1 /*-
2 * Copyright (c) 2009 Rick Macklem, University of Guelph
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 */
27
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30
31 #ifndef APPLEKEXT
32 #include <fs/nfs/nfsport.h>
33
34 struct nfsrv_stablefirst nfsrv_stablefirst;
35 int nfsrv_issuedelegs = 0;
36 int nfsrv_dolocallocks = 0;
37 struct nfsv4lock nfsv4rootfs_lock;
38
39 extern int newnfs_numnfsd;
40 extern struct nfsstats newnfsstats;
41 extern int nfsrv_lease;
42 extern struct timeval nfsboottime;
43 extern u_int32_t newnfs_true, newnfs_false;
44 NFSV4ROOTLOCKMUTEX;
45 NFSSTATESPINLOCK;
46
47 /*
48 * Hash lists for nfs V4.
49 * (Some would put them in the .h file, but I don't like declaring storage
50 * in a .h)
51 */
52 struct nfsclienthashhead nfsclienthash[NFSCLIENTHASHSIZE];
53 struct nfslockhashhead nfslockhash[NFSLOCKHASHSIZE];
54 #endif /* !APPLEKEXT */
55
56 static u_int32_t nfsrv_openpluslock = 0, nfsrv_delegatecnt = 0;
57 static time_t nfsrvboottime;
58 static int nfsrv_writedelegifpos = 1;
59 static int nfsrv_returnoldstateid = 0, nfsrv_clients = 0;
60 static int nfsrv_clienthighwater = NFSRV_CLIENTHIGHWATER;
61 static int nfsrv_nogsscallback = 0;
62
63 /* local functions */
64 static void nfsrv_dumpaclient(struct nfsclient *clp,
65 struct nfsd_dumpclients *dumpp);
66 static void nfsrv_freeopenowner(struct nfsstate *stp, int cansleep,
67 NFSPROC_T *p);
68 static int nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep,
69 NFSPROC_T *p);
70 static void nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
71 NFSPROC_T *p);
72 static void nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp,
73 int cansleep, NFSPROC_T *p);
74 static void nfsrv_freenfslock(struct nfslock *lop);
75 static void nfsrv_freenfslockfile(struct nfslockfile *lfp);
76 static void nfsrv_freedeleg(struct nfsstate *);
77 static int nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp,
78 u_int32_t flags, struct nfsstate **stpp);
79 static void nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
80 struct nfsstate **stpp);
81 static int nfsrv_getlockfh(vnode_t vp, u_short flags,
82 struct nfslockfile *new_lfp, fhandle_t *nfhp, NFSPROC_T *p);
83 static int nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
84 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit);
85 static void nfsrv_insertlock(struct nfslock *new_lop,
86 struct nfslock *insert_lop, struct nfsstate *stp, struct nfslockfile *lfp);
87 static void nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
88 struct nfslock **other_lopp, struct nfslockfile *lfp);
89 static int nfsrv_getipnumber(u_char *cp);
90 static int nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
91 nfsv4stateid_t *stateidp, int specialid);
92 static int nfsrv_checkgrace(u_int32_t flags);
93 static int nfsrv_docallback(struct nfsclient *clp, int procnum,
94 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
95 struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p);
96 static u_int32_t nfsrv_nextclientindex(void);
97 static u_int32_t nfsrv_nextstateindex(struct nfsclient *clp);
98 static void nfsrv_markstable(struct nfsclient *clp);
99 static int nfsrv_checkstable(struct nfsclient *clp);
100 static int nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, struct
101 vnode *vp, NFSPROC_T *p);
102 static int nfsrv_delegconflict(struct nfsstate *stp, int *haslockp,
103 NFSPROC_T *p, vnode_t vp);
104 static int nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
105 struct nfsclient *clp, int *haslockp, NFSPROC_T *p);
106 static int nfsrv_notsamecredname(struct nfsrv_descript *nd,
107 struct nfsclient *clp);
108 static time_t nfsrv_leaseexpiry(void);
109 static void nfsrv_delaydelegtimeout(struct nfsstate *stp);
110 static int nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
111 struct nfsstate *stp, struct nfsrvcache *op);
112 static int nfsrv_nootherstate(struct nfsstate *stp);
113 static int nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
114 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p);
115 static void nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp,
116 uint64_t init_first, uint64_t init_end, NFSPROC_T *p);
117 static int nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags,
118 int oldflags, uint64_t first, uint64_t end, struct nfslockconflict *cfp,
119 NFSPROC_T *p);
120 static void nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp,
121 NFSPROC_T *p);
122 static void nfsrv_locallock_commit(struct nfslockfile *lfp, int flags,
123 uint64_t first, uint64_t end);
124 static void nfsrv_locklf(struct nfslockfile *lfp);
125 static void nfsrv_unlocklf(struct nfslockfile *lfp);
126
127 /*
128 * Scan the client list for a match and either return the current one,
129 * create a new entry or return an error.
130 * If returning a non-error, the clp structure must either be linked into
131 * the client list or free'd.
132 */
133 APPLESTATIC int
134 nfsrv_setclient(struct nfsrv_descript *nd, struct nfsclient **new_clpp,
135 nfsquad_t *clientidp, nfsquad_t *confirmp, NFSPROC_T *p)
136 {
137 struct nfsclient *clp = NULL, *new_clp = *new_clpp;
138 int i, error = 0;
139 struct nfsstate *stp, *tstp;
140 struct sockaddr_in *sad, *rad;
141 int zapit = 0, gotit, hasstate = 0, igotlock;
142 static u_int64_t confirm_index = 0;
143
144 /*
145 * Check for state resource limit exceeded.
146 */
147 if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
148 error = NFSERR_RESOURCE;
149 goto out;
150 }
151
152 if (nfsrv_issuedelegs == 0 ||
153 ((nd->nd_flag & ND_GSS) != 0 && nfsrv_nogsscallback != 0))
154 /*
155 * Don't do callbacks when delegations are disabled or
156 * for AUTH_GSS unless enabled via nfsrv_nogsscallback.
157 * If establishing a callback connection is attempted
158 * when a firewall is blocking the callback path, the
159 * server may wait too long for the connect attempt to
160 * succeed during the Open. Some clients, such as Linux,
161 * may timeout and give up on the Open before the server
162 * replies. Also, since AUTH_GSS callbacks are not
163 * yet interoperability tested, they might cause the
164 * server to crap out, if they get past the Init call to
165 * the client.
166 */
167 new_clp->lc_program = 0;
168
169 /* Lock out other nfsd threads */
170 NFSLOCKV4ROOTMUTEX();
171 nfsv4_relref(&nfsv4rootfs_lock);
172 do {
173 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
174 NFSV4ROOTLOCKMUTEXPTR, NULL);
175 } while (!igotlock);
176 NFSUNLOCKV4ROOTMUTEX();
177
178 /*
179 * Search for a match in the client list.
180 */
181 gotit = i = 0;
182 while (i < NFSCLIENTHASHSIZE && !gotit) {
183 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
184 if (new_clp->lc_idlen == clp->lc_idlen &&
185 !NFSBCMP(new_clp->lc_id, clp->lc_id, clp->lc_idlen)) {
186 gotit = 1;
187 break;
188 }
189 }
190 if (gotit == 0)
191 i++;
192 }
193 if (!gotit ||
194 (clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_ADMINREVOKED))) {
195 /*
196 * Get rid of the old one.
197 */
198 if (i != NFSCLIENTHASHSIZE) {
199 LIST_REMOVE(clp, lc_hash);
200 nfsrv_cleanclient(clp, p);
201 nfsrv_freedeleglist(&clp->lc_deleg);
202 nfsrv_freedeleglist(&clp->lc_olddeleg);
203 zapit = 1;
204 }
205 /*
206 * Add it after assigning a client id to it.
207 */
208 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
209 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
210 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
211 (u_int32_t)nfsrvboottime;
212 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
213 nfsrv_nextclientindex();
214 new_clp->lc_stateindex = 0;
215 new_clp->lc_statemaxindex = 0;
216 new_clp->lc_cbref = 0;
217 new_clp->lc_expiry = nfsrv_leaseexpiry();
218 LIST_INIT(&new_clp->lc_open);
219 LIST_INIT(&new_clp->lc_deleg);
220 LIST_INIT(&new_clp->lc_olddeleg);
221 for (i = 0; i < NFSSTATEHASHSIZE; i++)
222 LIST_INIT(&new_clp->lc_stateid[i]);
223 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
224 lc_hash);
225 newnfsstats.srvclients++;
226 nfsrv_openpluslock++;
227 nfsrv_clients++;
228 NFSLOCKV4ROOTMUTEX();
229 nfsv4_unlock(&nfsv4rootfs_lock, 1);
230 NFSUNLOCKV4ROOTMUTEX();
231 if (zapit)
232 nfsrv_zapclient(clp, p);
233 *new_clpp = NULL;
234 goto out;
235 }
236
237 /*
238 * Now, handle the cases where the id is already issued.
239 */
240 if (nfsrv_notsamecredname(nd, clp)) {
241 /*
242 * Check to see if there is expired state that should go away.
243 */
244 if (clp->lc_expiry < NFSD_MONOSEC &&
245 (!LIST_EMPTY(&clp->lc_open) || !LIST_EMPTY(&clp->lc_deleg))) {
246 nfsrv_cleanclient(clp, p);
247 nfsrv_freedeleglist(&clp->lc_deleg);
248 }
249
250 /*
251 * If there is outstanding state, then reply NFSERR_CLIDINUSE per
252 * RFC3530 Sec. 8.1.2 last para.
253 */
254 if (!LIST_EMPTY(&clp->lc_deleg)) {
255 hasstate = 1;
256 } else if (LIST_EMPTY(&clp->lc_open)) {
257 hasstate = 0;
258 } else {
259 hasstate = 0;
260 /* Look for an Open on the OpenOwner */
261 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
262 if (!LIST_EMPTY(&stp->ls_open)) {
263 hasstate = 1;
264 break;
265 }
266 }
267 }
268 if (hasstate) {
269 /*
270 * If the uid doesn't match, return NFSERR_CLIDINUSE after
271 * filling out the correct ipaddr and portnum.
272 */
273 sad = NFSSOCKADDR(new_clp->lc_req.nr_nam, struct sockaddr_in *);
274 rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
275 sad->sin_addr.s_addr = rad->sin_addr.s_addr;
276 sad->sin_port = rad->sin_port;
277 NFSLOCKV4ROOTMUTEX();
278 nfsv4_unlock(&nfsv4rootfs_lock, 1);
279 NFSUNLOCKV4ROOTMUTEX();
280 error = NFSERR_CLIDINUSE;
281 goto out;
282 }
283 }
284
285 if (NFSBCMP(new_clp->lc_verf, clp->lc_verf, NFSX_VERF)) {
286 /*
287 * If the verifier has changed, the client has rebooted
288 * and a new client id is issued. The old state info
289 * can be thrown away once the SETCLIENTID_CONFIRM occurs.
290 */
291 LIST_REMOVE(clp, lc_hash);
292 new_clp->lc_flags |= LCL_NEEDSCONFIRM;
293 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
294 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
295 nfsrvboottime;
296 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
297 nfsrv_nextclientindex();
298 new_clp->lc_stateindex = 0;
299 new_clp->lc_statemaxindex = 0;
300 new_clp->lc_cbref = 0;
301 new_clp->lc_expiry = nfsrv_leaseexpiry();
302
303 /*
304 * Save the state until confirmed.
305 */
306 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
307 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
308 tstp->ls_clp = new_clp;
309 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
310 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
311 tstp->ls_clp = new_clp;
312 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg,
313 ls_list);
314 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
315 tstp->ls_clp = new_clp;
316 for (i = 0; i < NFSSTATEHASHSIZE; i++) {
317 LIST_NEWHEAD(&new_clp->lc_stateid[i],
318 &clp->lc_stateid[i], ls_hash);
319 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
320 tstp->ls_clp = new_clp;
321 }
322 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
323 lc_hash);
324 newnfsstats.srvclients++;
325 nfsrv_openpluslock++;
326 nfsrv_clients++;
327 NFSLOCKV4ROOTMUTEX();
328 nfsv4_unlock(&nfsv4rootfs_lock, 1);
329 NFSUNLOCKV4ROOTMUTEX();
330
331 /*
332 * Must wait until any outstanding callback on the old clp
333 * completes.
334 */
335 NFSLOCKSTATE();
336 while (clp->lc_cbref) {
337 clp->lc_flags |= LCL_WAKEUPWANTED;
338 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1,
339 "nfsd clp", 10 * hz);
340 }
341 NFSUNLOCKSTATE();
342 nfsrv_zapclient(clp, p);
343 *new_clpp = NULL;
344 goto out;
345 }
346 /*
347 * id and verifier match, so update the net address info
348 * and get rid of any existing callback authentication
349 * handle, so a new one will be acquired.
350 */
351 LIST_REMOVE(clp, lc_hash);
352 new_clp->lc_flags |= (LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
353 new_clp->lc_expiry = nfsrv_leaseexpiry();
354 confirmp->qval = new_clp->lc_confirm.qval = ++confirm_index;
355 clientidp->lval[0] = new_clp->lc_clientid.lval[0] =
356 clp->lc_clientid.lval[0];
357 clientidp->lval[1] = new_clp->lc_clientid.lval[1] =
358 clp->lc_clientid.lval[1];
359 new_clp->lc_delegtime = clp->lc_delegtime;
360 new_clp->lc_stateindex = clp->lc_stateindex;
361 new_clp->lc_statemaxindex = clp->lc_statemaxindex;
362 new_clp->lc_cbref = 0;
363 LIST_NEWHEAD(&new_clp->lc_open, &clp->lc_open, ls_list);
364 LIST_FOREACH(tstp, &new_clp->lc_open, ls_list)
365 tstp->ls_clp = new_clp;
366 LIST_NEWHEAD(&new_clp->lc_deleg, &clp->lc_deleg, ls_list);
367 LIST_FOREACH(tstp, &new_clp->lc_deleg, ls_list)
368 tstp->ls_clp = new_clp;
369 LIST_NEWHEAD(&new_clp->lc_olddeleg, &clp->lc_olddeleg, ls_list);
370 LIST_FOREACH(tstp, &new_clp->lc_olddeleg, ls_list)
371 tstp->ls_clp = new_clp;
372 for (i = 0; i < NFSSTATEHASHSIZE; i++) {
373 LIST_NEWHEAD(&new_clp->lc_stateid[i], &clp->lc_stateid[i],
374 ls_hash);
375 LIST_FOREACH(tstp, &new_clp->lc_stateid[i], ls_hash)
376 tstp->ls_clp = new_clp;
377 }
378 LIST_INSERT_HEAD(NFSCLIENTHASH(new_clp->lc_clientid), new_clp,
379 lc_hash);
380 newnfsstats.srvclients++;
381 nfsrv_openpluslock++;
382 nfsrv_clients++;
383 NFSLOCKV4ROOTMUTEX();
384 nfsv4_unlock(&nfsv4rootfs_lock, 1);
385 NFSUNLOCKV4ROOTMUTEX();
386
387 /*
388 * Must wait until any outstanding callback on the old clp
389 * completes.
390 */
391 NFSLOCKSTATE();
392 while (clp->lc_cbref) {
393 clp->lc_flags |= LCL_WAKEUPWANTED;
394 (void)mtx_sleep(clp, NFSSTATEMUTEXPTR, PZERO - 1, "nfsd clp",
395 10 * hz);
396 }
397 NFSUNLOCKSTATE();
398 nfsrv_zapclient(clp, p);
399 *new_clpp = NULL;
400
401 out:
402 NFSEXITCODE2(error, nd);
403 return (error);
404 }
405
406 /*
407 * Check to see if the client id exists and optionally confirm it.
408 */
409 APPLESTATIC int
410 nfsrv_getclient(nfsquad_t clientid, int opflags, struct nfsclient **clpp,
411 nfsquad_t confirm, struct nfsrv_descript *nd, NFSPROC_T *p)
412 {
413 struct nfsclient *clp;
414 struct nfsstate *stp;
415 int i;
416 struct nfsclienthashhead *hp;
417 int error = 0, igotlock, doneok;
418
419 if (clpp)
420 *clpp = NULL;
421 if (nfsrvboottime != clientid.lval[0]) {
422 error = NFSERR_STALECLIENTID;
423 goto out;
424 }
425
426 /*
427 * If called with opflags == CLOPS_RENEW, the State Lock is
428 * already held. Otherwise, we need to get either that or,
429 * for the case of Confirm, lock out the nfsd threads.
430 */
431 if (opflags & CLOPS_CONFIRM) {
432 NFSLOCKV4ROOTMUTEX();
433 nfsv4_relref(&nfsv4rootfs_lock);
434 do {
435 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
436 NFSV4ROOTLOCKMUTEXPTR, NULL);
437 } while (!igotlock);
438 NFSUNLOCKV4ROOTMUTEX();
439 } else if (opflags != CLOPS_RENEW) {
440 NFSLOCKSTATE();
441 }
442
443 hp = NFSCLIENTHASH(clientid);
444 LIST_FOREACH(clp, hp, lc_hash) {
445 if (clp->lc_clientid.lval[1] == clientid.lval[1])
446 break;
447 }
448 if (clp == LIST_END(hp)) {
449 if (opflags & CLOPS_CONFIRM)
450 error = NFSERR_STALECLIENTID;
451 else
452 error = NFSERR_EXPIRED;
453 } else if (clp->lc_flags & LCL_ADMINREVOKED) {
454 /*
455 * If marked admin revoked, just return the error.
456 */
457 error = NFSERR_ADMINREVOKED;
458 }
459 if (error) {
460 if (opflags & CLOPS_CONFIRM) {
461 NFSLOCKV4ROOTMUTEX();
462 nfsv4_unlock(&nfsv4rootfs_lock, 1);
463 NFSUNLOCKV4ROOTMUTEX();
464 } else if (opflags != CLOPS_RENEW) {
465 NFSUNLOCKSTATE();
466 }
467 goto out;
468 }
469
470 /*
471 * Perform any operations specified by the opflags.
472 */
473 if (opflags & CLOPS_CONFIRM) {
474 if (clp->lc_confirm.qval != confirm.qval)
475 error = NFSERR_STALECLIENTID;
476 else if (nfsrv_notsamecredname(nd, clp))
477 error = NFSERR_CLIDINUSE;
478
479 if (!error) {
480 if ((clp->lc_flags & (LCL_NEEDSCONFIRM | LCL_DONTCLEAN)) ==
481 LCL_NEEDSCONFIRM) {
482 /*
483 * Hang onto the delegations (as old delegations)
484 * for an Open with CLAIM_DELEGATE_PREV unless in
485 * grace, but get rid of the rest of the state.
486 */
487 nfsrv_cleanclient(clp, p);
488 nfsrv_freedeleglist(&clp->lc_olddeleg);
489 if (nfsrv_checkgrace(0)) {
490 /* In grace, so just delete delegations */
491 nfsrv_freedeleglist(&clp->lc_deleg);
492 } else {
493 LIST_FOREACH(stp, &clp->lc_deleg, ls_list)
494 stp->ls_flags |= NFSLCK_OLDDELEG;
495 clp->lc_delegtime = NFSD_MONOSEC +
496 nfsrv_lease + NFSRV_LEASEDELTA;
497 LIST_NEWHEAD(&clp->lc_olddeleg, &clp->lc_deleg,
498 ls_list);
499 }
500 }
501 clp->lc_flags &= ~(LCL_NEEDSCONFIRM | LCL_DONTCLEAN);
502 if (clp->lc_program)
503 clp->lc_flags |= LCL_NEEDSCBNULL;
504 }
505 } else if (clp->lc_flags & LCL_NEEDSCONFIRM) {
506 error = NFSERR_EXPIRED;
507 }
508
509 /*
510 * If called by the Renew Op, we must check the principal.
511 */
512 if (!error && (opflags & CLOPS_RENEWOP)) {
513 if (nfsrv_notsamecredname(nd, clp)) {
514 doneok = 0;
515 for (i = 0; i < NFSSTATEHASHSIZE && doneok == 0; i++) {
516 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
517 if ((stp->ls_flags & NFSLCK_OPEN) &&
518 stp->ls_uid == nd->nd_cred->cr_uid) {
519 doneok = 1;
520 break;
521 }
522 }
523 }
524 if (!doneok)
525 error = NFSERR_ACCES;
526 }
527 if (!error && (clp->lc_flags & LCL_CBDOWN))
528 error = NFSERR_CBPATHDOWN;
529 }
530 if ((!error || error == NFSERR_CBPATHDOWN) &&
531 (opflags & CLOPS_RENEW)) {
532 clp->lc_expiry = nfsrv_leaseexpiry();
533 }
534 if (opflags & CLOPS_CONFIRM) {
535 NFSLOCKV4ROOTMUTEX();
536 nfsv4_unlock(&nfsv4rootfs_lock, 1);
537 NFSUNLOCKV4ROOTMUTEX();
538 } else if (opflags != CLOPS_RENEW) {
539 NFSUNLOCKSTATE();
540 }
541 if (clpp)
542 *clpp = clp;
543
544 out:
545 NFSEXITCODE2(error, nd);
546 return (error);
547 }
548
549 /*
550 * Called from the new nfssvc syscall to admin revoke a clientid.
551 * Returns 0 for success, error otherwise.
552 */
553 APPLESTATIC int
554 nfsrv_adminrevoke(struct nfsd_clid *revokep, NFSPROC_T *p)
555 {
556 struct nfsclient *clp = NULL;
557 int i, error = 0;
558 int gotit, igotlock;
559
560 /*
561 * First, lock out the nfsd so that state won't change while the
562 * revocation record is being written to the stable storage restart
563 * file.
564 */
565 NFSLOCKV4ROOTMUTEX();
566 do {
567 igotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
568 NFSV4ROOTLOCKMUTEXPTR, NULL);
569 } while (!igotlock);
570 NFSUNLOCKV4ROOTMUTEX();
571
572 /*
573 * Search for a match in the client list.
574 */
575 gotit = i = 0;
576 while (i < NFSCLIENTHASHSIZE && !gotit) {
577 LIST_FOREACH(clp, &nfsclienthash[i], lc_hash) {
578 if (revokep->nclid_idlen == clp->lc_idlen &&
579 !NFSBCMP(revokep->nclid_id, clp->lc_id, clp->lc_idlen)) {
580 gotit = 1;
581 break;
582 }
583 }
584 i++;
585 }
586 if (!gotit) {
587 NFSLOCKV4ROOTMUTEX();
588 nfsv4_unlock(&nfsv4rootfs_lock, 0);
589 NFSUNLOCKV4ROOTMUTEX();
590 error = EPERM;
591 goto out;
592 }
593
594 /*
595 * Now, write out the revocation record
596 */
597 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
598 nfsrv_backupstable();
599
600 /*
601 * and clear out the state, marking the clientid revoked.
602 */
603 clp->lc_flags &= ~LCL_CALLBACKSON;
604 clp->lc_flags |= LCL_ADMINREVOKED;
605 nfsrv_cleanclient(clp, p);
606 nfsrv_freedeleglist(&clp->lc_deleg);
607 nfsrv_freedeleglist(&clp->lc_olddeleg);
608 NFSLOCKV4ROOTMUTEX();
609 nfsv4_unlock(&nfsv4rootfs_lock, 0);
610 NFSUNLOCKV4ROOTMUTEX();
611
612 out:
613 NFSEXITCODE(error);
614 return (error);
615 }
616
617 /*
618 * Dump out stats for all clients. Called from nfssvc(2), that is used
619 * newnfsstats.
620 */
621 APPLESTATIC void
622 nfsrv_dumpclients(struct nfsd_dumpclients *dumpp, int maxcnt)
623 {
624 struct nfsclient *clp;
625 int i = 0, cnt = 0;
626
627 /*
628 * First, get a reference on the nfsv4rootfs_lock so that an
629 * exclusive lock cannot be acquired while dumping the clients.
630 */
631 NFSLOCKV4ROOTMUTEX();
632 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
633 NFSUNLOCKV4ROOTMUTEX();
634 NFSLOCKSTATE();
635 /*
636 * Rattle through the client lists until done.
637 */
638 while (i < NFSCLIENTHASHSIZE && cnt < maxcnt) {
639 clp = LIST_FIRST(&nfsclienthash[i]);
640 while (clp != LIST_END(&nfsclienthash[i]) && cnt < maxcnt) {
641 nfsrv_dumpaclient(clp, &dumpp[cnt]);
642 cnt++;
643 clp = LIST_NEXT(clp, lc_hash);
644 }
645 i++;
646 }
647 if (cnt < maxcnt)
648 dumpp[cnt].ndcl_clid.nclid_idlen = 0;
649 NFSUNLOCKSTATE();
650 NFSLOCKV4ROOTMUTEX();
651 nfsv4_relref(&nfsv4rootfs_lock);
652 NFSUNLOCKV4ROOTMUTEX();
653 }
654
655 /*
656 * Dump stats for a client. Must be called with the NFSSTATELOCK and spl'd.
657 */
658 static void
659 nfsrv_dumpaclient(struct nfsclient *clp, struct nfsd_dumpclients *dumpp)
660 {
661 struct nfsstate *stp, *openstp, *lckownstp;
662 struct nfslock *lop;
663 struct sockaddr *sad;
664 struct sockaddr_in *rad;
665 struct sockaddr_in6 *rad6;
666
667 dumpp->ndcl_nopenowners = dumpp->ndcl_nlockowners = 0;
668 dumpp->ndcl_nopens = dumpp->ndcl_nlocks = 0;
669 dumpp->ndcl_ndelegs = dumpp->ndcl_nolddelegs = 0;
670 dumpp->ndcl_flags = clp->lc_flags;
671 dumpp->ndcl_clid.nclid_idlen = clp->lc_idlen;
672 NFSBCOPY(clp->lc_id, dumpp->ndcl_clid.nclid_id, clp->lc_idlen);
673 sad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr *);
674 dumpp->ndcl_addrfam = sad->sa_family;
675 if (sad->sa_family == AF_INET) {
676 rad = (struct sockaddr_in *)sad;
677 dumpp->ndcl_cbaddr.sin_addr = rad->sin_addr;
678 } else {
679 rad6 = (struct sockaddr_in6 *)sad;
680 dumpp->ndcl_cbaddr.sin6_addr = rad6->sin6_addr;
681 }
682
683 /*
684 * Now, scan the state lists and total up the opens and locks.
685 */
686 LIST_FOREACH(stp, &clp->lc_open, ls_list) {
687 dumpp->ndcl_nopenowners++;
688 LIST_FOREACH(openstp, &stp->ls_open, ls_list) {
689 dumpp->ndcl_nopens++;
690 LIST_FOREACH(lckownstp, &openstp->ls_open, ls_list) {
691 dumpp->ndcl_nlockowners++;
692 LIST_FOREACH(lop, &lckownstp->ls_lock, lo_lckowner) {
693 dumpp->ndcl_nlocks++;
694 }
695 }
696 }
697 }
698
699 /*
700 * and the delegation lists.
701 */
702 LIST_FOREACH(stp, &clp->lc_deleg, ls_list) {
703 dumpp->ndcl_ndelegs++;
704 }
705 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
706 dumpp->ndcl_nolddelegs++;
707 }
708 }
709
710 /*
711 * Dump out lock stats for a file.
712 */
713 APPLESTATIC void
714 nfsrv_dumplocks(vnode_t vp, struct nfsd_dumplocks *ldumpp, int maxcnt,
715 NFSPROC_T *p)
716 {
717 struct nfsstate *stp;
718 struct nfslock *lop;
719 int cnt = 0;
720 struct nfslockfile *lfp;
721 struct sockaddr *sad;
722 struct sockaddr_in *rad;
723 struct sockaddr_in6 *rad6;
724 int ret;
725 fhandle_t nfh;
726
727 ret = nfsrv_getlockfh(vp, 0, NULL, &nfh, p);
728 /*
729 * First, get a reference on the nfsv4rootfs_lock so that an
730 * exclusive lock on it cannot be acquired while dumping the locks.
731 */
732 NFSLOCKV4ROOTMUTEX();
733 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
734 NFSUNLOCKV4ROOTMUTEX();
735 NFSLOCKSTATE();
736 if (!ret)
737 ret = nfsrv_getlockfile(0, NULL, &lfp, &nfh, 0);
738 if (ret) {
739 ldumpp[0].ndlck_clid.nclid_idlen = 0;
740 NFSUNLOCKSTATE();
741 NFSLOCKV4ROOTMUTEX();
742 nfsv4_relref(&nfsv4rootfs_lock);
743 NFSUNLOCKV4ROOTMUTEX();
744 return;
745 }
746
747 /*
748 * For each open share on file, dump it out.
749 */
750 stp = LIST_FIRST(&lfp->lf_open);
751 while (stp != LIST_END(&lfp->lf_open) && cnt < maxcnt) {
752 ldumpp[cnt].ndlck_flags = stp->ls_flags;
753 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
754 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
755 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
756 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
757 ldumpp[cnt].ndlck_owner.nclid_idlen =
758 stp->ls_openowner->ls_ownerlen;
759 NFSBCOPY(stp->ls_openowner->ls_owner,
760 ldumpp[cnt].ndlck_owner.nclid_id,
761 stp->ls_openowner->ls_ownerlen);
762 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
763 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
764 stp->ls_clp->lc_idlen);
765 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
766 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
767 if (sad->sa_family == AF_INET) {
768 rad = (struct sockaddr_in *)sad;
769 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
770 } else {
771 rad6 = (struct sockaddr_in6 *)sad;
772 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
773 }
774 stp = LIST_NEXT(stp, ls_file);
775 cnt++;
776 }
777
778 /*
779 * and all locks.
780 */
781 lop = LIST_FIRST(&lfp->lf_lock);
782 while (lop != LIST_END(&lfp->lf_lock) && cnt < maxcnt) {
783 stp = lop->lo_stp;
784 ldumpp[cnt].ndlck_flags = lop->lo_flags;
785 ldumpp[cnt].ndlck_first = lop->lo_first;
786 ldumpp[cnt].ndlck_end = lop->lo_end;
787 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
788 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
789 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
790 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
791 ldumpp[cnt].ndlck_owner.nclid_idlen = stp->ls_ownerlen;
792 NFSBCOPY(stp->ls_owner, ldumpp[cnt].ndlck_owner.nclid_id,
793 stp->ls_ownerlen);
794 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
795 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
796 stp->ls_clp->lc_idlen);
797 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
798 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
799 if (sad->sa_family == AF_INET) {
800 rad = (struct sockaddr_in *)sad;
801 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
802 } else {
803 rad6 = (struct sockaddr_in6 *)sad;
804 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
805 }
806 lop = LIST_NEXT(lop, lo_lckfile);
807 cnt++;
808 }
809
810 /*
811 * and the delegations.
812 */
813 stp = LIST_FIRST(&lfp->lf_deleg);
814 while (stp != LIST_END(&lfp->lf_deleg) && cnt < maxcnt) {
815 ldumpp[cnt].ndlck_flags = stp->ls_flags;
816 ldumpp[cnt].ndlck_stateid.seqid = stp->ls_stateid.seqid;
817 ldumpp[cnt].ndlck_stateid.other[0] = stp->ls_stateid.other[0];
818 ldumpp[cnt].ndlck_stateid.other[1] = stp->ls_stateid.other[1];
819 ldumpp[cnt].ndlck_stateid.other[2] = stp->ls_stateid.other[2];
820 ldumpp[cnt].ndlck_owner.nclid_idlen = 0;
821 ldumpp[cnt].ndlck_clid.nclid_idlen = stp->ls_clp->lc_idlen;
822 NFSBCOPY(stp->ls_clp->lc_id, ldumpp[cnt].ndlck_clid.nclid_id,
823 stp->ls_clp->lc_idlen);
824 sad=NFSSOCKADDR(stp->ls_clp->lc_req.nr_nam, struct sockaddr *);
825 ldumpp[cnt].ndlck_addrfam = sad->sa_family;
826 if (sad->sa_family == AF_INET) {
827 rad = (struct sockaddr_in *)sad;
828 ldumpp[cnt].ndlck_cbaddr.sin_addr = rad->sin_addr;
829 } else {
830 rad6 = (struct sockaddr_in6 *)sad;
831 ldumpp[cnt].ndlck_cbaddr.sin6_addr = rad6->sin6_addr;
832 }
833 stp = LIST_NEXT(stp, ls_file);
834 cnt++;
835 }
836
837 /*
838 * If list isn't full, mark end of list by setting the client name
839 * to zero length.
840 */
841 if (cnt < maxcnt)
842 ldumpp[cnt].ndlck_clid.nclid_idlen = 0;
843 NFSUNLOCKSTATE();
844 NFSLOCKV4ROOTMUTEX();
845 nfsv4_relref(&nfsv4rootfs_lock);
846 NFSUNLOCKV4ROOTMUTEX();
847 }
848
849 /*
850 * Server timer routine. It can scan any linked list, so long
851 * as it holds the spin/mutex lock and there is no exclusive lock on
852 * nfsv4rootfs_lock.
853 * (For OpenBSD, a kthread is ok. For FreeBSD, I think it is ok
854 * to do this from a callout, since the spin locks work. For
855 * Darwin, I'm not sure what will work correctly yet.)
856 * Should be called once per second.
857 */
858 APPLESTATIC void
859 nfsrv_servertimer(void)
860 {
861 struct nfsclient *clp, *nclp;
862 struct nfsstate *stp, *nstp;
863 int got_ref, i;
864
865 /*
866 * Make sure nfsboottime is set. This is used by V3 as well
867 * as V4. Note that nfsboottime is not nfsrvboottime, which is
868 * only used by the V4 server for leases.
869 */
870 if (nfsboottime.tv_sec == 0)
871 NFSSETBOOTTIME(nfsboottime);
872
873 /*
874 * If server hasn't started yet, just return.
875 */
876 NFSLOCKSTATE();
877 if (nfsrv_stablefirst.nsf_eograce == 0) {
878 NFSUNLOCKSTATE();
879 return;
880 }
881 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE)) {
882 if (!(nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) &&
883 NFSD_MONOSEC > nfsrv_stablefirst.nsf_eograce)
884 nfsrv_stablefirst.nsf_flags |=
885 (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
886 NFSUNLOCKSTATE();
887 return;
888 }
889
890 /*
891 * Try and get a reference count on the nfsv4rootfs_lock so that
892 * no nfsd thread can acquire an exclusive lock on it before this
893 * call is done. If it is already exclusively locked, just return.
894 */
895 NFSLOCKV4ROOTMUTEX();
896 got_ref = nfsv4_getref_nonblock(&nfsv4rootfs_lock);
897 NFSUNLOCKV4ROOTMUTEX();
898 if (got_ref == 0) {
899 NFSUNLOCKSTATE();
900 return;
901 }
902
903 /*
904 * For each client...
905 */
906 for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
907 clp = LIST_FIRST(&nfsclienthash[i]);
908 while (clp != LIST_END(&nfsclienthash[i])) {
909 nclp = LIST_NEXT(clp, lc_hash);
910 if (!(clp->lc_flags & LCL_EXPIREIT)) {
911 if (((clp->lc_expiry + NFSRV_STALELEASE) < NFSD_MONOSEC
912 && ((LIST_EMPTY(&clp->lc_deleg)
913 && LIST_EMPTY(&clp->lc_open)) ||
914 nfsrv_clients > nfsrv_clienthighwater)) ||
915 (clp->lc_expiry + NFSRV_MOULDYLEASE) < NFSD_MONOSEC ||
916 (clp->lc_expiry < NFSD_MONOSEC &&
917 (nfsrv_openpluslock * 10 / 9) > NFSRV_V4STATELIMIT)) {
918 /*
919 * Lease has expired several nfsrv_lease times ago:
920 * PLUS
921 * - no state is associated with it
922 * OR
923 * - above high water mark for number of clients
924 * (nfsrv_clienthighwater should be large enough
925 * that this only occurs when clients fail to
926 * use the same nfs_client_id4.id. Maybe somewhat
927 * higher that the maximum number of clients that
928 * will mount this server?)
929 * OR
930 * Lease has expired a very long time ago
931 * OR
932 * Lease has expired PLUS the number of opens + locks
933 * has exceeded 90% of capacity
934 *
935 * --> Mark for expiry. The actual expiry will be done
936 * by an nfsd sometime soon.
937 */
938 clp->lc_flags |= LCL_EXPIREIT;
939 nfsrv_stablefirst.nsf_flags |=
940 (NFSNSF_NEEDLOCK | NFSNSF_EXPIREDCLIENT);
941 } else {
942 /*
943 * If there are no opens, increment no open tick cnt
944 * If time exceeds NFSNOOPEN, mark it to be thrown away
945 * otherwise, if there is an open, reset no open time
946 * Hopefully, this will avoid excessive re-creation
947 * of open owners and subsequent open confirms.
948 */
949 stp = LIST_FIRST(&clp->lc_open);
950 while (stp != LIST_END(&clp->lc_open)) {
951 nstp = LIST_NEXT(stp, ls_list);
952 if (LIST_EMPTY(&stp->ls_open)) {
953 stp->ls_noopens++;
954 if (stp->ls_noopens > NFSNOOPEN ||
955 (nfsrv_openpluslock * 2) >
956 NFSRV_V4STATELIMIT)
957 nfsrv_stablefirst.nsf_flags |=
958 NFSNSF_NOOPENS;
959 } else {
960 stp->ls_noopens = 0;
961 }
962 stp = nstp;
963 }
964 }
965 }
966 clp = nclp;
967 }
968 }
969 NFSUNLOCKSTATE();
970 NFSLOCKV4ROOTMUTEX();
971 nfsv4_relref(&nfsv4rootfs_lock);
972 NFSUNLOCKV4ROOTMUTEX();
973 }
974
975 /*
976 * The following set of functions free up the various data structures.
977 */
978 /*
979 * Clear out all open/lock state related to this nfsclient.
980 * Caller must hold an exclusive lock on nfsv4rootfs_lock, so that
981 * there are no other active nfsd threads.
982 */
983 APPLESTATIC void
984 nfsrv_cleanclient(struct nfsclient *clp, NFSPROC_T *p)
985 {
986 struct nfsstate *stp, *nstp;
987
988 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp)
989 nfsrv_freeopenowner(stp, 1, p);
990 }
991
992 /*
993 * Free a client that has been cleaned. It should also already have been
994 * removed from the lists.
995 * (Just to be safe w.r.t. newnfs_disconnect(), call this function when
996 * softclock interrupts are enabled.)
997 */
998 APPLESTATIC void
999 nfsrv_zapclient(struct nfsclient *clp, NFSPROC_T *p)
1000 {
1001
1002 #ifdef notyet
1003 if ((clp->lc_flags & (LCL_GSS | LCL_CALLBACKSON)) ==
1004 (LCL_GSS | LCL_CALLBACKSON) &&
1005 (clp->lc_hand.nfsh_flag & NFSG_COMPLETE) &&
1006 clp->lc_handlelen > 0) {
1007 clp->lc_hand.nfsh_flag &= ~NFSG_COMPLETE;
1008 clp->lc_hand.nfsh_flag |= NFSG_DESTROYED;
1009 (void) nfsrv_docallback(clp, NFSV4PROC_CBNULL,
1010 NULL, 0, NULL, NULL, NULL, p);
1011 }
1012 #endif
1013 newnfs_disconnect(&clp->lc_req);
1014 NFSSOCKADDRFREE(clp->lc_req.nr_nam);
1015 NFSFREEMUTEX(&clp->lc_req.nr_mtx);
1016 free((caddr_t)clp, M_NFSDCLIENT);
1017 NFSLOCKSTATE();
1018 newnfsstats.srvclients--;
1019 nfsrv_openpluslock--;
1020 nfsrv_clients--;
1021 NFSUNLOCKSTATE();
1022 }
1023
1024 /*
1025 * Free a list of delegation state structures.
1026 * (This function will also free all nfslockfile structures that no
1027 * longer have associated state.)
1028 */
1029 APPLESTATIC void
1030 nfsrv_freedeleglist(struct nfsstatehead *sthp)
1031 {
1032 struct nfsstate *stp, *nstp;
1033
1034 LIST_FOREACH_SAFE(stp, sthp, ls_list, nstp) {
1035 nfsrv_freedeleg(stp);
1036 }
1037 LIST_INIT(sthp);
1038 }
1039
1040 /*
1041 * Free up a delegation.
1042 */
1043 static void
1044 nfsrv_freedeleg(struct nfsstate *stp)
1045 {
1046 struct nfslockfile *lfp;
1047
1048 LIST_REMOVE(stp, ls_hash);
1049 LIST_REMOVE(stp, ls_list);
1050 LIST_REMOVE(stp, ls_file);
1051 lfp = stp->ls_lfp;
1052 if (LIST_EMPTY(&lfp->lf_open) &&
1053 LIST_EMPTY(&lfp->lf_lock) && LIST_EMPTY(&lfp->lf_deleg) &&
1054 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1055 lfp->lf_usecount == 0 &&
1056 nfsv4_testlock(&lfp->lf_locallock_lck) == 0)
1057 nfsrv_freenfslockfile(lfp);
1058 FREE((caddr_t)stp, M_NFSDSTATE);
1059 newnfsstats.srvdelegates--;
1060 nfsrv_openpluslock--;
1061 nfsrv_delegatecnt--;
1062 }
1063
1064 /*
1065 * This function frees an open owner and all associated opens.
1066 */
1067 static void
1068 nfsrv_freeopenowner(struct nfsstate *stp, int cansleep, NFSPROC_T *p)
1069 {
1070 struct nfsstate *nstp, *tstp;
1071
1072 LIST_REMOVE(stp, ls_list);
1073 /*
1074 * Now, free all associated opens.
1075 */
1076 nstp = LIST_FIRST(&stp->ls_open);
1077 while (nstp != LIST_END(&stp->ls_open)) {
1078 tstp = nstp;
1079 nstp = LIST_NEXT(nstp, ls_list);
1080 (void) nfsrv_freeopen(tstp, NULL, cansleep, p);
1081 }
1082 if (stp->ls_op)
1083 nfsrvd_derefcache(stp->ls_op);
1084 FREE((caddr_t)stp, M_NFSDSTATE);
1085 newnfsstats.srvopenowners--;
1086 nfsrv_openpluslock--;
1087 }
1088
1089 /*
1090 * This function frees an open (nfsstate open structure) with all associated
1091 * lock_owners and locks. It also frees the nfslockfile structure iff there
1092 * are no other opens on the file.
1093 * Returns 1 if it free'd the nfslockfile, 0 otherwise.
1094 */
1095 static int
1096 nfsrv_freeopen(struct nfsstate *stp, vnode_t vp, int cansleep, NFSPROC_T *p)
1097 {
1098 struct nfsstate *nstp, *tstp;
1099 struct nfslockfile *lfp;
1100 int ret;
1101
1102 LIST_REMOVE(stp, ls_hash);
1103 LIST_REMOVE(stp, ls_list);
1104 LIST_REMOVE(stp, ls_file);
1105
1106 lfp = stp->ls_lfp;
1107 /*
1108 * Now, free all lockowners associated with this open.
1109 */
1110 LIST_FOREACH_SAFE(tstp, &stp->ls_open, ls_list, nstp)
1111 nfsrv_freelockowner(tstp, vp, cansleep, p);
1112
1113 /*
1114 * The nfslockfile is freed here if there are no locks
1115 * associated with the open.
1116 * If there are locks associated with the open, the
1117 * nfslockfile structure can be freed via nfsrv_freelockowner().
1118 * Acquire the state mutex to avoid races with calls to
1119 * nfsrv_getlockfile().
1120 */
1121 if (cansleep != 0)
1122 NFSLOCKSTATE();
1123 if (lfp != NULL && LIST_EMPTY(&lfp->lf_open) &&
1124 LIST_EMPTY(&lfp->lf_deleg) && LIST_EMPTY(&lfp->lf_lock) &&
1125 LIST_EMPTY(&lfp->lf_locallock) && LIST_EMPTY(&lfp->lf_rollback) &&
1126 lfp->lf_usecount == 0 &&
1127 (cansleep != 0 || nfsv4_testlock(&lfp->lf_locallock_lck) == 0)) {
1128 nfsrv_freenfslockfile(lfp);
1129 ret = 1;
1130 } else
1131 ret = 0;
1132 if (cansleep != 0)
1133 NFSUNLOCKSTATE();
1134 FREE((caddr_t)stp, M_NFSDSTATE);
1135 newnfsstats.srvopens--;
1136 nfsrv_openpluslock--;
1137 return (ret);
1138 }
1139
1140 /*
1141 * Frees a lockowner and all associated locks.
1142 */
1143 static void
1144 nfsrv_freelockowner(struct nfsstate *stp, vnode_t vp, int cansleep,
1145 NFSPROC_T *p)
1146 {
1147
1148 LIST_REMOVE(stp, ls_hash);
1149 LIST_REMOVE(stp, ls_list);
1150 nfsrv_freeallnfslocks(stp, vp, cansleep, p);
1151 if (stp->ls_op)
1152 nfsrvd_derefcache(stp->ls_op);
1153 FREE((caddr_t)stp, M_NFSDSTATE);
1154 newnfsstats.srvlockowners--;
1155 nfsrv_openpluslock--;
1156 }
1157
1158 /*
1159 * Free all the nfs locks on a lockowner.
1160 */
1161 static void
1162 nfsrv_freeallnfslocks(struct nfsstate *stp, vnode_t vp, int cansleep,
1163 NFSPROC_T *p)
1164 {
1165 struct nfslock *lop, *nlop;
1166 struct nfsrollback *rlp, *nrlp;
1167 struct nfslockfile *lfp = NULL;
1168 int gottvp = 0;
1169 vnode_t tvp = NULL;
1170 uint64_t first, end;
1171
1172 if (vp != NULL)
1173 ASSERT_VOP_UNLOCKED(vp, "nfsrv_freeallnfslocks: vnode locked");
1174 lop = LIST_FIRST(&stp->ls_lock);
1175 while (lop != LIST_END(&stp->ls_lock)) {
1176 nlop = LIST_NEXT(lop, lo_lckowner);
1177 /*
1178 * Since all locks should be for the same file, lfp should
1179 * not change.
1180 */
1181 if (lfp == NULL)
1182 lfp = lop->lo_lfp;
1183 else if (lfp != lop->lo_lfp)
1184 panic("allnfslocks");
1185 /*
1186 * If vp is NULL and cansleep != 0, a vnode must be acquired
1187 * from the file handle. This only occurs when called from
1188 * nfsrv_cleanclient().
1189 */
1190 if (gottvp == 0) {
1191 if (nfsrv_dolocallocks == 0)
1192 tvp = NULL;
1193 else if (vp == NULL && cansleep != 0) {
1194 tvp = nfsvno_getvp(&lfp->lf_fh);
1195 NFSVOPUNLOCK(tvp, 0);
1196 } else
1197 tvp = vp;
1198 gottvp = 1;
1199 }
1200
1201 if (tvp != NULL) {
1202 if (cansleep == 0)
1203 panic("allnfs2");
1204 first = lop->lo_first;
1205 end = lop->lo_end;
1206 nfsrv_freenfslock(lop);
1207 nfsrv_localunlock(tvp, lfp, first, end, p);
1208 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list,
1209 nrlp)
1210 free(rlp, M_NFSDROLLBACK);
1211 LIST_INIT(&lfp->lf_rollback);
1212 } else
1213 nfsrv_freenfslock(lop);
1214 lop = nlop;
1215 }
1216 if (vp == NULL && tvp != NULL)
1217 vrele(tvp);
1218 }
1219
1220 /*
1221 * Free an nfslock structure.
1222 */
1223 static void
1224 nfsrv_freenfslock(struct nfslock *lop)
1225 {
1226
1227 if (lop->lo_lckfile.le_prev != NULL) {
1228 LIST_REMOVE(lop, lo_lckfile);
1229 newnfsstats.srvlocks--;
1230 nfsrv_openpluslock--;
1231 }
1232 LIST_REMOVE(lop, lo_lckowner);
1233 FREE((caddr_t)lop, M_NFSDLOCK);
1234 }
1235
1236 /*
1237 * This function frees an nfslockfile structure.
1238 */
1239 static void
1240 nfsrv_freenfslockfile(struct nfslockfile *lfp)
1241 {
1242
1243 LIST_REMOVE(lfp, lf_hash);
1244 FREE((caddr_t)lfp, M_NFSDLOCKFILE);
1245 }
1246
1247 /*
1248 * This function looks up an nfsstate structure via stateid.
1249 */
1250 static int
1251 nfsrv_getstate(struct nfsclient *clp, nfsv4stateid_t *stateidp, __unused u_int32_t flags,
1252 struct nfsstate **stpp)
1253 {
1254 struct nfsstate *stp;
1255 struct nfsstatehead *hp;
1256 int error = 0;
1257
1258 *stpp = NULL;
1259 hp = NFSSTATEHASH(clp, *stateidp);
1260 LIST_FOREACH(stp, hp, ls_hash) {
1261 if (!NFSBCMP(stp->ls_stateid.other, stateidp->other,
1262 NFSX_STATEIDOTHER))
1263 break;
1264 }
1265
1266 /*
1267 * If no state id in list, return NFSERR_BADSTATEID.
1268 */
1269 if (stp == LIST_END(hp)) {
1270 error = NFSERR_BADSTATEID;
1271 goto out;
1272 }
1273 *stpp = stp;
1274
1275 out:
1276 NFSEXITCODE(error);
1277 return (error);
1278 }
1279
1280 /*
1281 * This function gets an nfsstate structure via owner string.
1282 */
1283 static void
1284 nfsrv_getowner(struct nfsstatehead *hp, struct nfsstate *new_stp,
1285 struct nfsstate **stpp)
1286 {
1287 struct nfsstate *stp;
1288
1289 *stpp = NULL;
1290 LIST_FOREACH(stp, hp, ls_list) {
1291 if (new_stp->ls_ownerlen == stp->ls_ownerlen &&
1292 !NFSBCMP(new_stp->ls_owner,stp->ls_owner,stp->ls_ownerlen)) {
1293 *stpp = stp;
1294 return;
1295 }
1296 }
1297 }
1298
1299 /*
1300 * Lock control function called to update lock status.
1301 * Returns 0 upon success, -1 if there is no lock and the flags indicate
1302 * that one isn't to be created and an NFSERR_xxx for other errors.
1303 * The structures new_stp and new_lop are passed in as pointers that should
1304 * be set to NULL if the structure is used and shouldn't be free'd.
1305 * For the NFSLCK_TEST and NFSLCK_CHECK cases, the structures are
1306 * never used and can safely be allocated on the stack. For all other
1307 * cases, *new_stpp and *new_lopp should be malloc'd before the call,
1308 * in case they are used.
1309 */
1310 APPLESTATIC int
1311 nfsrv_lockctrl(vnode_t vp, struct nfsstate **new_stpp,
1312 struct nfslock **new_lopp, struct nfslockconflict *cfp,
1313 nfsquad_t clientid, nfsv4stateid_t *stateidp,
1314 __unused struct nfsexstuff *exp,
1315 struct nfsrv_descript *nd, NFSPROC_T *p)
1316 {
1317 struct nfslock *lop;
1318 struct nfsstate *new_stp = *new_stpp;
1319 struct nfslock *new_lop = *new_lopp;
1320 struct nfsstate *tstp, *mystp, *nstp;
1321 int specialid = 0;
1322 struct nfslockfile *lfp;
1323 struct nfslock *other_lop = NULL;
1324 struct nfsstate *stp, *lckstp = NULL;
1325 struct nfsclient *clp = NULL;
1326 u_int32_t bits;
1327 int error = 0, haslock = 0, ret, reterr;
1328 int getlckret, delegation = 0, filestruct_locked, vnode_unlocked = 0;
1329 fhandle_t nfh;
1330 uint64_t first, end;
1331 uint32_t lock_flags;
1332
1333 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1334 /*
1335 * Note the special cases of "all 1s" or "all 0s" stateids and
1336 * let reads with all 1s go ahead.
1337 */
1338 if (new_stp->ls_stateid.seqid == 0x0 &&
1339 new_stp->ls_stateid.other[0] == 0x0 &&
1340 new_stp->ls_stateid.other[1] == 0x0 &&
1341 new_stp->ls_stateid.other[2] == 0x0)
1342 specialid = 1;
1343 else if (new_stp->ls_stateid.seqid == 0xffffffff &&
1344 new_stp->ls_stateid.other[0] == 0xffffffff &&
1345 new_stp->ls_stateid.other[1] == 0xffffffff &&
1346 new_stp->ls_stateid.other[2] == 0xffffffff)
1347 specialid = 2;
1348 }
1349
1350 /*
1351 * Check for restart conditions (client and server).
1352 */
1353 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
1354 &new_stp->ls_stateid, specialid);
1355 if (error)
1356 goto out;
1357
1358 /*
1359 * Check for state resource limit exceeded.
1360 */
1361 if ((new_stp->ls_flags & NFSLCK_LOCK) &&
1362 nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
1363 error = NFSERR_RESOURCE;
1364 goto out;
1365 }
1366
1367 /*
1368 * For the lock case, get another nfslock structure,
1369 * just in case we need it.
1370 * Malloc now, before we start sifting through the linked lists,
1371 * in case we have to wait for memory.
1372 */
1373 tryagain:
1374 if (new_stp->ls_flags & NFSLCK_LOCK)
1375 MALLOC(other_lop, struct nfslock *, sizeof (struct nfslock),
1376 M_NFSDLOCK, M_WAITOK);
1377 filestruct_locked = 0;
1378 reterr = 0;
1379 lfp = NULL;
1380
1381 /*
1382 * Get the lockfile structure for CFH now, so we can do a sanity
1383 * check against the stateid, before incrementing the seqid#, since
1384 * we want to return NFSERR_BADSTATEID on failure and the seqid#
1385 * shouldn't be incremented for this case.
1386 * If nfsrv_getlockfile() returns -1, it means "not found", which
1387 * will be handled later.
1388 * If we are doing Lock/LockU and local locking is enabled, sleep
1389 * lock the nfslockfile structure.
1390 */
1391 getlckret = nfsrv_getlockfh(vp, new_stp->ls_flags, NULL, &nfh, p);
1392 NFSLOCKSTATE();
1393 if (getlckret == 0) {
1394 if ((new_stp->ls_flags & (NFSLCK_LOCK | NFSLCK_UNLOCK)) != 0 &&
1395 nfsrv_dolocallocks != 0 && nd->nd_repstat == 0) {
1396 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1397 &lfp, &nfh, 1);
1398 if (getlckret == 0)
1399 filestruct_locked = 1;
1400 } else
1401 getlckret = nfsrv_getlockfile(new_stp->ls_flags, NULL,
1402 &lfp, &nfh, 0);
1403 }
1404 if (getlckret != 0 && getlckret != -1)
1405 reterr = getlckret;
1406
1407 if (filestruct_locked != 0) {
1408 LIST_INIT(&lfp->lf_rollback);
1409 if ((new_stp->ls_flags & NFSLCK_LOCK)) {
1410 /*
1411 * For local locking, do the advisory locking now, so
1412 * that any conflict can be detected. A failure later
1413 * can be rolled back locally. If an error is returned,
1414 * struct nfslockfile has been unlocked and any local
1415 * locking rolled back.
1416 */
1417 NFSUNLOCKSTATE();
1418 if (vnode_unlocked == 0) {
1419 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl1");
1420 vnode_unlocked = 1;
1421 NFSVOPUNLOCK(vp, 0);
1422 }
1423 reterr = nfsrv_locallock(vp, lfp,
1424 (new_lop->lo_flags & (NFSLCK_READ | NFSLCK_WRITE)),
1425 new_lop->lo_first, new_lop->lo_end, cfp, p);
1426 NFSLOCKSTATE();
1427 }
1428 }
1429
1430 if (specialid == 0) {
1431 if (new_stp->ls_flags & NFSLCK_TEST) {
1432 /*
1433 * RFC 3530 does not list LockT as an op that renews a
1434 * lease, but the concensus seems to be that it is ok
1435 * for a server to do so.
1436 */
1437 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1438 (nfsquad_t)((u_quad_t)0), NULL, p);
1439
1440 /*
1441 * Since NFSERR_EXPIRED, NFSERR_ADMINREVOKED are not valid
1442 * error returns for LockT, just go ahead and test for a lock,
1443 * since there are no locks for this client, but other locks
1444 * can conflict. (ie. same client will always be false)
1445 */
1446 if (error == NFSERR_EXPIRED || error == NFSERR_ADMINREVOKED)
1447 error = 0;
1448 lckstp = new_stp;
1449 } else {
1450 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
1451 (nfsquad_t)((u_quad_t)0), NULL, p);
1452 if (error == 0)
1453 /*
1454 * Look up the stateid
1455 */
1456 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
1457 new_stp->ls_flags, &stp);
1458 /*
1459 * do some sanity checks for an unconfirmed open or a
1460 * stateid that refers to the wrong file, for an open stateid
1461 */
1462 if (error == 0 && (stp->ls_flags & NFSLCK_OPEN) &&
1463 ((stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM) ||
1464 (getlckret == 0 && stp->ls_lfp != lfp)))
1465 error = NFSERR_BADSTATEID;
1466 if (error == 0 &&
1467 (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) &&
1468 getlckret == 0 && stp->ls_lfp != lfp)
1469 error = NFSERR_BADSTATEID;
1470
1471 /*
1472 * If the lockowner stateid doesn't refer to the same file,
1473 * I believe that is considered ok, since some clients will
1474 * only create a single lockowner and use that for all locks
1475 * on all files.
1476 * For now, log it as a diagnostic, instead of considering it
1477 * a BadStateid.
1478 */
1479 if (error == 0 && (stp->ls_flags &
1480 (NFSLCK_OPEN | NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) == 0 &&
1481 getlckret == 0 && stp->ls_lfp != lfp) {
1482 #ifdef DIAGNOSTIC
1483 printf("Got a lock statid for different file open\n");
1484 #endif
1485 /*
1486 error = NFSERR_BADSTATEID;
1487 */
1488 }
1489
1490 if (error == 0) {
1491 if (new_stp->ls_flags & NFSLCK_OPENTOLOCK) {
1492 /*
1493 * If haslock set, we've already checked the seqid.
1494 */
1495 if (!haslock) {
1496 if (stp->ls_flags & NFSLCK_OPEN)
1497 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1498 stp->ls_openowner, new_stp->ls_op);
1499 else
1500 error = NFSERR_BADSTATEID;
1501 }
1502 if (!error)
1503 nfsrv_getowner(&stp->ls_open, new_stp, &lckstp);
1504 if (lckstp)
1505 /*
1506 * I believe this should be an error, but it
1507 * isn't obvious what NFSERR_xxx would be
1508 * appropriate, so I'll use NFSERR_INVAL for now.
1509 */
1510 error = NFSERR_INVAL;
1511 else
1512 lckstp = new_stp;
1513 } else if (new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK)) {
1514 /*
1515 * If haslock set, ditto above.
1516 */
1517 if (!haslock) {
1518 if (stp->ls_flags & NFSLCK_OPEN)
1519 error = NFSERR_BADSTATEID;
1520 else
1521 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
1522 stp, new_stp->ls_op);
1523 }
1524 lckstp = stp;
1525 } else {
1526 lckstp = stp;
1527 }
1528 }
1529 /*
1530 * If the seqid part of the stateid isn't the same, return
1531 * NFSERR_OLDSTATEID for cases other than I/O Ops.
1532 * For I/O Ops, only return NFSERR_OLDSTATEID if
1533 * nfsrv_returnoldstateid is set. (The concensus on the email
1534 * list was that most clients would prefer to not receive
1535 * NFSERR_OLDSTATEID for I/O Ops, but the RFC suggests that that
1536 * is what will happen, so I use the nfsrv_returnoldstateid to
1537 * allow for either server configuration.)
1538 */
1539 if (!error && stp->ls_stateid.seqid!=new_stp->ls_stateid.seqid &&
1540 (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1541 nfsrv_returnoldstateid))
1542 error = NFSERR_OLDSTATEID;
1543 }
1544 }
1545
1546 /*
1547 * Now we can check for grace.
1548 */
1549 if (!error)
1550 error = nfsrv_checkgrace(new_stp->ls_flags);
1551 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
1552 nfsrv_checkstable(clp))
1553 error = NFSERR_NOGRACE;
1554 /*
1555 * If we successfully Reclaimed state, note that.
1556 */
1557 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error)
1558 nfsrv_markstable(clp);
1559
1560 /*
1561 * At this point, either error == NFSERR_BADSTATEID or the
1562 * seqid# has been updated, so we can return any error.
1563 * If error == 0, there may be an error in:
1564 * nd_repstat - Set by the calling function.
1565 * reterr - Set above, if getting the nfslockfile structure
1566 * or acquiring the local lock failed.
1567 * (If both of these are set, nd_repstat should probably be
1568 * returned, since that error was detected before this
1569 * function call.)
1570 */
1571 if (error != 0 || nd->nd_repstat != 0 || reterr != 0) {
1572 if (error == 0) {
1573 if (nd->nd_repstat != 0)
1574 error = nd->nd_repstat;
1575 else
1576 error = reterr;
1577 }
1578 if (filestruct_locked != 0) {
1579 /* Roll back local locks. */
1580 NFSUNLOCKSTATE();
1581 if (vnode_unlocked == 0) {
1582 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl2");
1583 vnode_unlocked = 1;
1584 NFSVOPUNLOCK(vp, 0);
1585 }
1586 nfsrv_locallock_rollback(vp, lfp, p);
1587 NFSLOCKSTATE();
1588 nfsrv_unlocklf(lfp);
1589 }
1590 NFSUNLOCKSTATE();
1591 goto out;
1592 }
1593
1594 /*
1595 * Check the nfsrv_getlockfile return.
1596 * Returned -1 if no structure found.
1597 */
1598 if (getlckret == -1) {
1599 error = NFSERR_EXPIRED;
1600 /*
1601 * Called from lockt, so no lock is OK.
1602 */
1603 if (new_stp->ls_flags & NFSLCK_TEST) {
1604 error = 0;
1605 } else if (new_stp->ls_flags &
1606 (NFSLCK_CHECK | NFSLCK_SETATTR)) {
1607 /*
1608 * Called to check for a lock, OK if the stateid is all
1609 * 1s or all 0s, but there should be an nfsstate
1610 * otherwise.
1611 * (ie. If there is no open, I'll assume no share
1612 * deny bits.)
1613 */
1614 if (specialid)
1615 error = 0;
1616 else
1617 error = NFSERR_BADSTATEID;
1618 }
1619 NFSUNLOCKSTATE();
1620 goto out;
1621 }
1622
1623 /*
1624 * For NFSLCK_CHECK and NFSLCK_LOCK, test for a share conflict.
1625 * For NFSLCK_CHECK, allow a read if write access is granted,
1626 * but check for a deny. For NFSLCK_LOCK, require correct access,
1627 * which implies a conflicting deny can't exist.
1628 */
1629 if (new_stp->ls_flags & (NFSLCK_CHECK | NFSLCK_LOCK)) {
1630 /*
1631 * Four kinds of state id:
1632 * - specialid (all 0s or all 1s), only for NFSLCK_CHECK
1633 * - stateid for an open
1634 * - stateid for a delegation
1635 * - stateid for a lock owner
1636 */
1637 if (!specialid) {
1638 if (stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
1639 delegation = 1;
1640 mystp = stp;
1641 nfsrv_delaydelegtimeout(stp);
1642 } else if (stp->ls_flags & NFSLCK_OPEN) {
1643 mystp = stp;
1644 } else {
1645 mystp = stp->ls_openstp;
1646 }
1647 /*
1648 * If locking or checking, require correct access
1649 * bit set.
1650 */
1651 if (((new_stp->ls_flags & NFSLCK_LOCK) &&
1652 !((new_lop->lo_flags >> NFSLCK_LOCKSHIFT) &
1653 mystp->ls_flags & NFSLCK_ACCESSBITS)) ||
1654 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_READACCESS)) ==
1655 (NFSLCK_CHECK | NFSLCK_READACCESS) &&
1656 !(mystp->ls_flags & NFSLCK_READACCESS)) ||
1657 ((new_stp->ls_flags & (NFSLCK_CHECK|NFSLCK_WRITEACCESS)) ==
1658 (NFSLCK_CHECK | NFSLCK_WRITEACCESS) &&
1659 !(mystp->ls_flags & NFSLCK_WRITEACCESS))) {
1660 if (filestruct_locked != 0) {
1661 /* Roll back local locks. */
1662 NFSUNLOCKSTATE();
1663 if (vnode_unlocked == 0) {
1664 ASSERT_VOP_ELOCKED(vp,
1665 "nfsrv_lockctrl3");
1666 vnode_unlocked = 1;
1667 NFSVOPUNLOCK(vp, 0);
1668 }
1669 nfsrv_locallock_rollback(vp, lfp, p);
1670 NFSLOCKSTATE();
1671 nfsrv_unlocklf(lfp);
1672 }
1673 NFSUNLOCKSTATE();
1674 error = NFSERR_OPENMODE;
1675 goto out;
1676 }
1677 } else
1678 mystp = NULL;
1679 if ((new_stp->ls_flags & NFSLCK_CHECK) && !delegation) {
1680 /*
1681 * Check for a conflicting deny bit.
1682 */
1683 LIST_FOREACH(tstp, &lfp->lf_open, ls_file) {
1684 if (tstp != mystp) {
1685 bits = tstp->ls_flags;
1686 bits >>= NFSLCK_SHIFT;
1687 if (new_stp->ls_flags & bits & NFSLCK_ACCESSBITS) {
1688 KASSERT(vnode_unlocked == 0,
1689 ("nfsrv_lockctrl: vnode unlocked1"));
1690 ret = nfsrv_clientconflict(tstp->ls_clp, &haslock,
1691 vp, p);
1692 if (ret == 1) {
1693 /*
1694 * nfsrv_clientconflict unlocks state
1695 * when it returns non-zero.
1696 */
1697 lckstp = NULL;
1698 goto tryagain;
1699 }
1700 if (ret == 0)
1701 NFSUNLOCKSTATE();
1702 if (ret == 2)
1703 error = NFSERR_PERM;
1704 else
1705 error = NFSERR_OPENMODE;
1706 goto out;
1707 }
1708 }
1709 }
1710
1711 /* We're outta here */
1712 NFSUNLOCKSTATE();
1713 goto out;
1714 }
1715 }
1716
1717 /*
1718 * For setattr, just get rid of all the Delegations for other clients.
1719 */
1720 if (new_stp->ls_flags & NFSLCK_SETATTR) {
1721 KASSERT(vnode_unlocked == 0,
1722 ("nfsrv_lockctrl: vnode unlocked2"));
1723 ret = nfsrv_cleandeleg(vp, lfp, clp, &haslock, p);
1724 if (ret) {
1725 /*
1726 * nfsrv_cleandeleg() unlocks state when it
1727 * returns non-zero.
1728 */
1729 if (ret == -1) {
1730 lckstp = NULL;
1731 goto tryagain;
1732 }
1733 error = ret;
1734 goto out;
1735 }
1736 if (!(new_stp->ls_flags & NFSLCK_CHECK) ||
1737 (LIST_EMPTY(&lfp->lf_open) && LIST_EMPTY(&lfp->lf_lock) &&
1738 LIST_EMPTY(&lfp->lf_deleg))) {
1739 NFSUNLOCKSTATE();
1740 goto out;
1741 }
1742 }
1743
1744 /*
1745 * Check for a conflicting delegation. If one is found, call
1746 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
1747 * been set yet, it will get the lock. Otherwise, it will recall
1748 * the delegation. Then, we try try again...
1749 * I currently believe the conflict algorithm to be:
1750 * For Lock Ops (Lock/LockT/LockU)
1751 * - there is a conflict iff a different client has a write delegation
1752 * For Reading (Read Op)
1753 * - there is a conflict iff a different client has a write delegation
1754 * (the specialids are always a different client)
1755 * For Writing (Write/Setattr of size)
1756 * - there is a conflict if a different client has any delegation
1757 * - there is a conflict if the same client has a read delegation
1758 * (I don't understand why this isn't allowed, but that seems to be
1759 * the current concensus?)
1760 */
1761 tstp = LIST_FIRST(&lfp->lf_deleg);
1762 while (tstp != LIST_END(&lfp->lf_deleg)) {
1763 nstp = LIST_NEXT(tstp, ls_file);
1764 if ((((new_stp->ls_flags&(NFSLCK_LOCK|NFSLCK_UNLOCK|NFSLCK_TEST))||
1765 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1766 (new_lop->lo_flags & NFSLCK_READ))) &&
1767 clp != tstp->ls_clp &&
1768 (tstp->ls_flags & NFSLCK_DELEGWRITE)) ||
1769 ((new_stp->ls_flags & NFSLCK_CHECK) &&
1770 (new_lop->lo_flags & NFSLCK_WRITE) &&
1771 (clp != tstp->ls_clp ||
1772 (tstp->ls_flags & NFSLCK_DELEGREAD)))) {
1773 ret = 0;
1774 if (filestruct_locked != 0) {
1775 /* Roll back local locks. */
1776 NFSUNLOCKSTATE();
1777 if (vnode_unlocked == 0) {
1778 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl4");
1779 NFSVOPUNLOCK(vp, 0);
1780 }
1781 nfsrv_locallock_rollback(vp, lfp, p);
1782 NFSLOCKSTATE();
1783 nfsrv_unlocklf(lfp);
1784 NFSUNLOCKSTATE();
1785 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1786 vnode_unlocked = 0;
1787 if ((vp->v_iflag & VI_DOOMED) != 0)
1788 ret = NFSERR_SERVERFAULT;
1789 NFSLOCKSTATE();
1790 }
1791 if (ret == 0)
1792 ret = nfsrv_delegconflict(tstp, &haslock, p, vp);
1793 if (ret) {
1794 /*
1795 * nfsrv_delegconflict unlocks state when it
1796 * returns non-zero, which it always does.
1797 */
1798 if (other_lop) {
1799 FREE((caddr_t)other_lop, M_NFSDLOCK);
1800 other_lop = NULL;
1801 }
1802 if (ret == -1) {
1803 lckstp = NULL;
1804 goto tryagain;
1805 }
1806 error = ret;
1807 goto out;
1808 }
1809 /* Never gets here. */
1810 }
1811 tstp = nstp;
1812 }
1813
1814 /*
1815 * Handle the unlock case by calling nfsrv_updatelock().
1816 * (Should I have done some access checking above for unlock? For now,
1817 * just let it happen.)
1818 */
1819 if (new_stp->ls_flags & NFSLCK_UNLOCK) {
1820 first = new_lop->lo_first;
1821 end = new_lop->lo_end;
1822 nfsrv_updatelock(stp, new_lopp, &other_lop, lfp);
1823 stateidp->seqid = ++(stp->ls_stateid.seqid);
1824 stateidp->other[0] = stp->ls_stateid.other[0];
1825 stateidp->other[1] = stp->ls_stateid.other[1];
1826 stateidp->other[2] = stp->ls_stateid.other[2];
1827 if (filestruct_locked != 0) {
1828 NFSUNLOCKSTATE();
1829 if (vnode_unlocked == 0) {
1830 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl5");
1831 vnode_unlocked = 1;
1832 NFSVOPUNLOCK(vp, 0);
1833 }
1834 /* Update the local locks. */
1835 nfsrv_localunlock(vp, lfp, first, end, p);
1836 NFSLOCKSTATE();
1837 nfsrv_unlocklf(lfp);
1838 }
1839 NFSUNLOCKSTATE();
1840 goto out;
1841 }
1842
1843 /*
1844 * Search for a conflicting lock. A lock conflicts if:
1845 * - the lock range overlaps and
1846 * - at least one lock is a write lock and
1847 * - it is not owned by the same lock owner
1848 */
1849 if (!delegation) {
1850 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
1851 if (new_lop->lo_end > lop->lo_first &&
1852 new_lop->lo_first < lop->lo_end &&
1853 (new_lop->lo_flags == NFSLCK_WRITE ||
1854 lop->lo_flags == NFSLCK_WRITE) &&
1855 lckstp != lop->lo_stp &&
1856 (clp != lop->lo_stp->ls_clp ||
1857 lckstp->ls_ownerlen != lop->lo_stp->ls_ownerlen ||
1858 NFSBCMP(lckstp->ls_owner, lop->lo_stp->ls_owner,
1859 lckstp->ls_ownerlen))) {
1860 if (other_lop) {
1861 FREE((caddr_t)other_lop, M_NFSDLOCK);
1862 other_lop = NULL;
1863 }
1864 if (vnode_unlocked != 0)
1865 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
1866 NULL, p);
1867 else
1868 ret = nfsrv_clientconflict(lop->lo_stp->ls_clp, &haslock,
1869 vp, p);
1870 if (ret == 1) {
1871 if (filestruct_locked != 0) {
1872 if (vnode_unlocked == 0) {
1873 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl6");
1874 NFSVOPUNLOCK(vp, 0);
1875 }
1876 /* Roll back local locks. */
1877 nfsrv_locallock_rollback(vp, lfp, p);
1878 NFSLOCKSTATE();
1879 nfsrv_unlocklf(lfp);
1880 NFSUNLOCKSTATE();
1881 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1882 vnode_unlocked = 0;
1883 if ((vp->v_iflag & VI_DOOMED) != 0) {
1884 error = NFSERR_SERVERFAULT;
1885 goto out;
1886 }
1887 }
1888 /*
1889 * nfsrv_clientconflict() unlocks state when it
1890 * returns non-zero.
1891 */
1892 lckstp = NULL;
1893 goto tryagain;
1894 }
1895 /*
1896 * Found a conflicting lock, so record the conflict and
1897 * return the error.
1898 */
1899 if (cfp != NULL && ret == 0) {
1900 cfp->cl_clientid.lval[0]=lop->lo_stp->ls_stateid.other[0];
1901 cfp->cl_clientid.lval[1]=lop->lo_stp->ls_stateid.other[1];
1902 cfp->cl_first = lop->lo_first;
1903 cfp->cl_end = lop->lo_end;
1904 cfp->cl_flags = lop->lo_flags;
1905 cfp->cl_ownerlen = lop->lo_stp->ls_ownerlen;
1906 NFSBCOPY(lop->lo_stp->ls_owner, cfp->cl_owner,
1907 cfp->cl_ownerlen);
1908 }
1909 if (ret == 2)
1910 error = NFSERR_PERM;
1911 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
1912 error = NFSERR_RECLAIMCONFLICT;
1913 else if (new_stp->ls_flags & NFSLCK_CHECK)
1914 error = NFSERR_LOCKED;
1915 else
1916 error = NFSERR_DENIED;
1917 if (filestruct_locked != 0 && ret == 0) {
1918 /* Roll back local locks. */
1919 NFSUNLOCKSTATE();
1920 if (vnode_unlocked == 0) {
1921 ASSERT_VOP_ELOCKED(vp, "nfsrv_lockctrl7");
1922 vnode_unlocked = 1;
1923 NFSVOPUNLOCK(vp, 0);
1924 }
1925 nfsrv_locallock_rollback(vp, lfp, p);
1926 NFSLOCKSTATE();
1927 nfsrv_unlocklf(lfp);
1928 }
1929 if (ret == 0)
1930 NFSUNLOCKSTATE();
1931 goto out;
1932 }
1933 }
1934 }
1935
1936 /*
1937 * We only get here if there was no lock that conflicted.
1938 */
1939 if (new_stp->ls_flags & (NFSLCK_TEST | NFSLCK_CHECK)) {
1940 NFSUNLOCKSTATE();
1941 goto out;
1942 }
1943
1944 /*
1945 * We only get here when we are creating or modifying a lock.
1946 * There are two variants:
1947 * - exist_lock_owner where lock_owner exists
1948 * - open_to_lock_owner with new lock_owner
1949 */
1950 first = new_lop->lo_first;
1951 end = new_lop->lo_end;
1952 lock_flags = new_lop->lo_flags;
1953 if (!(new_stp->ls_flags & NFSLCK_OPENTOLOCK)) {
1954 nfsrv_updatelock(lckstp, new_lopp, &other_lop, lfp);
1955 stateidp->seqid = ++(lckstp->ls_stateid.seqid);
1956 stateidp->other[0] = lckstp->ls_stateid.other[0];
1957 stateidp->other[1] = lckstp->ls_stateid.other[1];
1958 stateidp->other[2] = lckstp->ls_stateid.other[2];
1959 } else {
1960 /*
1961 * The new open_to_lock_owner case.
1962 * Link the new nfsstate into the lists.
1963 */
1964 new_stp->ls_seq = new_stp->ls_opentolockseq;
1965 nfsrvd_refcache(new_stp->ls_op);
1966 stateidp->seqid = new_stp->ls_stateid.seqid = 1;
1967 stateidp->other[0] = new_stp->ls_stateid.other[0] =
1968 clp->lc_clientid.lval[0];
1969 stateidp->other[1] = new_stp->ls_stateid.other[1] =
1970 clp->lc_clientid.lval[1];
1971 stateidp->other[2] = new_stp->ls_stateid.other[2] =
1972 nfsrv_nextstateindex(clp);
1973 new_stp->ls_clp = clp;
1974 LIST_INIT(&new_stp->ls_lock);
1975 new_stp->ls_openstp = stp;
1976 new_stp->ls_lfp = lfp;
1977 nfsrv_insertlock(new_lop, (struct nfslock *)new_stp, new_stp,
1978 lfp);
1979 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_stp->ls_stateid),
1980 new_stp, ls_hash);
1981 LIST_INSERT_HEAD(&stp->ls_open, new_stp, ls_list);
1982 *new_lopp = NULL;
1983 *new_stpp = NULL;
1984 newnfsstats.srvlockowners++;
1985 nfsrv_openpluslock++;
1986 }
1987 if (filestruct_locked != 0) {
1988 NFSUNLOCKSTATE();
1989 nfsrv_locallock_commit(lfp, lock_flags, first, end);
1990 NFSLOCKSTATE();
1991 nfsrv_unlocklf(lfp);
1992 }
1993 NFSUNLOCKSTATE();
1994
1995 out:
1996 if (haslock) {
1997 NFSLOCKV4ROOTMUTEX();
1998 nfsv4_unlock(&nfsv4rootfs_lock, 1);
1999 NFSUNLOCKV4ROOTMUTEX();
2000 }
2001 if (vnode_unlocked != 0) {
2002 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
2003 if (error == 0 && (vp->v_iflag & VI_DOOMED) != 0)
2004 error = NFSERR_SERVERFAULT;
2005 }
2006 if (other_lop)
2007 FREE((caddr_t)other_lop, M_NFSDLOCK);
2008 NFSEXITCODE2(error, nd);
2009 return (error);
2010 }
2011
2012 /*
2013 * Check for state errors for Open.
2014 * repstat is passed back out as an error if more critical errors
2015 * are not detected.
2016 */
2017 APPLESTATIC int
2018 nfsrv_opencheck(nfsquad_t clientid, nfsv4stateid_t *stateidp,
2019 struct nfsstate *new_stp, vnode_t vp, struct nfsrv_descript *nd,
2020 NFSPROC_T *p, int repstat)
2021 {
2022 struct nfsstate *stp, *nstp;
2023 struct nfsclient *clp;
2024 struct nfsstate *ownerstp;
2025 struct nfslockfile *lfp, *new_lfp;
2026 int error = 0, haslock = 0, ret, readonly = 0, getfhret = 0;
2027
2028 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2029 readonly = 1;
2030 /*
2031 * Check for restart conditions (client and server).
2032 */
2033 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2034 &new_stp->ls_stateid, 0);
2035 if (error)
2036 goto out;
2037
2038 /*
2039 * Check for state resource limit exceeded.
2040 * Technically this should be SMP protected, but the worst
2041 * case error is "out by one or two" on the count when it
2042 * returns NFSERR_RESOURCE and the limit is just a rather
2043 * arbitrary high water mark, so no harm is done.
2044 */
2045 if (nfsrv_openpluslock > NFSRV_V4STATELIMIT) {
2046 error = NFSERR_RESOURCE;
2047 goto out;
2048 }
2049
2050 tryagain:
2051 MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2052 M_NFSDLOCKFILE, M_WAITOK);
2053 if (vp)
2054 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2055 NULL, p);
2056 NFSLOCKSTATE();
2057 /*
2058 * Get the nfsclient structure.
2059 */
2060 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2061 (nfsquad_t)((u_quad_t)0), NULL, p);
2062
2063 /*
2064 * Look up the open owner. See if it needs confirmation and
2065 * check the seq#, as required.
2066 */
2067 if (!error)
2068 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2069
2070 if (!error && ownerstp) {
2071 error = nfsrv_checkseqid(nd, new_stp->ls_seq, ownerstp,
2072 new_stp->ls_op);
2073 /*
2074 * If the OpenOwner hasn't been confirmed, assume the
2075 * old one was a replay and this one is ok.
2076 * See: RFC3530 Sec. 14.2.18.
2077 */
2078 if (error == NFSERR_BADSEQID &&
2079 (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM))
2080 error = 0;
2081 }
2082
2083 /*
2084 * Check for grace.
2085 */
2086 if (!error)
2087 error = nfsrv_checkgrace(new_stp->ls_flags);
2088 if ((new_stp->ls_flags & NFSLCK_RECLAIM) && !error &&
2089 nfsrv_checkstable(clp))
2090 error = NFSERR_NOGRACE;
2091
2092 /*
2093 * If none of the above errors occurred, let repstat be
2094 * returned.
2095 */
2096 if (repstat && !error)
2097 error = repstat;
2098 if (error) {
2099 NFSUNLOCKSTATE();
2100 if (haslock) {
2101 NFSLOCKV4ROOTMUTEX();
2102 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2103 NFSUNLOCKV4ROOTMUTEX();
2104 }
2105 free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2106 goto out;
2107 }
2108
2109 /*
2110 * If vp == NULL, the file doesn't exist yet, so return ok.
2111 * (This always happens on the first pass, so haslock must be 0.)
2112 */
2113 if (vp == NULL) {
2114 NFSUNLOCKSTATE();
2115 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2116 goto out;
2117 }
2118
2119 /*
2120 * Get the structure for the underlying file.
2121 */
2122 if (getfhret)
2123 error = getfhret;
2124 else
2125 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2126 NULL, 0);
2127 if (new_lfp)
2128 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2129 if (error) {
2130 NFSUNLOCKSTATE();
2131 if (haslock) {
2132 NFSLOCKV4ROOTMUTEX();
2133 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2134 NFSUNLOCKV4ROOTMUTEX();
2135 }
2136 goto out;
2137 }
2138
2139 /*
2140 * Search for a conflicting open/share.
2141 */
2142 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2143 /*
2144 * For Delegate_Cur, search for the matching Delegation,
2145 * which indicates no conflict.
2146 * An old delegation should have been recovered by the
2147 * client doing a Claim_DELEGATE_Prev, so I won't let
2148 * it match and return NFSERR_EXPIRED. Should I let it
2149 * match?
2150 */
2151 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2152 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2153 stateidp->seqid == stp->ls_stateid.seqid &&
2154 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2155 NFSX_STATEIDOTHER))
2156 break;
2157 }
2158 if (stp == LIST_END(&lfp->lf_deleg) ||
2159 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2160 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2161 NFSUNLOCKSTATE();
2162 if (haslock) {
2163 NFSLOCKV4ROOTMUTEX();
2164 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2165 NFSUNLOCKV4ROOTMUTEX();
2166 }
2167 error = NFSERR_EXPIRED;
2168 goto out;
2169 }
2170 }
2171
2172 /*
2173 * Check for access/deny bit conflicts. I check for the same
2174 * owner as well, in case the client didn't bother.
2175 */
2176 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2177 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR) &&
2178 (((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2179 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2180 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2181 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS)))){
2182 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2183 if (ret == 1) {
2184 /*
2185 * nfsrv_clientconflict() unlocks
2186 * state when it returns non-zero.
2187 */
2188 goto tryagain;
2189 }
2190 if (ret == 2)
2191 error = NFSERR_PERM;
2192 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2193 error = NFSERR_RECLAIMCONFLICT;
2194 else
2195 error = NFSERR_SHAREDENIED;
2196 if (ret == 0)
2197 NFSUNLOCKSTATE();
2198 if (haslock) {
2199 NFSLOCKV4ROOTMUTEX();
2200 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2201 NFSUNLOCKV4ROOTMUTEX();
2202 }
2203 goto out;
2204 }
2205 }
2206
2207 /*
2208 * Check for a conflicting delegation. If one is found, call
2209 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2210 * been set yet, it will get the lock. Otherwise, it will recall
2211 * the delegation. Then, we try try again...
2212 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2213 * isn't a conflict.)
2214 * I currently believe the conflict algorithm to be:
2215 * For Open with Read Access and Deny None
2216 * - there is a conflict iff a different client has a write delegation
2217 * For Open with other Write Access or any Deny except None
2218 * - there is a conflict if a different client has any delegation
2219 * - there is a conflict if the same client has a read delegation
2220 * (The current concensus is that this last case should be
2221 * considered a conflict since the client with a read delegation
2222 * could have done an Open with ReadAccess and WriteDeny
2223 * locally and then not have checked for the WriteDeny.)
2224 * Don't check for a Reclaim, since that will be dealt with
2225 * by nfsrv_openctrl().
2226 */
2227 if (!(new_stp->ls_flags &
2228 (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR | NFSLCK_RECLAIM))) {
2229 stp = LIST_FIRST(&lfp->lf_deleg);
2230 while (stp != LIST_END(&lfp->lf_deleg)) {
2231 nstp = LIST_NEXT(stp, ls_file);
2232 if ((readonly && stp->ls_clp != clp &&
2233 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2234 (!readonly && (stp->ls_clp != clp ||
2235 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2236 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2237 if (ret) {
2238 /*
2239 * nfsrv_delegconflict() unlocks state
2240 * when it returns non-zero.
2241 */
2242 if (ret == -1)
2243 goto tryagain;
2244 error = ret;
2245 goto out;
2246 }
2247 }
2248 stp = nstp;
2249 }
2250 }
2251 NFSUNLOCKSTATE();
2252 if (haslock) {
2253 NFSLOCKV4ROOTMUTEX();
2254 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2255 NFSUNLOCKV4ROOTMUTEX();
2256 }
2257
2258 out:
2259 NFSEXITCODE2(error, nd);
2260 return (error);
2261 }
2262
2263 /*
2264 * Open control function to create/update open state for an open.
2265 */
2266 APPLESTATIC int
2267 nfsrv_openctrl(struct nfsrv_descript *nd, vnode_t vp,
2268 struct nfsstate **new_stpp, nfsquad_t clientid, nfsv4stateid_t *stateidp,
2269 nfsv4stateid_t *delegstateidp, u_int32_t *rflagsp, struct nfsexstuff *exp,
2270 NFSPROC_T *p, u_quad_t filerev)
2271 {
2272 struct nfsstate *new_stp = *new_stpp;
2273 struct nfsstate *stp, *nstp;
2274 struct nfsstate *openstp = NULL, *new_open, *ownerstp, *new_deleg;
2275 struct nfslockfile *lfp, *new_lfp;
2276 struct nfsclient *clp;
2277 int error = 0, haslock = 0, ret, delegate = 1, writedeleg = 1;
2278 int readonly = 0, cbret = 1, getfhret = 0;
2279
2280 if ((new_stp->ls_flags & NFSLCK_SHAREBITS) == NFSLCK_READACCESS)
2281 readonly = 1;
2282 /*
2283 * Check for restart conditions (client and server).
2284 * (Paranoia, should have been detected by nfsrv_opencheck().)
2285 * If an error does show up, return NFSERR_EXPIRED, since the
2286 * the seqid# has already been incremented.
2287 */
2288 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2289 &new_stp->ls_stateid, 0);
2290 if (error) {
2291 printf("Nfsd: openctrl unexpected restart err=%d\n",
2292 error);
2293 error = NFSERR_EXPIRED;
2294 goto out;
2295 }
2296
2297 tryagain:
2298 MALLOC(new_lfp, struct nfslockfile *, sizeof (struct nfslockfile),
2299 M_NFSDLOCKFILE, M_WAITOK);
2300 MALLOC(new_open, struct nfsstate *, sizeof (struct nfsstate),
2301 M_NFSDSTATE, M_WAITOK);
2302 MALLOC(new_deleg, struct nfsstate *, sizeof (struct nfsstate),
2303 M_NFSDSTATE, M_WAITOK);
2304 getfhret = nfsrv_getlockfh(vp, new_stp->ls_flags, new_lfp,
2305 NULL, p);
2306 NFSLOCKSTATE();
2307 /*
2308 * Get the client structure. Since the linked lists could be changed
2309 * by other nfsd processes if this process does a tsleep(), one of
2310 * two things must be done.
2311 * 1 - don't tsleep()
2312 * or
2313 * 2 - get the nfsv4_lock() { indicated by haslock == 1 }
2314 * before using the lists, since this lock stops the other
2315 * nfsd. This should only be used for rare cases, since it
2316 * essentially single threads the nfsd.
2317 * At this time, it is only done for cases where the stable
2318 * storage file must be written prior to completion of state
2319 * expiration.
2320 */
2321 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2322 (nfsquad_t)((u_quad_t)0), NULL, p);
2323 if (!error && (clp->lc_flags & LCL_NEEDSCBNULL) &&
2324 clp->lc_program) {
2325 /*
2326 * This happens on the first open for a client
2327 * that supports callbacks.
2328 */
2329 NFSUNLOCKSTATE();
2330 /*
2331 * Although nfsrv_docallback() will sleep, clp won't
2332 * go away, since they are only removed when the
2333 * nfsv4_lock() has blocked the nfsd threads. The
2334 * fields in clp can change, but having multiple
2335 * threads do this Null callback RPC should be
2336 * harmless.
2337 */
2338 cbret = nfsrv_docallback(clp, NFSV4PROC_CBNULL,
2339 NULL, 0, NULL, NULL, NULL, p);
2340 NFSLOCKSTATE();
2341 clp->lc_flags &= ~LCL_NEEDSCBNULL;
2342 if (!cbret)
2343 clp->lc_flags |= LCL_CALLBACKSON;
2344 }
2345
2346 /*
2347 * Look up the open owner. See if it needs confirmation and
2348 * check the seq#, as required.
2349 */
2350 if (!error)
2351 nfsrv_getowner(&clp->lc_open, new_stp, &ownerstp);
2352
2353 if (error) {
2354 NFSUNLOCKSTATE();
2355 printf("Nfsd: openctrl unexpected state err=%d\n",
2356 error);
2357 free((caddr_t)new_lfp, M_NFSDLOCKFILE);
2358 free((caddr_t)new_open, M_NFSDSTATE);
2359 free((caddr_t)new_deleg, M_NFSDSTATE);
2360 if (haslock) {
2361 NFSLOCKV4ROOTMUTEX();
2362 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2363 NFSUNLOCKV4ROOTMUTEX();
2364 }
2365 error = NFSERR_EXPIRED;
2366 goto out;
2367 }
2368
2369 if (new_stp->ls_flags & NFSLCK_RECLAIM)
2370 nfsrv_markstable(clp);
2371
2372 /*
2373 * Get the structure for the underlying file.
2374 */
2375 if (getfhret)
2376 error = getfhret;
2377 else
2378 error = nfsrv_getlockfile(new_stp->ls_flags, &new_lfp, &lfp,
2379 NULL, 0);
2380 if (new_lfp)
2381 FREE((caddr_t)new_lfp, M_NFSDLOCKFILE);
2382 if (error) {
2383 NFSUNLOCKSTATE();
2384 printf("Nfsd openctrl unexpected getlockfile err=%d\n",
2385 error);
2386 free((caddr_t)new_open, M_NFSDSTATE);
2387 free((caddr_t)new_deleg, M_NFSDSTATE);
2388 if (haslock) {
2389 NFSLOCKV4ROOTMUTEX();
2390 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2391 NFSUNLOCKV4ROOTMUTEX();
2392 }
2393 goto out;
2394 }
2395
2396 /*
2397 * Search for a conflicting open/share.
2398 */
2399 if (new_stp->ls_flags & NFSLCK_DELEGCUR) {
2400 /*
2401 * For Delegate_Cur, search for the matching Delegation,
2402 * which indicates no conflict.
2403 * An old delegation should have been recovered by the
2404 * client doing a Claim_DELEGATE_Prev, so I won't let
2405 * it match and return NFSERR_EXPIRED. Should I let it
2406 * match?
2407 */
2408 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2409 if (!(stp->ls_flags & NFSLCK_OLDDELEG) &&
2410 stateidp->seqid == stp->ls_stateid.seqid &&
2411 !NFSBCMP(stateidp->other, stp->ls_stateid.other,
2412 NFSX_STATEIDOTHER))
2413 break;
2414 }
2415 if (stp == LIST_END(&lfp->lf_deleg) ||
2416 ((new_stp->ls_flags & NFSLCK_WRITEACCESS) &&
2417 (stp->ls_flags & NFSLCK_DELEGREAD))) {
2418 NFSUNLOCKSTATE();
2419 printf("Nfsd openctrl unexpected expiry\n");
2420 free((caddr_t)new_open, M_NFSDSTATE);
2421 free((caddr_t)new_deleg, M_NFSDSTATE);
2422 if (haslock) {
2423 NFSLOCKV4ROOTMUTEX();
2424 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2425 NFSUNLOCKV4ROOTMUTEX();
2426 }
2427 error = NFSERR_EXPIRED;
2428 goto out;
2429 }
2430
2431 /*
2432 * Don't issue a Delegation, since one already exists and
2433 * delay delegation timeout, as required.
2434 */
2435 delegate = 0;
2436 nfsrv_delaydelegtimeout(stp);
2437 }
2438
2439 /*
2440 * Check for access/deny bit conflicts. I also check for the
2441 * same owner, since the client might not have bothered to check.
2442 * Also, note an open for the same file and owner, if found,
2443 * which is all we do here for Delegate_Cur, since conflict
2444 * checking is already done.
2445 */
2446 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
2447 if (ownerstp && stp->ls_openowner == ownerstp)
2448 openstp = stp;
2449 if (!(new_stp->ls_flags & NFSLCK_DELEGCUR)) {
2450 /*
2451 * If another client has the file open, the only
2452 * delegation that can be issued is a Read delegation
2453 * and only if it is a Read open with Deny none.
2454 */
2455 if (clp != stp->ls_clp) {
2456 if ((stp->ls_flags & NFSLCK_SHAREBITS) ==
2457 NFSLCK_READACCESS)
2458 writedeleg = 0;
2459 else
2460 delegate = 0;
2461 }
2462 if(((new_stp->ls_flags & NFSLCK_ACCESSBITS) &
2463 ((stp->ls_flags>>NFSLCK_SHIFT) & NFSLCK_ACCESSBITS))||
2464 ((stp->ls_flags & NFSLCK_ACCESSBITS) &
2465 ((new_stp->ls_flags>>NFSLCK_SHIFT)&NFSLCK_ACCESSBITS))){
2466 ret = nfsrv_clientconflict(stp->ls_clp,&haslock,vp,p);
2467 if (ret == 1) {
2468 /*
2469 * nfsrv_clientconflict() unlocks state
2470 * when it returns non-zero.
2471 */
2472 free((caddr_t)new_open, M_NFSDSTATE);
2473 free((caddr_t)new_deleg, M_NFSDSTATE);
2474 openstp = NULL;
2475 goto tryagain;
2476 }
2477 if (ret == 2)
2478 error = NFSERR_PERM;
2479 else if (new_stp->ls_flags & NFSLCK_RECLAIM)
2480 error = NFSERR_RECLAIMCONFLICT;
2481 else
2482 error = NFSERR_SHAREDENIED;
2483 if (ret == 0)
2484 NFSUNLOCKSTATE();
2485 if (haslock) {
2486 NFSLOCKV4ROOTMUTEX();
2487 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2488 NFSUNLOCKV4ROOTMUTEX();
2489 }
2490 free((caddr_t)new_open, M_NFSDSTATE);
2491 free((caddr_t)new_deleg, M_NFSDSTATE);
2492 printf("nfsd openctrl unexpected client cnfl\n");
2493 goto out;
2494 }
2495 }
2496 }
2497
2498 /*
2499 * Check for a conflicting delegation. If one is found, call
2500 * nfsrv_delegconflict() to handle it. If the v4root lock hasn't
2501 * been set yet, it will get the lock. Otherwise, it will recall
2502 * the delegation. Then, we try try again...
2503 * (If NFSLCK_DELEGCUR is set, it has a delegation, so there
2504 * isn't a conflict.)
2505 * I currently believe the conflict algorithm to be:
2506 * For Open with Read Access and Deny None
2507 * - there is a conflict iff a different client has a write delegation
2508 * For Open with other Write Access or any Deny except None
2509 * - there is a conflict if a different client has any delegation
2510 * - there is a conflict if the same client has a read delegation
2511 * (The current concensus is that this last case should be
2512 * considered a conflict since the client with a read delegation
2513 * could have done an Open with ReadAccess and WriteDeny
2514 * locally and then not have checked for the WriteDeny.)
2515 */
2516 if (!(new_stp->ls_flags & (NFSLCK_DELEGPREV | NFSLCK_DELEGCUR))) {
2517 stp = LIST_FIRST(&lfp->lf_deleg);
2518 while (stp != LIST_END(&lfp->lf_deleg)) {
2519 nstp = LIST_NEXT(stp, ls_file);
2520 if (stp->ls_clp != clp && (stp->ls_flags & NFSLCK_DELEGREAD))
2521 writedeleg = 0;
2522 else
2523 delegate = 0;
2524 if ((readonly && stp->ls_clp != clp &&
2525 (stp->ls_flags & NFSLCK_DELEGWRITE)) ||
2526 (!readonly && (stp->ls_clp != clp ||
2527 (stp->ls_flags & NFSLCK_DELEGREAD)))) {
2528 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2529 delegate = 2;
2530 } else {
2531 ret = nfsrv_delegconflict(stp, &haslock, p, vp);
2532 if (ret) {
2533 /*
2534 * nfsrv_delegconflict() unlocks state
2535 * when it returns non-zero.
2536 */
2537 printf("Nfsd openctrl unexpected deleg cnfl\n");
2538 free((caddr_t)new_open, M_NFSDSTATE);
2539 free((caddr_t)new_deleg, M_NFSDSTATE);
2540 if (ret == -1) {
2541 openstp = NULL;
2542 goto tryagain;
2543 }
2544 error = ret;
2545 goto out;
2546 }
2547 }
2548 }
2549 stp = nstp;
2550 }
2551 }
2552
2553 /*
2554 * We only get here if there was no open that conflicted.
2555 * If an open for the owner exists, or in the access/deny bits.
2556 * Otherwise it is a new open. If the open_owner hasn't been
2557 * confirmed, replace the open with the new one needing confirmation,
2558 * otherwise add the open.
2559 */
2560 if (new_stp->ls_flags & NFSLCK_DELEGPREV) {
2561 /*
2562 * Handle NFSLCK_DELEGPREV by searching the old delegations for
2563 * a match. If found, just move the old delegation to the current
2564 * delegation list and issue open. If not found, return
2565 * NFSERR_EXPIRED.
2566 */
2567 LIST_FOREACH(stp, &clp->lc_olddeleg, ls_list) {
2568 if (stp->ls_lfp == lfp) {
2569 /* Found it */
2570 if (stp->ls_clp != clp)
2571 panic("olddeleg clp");
2572 LIST_REMOVE(stp, ls_list);
2573 LIST_REMOVE(stp, ls_hash);
2574 stp->ls_flags &= ~NFSLCK_OLDDELEG;
2575 stp->ls_stateid.seqid = delegstateidp->seqid = 0;
2576 stp->ls_stateid.other[0] = delegstateidp->other[0] =
2577 clp->lc_clientid.lval[0];
2578 stp->ls_stateid.other[1] = delegstateidp->other[1] =
2579 clp->lc_clientid.lval[1];
2580 stp->ls_stateid.other[2] = delegstateidp->other[2] =
2581 nfsrv_nextstateindex(clp);
2582 stp->ls_compref = nd->nd_compref;
2583 LIST_INSERT_HEAD(&clp->lc_deleg, stp, ls_list);
2584 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2585 stp->ls_stateid), stp, ls_hash);
2586 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2587 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2588 else
2589 *rflagsp |= NFSV4OPEN_READDELEGATE;
2590 clp->lc_delegtime = NFSD_MONOSEC +
2591 nfsrv_lease + NFSRV_LEASEDELTA;
2592
2593 /*
2594 * Now, do the associated open.
2595 */
2596 new_open->ls_stateid.seqid = 0;
2597 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2598 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2599 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2600 new_open->ls_flags = (new_stp->ls_flags&NFSLCK_DENYBITS)|
2601 NFSLCK_OPEN;
2602 if (stp->ls_flags & NFSLCK_DELEGWRITE)
2603 new_open->ls_flags |= (NFSLCK_READACCESS |
2604 NFSLCK_WRITEACCESS);
2605 else
2606 new_open->ls_flags |= NFSLCK_READACCESS;
2607 new_open->ls_uid = new_stp->ls_uid;
2608 new_open->ls_lfp = lfp;
2609 new_open->ls_clp = clp;
2610 LIST_INIT(&new_open->ls_open);
2611 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2612 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2613 new_open, ls_hash);
2614 /*
2615 * and handle the open owner
2616 */
2617 if (ownerstp) {
2618 new_open->ls_openowner = ownerstp;
2619 LIST_INSERT_HEAD(&ownerstp->ls_open,new_open,ls_list);
2620 } else {
2621 new_open->ls_openowner = new_stp;
2622 new_stp->ls_flags = 0;
2623 nfsrvd_refcache(new_stp->ls_op);
2624 new_stp->ls_noopens = 0;
2625 LIST_INIT(&new_stp->ls_open);
2626 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2627 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2628 *new_stpp = NULL;
2629 newnfsstats.srvopenowners++;
2630 nfsrv_openpluslock++;
2631 }
2632 openstp = new_open;
2633 new_open = NULL;
2634 newnfsstats.srvopens++;
2635 nfsrv_openpluslock++;
2636 break;
2637 }
2638 }
2639 if (stp == LIST_END(&clp->lc_olddeleg))
2640 error = NFSERR_EXPIRED;
2641 } else if (new_stp->ls_flags & (NFSLCK_DELEGREAD | NFSLCK_DELEGWRITE)) {
2642 /*
2643 * Scan to see that no delegation for this client and file
2644 * doesn't already exist.
2645 * There also shouldn't yet be an Open for this file and
2646 * openowner.
2647 */
2648 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
2649 if (stp->ls_clp == clp)
2650 break;
2651 }
2652 if (stp == LIST_END(&lfp->lf_deleg) && openstp == NULL) {
2653 /*
2654 * This is the Claim_Previous case with a delegation
2655 * type != Delegate_None.
2656 */
2657 /*
2658 * First, add the delegation. (Although we must issue the
2659 * delegation, we can also ask for an immediate return.)
2660 */
2661 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2662 new_deleg->ls_stateid.other[0] = delegstateidp->other[0] =
2663 clp->lc_clientid.lval[0];
2664 new_deleg->ls_stateid.other[1] = delegstateidp->other[1] =
2665 clp->lc_clientid.lval[1];
2666 new_deleg->ls_stateid.other[2] = delegstateidp->other[2] =
2667 nfsrv_nextstateindex(clp);
2668 if (new_stp->ls_flags & NFSLCK_DELEGWRITE) {
2669 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2670 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2671 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2672 } else {
2673 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2674 NFSLCK_READACCESS);
2675 *rflagsp |= NFSV4OPEN_READDELEGATE;
2676 }
2677 new_deleg->ls_uid = new_stp->ls_uid;
2678 new_deleg->ls_lfp = lfp;
2679 new_deleg->ls_clp = clp;
2680 new_deleg->ls_filerev = filerev;
2681 new_deleg->ls_compref = nd->nd_compref;
2682 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2683 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2684 new_deleg->ls_stateid), new_deleg, ls_hash);
2685 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2686 new_deleg = NULL;
2687 if (delegate == 2 || nfsrv_issuedelegs == 0 ||
2688 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) !=
2689 LCL_CALLBACKSON ||
2690 NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) ||
2691 !NFSVNO_DELEGOK(vp))
2692 *rflagsp |= NFSV4OPEN_RECALL;
2693 newnfsstats.srvdelegates++;
2694 nfsrv_openpluslock++;
2695 nfsrv_delegatecnt++;
2696
2697 /*
2698 * Now, do the associated open.
2699 */
2700 new_open->ls_stateid.seqid = 0;
2701 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2702 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2703 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2704 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_DENYBITS) |
2705 NFSLCK_OPEN;
2706 if (new_stp->ls_flags & NFSLCK_DELEGWRITE)
2707 new_open->ls_flags |= (NFSLCK_READACCESS |
2708 NFSLCK_WRITEACCESS);
2709 else
2710 new_open->ls_flags |= NFSLCK_READACCESS;
2711 new_open->ls_uid = new_stp->ls_uid;
2712 new_open->ls_lfp = lfp;
2713 new_open->ls_clp = clp;
2714 LIST_INIT(&new_open->ls_open);
2715 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2716 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2717 new_open, ls_hash);
2718 /*
2719 * and handle the open owner
2720 */
2721 if (ownerstp) {
2722 new_open->ls_openowner = ownerstp;
2723 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2724 } else {
2725 new_open->ls_openowner = new_stp;
2726 new_stp->ls_flags = 0;
2727 nfsrvd_refcache(new_stp->ls_op);
2728 new_stp->ls_noopens = 0;
2729 LIST_INIT(&new_stp->ls_open);
2730 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2731 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2732 *new_stpp = NULL;
2733 newnfsstats.srvopenowners++;
2734 nfsrv_openpluslock++;
2735 }
2736 openstp = new_open;
2737 new_open = NULL;
2738 newnfsstats.srvopens++;
2739 nfsrv_openpluslock++;
2740 } else {
2741 error = NFSERR_RECLAIMCONFLICT;
2742 }
2743 } else if (ownerstp) {
2744 if (ownerstp->ls_flags & NFSLCK_NEEDSCONFIRM) {
2745 /* Replace the open */
2746 if (ownerstp->ls_op)
2747 nfsrvd_derefcache(ownerstp->ls_op);
2748 ownerstp->ls_op = new_stp->ls_op;
2749 nfsrvd_refcache(ownerstp->ls_op);
2750 ownerstp->ls_seq = new_stp->ls_seq;
2751 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2752 stp = LIST_FIRST(&ownerstp->ls_open);
2753 stp->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2754 NFSLCK_OPEN;
2755 stp->ls_stateid.seqid = 0;
2756 stp->ls_uid = new_stp->ls_uid;
2757 if (lfp != stp->ls_lfp) {
2758 LIST_REMOVE(stp, ls_file);
2759 LIST_INSERT_HEAD(&lfp->lf_open, stp, ls_file);
2760 stp->ls_lfp = lfp;
2761 }
2762 openstp = stp;
2763 } else if (openstp) {
2764 openstp->ls_flags |= (new_stp->ls_flags & NFSLCK_SHAREBITS);
2765 openstp->ls_stateid.seqid++;
2766
2767 /*
2768 * This is where we can choose to issue a delegation.
2769 */
2770 if (delegate && nfsrv_issuedelegs &&
2771 writedeleg && !NFSVNO_EXRDONLY(exp) &&
2772 (nfsrv_writedelegifpos || !readonly) &&
2773 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2774 LCL_CALLBACKSON &&
2775 !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2776 NFSVNO_DELEGOK(vp)) {
2777 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2778 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2779 = clp->lc_clientid.lval[0];
2780 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2781 = clp->lc_clientid.lval[1];
2782 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2783 = nfsrv_nextstateindex(clp);
2784 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2785 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2786 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2787 new_deleg->ls_uid = new_stp->ls_uid;
2788 new_deleg->ls_lfp = lfp;
2789 new_deleg->ls_clp = clp;
2790 new_deleg->ls_filerev = filerev;
2791 new_deleg->ls_compref = nd->nd_compref;
2792 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2793 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2794 new_deleg->ls_stateid), new_deleg, ls_hash);
2795 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2796 new_deleg = NULL;
2797 newnfsstats.srvdelegates++;
2798 nfsrv_openpluslock++;
2799 nfsrv_delegatecnt++;
2800 }
2801 } else {
2802 new_open->ls_stateid.seqid = 0;
2803 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2804 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2805 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2806 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS)|
2807 NFSLCK_OPEN;
2808 new_open->ls_uid = new_stp->ls_uid;
2809 new_open->ls_openowner = ownerstp;
2810 new_open->ls_lfp = lfp;
2811 new_open->ls_clp = clp;
2812 LIST_INIT(&new_open->ls_open);
2813 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2814 LIST_INSERT_HEAD(&ownerstp->ls_open, new_open, ls_list);
2815 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2816 new_open, ls_hash);
2817 openstp = new_open;
2818 new_open = NULL;
2819 newnfsstats.srvopens++;
2820 nfsrv_openpluslock++;
2821
2822 /*
2823 * This is where we can choose to issue a delegation.
2824 */
2825 if (delegate && nfsrv_issuedelegs &&
2826 (writedeleg || readonly) &&
2827 (clp->lc_flags & (LCL_CALLBACKSON | LCL_CBDOWN)) ==
2828 LCL_CALLBACKSON &&
2829 !NFSRV_V4DELEGLIMIT(nfsrv_delegatecnt) &&
2830 NFSVNO_DELEGOK(vp)) {
2831 new_deleg->ls_stateid.seqid = delegstateidp->seqid = 0;
2832 new_deleg->ls_stateid.other[0] = delegstateidp->other[0]
2833 = clp->lc_clientid.lval[0];
2834 new_deleg->ls_stateid.other[1] = delegstateidp->other[1]
2835 = clp->lc_clientid.lval[1];
2836 new_deleg->ls_stateid.other[2] = delegstateidp->other[2]
2837 = nfsrv_nextstateindex(clp);
2838 if (writedeleg && !NFSVNO_EXRDONLY(exp) &&
2839 (nfsrv_writedelegifpos || !readonly)) {
2840 new_deleg->ls_flags = (NFSLCK_DELEGWRITE |
2841 NFSLCK_READACCESS | NFSLCK_WRITEACCESS);
2842 *rflagsp |= NFSV4OPEN_WRITEDELEGATE;
2843 } else {
2844 new_deleg->ls_flags = (NFSLCK_DELEGREAD |
2845 NFSLCK_READACCESS);
2846 *rflagsp |= NFSV4OPEN_READDELEGATE;
2847 }
2848 new_deleg->ls_uid = new_stp->ls_uid;
2849 new_deleg->ls_lfp = lfp;
2850 new_deleg->ls_clp = clp;
2851 new_deleg->ls_filerev = filerev;
2852 new_deleg->ls_compref = nd->nd_compref;
2853 LIST_INSERT_HEAD(&lfp->lf_deleg, new_deleg, ls_file);
2854 LIST_INSERT_HEAD(NFSSTATEHASH(clp,
2855 new_deleg->ls_stateid), new_deleg, ls_hash);
2856 LIST_INSERT_HEAD(&clp->lc_deleg, new_deleg, ls_list);
2857 new_deleg = NULL;
2858 newnfsstats.srvdelegates++;
2859 nfsrv_openpluslock++;
2860 nfsrv_delegatecnt++;
2861 }
2862 }
2863 } else {
2864 /*
2865 * New owner case. Start the open_owner sequence with a
2866 * Needs confirmation (unless a reclaim) and hang the
2867 * new open off it.
2868 */
2869 new_open->ls_stateid.seqid = 0;
2870 new_open->ls_stateid.other[0] = clp->lc_clientid.lval[0];
2871 new_open->ls_stateid.other[1] = clp->lc_clientid.lval[1];
2872 new_open->ls_stateid.other[2] = nfsrv_nextstateindex(clp);
2873 new_open->ls_flags = (new_stp->ls_flags & NFSLCK_SHAREBITS) |
2874 NFSLCK_OPEN;
2875 new_open->ls_uid = new_stp->ls_uid;
2876 LIST_INIT(&new_open->ls_open);
2877 new_open->ls_openowner = new_stp;
2878 new_open->ls_lfp = lfp;
2879 new_open->ls_clp = clp;
2880 LIST_INSERT_HEAD(&lfp->lf_open, new_open, ls_file);
2881 if (new_stp->ls_flags & NFSLCK_RECLAIM) {
2882 new_stp->ls_flags = 0;
2883 } else {
2884 *rflagsp |= NFSV4OPEN_RESULTCONFIRM;
2885 new_stp->ls_flags = NFSLCK_NEEDSCONFIRM;
2886 }
2887 nfsrvd_refcache(new_stp->ls_op);
2888 new_stp->ls_noopens = 0;
2889 LIST_INIT(&new_stp->ls_open);
2890 LIST_INSERT_HEAD(&new_stp->ls_open, new_open, ls_list);
2891 LIST_INSERT_HEAD(&clp->lc_open, new_stp, ls_list);
2892 LIST_INSERT_HEAD(NFSSTATEHASH(clp, new_open->ls_stateid),
2893 new_open, ls_hash);
2894 openstp = new_open;
2895 new_open = NULL;
2896 *new_stpp = NULL;
2897 newnfsstats.srvopens++;
2898 nfsrv_openpluslock++;
2899 newnfsstats.srvopenowners++;
2900 nfsrv_openpluslock++;
2901 }
2902 if (!error) {
2903 stateidp->seqid = openstp->ls_stateid.seqid;
2904 stateidp->other[0] = openstp->ls_stateid.other[0];
2905 stateidp->other[1] = openstp->ls_stateid.other[1];
2906 stateidp->other[2] = openstp->ls_stateid.other[2];
2907 }
2908 NFSUNLOCKSTATE();
2909 if (haslock) {
2910 NFSLOCKV4ROOTMUTEX();
2911 nfsv4_unlock(&nfsv4rootfs_lock, 1);
2912 NFSUNLOCKV4ROOTMUTEX();
2913 }
2914 if (new_open)
2915 FREE((caddr_t)new_open, M_NFSDSTATE);
2916 if (new_deleg)
2917 FREE((caddr_t)new_deleg, M_NFSDSTATE);
2918
2919 out:
2920 NFSEXITCODE2(error, nd);
2921 return (error);
2922 }
2923
2924 /*
2925 * Open update. Does the confirm, downgrade and close.
2926 */
2927 APPLESTATIC int
2928 nfsrv_openupdate(vnode_t vp, struct nfsstate *new_stp, nfsquad_t clientid,
2929 nfsv4stateid_t *stateidp, struct nfsrv_descript *nd, NFSPROC_T *p)
2930 {
2931 struct nfsstate *stp, *ownerstp;
2932 struct nfsclient *clp;
2933 struct nfslockfile *lfp;
2934 u_int32_t bits;
2935 int error = 0, gotstate = 0, len = 0;
2936 u_char client[NFSV4_OPAQUELIMIT];
2937
2938 /*
2939 * Check for restart conditions (client and server).
2940 */
2941 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
2942 &new_stp->ls_stateid, 0);
2943 if (error)
2944 goto out;
2945
2946 NFSLOCKSTATE();
2947 /*
2948 * Get the open structure via clientid and stateid.
2949 */
2950 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
2951 (nfsquad_t)((u_quad_t)0), NULL, p);
2952 if (!error)
2953 error = nfsrv_getstate(clp, &new_stp->ls_stateid,
2954 new_stp->ls_flags, &stp);
2955
2956 /*
2957 * Sanity check the open.
2958 */
2959 if (!error && (!(stp->ls_flags & NFSLCK_OPEN) ||
2960 (!(new_stp->ls_flags & NFSLCK_CONFIRM) &&
2961 (stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)) ||
2962 ((new_stp->ls_flags & NFSLCK_CONFIRM) &&
2963 (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM)))))
2964 error = NFSERR_BADSTATEID;
2965
2966 if (!error)
2967 error = nfsrv_checkseqid(nd, new_stp->ls_seq,
2968 stp->ls_openowner, new_stp->ls_op);
2969 if (!error && stp->ls_stateid.seqid != new_stp->ls_stateid.seqid &&
2970 !(new_stp->ls_flags & NFSLCK_CONFIRM))
2971 error = NFSERR_OLDSTATEID;
2972 if (!error && vnode_vtype(vp) != VREG) {
2973 if (vnode_vtype(vp) == VDIR)
2974 error = NFSERR_ISDIR;
2975 else
2976 error = NFSERR_INVAL;
2977 }
2978
2979 if (error) {
2980 /*
2981 * If a client tries to confirm an Open with a bad
2982 * seqid# and there are no byte range locks or other Opens
2983 * on the openowner, just throw it away, so the next use of the
2984 * openowner will start a fresh seq#.
2985 */
2986 if (error == NFSERR_BADSEQID &&
2987 (new_stp->ls_flags & NFSLCK_CONFIRM) &&
2988 nfsrv_nootherstate(stp))
2989 nfsrv_freeopenowner(stp->ls_openowner, 0, p);
2990 NFSUNLOCKSTATE();
2991 goto out;
2992 }
2993
2994 /*
2995 * Set the return stateid.
2996 */
2997 stateidp->seqid = stp->ls_stateid.seqid + 1;
2998 stateidp->other[0] = stp->ls_stateid.other[0];
2999 stateidp->other[1] = stp->ls_stateid.other[1];
3000 stateidp->other[2] = stp->ls_stateid.other[2];
3001 /*
3002 * Now, handle the three cases.
3003 */
3004 if (new_stp->ls_flags & NFSLCK_CONFIRM) {
3005 /*
3006 * If the open doesn't need confirmation, it seems to me that
3007 * there is a client error, but I'll just log it and keep going?
3008 */
3009 if (!(stp->ls_openowner->ls_flags & NFSLCK_NEEDSCONFIRM))
3010 printf("Nfsv4d: stray open confirm\n");
3011 stp->ls_openowner->ls_flags = 0;
3012 stp->ls_stateid.seqid++;
3013 if (!(clp->lc_flags & LCL_STAMPEDSTABLE)) {
3014 clp->lc_flags |= LCL_STAMPEDSTABLE;
3015 len = clp->lc_idlen;
3016 NFSBCOPY(clp->lc_id, client, len);
3017 gotstate = 1;
3018 }
3019 NFSUNLOCKSTATE();
3020 } else if (new_stp->ls_flags & NFSLCK_CLOSE) {
3021 ownerstp = stp->ls_openowner;
3022 lfp = stp->ls_lfp;
3023 if (nfsrv_dolocallocks != 0 && !LIST_EMPTY(&stp->ls_open)) {
3024 /* Get the lf lock */
3025 nfsrv_locklf(lfp);
3026 NFSUNLOCKSTATE();
3027 ASSERT_VOP_ELOCKED(vp, "nfsrv_openupdate");
3028 NFSVOPUNLOCK(vp, 0);
3029 if (nfsrv_freeopen(stp, vp, 1, p) == 0) {
3030 NFSLOCKSTATE();
3031 nfsrv_unlocklf(lfp);
3032 NFSUNLOCKSTATE();
3033 }
3034 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
3035 } else {
3036 (void) nfsrv_freeopen(stp, NULL, 0, p);
3037 NFSUNLOCKSTATE();
3038 }
3039 } else {
3040 /*
3041 * Update the share bits, making sure that the new set are a
3042 * subset of the old ones.
3043 */
3044 bits = (new_stp->ls_flags & NFSLCK_SHAREBITS);
3045 if (~(stp->ls_flags) & bits) {
3046 NFSUNLOCKSTATE();
3047 error = NFSERR_INVAL;
3048 goto out;
3049 }
3050 stp->ls_flags = (bits | NFSLCK_OPEN);
3051 stp->ls_stateid.seqid++;
3052 NFSUNLOCKSTATE();
3053 }
3054
3055 /*
3056 * If the client just confirmed its first open, write a timestamp
3057 * to the stable storage file.
3058 */
3059 if (gotstate != 0) {
3060 nfsrv_writestable(client, len, NFSNST_NEWSTATE, p);
3061 nfsrv_backupstable();
3062 }
3063
3064 out:
3065 NFSEXITCODE2(error, nd);
3066 return (error);
3067 }
3068
3069 /*
3070 * Delegation update. Does the purge and return.
3071 */
3072 APPLESTATIC int
3073 nfsrv_delegupdate(nfsquad_t clientid, nfsv4stateid_t *stateidp,
3074 vnode_t vp, int op, struct ucred *cred, NFSPROC_T *p)
3075 {
3076 struct nfsstate *stp;
3077 struct nfsclient *clp;
3078 int error = 0;
3079 fhandle_t fh;
3080
3081 /*
3082 * Do a sanity check against the file handle for DelegReturn.
3083 */
3084 if (vp) {
3085 error = nfsvno_getfh(vp, &fh, p);
3086 if (error)
3087 goto out;
3088 }
3089 /*
3090 * Check for restart conditions (client and server).
3091 */
3092 if (op == NFSV4OP_DELEGRETURN)
3093 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGRETURN,
3094 stateidp, 0);
3095 else
3096 error = nfsrv_checkrestart(clientid, NFSLCK_DELEGPURGE,
3097 stateidp, 0);
3098
3099 NFSLOCKSTATE();
3100 /*
3101 * Get the open structure via clientid and stateid.
3102 */
3103 if (!error)
3104 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3105 (nfsquad_t)((u_quad_t)0), NULL, p);
3106 if (error) {
3107 if (error == NFSERR_CBPATHDOWN)
3108 error = 0;
3109 if (error == NFSERR_STALECLIENTID && op == NFSV4OP_DELEGRETURN)
3110 error = NFSERR_STALESTATEID;
3111 }
3112 if (!error && op == NFSV4OP_DELEGRETURN) {
3113 error = nfsrv_getstate(clp, stateidp, NFSLCK_DELEGRETURN, &stp);
3114 if (!error && stp->ls_stateid.seqid != stateidp->seqid)
3115 error = NFSERR_OLDSTATEID;
3116 }
3117 /*
3118 * NFSERR_EXPIRED means that the state has gone away,
3119 * so Delegations have been purged. Just return ok.
3120 */
3121 if (error == NFSERR_EXPIRED && op == NFSV4OP_DELEGPURGE) {
3122 NFSUNLOCKSTATE();
3123 error = 0;
3124 goto out;
3125 }
3126 if (error) {
3127 NFSUNLOCKSTATE();
3128 goto out;
3129 }
3130
3131 if (op == NFSV4OP_DELEGRETURN) {
3132 if (NFSBCMP((caddr_t)&fh, (caddr_t)&stp->ls_lfp->lf_fh,
3133 sizeof (fhandle_t))) {
3134 NFSUNLOCKSTATE();
3135 error = NFSERR_BADSTATEID;
3136 goto out;
3137 }
3138 nfsrv_freedeleg(stp);
3139 } else {
3140 nfsrv_freedeleglist(&clp->lc_olddeleg);
3141 }
3142 NFSUNLOCKSTATE();
3143 error = 0;
3144
3145 out:
3146 NFSEXITCODE(error);
3147 return (error);
3148 }
3149
3150 /*
3151 * Release lock owner.
3152 */
3153 APPLESTATIC int
3154 nfsrv_releaselckown(struct nfsstate *new_stp, nfsquad_t clientid,
3155 NFSPROC_T *p)
3156 {
3157 struct nfsstate *stp, *nstp, *openstp, *ownstp;
3158 struct nfsclient *clp;
3159 int error = 0;
3160
3161 /*
3162 * Check for restart conditions (client and server).
3163 */
3164 error = nfsrv_checkrestart(clientid, new_stp->ls_flags,
3165 &new_stp->ls_stateid, 0);
3166 if (error)
3167 goto out;
3168
3169 NFSLOCKSTATE();
3170 /*
3171 * Get the lock owner by name.
3172 */
3173 error = nfsrv_getclient(clientid, CLOPS_RENEW, &clp,
3174 (nfsquad_t)((u_quad_t)0), NULL, p);
3175 if (error) {
3176 NFSUNLOCKSTATE();
3177 goto out;
3178 }
3179 LIST_FOREACH(ownstp, &clp->lc_open, ls_list) {
3180 LIST_FOREACH(openstp, &ownstp->ls_open, ls_list) {
3181 stp = LIST_FIRST(&openstp->ls_open);
3182 while (stp != LIST_END(&openstp->ls_open)) {
3183 nstp = LIST_NEXT(stp, ls_list);
3184 /*
3185 * If the owner matches, check for locks and
3186 * then free or return an error.
3187 */
3188 if (stp->ls_ownerlen == new_stp->ls_ownerlen &&
3189 !NFSBCMP(stp->ls_owner, new_stp->ls_owner,
3190 stp->ls_ownerlen)){
3191 if (LIST_EMPTY(&stp->ls_lock)) {
3192 nfsrv_freelockowner(stp, NULL, 0, p);
3193 } else {
3194 NFSUNLOCKSTATE();
3195 error = NFSERR_LOCKSHELD;
3196 goto out;
3197 }
3198 }
3199 stp = nstp;
3200 }
3201 }
3202 }
3203 NFSUNLOCKSTATE();
3204
3205 out:
3206 NFSEXITCODE(error);
3207 return (error);
3208 }
3209
3210 /*
3211 * Get the file handle for a lock structure.
3212 */
3213 static int
3214 nfsrv_getlockfh(vnode_t vp, u_short flags, struct nfslockfile *new_lfp,
3215 fhandle_t *nfhp, NFSPROC_T *p)
3216 {
3217 fhandle_t *fhp = NULL;
3218 int error;
3219
3220 /*
3221 * For lock, use the new nfslock structure, otherwise just
3222 * a fhandle_t on the stack.
3223 */
3224 if (flags & NFSLCK_OPEN) {
3225 KASSERT(new_lfp != NULL, ("nfsrv_getlockfh: new_lfp NULL"));
3226 fhp = &new_lfp->lf_fh;
3227 } else if (nfhp) {
3228 fhp = nfhp;
3229 } else {
3230 panic("nfsrv_getlockfh");
3231 }
3232 error = nfsvno_getfh(vp, fhp, p);
3233 NFSEXITCODE(error);
3234 return (error);
3235 }
3236
3237 /*
3238 * Get an nfs lock structure. Allocate one, as required, and return a
3239 * pointer to it.
3240 * Returns an NFSERR_xxx upon failure or -1 to indicate no current lock.
3241 */
3242 static int
3243 nfsrv_getlockfile(u_short flags, struct nfslockfile **new_lfpp,
3244 struct nfslockfile **lfpp, fhandle_t *nfhp, int lockit)
3245 {
3246 struct nfslockfile *lfp;
3247 fhandle_t *fhp = NULL, *tfhp;
3248 struct nfslockhashhead *hp;
3249 struct nfslockfile *new_lfp = NULL;
3250
3251 /*
3252 * For lock, use the new nfslock structure, otherwise just
3253 * a fhandle_t on the stack.
3254 */
3255 if (flags & NFSLCK_OPEN) {
3256 new_lfp = *new_lfpp;
3257 fhp = &new_lfp->lf_fh;
3258 } else if (nfhp) {
3259 fhp = nfhp;
3260 } else {
3261 panic("nfsrv_getlockfile");
3262 }
3263
3264 hp = NFSLOCKHASH(fhp);
3265 LIST_FOREACH(lfp, hp, lf_hash) {
3266 tfhp = &lfp->lf_fh;
3267 if (NFSVNO_CMPFH(fhp, tfhp)) {
3268 if (lockit)
3269 nfsrv_locklf(lfp);
3270 *lfpp = lfp;
3271 return (0);
3272 }
3273 }
3274 if (!(flags & NFSLCK_OPEN))
3275 return (-1);
3276
3277 /*
3278 * No match, so chain the new one into the list.
3279 */
3280 LIST_INIT(&new_lfp->lf_open);
3281 LIST_INIT(&new_lfp->lf_lock);
3282 LIST_INIT(&new_lfp->lf_deleg);
3283 LIST_INIT(&new_lfp->lf_locallock);
3284 LIST_INIT(&new_lfp->lf_rollback);
3285 new_lfp->lf_locallock_lck.nfslock_usecnt = 0;
3286 new_lfp->lf_locallock_lck.nfslock_lock = 0;
3287 new_lfp->lf_usecount = 0;
3288 LIST_INSERT_HEAD(hp, new_lfp, lf_hash);
3289 *lfpp = new_lfp;
3290 *new_lfpp = NULL;
3291 return (0);
3292 }
3293
3294 /*
3295 * This function adds a nfslock lock structure to the list for the associated
3296 * nfsstate and nfslockfile structures. It will be inserted after the
3297 * entry pointed at by insert_lop.
3298 */
3299 static void
3300 nfsrv_insertlock(struct nfslock *new_lop, struct nfslock *insert_lop,
3301 struct nfsstate *stp, struct nfslockfile *lfp)
3302 {
3303 struct nfslock *lop, *nlop;
3304
3305 new_lop->lo_stp = stp;
3306 new_lop->lo_lfp = lfp;
3307
3308 if (stp != NULL) {
3309 /* Insert in increasing lo_first order */
3310 lop = LIST_FIRST(&lfp->lf_lock);
3311 if (lop == LIST_END(&lfp->lf_lock) ||
3312 new_lop->lo_first <= lop->lo_first) {
3313 LIST_INSERT_HEAD(&lfp->lf_lock, new_lop, lo_lckfile);
3314 } else {
3315 nlop = LIST_NEXT(lop, lo_lckfile);
3316 while (nlop != LIST_END(&lfp->lf_lock) &&
3317 nlop->lo_first < new_lop->lo_first) {
3318 lop = nlop;
3319 nlop = LIST_NEXT(lop, lo_lckfile);
3320 }
3321 LIST_INSERT_AFTER(lop, new_lop, lo_lckfile);
3322 }
3323 } else {
3324 new_lop->lo_lckfile.le_prev = NULL; /* list not used */
3325 }
3326
3327 /*
3328 * Insert after insert_lop, which is overloaded as stp or lfp for
3329 * an empty list.
3330 */
3331 if (stp == NULL && (struct nfslockfile *)insert_lop == lfp)
3332 LIST_INSERT_HEAD(&lfp->lf_locallock, new_lop, lo_lckowner);
3333 else if ((struct nfsstate *)insert_lop == stp)
3334 LIST_INSERT_HEAD(&stp->ls_lock, new_lop, lo_lckowner);
3335 else
3336 LIST_INSERT_AFTER(insert_lop, new_lop, lo_lckowner);
3337 if (stp != NULL) {
3338 newnfsstats.srvlocks++;
3339 nfsrv_openpluslock++;
3340 }
3341 }
3342
3343 /*
3344 * This function updates the locking for a lock owner and given file. It
3345 * maintains a list of lock ranges ordered on increasing file offset that
3346 * are NFSLCK_READ or NFSLCK_WRITE and non-overlapping (aka POSIX style).
3347 * It always adds new_lop to the list and sometimes uses the one pointed
3348 * at by other_lopp.
3349 */
3350 static void
3351 nfsrv_updatelock(struct nfsstate *stp, struct nfslock **new_lopp,
3352 struct nfslock **other_lopp, struct nfslockfile *lfp)
3353 {
3354 struct nfslock *new_lop = *new_lopp;
3355 struct nfslock *lop, *tlop, *ilop;
3356 struct nfslock *other_lop = *other_lopp;
3357 int unlock = 0, myfile = 0;
3358 u_int64_t tmp;
3359
3360 /*
3361 * Work down the list until the lock is merged.
3362 */
3363 if (new_lop->lo_flags & NFSLCK_UNLOCK)
3364 unlock = 1;
3365 if (stp != NULL) {
3366 ilop = (struct nfslock *)stp;
3367 lop = LIST_FIRST(&stp->ls_lock);
3368 } else {
3369 ilop = (struct nfslock *)lfp;
3370 lop = LIST_FIRST(&lfp->lf_locallock);
3371 }
3372 while (lop != NULL) {
3373 /*
3374 * Only check locks for this file that aren't before the start of
3375 * new lock's range.
3376 */
3377 if (lop->lo_lfp == lfp) {
3378 myfile = 1;
3379 if (lop->lo_end >= new_lop->lo_first) {
3380 if (new_lop->lo_end < lop->lo_first) {
3381 /*
3382 * If the new lock ends before the start of the
3383 * current lock's range, no merge, just insert
3384 * the new lock.
3385 */
3386 break;
3387 }
3388 if (new_lop->lo_flags == lop->lo_flags ||
3389 (new_lop->lo_first <= lop->lo_first &&
3390 new_lop->lo_end >= lop->lo_end)) {
3391 /*
3392 * This lock can be absorbed by the new lock/unlock.
3393 * This happens when it covers the entire range
3394 * of the old lock or is contiguous
3395 * with the old lock and is of the same type or an
3396 * unlock.
3397 */
3398 if (lop->lo_first < new_lop->lo_first)
3399 new_lop->lo_first = lop->lo_first;
3400 if (lop->lo_end > new_lop->lo_end)
3401 new_lop->lo_end = lop->lo_end;
3402 tlop = lop;
3403 lop = LIST_NEXT(lop, lo_lckowner);
3404 nfsrv_freenfslock(tlop);
3405 continue;
3406 }
3407
3408 /*
3409 * All these cases are for contiguous locks that are not the
3410 * same type, so they can't be merged.
3411 */
3412 if (new_lop->lo_first <= lop->lo_first) {
3413 /*
3414 * This case is where the new lock overlaps with the
3415 * first part of the old lock. Move the start of the
3416 * old lock to just past the end of the new lock. The
3417 * new lock will be inserted in front of the old, since
3418 * ilop hasn't been updated. (We are done now.)
3419 */
3420 lop->lo_first = new_lop->lo_end;
3421 break;
3422 }
3423 if (new_lop->lo_end >= lop->lo_end) {
3424 /*
3425 * This case is where the new lock overlaps with the
3426 * end of the old lock's range. Move the old lock's
3427 * end to just before the new lock's first and insert
3428 * the new lock after the old lock.
3429 * Might not be done yet, since the new lock could
3430 * overlap further locks with higher ranges.
3431 */
3432 lop->lo_end = new_lop->lo_first;
3433 ilop = lop;
3434 lop = LIST_NEXT(lop, lo_lckowner);
3435 continue;
3436 }
3437 /*
3438 * The final case is where the new lock's range is in the
3439 * middle of the current lock's and splits the current lock
3440 * up. Use *other_lopp to handle the second part of the
3441 * split old lock range. (We are done now.)
3442 * For unlock, we use new_lop as other_lop and tmp, since
3443 * other_lop and new_lop are the same for this case.
3444 * We noted the unlock case above, so we don't need
3445 * new_lop->lo_flags any longer.
3446 */
3447 tmp = new_lop->lo_first;
3448 if (other_lop == NULL) {
3449 if (!unlock)
3450 panic("nfsd srv update unlock");
3451 other_lop = new_lop;
3452 *new_lopp = NULL;
3453 }
3454 other_lop->lo_first = new_lop->lo_end;
3455 other_lop->lo_end = lop->lo_end;
3456 other_lop->lo_flags = lop->lo_flags;
3457 other_lop->lo_stp = stp;
3458 other_lop->lo_lfp = lfp;
3459 lop->lo_end = tmp;
3460 nfsrv_insertlock(other_lop, lop, stp, lfp);
3461 *other_lopp = NULL;
3462 ilop = lop;
3463 break;
3464 }
3465 }
3466 ilop = lop;
3467 lop = LIST_NEXT(lop, lo_lckowner);
3468 if (myfile && (lop == NULL || lop->lo_lfp != lfp))
3469 break;
3470 }
3471
3472 /*
3473 * Insert the new lock in the list at the appropriate place.
3474 */
3475 if (!unlock) {
3476 nfsrv_insertlock(new_lop, ilop, stp, lfp);
3477 *new_lopp = NULL;
3478 }
3479 }
3480
3481 /*
3482 * This function handles sequencing of locks, etc.
3483 * It returns an error that indicates what the caller should do.
3484 */
3485 static int
3486 nfsrv_checkseqid(struct nfsrv_descript *nd, u_int32_t seqid,
3487 struct nfsstate *stp, struct nfsrvcache *op)
3488 {
3489 int error = 0;
3490
3491 if (op != nd->nd_rp)
3492 panic("nfsrvstate checkseqid");
3493 if (!(op->rc_flag & RC_INPROG))
3494 panic("nfsrvstate not inprog");
3495 if (stp->ls_op && stp->ls_op->rc_refcnt <= 0) {
3496 printf("refcnt=%d\n", stp->ls_op->rc_refcnt);
3497 panic("nfsrvstate op refcnt");
3498 }
3499 if ((stp->ls_seq + 1) == seqid) {
3500 if (stp->ls_op)
3501 nfsrvd_derefcache(stp->ls_op);
3502 stp->ls_op = op;
3503 nfsrvd_refcache(op);
3504 stp->ls_seq = seqid;
3505 goto out;
3506 } else if (stp->ls_seq == seqid && stp->ls_op &&
3507 op->rc_xid == stp->ls_op->rc_xid &&
3508 op->rc_refcnt == 0 &&
3509 op->rc_reqlen == stp->ls_op->rc_reqlen &&
3510 op->rc_cksum == stp->ls_op->rc_cksum) {
3511 if (stp->ls_op->rc_flag & RC_INPROG) {
3512 error = NFSERR_DONTREPLY;
3513 goto out;
3514 }
3515 nd->nd_rp = stp->ls_op;
3516 nd->nd_rp->rc_flag |= RC_INPROG;
3517 nfsrvd_delcache(op);
3518 error = NFSERR_REPLYFROMCACHE;
3519 goto out;
3520 }
3521 error = NFSERR_BADSEQID;
3522
3523 out:
3524 NFSEXITCODE2(error, nd);
3525 return (error);
3526 }
3527
3528 /*
3529 * Get the client ip address for callbacks. If the strings can't be parsed,
3530 * just set lc_program to 0 to indicate no callbacks are possible.
3531 * (For cases where the address can't be parsed or is 0.0.0.0.0.0, set
3532 * the address to the client's transport address. This won't be used
3533 * for callbacks, but can be printed out by newnfsstats for info.)
3534 * Return error if the xdr can't be parsed, 0 otherwise.
3535 */
3536 APPLESTATIC int
3537 nfsrv_getclientipaddr(struct nfsrv_descript *nd, struct nfsclient *clp)
3538 {
3539 u_int32_t *tl;
3540 u_char *cp, *cp2;
3541 int i, j;
3542 struct sockaddr_in *rad, *sad;
3543 u_char protocol[5], addr[24];
3544 int error = 0, cantparse = 0;
3545 union {
3546 u_long ival;
3547 u_char cval[4];
3548 } ip;
3549 union {
3550 u_short sval;
3551 u_char cval[2];
3552 } port;
3553
3554 rad = NFSSOCKADDR(clp->lc_req.nr_nam, struct sockaddr_in *);
3555 rad->sin_family = AF_INET;
3556 rad->sin_len = sizeof (struct sockaddr_in);
3557 rad->sin_addr.s_addr = 0;
3558 rad->sin_port = 0;
3559 clp->lc_req.nr_client = NULL;
3560 clp->lc_req.nr_lock = 0;
3561 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3562 i = fxdr_unsigned(int, *tl);
3563 if (i >= 3 && i <= 4) {
3564 error = nfsrv_mtostr(nd, protocol, i);
3565 if (error)
3566 goto nfsmout;
3567 if (!strcmp(protocol, "tcp")) {
3568 clp->lc_flags |= LCL_TCPCALLBACK;
3569 clp->lc_req.nr_sotype = SOCK_STREAM;
3570 clp->lc_req.nr_soproto = IPPROTO_TCP;
3571 } else if (!strcmp(protocol, "udp")) {
3572 clp->lc_req.nr_sotype = SOCK_DGRAM;
3573 clp->lc_req.nr_soproto = IPPROTO_UDP;
3574 } else {
3575 cantparse = 1;
3576 }
3577 } else {
3578 cantparse = 1;
3579 if (i > 0) {
3580 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3581 if (error)
3582 goto nfsmout;
3583 }
3584 }
3585 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
3586 i = fxdr_unsigned(int, *tl);
3587 if (i < 0) {
3588 error = NFSERR_BADXDR;
3589 goto nfsmout;
3590 } else if (i == 0) {
3591 cantparse = 1;
3592 } else if (!cantparse && i <= 23 && i >= 11) {
3593 error = nfsrv_mtostr(nd, addr, i);
3594 if (error)
3595 goto nfsmout;
3596
3597 /*
3598 * Parse out the address fields. We expect 6 decimal numbers
3599 * separated by '.'s.
3600 */
3601 cp = addr;
3602 i = 0;
3603 while (*cp && i < 6) {
3604 cp2 = cp;
3605 while (*cp2 && *cp2 != '.')
3606 cp2++;
3607 if (*cp2)
3608 *cp2++ = '\0';
3609 else if (i != 5) {
3610 cantparse = 1;
3611 break;
3612 }
3613 j = nfsrv_getipnumber(cp);
3614 if (j >= 0) {
3615 if (i < 4)
3616 ip.cval[3 - i] = j;
3617 else
3618 port.cval[5 - i] = j;
3619 } else {
3620 cantparse = 1;
3621 break;
3622 }
3623 cp = cp2;
3624 i++;
3625 }
3626 if (!cantparse) {
3627 if (ip.ival != 0x0) {
3628 rad->sin_addr.s_addr = htonl(ip.ival);
3629 rad->sin_port = htons(port.sval);
3630 } else {
3631 cantparse = 1;
3632 }
3633 }
3634 } else {
3635 cantparse = 1;
3636 if (i > 0) {
3637 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
3638 if (error)
3639 goto nfsmout;
3640 }
3641 }
3642 if (cantparse) {
3643 sad = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3644 rad->sin_addr.s_addr = sad->sin_addr.s_addr;
3645 rad->sin_port = 0x0;
3646 clp->lc_program = 0;
3647 }
3648 nfsmout:
3649 NFSEXITCODE2(error, nd);
3650 return (error);
3651 }
3652
3653 /*
3654 * Turn a string of up to three decimal digits into a number. Return -1 upon
3655 * error.
3656 */
3657 static int
3658 nfsrv_getipnumber(u_char *cp)
3659 {
3660 int i = 0, j = 0;
3661
3662 while (*cp) {
3663 if (j > 2 || *cp < '' || *cp > '9')
3664 return (-1);
3665 i *= 10;
3666 i += (*cp - '');
3667 cp++;
3668 j++;
3669 }
3670 if (i < 256)
3671 return (i);
3672 return (-1);
3673 }
3674
3675 /*
3676 * This function checks for restart conditions.
3677 */
3678 static int
3679 nfsrv_checkrestart(nfsquad_t clientid, u_int32_t flags,
3680 nfsv4stateid_t *stateidp, int specialid)
3681 {
3682 int ret = 0;
3683
3684 /*
3685 * First check for a server restart. Open, LockT, ReleaseLockOwner
3686 * and DelegPurge have a clientid, the rest a stateid.
3687 */
3688 if (flags &
3689 (NFSLCK_OPEN | NFSLCK_TEST | NFSLCK_RELEASE | NFSLCK_DELEGPURGE)) {
3690 if (clientid.lval[0] != nfsrvboottime) {
3691 ret = NFSERR_STALECLIENTID;
3692 goto out;
3693 }
3694 } else if (stateidp->other[0] != nfsrvboottime &&
3695 specialid == 0) {
3696 ret = NFSERR_STALESTATEID;
3697 goto out;
3698 }
3699
3700 /*
3701 * Read, Write, Setattr and LockT can return NFSERR_GRACE and do
3702 * not use a lock/open owner seqid#, so the check can be done now.
3703 * (The others will be checked, as required, later.)
3704 */
3705 if (!(flags & (NFSLCK_CHECK | NFSLCK_TEST)))
3706 goto out;
3707
3708 NFSLOCKSTATE();
3709 ret = nfsrv_checkgrace(flags);
3710 NFSUNLOCKSTATE();
3711
3712 out:
3713 NFSEXITCODE(ret);
3714 return (ret);
3715 }
3716
3717 /*
3718 * Check for grace.
3719 */
3720 static int
3721 nfsrv_checkgrace(u_int32_t flags)
3722 {
3723 int error = 0;
3724
3725 if (nfsrv_stablefirst.nsf_flags & NFSNSF_GRACEOVER) {
3726 if (flags & NFSLCK_RECLAIM) {
3727 error = NFSERR_NOGRACE;
3728 goto out;
3729 }
3730 } else {
3731 if (!(flags & NFSLCK_RECLAIM)) {
3732 error = NFSERR_GRACE;
3733 goto out;
3734 }
3735
3736 /*
3737 * If grace is almost over and we are still getting Reclaims,
3738 * extend grace a bit.
3739 */
3740 if ((NFSD_MONOSEC + NFSRV_LEASEDELTA) >
3741 nfsrv_stablefirst.nsf_eograce)
3742 nfsrv_stablefirst.nsf_eograce = NFSD_MONOSEC +
3743 NFSRV_LEASEDELTA;
3744 }
3745
3746 out:
3747 NFSEXITCODE(error);
3748 return (error);
3749 }
3750
3751 /*
3752 * Do a server callback.
3753 */
3754 static int
3755 nfsrv_docallback(struct nfsclient *clp, int procnum,
3756 nfsv4stateid_t *stateidp, int trunc, fhandle_t *fhp,
3757 struct nfsvattr *nap, nfsattrbit_t *attrbitp, NFSPROC_T *p)
3758 {
3759 mbuf_t m;
3760 u_int32_t *tl;
3761 struct nfsrv_descript nfsd, *nd = &nfsd;
3762 struct ucred *cred;
3763 int error = 0;
3764 u_int32_t callback;
3765
3766 cred = newnfs_getcred();
3767 NFSLOCKSTATE(); /* mostly for lc_cbref++ */
3768 if (clp->lc_flags & LCL_NEEDSCONFIRM) {
3769 NFSUNLOCKSTATE();
3770 panic("docallb");
3771 }
3772 clp->lc_cbref++;
3773
3774 /*
3775 * Fill the callback program# and version into the request
3776 * structure for newnfs_connect() to use.
3777 */
3778 clp->lc_req.nr_prog = clp->lc_program;
3779 clp->lc_req.nr_vers = NFSV4_CBVERS;
3780
3781 /*
3782 * First, fill in some of the fields of nd and cr.
3783 */
3784 nd->nd_flag = ND_NFSV4;
3785 if (clp->lc_flags & LCL_GSS)
3786 nd->nd_flag |= ND_KERBV;
3787 nd->nd_repstat = 0;
3788 cred->cr_uid = clp->lc_uid;
3789 cred->cr_gid = clp->lc_gid;
3790 callback = clp->lc_callback;
3791 NFSUNLOCKSTATE();
3792 cred->cr_ngroups = 1;
3793
3794 /*
3795 * Get the first mbuf for the request.
3796 */
3797 MGET(m, M_WAIT, MT_DATA);
3798 mbuf_setlen(m, 0);
3799 nd->nd_mreq = nd->nd_mb = m;
3800 nd->nd_bpos = NFSMTOD(m, caddr_t);
3801
3802 /*
3803 * and build the callback request.
3804 */
3805 if (procnum == NFSV4OP_CBGETATTR) {
3806 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3807 (void) nfsm_strtom(nd, "CB Getattr", 10);
3808 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
3809 *tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3810 *tl++ = txdr_unsigned(callback);
3811 *tl++ = txdr_unsigned(1);
3812 *tl = txdr_unsigned(NFSV4OP_CBGETATTR);
3813 (void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3814 (void) nfsrv_putattrbit(nd, attrbitp);
3815 } else if (procnum == NFSV4OP_CBRECALL) {
3816 nd->nd_procnum = NFSV4PROC_CBCOMPOUND;
3817 (void) nfsm_strtom(nd, "CB Recall", 9);
3818 NFSM_BUILD(tl, u_int32_t *, 5 * NFSX_UNSIGNED + NFSX_STATEID);
3819 *tl++ = txdr_unsigned(NFSV4_MINORVERSION);
3820 *tl++ = txdr_unsigned(callback);
3821 *tl++ = txdr_unsigned(1);
3822 *tl++ = txdr_unsigned(NFSV4OP_CBRECALL);
3823 *tl++ = txdr_unsigned(stateidp->seqid);
3824 NFSBCOPY((caddr_t)stateidp->other, (caddr_t)tl,
3825 NFSX_STATEIDOTHER);
3826 tl += (NFSX_STATEIDOTHER / NFSX_UNSIGNED);
3827 if (trunc)
3828 *tl = newnfs_true;
3829 else
3830 *tl = newnfs_false;
3831 (void) nfsm_fhtom(nd, (u_int8_t *)fhp, NFSX_MYFH, 0);
3832 } else {
3833 nd->nd_procnum = NFSV4PROC_CBNULL;
3834 }
3835
3836 /*
3837 * Call newnfs_connect(), as required, and then newnfs_request().
3838 */
3839 (void) newnfs_sndlock(&clp->lc_req.nr_lock);
3840 if (clp->lc_req.nr_client == NULL) {
3841 if (nd->nd_procnum == NFSV4PROC_CBNULL)
3842 error = newnfs_connect(NULL, &clp->lc_req, cred,
3843 NULL, 1);
3844 else
3845 error = newnfs_connect(NULL, &clp->lc_req, cred,
3846 NULL, 3);
3847 }
3848 newnfs_sndunlock(&clp->lc_req.nr_lock);
3849 if (!error) {
3850 error = newnfs_request(nd, NULL, clp, &clp->lc_req, NULL,
3851 NULL, cred, clp->lc_program, NFSV4_CBVERS, NULL, 1, NULL);
3852 }
3853 NFSFREECRED(cred);
3854
3855 /*
3856 * If error is set here, the Callback path isn't working
3857 * properly, so twiddle the appropriate LCL_ flags.
3858 * (nd_repstat != 0 indicates the Callback path is working,
3859 * but the callback failed on the client.)
3860 */
3861 if (error) {
3862 /*
3863 * Mark the callback pathway down, which disabled issuing
3864 * of delegations and gets Renew to return NFSERR_CBPATHDOWN.
3865 */
3866 NFSLOCKSTATE();
3867 clp->lc_flags |= LCL_CBDOWN;
3868 NFSUNLOCKSTATE();
3869 } else {
3870 /*
3871 * Callback worked. If the callback path was down, disable
3872 * callbacks, so no more delegations will be issued. (This
3873 * is done on the assumption that the callback pathway is
3874 * flakey.)
3875 */
3876 NFSLOCKSTATE();
3877 if (clp->lc_flags & LCL_CBDOWN)
3878 clp->lc_flags &= ~(LCL_CBDOWN | LCL_CALLBACKSON);
3879 NFSUNLOCKSTATE();
3880 if (nd->nd_repstat)
3881 error = nd->nd_repstat;
3882 else if (procnum == NFSV4OP_CBGETATTR)
3883 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
3884 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
3885 p, NULL);
3886 mbuf_freem(nd->nd_mrep);
3887 }
3888 NFSLOCKSTATE();
3889 clp->lc_cbref--;
3890 if ((clp->lc_flags & LCL_WAKEUPWANTED) && clp->lc_cbref == 0) {
3891 clp->lc_flags &= ~LCL_WAKEUPWANTED;
3892 wakeup(clp);
3893 }
3894 NFSUNLOCKSTATE();
3895
3896 NFSEXITCODE(error);
3897 return (error);
3898 }
3899
3900 /*
3901 * Return the next index# for a clientid. Mostly just increment and return
3902 * the next one, but... if the 32bit unsigned does actually wrap around,
3903 * it should be rebooted.
3904 * At an average rate of one new client per second, it will wrap around in
3905 * approximately 136 years. (I think the server will have been shut
3906 * down or rebooted before then.)
3907 */
3908 static u_int32_t
3909 nfsrv_nextclientindex(void)
3910 {
3911 static u_int32_t client_index = 0;
3912
3913 client_index++;
3914 if (client_index != 0)
3915 return (client_index);
3916
3917 printf("%s: out of clientids\n", __func__);
3918 return (client_index);
3919 }
3920
3921 /*
3922 * Return the next index# for a stateid. Mostly just increment and return
3923 * the next one, but... if the 32bit unsigned does actually wrap around
3924 * (will a BSD server stay up that long?), find
3925 * new start and end values.
3926 */
3927 static u_int32_t
3928 nfsrv_nextstateindex(struct nfsclient *clp)
3929 {
3930 struct nfsstate *stp;
3931 int i;
3932 u_int32_t canuse, min_index, max_index;
3933
3934 if (!(clp->lc_flags & LCL_INDEXNOTOK)) {
3935 clp->lc_stateindex++;
3936 if (clp->lc_stateindex != clp->lc_statemaxindex)
3937 return (clp->lc_stateindex);
3938 }
3939
3940 /*
3941 * Yuck, we've hit the end.
3942 * Look for a new min and max.
3943 */
3944 min_index = 0;
3945 max_index = 0xffffffff;
3946 for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3947 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3948 if (stp->ls_stateid.other[2] > 0x80000000) {
3949 if (stp->ls_stateid.other[2] < max_index)
3950 max_index = stp->ls_stateid.other[2];
3951 } else {
3952 if (stp->ls_stateid.other[2] > min_index)
3953 min_index = stp->ls_stateid.other[2];
3954 }
3955 }
3956 }
3957
3958 /*
3959 * Yikes, highly unlikely, but I'll handle it anyhow.
3960 */
3961 if (min_index == 0x80000000 && max_index == 0x80000001) {
3962 canuse = 0;
3963 /*
3964 * Loop around until we find an unused entry. Return that
3965 * and set LCL_INDEXNOTOK, so the search will continue next time.
3966 * (This is one of those rare cases where a goto is the
3967 * cleanest way to code the loop.)
3968 */
3969 tryagain:
3970 for (i = 0; i < NFSSTATEHASHSIZE; i++) {
3971 LIST_FOREACH(stp, &clp->lc_stateid[i], ls_hash) {
3972 if (stp->ls_stateid.other[2] == canuse) {
3973 canuse++;
3974 goto tryagain;
3975 }
3976 }
3977 }
3978 clp->lc_flags |= LCL_INDEXNOTOK;
3979 return (canuse);
3980 }
3981
3982 /*
3983 * Ok to start again from min + 1.
3984 */
3985 clp->lc_stateindex = min_index + 1;
3986 clp->lc_statemaxindex = max_index;
3987 clp->lc_flags &= ~LCL_INDEXNOTOK;
3988 return (clp->lc_stateindex);
3989 }
3990
3991 /*
3992 * The following functions handle the stable storage file that deals with
3993 * the edge conditions described in RFC3530 Sec. 8.6.3.
3994 * The file is as follows:
3995 * - a single record at the beginning that has the lease time of the
3996 * previous server instance (before the last reboot) and the nfsrvboottime
3997 * values for the previous server boots.
3998 * These previous boot times are used to ensure that the current
3999 * nfsrvboottime does not, somehow, get set to a previous one.
4000 * (This is important so that Stale ClientIDs and StateIDs can
4001 * be recognized.)
4002 * The number of previous nfsvrboottime values preceeds the list.
4003 * - followed by some number of appended records with:
4004 * - client id string
4005 * - flag that indicates it is a record revoking state via lease
4006 * expiration or similar
4007 * OR has successfully acquired state.
4008 * These structures vary in length, with the client string at the end, up
4009 * to NFSV4_OPAQUELIMIT in size.
4010 *
4011 * At the end of the grace period, the file is truncated, the first
4012 * record is rewritten with updated information and any acquired state
4013 * records for successful reclaims of state are written.
4014 *
4015 * Subsequent records are appended when the first state is issued to
4016 * a client and when state is revoked for a client.
4017 *
4018 * When reading the file in, state issued records that come later in
4019 * the file override older ones, since the append log is in cronological order.
4020 * If, for some reason, the file can't be read, the grace period is
4021 * immediately terminated and all reclaims get NFSERR_NOGRACE.
4022 */
4023
4024 /*
4025 * Read in the stable storage file. Called by nfssvc() before the nfsd
4026 * processes start servicing requests.
4027 */
4028 APPLESTATIC void
4029 nfsrv_setupstable(NFSPROC_T *p)
4030 {
4031 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4032 struct nfsrv_stable *sp, *nsp;
4033 struct nfst_rec *tsp;
4034 int error, i, tryagain;
4035 off_t off = 0;
4036 ssize_t aresid, len;
4037
4038 /*
4039 * If NFSNSF_UPDATEDONE is set, this is a restart of the nfsds without
4040 * a reboot, so state has not been lost.
4041 */
4042 if (sf->nsf_flags & NFSNSF_UPDATEDONE)
4043 return;
4044 /*
4045 * Set Grace over just until the file reads successfully.
4046 */
4047 nfsrvboottime = time_second;
4048 LIST_INIT(&sf->nsf_head);
4049 sf->nsf_flags = (NFSNSF_GRACEOVER | NFSNSF_NEEDLOCK);
4050 sf->nsf_eograce = NFSD_MONOSEC + NFSRV_LEASEDELTA;
4051 if (sf->nsf_fp == NULL)
4052 return;
4053 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4054 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), off, UIO_SYSSPACE,
4055 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4056 if (error || aresid || sf->nsf_numboots == 0 ||
4057 sf->nsf_numboots > NFSNSF_MAXNUMBOOTS)
4058 return;
4059
4060 /*
4061 * Now, read in the boottimes.
4062 */
4063 sf->nsf_bootvals = (time_t *)malloc((sf->nsf_numboots + 1) *
4064 sizeof (time_t), M_TEMP, M_WAITOK);
4065 off = sizeof (struct nfsf_rec);
4066 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4067 (caddr_t)sf->nsf_bootvals, sf->nsf_numboots * sizeof (time_t), off,
4068 UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4069 if (error || aresid) {
4070 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4071 sf->nsf_bootvals = NULL;
4072 return;
4073 }
4074
4075 /*
4076 * Make sure this nfsrvboottime is different from all recorded
4077 * previous ones.
4078 */
4079 do {
4080 tryagain = 0;
4081 for (i = 0; i < sf->nsf_numboots; i++) {
4082 if (nfsrvboottime == sf->nsf_bootvals[i]) {
4083 nfsrvboottime++;
4084 tryagain = 1;
4085 break;
4086 }
4087 }
4088 } while (tryagain);
4089
4090 sf->nsf_flags |= NFSNSF_OK;
4091 off += (sf->nsf_numboots * sizeof (time_t));
4092
4093 /*
4094 * Read through the file, building a list of records for grace
4095 * checking.
4096 * Each record is between sizeof (struct nfst_rec) and
4097 * sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1
4098 * and is actually sizeof (struct nfst_rec) + nst_len - 1.
4099 */
4100 tsp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4101 NFSV4_OPAQUELIMIT - 1, M_TEMP, M_WAITOK);
4102 do {
4103 error = NFSD_RDWR(UIO_READ, NFSFPVNODE(sf->nsf_fp),
4104 (caddr_t)tsp, sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1,
4105 off, UIO_SYSSPACE, 0, NFSFPCRED(sf->nsf_fp), &aresid, p);
4106 len = (sizeof (struct nfst_rec) + NFSV4_OPAQUELIMIT - 1) - aresid;
4107 if (error || (len > 0 && (len < sizeof (struct nfst_rec) ||
4108 len < (sizeof (struct nfst_rec) + tsp->len - 1)))) {
4109 /*
4110 * Yuck, the file has been corrupted, so just return
4111 * after clearing out any restart state, so the grace period
4112 * is over.
4113 */
4114 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4115 LIST_REMOVE(sp, nst_list);
4116 free((caddr_t)sp, M_TEMP);
4117 }
4118 free((caddr_t)tsp, M_TEMP);
4119 sf->nsf_flags &= ~NFSNSF_OK;
4120 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4121 sf->nsf_bootvals = NULL;
4122 return;
4123 }
4124 if (len > 0) {
4125 off += sizeof (struct nfst_rec) + tsp->len - 1;
4126 /*
4127 * Search the list for a matching client.
4128 */
4129 LIST_FOREACH(sp, &sf->nsf_head, nst_list) {
4130 if (tsp->len == sp->nst_len &&
4131 !NFSBCMP(tsp->client, sp->nst_client, tsp->len))
4132 break;
4133 }
4134 if (sp == LIST_END(&sf->nsf_head)) {
4135 sp = (struct nfsrv_stable *)malloc(tsp->len +
4136 sizeof (struct nfsrv_stable) - 1, M_TEMP,
4137 M_WAITOK);
4138 NFSBCOPY((caddr_t)tsp, (caddr_t)&sp->nst_rec,
4139 sizeof (struct nfst_rec) + tsp->len - 1);
4140 LIST_INSERT_HEAD(&sf->nsf_head, sp, nst_list);
4141 } else {
4142 if (tsp->flag == NFSNST_REVOKE)
4143 sp->nst_flag |= NFSNST_REVOKE;
4144 else
4145 /*
4146 * A subsequent timestamp indicates the client
4147 * did a setclientid/confirm and any previous
4148 * revoke is no longer relevant.
4149 */
4150 sp->nst_flag &= ~NFSNST_REVOKE;
4151 }
4152 }
4153 } while (len > 0);
4154 free((caddr_t)tsp, M_TEMP);
4155 sf->nsf_flags = NFSNSF_OK;
4156 sf->nsf_eograce = NFSD_MONOSEC + sf->nsf_lease +
4157 NFSRV_LEASEDELTA;
4158 }
4159
4160 /*
4161 * Update the stable storage file, now that the grace period is over.
4162 */
4163 APPLESTATIC void
4164 nfsrv_updatestable(NFSPROC_T *p)
4165 {
4166 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4167 struct nfsrv_stable *sp, *nsp;
4168 int i;
4169 struct nfsvattr nva;
4170 vnode_t vp;
4171 #if defined(__FreeBSD_version) && (__FreeBSD_version >= 500000)
4172 mount_t mp = NULL;
4173 #endif
4174 int error;
4175
4176 if (sf->nsf_fp == NULL || (sf->nsf_flags & NFSNSF_UPDATEDONE))
4177 return;
4178 sf->nsf_flags |= NFSNSF_UPDATEDONE;
4179 /*
4180 * Ok, we need to rewrite the stable storage file.
4181 * - truncate to 0 length
4182 * - write the new first structure
4183 * - loop through the data structures, writing out any that
4184 * have timestamps older than the old boot
4185 */
4186 if (sf->nsf_bootvals) {
4187 sf->nsf_numboots++;
4188 for (i = sf->nsf_numboots - 2; i >= 0; i--)
4189 sf->nsf_bootvals[i + 1] = sf->nsf_bootvals[i];
4190 } else {
4191 sf->nsf_numboots = 1;
4192 sf->nsf_bootvals = (time_t *)malloc(sizeof (time_t),
4193 M_TEMP, M_WAITOK);
4194 }
4195 sf->nsf_bootvals[0] = nfsrvboottime;
4196 sf->nsf_lease = nfsrv_lease;
4197 NFSVNO_ATTRINIT(&nva);
4198 NFSVNO_SETATTRVAL(&nva, size, 0);
4199 vp = NFSFPVNODE(sf->nsf_fp);
4200 vn_start_write(vp, &mp, V_WAIT);
4201 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4202 error = nfsvno_setattr(vp, &nva, NFSFPCRED(sf->nsf_fp), p,
4203 NULL);
4204 NFSVOPUNLOCK(vp, 0);
4205 } else
4206 error = EPERM;
4207 vn_finished_write(mp);
4208 if (!error)
4209 error = NFSD_RDWR(UIO_WRITE, vp,
4210 (caddr_t)&sf->nsf_rec, sizeof (struct nfsf_rec), (off_t)0,
4211 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4212 if (!error)
4213 error = NFSD_RDWR(UIO_WRITE, vp,
4214 (caddr_t)sf->nsf_bootvals,
4215 sf->nsf_numboots * sizeof (time_t),
4216 (off_t)(sizeof (struct nfsf_rec)),
4217 UIO_SYSSPACE, IO_SYNC, NFSFPCRED(sf->nsf_fp), NULL, p);
4218 free((caddr_t)sf->nsf_bootvals, M_TEMP);
4219 sf->nsf_bootvals = NULL;
4220 if (error) {
4221 sf->nsf_flags &= ~NFSNSF_OK;
4222 printf("EEK! Can't write NfsV4 stable storage file\n");
4223 return;
4224 }
4225 sf->nsf_flags |= NFSNSF_OK;
4226
4227 /*
4228 * Loop through the list and write out timestamp records for
4229 * any clients that successfully reclaimed state.
4230 */
4231 LIST_FOREACH_SAFE(sp, &sf->nsf_head, nst_list, nsp) {
4232 if (sp->nst_flag & NFSNST_GOTSTATE) {
4233 nfsrv_writestable(sp->nst_client, sp->nst_len,
4234 NFSNST_NEWSTATE, p);
4235 sp->nst_clp->lc_flags |= LCL_STAMPEDSTABLE;
4236 }
4237 LIST_REMOVE(sp, nst_list);
4238 free((caddr_t)sp, M_TEMP);
4239 }
4240 nfsrv_backupstable();
4241 }
4242
4243 /*
4244 * Append a record to the stable storage file.
4245 */
4246 APPLESTATIC void
4247 nfsrv_writestable(u_char *client, int len, int flag, NFSPROC_T *p)
4248 {
4249 struct nfsrv_stablefirst *sf = &nfsrv_stablefirst;
4250 struct nfst_rec *sp;
4251 int error;
4252
4253 if (!(sf->nsf_flags & NFSNSF_OK) || sf->nsf_fp == NULL)
4254 return;
4255 sp = (struct nfst_rec *)malloc(sizeof (struct nfst_rec) +
4256 len - 1, M_TEMP, M_WAITOK);
4257 sp->len = len;
4258 NFSBCOPY(client, sp->client, len);
4259 sp->flag = flag;
4260 error = NFSD_RDWR(UIO_WRITE, NFSFPVNODE(sf->nsf_fp),
4261 (caddr_t)sp, sizeof (struct nfst_rec) + len - 1, (off_t)0,
4262 UIO_SYSSPACE, (IO_SYNC | IO_APPEND), NFSFPCRED(sf->nsf_fp), NULL, p);
4263 free((caddr_t)sp, M_TEMP);
4264 if (error) {
4265 sf->nsf_flags &= ~NFSNSF_OK;
4266 printf("EEK! Can't write NfsV4 stable storage file\n");
4267 }
4268 }
4269
4270 /*
4271 * This function is called during the grace period to mark a client
4272 * that successfully reclaimed state.
4273 */
4274 static void
4275 nfsrv_markstable(struct nfsclient *clp)
4276 {
4277 struct nfsrv_stable *sp;
4278
4279 /*
4280 * First find the client structure.
4281 */
4282 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4283 if (sp->nst_len == clp->lc_idlen &&
4284 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4285 break;
4286 }
4287 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head))
4288 return;
4289
4290 /*
4291 * Now, just mark it and set the nfsclient back pointer.
4292 */
4293 sp->nst_flag |= NFSNST_GOTSTATE;
4294 sp->nst_clp = clp;
4295 }
4296
4297 /*
4298 * This function is called for a reclaim, to see if it gets grace.
4299 * It returns 0 if a reclaim is allowed, 1 otherwise.
4300 */
4301 static int
4302 nfsrv_checkstable(struct nfsclient *clp)
4303 {
4304 struct nfsrv_stable *sp;
4305
4306 /*
4307 * First, find the entry for the client.
4308 */
4309 LIST_FOREACH(sp, &nfsrv_stablefirst.nsf_head, nst_list) {
4310 if (sp->nst_len == clp->lc_idlen &&
4311 !NFSBCMP(sp->nst_client, clp->lc_id, sp->nst_len))
4312 break;
4313 }
4314
4315 /*
4316 * If not in the list, state was revoked or no state was issued
4317 * since the previous reboot, a reclaim is denied.
4318 */
4319 if (sp == LIST_END(&nfsrv_stablefirst.nsf_head) ||
4320 (sp->nst_flag & NFSNST_REVOKE) ||
4321 !(nfsrv_stablefirst.nsf_flags & NFSNSF_OK))
4322 return (1);
4323 return (0);
4324 }
4325
4326 /*
4327 * Test for and try to clear out a conflicting client. This is called by
4328 * nfsrv_lockctrl() and nfsrv_openctrl() when conflicts with other clients
4329 * a found.
4330 * The trick here is that it can't revoke a conflicting client with an
4331 * expired lease unless it holds the v4root lock, so...
4332 * If no v4root lock, get the lock and return 1 to indicate "try again".
4333 * Return 0 to indicate the conflict can't be revoked and 1 to indicate
4334 * the revocation worked and the conflicting client is "bye, bye", so it
4335 * can be tried again.
4336 * Return 2 to indicate that the vnode is VI_DOOMED after NFSVOPLOCK().
4337 * Unlocks State before a non-zero value is returned.
4338 */
4339 static int
4340 nfsrv_clientconflict(struct nfsclient *clp, int *haslockp, vnode_t vp,
4341 NFSPROC_T *p)
4342 {
4343 int gotlock, lktype = 0;
4344
4345 /*
4346 * If lease hasn't expired, we can't fix it.
4347 */
4348 if (clp->lc_expiry >= NFSD_MONOSEC ||
4349 !(nfsrv_stablefirst.nsf_flags & NFSNSF_UPDATEDONE))
4350 return (0);
4351 if (*haslockp == 0) {
4352 NFSUNLOCKSTATE();
4353 if (vp != NULL) {
4354 lktype = NFSVOPISLOCKED(vp);
4355 NFSVOPUNLOCK(vp, 0);
4356 }
4357 NFSLOCKV4ROOTMUTEX();
4358 nfsv4_relref(&nfsv4rootfs_lock);
4359 do {
4360 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4361 NFSV4ROOTLOCKMUTEXPTR, NULL);
4362 } while (!gotlock);
4363 NFSUNLOCKV4ROOTMUTEX();
4364 *haslockp = 1;
4365 if (vp != NULL) {
4366 NFSVOPLOCK(vp, lktype | LK_RETRY);
4367 if ((vp->v_iflag & VI_DOOMED) != 0)
4368 return (2);
4369 }
4370 return (1);
4371 }
4372 NFSUNLOCKSTATE();
4373
4374 /*
4375 * Ok, we can expire the conflicting client.
4376 */
4377 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4378 nfsrv_backupstable();
4379 nfsrv_cleanclient(clp, p);
4380 nfsrv_freedeleglist(&clp->lc_deleg);
4381 nfsrv_freedeleglist(&clp->lc_olddeleg);
4382 LIST_REMOVE(clp, lc_hash);
4383 nfsrv_zapclient(clp, p);
4384 return (1);
4385 }
4386
4387 /*
4388 * Resolve a delegation conflict.
4389 * Returns 0 to indicate the conflict was resolved without sleeping.
4390 * Return -1 to indicate that the caller should check for conflicts again.
4391 * Return > 0 for an error that should be returned, normally NFSERR_DELAY.
4392 *
4393 * Also, manipulate the nfsv4root_lock, as required. It isn't changed
4394 * for a return of 0, since there was no sleep and it could be required
4395 * later. It is released for a return of NFSERR_DELAY, since the caller
4396 * will return that error. It is released when a sleep was done waiting
4397 * for the delegation to be returned or expire (so that other nfsds can
4398 * handle ops). Then, it must be acquired for the write to stable storage.
4399 * (This function is somewhat similar to nfsrv_clientconflict(), but
4400 * the semantics differ in a couple of subtle ways. The return of 0
4401 * indicates the conflict was resolved without sleeping here, not
4402 * that the conflict can't be resolved and the handling of nfsv4root_lock
4403 * differs, as noted above.)
4404 * Unlocks State before returning a non-zero value.
4405 */
4406 static int
4407 nfsrv_delegconflict(struct nfsstate *stp, int *haslockp, NFSPROC_T *p,
4408 vnode_t vp)
4409 {
4410 struct nfsclient *clp = stp->ls_clp;
4411 int gotlock, error, lktype = 0, retrycnt, zapped_clp;
4412 nfsv4stateid_t tstateid;
4413 fhandle_t tfh;
4414
4415 /*
4416 * If the conflict is with an old delegation...
4417 */
4418 if (stp->ls_flags & NFSLCK_OLDDELEG) {
4419 /*
4420 * You can delete it, if it has expired.
4421 */
4422 if (clp->lc_delegtime < NFSD_MONOSEC) {
4423 nfsrv_freedeleg(stp);
4424 NFSUNLOCKSTATE();
4425 error = -1;
4426 goto out;
4427 }
4428 NFSUNLOCKSTATE();
4429 /*
4430 * During this delay, the old delegation could expire or it
4431 * could be recovered by the client via an Open with
4432 * CLAIM_DELEGATE_PREV.
4433 * Release the nfsv4root_lock, if held.
4434 */
4435 if (*haslockp) {
4436 *haslockp = 0;
4437 NFSLOCKV4ROOTMUTEX();
4438 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4439 NFSUNLOCKV4ROOTMUTEX();
4440 }
4441 error = NFSERR_DELAY;
4442 goto out;
4443 }
4444
4445 /*
4446 * It's a current delegation, so:
4447 * - check to see if the delegation has expired
4448 * - if so, get the v4root lock and then expire it
4449 */
4450 if (!(stp->ls_flags & NFSLCK_DELEGRECALL)) {
4451 /*
4452 * - do a recall callback, since not yet done
4453 * For now, never allow truncate to be set. To use
4454 * truncate safely, it must be guaranteed that the
4455 * Remove, Rename or Setattr with size of 0 will
4456 * succeed and that would require major changes to
4457 * the VFS/Vnode OPs.
4458 * Set the expiry time large enough so that it won't expire
4459 * until after the callback, then set it correctly, once
4460 * the callback is done. (The delegation will now time
4461 * out whether or not the Recall worked ok. The timeout
4462 * will be extended when ops are done on the delegation
4463 * stateid, up to the timelimit.)
4464 */
4465 stp->ls_delegtime = NFSD_MONOSEC + (2 * nfsrv_lease) +
4466 NFSRV_LEASEDELTA;
4467 stp->ls_delegtimelimit = NFSD_MONOSEC + (6 * nfsrv_lease) +
4468 NFSRV_LEASEDELTA;
4469 stp->ls_flags |= NFSLCK_DELEGRECALL;
4470
4471 /*
4472 * Loop NFSRV_CBRETRYCNT times while the CBRecall replies
4473 * NFSERR_BADSTATEID or NFSERR_BADHANDLE. This is done
4474 * in order to try and avoid a race that could happen
4475 * when a CBRecall request passed the Open reply with
4476 * the delegation in it when transitting the network.
4477 * Since nfsrv_docallback will sleep, don't use stp after
4478 * the call.
4479 */
4480 NFSBCOPY((caddr_t)&stp->ls_stateid, (caddr_t)&tstateid,
4481 sizeof (tstateid));
4482 NFSBCOPY((caddr_t)&stp->ls_lfp->lf_fh, (caddr_t)&tfh,
4483 sizeof (tfh));
4484 NFSUNLOCKSTATE();
4485 if (*haslockp) {
4486 *haslockp = 0;
4487 NFSLOCKV4ROOTMUTEX();
4488 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4489 NFSUNLOCKV4ROOTMUTEX();
4490 }
4491 retrycnt = 0;
4492 do {
4493 error = nfsrv_docallback(clp, NFSV4OP_CBRECALL,
4494 &tstateid, 0, &tfh, NULL, NULL, p);
4495 retrycnt++;
4496 } while ((error == NFSERR_BADSTATEID ||
4497 error == NFSERR_BADHANDLE) && retrycnt < NFSV4_CBRETRYCNT);
4498 error = NFSERR_DELAY;
4499 goto out;
4500 }
4501
4502 if (clp->lc_expiry >= NFSD_MONOSEC &&
4503 stp->ls_delegtime >= NFSD_MONOSEC) {
4504 NFSUNLOCKSTATE();
4505 /*
4506 * A recall has been done, but it has not yet expired.
4507 * So, RETURN_DELAY.
4508 */
4509 if (*haslockp) {
4510 *haslockp = 0;
4511 NFSLOCKV4ROOTMUTEX();
4512 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4513 NFSUNLOCKV4ROOTMUTEX();
4514 }
4515 error = NFSERR_DELAY;
4516 goto out;
4517 }
4518
4519 /*
4520 * If we don't yet have the lock, just get it and then return,
4521 * since we need that before deleting expired state, such as
4522 * this delegation.
4523 * When getting the lock, unlock the vnode, so other nfsds that
4524 * are in progress, won't get stuck waiting for the vnode lock.
4525 */
4526 if (*haslockp == 0) {
4527 NFSUNLOCKSTATE();
4528 if (vp != NULL) {
4529 lktype = NFSVOPISLOCKED(vp);
4530 NFSVOPUNLOCK(vp, 0);
4531 }
4532 NFSLOCKV4ROOTMUTEX();
4533 nfsv4_relref(&nfsv4rootfs_lock);
4534 do {
4535 gotlock = nfsv4_lock(&nfsv4rootfs_lock, 1, NULL,
4536 NFSV4ROOTLOCKMUTEXPTR, NULL);
4537 } while (!gotlock);
4538 NFSUNLOCKV4ROOTMUTEX();
4539 *haslockp = 1;
4540 if (vp != NULL) {
4541 NFSVOPLOCK(vp, lktype | LK_RETRY);
4542 if ((vp->v_iflag & VI_DOOMED) != 0) {
4543 *haslockp = 0;
4544 NFSLOCKV4ROOTMUTEX();
4545 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4546 NFSUNLOCKV4ROOTMUTEX();
4547 error = NFSERR_PERM;
4548 goto out;
4549 }
4550 }
4551 error = -1;
4552 goto out;
4553 }
4554
4555 NFSUNLOCKSTATE();
4556 /*
4557 * Ok, we can delete the expired delegation.
4558 * First, write the Revoke record to stable storage and then
4559 * clear out the conflict.
4560 * Since all other nfsd threads are now blocked, we can safely
4561 * sleep without the state changing.
4562 */
4563 nfsrv_writestable(clp->lc_id, clp->lc_idlen, NFSNST_REVOKE, p);
4564 nfsrv_backupstable();
4565 if (clp->lc_expiry < NFSD_MONOSEC) {
4566 nfsrv_cleanclient(clp, p);
4567 nfsrv_freedeleglist(&clp->lc_deleg);
4568 nfsrv_freedeleglist(&clp->lc_olddeleg);
4569 LIST_REMOVE(clp, lc_hash);
4570 zapped_clp = 1;
4571 } else {
4572 nfsrv_freedeleg(stp);
4573 zapped_clp = 0;
4574 }
4575 if (zapped_clp)
4576 nfsrv_zapclient(clp, p);
4577 error = -1;
4578
4579 out:
4580 NFSEXITCODE(error);
4581 return (error);
4582 }
4583
4584 /*
4585 * Check for a remove allowed, if remove is set to 1 and get rid of
4586 * delegations.
4587 */
4588 APPLESTATIC int
4589 nfsrv_checkremove(vnode_t vp, int remove, NFSPROC_T *p)
4590 {
4591 struct nfsstate *stp;
4592 struct nfslockfile *lfp;
4593 int error, haslock = 0;
4594 fhandle_t nfh;
4595
4596 /*
4597 * First, get the lock file structure.
4598 * (A return of -1 means no associated state, so remove ok.)
4599 */
4600 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4601 tryagain:
4602 NFSLOCKSTATE();
4603 if (!error)
4604 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4605 if (error) {
4606 NFSUNLOCKSTATE();
4607 if (haslock) {
4608 NFSLOCKV4ROOTMUTEX();
4609 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4610 NFSUNLOCKV4ROOTMUTEX();
4611 }
4612 if (error == -1)
4613 error = 0;
4614 goto out;
4615 }
4616
4617 /*
4618 * Now, we must Recall any delegations.
4619 */
4620 error = nfsrv_cleandeleg(vp, lfp, NULL, &haslock, p);
4621 if (error) {
4622 /*
4623 * nfsrv_cleandeleg() unlocks state for non-zero
4624 * return.
4625 */
4626 if (error == -1)
4627 goto tryagain;
4628 if (haslock) {
4629 NFSLOCKV4ROOTMUTEX();
4630 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4631 NFSUNLOCKV4ROOTMUTEX();
4632 }
4633 goto out;
4634 }
4635
4636 /*
4637 * Now, look for a conflicting open share.
4638 */
4639 if (remove) {
4640 LIST_FOREACH(stp, &lfp->lf_open, ls_file) {
4641 if (stp->ls_flags & NFSLCK_WRITEDENY) {
4642 error = NFSERR_FILEOPEN;
4643 break;
4644 }
4645 }
4646 }
4647
4648 NFSUNLOCKSTATE();
4649 if (haslock) {
4650 NFSLOCKV4ROOTMUTEX();
4651 nfsv4_unlock(&nfsv4rootfs_lock, 1);
4652 NFSUNLOCKV4ROOTMUTEX();
4653 }
4654
4655 out:
4656 NFSEXITCODE(error);
4657 return (error);
4658 }
4659
4660 /*
4661 * Clear out all delegations for the file referred to by lfp.
4662 * May return NFSERR_DELAY, if there will be a delay waiting for
4663 * delegations to expire.
4664 * Returns -1 to indicate it slept while recalling a delegation.
4665 * This function has the side effect of deleting the nfslockfile structure,
4666 * if it no longer has associated state and didn't have to sleep.
4667 * Unlocks State before a non-zero value is returned.
4668 */
4669 static int
4670 nfsrv_cleandeleg(vnode_t vp, struct nfslockfile *lfp,
4671 struct nfsclient *clp, int *haslockp, NFSPROC_T *p)
4672 {
4673 struct nfsstate *stp, *nstp;
4674 int ret = 0;
4675
4676 stp = LIST_FIRST(&lfp->lf_deleg);
4677 while (stp != LIST_END(&lfp->lf_deleg)) {
4678 nstp = LIST_NEXT(stp, ls_file);
4679 if (stp->ls_clp != clp) {
4680 ret = nfsrv_delegconflict(stp, haslockp, p, vp);
4681 if (ret) {
4682 /*
4683 * nfsrv_delegconflict() unlocks state
4684 * when it returns non-zero.
4685 */
4686 goto out;
4687 }
4688 }
4689 stp = nstp;
4690 }
4691 out:
4692 NFSEXITCODE(ret);
4693 return (ret);
4694 }
4695
4696 /*
4697 * There are certain operations that, when being done outside of NFSv4,
4698 * require that any NFSv4 delegation for the file be recalled.
4699 * This function is to be called for those cases:
4700 * VOP_RENAME() - When a delegation is being recalled for any reason,
4701 * the client may have to do Opens against the server, using the file's
4702 * final component name. If the file has been renamed on the server,
4703 * that component name will be incorrect and the Open will fail.
4704 * VOP_REMOVE() - Theoretically, a client could Open a file after it has
4705 * been removed on the server, if there is a delegation issued to
4706 * that client for the file. I say "theoretically" since clients
4707 * normally do an Access Op before the Open and that Access Op will
4708 * fail with ESTALE. Note that NFSv2 and 3 don't even do Opens, so
4709 * they will detect the file's removal in the same manner. (There is
4710 * one case where RFC3530 allows a client to do an Open without first
4711 * doing an Access Op, which is passage of a check against the ACE
4712 * returned with a Write delegation, but current practice is to ignore
4713 * the ACE and always do an Access Op.)
4714 * Since the functions can only be called with an unlocked vnode, this
4715 * can't be done at this time.
4716 * VOP_ADVLOCK() - When a client holds a delegation, it can issue byte range
4717 * locks locally in the client, which are not visible to the server. To
4718 * deal with this, issuing of delegations for a vnode must be disabled
4719 * and all delegations for the vnode recalled. This is done via the
4720 * second function, using the VV_DISABLEDELEG vflag on the vnode.
4721 */
4722 APPLESTATIC void
4723 nfsd_recalldelegation(vnode_t vp, NFSPROC_T *p)
4724 {
4725 time_t starttime;
4726 int error;
4727
4728 /*
4729 * First, check to see if the server is currently running and it has
4730 * been called for a regular file when issuing delegations.
4731 */
4732 if (newnfs_numnfsd == 0 || vp->v_type != VREG ||
4733 nfsrv_issuedelegs == 0)
4734 return;
4735
4736 KASSERT((NFSVOPISLOCKED(vp) != LK_EXCLUSIVE), ("vp %p is locked", vp));
4737 /*
4738 * First, get a reference on the nfsv4rootfs_lock so that an
4739 * exclusive lock cannot be acquired by another thread.
4740 */
4741 NFSLOCKV4ROOTMUTEX();
4742 nfsv4_getref(&nfsv4rootfs_lock, NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
4743 NFSUNLOCKV4ROOTMUTEX();
4744
4745 /*
4746 * Now, call nfsrv_checkremove() in a loop while it returns
4747 * NFSERR_DELAY. Return upon any other error or when timed out.
4748 */
4749 starttime = NFSD_MONOSEC;
4750 do {
4751 if (NFSVOPLOCK(vp, LK_EXCLUSIVE) == 0) {
4752 error = nfsrv_checkremove(vp, 0, p);
4753 NFSVOPUNLOCK(vp, 0);
4754 } else
4755 error = EPERM;
4756 if (error == NFSERR_DELAY) {
4757 if (NFSD_MONOSEC - starttime > NFS_REMOVETIMEO)
4758 break;
4759 /* Sleep for a short period of time */
4760 (void) nfs_catnap(PZERO, 0, "nfsremove");
4761 }
4762 } while (error == NFSERR_DELAY);
4763 NFSLOCKV4ROOTMUTEX();
4764 nfsv4_relref(&nfsv4rootfs_lock);
4765 NFSUNLOCKV4ROOTMUTEX();
4766 }
4767
4768 APPLESTATIC void
4769 nfsd_disabledelegation(vnode_t vp, NFSPROC_T *p)
4770 {
4771
4772 #ifdef VV_DISABLEDELEG
4773 /*
4774 * First, flag issuance of delegations disabled.
4775 */
4776 atomic_set_long(&vp->v_vflag, VV_DISABLEDELEG);
4777 #endif
4778
4779 /*
4780 * Then call nfsd_recalldelegation() to get rid of all extant
4781 * delegations.
4782 */
4783 nfsd_recalldelegation(vp, p);
4784 }
4785
4786 /*
4787 * Check for conflicting locks, etc. and then get rid of delegations.
4788 * (At one point I thought that I should get rid of delegations for any
4789 * Setattr, since it could potentially disallow the I/O op (read or write)
4790 * allowed by the delegation. However, Setattr Ops that aren't changing
4791 * the size get a stateid of all 0s, so you can't tell if it is a delegation
4792 * for the same client or a different one, so I decided to only get rid
4793 * of delegations for other clients when the size is being changed.)
4794 * In general, a Setattr can disable NFS I/O Ops that are outstanding, such
4795 * as Write backs, even if there is no delegation, so it really isn't any
4796 * different?)
4797 */
4798 APPLESTATIC int
4799 nfsrv_checksetattr(vnode_t vp, struct nfsrv_descript *nd,
4800 nfsv4stateid_t *stateidp, struct nfsvattr *nvap, nfsattrbit_t *attrbitp,
4801 struct nfsexstuff *exp, NFSPROC_T *p)
4802 {
4803 struct nfsstate st, *stp = &st;
4804 struct nfslock lo, *lop = &lo;
4805 int error = 0;
4806 nfsquad_t clientid;
4807
4808 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE)) {
4809 stp->ls_flags = (NFSLCK_CHECK | NFSLCK_WRITEACCESS);
4810 lop->lo_first = nvap->na_size;
4811 } else {
4812 stp->ls_flags = 0;
4813 lop->lo_first = 0;
4814 }
4815 if (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNER) ||
4816 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_OWNERGROUP) ||
4817 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_MODE) ||
4818 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_ACL))
4819 stp->ls_flags |= NFSLCK_SETATTR;
4820 if (stp->ls_flags == 0)
4821 goto out;
4822 lop->lo_end = NFS64BITSSET;
4823 lop->lo_flags = NFSLCK_WRITE;
4824 stp->ls_ownerlen = 0;
4825 stp->ls_op = NULL;
4826 stp->ls_uid = nd->nd_cred->cr_uid;
4827 stp->ls_stateid.seqid = stateidp->seqid;
4828 clientid.lval[0] = stp->ls_stateid.other[0] = stateidp->other[0];
4829 clientid.lval[1] = stp->ls_stateid.other[1] = stateidp->other[1];
4830 stp->ls_stateid.other[2] = stateidp->other[2];
4831 error = nfsrv_lockctrl(vp, &stp, &lop, NULL, clientid,
4832 stateidp, exp, nd, p);
4833
4834 out:
4835 NFSEXITCODE2(error, nd);
4836 return (error);
4837 }
4838
4839 /*
4840 * Check for a write delegation and do a CBGETATTR if there is one, updating
4841 * the attributes, as required.
4842 * Should I return an error if I can't get the attributes? (For now, I'll
4843 * just return ok.
4844 */
4845 APPLESTATIC int
4846 nfsrv_checkgetattr(struct nfsrv_descript *nd, vnode_t vp,
4847 struct nfsvattr *nvap, nfsattrbit_t *attrbitp, struct ucred *cred,
4848 NFSPROC_T *p)
4849 {
4850 struct nfsstate *stp;
4851 struct nfslockfile *lfp;
4852 struct nfsclient *clp;
4853 struct nfsvattr nva;
4854 fhandle_t nfh;
4855 int error = 0;
4856 nfsattrbit_t cbbits;
4857 u_quad_t delegfilerev;
4858
4859 NFSCBGETATTR_ATTRBIT(attrbitp, &cbbits);
4860 if (!NFSNONZERO_ATTRBIT(&cbbits))
4861 goto out;
4862
4863 /*
4864 * Get the lock file structure.
4865 * (A return of -1 means no associated state, so return ok.)
4866 */
4867 error = nfsrv_getlockfh(vp, NFSLCK_CHECK, NULL, &nfh, p);
4868 NFSLOCKSTATE();
4869 if (!error)
4870 error = nfsrv_getlockfile(NFSLCK_CHECK, NULL, &lfp, &nfh, 0);
4871 if (error) {
4872 NFSUNLOCKSTATE();
4873 if (error == -1)
4874 error = 0;
4875 goto out;
4876 }
4877
4878 /*
4879 * Now, look for a write delegation.
4880 */
4881 LIST_FOREACH(stp, &lfp->lf_deleg, ls_file) {
4882 if (stp->ls_flags & NFSLCK_DELEGWRITE)
4883 break;
4884 }
4885 if (stp == LIST_END(&lfp->lf_deleg)) {
4886 NFSUNLOCKSTATE();
4887 goto out;
4888 }
4889 clp = stp->ls_clp;
4890 delegfilerev = stp->ls_filerev;
4891
4892 /*
4893 * If the Write delegation was issued as a part of this Compound RPC
4894 * or if we have an Implied Clientid (used in a previous Op in this
4895 * compound) and it is the client the delegation was issued to,
4896 * just return ok.
4897 * I also assume that it is from the same client iff the network
4898 * host IP address is the same as the callback address. (Not
4899 * exactly correct by the RFC, but avoids a lot of Getattr
4900 * callbacks.)
4901 */
4902 if (nd->nd_compref == stp->ls_compref ||
4903 ((nd->nd_flag & ND_IMPLIEDCLID) &&
4904 clp->lc_clientid.qval == nd->nd_clientid.qval) ||
4905 nfsaddr2_match(clp->lc_req.nr_nam, nd->nd_nam)) {
4906 NFSUNLOCKSTATE();
4907 goto out;
4908 }
4909
4910 /*
4911 * We are now done with the delegation state structure,
4912 * so the statelock can be released and we can now tsleep().
4913 */
4914
4915 /*
4916 * Now, we must do the CB Getattr callback, to see if Change or Size
4917 * has changed.
4918 */
4919 if (clp->lc_expiry >= NFSD_MONOSEC) {
4920 NFSUNLOCKSTATE();
4921 NFSVNO_ATTRINIT(&nva);
4922 nva.na_filerev = NFS64BITSSET;
4923 error = nfsrv_docallback(clp, NFSV4OP_CBGETATTR, NULL,
4924 0, &nfh, &nva, &cbbits, p);
4925 if (!error) {
4926 if ((nva.na_filerev != NFS64BITSSET &&
4927 nva.na_filerev > delegfilerev) ||
4928 (NFSVNO_ISSETSIZE(&nva) &&
4929 nva.na_size != nvap->na_size)) {
4930 error = nfsvno_updfilerev(vp, nvap, cred, p);
4931 if (NFSVNO_ISSETSIZE(&nva))
4932 nvap->na_size = nva.na_size;
4933 }
4934 } else
4935 error = 0; /* Ignore callback errors for now. */
4936 } else {
4937 NFSUNLOCKSTATE();
4938 }
4939
4940 out:
4941 NFSEXITCODE2(error, nd);
4942 return (error);
4943 }
4944
4945 /*
4946 * This function looks for openowners that haven't had any opens for
4947 * a while and throws them away. Called by an nfsd when NFSNSF_NOOPENS
4948 * is set.
4949 */
4950 APPLESTATIC void
4951 nfsrv_throwawayopens(NFSPROC_T *p)
4952 {
4953 struct nfsclient *clp, *nclp;
4954 struct nfsstate *stp, *nstp;
4955 int i;
4956
4957 NFSLOCKSTATE();
4958 nfsrv_stablefirst.nsf_flags &= ~NFSNSF_NOOPENS;
4959 /*
4960 * For each client...
4961 */
4962 for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
4963 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
4964 LIST_FOREACH_SAFE(stp, &clp->lc_open, ls_list, nstp) {
4965 if (LIST_EMPTY(&stp->ls_open) &&
4966 (stp->ls_noopens > NFSNOOPEN ||
4967 (nfsrv_openpluslock * 2) >
4968 NFSRV_V4STATELIMIT))
4969 nfsrv_freeopenowner(stp, 0, p);
4970 }
4971 }
4972 }
4973 NFSUNLOCKSTATE();
4974 }
4975
4976 /*
4977 * This function checks to see if the credentials are the same.
4978 * Returns 1 for not same, 0 otherwise.
4979 */
4980 static int
4981 nfsrv_notsamecredname(struct nfsrv_descript *nd, struct nfsclient *clp)
4982 {
4983
4984 if (nd->nd_flag & ND_GSS) {
4985 if (!(clp->lc_flags & LCL_GSS))
4986 return (1);
4987 if (clp->lc_flags & LCL_NAME) {
4988 if (nd->nd_princlen != clp->lc_namelen ||
4989 NFSBCMP(nd->nd_principal, clp->lc_name,
4990 clp->lc_namelen))
4991 return (1);
4992 else
4993 return (0);
4994 }
4995 if (nd->nd_cred->cr_uid == clp->lc_uid)
4996 return (0);
4997 else
4998 return (1);
4999 } else if (clp->lc_flags & LCL_GSS)
5000 return (1);
5001 /*
5002 * For AUTH_SYS, allow the same uid or root. (This is underspecified
5003 * in RFC3530, which talks about principals, but doesn't say anything
5004 * about uids for AUTH_SYS.)
5005 */
5006 if (nd->nd_cred->cr_uid == clp->lc_uid || nd->nd_cred->cr_uid == 0)
5007 return (0);
5008 else
5009 return (1);
5010 }
5011
5012 /*
5013 * Calculate the lease expiry time.
5014 */
5015 static time_t
5016 nfsrv_leaseexpiry(void)
5017 {
5018
5019 if (nfsrv_stablefirst.nsf_eograce > NFSD_MONOSEC)
5020 return (NFSD_MONOSEC + 2 * (nfsrv_lease + NFSRV_LEASEDELTA));
5021 return (NFSD_MONOSEC + nfsrv_lease + NFSRV_LEASEDELTA);
5022 }
5023
5024 /*
5025 * Delay the delegation timeout as far as ls_delegtimelimit, as required.
5026 */
5027 static void
5028 nfsrv_delaydelegtimeout(struct nfsstate *stp)
5029 {
5030
5031 if ((stp->ls_flags & NFSLCK_DELEGRECALL) == 0)
5032 return;
5033
5034 if ((stp->ls_delegtime + 15) > NFSD_MONOSEC &&
5035 stp->ls_delegtime < stp->ls_delegtimelimit) {
5036 stp->ls_delegtime += nfsrv_lease;
5037 if (stp->ls_delegtime > stp->ls_delegtimelimit)
5038 stp->ls_delegtime = stp->ls_delegtimelimit;
5039 }
5040 }
5041
5042 /*
5043 * This function checks to see if there is any other state associated
5044 * with the openowner for this Open.
5045 * It returns 1 if there is no other state, 0 otherwise.
5046 */
5047 static int
5048 nfsrv_nootherstate(struct nfsstate *stp)
5049 {
5050 struct nfsstate *tstp;
5051
5052 LIST_FOREACH(tstp, &stp->ls_openowner->ls_open, ls_list) {
5053 if (tstp != stp || !LIST_EMPTY(&tstp->ls_lock))
5054 return (0);
5055 }
5056 return (1);
5057 }
5058
5059 /*
5060 * Create a list of lock deltas (changes to local byte range locking
5061 * that can be rolled back using the list) and apply the changes via
5062 * nfsvno_advlock(). Optionally, lock the list. It is expected that either
5063 * the rollback or update function will be called after this.
5064 * It returns an error (and rolls back, as required), if any nfsvno_advlock()
5065 * call fails. If it returns an error, it will unlock the list.
5066 */
5067 static int
5068 nfsrv_locallock(vnode_t vp, struct nfslockfile *lfp, int flags,
5069 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5070 {
5071 struct nfslock *lop, *nlop;
5072 int error = 0;
5073
5074 /* Loop through the list of locks. */
5075 lop = LIST_FIRST(&lfp->lf_locallock);
5076 while (first < end && lop != NULL) {
5077 nlop = LIST_NEXT(lop, lo_lckowner);
5078 if (first >= lop->lo_end) {
5079 /* not there yet */
5080 lop = nlop;
5081 } else if (first < lop->lo_first) {
5082 /* new one starts before entry in list */
5083 if (end <= lop->lo_first) {
5084 /* no overlap between old and new */
5085 error = nfsrv_dolocal(vp, lfp, flags,
5086 NFSLCK_UNLOCK, first, end, cfp, p);
5087 if (error != 0)
5088 break;
5089 first = end;
5090 } else {
5091 /* handle fragment overlapped with new one */
5092 error = nfsrv_dolocal(vp, lfp, flags,
5093 NFSLCK_UNLOCK, first, lop->lo_first, cfp,
5094 p);
5095 if (error != 0)
5096 break;
5097 first = lop->lo_first;
5098 }
5099 } else {
5100 /* new one overlaps this entry in list */
5101 if (end <= lop->lo_end) {
5102 /* overlaps all of new one */
5103 error = nfsrv_dolocal(vp, lfp, flags,
5104 lop->lo_flags, first, end, cfp, p);
5105 if (error != 0)
5106 break;
5107 first = end;
5108 } else {
5109 /* handle fragment overlapped with new one */
5110 error = nfsrv_dolocal(vp, lfp, flags,
5111 lop->lo_flags, first, lop->lo_end, cfp, p);
5112 if (error != 0)
5113 break;
5114 first = lop->lo_end;
5115 lop = nlop;
5116 }
5117 }
5118 }
5119 if (first < end && error == 0)
5120 /* handle fragment past end of list */
5121 error = nfsrv_dolocal(vp, lfp, flags, NFSLCK_UNLOCK, first,
5122 end, cfp, p);
5123
5124 NFSEXITCODE(error);
5125 return (error);
5126 }
5127
5128 /*
5129 * Local lock unlock. Unlock all byte ranges that are no longer locked
5130 * by NFSv4. To do this, unlock any subranges of first-->end that
5131 * do not overlap with the byte ranges of any lock in the lfp->lf_lock
5132 * list. This list has all locks for the file held by other
5133 * <clientid, lockowner> tuples. The list is ordered by increasing
5134 * lo_first value, but may have entries that overlap each other, for
5135 * the case of read locks.
5136 */
5137 static void
5138 nfsrv_localunlock(vnode_t vp, struct nfslockfile *lfp, uint64_t init_first,
5139 uint64_t init_end, NFSPROC_T *p)
5140 {
5141 struct nfslock *lop;
5142 uint64_t first, end, prevfirst;
5143
5144 first = init_first;
5145 end = init_end;
5146 while (first < init_end) {
5147 /* Loop through all nfs locks, adjusting first and end */
5148 prevfirst = 0;
5149 LIST_FOREACH(lop, &lfp->lf_lock, lo_lckfile) {
5150 KASSERT(prevfirst <= lop->lo_first,
5151 ("nfsv4 locks out of order"));
5152 KASSERT(lop->lo_first < lop->lo_end,
5153 ("nfsv4 bogus lock"));
5154 prevfirst = lop->lo_first;
5155 if (first >= lop->lo_first &&
5156 first < lop->lo_end)
5157 /*
5158 * Overlaps with initial part, so trim
5159 * off that initial part by moving first past
5160 * it.
5161 */
5162 first = lop->lo_end;
5163 else if (end > lop->lo_first &&
5164 lop->lo_first > first) {
5165 /*
5166 * This lock defines the end of the
5167 * segment to unlock, so set end to the
5168 * start of it and break out of the loop.
5169 */
5170 end = lop->lo_first;
5171 break;
5172 }
5173 if (first >= end)
5174 /*
5175 * There is no segment left to do, so
5176 * break out of this loop and then exit
5177 * the outer while() since first will be set
5178 * to end, which must equal init_end here.
5179 */
5180 break;
5181 }
5182 if (first < end) {
5183 /* Unlock this segment */
5184 (void) nfsrv_dolocal(vp, lfp, NFSLCK_UNLOCK,
5185 NFSLCK_READ, first, end, NULL, p);
5186 nfsrv_locallock_commit(lfp, NFSLCK_UNLOCK,
5187 first, end);
5188 }
5189 /*
5190 * Now move past this segment and look for any further
5191 * segment in the range, if there is one.
5192 */
5193 first = end;
5194 end = init_end;
5195 }
5196 }
5197
5198 /*
5199 * Do the local lock operation and update the rollback list, as required.
5200 * Perform the rollback and return the error if nfsvno_advlock() fails.
5201 */
5202 static int
5203 nfsrv_dolocal(vnode_t vp, struct nfslockfile *lfp, int flags, int oldflags,
5204 uint64_t first, uint64_t end, struct nfslockconflict *cfp, NFSPROC_T *p)
5205 {
5206 struct nfsrollback *rlp;
5207 int error = 0, ltype, oldltype;
5208
5209 if (flags & NFSLCK_WRITE)
5210 ltype = F_WRLCK;
5211 else if (flags & NFSLCK_READ)
5212 ltype = F_RDLCK;
5213 else
5214 ltype = F_UNLCK;
5215 if (oldflags & NFSLCK_WRITE)
5216 oldltype = F_WRLCK;
5217 else if (oldflags & NFSLCK_READ)
5218 oldltype = F_RDLCK;
5219 else
5220 oldltype = F_UNLCK;
5221 if (ltype == oldltype || (oldltype == F_WRLCK && ltype == F_RDLCK))
5222 /* nothing to do */
5223 goto out;
5224 error = nfsvno_advlock(vp, ltype, first, end, p);
5225 if (error != 0) {
5226 if (cfp != NULL) {
5227 cfp->cl_clientid.lval[0] = 0;
5228 cfp->cl_clientid.lval[1] = 0;
5229 cfp->cl_first = 0;
5230 cfp->cl_end = NFS64BITSSET;
5231 cfp->cl_flags = NFSLCK_WRITE;
5232 cfp->cl_ownerlen = 5;
5233 NFSBCOPY("LOCAL", cfp->cl_owner, 5);
5234 }
5235 nfsrv_locallock_rollback(vp, lfp, p);
5236 } else if (ltype != F_UNLCK) {
5237 rlp = malloc(sizeof (struct nfsrollback), M_NFSDROLLBACK,
5238 M_WAITOK);
5239 rlp->rlck_first = first;
5240 rlp->rlck_end = end;
5241 rlp->rlck_type = oldltype;
5242 LIST_INSERT_HEAD(&lfp->lf_rollback, rlp, rlck_list);
5243 }
5244
5245 out:
5246 NFSEXITCODE(error);
5247 return (error);
5248 }
5249
5250 /*
5251 * Roll back local lock changes and free up the rollback list.
5252 */
5253 static void
5254 nfsrv_locallock_rollback(vnode_t vp, struct nfslockfile *lfp, NFSPROC_T *p)
5255 {
5256 struct nfsrollback *rlp, *nrlp;
5257
5258 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp) {
5259 (void) nfsvno_advlock(vp, rlp->rlck_type, rlp->rlck_first,
5260 rlp->rlck_end, p);
5261 free(rlp, M_NFSDROLLBACK);
5262 }
5263 LIST_INIT(&lfp->lf_rollback);
5264 }
5265
5266 /*
5267 * Update local lock list and delete rollback list (ie now committed to the
5268 * local locks). Most of the work is done by the internal function.
5269 */
5270 static void
5271 nfsrv_locallock_commit(struct nfslockfile *lfp, int flags, uint64_t first,
5272 uint64_t end)
5273 {
5274 struct nfsrollback *rlp, *nrlp;
5275 struct nfslock *new_lop, *other_lop;
5276
5277 new_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK, M_WAITOK);
5278 if (flags & (NFSLCK_READ | NFSLCK_WRITE))
5279 other_lop = malloc(sizeof (struct nfslock), M_NFSDLOCK,
5280 M_WAITOK);
5281 else
5282 other_lop = NULL;
5283 new_lop->lo_flags = flags;
5284 new_lop->lo_first = first;
5285 new_lop->lo_end = end;
5286 nfsrv_updatelock(NULL, &new_lop, &other_lop, lfp);
5287 if (new_lop != NULL)
5288 free(new_lop, M_NFSDLOCK);
5289 if (other_lop != NULL)
5290 free(other_lop, M_NFSDLOCK);
5291
5292 /* and get rid of the rollback list */
5293 LIST_FOREACH_SAFE(rlp, &lfp->lf_rollback, rlck_list, nrlp)
5294 free(rlp, M_NFSDROLLBACK);
5295 LIST_INIT(&lfp->lf_rollback);
5296 }
5297
5298 /*
5299 * Lock the struct nfslockfile for local lock updating.
5300 */
5301 static void
5302 nfsrv_locklf(struct nfslockfile *lfp)
5303 {
5304 int gotlock;
5305
5306 /* lf_usecount ensures *lfp won't be free'd */
5307 lfp->lf_usecount++;
5308 do {
5309 gotlock = nfsv4_lock(&lfp->lf_locallock_lck, 1, NULL,
5310 NFSSTATEMUTEXPTR, NULL);
5311 } while (gotlock == 0);
5312 lfp->lf_usecount--;
5313 }
5314
5315 /*
5316 * Unlock the struct nfslockfile after local lock updating.
5317 */
5318 static void
5319 nfsrv_unlocklf(struct nfslockfile *lfp)
5320 {
5321
5322 nfsv4_unlock(&lfp->lf_locallock_lck, 0);
5323 }
5324
5325 /*
5326 * Clear out all state for the NFSv4 server.
5327 * Must be called by a thread that can sleep when no nfsds are running.
5328 */
5329 void
5330 nfsrv_throwawayallstate(NFSPROC_T *p)
5331 {
5332 struct nfsclient *clp, *nclp;
5333 struct nfslockfile *lfp, *nlfp;
5334 int i;
5335
5336 /*
5337 * For each client, clean out the state and then free the structure.
5338 */
5339 for (i = 0; i < NFSCLIENTHASHSIZE; i++) {
5340 LIST_FOREACH_SAFE(clp, &nfsclienthash[i], lc_hash, nclp) {
5341 nfsrv_cleanclient(clp, p);
5342 nfsrv_freedeleglist(&clp->lc_deleg);
5343 nfsrv_freedeleglist(&clp->lc_olddeleg);
5344 free(clp, M_NFSDCLIENT);
5345 }
5346 }
5347
5348 /*
5349 * Also, free up any remaining lock file structures.
5350 */
5351 for (i = 0; i < NFSLOCKHASHSIZE; i++) {
5352 LIST_FOREACH_SAFE(lfp, &nfslockhash[i], lf_hash, nlfp) {
5353 printf("nfsd unload: fnd a lock file struct\n");
5354 nfsrv_freenfslockfile(lfp);
5355 }
5356 }
5357 }
5358
Cache object: 22d25a31d182f01ae5c15a842a8d7c37
|