network - Tokenize NFS, fix MP races
[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 <sys/mutex.h>
62 #include <vm/vm_zone.h>
63
64 #include <sys/mutex2.h>
65 #include <sys/thread2.h>
66
67 #include <netinet/in.h>
68 #include <netinet/tcp.h>
69 #include "xdr_subs.h"
70 #include "rpcv2.h"
71 #include "nfsproto.h"
72 #include "nfs.h"
73 #include "nfsm_subs.h"
74 #include "nfsrvcache.h"
75 #include "nfsmount.h"
76 #include "nfsnode.h"
77 #include "nfsrtt.h"
78
79 #include <sys/thread2.h>
80
81 static MALLOC_DEFINE(M_NFSSVC, "NFS srvsock", "Nfs server structure");
82
83 static int nuidhash_max = NFS_MAXUIDHASH;
84
85 #ifndef NFS_NOSERVER
86 static void     nfsrv_zapsock (struct nfssvc_sock *slp);
87 #endif
88
89 #define TRUE    1
90 #define FALSE   0
91
92 SYSCTL_DECL(_vfs_nfs);
93
94 #ifndef NFS_NOSERVER
95 int nfsd_waiting = 0;
96 static struct nfsdrt nfsdrt;
97 static int nfs_numnfsd = 0;
98 static void     nfsd_rt (int sotype, struct nfsrv_descript *nd,
99                              int cacherep);
100 static int      nfssvc_addsock (struct file *, struct sockaddr *,
101                                     struct thread *);
102 static int      nfssvc_nfsd (struct nfsd_srvargs *,caddr_t,struct thread *);
103
104 static int nfs_privport = 0;
105 SYSCTL_INT(_vfs_nfs, NFS_NFSPRIVPORT, nfs_privport, CTLFLAG_RW, &nfs_privport, 0, "");
106 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay, CTLFLAG_RW, &nfsrvw_procrastinate, 0, "");
107 SYSCTL_INT(_vfs_nfs, OID_AUTO, gatherdelay_v3, CTLFLAG_RW, &nfsrvw_procrastinate_v3, 0, "");
108 int     nfs_soreserve = NFS_MAXPACKET * NFS_MAXASYNCBIO;
109 SYSCTL_INT(_vfs_nfs, OID_AUTO, soreserve, CTLFLAG_RW, &nfs_soreserve, 0, "");
110
111 /*
112  * NFS server system calls
113  */
114
115 #endif /* NFS_NOSERVER */
116 /*
117  * nfssvc_args(int flag, caddr_t argp)
118  *
119  * Nfs server psuedo system call for the nfsd's
120  * Based on the flag value it either:
121  * - adds a socket to the selection list
122  * - remains in the kernel as an nfsd
123  * - remains in the kernel as an nfsiod
124  *
125  * MPALMOSTSAFE
126  */
127 int
128 sys_nfssvc(struct nfssvc_args *uap)
129 {
130 #ifndef NFS_NOSERVER
131         struct nlookupdata nd;
132         struct file *fp;
133         struct sockaddr *nam;
134         struct nfsd_args nfsdarg;
135         struct nfsd_srvargs nfsd_srvargs, *nsd = &nfsd_srvargs;
136         struct nfsd_cargs ncd;
137         struct nfsd *nfsd;
138         struct nfssvc_sock *slp;
139         struct nfsuid *nuidp;
140         struct nfsmount *nmp;
141         struct vnode *vp;
142 #endif /* NFS_NOSERVER */
143         int error;
144         struct thread *td = curthread;
145
146         /*
147          * Must be super user
148          */
149         error = priv_check(td, PRIV_ROOT);
150         if (error)
151                 return (error);
152
153         lwkt_gettoken(&nfs_token);
154
155         while (nfssvc_sockhead_flag & SLP_INIT) {
156                 nfssvc_sockhead_flag |= SLP_WANTINIT;
157                 tsleep((caddr_t)&nfssvc_sockhead, 0, "nfsd init", 0);
158         }
159         if (uap->flag & NFSSVC_BIOD)
160                 error = ENXIO;          /* no longer need nfsiod's */
161 #ifdef NFS_NOSERVER
162         else
163                 error = ENXIO;
164 #else /* !NFS_NOSERVER */
165         else if (uap->flag & NFSSVC_MNTD) {
166                 error = copyin(uap->argp, (caddr_t)&ncd, sizeof (ncd));
167                 if (error)
168                         goto done;
169                 vp = NULL;
170                 error = nlookup_init(&nd, ncd.ncd_dirp, UIO_USERSPACE, 
171                                         NLC_FOLLOW);
172                 if (error == 0)
173                         error = nlookup(&nd);
174                 if (error == 0)
175                         error = cache_vget(&nd.nl_nch, nd.nl_cred, LK_EXCLUSIVE, &vp);   
176                 nlookup_done(&nd);
177                 if (error)
178                         goto done;
179
180                 if ((vp->v_flag & VROOT) == 0)
181                         error = EINVAL;
182                 nmp = VFSTONFS(vp->v_mount);
183                 vput(vp);
184                 if (error)
185                         goto done;
186                 if ((nmp->nm_state & NFSSTA_MNTD) &&
187                         (uap->flag & NFSSVC_GOTAUTH) == 0) {
188                         error = 0;
189                         goto done;
190                 }
191                 nmp->nm_state |= NFSSTA_MNTD;
192                 error = nfs_clientd(nmp, td->td_ucred, &ncd, uap->flag,
193                                     uap->argp, td);
194         } else if (uap->flag & NFSSVC_ADDSOCK) {
195                 error = copyin(uap->argp, (caddr_t)&nfsdarg, sizeof(nfsdarg));
196                 if (error)
197                         goto done;
198                 error = holdsock(td->td_proc->p_fd, nfsdarg.sock, &fp);
199                 if (error)
200                         goto done;
201                 /*
202                  * Get the client address for connected sockets.
203                  */
204                 if (nfsdarg.name == NULL || nfsdarg.namelen == 0)
205                         nam = NULL;
206                 else {
207                         error = getsockaddr(&nam, nfsdarg.name,
208                                             nfsdarg.namelen);
209                         if (error) {
210                                 fdrop(fp);
211                                 goto done;
212                         }
213                 }
214                 error = nfssvc_addsock(fp, nam, td);
215                 fdrop(fp);
216         } else {
217                 error = copyin(uap->argp, (caddr_t)nsd, sizeof (*nsd));
218                 if (error)
219                         goto done;
220                 if ((uap->flag & NFSSVC_AUTHIN) &&
221                     ((nfsd = nsd->nsd_nfsd)) != NULL &&
222                     (nfsd->nfsd_slp->ns_flag & SLP_VALID)) {
223                         slp = nfsd->nfsd_slp;
224
225                         /*
226                          * First check to see if another nfsd has already
227                          * added this credential.
228                          */
229                         for (nuidp = NUIDHASH(slp,nsd->nsd_cr.cr_uid)->lh_first;
230                             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
231                                 if (nuidp->nu_cr.cr_uid == nsd->nsd_cr.cr_uid &&
232                                     (!nfsd->nfsd_nd->nd_nam2 ||
233                                      netaddr_match(NU_NETFAM(nuidp),
234                                      &nuidp->nu_haddr, nfsd->nfsd_nd->nd_nam2)))
235                                         break;
236                         }
237                         if (nuidp) {
238                             nfsrv_setcred(&nuidp->nu_cr,&nfsd->nfsd_nd->nd_cr);
239                             nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
240                         } else {
241                             /*
242                              * Nope, so we will.
243                              */
244                             if (slp->ns_numuids < nuidhash_max) {
245                                 slp->ns_numuids++;
246                                 nuidp = (struct nfsuid *)
247                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
248                                         M_WAITOK);
249                             } else
250                                 nuidp = NULL;
251                             if ((slp->ns_flag & SLP_VALID) == 0) {
252                                 if (nuidp)
253                                     kfree((caddr_t)nuidp, M_NFSUID);
254                             } else {
255                                 if (nuidp == NULL) {
256                                     nuidp = TAILQ_FIRST(&slp->ns_uidlruhead);
257                                     LIST_REMOVE(nuidp, nu_hash);
258                                     TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp,
259                                         nu_lru);
260                                     if (nuidp->nu_flag & NU_NAM)
261                                         FREE(nuidp->nu_nam, M_SONAME);
262                                 }
263                                 nuidp->nu_flag = 0;
264                                 nuidp->nu_cr = nsd->nsd_cr;
265                                 if (nuidp->nu_cr.cr_ngroups > NGROUPS)
266                                     nuidp->nu_cr.cr_ngroups = NGROUPS;
267                                 nuidp->nu_cr.cr_ref = 1;
268                                 nuidp->nu_timestamp = nsd->nsd_timestamp;
269                                 nuidp->nu_expire = time_second + nsd->nsd_ttl;
270                                 /*
271                                  * and save the session key in nu_key.
272                                  */
273                                 bcopy(nsd->nsd_key, nuidp->nu_key,
274                                     sizeof (nsd->nsd_key));
275                                 if (nfsd->nfsd_nd->nd_nam2) {
276                                     struct sockaddr_in *saddr;
277
278                                     saddr = (struct sockaddr_in *)
279                                             nfsd->nfsd_nd->nd_nam2;
280                                     switch (saddr->sin_family) {
281                                     case AF_INET:
282                                         nuidp->nu_flag |= NU_INETADDR;
283                                         nuidp->nu_inetaddr =
284                                              saddr->sin_addr.s_addr;
285                                         break;
286                                     case AF_ISO:
287                                     default:
288                                         nuidp->nu_flag |= NU_NAM;
289                                         nuidp->nu_nam = 
290                                           dup_sockaddr(nfsd->nfsd_nd->nd_nam2);
291                                         break;
292                                     };
293                                 }
294                                 TAILQ_INSERT_TAIL(&slp->ns_uidlruhead, nuidp,
295                                         nu_lru);
296                                 LIST_INSERT_HEAD(NUIDHASH(slp, nsd->nsd_uid),
297                                         nuidp, nu_hash);
298                                 nfsrv_setcred(&nuidp->nu_cr,
299                                     &nfsd->nfsd_nd->nd_cr);
300                                 nfsd->nfsd_nd->nd_flag |= ND_KERBFULL;
301                             }
302                         }
303                 }
304                 if ((uap->flag & NFSSVC_AUTHINFAIL) && (nfsd = nsd->nsd_nfsd))
305                         nfsd->nfsd_flag |= NFSD_AUTHFAIL;
306                 error = nfssvc_nfsd(nsd, uap->argp, td);
307         }
308 #endif /* NFS_NOSERVER */
309         if (error == EINTR || error == ERESTART)
310                 error = 0;
311 done:
312         lwkt_reltoken(&nfs_token);
313         return (error);
314 }
315
316 #ifndef NFS_NOSERVER
317 /*
318  * Adds a socket to the list for servicing by nfsds.
319  */
320 static int
321 nfssvc_addsock(struct file *fp, struct sockaddr *mynam, struct thread *td)
322 {
323         int siz;
324         struct nfssvc_sock *slp;
325         struct socket *so;
326         int error;
327
328         so = (struct socket *)fp->f_data;
329 #if 0
330         tslp = NULL;
331         /*
332          * Add it to the list, as required.
333          */
334         if (so->so_proto->pr_protocol == IPPROTO_UDP) {
335                 tslp = nfs_udpsock;
336                 if (tslp->ns_flag & SLP_VALID) {
337                         if (mynam != NULL)
338                                 FREE(mynam, M_SONAME);
339                         return (EPERM);
340                 }
341         }
342 #endif
343         /*
344          * Reserve buffer space in the socket.  Note that due to bugs in
345          * Linux's delayed-ack code, serious performance degredation may
346          * occur with linux hosts if the minimum is used.
347          *
348          * NFS sockets are not limited to the standard sb_max or by
349          * resource limits.
350          */
351         if (so->so_type == SOCK_STREAM)
352                 siz = NFS_MAXPACKET + sizeof (u_long);
353         else
354                 siz = NFS_MAXPACKET;
355         if (siz < nfs_soreserve)
356             siz = nfs_soreserve;
357
358         error = soreserve(so, siz, siz, NULL);
359         if (error) {
360                 if (mynam != NULL)
361                         FREE(mynam, M_SONAME);
362                 return (error);
363         }
364
365         /*
366          * Set protocol specific options { for now TCP only } and
367          * reserve some space. For datagram sockets, this can get called
368          * repeatedly for the same socket, but that isn't harmful.
369          */
370         if (so->so_type == SOCK_STREAM) {
371                 struct sockopt sopt;
372                 int val;
373
374                 bzero(&sopt, sizeof sopt);
375                 sopt.sopt_level = SOL_SOCKET;
376                 sopt.sopt_name = SO_KEEPALIVE;
377                 sopt.sopt_val = &val;
378                 sopt.sopt_valsize = sizeof val;
379                 val = 1;
380                 sosetopt(so, &sopt);
381         }
382         if (so->so_proto->pr_domain->dom_family == AF_INET &&
383             so->so_proto->pr_protocol == IPPROTO_TCP) {
384                 struct sockopt sopt;
385                 int val;
386
387                 bzero(&sopt, sizeof sopt);
388                 sopt.sopt_level = IPPROTO_TCP;
389                 sopt.sopt_name = TCP_NODELAY;
390                 sopt.sopt_val = &val;
391                 sopt.sopt_valsize = sizeof val;
392                 val = 1;
393                 sosetopt(so, &sopt);
394         }
395         atomic_clear_int(&so->so_rcv.ssb_flags, SSB_NOINTR);
396         so->so_rcv.ssb_timeo = 0;
397         atomic_clear_int(&so->so_snd.ssb_flags, SSB_NOINTR);
398         so->so_snd.ssb_timeo = 0;
399
400         slp = (struct nfssvc_sock *)kmalloc(sizeof (struct nfssvc_sock),
401             M_NFSSVC, M_WAITOK | M_ZERO);
402         mtx_init(&slp->ns_solock);
403         STAILQ_INIT(&slp->ns_rec);
404         TAILQ_INIT(&slp->ns_uidlruhead);
405         lwkt_token_init(&slp->ns_token, 1, "nfssrv_token");
406
407         lwkt_gettoken(&nfs_token);
408         TAILQ_INSERT_TAIL(&nfssvc_sockhead, slp, ns_chain);
409         nfsrv_slpref(slp);
410         lwkt_gettoken(&slp->ns_token);
411
412         slp->ns_so = so;
413         slp->ns_nam = mynam;
414         fp->f_count++;
415         slp->ns_fp = fp;
416
417         so->so_upcallarg = (caddr_t)slp;
418         so->so_upcall = nfsrv_rcv_upcall;
419         atomic_set_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
420         slp->ns_flag = (SLP_VALID | SLP_NEEDQ);
421         nfsrv_wakenfsd(slp, 1);
422
423         lwkt_reltoken(&slp->ns_token);
424         lwkt_reltoken(&nfs_token);
425
426         return (0);
427 }
428
429 /*
430  * Called by nfssvc() for nfsds. Just loops around servicing rpc requests
431  * until it is killed by a signal.
432  */
433 static int
434 nfssvc_nfsd(struct nfsd_srvargs *nsd, caddr_t argp, struct thread *td)
435 {
436         int siz;
437         struct nfssvc_sock *slp;
438         struct nfsd *nfsd = nsd->nsd_nfsd;
439         struct nfsrv_descript *nd = NULL;
440         struct mbuf *m, *mreq;
441         int error = 0, cacherep, sotype, writes_todo;
442         int procrastinate;
443         u_quad_t cur_usec;
444
445 #ifndef nolint
446         cacherep = RC_DOIT;
447         writes_todo = 0;
448 #endif
449         lwkt_gettoken(&nfs_token);
450
451         if (nfsd == NULL) {
452                 nsd->nsd_nfsd = nfsd = (struct nfsd *)
453                         kmalloc(sizeof (struct nfsd), M_NFSD, M_WAITOK|M_ZERO);
454                 nfsd->nfsd_td = td;
455                 TAILQ_INSERT_TAIL(&nfsd_head, nfsd, nfsd_chain);
456                 nfs_numnfsd++;
457         }
458
459         /*
460          * Loop getting rpc requests until SIGKILL.
461          */
462         for (;;) {
463                 if ((nfsd->nfsd_flag & NFSD_REQINPROG) == 0) {
464                         while (nfsd->nfsd_slp == NULL &&
465                             (nfsd_head_flag & NFSD_CHECKSLP) == 0) {
466                                 nfsd->nfsd_flag |= NFSD_WAITING;
467                                 nfsd_waiting++;
468                                 error = tsleep((caddr_t)nfsd, PCATCH, "nfsd", 0);
469                                 nfsd_waiting--;
470                                 if (error)
471                                         goto done;
472                         }
473                         if (nfsd->nfsd_slp == NULL &&
474                             (nfsd_head_flag & NFSD_CHECKSLP) != 0) {
475                                 TAILQ_FOREACH(slp, &nfssvc_sockhead, ns_chain) {
476                                     if ((slp->ns_flag & (SLP_VALID | SLP_DOREC))
477                                         == (SLP_VALID | SLP_DOREC)) {
478                                             slp->ns_flag &= ~SLP_DOREC;
479                                             nfsrv_slpref(slp);
480                                             nfsd->nfsd_slp = slp;
481                                             break;
482                                     }
483                                 }
484                                 if (slp == 0)
485                                         nfsd_head_flag &= ~NFSD_CHECKSLP;
486                         }
487                         if ((slp = nfsd->nfsd_slp) == NULL)
488                                 continue;
489
490                         lwkt_reltoken(&nfs_token);
491                         lwkt_gettoken(&slp->ns_token);
492
493                         if (slp->ns_flag & SLP_VALID) {
494                                 if (slp->ns_flag & SLP_DISCONN)
495                                         nfsrv_zapsock(slp);
496                                 else if (slp->ns_flag & SLP_NEEDQ) {
497                                         slp->ns_flag &= ~SLP_NEEDQ;
498                                         (void) nfs_slplock(slp, 1);
499                                         nfsrv_rcv(slp->ns_so, (caddr_t)slp,
500                                                 MB_WAIT);
501                                         nfs_slpunlock(slp);
502                                 }
503                                 error = nfsrv_dorec(slp, nfsd, &nd);
504                                 cur_usec = nfs_curusec();
505                                 if (error && slp->ns_tq.lh_first &&
506                                     slp->ns_tq.lh_first->nd_time <= cur_usec) {
507                                         error = 0;
508                                         cacherep = RC_DOIT;
509                                         writes_todo = 1;
510                                 } else
511                                         writes_todo = 0;
512                                 nfsd->nfsd_flag |= NFSD_REQINPROG;
513                         }
514                 } else {
515                         error = 0;
516                         slp = nfsd->nfsd_slp;
517                         lwkt_reltoken(&nfs_token);
518                         lwkt_gettoken(&slp->ns_token);
519                 }
520
521                 /*
522                  * nfs_token not held here.  slp token is held.
523                  */
524                 if (error || (slp->ns_flag & SLP_VALID) == 0) {
525                         if (nd) {
526                                 kfree((caddr_t)nd, M_NFSRVDESC);
527                                 nd = NULL;
528                         }
529                         nfsd->nfsd_slp = NULL;
530                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
531                         lwkt_reltoken(&slp->ns_token);
532                         lwkt_gettoken(&nfs_token);
533                         nfsrv_slpderef(slp);
534                         continue;
535                 }
536
537                 /*
538                  * nfs_token not held here.  slp token is held.
539                  */
540                 sotype = slp->ns_so->so_type;
541                 if (nd) {
542                     getmicrotime(&nd->nd_starttime);
543                     if (nd->nd_nam2)
544                         nd->nd_nam = nd->nd_nam2;
545                     else
546                         nd->nd_nam = slp->ns_nam;
547
548                     /*
549                      * Check to see if authorization is needed.
550                      */
551                     if (nfsd->nfsd_flag & NFSD_NEEDAUTH) {
552                         nfsd->nfsd_flag &= ~NFSD_NEEDAUTH;
553                         nsd->nsd_haddr = 
554                                 ((struct sockaddr_in *)
555                                  nd->nd_nam)->sin_addr.s_addr;
556                         nsd->nsd_authlen = nfsd->nfsd_authlen;
557                         nsd->nsd_verflen = nfsd->nfsd_verflen;
558                         if (!copyout(nfsd->nfsd_authstr,nsd->nsd_authstr,
559                                 nfsd->nfsd_authlen) &&
560                             !copyout(nfsd->nfsd_verfstr, nsd->nsd_verfstr,
561                                 nfsd->nfsd_verflen) &&
562                             !copyout((caddr_t)nsd, argp, sizeof (*nsd)))
563                             lwkt_reltoken(&slp->ns_token);
564                             return (ENEEDAUTH);
565                         cacherep = RC_DROPIT;
566                     } else {
567                         cacherep = nfsrv_getcache(nd, slp, &mreq);
568                     }
569
570                     if (nfsd->nfsd_flag & NFSD_AUTHFAIL) {
571                         nfsd->nfsd_flag &= ~NFSD_AUTHFAIL;
572                         nd->nd_procnum = NFSPROC_NOOP;
573                         nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
574                         cacherep = RC_DOIT;
575                     } else if (nfs_privport) {
576                         /* Check if source port is privileged */
577                         u_short port;
578                         struct sockaddr *nam = nd->nd_nam;
579                         struct sockaddr_in *sin;
580
581                         sin = (struct sockaddr_in *)nam;
582                         port = ntohs(sin->sin_port);
583                         if (port >= IPPORT_RESERVED && 
584                             nd->nd_procnum != NFSPROC_NULL) {
585                             nd->nd_procnum = NFSPROC_NOOP;
586                             nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
587                             cacherep = RC_DOIT;
588                             kprintf("NFS request from unprivileged port (%s:%d)\n",
589                                    inet_ntoa(sin->sin_addr), port);
590                         }
591                     }
592
593                 }
594
595                 /*
596                  * Loop to get all the write rpc replies that have been
597                  * gathered together.
598                  *
599                  * nfs_token not held here.  slp token is held.
600                  */
601                 do {
602                     switch (cacherep) {
603                     case RC_DOIT:
604                         if (nd && (nd->nd_flag & ND_NFSV3))
605                             procrastinate = nfsrvw_procrastinate_v3;
606                         else
607                             procrastinate = nfsrvw_procrastinate;
608                         if (writes_todo || (nd->nd_procnum == NFSPROC_WRITE &&
609                             procrastinate > 0)
610                         ) {
611                             error = nfsrv_writegather(&nd, slp,
612                                 nfsd->nfsd_td, &mreq);
613                         } else {
614                             error = (*(nfsrv3_procs[nd->nd_procnum]))(nd,
615                                 slp, nfsd->nfsd_td, &mreq);
616                         }
617                         if (mreq == NULL)
618                                 break;
619                         if (error != 0 && error != NFSERR_RETVOID) {
620                                 if (nd->nd_procnum != NQNFSPROC_VACATED)
621                                         nfsstats.srv_errs++;
622                                 nfsrv_updatecache(nd, FALSE, mreq);
623                                 if (nd->nd_nam2)
624                                         FREE(nd->nd_nam2, M_SONAME);
625                                 break;
626                         }
627                         nfsstats.srvrpccnt[nd->nd_procnum]++;
628                         nfsrv_updatecache(nd, TRUE, mreq);
629                         nd->nd_mrep = NULL;
630                     case RC_REPLY:
631                         m = mreq;
632                         siz = 0;
633                         while (m) {
634                                 siz += m->m_len;
635                                 m = m->m_next;
636                         }
637                         if (siz <= 0 || siz > NFS_MAXPACKET) {
638                                 kprintf("mbuf siz=%d\n",siz);
639                                 panic("Bad nfs svc reply");
640                         }
641                         m = mreq;
642                         m->m_pkthdr.len = siz;
643                         m->m_pkthdr.rcvif = NULL;
644                         /*
645                          * For stream protocols, prepend a Sun RPC
646                          * Record Mark.
647                          */
648                         if (sotype == SOCK_STREAM) {
649                                 M_PREPEND(m, NFSX_UNSIGNED, MB_WAIT);
650                                 if (m == NULL)
651                                         return (ENOBUFS);
652                                 *mtod(m, u_int32_t *) = htonl(0x80000000 | siz);
653                         }
654                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
655                                 (void) nfs_slplock(slp, 1);
656                         if (slp->ns_flag & SLP_VALID)
657                             error = nfs_send(slp->ns_so, nd->nd_nam2, m, NULL);
658                         else {
659                             error = EPIPE;
660                             m_freem(m);
661                         }
662                         if (nfsrtton)
663                                 nfsd_rt(sotype, nd, cacherep);
664                         if (nd->nd_nam2)
665                                 FREE(nd->nd_nam2, M_SONAME);
666                         if (nd->nd_mrep)
667                                 m_freem(nd->nd_mrep);
668                         if (error == EPIPE)
669                                 nfsrv_zapsock(slp);
670                         if (slp->ns_so->so_proto->pr_flags & PR_CONNREQUIRED)
671                                 nfs_slpunlock(slp);
672                         if (error == EINTR || error == ERESTART) {
673                                 kfree((caddr_t)nd, M_NFSRVDESC);
674                                 lwkt_reltoken(&slp->ns_token);
675                                 lwkt_gettoken(&nfs_token);
676                                 nfsrv_slpderef(slp);
677                                 goto done;
678                         }
679                         break;
680                     case RC_DROPIT:
681                         if (nfsrtton)
682                                 nfsd_rt(sotype, nd, cacherep);
683                         m_freem(nd->nd_mrep);
684                         if (nd->nd_nam2)
685                                 FREE(nd->nd_nam2, M_SONAME);
686                         break;
687                     };
688                     if (nd) {
689                         FREE((caddr_t)nd, M_NFSRVDESC);
690                         nd = NULL;
691                     }
692
693                     /*
694                      * Check to see if there are outstanding writes that
695                      * need to be serviced.
696                      */
697                     cur_usec = nfs_curusec();
698                     if (slp->ns_tq.lh_first &&
699                         slp->ns_tq.lh_first->nd_time <= cur_usec) {
700                         cacherep = RC_DOIT;
701                         writes_todo = 1;
702                     } else {
703                         writes_todo = 0;
704                     }
705                 } while (writes_todo);
706
707                 /*
708                  * nfs_token not held here.  slp token is held.
709                  */
710                 if (nfsrv_dorec(slp, nfsd, &nd)) {
711                         nfsd->nfsd_flag &= ~NFSD_REQINPROG;
712                         nfsd->nfsd_slp = NULL;
713                         lwkt_reltoken(&slp->ns_token);
714                         lwkt_gettoken(&nfs_token);
715                         nfsrv_slpderef(slp);
716                 } else {
717                         lwkt_reltoken(&slp->ns_token);
718                         lwkt_gettoken(&nfs_token);
719                 }
720         }
721 done:
722         TAILQ_REMOVE(&nfsd_head, nfsd, nfsd_chain);
723         kfree((caddr_t)nfsd, M_NFSD);
724         nsd->nsd_nfsd = NULL;
725         if (--nfs_numnfsd == 0)
726                 nfsrv_init(TRUE);       /* Reinitialize everything */
727
728         lwkt_reltoken(&nfs_token);
729         return (error);
730 }
731
732 /*
733  * Shut down a socket associated with an nfssvc_sock structure.
734  * Should be called with the send lock set, if required.
735  * The trick here is to increment the sref at the start, so that the nfsds
736  * will stop using it and clear ns_flag at the end so that it will not be
737  * reassigned during cleanup.
738  */
739 static void
740 nfsrv_zapsock(struct nfssvc_sock *slp)
741 {
742         struct nfsuid *nuidp, *nnuidp;
743         struct nfsrv_descript *nwp, *nnwp;
744         struct socket *so;
745         struct file *fp;
746         struct nfsrv_rec *rec;
747
748         slp->ns_flag &= ~SLP_ALLFLAGS;
749         fp = slp->ns_fp;
750         if (fp) {
751                 slp->ns_fp = NULL;
752                 so = slp->ns_so;
753                 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_UPCALL);
754                 so->so_upcall = NULL;
755                 so->so_upcallarg = NULL;
756                 soshutdown(so, SHUT_RDWR);
757                 closef(fp, NULL);
758                 if (slp->ns_nam)
759                         FREE(slp->ns_nam, M_SONAME);
760                 m_freem(slp->ns_raw);
761                 while ((rec = STAILQ_FIRST(&slp->ns_rec)) != NULL) {
762                         --slp->ns_numrec;
763                         STAILQ_REMOVE_HEAD(&slp->ns_rec, nr_link);
764                         if (rec->nr_address)
765                                 FREE(rec->nr_address, M_SONAME);
766                         m_freem(rec->nr_packet);
767                         kfree(rec, M_NFSRVDESC);
768                 }
769                 TAILQ_FOREACH_MUTABLE(nuidp, &slp->ns_uidlruhead, nu_lru,
770                                       nnuidp) {
771                         LIST_REMOVE(nuidp, nu_hash);
772                         TAILQ_REMOVE(&slp->ns_uidlruhead, nuidp, nu_lru);
773                         if (nuidp->nu_flag & NU_NAM)
774                                 FREE(nuidp->nu_nam, M_SONAME);
775                         kfree((caddr_t)nuidp, M_NFSUID);
776                 }
777                 crit_enter();
778                 for (nwp = slp->ns_tq.lh_first; nwp; nwp = nnwp) {
779                         nnwp = nwp->nd_tq.le_next;
780                         LIST_REMOVE(nwp, nd_tq);
781                         kfree((caddr_t)nwp, M_NFSRVDESC);
782                 }
783                 LIST_INIT(&slp->ns_tq);
784                 crit_exit();
785         }
786 }
787
788 /*
789  * Derefence a server socket structure. If it has no more references and
790  * is no longer valid, you can throw it away.
791  *
792  * Must be holding nfs_token!
793  */
794 void
795 nfsrv_slpderef(struct nfssvc_sock *slp)
796 {
797         ASSERT_LWKT_TOKEN_HELD(&nfs_token);
798         if (--slp->ns_sref == 0 && (slp->ns_flag & SLP_VALID) == 0) {
799                 TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
800                 kfree((caddr_t)slp, M_NFSSVC);
801         }
802 }
803
804 void
805 nfsrv_slpref(struct nfssvc_sock *slp)
806 {
807         ASSERT_LWKT_TOKEN_HELD(&nfs_token);
808         ++slp->ns_sref;
809 }
810
811 /*
812  * Lock a socket against others.
813  *
814  * Returns 0 on failure, 1 on success.
815  */
816 int
817 nfs_slplock(struct nfssvc_sock *slp, int wait)
818 {
819         mtx_t mtx = &slp->ns_solock;
820
821         if (wait) {
822                 mtx_lock_ex(mtx, "nfsslplck", 0, 0);
823                 return(1);
824         } else if (mtx_lock_ex_try(mtx) == 0) {
825                 return(1);
826         } else {
827                 return(0);
828         }
829 }
830
831 /*
832  * Unlock the stream socket for others.
833  */
834 void
835 nfs_slpunlock(struct nfssvc_sock *slp)
836 {
837         mtx_t mtx = &slp->ns_solock;
838
839         mtx_unlock(mtx);
840 }
841
842 /*
843  * Initialize the data structures for the server.
844  * Handshake with any new nfsds starting up to avoid any chance of
845  * corruption.
846  */
847 void
848 nfsrv_init(int terminating)
849 {
850         struct nfssvc_sock *slp, *nslp;
851
852         lwkt_gettoken(&nfs_token);
853         if (nfssvc_sockhead_flag & SLP_INIT)
854                 panic("nfsd init");
855         nfssvc_sockhead_flag |= SLP_INIT;
856
857         if (terminating) {
858                 TAILQ_FOREACH_MUTABLE(slp, &nfssvc_sockhead, ns_chain, nslp) {
859                         if (slp->ns_flag & SLP_VALID)
860                                 nfsrv_zapsock(slp);
861                         TAILQ_REMOVE(&nfssvc_sockhead, slp, ns_chain);
862                         kfree((caddr_t)slp, M_NFSSVC);
863                 }
864                 nfsrv_cleancache();     /* And clear out server cache */
865         } else {
866                 nfs_pub.np_valid = 0;
867         }
868
869         TAILQ_INIT(&nfssvc_sockhead);
870         nfssvc_sockhead_flag &= ~SLP_INIT;
871         if (nfssvc_sockhead_flag & SLP_WANTINIT) {
872                 nfssvc_sockhead_flag &= ~SLP_WANTINIT;
873                 wakeup((caddr_t)&nfssvc_sockhead);
874         }
875
876         TAILQ_INIT(&nfsd_head);
877         nfsd_head_flag &= ~NFSD_CHECKSLP;
878
879         lwkt_reltoken(&nfs_token);
880
881 #if 0
882         nfs_udpsock = (struct nfssvc_sock *)
883             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
884         mtx_init(&nfs_udpsock->ns_solock);
885         STAILQ_INIT(&nfs_udpsock->ns_rec);
886         TAILQ_INIT(&nfs_udpsock->ns_uidlruhead);
887         TAILQ_INSERT_HEAD(&nfssvc_sockhead, nfs_udpsock, ns_chain);
888
889         nfs_cltpsock = (struct nfssvc_sock *)
890             kmalloc(sizeof (struct nfssvc_sock), M_NFSSVC, M_WAITOK | M_ZERO);
891         mtx_init(&nfs_cltpsock->ns_solock);
892         STAILQ_INIT(&nfs_cltpsock->ns_rec);
893         TAILQ_INIT(&nfs_cltpsock->ns_uidlruhead);
894         TAILQ_INSERT_TAIL(&nfssvc_sockhead, nfs_cltpsock, ns_chain);
895 #endif
896 }
897
898 /*
899  * Add entries to the server monitor log.
900  */
901 static void
902 nfsd_rt(int sotype, struct nfsrv_descript *nd, int cacherep)
903 {
904         struct drt *rt;
905
906         rt = &nfsdrt.drt[nfsdrt.pos];
907         if (cacherep == RC_DOIT)
908                 rt->flag = 0;
909         else if (cacherep == RC_REPLY)
910                 rt->flag = DRT_CACHEREPLY;
911         else
912                 rt->flag = DRT_CACHEDROP;
913         if (sotype == SOCK_STREAM)
914                 rt->flag |= DRT_TCP;
915         if (nd->nd_flag & ND_NFSV3)
916                 rt->flag |= DRT_NFSV3;
917         rt->proc = nd->nd_procnum;
918         if (nd->nd_nam->sa_family == AF_INET)
919             rt->ipadr = ((struct sockaddr_in *)nd->nd_nam)->sin_addr.s_addr;
920         else
921             rt->ipadr = INADDR_ANY;
922         rt->resptime = nfs_curusec() - (nd->nd_starttime.tv_sec * 1000000 + nd->nd_starttime.tv_usec);
923         getmicrotime(&rt->tstamp);
924         nfsdrt.pos = (nfsdrt.pos + 1) % NFSRTTLOGSIZ;
925 }
926 #endif /* NFS_NOSERVER */
927
928 static int nfs_defect = 0;
929 SYSCTL_INT(_vfs_nfs, OID_AUTO, defect, CTLFLAG_RW, &nfs_defect, 0, "");
930
931 /*
932  * Get an authorization string for the uid by having the mount_nfs sitting
933  * on this mount point porpous out of the kernel and do it.
934  */
935 int
936 nfs_getauth(struct nfsmount *nmp, struct nfsreq *rep,
937             struct ucred *cred, char **auth_str, int *auth_len, char *verf_str,
938             int *verf_len, NFSKERBKEY_T key /* return session key */)
939 {
940         int error = 0;
941
942         while ((nmp->nm_state & NFSSTA_WAITAUTH) == 0) {
943                 nmp->nm_state |= NFSSTA_WANTAUTH;
944                 (void) tsleep((caddr_t)&nmp->nm_authtype, 0,
945                         "nfsauth1", 2 * hz);
946                 error = nfs_sigintr(nmp, rep, rep->r_td);
947                 if (error) {
948                         nmp->nm_state &= ~NFSSTA_WANTAUTH;
949                         return (error);
950                 }
951         }
952         nmp->nm_state &= ~(NFSSTA_WAITAUTH | NFSSTA_WANTAUTH);
953         nmp->nm_authstr = *auth_str = (char *)kmalloc(RPCAUTH_MAXSIZ, M_TEMP, M_WAITOK);
954         nmp->nm_authlen = RPCAUTH_MAXSIZ;
955         nmp->nm_verfstr = verf_str;
956         nmp->nm_verflen = *verf_len;
957         nmp->nm_authuid = cred->cr_uid;
958         wakeup((caddr_t)&nmp->nm_authstr);
959
960         /*
961          * And wait for mount_nfs to do its stuff.
962          */
963         while ((nmp->nm_state & NFSSTA_HASAUTH) == 0 && error == 0) {
964                 (void) tsleep((caddr_t)&nmp->nm_authlen, 0,
965                         "nfsauth2", 2 * hz);
966                 error = nfs_sigintr(nmp, rep, rep->r_td);
967         }
968         if (nmp->nm_state & NFSSTA_AUTHERR) {
969                 nmp->nm_state &= ~NFSSTA_AUTHERR;
970                 error = EAUTH;
971         }
972         if (error)
973                 kfree((caddr_t)*auth_str, M_TEMP);
974         else {
975                 *auth_len = nmp->nm_authlen;
976                 *verf_len = nmp->nm_verflen;
977                 bcopy((caddr_t)nmp->nm_key, (caddr_t)key, sizeof (key));
978         }
979         nmp->nm_state &= ~NFSSTA_HASAUTH;
980         nmp->nm_state |= NFSSTA_WAITAUTH;
981         if (nmp->nm_state & NFSSTA_WANTAUTH) {
982                 nmp->nm_state &= ~NFSSTA_WANTAUTH;
983                 wakeup((caddr_t)&nmp->nm_authtype);
984         }
985         return (error);
986 }
987
988 /*
989  * Get a nickname authenticator and verifier.
990  */
991 int
992 nfs_getnickauth(struct nfsmount *nmp, struct ucred *cred, char **auth_str,
993                 int *auth_len, char *verf_str, int verf_len)
994 {
995         struct nfsuid *nuidp;
996         u_int32_t *nickp, *verfp;
997         struct timeval ktvin, ktvout;
998
999 #ifdef DIAGNOSTIC
1000         if (verf_len < (4 * NFSX_UNSIGNED))
1001                 panic("nfs_getnickauth verf too small");
1002 #endif
1003         for (nuidp = NMUIDHASH(nmp, cred->cr_uid)->lh_first;
1004             nuidp != 0; nuidp = nuidp->nu_hash.le_next) {
1005                 if (nuidp->nu_cr.cr_uid == cred->cr_uid)
1006                         break;
1007         }
1008         if (!nuidp || nuidp->nu_expire < time_second)
1009                 return (EACCES);
1010
1011         /*
1012          * Move to the end of the lru list (end of lru == most recently used).
1013          */
1014         TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp, nu_lru);
1015         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp, nu_lru);
1016
1017         nickp = (u_int32_t *)kmalloc(2 * NFSX_UNSIGNED, M_TEMP, M_WAITOK);
1018         *nickp++ = txdr_unsigned(RPCAKN_NICKNAME);
1019         *nickp = txdr_unsigned(nuidp->nu_nickname);
1020         *auth_str = (char *)nickp;
1021         *auth_len = 2 * NFSX_UNSIGNED;
1022
1023         /*
1024          * Now we must encrypt the verifier and package it up.
1025          */
1026         verfp = (u_int32_t *)verf_str;
1027         *verfp++ = txdr_unsigned(RPCAKN_NICKNAME);
1028         if (time_second > nuidp->nu_timestamp.tv_sec ||
1029             (time_second == nuidp->nu_timestamp.tv_sec &&
1030              time_second > nuidp->nu_timestamp.tv_usec))
1031                 getmicrotime(&nuidp->nu_timestamp);
1032         else
1033                 nuidp->nu_timestamp.tv_usec++;
1034         ktvin.tv_sec = txdr_unsigned(nuidp->nu_timestamp.tv_sec);
1035         ktvin.tv_usec = txdr_unsigned(nuidp->nu_timestamp.tv_usec);
1036
1037         /*
1038          * Now encrypt the timestamp verifier in ecb mode using the session
1039          * key.
1040          */
1041 #ifdef NFSKERB
1042         XXX
1043 #else
1044         ktvout.tv_sec = 0;
1045         ktvout.tv_usec = 0;
1046 #endif
1047
1048         *verfp++ = ktvout.tv_sec;
1049         *verfp++ = ktvout.tv_usec;
1050         *verfp = 0;
1051         return (0);
1052 }
1053
1054 /*
1055  * Save the current nickname in a hash list entry on the mount point.
1056  */
1057 int
1058 nfs_savenickauth(struct nfsmount *nmp, struct ucred *cred, int len,
1059                  NFSKERBKEY_T key, struct mbuf **mdp, char **dposp,
1060                  struct mbuf *mrep)
1061 {
1062         struct nfsuid *nuidp;
1063         u_int32_t *tl;
1064         struct timeval ktvin, ktvout;
1065         u_int32_t nick;
1066         int deltasec, error = 0;
1067         struct nfsm_info info;
1068
1069         info.md = *mdp;
1070         info.dpos = *dposp;
1071         info.mrep = mrep;
1072
1073         if (len == (3 * NFSX_UNSIGNED)) {
1074                 NULLOUT(tl = nfsm_dissect(&info, 3 * NFSX_UNSIGNED));
1075                 ktvin.tv_sec = *tl++;
1076                 ktvin.tv_usec = *tl++;
1077                 nick = fxdr_unsigned(u_int32_t, *tl);
1078
1079                 /*
1080                  * Decrypt the timestamp in ecb mode.
1081                  */
1082 #ifdef NFSKERB
1083                 XXX
1084 #else
1085                 ktvout.tv_sec = 0;
1086                 ktvout.tv_usec = 0;
1087 #endif
1088                 ktvout.tv_sec = fxdr_unsigned(long, ktvout.tv_sec);
1089                 ktvout.tv_usec = fxdr_unsigned(long, ktvout.tv_usec);
1090                 deltasec = time_second - ktvout.tv_sec;
1091                 if (deltasec < 0)
1092                         deltasec = -deltasec;
1093                 /*
1094                  * If ok, add it to the hash list for the mount point.
1095                  */
1096                 if (deltasec <= NFS_KERBCLOCKSKEW) {
1097                         if (nmp->nm_numuids < nuidhash_max) {
1098                                 nmp->nm_numuids++;
1099                                 nuidp = (struct nfsuid *)
1100                                    kmalloc(sizeof (struct nfsuid), M_NFSUID,
1101                                         M_WAITOK);
1102                         } else {
1103                                 nuidp = TAILQ_FIRST(&nmp->nm_uidlruhead);
1104                                 LIST_REMOVE(nuidp, nu_hash);
1105                                 TAILQ_REMOVE(&nmp->nm_uidlruhead, nuidp,
1106                                         nu_lru);
1107                         }
1108                         nuidp->nu_flag = 0;
1109                         nuidp->nu_cr.cr_uid = cred->cr_uid;
1110                         nuidp->nu_expire = time_second + NFS_KERBTTL;
1111                         nuidp->nu_timestamp = ktvout;
1112                         nuidp->nu_nickname = nick;
1113                         bcopy(key, nuidp->nu_key, sizeof (key));
1114                         TAILQ_INSERT_TAIL(&nmp->nm_uidlruhead, nuidp,
1115                                 nu_lru);
1116                         LIST_INSERT_HEAD(NMUIDHASH(nmp, cred->cr_uid),
1117                                 nuidp, nu_hash);
1118                 }
1119         } else {
1120                 ERROROUT(nfsm_adv(&info, nfsm_rndup(len)));
1121         }
1122 nfsmout:
1123         *mdp = info.md;
1124         *dposp = info.dpos;
1125         return (error);
1126 }