to see if another nfsd has already
226                          * added this credential.
227                          */
228                         for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
229                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
230                                 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
231                                     (!nfsd->nfsd_nd->nd_nam2 ||
232                                      netaddr_match(NU_NETFAM(nuidp),
233                                      &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
234                                         break;
235                         }
236                         if (nuidp) {
237                             nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
238                             nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
239                         } else {
240                             /*
241                              * Nope, so we will.
242                              */
243                             if (slp->ns_numuids < nuidhash_max) {
244                                 slp->ns_numuids++;
245                                 nuidp = (struct nfsuid *)
246                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
247                                         M_WAITOK);
248                             } else
249                                 nuidp = NULL;
250                             if ((slp->ns_flag & SLP_VALID) == 0) {
251                                 if (nuidp)
252                                     kfree((caddr_t)nuidp, M_NFSUID);
253                             } else {
254                                 if (nuidp == NULL) {
255                                     nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
256                                     LIST_REMOVE(nuidp, nu_hash);
257                                     TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
258                                         nu_lru);
259                                     if (nuidp->nu_flag & NU_NAM)
260                                         FREE(nuidp->nu_nam, M_SONAME);
261                                 }
262                                 nuidp->nu_flag = 0;
263                                 nuidp->nu_cr = nsd->nsd_cr;
264                                 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
265                                     nuidp->nu_cr.cr_ngroups = NGROUPS;
266                                 nuidp->nu_cr.cr_ref = 1;
267                                 nuidp->nu_timestamp = nsd->nsd_timestamp;
268                                 nuidp->nu_expire = time_second + nsd->nsd_ttl;
269                                 /*
270                                  * and save the session key in nu_key.
271                                  */
272                                 bcopy(nsd->nsd_key, nuidp->nu_key,
273                                     sizeof (nsd->nsd_key));
274                                 if (nfsd->nfsd_nd->nd_nam2) {
275                                     struct sockaddr_in *saddr;
276
277                                     saddr = (struct sockaddr_in *)
278                                             nfsd->nfsd_nd->nd_nam2;
279                                     switch (saddr->sin_family) {
280                                     case AF_INET:
281                                         nuidp->nu_flag |= NU_INETADDR;
282                                         nuidp->nu_inetaddr =
283                                              saddr->sin_addr.s_addr;
284                                         break;
285                                     case AF_ISO:
286                                     default:
287                                         nuidp->nu_flag |= NU_NAM;
288                                         nuidp->nu_nam = 
289                                           dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
290                                         break;
291                                     };
292                                 }
293                                 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
294                                         nu_lru);
295                                 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
296                                         nuidp, nu_hash);
297                                 nfsrv_setcred(&nuidp->nu_cr,
298                                     &nfsd->nfsd_nd->nd_cr);
299                                 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
300                             }
301                         }
302                 }
303                 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
304                         nfsd->nfsd_flag |= NFSD_AUTHFAIL;
305                 error = nfssvc_nfsd(nsd, uap->argp, td);
306         }
307 #endif /* NFS_NOSERVER */
308         if (error == EINTR || error == ERESTART)
309                 error = 0;
310 done:
311         lwkt_reltoken(&nfs_token);
312         return (error);
313 }
314
315 #ifndef NFS_NOSERVER
316 /*
317  * Adds a socket to the list for servicing by nfsds.
318  */
319 static int
320 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
321 {
322         int siz;
323         struct nfssvc_sock *slp;
324         struct socket *so;
325         int error;
326
327         so = (struct socket *)fp->f_data;
328 #if 0
329         tslp = NULL;
330         /*
331          * Add it to the list, as required.
332          */
333         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
334                 tslp = nfs_udpsock;
335                 if (tslp->ns_flag & SLP_VALID) {
336                         if (mynam != NULL)
337                                 FREE(mynam, M_SONAME);
338                         return (EPERM);
339                 }
340         }
341 #endif
342         /*
343          * Reserve buffer space in the socket.  Note that due to bugs in
344          * Linux's delayed-ack code, serious performance degredation may
345          * occur with linux hosts if the minimum is used.
346          *
347          * NFS sockets are not limited to the standard sb_max or by
348          * resource limits.
349          */
350         if (so->so_type == SOCK_STREAM)
351                 siz = NFS_MAXPACKET + sizeof (u_long);
352         else
353                 siz = NFS_MAXPACKET;
354         if (siz < nfs_soreserve)
355             siz = nfs_soreserve;
356
357         error = soreserve(so, siz, siz, NULL);
358         if (error) {
359                 if (mynam != NULL)
360                         FREE(mynam, M_SONAME);
361                 return (error);
362         }
363
364         /*
365          * Set protocol specific options { for now TCP only } and
366          * reserve some space. For datagram sockets, this can get called
367          * repeatedly for the same socket, but that isn't harmful.
368          */
369         if (so->so_type == SOCK_STREAM) {
370                 struct sockopt sopt;
371                 int val;
372
373                 bzero(&sopt, sizeof sopt);
374                 sopt.sopt_level = SOL_SOCKET;
375                 sopt.sopt_name = SO_KEEPALIVE;
376                 sopt.sopt_val = &val;
377                 sopt.sopt_valsize = sizeof val;
378                 val = 1;
379                 sosetopt(so, &sopt);
380         }
381         if (so->so_proto->pr_domain->dom_family == AF_INET &&
382             so->so_proto->pr_protocol == IPPROTO_TCP) {
383                 struct sockopt sopt;
384                 int val;
385
386                 bzero(&sopt, sizeof sopt);
387                 sopt.sopt_level = IPPROTO_TCP;
388                 sopt.sopt_name = TCP_NODELAY;
389                 sopt.sopt_val = &val;
390                 sopt.sopt_valsize = sizeof val;
391                 val = 1;
392                 sosetopt(so, &sopt);
393         }
394         atomic_clear_int(&so->so_rcv.ssb_flags, SSB_NOINTR);
395         so->so_rcv.ssb_timeo = 0;
396         atomic_clear_int(&so->so_snd.ssb_flags, SSB_NOINTR);
397         so->so_snd.ssb_timeo = 0;
398
399         slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
400             M_NFSSVC, M_WAITOK | M_ZERO);
401         mtx_init(&slp->ns_solock);
402         STAILQ_INIT(&slp->ns_rec);
403         TAILQ_INIT(&slp->ns_uidlruhead);
404         lwkt_token_init(&slp->ns_token, 1, "nfssrv_token");
405
406         lwkt_gettoken(&nfs_token);
407         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
408         nfsrv_slpref(slp);
409         lwkt_gettoken(&slp->ns_token);
410
411         slp->ns_so = so;
412         slp->ns_nam = mynam;
413         fp->f_count++;
414         slp->ns_fp = fp;
415
416         so->so_upcallarg = (caddr_t)slp;
417         so->so_upcall = nfsrv_rcv_upcall;
418         atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
419         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
420         nfsrv_wakenfsd(slp, 1);
421
422         lwkt_reltoken(&slp->ns_token);
423         lwkt_reltoken(&nfs_token);
424
425         return (0);
426 }
427
428 /*
429  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
430  * until it is killed by a signal.
431  */
432 static int
433 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
434 {
435         int siz;
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;
441         int procrastinate;
442         u_quad_t cur_usec;
443
444 #ifndef nolint
445         cacherep = RC_DOIT;
446         writes_todo = 0;
447 #endif
448         lwkt_gettoken(&nfs_token);
449
450         if (nfsd == NULL) {
451                 nsd->nsd_nfsd = nfsd = (struct nfsd *)
452                         kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
453                 nfsd->nfsd_td = td;
454                 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
455                 nfs_numnfsd++;
456         }
457
458         /*
459          * Loop getting rpc requests until SIGKILL.
460          */
461         for (;;) {
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;
466                                 nfsd_waiting++;
467                                 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
468                                 nfsd_waiting--;
469                                 if (error)
470                                         goto done;
471                         }
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;
478                                             nfsrv_slpref(slp);
479                                             nfsd->nfsd_slp = slp;
480                                             break;
481                                     }
482                                 }
483                                 if (slp == 0)
484                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
485                         }
486                         if ((slp = nfsd->nfsd_slp) == NULL)
487                                 continue;
488
489                         lwkt_reltoken(&nfs_token);
490                         lwkt_gettoken(&slp->ns_token);
491
492                         if (slp->ns_flag & SLP_VALID) {
493                                 if (slp->ns_flag & SLP_DISCONN)
494                                         nfsrv_zapsock(slp);
495                                 else if (slp->ns_flag & SLP_NEEDQ) {
496                                         slp->ns_flag &= ~SLP_NEEDQ;
497                                         (void) nfs_slplock(slp, 1);
498                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
499                                                 MB_WAIT);
500                                         nfs_slpunlock(slp);
501                                 }
502                                 error = nfsrv_dorec(slp, nfsd, &nd);
503                                 cur_usec = nfs_curusec();
504                                 if (error && slp->ns_tq.lh_first &&
505                                     slp->ns_tq.lh_first->nd_time <= cur_usec) {
506                                         error = 0;
507                                         cacherep = RC_DOIT;
508                                         writes_todo = 1;
509                                 } else
510                                         writes_todo = 0;
511                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
512                         }
513                 } else {
514                         error = 0;
515                         slp = nfsd->nfsd_slp;
516                         lwkt_reltoken(&nfs_token);
517                         lwkt_gettoken(&slp->ns_token);
518                 }
519
520                 /*
521                  * nfs_token not held here.  slp token is held.
522                  */
523                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
524                         if (nd) {
525                                 kfree((caddr_t)nd, M_NFSRVDESC);
526                                 nd = NULL;
527                         }
528                         nfsd->nfsd_slp = NULL;
529                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
530                         lwkt_reltoken(&slp->ns_token);
531                         lwkt_gettoken(&nfs_token);
532                         nfsrv_slpderef(slp);
533                         continue;
534                 }
535
536                 /*
537                  * nfs_token not held here.  slp token is held.
538                  */
539                 sotype = slp->ns_so->so_type;
540                 if (nd) {
541                     getmicrotime(&nd->nd_starttime);
542                     if (nd->nd_nam2)
543                         nd->nd_nam = nd->nd_nam2;
544                     else
545                         nd->nd_nam = slp->ns_nam;
546
547                     /*
548                      * Check to see if authorization is needed.
549                      */
550                     if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
551                         nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
552                         nsd->nsd_haddr = 
553                                 ((struct sockaddr_in *)
554                                  nd->nd_nam)->sin_addr.s_addr;
555                         nsd->nsd_authlen = nfsd->nfsd_authlen;
556                         nsd->nsd_verflen = nfsd->nfsd_verflen;
557                         if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
558                                 nfsd->nfsd_authlen) &&
559                             !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
560                                 nfsd->nfsd_verflen) &&
561                             !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
562                             lwkt_reltoken(&slp->ns_token);
563                             return (ENEEDAUTH);
564                         cacherep = RC_DROPIT;
565                     } else {
566                         cacherep = nfsrv_getcache(nd, slp, &mreq);
567                     }
568
569                     if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
570                         nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
571                         nd->nd_procnum = NFSPROC_NOOP;
572                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
573                         cacherep = RC_DOIT;
574                     } else if (nfs_privport) {
575                         /* Check if source port is privileged */
576                         u_short port;
577                         struct sockaddr *nam = nd->nd_nam;
578                         struct sockaddr_in *sin;
579
580                         sin = (struct sockaddr_in *)nam;
581                         port = ntohs(sin->sin_port);
582                         if (port >= IPPORT_RESERVED && 
583                             nd->nd_procnum != NFSPROC_NULL) {
584                             nd->nd_procnum = NFSPROC_NOOP;
585                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
586                             cacherep = RC_DOIT;
587                             kprintf("NFS request from unprivileged port (%s:%d)\n",
588                                    inet_ntoa(sin->sin_addr), port);
589                         }
590                     }
591
592                 }
593
594                 /*
595                  * Loop to get all the write rpc replies that have been
596                  * gathered together.
597                  *
598                  * nfs_token not held here.  slp token is held.
599                  */
600                 do {
601                     switch (cacherep) {
602                     case RC_DOIT:
603                         if (nd && (nd->nd_flag & ND_NFSV3))
604                             procrastinate = nfsrvw_procrastinate_v3;
605                         else
606                             procrastinate = nfsrvw_procrastinate;
607                         if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
608                             procrastinate > 0)
609                         ) {
610                             error = nfsrv_writegather(&nd, slp,
611                                 nfsd->nfsd_td, &mreq);
612                         } else {
613                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
614                                 slp, nfsd->nfsd_td, &mreq);
615                         }
616                         if (mreq == NULL)
617                                 break;
618                         if (error != 0 && error != NFSERR_RETVOID) {
619                                 if (nd->nd_procnum != NQNFSPROC_VACATED)
620                                         nfsstats.srv_errs++;
621                                 nfsrv_updatecache(nd, FALSE, mreq);
622                                 if (nd->nd_nam2)
623                                         FREE(nd->nd_nam2, M_SONAME);
624                                 break;
625                         }
626                         nfsstats.srvrpccnt[nd->nd_procnum]++;
627                         nfsrv_updatecache(nd, TRUE, mreq);
628                         nd->nd_mrep = NULL;
629                     case RC_REPLY:
630                         m = mreq;
631                         siz = 0;
632                         while (m) {
633                                 siz += m->m_len;
634                                 m = m->m_next;
635                         }
636                         if (siz <= 0 || siz > NFS_MAXPACKET) {
637                                 kprintf("mbuf siz=%d\n",siz);
638                                 panic("Bad nfs svc reply");
639                         }
640                         m = mreq;
641                         m->m_pkthdr.len = siz;
642                         m->m_pkthdr.rcvif = NULL;
643                         /*
644                          * For stream protocols, prepend a Sun RPC
645                          * Record Mark.
646                          */
647                         if (sotype == SOCK_STREAM) {
648                                 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
649                                 if (m == NULL)
650                                         return (ENOBUFS);
651                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
652                         }
653                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
654                                 (void) nfs_slplock(slp, 1);
655                         if (slp->ns_flag & SLP_VALID)
656                             error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
657                         else {
658                             error = EPIPE;
659                             m_freem(m);
660                         }
661                         if (nfsrtton)
662                                 nfsd_rt(sotype, nd, cacherep);
663                         if (nd->nd_nam2)
664                                 FREE(nd->nd_nam2, M_SONAME);
665                         if (nd->nd_mrep)
666                                 m_freem(nd->nd_mrep);
667                         if (error == EPIPE)
668                                 nfsrv_zapsock(slp);
669                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
670                                 nfs_slpunlock(slp);
671                         if (error == EINTR || error == ERESTART) {
672                                 kfree((caddr_t)nd, M_NFSRVDESC);
673                                 lwkt_reltoken(&slp->ns_token);
674                                 lwkt_gettoken(&nfs_token);
675                                 nfsrv_slpderef(slp);
676                                 goto done;
677                         }
678                         break;
679                     case RC_DROPIT:
680                         if (nfsrtton)
681                                 nfsd_rt(sotype, nd, cacherep);
682                         m_freem(nd->nd_mrep);
683                         if (nd->nd_nam2)
684                                 FREE(nd->nd_nam2, M_SONAME);
685                         break;
686                     };
687                     if (nd) {
688                         FREE((caddr_t)nd, M_NFSRVDESC);
689                         nd = NULL;
690                     }
691
692                     /*
693                      * Check to see if there are outstanding writes that
694                      * need to be serviced.
695                      */
696                     cur_usec = nfs_curusec();
697                     if (slp->ns_tq.lh_first &&
698                         slp->ns_tq.lh_first->nd_time <= cur_usec) {
699                         cacherep = RC_DOIT;
700                         writes_todo = 1;
701                     } else {
702                         writes_todo = 0;
703                     }
704                 } while (writes_todo);
705
706                 /*
707                  * nfs_token not held here.  slp token is held.
708                  */
709                 if (nfsrv_dorec(slp, nfsd, &nd)) {
710                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
711                         nfsd->nfsd_slp = NULL;
712                         lwkt_reltoken(&slp->ns_token);
713                         lwkt_gettoken(&nfs_token);
714                         nfsrv_slpderef(slp);
715                 } else {
716                         lwkt_reltoken(&slp->ns_token);
717                         lwkt_gettoken(&nfs_token);
718                 }
719         }
720 done:
721         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
722         kfree((caddr_t)nfsd, M_NFSD);
723         nsd->nsd_nfsd = NULL;
724         if (--nfs_numnfsd == 0)
725                 nfsrv_init(TRUE);       /* Reinitialize everything */
726
727         lwkt_reltoken(&nfs_token);
728         return (error);
729 }
730
731 /*
732  * Shut down a socket associated with an nfssvc_sock structure.
733  * Should be called with the send lock set, if required.
734  * The trick here is to increment the sref at the start, so that the nfsds
735  * will stop using it and clear ns_flag at the end so that it will not be
736  * reassigned during cleanup.
737  */
738 static void
739 nfsrv_zapsock(struct nfssvc_sock *slp)
740 {
741         struct nfsuid *nuidp, *nnuidp;
742         struct nfsrv_descript *nwp, *nnwp;
743         struct socket *so;
744         struct file *fp;
745         struct nfsrv_rec *rec;
746
747         slp->ns_flag &= ~SLP_ALLFLAGS;
748         fp = slp->ns_fp;
749         if (fp) {
750                 slp->ns_fp = NULL;
751                 so = slp->ns_so;
752                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
753                 so->so_upcall = NULL;
754                 so->so_upcallarg = NULL;
755                 soshutdown(so, SHUT_RDWR);
756                 closef(fp, NULL);
757                 if (slp->ns_nam)
758                         FREE(slp->ns_nam, M_SONAME);
759                 m_freem(slp->ns_raw);
760                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
761                         --slp->ns_numrec;
762                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
763                         if (rec->nr_address)
764                                 FREE(rec->nr_address, M_SONAME);
765                         m_freem(rec->nr_packet);
766                         kfree(rec, M_NFSRVDESC);
767                 }
768                 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
769                                       nnuidp) {
770                         LIST_REMOVE(nuidp, nu_hash);
771                         TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
772                         if (nuidp->nu_flag & NU_NAM)
773                                 FREE(nuidp->nu_nam, M_SONAME);
774                         kfree((caddr_t)nuidp, M_NFSUID);
775                 }
776                 crit_enter();
777                 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
778                         nnwp = nwp->nd_tq.le_next;
779                         LIST_REMOVE(nwp, nd_tq);
780                         kfree((caddr_t)nwp, M_NFSRVDESC);
781                 }
782                 LIST_INIT(&slp->ns_tq);
783                 crit_exit();
784         }
785 }
786
787 /*
788  * Derefence a server socket structure. If it has no more references and
789  * is no longer valid, you can throw it away.
790  *
791  * Must be holding nfs_token!
792  */
793 void
794 nfsrv_slpderef(struct nfssvc_sock *slp)
795 {
796         ASSERT_LWKT_TOKEN_HELD(&nfs_token);
797         if (--slp->ns_sref == 0 && (slp->ns_flag & SLP_VALID) == 0) {
798                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
799                 kfree((caddr_t)slp, M_NFSSVC);
800         }
801 }
802
803 void
804 nfsrv_slpref(struct nfssvc_sock *slp)
805 {
806         ASSERT_LWKT_TOKEN_HELD(&nfs_token);
807         ++slp->ns_sref;
808 }
809
810 /*
811  * Lock a socket against others.
812  *
813  * Returns 0 on failure, 1 on success.
814  */
815 int
816 nfs_slplock(struct nfssvc_sock *slp, int wait)
817 {
818         mtx_t mtx = &slp->ns_solock;
819
820         if (wait) {
821                 mtx_lock_ex(mtx, "nfsslplck", 0, 0);
822                 return(1);
823         } else if (mtx_lock_ex_try(mtx) == 0) {
824                 return(1);
825         } else {
826                 return(0);
827         }
828 }
829
830 /*
831  * Unlock the stream socket for others.
832  */
833 void
834 nfs_slpunlock(struct nfssvc_sock *slp)
835 {
836         mtx_t mtx = &slp->ns_solock;
837
838         mtx_unlock(mtx);
839 }
840
841 /*
842  * Initialize the data structures for the server.
843  * Handshake with any new nfsds starting up to avoid any chance of
844  * corruption.
845  */
846 void
847 nfsrv_init(int terminating)
848 {
849         struct nfssvc_sock *slp, *nslp;
850
851         lwkt_gettoken(&nfs_token);
852         if (nfssvc_sockhead_flag & SLP_INIT)
853                 panic("nfsd init");
854         nfssvc_sockhead_flag |= SLP_INIT;
855
856         if (terminating) {
857                 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
858                         if (slp->ns_flag & SLP_VALID)
859                                 nfsrv_zapsock(slp);
860                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
861                         kfree((caddr_t)slp, M_NFSSVC);
862                 }
863                 nfsrv_cleancache();     /* And clear out server cache */
864         } else {
865                 nfs_pub.np_valid = 0;
866         }
867
868         TAILQ_INIT(&nfssvc_sockhead);
869         nfssvc_sockhead_flag &= ~SLP_INIT;
870         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
871                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
872                 wakeup((caddr_t)&nfssvc_sockhead);
873         }
874
875         TAILQ_INIT(&nfsd_head);
876         nfsd_head_flag &= ~NFSD_CHECKSLP;
877
878         lwkt_reltoken(&nfs_token);
879
880 #if 0
881         nfs_udpsock = (struct nfssvc_sock *)
882             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
883         mtx_init(&nfs_udpsock->ns_solock);
884         STAILQ_INIT(&nfs_udpsock->ns_rec);
885         TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
886         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
887
888         nfs_cltpsock = (struct nfssvc_sock *)
889             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
890         mtx_init(&nfs_cltpsock->ns_solock);
891         STAILQ_INIT(&nfs_cltpsock->ns_rec);
892         TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
893         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
894 #endif
895 }
896
897 /*
898  * Add entries to the server monitor log.
899  */
900 static void
901 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
902 {
903         struct drt *rt;
904
905         rt = &nfsdrt.drt[nfsdrt.pos];
906         if (cacherep == RC_DOIT)
907                 rt->flag = 0;
908         else if (cacherep == RC_REPLY)
909                 rt->flag = DRT_CACHEREPLY;
910         else
911                 rt->flag = DRT_CACHEDROP;
912         if (sotype == SOCK_STREAM)
913                 rt->flag |= DRT_TCP;
914         if (nd->nd_flag & ND_NFSV3)
915                 rt->flag |= DRT_NFSV3;
916         rt->proc = nd->nd_procnum;
917         if (nd->nd_nam->sa_family == AF_INET)
918             rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
919         else
920             rt->ipadr = INADDR_ANY;
921         rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
922         getmicrotime(&rt->tstamp);
923         nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
924 }
925 #endif /* NFS_NOSERVER */
926
927 static int nfs_defect = 0;
928 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
929
930 /*
931  * Get an authorization string for the uid by having the mount_nfs sitting
932  * on this mount point porpous out of the kernel and do it.
933  */
934 int
935 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
936             struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
937             int *verf_len, NFSKERBKEY_T key /* return session key */)
938 {
939         int error = 0;
940
941         while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
942                 nmp->nm_state |= NFSSTA_WANTAUTH;
943                 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
944                         "nfsauth1", 2 * hz);
945                 error = nfs_sigintr(nmp, rep, rep->r_td);
946                 if (error) {
947                         nmp->nm_state &= ~NFSSTA_WANTAUTH;
948                         return (error);
949                 }
950         }
951         nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
952         nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
953         nmp->nm_authlen = RPCAUTH_MAXSIZ;
954         nmp->nm_verfstr = verf_str;
955         nmp->nm_verflen = *verf_len;
956         nmp->nm_authuid = cred->cr_uid;
957         wakeup((caddr_t)&nmp->nm_authstr);
958
959         /*
960          * And wait for mount_nfs to do its stuff.
961          */
962         while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
963                 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
964                         "nfsauth2", 2 * hz);
965                 error = nfs_sigintr(nmp, rep, rep->r_td);
966         }
967         if (nmp->nm_state & NFSSTA_AUTHERR) {
968                 nmp->nm_state &= ~NFSSTA_AUTHERR;
969                 error = EAUTH;
970         }
971         if (error)
972                 kfree((caddr_t)*auth_str, M_TEMP);
973         else {
974                 *auth_len = nmp->nm_authlen;
975                 *verf_len = nmp->nm_verflen;
976                 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
977         }
978         nmp->nm_state &= ~NFSSTA_HASAUTH;
979         nmp->nm_state |= NFSSTA_WAITAUTH;
980         if (nmp->nm_state & NFSSTA_WANTAUTH) {
981                 nmp->nm_state &= ~NFSSTA_WANTAUTH;
982                 wakeup((caddr_t)&nmp->nm_authtype);
983         }
984         return (error);
985 }
986
987 /*
988  * Get a nickname authenticator and verifier.
989  */
990 int
991 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
992                 int *auth_len, char *verf_str, int verf_len)
993 {
994         struct nfsuid *nuidp;
995         u_int32_t *nickp, *verfp;
996         struct timeval ktvin, ktvout;
997
998 #ifdef DIAGNOSTIC
999         if (verf_len < (4 * NFSX_UNSIGNED))
1000                 panic("nfs_getnickauth verf too small");
1001 #endif
1002         for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
1003             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1004                 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
1005                         break;
1006         }
1007         if (!nuidp || nuidp->nu_expire < time_second)
1008                 return (EACCES);
1009
1010         /*
1011          * Move to the end of the lru list (end of lru == most recently used).
1012          */
1013         TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1014         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1015
1016         nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1017         *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1018         *nickp = txdr_unsigned(nuidp->nu_nickname);
1019         *auth_str = (char *)nickp;
1020         *auth_len = 2 * NFSX_UNSIGNED;
1021
1022         /*
1023          * Now we must encrypt the verifier and package it up.
1024          */
1025         verfp = (u_int32_t *)verf_str;
1026         *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1027         if (time_second > nuidp->nu_timestamp.tv_sec ||
1028             (time_second == nuidp->nu_timestamp.tv_sec &&
1029              time_second > nuidp->nu_timestamp.tv_usec))
1030                 getmicrotime(&nuidp->nu_timestamp);
1031         else
1032                 nuidp->nu_timestamp.tv_usec++;
1033         ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
1034         ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1035
1036         /*
1037          * Now encrypt the timestamp verifier in ecb mode using the session
1038          * key.
1039          */
1040 #ifdef NFSKERB
1041         XXX
1042 #else
1043         ktvout.tv_sec = 0;
1044         ktvout.tv_usec = 0;
1045 #endif
1046
1047         *verfp++ = ktvout.tv_sec;
1048         *verfp++ = ktvout.tv_usec;
1049         *verfp = 0;
1050         return (0);
1051 }
1052
1053 /*
1054  * Save the current nickname in a hash list entry on the mount point.
1055  */
1056 int
1057 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1058                  NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1059                  struct mbuf *mrep)
1060 {
1061         struct nfsuid *nuidp;
1062         u_int32_t *tl;
1063         struct timeval ktvin, ktvout;
1064         u_int32_t nick;
1065         int deltasec, error = 0;
1066         struct nfsm_info info;
1067
1068         info.md = *mdp;
1069         info.dpos = *dposp;
1070         info.mrep = mrep;
1071
1072         if (len == (3 * NFSX_UNSIGNED)) {
1073                 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
1074                 ktvin.tv_sec = *tl++;
1075                 ktvin.tv_usec = *tl++;
1076                 nick = fxdr_unsigned(u_int32_t, *tl);
1077
1078                 /*
1079                  * Decrypt the timestamp in ecb mode.
1080                  */
1081 #ifdef NFSKERB
1082                 XXX
1083 #else
1084                 ktvout.tv_sec = 0;
1085                 ktvout.tv_usec = 0;
1086 #endif
1087                 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1088                 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1089                 deltasec = time_second - ktvout.tv_sec;
1090                 if (deltasec < 0)
1091                         deltasec = -deltasec;
1092                 /*
1093                  * If ok, add it to the hash list for the mount point.
1094                  */
1095                 if (deltasec <= NFS_KERBCLOCKSKEW) {
1096                         if (nmp->nm_numuids < nuidhash_max) {
1097                                 nmp->nm_numuids++;
1098                                 nuidp = (struct nfsuid *)
1099                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
1100                                         M_WAITOK);
1101                         } else {
1102                                 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1103                                 LIST_REMOVE(nuidp, nu_hash);
1104                                 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1105                                         nu_lru);
1106                         }
1107                         nuidp->nu_flag = 0;
1108                         nuidp->nu_cr.cr_uid = cred->cr_uid;
1109                         nuidp->nu_expire = time_second + NFS_KERBTTL;
1110                         nuidp->nu_timestamp = ktvout;
1111                         nuidp->nu_nickname = nick;
1112                         bcopy(key, nuidp->nu_key, sizeof (key));
1113                         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1114                                 nu_lru);
1115                         LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1116                                 nuidp, nu_hash);
1117                 }
1118         } else {
1119                 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
1120         }
1121 nfsmout:
1122         *mdp = info.md;
1123         *dposp = info.dpos;
1124         return (error);
1125 }