]> git.ipfire.org Git - thirdparty/linux.git/blob - fs/nfsd/nfs4state.c
gpu: host1x: Use SMMU on Tegra124 and Tegra210
[thirdparty/linux.git] / fs / nfsd / nfs4state.c
1 /*
2 * Copyright (c) 2001 The Regents of the University of Michigan.
3 * All rights reserved.
4 *
5 * Kendrick Smith <kmsmith@umich.edu>
6 * Andy Adamson <kandros@umich.edu>
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 *
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 * DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
45 #include <linux/string_helpers.h>
46 #include "xdr4.h"
47 #include "xdr4cb.h"
48 #include "vfs.h"
49 #include "current_stateid.h"
50
51 #include "netns.h"
52 #include "pnfs.h"
53 #include "filecache.h"
54
55 #define NFSDDBG_FACILITY NFSDDBG_PROC
56
57 #define all_ones {{~0,~0},~0}
58 static const stateid_t one_stateid = {
59 .si_generation = ~0,
60 .si_opaque = all_ones,
61 };
62 static const stateid_t zero_stateid = {
63 /* all fields zero */
64 };
65 static const stateid_t currentstateid = {
66 .si_generation = 1,
67 };
68 static const stateid_t close_stateid = {
69 .si_generation = 0xffffffffU,
70 };
71
72 static u64 current_sessionid = 1;
73
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
76 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
77 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
78
79 /* forward declarations */
80 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
81 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
82 void nfsd4_end_grace(struct nfsd_net *nn);
83 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
84
85 /* Locking: */
86
87 /*
88 * Currently used for the del_recall_lru and file hash table. In an
89 * effort to decrease the scope of the client_mutex, this spinlock may
90 * eventually cover more:
91 */
92 static DEFINE_SPINLOCK(state_lock);
93
94 enum nfsd4_st_mutex_lock_subclass {
95 OPEN_STATEID_MUTEX = 0,
96 LOCK_STATEID_MUTEX = 1,
97 };
98
99 /*
100 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
101 * the refcount on the open stateid to drop.
102 */
103 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
104
105 /*
106 * A waitqueue where a writer to clients/#/ctl destroying a client can
107 * wait for cl_rpc_users to drop to 0 and then for the client to be
108 * unhashed.
109 */
110 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
111
112 static struct kmem_cache *client_slab;
113 static struct kmem_cache *openowner_slab;
114 static struct kmem_cache *lockowner_slab;
115 static struct kmem_cache *file_slab;
116 static struct kmem_cache *stateid_slab;
117 static struct kmem_cache *deleg_slab;
118 static struct kmem_cache *odstate_slab;
119
120 static void free_session(struct nfsd4_session *);
121
122 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
123 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
124
125 static bool is_session_dead(struct nfsd4_session *ses)
126 {
127 return ses->se_flags & NFS4_SESSION_DEAD;
128 }
129
130 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
131 {
132 if (atomic_read(&ses->se_ref) > ref_held_by_me)
133 return nfserr_jukebox;
134 ses->se_flags |= NFS4_SESSION_DEAD;
135 return nfs_ok;
136 }
137
138 static bool is_client_expired(struct nfs4_client *clp)
139 {
140 return clp->cl_time == 0;
141 }
142
143 static __be32 get_client_locked(struct nfs4_client *clp)
144 {
145 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
146
147 lockdep_assert_held(&nn->client_lock);
148
149 if (is_client_expired(clp))
150 return nfserr_expired;
151 atomic_inc(&clp->cl_rpc_users);
152 return nfs_ok;
153 }
154
155 /* must be called under the client_lock */
156 static inline void
157 renew_client_locked(struct nfs4_client *clp)
158 {
159 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
160
161 if (is_client_expired(clp)) {
162 WARN_ON(1);
163 printk("%s: client (clientid %08x/%08x) already expired\n",
164 __func__,
165 clp->cl_clientid.cl_boot,
166 clp->cl_clientid.cl_id);
167 return;
168 }
169
170 dprintk("renewing client (clientid %08x/%08x)\n",
171 clp->cl_clientid.cl_boot,
172 clp->cl_clientid.cl_id);
173 list_move_tail(&clp->cl_lru, &nn->client_lru);
174 clp->cl_time = ktime_get_boottime_seconds();
175 }
176
177 static void put_client_renew_locked(struct nfs4_client *clp)
178 {
179 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
180
181 lockdep_assert_held(&nn->client_lock);
182
183 if (!atomic_dec_and_test(&clp->cl_rpc_users))
184 return;
185 if (!is_client_expired(clp))
186 renew_client_locked(clp);
187 else
188 wake_up_all(&expiry_wq);
189 }
190
191 static void put_client_renew(struct nfs4_client *clp)
192 {
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
194
195 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
196 return;
197 if (!is_client_expired(clp))
198 renew_client_locked(clp);
199 else
200 wake_up_all(&expiry_wq);
201 spin_unlock(&nn->client_lock);
202 }
203
204 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
205 {
206 __be32 status;
207
208 if (is_session_dead(ses))
209 return nfserr_badsession;
210 status = get_client_locked(ses->se_client);
211 if (status)
212 return status;
213 atomic_inc(&ses->se_ref);
214 return nfs_ok;
215 }
216
217 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
218 {
219 struct nfs4_client *clp = ses->se_client;
220 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
221
222 lockdep_assert_held(&nn->client_lock);
223
224 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
225 free_session(ses);
226 put_client_renew_locked(clp);
227 }
228
229 static void nfsd4_put_session(struct nfsd4_session *ses)
230 {
231 struct nfs4_client *clp = ses->se_client;
232 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
233
234 spin_lock(&nn->client_lock);
235 nfsd4_put_session_locked(ses);
236 spin_unlock(&nn->client_lock);
237 }
238
239 static struct nfsd4_blocked_lock *
240 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
241 struct nfsd_net *nn)
242 {
243 struct nfsd4_blocked_lock *cur, *found = NULL;
244
245 spin_lock(&nn->blocked_locks_lock);
246 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
247 if (fh_match(fh, &cur->nbl_fh)) {
248 list_del_init(&cur->nbl_list);
249 list_del_init(&cur->nbl_lru);
250 found = cur;
251 break;
252 }
253 }
254 spin_unlock(&nn->blocked_locks_lock);
255 if (found)
256 locks_delete_block(&found->nbl_lock);
257 return found;
258 }
259
260 static struct nfsd4_blocked_lock *
261 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
262 struct nfsd_net *nn)
263 {
264 struct nfsd4_blocked_lock *nbl;
265
266 nbl = find_blocked_lock(lo, fh, nn);
267 if (!nbl) {
268 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
269 if (nbl) {
270 fh_copy_shallow(&nbl->nbl_fh, fh);
271 locks_init_lock(&nbl->nbl_lock);
272 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
273 &nfsd4_cb_notify_lock_ops,
274 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
275 }
276 }
277 return nbl;
278 }
279
280 static void
281 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
282 {
283 locks_delete_block(&nbl->nbl_lock);
284 locks_release_private(&nbl->nbl_lock);
285 kfree(nbl);
286 }
287
288 static void
289 remove_blocked_locks(struct nfs4_lockowner *lo)
290 {
291 struct nfs4_client *clp = lo->lo_owner.so_client;
292 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
293 struct nfsd4_blocked_lock *nbl;
294 LIST_HEAD(reaplist);
295
296 /* Dequeue all blocked locks */
297 spin_lock(&nn->blocked_locks_lock);
298 while (!list_empty(&lo->lo_blocked)) {
299 nbl = list_first_entry(&lo->lo_blocked,
300 struct nfsd4_blocked_lock,
301 nbl_list);
302 list_del_init(&nbl->nbl_list);
303 list_move(&nbl->nbl_lru, &reaplist);
304 }
305 spin_unlock(&nn->blocked_locks_lock);
306
307 /* Now free them */
308 while (!list_empty(&reaplist)) {
309 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
310 nbl_lru);
311 list_del_init(&nbl->nbl_lru);
312 free_blocked_lock(nbl);
313 }
314 }
315
316 static void
317 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
318 {
319 struct nfsd4_blocked_lock *nbl = container_of(cb,
320 struct nfsd4_blocked_lock, nbl_cb);
321 locks_delete_block(&nbl->nbl_lock);
322 }
323
324 static int
325 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
326 {
327 /*
328 * Since this is just an optimization, we don't try very hard if it
329 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
330 * just quit trying on anything else.
331 */
332 switch (task->tk_status) {
333 case -NFS4ERR_DELAY:
334 rpc_delay(task, 1 * HZ);
335 return 0;
336 default:
337 return 1;
338 }
339 }
340
341 static void
342 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
343 {
344 struct nfsd4_blocked_lock *nbl = container_of(cb,
345 struct nfsd4_blocked_lock, nbl_cb);
346
347 free_blocked_lock(nbl);
348 }
349
350 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
351 .prepare = nfsd4_cb_notify_lock_prepare,
352 .done = nfsd4_cb_notify_lock_done,
353 .release = nfsd4_cb_notify_lock_release,
354 };
355
356 static inline struct nfs4_stateowner *
357 nfs4_get_stateowner(struct nfs4_stateowner *sop)
358 {
359 atomic_inc(&sop->so_count);
360 return sop;
361 }
362
363 static int
364 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
365 {
366 return (sop->so_owner.len == owner->len) &&
367 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
368 }
369
370 static struct nfs4_openowner *
371 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
372 struct nfs4_client *clp)
373 {
374 struct nfs4_stateowner *so;
375
376 lockdep_assert_held(&clp->cl_lock);
377
378 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
379 so_strhash) {
380 if (!so->so_is_open_owner)
381 continue;
382 if (same_owner_str(so, &open->op_owner))
383 return openowner(nfs4_get_stateowner(so));
384 }
385 return NULL;
386 }
387
388 static struct nfs4_openowner *
389 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
390 struct nfs4_client *clp)
391 {
392 struct nfs4_openowner *oo;
393
394 spin_lock(&clp->cl_lock);
395 oo = find_openstateowner_str_locked(hashval, open, clp);
396 spin_unlock(&clp->cl_lock);
397 return oo;
398 }
399
400 static inline u32
401 opaque_hashval(const void *ptr, int nbytes)
402 {
403 unsigned char *cptr = (unsigned char *) ptr;
404
405 u32 x = 0;
406 while (nbytes--) {
407 x *= 37;
408 x += *cptr++;
409 }
410 return x;
411 }
412
413 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
414 {
415 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
416
417 kmem_cache_free(file_slab, fp);
418 }
419
420 void
421 put_nfs4_file(struct nfs4_file *fi)
422 {
423 might_lock(&state_lock);
424
425 if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
426 hlist_del_rcu(&fi->fi_hash);
427 spin_unlock(&state_lock);
428 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
429 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
430 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
431 }
432 }
433
434 static struct nfsd_file *
435 __nfs4_get_fd(struct nfs4_file *f, int oflag)
436 {
437 if (f->fi_fds[oflag])
438 return nfsd_file_get(f->fi_fds[oflag]);
439 return NULL;
440 }
441
442 static struct nfsd_file *
443 find_writeable_file_locked(struct nfs4_file *f)
444 {
445 struct nfsd_file *ret;
446
447 lockdep_assert_held(&f->fi_lock);
448
449 ret = __nfs4_get_fd(f, O_WRONLY);
450 if (!ret)
451 ret = __nfs4_get_fd(f, O_RDWR);
452 return ret;
453 }
454
455 static struct nfsd_file *
456 find_writeable_file(struct nfs4_file *f)
457 {
458 struct nfsd_file *ret;
459
460 spin_lock(&f->fi_lock);
461 ret = find_writeable_file_locked(f);
462 spin_unlock(&f->fi_lock);
463
464 return ret;
465 }
466
467 static struct nfsd_file *
468 find_readable_file_locked(struct nfs4_file *f)
469 {
470 struct nfsd_file *ret;
471
472 lockdep_assert_held(&f->fi_lock);
473
474 ret = __nfs4_get_fd(f, O_RDONLY);
475 if (!ret)
476 ret = __nfs4_get_fd(f, O_RDWR);
477 return ret;
478 }
479
480 static struct nfsd_file *
481 find_readable_file(struct nfs4_file *f)
482 {
483 struct nfsd_file *ret;
484
485 spin_lock(&f->fi_lock);
486 ret = find_readable_file_locked(f);
487 spin_unlock(&f->fi_lock);
488
489 return ret;
490 }
491
492 struct nfsd_file *
493 find_any_file(struct nfs4_file *f)
494 {
495 struct nfsd_file *ret;
496
497 if (!f)
498 return NULL;
499 spin_lock(&f->fi_lock);
500 ret = __nfs4_get_fd(f, O_RDWR);
501 if (!ret) {
502 ret = __nfs4_get_fd(f, O_WRONLY);
503 if (!ret)
504 ret = __nfs4_get_fd(f, O_RDONLY);
505 }
506 spin_unlock(&f->fi_lock);
507 return ret;
508 }
509
510 static atomic_long_t num_delegations;
511 unsigned long max_delegations;
512
513 /*
514 * Open owner state (share locks)
515 */
516
517 /* hash tables for lock and open owners */
518 #define OWNER_HASH_BITS 8
519 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
520 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
521
522 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
523 {
524 unsigned int ret;
525
526 ret = opaque_hashval(ownername->data, ownername->len);
527 return ret & OWNER_HASH_MASK;
528 }
529
530 /* hash table for nfs4_file */
531 #define FILE_HASH_BITS 8
532 #define FILE_HASH_SIZE (1 << FILE_HASH_BITS)
533
534 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
535 {
536 return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
537 }
538
539 static unsigned int file_hashval(struct knfsd_fh *fh)
540 {
541 return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
542 }
543
544 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
545
546 static void
547 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
548 {
549 lockdep_assert_held(&fp->fi_lock);
550
551 if (access & NFS4_SHARE_ACCESS_WRITE)
552 atomic_inc(&fp->fi_access[O_WRONLY]);
553 if (access & NFS4_SHARE_ACCESS_READ)
554 atomic_inc(&fp->fi_access[O_RDONLY]);
555 }
556
557 static __be32
558 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
559 {
560 lockdep_assert_held(&fp->fi_lock);
561
562 /* Does this access mode make sense? */
563 if (access & ~NFS4_SHARE_ACCESS_BOTH)
564 return nfserr_inval;
565
566 /* Does it conflict with a deny mode already set? */
567 if ((access & fp->fi_share_deny) != 0)
568 return nfserr_share_denied;
569
570 __nfs4_file_get_access(fp, access);
571 return nfs_ok;
572 }
573
574 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
575 {
576 /* Common case is that there is no deny mode. */
577 if (deny) {
578 /* Does this deny mode make sense? */
579 if (deny & ~NFS4_SHARE_DENY_BOTH)
580 return nfserr_inval;
581
582 if ((deny & NFS4_SHARE_DENY_READ) &&
583 atomic_read(&fp->fi_access[O_RDONLY]))
584 return nfserr_share_denied;
585
586 if ((deny & NFS4_SHARE_DENY_WRITE) &&
587 atomic_read(&fp->fi_access[O_WRONLY]))
588 return nfserr_share_denied;
589 }
590 return nfs_ok;
591 }
592
593 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
594 {
595 might_lock(&fp->fi_lock);
596
597 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
598 struct nfsd_file *f1 = NULL;
599 struct nfsd_file *f2 = NULL;
600
601 swap(f1, fp->fi_fds[oflag]);
602 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
603 swap(f2, fp->fi_fds[O_RDWR]);
604 spin_unlock(&fp->fi_lock);
605 if (f1)
606 nfsd_file_put(f1);
607 if (f2)
608 nfsd_file_put(f2);
609 }
610 }
611
612 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
613 {
614 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
615
616 if (access & NFS4_SHARE_ACCESS_WRITE)
617 __nfs4_file_put_access(fp, O_WRONLY);
618 if (access & NFS4_SHARE_ACCESS_READ)
619 __nfs4_file_put_access(fp, O_RDONLY);
620 }
621
622 /*
623 * Allocate a new open/delegation state counter. This is needed for
624 * pNFS for proper return on close semantics.
625 *
626 * Note that we only allocate it for pNFS-enabled exports, otherwise
627 * all pointers to struct nfs4_clnt_odstate are always NULL.
628 */
629 static struct nfs4_clnt_odstate *
630 alloc_clnt_odstate(struct nfs4_client *clp)
631 {
632 struct nfs4_clnt_odstate *co;
633
634 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
635 if (co) {
636 co->co_client = clp;
637 refcount_set(&co->co_odcount, 1);
638 }
639 return co;
640 }
641
642 static void
643 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
644 {
645 struct nfs4_file *fp = co->co_file;
646
647 lockdep_assert_held(&fp->fi_lock);
648 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
649 }
650
651 static inline void
652 get_clnt_odstate(struct nfs4_clnt_odstate *co)
653 {
654 if (co)
655 refcount_inc(&co->co_odcount);
656 }
657
658 static void
659 put_clnt_odstate(struct nfs4_clnt_odstate *co)
660 {
661 struct nfs4_file *fp;
662
663 if (!co)
664 return;
665
666 fp = co->co_file;
667 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
668 list_del(&co->co_perfile);
669 spin_unlock(&fp->fi_lock);
670
671 nfsd4_return_all_file_layouts(co->co_client, fp);
672 kmem_cache_free(odstate_slab, co);
673 }
674 }
675
676 static struct nfs4_clnt_odstate *
677 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
678 {
679 struct nfs4_clnt_odstate *co;
680 struct nfs4_client *cl;
681
682 if (!new)
683 return NULL;
684
685 cl = new->co_client;
686
687 spin_lock(&fp->fi_lock);
688 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
689 if (co->co_client == cl) {
690 get_clnt_odstate(co);
691 goto out;
692 }
693 }
694 co = new;
695 co->co_file = fp;
696 hash_clnt_odstate_locked(new);
697 out:
698 spin_unlock(&fp->fi_lock);
699 return co;
700 }
701
702 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
703 void (*sc_free)(struct nfs4_stid *))
704 {
705 struct nfs4_stid *stid;
706 int new_id;
707
708 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
709 if (!stid)
710 return NULL;
711
712 idr_preload(GFP_KERNEL);
713 spin_lock(&cl->cl_lock);
714 /* Reserving 0 for start of file in nfsdfs "states" file: */
715 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
716 spin_unlock(&cl->cl_lock);
717 idr_preload_end();
718 if (new_id < 0)
719 goto out_free;
720
721 stid->sc_free = sc_free;
722 stid->sc_client = cl;
723 stid->sc_stateid.si_opaque.so_id = new_id;
724 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
725 /* Will be incremented before return to client: */
726 refcount_set(&stid->sc_count, 1);
727 spin_lock_init(&stid->sc_lock);
728 INIT_LIST_HEAD(&stid->sc_cp_list);
729
730 /*
731 * It shouldn't be a problem to reuse an opaque stateid value.
732 * I don't think it is for 4.1. But with 4.0 I worry that, for
733 * example, a stray write retransmission could be accepted by
734 * the server when it should have been rejected. Therefore,
735 * adopt a trick from the sctp code to attempt to maximize the
736 * amount of time until an id is reused, by ensuring they always
737 * "increase" (mod INT_MAX):
738 */
739 return stid;
740 out_free:
741 kmem_cache_free(slab, stid);
742 return NULL;
743 }
744
745 /*
746 * Create a unique stateid_t to represent each COPY.
747 */
748 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
749 unsigned char sc_type)
750 {
751 int new_id;
752
753 stid->stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
754 stid->stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
755 stid->sc_type = sc_type;
756
757 idr_preload(GFP_KERNEL);
758 spin_lock(&nn->s2s_cp_lock);
759 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
760 stid->stid.si_opaque.so_id = new_id;
761 spin_unlock(&nn->s2s_cp_lock);
762 idr_preload_end();
763 if (new_id < 0)
764 return 0;
765 return 1;
766 }
767
768 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
769 {
770 return nfs4_init_cp_state(nn, &copy->cp_stateid, NFS4_COPY_STID);
771 }
772
773 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
774 struct nfs4_stid *p_stid)
775 {
776 struct nfs4_cpntf_state *cps;
777
778 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
779 if (!cps)
780 return NULL;
781 cps->cpntf_time = ktime_get_boottime_seconds();
782 refcount_set(&cps->cp_stateid.sc_count, 1);
783 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
784 goto out_free;
785 spin_lock(&nn->s2s_cp_lock);
786 list_add(&cps->cp_list, &p_stid->sc_cp_list);
787 spin_unlock(&nn->s2s_cp_lock);
788 return cps;
789 out_free:
790 kfree(cps);
791 return NULL;
792 }
793
794 void nfs4_free_copy_state(struct nfsd4_copy *copy)
795 {
796 struct nfsd_net *nn;
797
798 WARN_ON_ONCE(copy->cp_stateid.sc_type != NFS4_COPY_STID);
799 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
800 spin_lock(&nn->s2s_cp_lock);
801 idr_remove(&nn->s2s_cp_stateids,
802 copy->cp_stateid.stid.si_opaque.so_id);
803 spin_unlock(&nn->s2s_cp_lock);
804 }
805
806 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
807 {
808 struct nfs4_cpntf_state *cps;
809 struct nfsd_net *nn;
810
811 nn = net_generic(net, nfsd_net_id);
812 spin_lock(&nn->s2s_cp_lock);
813 while (!list_empty(&stid->sc_cp_list)) {
814 cps = list_first_entry(&stid->sc_cp_list,
815 struct nfs4_cpntf_state, cp_list);
816 _free_cpntf_state_locked(nn, cps);
817 }
818 spin_unlock(&nn->s2s_cp_lock);
819 }
820
821 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
822 {
823 struct nfs4_stid *stid;
824
825 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
826 if (!stid)
827 return NULL;
828
829 return openlockstateid(stid);
830 }
831
832 static void nfs4_free_deleg(struct nfs4_stid *stid)
833 {
834 kmem_cache_free(deleg_slab, stid);
835 atomic_long_dec(&num_delegations);
836 }
837
838 /*
839 * When we recall a delegation, we should be careful not to hand it
840 * out again straight away.
841 * To ensure this we keep a pair of bloom filters ('new' and 'old')
842 * in which the filehandles of recalled delegations are "stored".
843 * If a filehandle appear in either filter, a delegation is blocked.
844 * When a delegation is recalled, the filehandle is stored in the "new"
845 * filter.
846 * Every 30 seconds we swap the filters and clear the "new" one,
847 * unless both are empty of course.
848 *
849 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
850 * low 3 bytes as hash-table indices.
851 *
852 * 'blocked_delegations_lock', which is always taken in block_delegations(),
853 * is used to manage concurrent access. Testing does not need the lock
854 * except when swapping the two filters.
855 */
856 static DEFINE_SPINLOCK(blocked_delegations_lock);
857 static struct bloom_pair {
858 int entries, old_entries;
859 time64_t swap_time;
860 int new; /* index into 'set' */
861 DECLARE_BITMAP(set[2], 256);
862 } blocked_delegations;
863
864 static int delegation_blocked(struct knfsd_fh *fh)
865 {
866 u32 hash;
867 struct bloom_pair *bd = &blocked_delegations;
868
869 if (bd->entries == 0)
870 return 0;
871 if (ktime_get_seconds() - bd->swap_time > 30) {
872 spin_lock(&blocked_delegations_lock);
873 if (ktime_get_seconds() - bd->swap_time > 30) {
874 bd->entries -= bd->old_entries;
875 bd->old_entries = bd->entries;
876 memset(bd->set[bd->new], 0,
877 sizeof(bd->set[0]));
878 bd->new = 1-bd->new;
879 bd->swap_time = ktime_get_seconds();
880 }
881 spin_unlock(&blocked_delegations_lock);
882 }
883 hash = jhash(&fh->fh_base, fh->fh_size, 0);
884 if (test_bit(hash&255, bd->set[0]) &&
885 test_bit((hash>>8)&255, bd->set[0]) &&
886 test_bit((hash>>16)&255, bd->set[0]))
887 return 1;
888
889 if (test_bit(hash&255, bd->set[1]) &&
890 test_bit((hash>>8)&255, bd->set[1]) &&
891 test_bit((hash>>16)&255, bd->set[1]))
892 return 1;
893
894 return 0;
895 }
896
897 static void block_delegations(struct knfsd_fh *fh)
898 {
899 u32 hash;
900 struct bloom_pair *bd = &blocked_delegations;
901
902 hash = jhash(&fh->fh_base, fh->fh_size, 0);
903
904 spin_lock(&blocked_delegations_lock);
905 __set_bit(hash&255, bd->set[bd->new]);
906 __set_bit((hash>>8)&255, bd->set[bd->new]);
907 __set_bit((hash>>16)&255, bd->set[bd->new]);
908 if (bd->entries == 0)
909 bd->swap_time = ktime_get_seconds();
910 bd->entries += 1;
911 spin_unlock(&blocked_delegations_lock);
912 }
913
914 static struct nfs4_delegation *
915 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
916 struct svc_fh *current_fh,
917 struct nfs4_clnt_odstate *odstate)
918 {
919 struct nfs4_delegation *dp;
920 long n;
921
922 dprintk("NFSD alloc_init_deleg\n");
923 n = atomic_long_inc_return(&num_delegations);
924 if (n < 0 || n > max_delegations)
925 goto out_dec;
926 if (delegation_blocked(&current_fh->fh_handle))
927 goto out_dec;
928 dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
929 if (dp == NULL)
930 goto out_dec;
931
932 /*
933 * delegation seqid's are never incremented. The 4.1 special
934 * meaning of seqid 0 isn't meaningful, really, but let's avoid
935 * 0 anyway just for consistency and use 1:
936 */
937 dp->dl_stid.sc_stateid.si_generation = 1;
938 INIT_LIST_HEAD(&dp->dl_perfile);
939 INIT_LIST_HEAD(&dp->dl_perclnt);
940 INIT_LIST_HEAD(&dp->dl_recall_lru);
941 dp->dl_clnt_odstate = odstate;
942 get_clnt_odstate(odstate);
943 dp->dl_type = NFS4_OPEN_DELEGATE_READ;
944 dp->dl_retries = 1;
945 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
946 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
947 get_nfs4_file(fp);
948 dp->dl_stid.sc_file = fp;
949 return dp;
950 out_dec:
951 atomic_long_dec(&num_delegations);
952 return NULL;
953 }
954
955 void
956 nfs4_put_stid(struct nfs4_stid *s)
957 {
958 struct nfs4_file *fp = s->sc_file;
959 struct nfs4_client *clp = s->sc_client;
960
961 might_lock(&clp->cl_lock);
962
963 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
964 wake_up_all(&close_wq);
965 return;
966 }
967 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
968 nfs4_free_cpntf_statelist(clp->net, s);
969 spin_unlock(&clp->cl_lock);
970 s->sc_free(s);
971 if (fp)
972 put_nfs4_file(fp);
973 }
974
975 void
976 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
977 {
978 stateid_t *src = &stid->sc_stateid;
979
980 spin_lock(&stid->sc_lock);
981 if (unlikely(++src->si_generation == 0))
982 src->si_generation = 1;
983 memcpy(dst, src, sizeof(*dst));
984 spin_unlock(&stid->sc_lock);
985 }
986
987 static void put_deleg_file(struct nfs4_file *fp)
988 {
989 struct nfsd_file *nf = NULL;
990
991 spin_lock(&fp->fi_lock);
992 if (--fp->fi_delegees == 0)
993 swap(nf, fp->fi_deleg_file);
994 spin_unlock(&fp->fi_lock);
995
996 if (nf)
997 nfsd_file_put(nf);
998 }
999
1000 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1001 {
1002 struct nfs4_file *fp = dp->dl_stid.sc_file;
1003 struct nfsd_file *nf = fp->fi_deleg_file;
1004
1005 WARN_ON_ONCE(!fp->fi_delegees);
1006
1007 vfs_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1008 put_deleg_file(fp);
1009 }
1010
1011 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1012 {
1013 put_clnt_odstate(dp->dl_clnt_odstate);
1014 nfs4_unlock_deleg_lease(dp);
1015 nfs4_put_stid(&dp->dl_stid);
1016 }
1017
1018 void nfs4_unhash_stid(struct nfs4_stid *s)
1019 {
1020 s->sc_type = 0;
1021 }
1022
1023 /**
1024 * nfs4_delegation_exists - Discover if this delegation already exists
1025 * @clp: a pointer to the nfs4_client we're granting a delegation to
1026 * @fp: a pointer to the nfs4_file we're granting a delegation on
1027 *
1028 * Return:
1029 * On success: true iff an existing delegation is found
1030 */
1031
1032 static bool
1033 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1034 {
1035 struct nfs4_delegation *searchdp = NULL;
1036 struct nfs4_client *searchclp = NULL;
1037
1038 lockdep_assert_held(&state_lock);
1039 lockdep_assert_held(&fp->fi_lock);
1040
1041 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1042 searchclp = searchdp->dl_stid.sc_client;
1043 if (clp == searchclp) {
1044 return true;
1045 }
1046 }
1047 return false;
1048 }
1049
1050 /**
1051 * hash_delegation_locked - Add a delegation to the appropriate lists
1052 * @dp: a pointer to the nfs4_delegation we are adding.
1053 * @fp: a pointer to the nfs4_file we're granting a delegation on
1054 *
1055 * Return:
1056 * On success: NULL if the delegation was successfully hashed.
1057 *
1058 * On error: -EAGAIN if one was previously granted to this
1059 * nfs4_client for this nfs4_file. Delegation is not hashed.
1060 *
1061 */
1062
1063 static int
1064 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1065 {
1066 struct nfs4_client *clp = dp->dl_stid.sc_client;
1067
1068 lockdep_assert_held(&state_lock);
1069 lockdep_assert_held(&fp->fi_lock);
1070
1071 if (nfs4_delegation_exists(clp, fp))
1072 return -EAGAIN;
1073 refcount_inc(&dp->dl_stid.sc_count);
1074 dp->dl_stid.sc_type = NFS4_DELEG_STID;
1075 list_add(&dp->dl_perfile, &fp->fi_delegations);
1076 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1077 return 0;
1078 }
1079
1080 static bool
1081 unhash_delegation_locked(struct nfs4_delegation *dp)
1082 {
1083 struct nfs4_file *fp = dp->dl_stid.sc_file;
1084
1085 lockdep_assert_held(&state_lock);
1086
1087 if (list_empty(&dp->dl_perfile))
1088 return false;
1089
1090 dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
1091 /* Ensure that deleg break won't try to requeue it */
1092 ++dp->dl_time;
1093 spin_lock(&fp->fi_lock);
1094 list_del_init(&dp->dl_perclnt);
1095 list_del_init(&dp->dl_recall_lru);
1096 list_del_init(&dp->dl_perfile);
1097 spin_unlock(&fp->fi_lock);
1098 return true;
1099 }
1100
1101 static void destroy_delegation(struct nfs4_delegation *dp)
1102 {
1103 bool unhashed;
1104
1105 spin_lock(&state_lock);
1106 unhashed = unhash_delegation_locked(dp);
1107 spin_unlock(&state_lock);
1108 if (unhashed)
1109 destroy_unhashed_deleg(dp);
1110 }
1111
1112 static void revoke_delegation(struct nfs4_delegation *dp)
1113 {
1114 struct nfs4_client *clp = dp->dl_stid.sc_client;
1115
1116 WARN_ON(!list_empty(&dp->dl_recall_lru));
1117
1118 if (clp->cl_minorversion) {
1119 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1120 refcount_inc(&dp->dl_stid.sc_count);
1121 spin_lock(&clp->cl_lock);
1122 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1123 spin_unlock(&clp->cl_lock);
1124 }
1125 destroy_unhashed_deleg(dp);
1126 }
1127
1128 /*
1129 * SETCLIENTID state
1130 */
1131
1132 static unsigned int clientid_hashval(u32 id)
1133 {
1134 return id & CLIENT_HASH_MASK;
1135 }
1136
1137 static unsigned int clientstr_hashval(struct xdr_netobj name)
1138 {
1139 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1140 }
1141
1142 /*
1143 * We store the NONE, READ, WRITE, and BOTH bits separately in the
1144 * st_{access,deny}_bmap field of the stateid, in order to track not
1145 * only what share bits are currently in force, but also what
1146 * combinations of share bits previous opens have used. This allows us
1147 * to enforce the recommendation of rfc 3530 14.2.19 that the server
1148 * return an error if the client attempt to downgrade to a combination
1149 * of share bits not explicable by closing some of its previous opens.
1150 *
1151 * XXX: This enforcement is actually incomplete, since we don't keep
1152 * track of access/deny bit combinations; so, e.g., we allow:
1153 *
1154 * OPEN allow read, deny write
1155 * OPEN allow both, deny none
1156 * DOWNGRADE allow read, deny none
1157 *
1158 * which we should reject.
1159 */
1160 static unsigned int
1161 bmap_to_share_mode(unsigned long bmap) {
1162 int i;
1163 unsigned int access = 0;
1164
1165 for (i = 1; i < 4; i++) {
1166 if (test_bit(i, &bmap))
1167 access |= i;
1168 }
1169 return access;
1170 }
1171
1172 /* set share access for a given stateid */
1173 static inline void
1174 set_access(u32 access, struct nfs4_ol_stateid *stp)
1175 {
1176 unsigned char mask = 1 << access;
1177
1178 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1179 stp->st_access_bmap |= mask;
1180 }
1181
1182 /* clear share access for a given stateid */
1183 static inline void
1184 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1185 {
1186 unsigned char mask = 1 << access;
1187
1188 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1189 stp->st_access_bmap &= ~mask;
1190 }
1191
1192 /* test whether a given stateid has access */
1193 static inline bool
1194 test_access(u32 access, struct nfs4_ol_stateid *stp)
1195 {
1196 unsigned char mask = 1 << access;
1197
1198 return (bool)(stp->st_access_bmap & mask);
1199 }
1200
1201 /* set share deny for a given stateid */
1202 static inline void
1203 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1204 {
1205 unsigned char mask = 1 << deny;
1206
1207 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1208 stp->st_deny_bmap |= mask;
1209 }
1210
1211 /* clear share deny for a given stateid */
1212 static inline void
1213 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1214 {
1215 unsigned char mask = 1 << deny;
1216
1217 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1218 stp->st_deny_bmap &= ~mask;
1219 }
1220
1221 /* test whether a given stateid is denying specific access */
1222 static inline bool
1223 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1224 {
1225 unsigned char mask = 1 << deny;
1226
1227 return (bool)(stp->st_deny_bmap & mask);
1228 }
1229
1230 static int nfs4_access_to_omode(u32 access)
1231 {
1232 switch (access & NFS4_SHARE_ACCESS_BOTH) {
1233 case NFS4_SHARE_ACCESS_READ:
1234 return O_RDONLY;
1235 case NFS4_SHARE_ACCESS_WRITE:
1236 return O_WRONLY;
1237 case NFS4_SHARE_ACCESS_BOTH:
1238 return O_RDWR;
1239 }
1240 WARN_ON_ONCE(1);
1241 return O_RDONLY;
1242 }
1243
1244 /*
1245 * A stateid that had a deny mode associated with it is being released
1246 * or downgraded. Recalculate the deny mode on the file.
1247 */
1248 static void
1249 recalculate_deny_mode(struct nfs4_file *fp)
1250 {
1251 struct nfs4_ol_stateid *stp;
1252
1253 spin_lock(&fp->fi_lock);
1254 fp->fi_share_deny = 0;
1255 list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1256 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1257 spin_unlock(&fp->fi_lock);
1258 }
1259
1260 static void
1261 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1262 {
1263 int i;
1264 bool change = false;
1265
1266 for (i = 1; i < 4; i++) {
1267 if ((i & deny) != i) {
1268 change = true;
1269 clear_deny(i, stp);
1270 }
1271 }
1272
1273 /* Recalculate per-file deny mode if there was a change */
1274 if (change)
1275 recalculate_deny_mode(stp->st_stid.sc_file);
1276 }
1277
1278 /* release all access and file references for a given stateid */
1279 static void
1280 release_all_access(struct nfs4_ol_stateid *stp)
1281 {
1282 int i;
1283 struct nfs4_file *fp = stp->st_stid.sc_file;
1284
1285 if (fp && stp->st_deny_bmap != 0)
1286 recalculate_deny_mode(fp);
1287
1288 for (i = 1; i < 4; i++) {
1289 if (test_access(i, stp))
1290 nfs4_file_put_access(stp->st_stid.sc_file, i);
1291 clear_access(i, stp);
1292 }
1293 }
1294
1295 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1296 {
1297 kfree(sop->so_owner.data);
1298 sop->so_ops->so_free(sop);
1299 }
1300
1301 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1302 {
1303 struct nfs4_client *clp = sop->so_client;
1304
1305 might_lock(&clp->cl_lock);
1306
1307 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1308 return;
1309 sop->so_ops->so_unhash(sop);
1310 spin_unlock(&clp->cl_lock);
1311 nfs4_free_stateowner(sop);
1312 }
1313
1314 static bool
1315 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1316 {
1317 return list_empty(&stp->st_perfile);
1318 }
1319
1320 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1321 {
1322 struct nfs4_file *fp = stp->st_stid.sc_file;
1323
1324 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1325
1326 if (list_empty(&stp->st_perfile))
1327 return false;
1328
1329 spin_lock(&fp->fi_lock);
1330 list_del_init(&stp->st_perfile);
1331 spin_unlock(&fp->fi_lock);
1332 list_del(&stp->st_perstateowner);
1333 return true;
1334 }
1335
1336 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1337 {
1338 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1339
1340 put_clnt_odstate(stp->st_clnt_odstate);
1341 release_all_access(stp);
1342 if (stp->st_stateowner)
1343 nfs4_put_stateowner(stp->st_stateowner);
1344 kmem_cache_free(stateid_slab, stid);
1345 }
1346
1347 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1348 {
1349 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1350 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1351 struct nfsd_file *nf;
1352
1353 nf = find_any_file(stp->st_stid.sc_file);
1354 if (nf) {
1355 get_file(nf->nf_file);
1356 filp_close(nf->nf_file, (fl_owner_t)lo);
1357 nfsd_file_put(nf);
1358 }
1359 nfs4_free_ol_stateid(stid);
1360 }
1361
1362 /*
1363 * Put the persistent reference to an already unhashed generic stateid, while
1364 * holding the cl_lock. If it's the last reference, then put it onto the
1365 * reaplist for later destruction.
1366 */
1367 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1368 struct list_head *reaplist)
1369 {
1370 struct nfs4_stid *s = &stp->st_stid;
1371 struct nfs4_client *clp = s->sc_client;
1372
1373 lockdep_assert_held(&clp->cl_lock);
1374
1375 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1376
1377 if (!refcount_dec_and_test(&s->sc_count)) {
1378 wake_up_all(&close_wq);
1379 return;
1380 }
1381
1382 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1383 list_add(&stp->st_locks, reaplist);
1384 }
1385
1386 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1387 {
1388 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1389
1390 if (!unhash_ol_stateid(stp))
1391 return false;
1392 list_del_init(&stp->st_locks);
1393 nfs4_unhash_stid(&stp->st_stid);
1394 return true;
1395 }
1396
1397 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1398 {
1399 struct nfs4_client *clp = stp->st_stid.sc_client;
1400 bool unhashed;
1401
1402 spin_lock(&clp->cl_lock);
1403 unhashed = unhash_lock_stateid(stp);
1404 spin_unlock(&clp->cl_lock);
1405 if (unhashed)
1406 nfs4_put_stid(&stp->st_stid);
1407 }
1408
1409 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1410 {
1411 struct nfs4_client *clp = lo->lo_owner.so_client;
1412
1413 lockdep_assert_held(&clp->cl_lock);
1414
1415 list_del_init(&lo->lo_owner.so_strhash);
1416 }
1417
1418 /*
1419 * Free a list of generic stateids that were collected earlier after being
1420 * fully unhashed.
1421 */
1422 static void
1423 free_ol_stateid_reaplist(struct list_head *reaplist)
1424 {
1425 struct nfs4_ol_stateid *stp;
1426 struct nfs4_file *fp;
1427
1428 might_sleep();
1429
1430 while (!list_empty(reaplist)) {
1431 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1432 st_locks);
1433 list_del(&stp->st_locks);
1434 fp = stp->st_stid.sc_file;
1435 stp->st_stid.sc_free(&stp->st_stid);
1436 if (fp)
1437 put_nfs4_file(fp);
1438 }
1439 }
1440
1441 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1442 struct list_head *reaplist)
1443 {
1444 struct nfs4_ol_stateid *stp;
1445
1446 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1447
1448 while (!list_empty(&open_stp->st_locks)) {
1449 stp = list_entry(open_stp->st_locks.next,
1450 struct nfs4_ol_stateid, st_locks);
1451 WARN_ON(!unhash_lock_stateid(stp));
1452 put_ol_stateid_locked(stp, reaplist);
1453 }
1454 }
1455
1456 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1457 struct list_head *reaplist)
1458 {
1459 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1460
1461 if (!unhash_ol_stateid(stp))
1462 return false;
1463 release_open_stateid_locks(stp, reaplist);
1464 return true;
1465 }
1466
1467 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1468 {
1469 LIST_HEAD(reaplist);
1470
1471 spin_lock(&stp->st_stid.sc_client->cl_lock);
1472 if (unhash_open_stateid(stp, &reaplist))
1473 put_ol_stateid_locked(stp, &reaplist);
1474 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1475 free_ol_stateid_reaplist(&reaplist);
1476 }
1477
1478 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1479 {
1480 struct nfs4_client *clp = oo->oo_owner.so_client;
1481
1482 lockdep_assert_held(&clp->cl_lock);
1483
1484 list_del_init(&oo->oo_owner.so_strhash);
1485 list_del_init(&oo->oo_perclient);
1486 }
1487
1488 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1489 {
1490 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1491 nfsd_net_id);
1492 struct nfs4_ol_stateid *s;
1493
1494 spin_lock(&nn->client_lock);
1495 s = oo->oo_last_closed_stid;
1496 if (s) {
1497 list_del_init(&oo->oo_close_lru);
1498 oo->oo_last_closed_stid = NULL;
1499 }
1500 spin_unlock(&nn->client_lock);
1501 if (s)
1502 nfs4_put_stid(&s->st_stid);
1503 }
1504
1505 static void release_openowner(struct nfs4_openowner *oo)
1506 {
1507 struct nfs4_ol_stateid *stp;
1508 struct nfs4_client *clp = oo->oo_owner.so_client;
1509 struct list_head reaplist;
1510
1511 INIT_LIST_HEAD(&reaplist);
1512
1513 spin_lock(&clp->cl_lock);
1514 unhash_openowner_locked(oo);
1515 while (!list_empty(&oo->oo_owner.so_stateids)) {
1516 stp = list_first_entry(&oo->oo_owner.so_stateids,
1517 struct nfs4_ol_stateid, st_perstateowner);
1518 if (unhash_open_stateid(stp, &reaplist))
1519 put_ol_stateid_locked(stp, &reaplist);
1520 }
1521 spin_unlock(&clp->cl_lock);
1522 free_ol_stateid_reaplist(&reaplist);
1523 release_last_closed_stateid(oo);
1524 nfs4_put_stateowner(&oo->oo_owner);
1525 }
1526
1527 static inline int
1528 hash_sessionid(struct nfs4_sessionid *sessionid)
1529 {
1530 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1531
1532 return sid->sequence % SESSION_HASH_SIZE;
1533 }
1534
1535 #ifdef CONFIG_SUNRPC_DEBUG
1536 static inline void
1537 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1538 {
1539 u32 *ptr = (u32 *)(&sessionid->data[0]);
1540 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1541 }
1542 #else
1543 static inline void
1544 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1545 {
1546 }
1547 #endif
1548
1549 /*
1550 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1551 * won't be used for replay.
1552 */
1553 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1554 {
1555 struct nfs4_stateowner *so = cstate->replay_owner;
1556
1557 if (nfserr == nfserr_replay_me)
1558 return;
1559
1560 if (!seqid_mutating_err(ntohl(nfserr))) {
1561 nfsd4_cstate_clear_replay(cstate);
1562 return;
1563 }
1564 if (!so)
1565 return;
1566 if (so->so_is_open_owner)
1567 release_last_closed_stateid(openowner(so));
1568 so->so_seqid++;
1569 return;
1570 }
1571
1572 static void
1573 gen_sessionid(struct nfsd4_session *ses)
1574 {
1575 struct nfs4_client *clp = ses->se_client;
1576 struct nfsd4_sessionid *sid;
1577
1578 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1579 sid->clientid = clp->cl_clientid;
1580 sid->sequence = current_sessionid++;
1581 sid->reserved = 0;
1582 }
1583
1584 /*
1585 * The protocol defines ca_maxresponssize_cached to include the size of
1586 * the rpc header, but all we need to cache is the data starting after
1587 * the end of the initial SEQUENCE operation--the rest we regenerate
1588 * each time. Therefore we can advertise a ca_maxresponssize_cached
1589 * value that is the number of bytes in our cache plus a few additional
1590 * bytes. In order to stay on the safe side, and not promise more than
1591 * we can cache, those additional bytes must be the minimum possible: 24
1592 * bytes of rpc header (xid through accept state, with AUTH_NULL
1593 * verifier), 12 for the compound header (with zero-length tag), and 44
1594 * for the SEQUENCE op response:
1595 */
1596 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1597
1598 static void
1599 free_session_slots(struct nfsd4_session *ses)
1600 {
1601 int i;
1602
1603 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1604 free_svc_cred(&ses->se_slots[i]->sl_cred);
1605 kfree(ses->se_slots[i]);
1606 }
1607 }
1608
1609 /*
1610 * We don't actually need to cache the rpc and session headers, so we
1611 * can allocate a little less for each slot:
1612 */
1613 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1614 {
1615 u32 size;
1616
1617 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1618 size = 0;
1619 else
1620 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1621 return size + sizeof(struct nfsd4_slot);
1622 }
1623
1624 /*
1625 * XXX: If we run out of reserved DRC memory we could (up to a point)
1626 * re-negotiate active sessions and reduce their slot usage to make
1627 * room for new connections. For now we just fail the create session.
1628 */
1629 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1630 {
1631 u32 slotsize = slot_bytes(ca);
1632 u32 num = ca->maxreqs;
1633 unsigned long avail, total_avail;
1634 unsigned int scale_factor;
1635
1636 spin_lock(&nfsd_drc_lock);
1637 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1638 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1639 else
1640 /* We have handed out more space than we chose in
1641 * set_max_drc() to allow. That isn't really a
1642 * problem as long as that doesn't make us think we
1643 * have lots more due to integer overflow.
1644 */
1645 total_avail = 0;
1646 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1647 /*
1648 * Never use more than a fraction of the remaining memory,
1649 * unless it's the only way to give this client a slot.
1650 * The chosen fraction is either 1/8 or 1/number of threads,
1651 * whichever is smaller. This ensures there are adequate
1652 * slots to support multiple clients per thread.
1653 * Give the client one slot even if that would require
1654 * over-allocation--it is better than failure.
1655 */
1656 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1657
1658 avail = clamp_t(unsigned long, avail, slotsize,
1659 total_avail/scale_factor);
1660 num = min_t(int, num, avail / slotsize);
1661 num = max_t(int, num, 1);
1662 nfsd_drc_mem_used += num * slotsize;
1663 spin_unlock(&nfsd_drc_lock);
1664
1665 return num;
1666 }
1667
1668 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1669 {
1670 int slotsize = slot_bytes(ca);
1671
1672 spin_lock(&nfsd_drc_lock);
1673 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1674 spin_unlock(&nfsd_drc_lock);
1675 }
1676
1677 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1678 struct nfsd4_channel_attrs *battrs)
1679 {
1680 int numslots = fattrs->maxreqs;
1681 int slotsize = slot_bytes(fattrs);
1682 struct nfsd4_session *new;
1683 int mem, i;
1684
1685 BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1686 + sizeof(struct nfsd4_session) > PAGE_SIZE);
1687 mem = numslots * sizeof(struct nfsd4_slot *);
1688
1689 new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1690 if (!new)
1691 return NULL;
1692 /* allocate each struct nfsd4_slot and data cache in one piece */
1693 for (i = 0; i < numslots; i++) {
1694 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1695 if (!new->se_slots[i])
1696 goto out_free;
1697 }
1698
1699 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1700 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1701
1702 return new;
1703 out_free:
1704 while (i--)
1705 kfree(new->se_slots[i]);
1706 kfree(new);
1707 return NULL;
1708 }
1709
1710 static void free_conn(struct nfsd4_conn *c)
1711 {
1712 svc_xprt_put(c->cn_xprt);
1713 kfree(c);
1714 }
1715
1716 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1717 {
1718 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1719 struct nfs4_client *clp = c->cn_session->se_client;
1720
1721 spin_lock(&clp->cl_lock);
1722 if (!list_empty(&c->cn_persession)) {
1723 list_del(&c->cn_persession);
1724 free_conn(c);
1725 }
1726 nfsd4_probe_callback(clp);
1727 spin_unlock(&clp->cl_lock);
1728 }
1729
1730 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1731 {
1732 struct nfsd4_conn *conn;
1733
1734 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1735 if (!conn)
1736 return NULL;
1737 svc_xprt_get(rqstp->rq_xprt);
1738 conn->cn_xprt = rqstp->rq_xprt;
1739 conn->cn_flags = flags;
1740 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1741 return conn;
1742 }
1743
1744 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1745 {
1746 conn->cn_session = ses;
1747 list_add(&conn->cn_persession, &ses->se_conns);
1748 }
1749
1750 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1751 {
1752 struct nfs4_client *clp = ses->se_client;
1753
1754 spin_lock(&clp->cl_lock);
1755 __nfsd4_hash_conn(conn, ses);
1756 spin_unlock(&clp->cl_lock);
1757 }
1758
1759 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1760 {
1761 conn->cn_xpt_user.callback = nfsd4_conn_lost;
1762 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1763 }
1764
1765 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1766 {
1767 int ret;
1768
1769 nfsd4_hash_conn(conn, ses);
1770 ret = nfsd4_register_conn(conn);
1771 if (ret)
1772 /* oops; xprt is already down: */
1773 nfsd4_conn_lost(&conn->cn_xpt_user);
1774 /* We may have gained or lost a callback channel: */
1775 nfsd4_probe_callback_sync(ses->se_client);
1776 }
1777
1778 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1779 {
1780 u32 dir = NFS4_CDFC4_FORE;
1781
1782 if (cses->flags & SESSION4_BACK_CHAN)
1783 dir |= NFS4_CDFC4_BACK;
1784 return alloc_conn(rqstp, dir);
1785 }
1786
1787 /* must be called under client_lock */
1788 static void nfsd4_del_conns(struct nfsd4_session *s)
1789 {
1790 struct nfs4_client *clp = s->se_client;
1791 struct nfsd4_conn *c;
1792
1793 spin_lock(&clp->cl_lock);
1794 while (!list_empty(&s->se_conns)) {
1795 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1796 list_del_init(&c->cn_persession);
1797 spin_unlock(&clp->cl_lock);
1798
1799 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1800 free_conn(c);
1801
1802 spin_lock(&clp->cl_lock);
1803 }
1804 spin_unlock(&clp->cl_lock);
1805 }
1806
1807 static void __free_session(struct nfsd4_session *ses)
1808 {
1809 free_session_slots(ses);
1810 kfree(ses);
1811 }
1812
1813 static void free_session(struct nfsd4_session *ses)
1814 {
1815 nfsd4_del_conns(ses);
1816 nfsd4_put_drc_mem(&ses->se_fchannel);
1817 __free_session(ses);
1818 }
1819
1820 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1821 {
1822 int idx;
1823 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1824
1825 new->se_client = clp;
1826 gen_sessionid(new);
1827
1828 INIT_LIST_HEAD(&new->se_conns);
1829
1830 new->se_cb_seq_nr = 1;
1831 new->se_flags = cses->flags;
1832 new->se_cb_prog = cses->callback_prog;
1833 new->se_cb_sec = cses->cb_sec;
1834 atomic_set(&new->se_ref, 0);
1835 idx = hash_sessionid(&new->se_sessionid);
1836 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1837 spin_lock(&clp->cl_lock);
1838 list_add(&new->se_perclnt, &clp->cl_sessions);
1839 spin_unlock(&clp->cl_lock);
1840
1841 {
1842 struct sockaddr *sa = svc_addr(rqstp);
1843 /*
1844 * This is a little silly; with sessions there's no real
1845 * use for the callback address. Use the peer address
1846 * as a reasonable default for now, but consider fixing
1847 * the rpc client not to require an address in the
1848 * future:
1849 */
1850 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1851 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1852 }
1853 }
1854
1855 /* caller must hold client_lock */
1856 static struct nfsd4_session *
1857 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1858 {
1859 struct nfsd4_session *elem;
1860 int idx;
1861 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1862
1863 lockdep_assert_held(&nn->client_lock);
1864
1865 dump_sessionid(__func__, sessionid);
1866 idx = hash_sessionid(sessionid);
1867 /* Search in the appropriate list */
1868 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1869 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1870 NFS4_MAX_SESSIONID_LEN)) {
1871 return elem;
1872 }
1873 }
1874
1875 dprintk("%s: session not found\n", __func__);
1876 return NULL;
1877 }
1878
1879 static struct nfsd4_session *
1880 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1881 __be32 *ret)
1882 {
1883 struct nfsd4_session *session;
1884 __be32 status = nfserr_badsession;
1885
1886 session = __find_in_sessionid_hashtbl(sessionid, net);
1887 if (!session)
1888 goto out;
1889 status = nfsd4_get_session_locked(session);
1890 if (status)
1891 session = NULL;
1892 out:
1893 *ret = status;
1894 return session;
1895 }
1896
1897 /* caller must hold client_lock */
1898 static void
1899 unhash_session(struct nfsd4_session *ses)
1900 {
1901 struct nfs4_client *clp = ses->se_client;
1902 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1903
1904 lockdep_assert_held(&nn->client_lock);
1905
1906 list_del(&ses->se_hash);
1907 spin_lock(&ses->se_client->cl_lock);
1908 list_del(&ses->se_perclnt);
1909 spin_unlock(&ses->se_client->cl_lock);
1910 }
1911
1912 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1913 static int
1914 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1915 {
1916 /*
1917 * We're assuming the clid was not given out from a boot
1918 * precisely 2^32 (about 136 years) before this one. That seems
1919 * a safe assumption:
1920 */
1921 if (clid->cl_boot == (u32)nn->boot_time)
1922 return 0;
1923 dprintk("NFSD stale clientid (%08x/%08x) boot_time %08llx\n",
1924 clid->cl_boot, clid->cl_id, nn->boot_time);
1925 return 1;
1926 }
1927
1928 /*
1929 * XXX Should we use a slab cache ?
1930 * This type of memory management is somewhat inefficient, but we use it
1931 * anyway since SETCLIENTID is not a common operation.
1932 */
1933 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1934 {
1935 struct nfs4_client *clp;
1936 int i;
1937
1938 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1939 if (clp == NULL)
1940 return NULL;
1941 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
1942 if (clp->cl_name.data == NULL)
1943 goto err_no_name;
1944 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
1945 sizeof(struct list_head),
1946 GFP_KERNEL);
1947 if (!clp->cl_ownerstr_hashtbl)
1948 goto err_no_hashtbl;
1949 for (i = 0; i < OWNER_HASH_SIZE; i++)
1950 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1951 INIT_LIST_HEAD(&clp->cl_sessions);
1952 idr_init(&clp->cl_stateids);
1953 atomic_set(&clp->cl_rpc_users, 0);
1954 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1955 INIT_LIST_HEAD(&clp->cl_idhash);
1956 INIT_LIST_HEAD(&clp->cl_openowners);
1957 INIT_LIST_HEAD(&clp->cl_delegations);
1958 INIT_LIST_HEAD(&clp->cl_lru);
1959 INIT_LIST_HEAD(&clp->cl_revoked);
1960 #ifdef CONFIG_NFSD_PNFS
1961 INIT_LIST_HEAD(&clp->cl_lo_states);
1962 #endif
1963 INIT_LIST_HEAD(&clp->async_copies);
1964 spin_lock_init(&clp->async_lock);
1965 spin_lock_init(&clp->cl_lock);
1966 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1967 return clp;
1968 err_no_hashtbl:
1969 kfree(clp->cl_name.data);
1970 err_no_name:
1971 kmem_cache_free(client_slab, clp);
1972 return NULL;
1973 }
1974
1975 static void __free_client(struct kref *k)
1976 {
1977 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
1978 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
1979
1980 free_svc_cred(&clp->cl_cred);
1981 kfree(clp->cl_ownerstr_hashtbl);
1982 kfree(clp->cl_name.data);
1983 kfree(clp->cl_nii_domain.data);
1984 kfree(clp->cl_nii_name.data);
1985 idr_destroy(&clp->cl_stateids);
1986 kmem_cache_free(client_slab, clp);
1987 }
1988
1989 static void drop_client(struct nfs4_client *clp)
1990 {
1991 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
1992 }
1993
1994 static void
1995 free_client(struct nfs4_client *clp)
1996 {
1997 while (!list_empty(&clp->cl_sessions)) {
1998 struct nfsd4_session *ses;
1999 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2000 se_perclnt);
2001 list_del(&ses->se_perclnt);
2002 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2003 free_session(ses);
2004 }
2005 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2006 if (clp->cl_nfsd_dentry) {
2007 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2008 clp->cl_nfsd_dentry = NULL;
2009 wake_up_all(&expiry_wq);
2010 }
2011 drop_client(clp);
2012 }
2013
2014 /* must be called under the client_lock */
2015 static void
2016 unhash_client_locked(struct nfs4_client *clp)
2017 {
2018 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2019 struct nfsd4_session *ses;
2020
2021 lockdep_assert_held(&nn->client_lock);
2022
2023 /* Mark the client as expired! */
2024 clp->cl_time = 0;
2025 /* Make it invisible */
2026 if (!list_empty(&clp->cl_idhash)) {
2027 list_del_init(&clp->cl_idhash);
2028 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2029 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2030 else
2031 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2032 }
2033 list_del_init(&clp->cl_lru);
2034 spin_lock(&clp->cl_lock);
2035 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2036 list_del_init(&ses->se_hash);
2037 spin_unlock(&clp->cl_lock);
2038 }
2039
2040 static void
2041 unhash_client(struct nfs4_client *clp)
2042 {
2043 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2044
2045 spin_lock(&nn->client_lock);
2046 unhash_client_locked(clp);
2047 spin_unlock(&nn->client_lock);
2048 }
2049
2050 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2051 {
2052 if (atomic_read(&clp->cl_rpc_users))
2053 return nfserr_jukebox;
2054 unhash_client_locked(clp);
2055 return nfs_ok;
2056 }
2057
2058 static void
2059 __destroy_client(struct nfs4_client *clp)
2060 {
2061 int i;
2062 struct nfs4_openowner *oo;
2063 struct nfs4_delegation *dp;
2064 struct list_head reaplist;
2065
2066 INIT_LIST_HEAD(&reaplist);
2067 spin_lock(&state_lock);
2068 while (!list_empty(&clp->cl_delegations)) {
2069 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2070 WARN_ON(!unhash_delegation_locked(dp));
2071 list_add(&dp->dl_recall_lru, &reaplist);
2072 }
2073 spin_unlock(&state_lock);
2074 while (!list_empty(&reaplist)) {
2075 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2076 list_del_init(&dp->dl_recall_lru);
2077 destroy_unhashed_deleg(dp);
2078 }
2079 while (!list_empty(&clp->cl_revoked)) {
2080 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2081 list_del_init(&dp->dl_recall_lru);
2082 nfs4_put_stid(&dp->dl_stid);
2083 }
2084 while (!list_empty(&clp->cl_openowners)) {
2085 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2086 nfs4_get_stateowner(&oo->oo_owner);
2087 release_openowner(oo);
2088 }
2089 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2090 struct nfs4_stateowner *so, *tmp;
2091
2092 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2093 so_strhash) {
2094 /* Should be no openowners at this point */
2095 WARN_ON_ONCE(so->so_is_open_owner);
2096 remove_blocked_locks(lockowner(so));
2097 }
2098 }
2099 nfsd4_return_all_client_layouts(clp);
2100 nfsd4_shutdown_copy(clp);
2101 nfsd4_shutdown_callback(clp);
2102 if (clp->cl_cb_conn.cb_xprt)
2103 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2104 free_client(clp);
2105 wake_up_all(&expiry_wq);
2106 }
2107
2108 static void
2109 destroy_client(struct nfs4_client *clp)
2110 {
2111 unhash_client(clp);
2112 __destroy_client(clp);
2113 }
2114
2115 static void inc_reclaim_complete(struct nfs4_client *clp)
2116 {
2117 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2118
2119 if (!nn->track_reclaim_completes)
2120 return;
2121 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2122 return;
2123 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2124 nn->reclaim_str_hashtbl_size) {
2125 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2126 clp->net->ns.inum);
2127 nfsd4_end_grace(nn);
2128 }
2129 }
2130
2131 static void expire_client(struct nfs4_client *clp)
2132 {
2133 unhash_client(clp);
2134 nfsd4_client_record_remove(clp);
2135 __destroy_client(clp);
2136 }
2137
2138 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2139 {
2140 memcpy(target->cl_verifier.data, source->data,
2141 sizeof(target->cl_verifier.data));
2142 }
2143
2144 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2145 {
2146 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2147 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2148 }
2149
2150 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2151 {
2152 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2153 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2154 GFP_KERNEL);
2155 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2156 if ((source->cr_principal && !target->cr_principal) ||
2157 (source->cr_raw_principal && !target->cr_raw_principal) ||
2158 (source->cr_targ_princ && !target->cr_targ_princ))
2159 return -ENOMEM;
2160
2161 target->cr_flavor = source->cr_flavor;
2162 target->cr_uid = source->cr_uid;
2163 target->cr_gid = source->cr_gid;
2164 target->cr_group_info = source->cr_group_info;
2165 get_group_info(target->cr_group_info);
2166 target->cr_gss_mech = source->cr_gss_mech;
2167 if (source->cr_gss_mech)
2168 gss_mech_get(source->cr_gss_mech);
2169 return 0;
2170 }
2171
2172 static int
2173 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2174 {
2175 if (o1->len < o2->len)
2176 return -1;
2177 if (o1->len > o2->len)
2178 return 1;
2179 return memcmp(o1->data, o2->data, o1->len);
2180 }
2181
2182 static int
2183 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2184 {
2185 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2186 }
2187
2188 static int
2189 same_clid(clientid_t *cl1, clientid_t *cl2)
2190 {
2191 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2192 }
2193
2194 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2195 {
2196 int i;
2197
2198 if (g1->ngroups != g2->ngroups)
2199 return false;
2200 for (i=0; i<g1->ngroups; i++)
2201 if (!gid_eq(g1->gid[i], g2->gid[i]))
2202 return false;
2203 return true;
2204 }
2205
2206 /*
2207 * RFC 3530 language requires clid_inuse be returned when the
2208 * "principal" associated with a requests differs from that previously
2209 * used. We use uid, gid's, and gss principal string as our best
2210 * approximation. We also don't want to allow non-gss use of a client
2211 * established using gss: in theory cr_principal should catch that
2212 * change, but in practice cr_principal can be null even in the gss case
2213 * since gssd doesn't always pass down a principal string.
2214 */
2215 static bool is_gss_cred(struct svc_cred *cr)
2216 {
2217 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2218 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2219 }
2220
2221
2222 static bool
2223 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2224 {
2225 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2226 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2227 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2228 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2229 return false;
2230 /* XXX: check that cr_targ_princ fields match ? */
2231 if (cr1->cr_principal == cr2->cr_principal)
2232 return true;
2233 if (!cr1->cr_principal || !cr2->cr_principal)
2234 return false;
2235 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2236 }
2237
2238 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2239 {
2240 struct svc_cred *cr = &rqstp->rq_cred;
2241 u32 service;
2242
2243 if (!cr->cr_gss_mech)
2244 return false;
2245 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2246 return service == RPC_GSS_SVC_INTEGRITY ||
2247 service == RPC_GSS_SVC_PRIVACY;
2248 }
2249
2250 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2251 {
2252 struct svc_cred *cr = &rqstp->rq_cred;
2253
2254 if (!cl->cl_mach_cred)
2255 return true;
2256 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2257 return false;
2258 if (!svc_rqst_integrity_protected(rqstp))
2259 return false;
2260 if (cl->cl_cred.cr_raw_principal)
2261 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2262 cr->cr_raw_principal);
2263 if (!cr->cr_principal)
2264 return false;
2265 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2266 }
2267
2268 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2269 {
2270 __be32 verf[2];
2271
2272 /*
2273 * This is opaque to client, so no need to byte-swap. Use
2274 * __force to keep sparse happy
2275 */
2276 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2277 verf[1] = (__force __be32)nn->clverifier_counter++;
2278 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2279 }
2280
2281 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2282 {
2283 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2284 clp->cl_clientid.cl_id = nn->clientid_counter++;
2285 gen_confirm(clp, nn);
2286 }
2287
2288 static struct nfs4_stid *
2289 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2290 {
2291 struct nfs4_stid *ret;
2292
2293 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2294 if (!ret || !ret->sc_type)
2295 return NULL;
2296 return ret;
2297 }
2298
2299 static struct nfs4_stid *
2300 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2301 {
2302 struct nfs4_stid *s;
2303
2304 spin_lock(&cl->cl_lock);
2305 s = find_stateid_locked(cl, t);
2306 if (s != NULL) {
2307 if (typemask & s->sc_type)
2308 refcount_inc(&s->sc_count);
2309 else
2310 s = NULL;
2311 }
2312 spin_unlock(&cl->cl_lock);
2313 return s;
2314 }
2315
2316 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2317 {
2318 struct nfsdfs_client *nc;
2319 nc = get_nfsdfs_client(inode);
2320 if (!nc)
2321 return NULL;
2322 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2323 }
2324
2325 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2326 {
2327 seq_printf(m, "\"");
2328 seq_escape_mem_ascii(m, data, len);
2329 seq_printf(m, "\"");
2330 }
2331
2332 static int client_info_show(struct seq_file *m, void *v)
2333 {
2334 struct inode *inode = m->private;
2335 struct nfs4_client *clp;
2336 u64 clid;
2337
2338 clp = get_nfsdfs_clp(inode);
2339 if (!clp)
2340 return -ENXIO;
2341 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2342 seq_printf(m, "clientid: 0x%llx\n", clid);
2343 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2344 seq_printf(m, "name: ");
2345 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2346 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2347 if (clp->cl_nii_domain.data) {
2348 seq_printf(m, "Implementation domain: ");
2349 seq_quote_mem(m, clp->cl_nii_domain.data,
2350 clp->cl_nii_domain.len);
2351 seq_printf(m, "\nImplementation name: ");
2352 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2353 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2354 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2355 }
2356 drop_client(clp);
2357
2358 return 0;
2359 }
2360
2361 static int client_info_open(struct inode *inode, struct file *file)
2362 {
2363 return single_open(file, client_info_show, inode);
2364 }
2365
2366 static const struct file_operations client_info_fops = {
2367 .open = client_info_open,
2368 .read = seq_read,
2369 .llseek = seq_lseek,
2370 .release = single_release,
2371 };
2372
2373 static void *states_start(struct seq_file *s, loff_t *pos)
2374 __acquires(&clp->cl_lock)
2375 {
2376 struct nfs4_client *clp = s->private;
2377 unsigned long id = *pos;
2378 void *ret;
2379
2380 spin_lock(&clp->cl_lock);
2381 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2382 *pos = id;
2383 return ret;
2384 }
2385
2386 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2387 {
2388 struct nfs4_client *clp = s->private;
2389 unsigned long id = *pos;
2390 void *ret;
2391
2392 id = *pos;
2393 id++;
2394 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2395 *pos = id;
2396 return ret;
2397 }
2398
2399 static void states_stop(struct seq_file *s, void *v)
2400 __releases(&clp->cl_lock)
2401 {
2402 struct nfs4_client *clp = s->private;
2403
2404 spin_unlock(&clp->cl_lock);
2405 }
2406
2407 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2408 {
2409 struct inode *inode = f->nf_inode;
2410
2411 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2412 MAJOR(inode->i_sb->s_dev),
2413 MINOR(inode->i_sb->s_dev),
2414 inode->i_ino);
2415 }
2416
2417 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2418 {
2419 seq_printf(s, "owner: ");
2420 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2421 }
2422
2423 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2424 {
2425 struct nfs4_ol_stateid *ols;
2426 struct nfs4_file *nf;
2427 struct nfsd_file *file;
2428 struct nfs4_stateowner *oo;
2429 unsigned int access, deny;
2430
2431 if (st->sc_type != NFS4_OPEN_STID && st->sc_type != NFS4_LOCK_STID)
2432 return 0; /* XXX: or SEQ_SKIP? */
2433 ols = openlockstateid(st);
2434 oo = ols->st_stateowner;
2435 nf = st->sc_file;
2436 file = find_any_file(nf);
2437
2438 seq_printf(s, "- 0x%16phN: { type: open, ", &st->sc_stateid);
2439
2440 access = bmap_to_share_mode(ols->st_access_bmap);
2441 deny = bmap_to_share_mode(ols->st_deny_bmap);
2442
2443 seq_printf(s, "access: %s%s, ",
2444 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2445 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2446 seq_printf(s, "deny: %s%s, ",
2447 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2448 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2449
2450 nfs4_show_superblock(s, file);
2451 seq_printf(s, ", ");
2452 nfs4_show_owner(s, oo);
2453 seq_printf(s, " }\n");
2454 nfsd_file_put(file);
2455
2456 return 0;
2457 }
2458
2459 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2460 {
2461 struct nfs4_ol_stateid *ols;
2462 struct nfs4_file *nf;
2463 struct nfsd_file *file;
2464 struct nfs4_stateowner *oo;
2465
2466 ols = openlockstateid(st);
2467 oo = ols->st_stateowner;
2468 nf = st->sc_file;
2469 file = find_any_file(nf);
2470
2471 seq_printf(s, "- 0x%16phN: { type: lock, ", &st->sc_stateid);
2472
2473 /*
2474 * Note: a lock stateid isn't really the same thing as a lock,
2475 * it's the locking state held by one owner on a file, and there
2476 * may be multiple (or no) lock ranges associated with it.
2477 * (Same for the matter is true of open stateids.)
2478 */
2479
2480 nfs4_show_superblock(s, file);
2481 /* XXX: open stateid? */
2482 seq_printf(s, ", ");
2483 nfs4_show_owner(s, oo);
2484 seq_printf(s, " }\n");
2485 nfsd_file_put(file);
2486
2487 return 0;
2488 }
2489
2490 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2491 {
2492 struct nfs4_delegation *ds;
2493 struct nfs4_file *nf;
2494 struct nfsd_file *file;
2495
2496 ds = delegstateid(st);
2497 nf = st->sc_file;
2498 file = nf->fi_deleg_file;
2499
2500 seq_printf(s, "- 0x%16phN: { type: deleg, ", &st->sc_stateid);
2501
2502 /* Kinda dead code as long as we only support read delegs: */
2503 seq_printf(s, "access: %s, ",
2504 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2505
2506 /* XXX: lease time, whether it's being recalled. */
2507
2508 nfs4_show_superblock(s, file);
2509 seq_printf(s, " }\n");
2510
2511 return 0;
2512 }
2513
2514 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2515 {
2516 struct nfs4_layout_stateid *ls;
2517 struct nfsd_file *file;
2518
2519 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2520 file = ls->ls_file;
2521
2522 seq_printf(s, "- 0x%16phN: { type: layout, ", &st->sc_stateid);
2523
2524 /* XXX: What else would be useful? */
2525
2526 nfs4_show_superblock(s, file);
2527 seq_printf(s, " }\n");
2528
2529 return 0;
2530 }
2531
2532 static int states_show(struct seq_file *s, void *v)
2533 {
2534 struct nfs4_stid *st = v;
2535
2536 switch (st->sc_type) {
2537 case NFS4_OPEN_STID:
2538 return nfs4_show_open(s, st);
2539 case NFS4_LOCK_STID:
2540 return nfs4_show_lock(s, st);
2541 case NFS4_DELEG_STID:
2542 return nfs4_show_deleg(s, st);
2543 case NFS4_LAYOUT_STID:
2544 return nfs4_show_layout(s, st);
2545 default:
2546 return 0; /* XXX: or SEQ_SKIP? */
2547 }
2548 /* XXX: copy stateids? */
2549 }
2550
2551 static struct seq_operations states_seq_ops = {
2552 .start = states_start,
2553 .next = states_next,
2554 .stop = states_stop,
2555 .show = states_show
2556 };
2557
2558 static int client_states_open(struct inode *inode, struct file *file)
2559 {
2560 struct seq_file *s;
2561 struct nfs4_client *clp;
2562 int ret;
2563
2564 clp = get_nfsdfs_clp(inode);
2565 if (!clp)
2566 return -ENXIO;
2567
2568 ret = seq_open(file, &states_seq_ops);
2569 if (ret)
2570 return ret;
2571 s = file->private_data;
2572 s->private = clp;
2573 return 0;
2574 }
2575
2576 static int client_opens_release(struct inode *inode, struct file *file)
2577 {
2578 struct seq_file *m = file->private_data;
2579 struct nfs4_client *clp = m->private;
2580
2581 /* XXX: alternatively, we could get/drop in seq start/stop */
2582 drop_client(clp);
2583 return 0;
2584 }
2585
2586 static const struct file_operations client_states_fops = {
2587 .open = client_states_open,
2588 .read = seq_read,
2589 .llseek = seq_lseek,
2590 .release = client_opens_release,
2591 };
2592
2593 /*
2594 * Normally we refuse to destroy clients that are in use, but here the
2595 * administrator is telling us to just do it. We also want to wait
2596 * so the caller has a guarantee that the client's locks are gone by
2597 * the time the write returns:
2598 */
2599 static void force_expire_client(struct nfs4_client *clp)
2600 {
2601 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2602 bool already_expired;
2603
2604 spin_lock(&clp->cl_lock);
2605 clp->cl_time = 0;
2606 spin_unlock(&clp->cl_lock);
2607
2608 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2609 spin_lock(&nn->client_lock);
2610 already_expired = list_empty(&clp->cl_lru);
2611 if (!already_expired)
2612 unhash_client_locked(clp);
2613 spin_unlock(&nn->client_lock);
2614
2615 if (!already_expired)
2616 expire_client(clp);
2617 else
2618 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
2619 }
2620
2621 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
2622 size_t size, loff_t *pos)
2623 {
2624 char *data;
2625 struct nfs4_client *clp;
2626
2627 data = simple_transaction_get(file, buf, size);
2628 if (IS_ERR(data))
2629 return PTR_ERR(data);
2630 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
2631 return -EINVAL;
2632 clp = get_nfsdfs_clp(file_inode(file));
2633 if (!clp)
2634 return -ENXIO;
2635 force_expire_client(clp);
2636 drop_client(clp);
2637 return 7;
2638 }
2639
2640 static const struct file_operations client_ctl_fops = {
2641 .write = client_ctl_write,
2642 .release = simple_transaction_release,
2643 };
2644
2645 static const struct tree_descr client_files[] = {
2646 [0] = {"info", &client_info_fops, S_IRUSR},
2647 [1] = {"states", &client_states_fops, S_IRUSR},
2648 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
2649 [3] = {""},
2650 };
2651
2652 static struct nfs4_client *create_client(struct xdr_netobj name,
2653 struct svc_rqst *rqstp, nfs4_verifier *verf)
2654 {
2655 struct nfs4_client *clp;
2656 struct sockaddr *sa = svc_addr(rqstp);
2657 int ret;
2658 struct net *net = SVC_NET(rqstp);
2659 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2660
2661 clp = alloc_client(name);
2662 if (clp == NULL)
2663 return NULL;
2664
2665 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2666 if (ret) {
2667 free_client(clp);
2668 return NULL;
2669 }
2670 gen_clid(clp, nn);
2671 kref_init(&clp->cl_nfsdfs.cl_ref);
2672 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2673 clp->cl_time = ktime_get_boottime_seconds();
2674 clear_bit(0, &clp->cl_cb_slot_busy);
2675 copy_verf(clp, verf);
2676 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
2677 clp->cl_cb_session = NULL;
2678 clp->net = net;
2679 clp->cl_nfsd_dentry = nfsd_client_mkdir(nn, &clp->cl_nfsdfs,
2680 clp->cl_clientid.cl_id - nn->clientid_base,
2681 client_files);
2682 if (!clp->cl_nfsd_dentry) {
2683 free_client(clp);
2684 return NULL;
2685 }
2686 return clp;
2687 }
2688
2689 static void
2690 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2691 {
2692 struct rb_node **new = &(root->rb_node), *parent = NULL;
2693 struct nfs4_client *clp;
2694
2695 while (*new) {
2696 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2697 parent = *new;
2698
2699 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2700 new = &((*new)->rb_left);
2701 else
2702 new = &((*new)->rb_right);
2703 }
2704
2705 rb_link_node(&new_clp->cl_namenode, parent, new);
2706 rb_insert_color(&new_clp->cl_namenode, root);
2707 }
2708
2709 static struct nfs4_client *
2710 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2711 {
2712 int cmp;
2713 struct rb_node *node = root->rb_node;
2714 struct nfs4_client *clp;
2715
2716 while (node) {
2717 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2718 cmp = compare_blob(&clp->cl_name, name);
2719 if (cmp > 0)
2720 node = node->rb_left;
2721 else if (cmp < 0)
2722 node = node->rb_right;
2723 else
2724 return clp;
2725 }
2726 return NULL;
2727 }
2728
2729 static void
2730 add_to_unconfirmed(struct nfs4_client *clp)
2731 {
2732 unsigned int idhashval;
2733 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2734
2735 lockdep_assert_held(&nn->client_lock);
2736
2737 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2738 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2739 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2740 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2741 renew_client_locked(clp);
2742 }
2743
2744 static void
2745 move_to_confirmed(struct nfs4_client *clp)
2746 {
2747 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2748 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2749
2750 lockdep_assert_held(&nn->client_lock);
2751
2752 dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2753 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2754 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2755 add_clp_to_name_tree(clp, &nn->conf_name_tree);
2756 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2757 renew_client_locked(clp);
2758 }
2759
2760 static struct nfs4_client *
2761 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2762 {
2763 struct nfs4_client *clp;
2764 unsigned int idhashval = clientid_hashval(clid->cl_id);
2765
2766 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2767 if (same_clid(&clp->cl_clientid, clid)) {
2768 if ((bool)clp->cl_minorversion != sessions)
2769 return NULL;
2770 renew_client_locked(clp);
2771 return clp;
2772 }
2773 }
2774 return NULL;
2775 }
2776
2777 static struct nfs4_client *
2778 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2779 {
2780 struct list_head *tbl = nn->conf_id_hashtbl;
2781
2782 lockdep_assert_held(&nn->client_lock);
2783 return find_client_in_id_table(tbl, clid, sessions);
2784 }
2785
2786 static struct nfs4_client *
2787 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2788 {
2789 struct list_head *tbl = nn->unconf_id_hashtbl;
2790
2791 lockdep_assert_held(&nn->client_lock);
2792 return find_client_in_id_table(tbl, clid, sessions);
2793 }
2794
2795 static bool clp_used_exchangeid(struct nfs4_client *clp)
2796 {
2797 return clp->cl_exchange_flags != 0;
2798 }
2799
2800 static struct nfs4_client *
2801 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2802 {
2803 lockdep_assert_held(&nn->client_lock);
2804 return find_clp_in_name_tree(name, &nn->conf_name_tree);
2805 }
2806
2807 static struct nfs4_client *
2808 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2809 {
2810 lockdep_assert_held(&nn->client_lock);
2811 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2812 }
2813
2814 static void
2815 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2816 {
2817 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2818 struct sockaddr *sa = svc_addr(rqstp);
2819 u32 scopeid = rpc_get_scope_id(sa);
2820 unsigned short expected_family;
2821
2822 /* Currently, we only support tcp and tcp6 for the callback channel */
2823 if (se->se_callback_netid_len == 3 &&
2824 !memcmp(se->se_callback_netid_val, "tcp", 3))
2825 expected_family = AF_INET;
2826 else if (se->se_callback_netid_len == 4 &&
2827 !memcmp(se->se_callback_netid_val, "tcp6", 4))
2828 expected_family = AF_INET6;
2829 else
2830 goto out_err;
2831
2832 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2833 se->se_callback_addr_len,
2834 (struct sockaddr *)&conn->cb_addr,
2835 sizeof(conn->cb_addr));
2836
2837 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2838 goto out_err;
2839
2840 if (conn->cb_addr.ss_family == AF_INET6)
2841 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2842
2843 conn->cb_prog = se->se_callback_prog;
2844 conn->cb_ident = se->se_callback_ident;
2845 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2846 return;
2847 out_err:
2848 conn->cb_addr.ss_family = AF_UNSPEC;
2849 conn->cb_addrlen = 0;
2850 dprintk("NFSD: this client (clientid %08x/%08x) "
2851 "will not receive delegations\n",
2852 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2853
2854 return;
2855 }
2856
2857 /*
2858 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2859 */
2860 static void
2861 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2862 {
2863 struct xdr_buf *buf = resp->xdr.buf;
2864 struct nfsd4_slot *slot = resp->cstate.slot;
2865 unsigned int base;
2866
2867 dprintk("--> %s slot %p\n", __func__, slot);
2868
2869 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2870 slot->sl_opcnt = resp->opcnt;
2871 slot->sl_status = resp->cstate.status;
2872 free_svc_cred(&slot->sl_cred);
2873 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2874
2875 if (!nfsd4_cache_this(resp)) {
2876 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2877 return;
2878 }
2879 slot->sl_flags |= NFSD4_SLOT_CACHED;
2880
2881 base = resp->cstate.data_offset;
2882 slot->sl_datalen = buf->len - base;
2883 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2884 WARN(1, "%s: sessions DRC could not cache compound\n",
2885 __func__);
2886 return;
2887 }
2888
2889 /*
2890 * Encode the replay sequence operation from the slot values.
2891 * If cachethis is FALSE encode the uncached rep error on the next
2892 * operation which sets resp->p and increments resp->opcnt for
2893 * nfs4svc_encode_compoundres.
2894 *
2895 */
2896 static __be32
2897 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2898 struct nfsd4_compoundres *resp)
2899 {
2900 struct nfsd4_op *op;
2901 struct nfsd4_slot *slot = resp->cstate.slot;
2902
2903 /* Encode the replayed sequence operation */
2904 op = &args->ops[resp->opcnt - 1];
2905 nfsd4_encode_operation(resp, op);
2906
2907 if (slot->sl_flags & NFSD4_SLOT_CACHED)
2908 return op->status;
2909 if (args->opcnt == 1) {
2910 /*
2911 * The original operation wasn't a solo sequence--we
2912 * always cache those--so this retry must not match the
2913 * original:
2914 */
2915 op->status = nfserr_seq_false_retry;
2916 } else {
2917 op = &args->ops[resp->opcnt++];
2918 op->status = nfserr_retry_uncached_rep;
2919 nfsd4_encode_operation(resp, op);
2920 }
2921 return op->status;
2922 }
2923
2924 /*
2925 * The sequence operation is not cached because we can use the slot and
2926 * session values.
2927 */
2928 static __be32
2929 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2930 struct nfsd4_sequence *seq)
2931 {
2932 struct nfsd4_slot *slot = resp->cstate.slot;
2933 struct xdr_stream *xdr = &resp->xdr;
2934 __be32 *p;
2935 __be32 status;
2936
2937 dprintk("--> %s slot %p\n", __func__, slot);
2938
2939 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2940 if (status)
2941 return status;
2942
2943 p = xdr_reserve_space(xdr, slot->sl_datalen);
2944 if (!p) {
2945 WARN_ON_ONCE(1);
2946 return nfserr_serverfault;
2947 }
2948 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2949 xdr_commit_encode(xdr);
2950
2951 resp->opcnt = slot->sl_opcnt;
2952 return slot->sl_status;
2953 }
2954
2955 /*
2956 * Set the exchange_id flags returned by the server.
2957 */
2958 static void
2959 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2960 {
2961 #ifdef CONFIG_NFSD_PNFS
2962 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2963 #else
2964 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2965 #endif
2966
2967 /* Referrals are supported, Migration is not. */
2968 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2969
2970 /* set the wire flags to return to client. */
2971 clid->flags = new->cl_exchange_flags;
2972 }
2973
2974 static bool client_has_openowners(struct nfs4_client *clp)
2975 {
2976 struct nfs4_openowner *oo;
2977
2978 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2979 if (!list_empty(&oo->oo_owner.so_stateids))
2980 return true;
2981 }
2982 return false;
2983 }
2984
2985 static bool client_has_state(struct nfs4_client *clp)
2986 {
2987 return client_has_openowners(clp)
2988 #ifdef CONFIG_NFSD_PNFS
2989 || !list_empty(&clp->cl_lo_states)
2990 #endif
2991 || !list_empty(&clp->cl_delegations)
2992 || !list_empty(&clp->cl_sessions)
2993 || !list_empty(&clp->async_copies);
2994 }
2995
2996 static __be32 copy_impl_id(struct nfs4_client *clp,
2997 struct nfsd4_exchange_id *exid)
2998 {
2999 if (!exid->nii_domain.data)
3000 return 0;
3001 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3002 if (!clp->cl_nii_domain.data)
3003 return nfserr_jukebox;
3004 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3005 if (!clp->cl_nii_name.data)
3006 return nfserr_jukebox;
3007 clp->cl_nii_time = exid->nii_time;
3008 return 0;
3009 }
3010
3011 __be32
3012 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3013 union nfsd4_op_u *u)
3014 {
3015 struct nfsd4_exchange_id *exid = &u->exchange_id;
3016 struct nfs4_client *conf, *new;
3017 struct nfs4_client *unconf = NULL;
3018 __be32 status;
3019 char addr_str[INET6_ADDRSTRLEN];
3020 nfs4_verifier verf = exid->verifier;
3021 struct sockaddr *sa = svc_addr(rqstp);
3022 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3023 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3024
3025 rpc_ntop(sa, addr_str, sizeof(addr_str));
3026 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3027 "ip_addr=%s flags %x, spa_how %d\n",
3028 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3029 addr_str, exid->flags, exid->spa_how);
3030
3031 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3032 return nfserr_inval;
3033
3034 new = create_client(exid->clname, rqstp, &verf);
3035 if (new == NULL)
3036 return nfserr_jukebox;
3037 status = copy_impl_id(new, exid);
3038 if (status)
3039 goto out_nolock;
3040
3041 switch (exid->spa_how) {
3042 case SP4_MACH_CRED:
3043 exid->spo_must_enforce[0] = 0;
3044 exid->spo_must_enforce[1] = (
3045 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3046 1 << (OP_EXCHANGE_ID - 32) |
3047 1 << (OP_CREATE_SESSION - 32) |
3048 1 << (OP_DESTROY_SESSION - 32) |
3049 1 << (OP_DESTROY_CLIENTID - 32));
3050
3051 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3052 1 << (OP_OPEN_DOWNGRADE) |
3053 1 << (OP_LOCKU) |
3054 1 << (OP_DELEGRETURN));
3055
3056 exid->spo_must_allow[1] &= (
3057 1 << (OP_TEST_STATEID - 32) |
3058 1 << (OP_FREE_STATEID - 32));
3059 if (!svc_rqst_integrity_protected(rqstp)) {
3060 status = nfserr_inval;
3061 goto out_nolock;
3062 }
3063 /*
3064 * Sometimes userspace doesn't give us a principal.
3065 * Which is a bug, really. Anyway, we can't enforce
3066 * MACH_CRED in that case, better to give up now:
3067 */
3068 if (!new->cl_cred.cr_principal &&
3069 !new->cl_cred.cr_raw_principal) {
3070 status = nfserr_serverfault;
3071 goto out_nolock;
3072 }
3073 new->cl_mach_cred = true;
3074 case SP4_NONE:
3075 break;
3076 default: /* checked by xdr code */
3077 WARN_ON_ONCE(1);
3078 /* fall through */
3079 case SP4_SSV:
3080 status = nfserr_encr_alg_unsupp;
3081 goto out_nolock;
3082 }
3083
3084 /* Cases below refer to rfc 5661 section 18.35.4: */
3085 spin_lock(&nn->client_lock);
3086 conf = find_confirmed_client_by_name(&exid->clname, nn);
3087 if (conf) {
3088 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3089 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3090
3091 if (update) {
3092 if (!clp_used_exchangeid(conf)) { /* buggy client */
3093 status = nfserr_inval;
3094 goto out;
3095 }
3096 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3097 status = nfserr_wrong_cred;
3098 goto out;
3099 }
3100 if (!creds_match) { /* case 9 */
3101 status = nfserr_perm;
3102 goto out;
3103 }
3104 if (!verfs_match) { /* case 8 */
3105 status = nfserr_not_same;
3106 goto out;
3107 }
3108 /* case 6 */
3109 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3110 goto out_copy;
3111 }
3112 if (!creds_match) { /* case 3 */
3113 if (client_has_state(conf)) {
3114 status = nfserr_clid_inuse;
3115 goto out;
3116 }
3117 goto out_new;
3118 }
3119 if (verfs_match) { /* case 2 */
3120 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3121 goto out_copy;
3122 }
3123 /* case 5, client reboot */
3124 conf = NULL;
3125 goto out_new;
3126 }
3127
3128 if (update) { /* case 7 */
3129 status = nfserr_noent;
3130 goto out;
3131 }
3132
3133 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3134 if (unconf) /* case 4, possible retry or client restart */
3135 unhash_client_locked(unconf);
3136
3137 /* case 1 (normal case) */
3138 out_new:
3139 if (conf) {
3140 status = mark_client_expired_locked(conf);
3141 if (status)
3142 goto out;
3143 }
3144 new->cl_minorversion = cstate->minorversion;
3145 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3146 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3147
3148 add_to_unconfirmed(new);
3149 swap(new, conf);
3150 out_copy:
3151 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3152 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3153
3154 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3155 nfsd4_set_ex_flags(conf, exid);
3156
3157 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3158 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3159 status = nfs_ok;
3160
3161 out:
3162 spin_unlock(&nn->client_lock);
3163 out_nolock:
3164 if (new)
3165 expire_client(new);
3166 if (unconf)
3167 expire_client(unconf);
3168 return status;
3169 }
3170
3171 static __be32
3172 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
3173 {
3174 dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
3175 slot_seqid);
3176
3177 /* The slot is in use, and no response has been sent. */
3178 if (slot_inuse) {
3179 if (seqid == slot_seqid)
3180 return nfserr_jukebox;
3181 else
3182 return nfserr_seq_misordered;
3183 }
3184 /* Note unsigned 32-bit arithmetic handles wraparound: */
3185 if (likely(seqid == slot_seqid + 1))
3186 return nfs_ok;
3187 if (seqid == slot_seqid)
3188 return nfserr_replay_cache;
3189 return nfserr_seq_misordered;
3190 }
3191
3192 /*
3193 * Cache the create session result into the create session single DRC
3194 * slot cache by saving the xdr structure. sl_seqid has been set.
3195 * Do this for solo or embedded create session operations.
3196 */
3197 static void
3198 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3199 struct nfsd4_clid_slot *slot, __be32 nfserr)
3200 {
3201 slot->sl_status = nfserr;
3202 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3203 }
3204
3205 static __be32
3206 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3207 struct nfsd4_clid_slot *slot)
3208 {
3209 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3210 return slot->sl_status;
3211 }
3212
3213 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3214 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3215 1 + /* MIN tag is length with zero, only length */ \
3216 3 + /* version, opcount, opcode */ \
3217 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3218 /* seqid, slotID, slotID, cache */ \
3219 4 ) * sizeof(__be32))
3220
3221 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3222 2 + /* verifier: AUTH_NULL, length 0 */\
3223 1 + /* status */ \
3224 1 + /* MIN tag is length with zero, only length */ \
3225 3 + /* opcount, opcode, opstatus*/ \
3226 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3227 /* seqid, slotID, slotID, slotID, status */ \
3228 5 ) * sizeof(__be32))
3229
3230 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3231 {
3232 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3233
3234 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3235 return nfserr_toosmall;
3236 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3237 return nfserr_toosmall;
3238 ca->headerpadsz = 0;
3239 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3240 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3241 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3242 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3243 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3244 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3245 /*
3246 * Note decreasing slot size below client's request may make it
3247 * difficult for client to function correctly, whereas
3248 * decreasing the number of slots will (just?) affect
3249 * performance. When short on memory we therefore prefer to
3250 * decrease number of slots instead of their size. Clients that
3251 * request larger slots than they need will get poor results:
3252 * Note that we always allow at least one slot, because our
3253 * accounting is soft and provides no guarantees either way.
3254 */
3255 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3256
3257 return nfs_ok;
3258 }
3259
3260 /*
3261 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3262 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3263 */
3264 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3265 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3266
3267 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3268 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3269
3270 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3271 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3272 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3273 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3274 sizeof(__be32))
3275
3276 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3277 {
3278 ca->headerpadsz = 0;
3279
3280 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3281 return nfserr_toosmall;
3282 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3283 return nfserr_toosmall;
3284 ca->maxresp_cached = 0;
3285 if (ca->maxops < 2)
3286 return nfserr_toosmall;
3287
3288 return nfs_ok;
3289 }
3290
3291 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3292 {
3293 switch (cbs->flavor) {
3294 case RPC_AUTH_NULL:
3295 case RPC_AUTH_UNIX:
3296 return nfs_ok;
3297 default:
3298 /*
3299 * GSS case: the spec doesn't allow us to return this
3300 * error. But it also doesn't allow us not to support
3301 * GSS.
3302 * I'd rather this fail hard than return some error the
3303 * client might think it can already handle:
3304 */
3305 return nfserr_encr_alg_unsupp;
3306 }
3307 }
3308
3309 __be32
3310 nfsd4_create_session(struct svc_rqst *rqstp,
3311 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3312 {
3313 struct nfsd4_create_session *cr_ses = &u->create_session;
3314 struct sockaddr *sa = svc_addr(rqstp);
3315 struct nfs4_client *conf, *unconf;
3316 struct nfs4_client *old = NULL;
3317 struct nfsd4_session *new;
3318 struct nfsd4_conn *conn;
3319 struct nfsd4_clid_slot *cs_slot = NULL;
3320 __be32 status = 0;
3321 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3322
3323 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3324 return nfserr_inval;
3325 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3326 if (status)
3327 return status;
3328 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3329 if (status)
3330 return status;
3331 status = check_backchannel_attrs(&cr_ses->back_channel);
3332 if (status)
3333 goto out_release_drc_mem;
3334 status = nfserr_jukebox;
3335 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3336 if (!new)
3337 goto out_release_drc_mem;
3338 conn = alloc_conn_from_crses(rqstp, cr_ses);
3339 if (!conn)
3340 goto out_free_session;
3341
3342 spin_lock(&nn->client_lock);
3343 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3344 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3345 WARN_ON_ONCE(conf && unconf);
3346
3347 if (conf) {
3348 status = nfserr_wrong_cred;
3349 if (!nfsd4_mach_creds_match(conf, rqstp))
3350 goto out_free_conn;
3351 cs_slot = &conf->cl_cs_slot;
3352 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3353 if (status) {
3354 if (status == nfserr_replay_cache)
3355 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3356 goto out_free_conn;
3357 }
3358 } else if (unconf) {
3359 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3360 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3361 status = nfserr_clid_inuse;
3362 goto out_free_conn;
3363 }
3364 status = nfserr_wrong_cred;
3365 if (!nfsd4_mach_creds_match(unconf, rqstp))
3366 goto out_free_conn;
3367 cs_slot = &unconf->cl_cs_slot;
3368 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3369 if (status) {
3370 /* an unconfirmed replay returns misordered */
3371 status = nfserr_seq_misordered;
3372 goto out_free_conn;
3373 }
3374 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3375 if (old) {
3376 status = mark_client_expired_locked(old);
3377 if (status) {
3378 old = NULL;
3379 goto out_free_conn;
3380 }
3381 }
3382 move_to_confirmed(unconf);
3383 conf = unconf;
3384 } else {
3385 status = nfserr_stale_clientid;
3386 goto out_free_conn;
3387 }
3388 status = nfs_ok;
3389 /* Persistent sessions are not supported */
3390 cr_ses->flags &= ~SESSION4_PERSIST;
3391 /* Upshifting from TCP to RDMA is not supported */
3392 cr_ses->flags &= ~SESSION4_RDMA;
3393
3394 init_session(rqstp, new, conf, cr_ses);
3395 nfsd4_get_session_locked(new);
3396
3397 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3398 NFS4_MAX_SESSIONID_LEN);
3399 cs_slot->sl_seqid++;
3400 cr_ses->seqid = cs_slot->sl_seqid;
3401
3402 /* cache solo and embedded create sessions under the client_lock */
3403 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3404 spin_unlock(&nn->client_lock);
3405 /* init connection and backchannel */
3406 nfsd4_init_conn(rqstp, conn, new);
3407 nfsd4_put_session(new);
3408 if (old)
3409 expire_client(old);
3410 return status;
3411 out_free_conn:
3412 spin_unlock(&nn->client_lock);
3413 free_conn(conn);
3414 if (old)
3415 expire_client(old);
3416 out_free_session:
3417 __free_session(new);
3418 out_release_drc_mem:
3419 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3420 return status;
3421 }
3422
3423 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3424 {
3425 switch (*dir) {
3426 case NFS4_CDFC4_FORE:
3427 case NFS4_CDFC4_BACK:
3428 return nfs_ok;
3429 case NFS4_CDFC4_FORE_OR_BOTH:
3430 case NFS4_CDFC4_BACK_OR_BOTH:
3431 *dir = NFS4_CDFC4_BOTH;
3432 return nfs_ok;
3433 }
3434 return nfserr_inval;
3435 }
3436
3437 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3438 struct nfsd4_compound_state *cstate,
3439 union nfsd4_op_u *u)
3440 {
3441 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3442 struct nfsd4_session *session = cstate->session;
3443 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3444 __be32 status;
3445
3446 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3447 if (status)
3448 return status;
3449 spin_lock(&nn->client_lock);
3450 session->se_cb_prog = bc->bc_cb_program;
3451 session->se_cb_sec = bc->bc_cb_sec;
3452 spin_unlock(&nn->client_lock);
3453
3454 nfsd4_probe_callback(session->se_client);
3455
3456 return nfs_ok;
3457 }
3458
3459 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
3460 struct nfsd4_compound_state *cstate,
3461 union nfsd4_op_u *u)
3462 {
3463 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
3464 __be32 status;
3465 struct nfsd4_conn *conn;
3466 struct nfsd4_session *session;
3467 struct net *net = SVC_NET(rqstp);
3468 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3469
3470 if (!nfsd4_last_compound_op(rqstp))
3471 return nfserr_not_only_op;
3472 spin_lock(&nn->client_lock);
3473 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
3474 spin_unlock(&nn->client_lock);
3475 if (!session)
3476 goto out_no_session;
3477 status = nfserr_wrong_cred;
3478 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
3479 goto out;
3480 status = nfsd4_map_bcts_dir(&bcts->dir);
3481 if (status)
3482 goto out;
3483 conn = alloc_conn(rqstp, bcts->dir);
3484 status = nfserr_jukebox;
3485 if (!conn)
3486 goto out;
3487 nfsd4_init_conn(rqstp, conn, session);
3488 status = nfs_ok;
3489 out:
3490 nfsd4_put_session(session);
3491 out_no_session:
3492 return status;
3493 }
3494
3495 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
3496 {
3497 if (!cstate->session)
3498 return false;
3499 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
3500 }
3501
3502 __be32
3503 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
3504 union nfsd4_op_u *u)
3505 {
3506 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
3507 struct nfsd4_session *ses;
3508 __be32 status;
3509 int ref_held_by_me = 0;
3510 struct net *net = SVC_NET(r);
3511 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3512
3513 status = nfserr_not_only_op;
3514 if (nfsd4_compound_in_session(cstate, sessionid)) {
3515 if (!nfsd4_last_compound_op(r))
3516 goto out;
3517 ref_held_by_me++;
3518 }
3519 dump_sessionid(__func__, sessionid);
3520 spin_lock(&nn->client_lock);
3521 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
3522 if (!ses)
3523 goto out_client_lock;
3524 status = nfserr_wrong_cred;
3525 if (!nfsd4_mach_creds_match(ses->se_client, r))
3526 goto out_put_session;
3527 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
3528 if (status)
3529 goto out_put_session;
3530 unhash_session(ses);
3531 spin_unlock(&nn->client_lock);
3532
3533 nfsd4_probe_callback_sync(ses->se_client);
3534
3535 spin_lock(&nn->client_lock);
3536 status = nfs_ok;
3537 out_put_session:
3538 nfsd4_put_session_locked(ses);
3539 out_client_lock:
3540 spin_unlock(&nn->client_lock);
3541 out:
3542 return status;
3543 }
3544
3545 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3546 {
3547 struct nfsd4_conn *c;
3548
3549 list_for_each_entry(c, &s->se_conns, cn_persession) {
3550 if (c->cn_xprt == xpt) {
3551 return c;
3552 }
3553 }
3554 return NULL;
3555 }
3556
3557 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3558 {
3559 struct nfs4_client *clp = ses->se_client;
3560 struct nfsd4_conn *c;
3561 __be32 status = nfs_ok;
3562 int ret;
3563
3564 spin_lock(&clp->cl_lock);
3565 c = __nfsd4_find_conn(new->cn_xprt, ses);
3566 if (c)
3567 goto out_free;
3568 status = nfserr_conn_not_bound_to_session;
3569 if (clp->cl_mach_cred)
3570 goto out_free;
3571 __nfsd4_hash_conn(new, ses);
3572 spin_unlock(&clp->cl_lock);
3573 ret = nfsd4_register_conn(new);
3574 if (ret)
3575 /* oops; xprt is already down: */
3576 nfsd4_conn_lost(&new->cn_xpt_user);
3577 return nfs_ok;
3578 out_free:
3579 spin_unlock(&clp->cl_lock);
3580 free_conn(new);
3581 return status;
3582 }
3583
3584 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3585 {
3586 struct nfsd4_compoundargs *args = rqstp->rq_argp;
3587
3588 return args->opcnt > session->se_fchannel.maxops;
3589 }
3590
3591 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3592 struct nfsd4_session *session)
3593 {
3594 struct xdr_buf *xb = &rqstp->rq_arg;
3595
3596 return xb->len > session->se_fchannel.maxreq_sz;
3597 }
3598
3599 static bool replay_matches_cache(struct svc_rqst *rqstp,
3600 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3601 {
3602 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3603
3604 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3605 (bool)seq->cachethis)
3606 return false;
3607 /*
3608 * If there's an error then the reply can have fewer ops than
3609 * the call.
3610 */
3611 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
3612 return false;
3613 /*
3614 * But if we cached a reply with *more* ops than the call you're
3615 * sending us now, then this new call is clearly not really a
3616 * replay of the old one:
3617 */
3618 if (slot->sl_opcnt > argp->opcnt)
3619 return false;
3620 /* This is the only check explicitly called by spec: */
3621 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3622 return false;
3623 /*
3624 * There may be more comparisons we could actually do, but the
3625 * spec doesn't require us to catch every case where the calls
3626 * don't match (that would require caching the call as well as
3627 * the reply), so we don't bother.
3628 */
3629 return true;
3630 }
3631
3632 __be32
3633 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3634 union nfsd4_op_u *u)
3635 {
3636 struct nfsd4_sequence *seq = &u->sequence;
3637 struct nfsd4_compoundres *resp = rqstp->rq_resp;
3638 struct xdr_stream *xdr = &resp->xdr;
3639 struct nfsd4_session *session;
3640 struct nfs4_client *clp;
3641 struct nfsd4_slot *slot;
3642 struct nfsd4_conn *conn;
3643 __be32 status;
3644 int buflen;
3645 struct net *net = SVC_NET(rqstp);
3646 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3647
3648 if (resp->opcnt != 1)
3649 return nfserr_sequence_pos;
3650
3651 /*
3652 * Will be either used or freed by nfsd4_sequence_check_conn
3653 * below.
3654 */
3655 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3656 if (!conn)
3657 return nfserr_jukebox;
3658
3659 spin_lock(&nn->client_lock);
3660 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3661 if (!session)
3662 goto out_no_session;
3663 clp = session->se_client;
3664
3665 status = nfserr_too_many_ops;
3666 if (nfsd4_session_too_many_ops(rqstp, session))
3667 goto out_put_session;
3668
3669 status = nfserr_req_too_big;
3670 if (nfsd4_request_too_big(rqstp, session))
3671 goto out_put_session;
3672
3673 status = nfserr_badslot;
3674 if (seq->slotid >= session->se_fchannel.maxreqs)
3675 goto out_put_session;
3676
3677 slot = session->se_slots[seq->slotid];
3678 dprintk("%s: slotid %d\n", __func__, seq->slotid);
3679
3680 /* We do not negotiate the number of slots yet, so set the
3681 * maxslots to the session maxreqs which is used to encode
3682 * sr_highest_slotid and the sr_target_slot id to maxslots */
3683 seq->maxslots = session->se_fchannel.maxreqs;
3684
3685 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3686 slot->sl_flags & NFSD4_SLOT_INUSE);
3687 if (status == nfserr_replay_cache) {
3688 status = nfserr_seq_misordered;
3689 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3690 goto out_put_session;
3691 status = nfserr_seq_false_retry;
3692 if (!replay_matches_cache(rqstp, seq, slot))
3693 goto out_put_session;
3694 cstate->slot = slot;
3695 cstate->session = session;
3696 cstate->clp = clp;
3697 /* Return the cached reply status and set cstate->status
3698 * for nfsd4_proc_compound processing */
3699 status = nfsd4_replay_cache_entry(resp, seq);
3700 cstate->status = nfserr_replay_cache;
3701 goto out;
3702 }
3703 if (status)
3704 goto out_put_session;
3705
3706 status = nfsd4_sequence_check_conn(conn, session);
3707 conn = NULL;
3708 if (status)
3709 goto out_put_session;
3710
3711 buflen = (seq->cachethis) ?
3712 session->se_fchannel.maxresp_cached :
3713 session->se_fchannel.maxresp_sz;
3714 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3715 nfserr_rep_too_big;
3716 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3717 goto out_put_session;
3718 svc_reserve(rqstp, buflen);
3719
3720 status = nfs_ok;
3721 /* Success! bump slot seqid */
3722 slot->sl_seqid = seq->seqid;
3723 slot->sl_flags |= NFSD4_SLOT_INUSE;
3724 if (seq->cachethis)
3725 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3726 else
3727 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3728
3729 cstate->slot = slot;
3730 cstate->session = session;
3731 cstate->clp = clp;
3732
3733 out:
3734 switch (clp->cl_cb_state) {
3735 case NFSD4_CB_DOWN:
3736 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3737 break;
3738 case NFSD4_CB_FAULT:
3739 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3740 break;
3741 default:
3742 seq->status_flags = 0;
3743 }
3744 if (!list_empty(&clp->cl_revoked))
3745 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3746 out_no_session:
3747 if (conn)
3748 free_conn(conn);
3749 spin_unlock(&nn->client_lock);
3750 return status;
3751 out_put_session:
3752 nfsd4_put_session_locked(session);
3753 goto out_no_session;
3754 }
3755
3756 void
3757 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3758 {
3759 struct nfsd4_compound_state *cs = &resp->cstate;
3760
3761 if (nfsd4_has_session(cs)) {
3762 if (cs->status != nfserr_replay_cache) {
3763 nfsd4_store_cache_entry(resp);
3764 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3765 }
3766 /* Drop session reference that was taken in nfsd4_sequence() */
3767 nfsd4_put_session(cs->session);
3768 } else if (cs->clp)
3769 put_client_renew(cs->clp);
3770 }
3771
3772 __be32
3773 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3774 struct nfsd4_compound_state *cstate,
3775 union nfsd4_op_u *u)
3776 {
3777 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3778 struct nfs4_client *conf, *unconf;
3779 struct nfs4_client *clp = NULL;
3780 __be32 status = 0;
3781 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3782
3783 spin_lock(&nn->client_lock);
3784 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3785 conf = find_confirmed_client(&dc->clientid, true, nn);
3786 WARN_ON_ONCE(conf && unconf);
3787
3788 if (conf) {
3789 if (client_has_state(conf)) {
3790 status = nfserr_clientid_busy;
3791 goto out;
3792 }
3793 status = mark_client_expired_locked(conf);
3794 if (status)
3795 goto out;
3796 clp = conf;
3797 } else if (unconf)
3798 clp = unconf;
3799 else {
3800 status = nfserr_stale_clientid;
3801 goto out;
3802 }
3803 if (!nfsd4_mach_creds_match(clp, rqstp)) {
3804 clp = NULL;
3805 status = nfserr_wrong_cred;
3806 goto out;
3807 }
3808 unhash_client_locked(clp);
3809 out:
3810 spin_unlock(&nn->client_lock);
3811 if (clp)
3812 expire_client(clp);
3813 return status;
3814 }
3815
3816 __be32
3817 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3818 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3819 {
3820 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3821 __be32 status = 0;
3822
3823 if (rc->rca_one_fs) {
3824 if (!cstate->current_fh.fh_dentry)
3825 return nfserr_nofilehandle;
3826 /*
3827 * We don't take advantage of the rca_one_fs case.
3828 * That's OK, it's optional, we can safely ignore it.
3829 */
3830 return nfs_ok;
3831 }
3832
3833 status = nfserr_complete_already;
3834 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3835 &cstate->session->se_client->cl_flags))
3836 goto out;
3837
3838 status = nfserr_stale_clientid;
3839 if (is_client_expired(cstate->session->se_client))
3840 /*
3841 * The following error isn't really legal.
3842 * But we only get here if the client just explicitly
3843 * destroyed the client. Surely it no longer cares what
3844 * error it gets back on an operation for the dead
3845 * client.
3846 */
3847 goto out;
3848
3849 status = nfs_ok;
3850 nfsd4_client_record_create(cstate->session->se_client);
3851 inc_reclaim_complete(cstate->session->se_client);
3852 out:
3853 return status;
3854 }
3855
3856 __be32
3857 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3858 union nfsd4_op_u *u)
3859 {
3860 struct nfsd4_setclientid *setclid = &u->setclientid;
3861 struct xdr_netobj clname = setclid->se_name;
3862 nfs4_verifier clverifier = setclid->se_verf;
3863 struct nfs4_client *conf, *new;
3864 struct nfs4_client *unconf = NULL;
3865 __be32 status;
3866 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3867
3868 new = create_client(clname, rqstp, &clverifier);
3869 if (new == NULL)
3870 return nfserr_jukebox;
3871 /* Cases below refer to rfc 3530 section 14.2.33: */
3872 spin_lock(&nn->client_lock);
3873 conf = find_confirmed_client_by_name(&clname, nn);
3874 if (conf && client_has_state(conf)) {
3875 /* case 0: */
3876 status = nfserr_clid_inuse;
3877 if (clp_used_exchangeid(conf))
3878 goto out;
3879 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3880 char addr_str[INET6_ADDRSTRLEN];
3881 rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3882 sizeof(addr_str));
3883 dprintk("NFSD: setclientid: string in use by client "
3884 "at %s\n", addr_str);
3885 goto out;
3886 }
3887 }
3888 unconf = find_unconfirmed_client_by_name(&clname, nn);
3889 if (unconf)
3890 unhash_client_locked(unconf);
3891 if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3892 /* case 1: probable callback update */
3893 copy_clid(new, conf);
3894 gen_confirm(new, nn);
3895 } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3896 ;
3897 new->cl_minorversion = 0;
3898 gen_callback(new, setclid, rqstp);
3899 add_to_unconfirmed(new);
3900 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3901 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3902 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3903 new = NULL;
3904 status = nfs_ok;
3905 out:
3906 spin_unlock(&nn->client_lock);
3907 if (new)
3908 free_client(new);
3909 if (unconf)
3910 expire_client(unconf);
3911 return status;
3912 }
3913
3914
3915 __be32
3916 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3917 struct nfsd4_compound_state *cstate,
3918 union nfsd4_op_u *u)
3919 {
3920 struct nfsd4_setclientid_confirm *setclientid_confirm =
3921 &u->setclientid_confirm;
3922 struct nfs4_client *conf, *unconf;
3923 struct nfs4_client *old = NULL;
3924 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3925 clientid_t * clid = &setclientid_confirm->sc_clientid;
3926 __be32 status;
3927 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3928
3929 if (STALE_CLIENTID(clid, nn))
3930 return nfserr_stale_clientid;
3931
3932 spin_lock(&nn->client_lock);
3933 conf = find_confirmed_client(clid, false, nn);
3934 unconf = find_unconfirmed_client(clid, false, nn);
3935 /*
3936 * We try hard to give out unique clientid's, so if we get an
3937 * attempt to confirm the same clientid with a different cred,
3938 * the client may be buggy; this should never happen.
3939 *
3940 * Nevertheless, RFC 7530 recommends INUSE for this case:
3941 */
3942 status = nfserr_clid_inuse;
3943 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3944 goto out;
3945 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3946 goto out;
3947 /* cases below refer to rfc 3530 section 14.2.34: */
3948 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3949 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3950 /* case 2: probable retransmit */
3951 status = nfs_ok;
3952 } else /* case 4: client hasn't noticed we rebooted yet? */
3953 status = nfserr_stale_clientid;
3954 goto out;
3955 }
3956 status = nfs_ok;
3957 if (conf) { /* case 1: callback update */
3958 old = unconf;
3959 unhash_client_locked(old);
3960 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3961 } else { /* case 3: normal case; new or rebooted client */
3962 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3963 if (old) {
3964 status = nfserr_clid_inuse;
3965 if (client_has_state(old)
3966 && !same_creds(&unconf->cl_cred,
3967 &old->cl_cred))
3968 goto out;
3969 status = mark_client_expired_locked(old);
3970 if (status) {
3971 old = NULL;
3972 goto out;
3973 }
3974 }
3975 move_to_confirmed(unconf);
3976 conf = unconf;
3977 }
3978 get_client_locked(conf);
3979 spin_unlock(&nn->client_lock);
3980 nfsd4_probe_callback(conf);
3981 spin_lock(&nn->client_lock);
3982 put_client_renew_locked(conf);
3983 out:
3984 spin_unlock(&nn->client_lock);
3985 if (old)
3986 expire_client(old);
3987 return status;
3988 }
3989
3990 static struct nfs4_file *nfsd4_alloc_file(void)
3991 {
3992 return kmem_cache_alloc(file_slab, GFP_KERNEL);
3993 }
3994
3995 /* OPEN Share state helper functions */
3996 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3997 struct nfs4_file *fp)
3998 {
3999 lockdep_assert_held(&state_lock);
4000
4001 refcount_set(&fp->fi_ref, 1);
4002 spin_lock_init(&fp->fi_lock);
4003 INIT_LIST_HEAD(&fp->fi_stateids);
4004 INIT_LIST_HEAD(&fp->fi_delegations);
4005 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4006 fh_copy_shallow(&fp->fi_fhandle, fh);
4007 fp->fi_deleg_file = NULL;
4008 fp->fi_had_conflict = false;
4009 fp->fi_share_deny = 0;
4010 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4011 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4012 #ifdef CONFIG_NFSD_PNFS
4013 INIT_LIST_HEAD(&fp->fi_lo_states);
4014 atomic_set(&fp->fi_lo_recalls, 0);
4015 #endif
4016 hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
4017 }
4018
4019 void
4020 nfsd4_free_slabs(void)
4021 {
4022 kmem_cache_destroy(client_slab);
4023 kmem_cache_destroy(openowner_slab);
4024 kmem_cache_destroy(lockowner_slab);
4025 kmem_cache_destroy(file_slab);
4026 kmem_cache_destroy(stateid_slab);
4027 kmem_cache_destroy(deleg_slab);
4028 kmem_cache_destroy(odstate_slab);
4029 }
4030
4031 int
4032 nfsd4_init_slabs(void)
4033 {
4034 client_slab = kmem_cache_create("nfsd4_clients",
4035 sizeof(struct nfs4_client), 0, 0, NULL);
4036 if (client_slab == NULL)
4037 goto out;
4038 openowner_slab = kmem_cache_create("nfsd4_openowners",
4039 sizeof(struct nfs4_openowner), 0, 0, NULL);
4040 if (openowner_slab == NULL)
4041 goto out_free_client_slab;
4042 lockowner_slab = kmem_cache_create("nfsd4_lockowners",
4043 sizeof(struct nfs4_lockowner), 0, 0, NULL);
4044 if (lockowner_slab == NULL)
4045 goto out_free_openowner_slab;
4046 file_slab = kmem_cache_create("nfsd4_files",
4047 sizeof(struct nfs4_file), 0, 0, NULL);
4048 if (file_slab == NULL)
4049 goto out_free_lockowner_slab;
4050 stateid_slab = kmem_cache_create("nfsd4_stateids",
4051 sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
4052 if (stateid_slab == NULL)
4053 goto out_free_file_slab;
4054 deleg_slab = kmem_cache_create("nfsd4_delegations",
4055 sizeof(struct nfs4_delegation), 0, 0, NULL);
4056 if (deleg_slab == NULL)
4057 goto out_free_stateid_slab;
4058 odstate_slab = kmem_cache_create("nfsd4_odstate",
4059 sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
4060 if (odstate_slab == NULL)
4061 goto out_free_deleg_slab;
4062 return 0;
4063
4064 out_free_deleg_slab:
4065 kmem_cache_destroy(deleg_slab);
4066 out_free_stateid_slab:
4067 kmem_cache_destroy(stateid_slab);
4068 out_free_file_slab:
4069 kmem_cache_destroy(file_slab);
4070 out_free_lockowner_slab:
4071 kmem_cache_destroy(lockowner_slab);
4072 out_free_openowner_slab:
4073 kmem_cache_destroy(openowner_slab);
4074 out_free_client_slab:
4075 kmem_cache_destroy(client_slab);
4076 out:
4077 dprintk("nfsd4: out of memory while initializing nfsv4\n");
4078 return -ENOMEM;
4079 }
4080
4081 static void init_nfs4_replay(struct nfs4_replay *rp)
4082 {
4083 rp->rp_status = nfserr_serverfault;
4084 rp->rp_buflen = 0;
4085 rp->rp_buf = rp->rp_ibuf;
4086 mutex_init(&rp->rp_mutex);
4087 }
4088
4089 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4090 struct nfs4_stateowner *so)
4091 {
4092 if (!nfsd4_has_session(cstate)) {
4093 mutex_lock(&so->so_replay.rp_mutex);
4094 cstate->replay_owner = nfs4_get_stateowner(so);
4095 }
4096 }
4097
4098 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4099 {
4100 struct nfs4_stateowner *so = cstate->replay_owner;
4101
4102 if (so != NULL) {
4103 cstate->replay_owner = NULL;
4104 mutex_unlock(&so->so_replay.rp_mutex);
4105 nfs4_put_stateowner(so);
4106 }
4107 }
4108
4109 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4110 {
4111 struct nfs4_stateowner *sop;
4112
4113 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4114 if (!sop)
4115 return NULL;
4116
4117 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4118 if (!sop->so_owner.data) {
4119 kmem_cache_free(slab, sop);
4120 return NULL;
4121 }
4122
4123 INIT_LIST_HEAD(&sop->so_stateids);
4124 sop->so_client = clp;
4125 init_nfs4_replay(&sop->so_replay);
4126 atomic_set(&sop->so_count, 1);
4127 return sop;
4128 }
4129
4130 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4131 {
4132 lockdep_assert_held(&clp->cl_lock);
4133
4134 list_add(&oo->oo_owner.so_strhash,
4135 &clp->cl_ownerstr_hashtbl[strhashval]);
4136 list_add(&oo->oo_perclient, &clp->cl_openowners);
4137 }
4138
4139 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4140 {
4141 unhash_openowner_locked(openowner(so));
4142 }
4143
4144 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4145 {
4146 struct nfs4_openowner *oo = openowner(so);
4147
4148 kmem_cache_free(openowner_slab, oo);
4149 }
4150
4151 static const struct nfs4_stateowner_operations openowner_ops = {
4152 .so_unhash = nfs4_unhash_openowner,
4153 .so_free = nfs4_free_openowner,
4154 };
4155
4156 static struct nfs4_ol_stateid *
4157 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4158 {
4159 struct nfs4_ol_stateid *local, *ret = NULL;
4160 struct nfs4_openowner *oo = open->op_openowner;
4161
4162 lockdep_assert_held(&fp->fi_lock);
4163
4164 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4165 /* ignore lock owners */
4166 if (local->st_stateowner->so_is_open_owner == 0)
4167 continue;
4168 if (local->st_stateowner != &oo->oo_owner)
4169 continue;
4170 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
4171 ret = local;
4172 refcount_inc(&ret->st_stid.sc_count);
4173 break;
4174 }
4175 }
4176 return ret;
4177 }
4178
4179 static __be32
4180 nfsd4_verify_open_stid(struct nfs4_stid *s)
4181 {
4182 __be32 ret = nfs_ok;
4183
4184 switch (s->sc_type) {
4185 default:
4186 break;
4187 case 0:
4188 case NFS4_CLOSED_STID:
4189 case NFS4_CLOSED_DELEG_STID:
4190 ret = nfserr_bad_stateid;
4191 break;
4192 case NFS4_REVOKED_DELEG_STID:
4193 ret = nfserr_deleg_revoked;
4194 }
4195 return ret;
4196 }
4197
4198 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4199 static __be32
4200 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4201 {
4202 __be32 ret;
4203
4204 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4205 ret = nfsd4_verify_open_stid(&stp->st_stid);
4206 if (ret != nfs_ok)
4207 mutex_unlock(&stp->st_mutex);
4208 return ret;
4209 }
4210
4211 static struct nfs4_ol_stateid *
4212 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4213 {
4214 struct nfs4_ol_stateid *stp;
4215 for (;;) {
4216 spin_lock(&fp->fi_lock);
4217 stp = nfsd4_find_existing_open(fp, open);
4218 spin_unlock(&fp->fi_lock);
4219 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4220 break;
4221 nfs4_put_stid(&stp->st_stid);
4222 }
4223 return stp;
4224 }
4225
4226 static struct nfs4_openowner *
4227 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4228 struct nfsd4_compound_state *cstate)
4229 {
4230 struct nfs4_client *clp = cstate->clp;
4231 struct nfs4_openowner *oo, *ret;
4232
4233 oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4234 if (!oo)
4235 return NULL;
4236 oo->oo_owner.so_ops = &openowner_ops;
4237 oo->oo_owner.so_is_open_owner = 1;
4238 oo->oo_owner.so_seqid = open->op_seqid;
4239 oo->oo_flags = 0;
4240 if (nfsd4_has_session(cstate))
4241 oo->oo_flags |= NFS4_OO_CONFIRMED;
4242 oo->oo_time = 0;
4243 oo->oo_last_closed_stid = NULL;
4244 INIT_LIST_HEAD(&oo->oo_close_lru);
4245 spin_lock(&clp->cl_lock);
4246 ret = find_openstateowner_str_locked(strhashval, open, clp);
4247 if (ret == NULL) {
4248 hash_openowner(oo, clp, strhashval);
4249 ret = oo;
4250 } else
4251 nfs4_free_stateowner(&oo->oo_owner);
4252
4253 spin_unlock(&clp->cl_lock);
4254 return ret;
4255 }
4256
4257 static struct nfs4_ol_stateid *
4258 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4259 {
4260
4261 struct nfs4_openowner *oo = open->op_openowner;
4262 struct nfs4_ol_stateid *retstp = NULL;
4263 struct nfs4_ol_stateid *stp;
4264
4265 stp = open->op_stp;
4266 /* We are moving these outside of the spinlocks to avoid the warnings */
4267 mutex_init(&stp->st_mutex);
4268 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4269
4270 retry:
4271 spin_lock(&oo->oo_owner.so_client->cl_lock);
4272 spin_lock(&fp->fi_lock);
4273
4274 retstp = nfsd4_find_existing_open(fp, open);
4275 if (retstp)
4276 goto out_unlock;
4277
4278 open->op_stp = NULL;
4279 refcount_inc(&stp->st_stid.sc_count);
4280 stp->st_stid.sc_type = NFS4_OPEN_STID;
4281 INIT_LIST_HEAD(&stp->st_locks);
4282 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4283 get_nfs4_file(fp);
4284 stp->st_stid.sc_file = fp;
4285 stp->st_access_bmap = 0;
4286 stp->st_deny_bmap = 0;
4287 stp->st_openstp = NULL;
4288 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4289 list_add(&stp->st_perfile, &fp->fi_stateids);
4290
4291 out_unlock:
4292 spin_unlock(&fp->fi_lock);
4293 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4294 if (retstp) {
4295 /* Handle races with CLOSE */
4296 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4297 nfs4_put_stid(&retstp->st_stid);
4298 goto retry;
4299 }
4300 /* To keep mutex tracking happy */
4301 mutex_unlock(&stp->st_mutex);
4302 stp = retstp;
4303 }
4304 return stp;
4305 }
4306
4307 /*
4308 * In the 4.0 case we need to keep the owners around a little while to handle
4309 * CLOSE replay. We still do need to release any file access that is held by
4310 * them before returning however.
4311 */
4312 static void
4313 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4314 {
4315 struct nfs4_ol_stateid *last;
4316 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4317 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4318 nfsd_net_id);
4319
4320 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4321
4322 /*
4323 * We know that we hold one reference via nfsd4_close, and another
4324 * "persistent" reference for the client. If the refcount is higher
4325 * than 2, then there are still calls in progress that are using this
4326 * stateid. We can't put the sc_file reference until they are finished.
4327 * Wait for the refcount to drop to 2. Since it has been unhashed,
4328 * there should be no danger of the refcount going back up again at
4329 * this point.
4330 */
4331 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
4332
4333 release_all_access(s);
4334 if (s->st_stid.sc_file) {
4335 put_nfs4_file(s->st_stid.sc_file);
4336 s->st_stid.sc_file = NULL;
4337 }
4338
4339 spin_lock(&nn->client_lock);
4340 last = oo->oo_last_closed_stid;
4341 oo->oo_last_closed_stid = s;
4342 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
4343 oo->oo_time = ktime_get_boottime_seconds();
4344 spin_unlock(&nn->client_lock);
4345 if (last)
4346 nfs4_put_stid(&last->st_stid);
4347 }
4348
4349 /* search file_hashtbl[] for file */
4350 static struct nfs4_file *
4351 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
4352 {
4353 struct nfs4_file *fp;
4354
4355 hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash,
4356 lockdep_is_held(&state_lock)) {
4357 if (fh_match(&fp->fi_fhandle, fh)) {
4358 if (refcount_inc_not_zero(&fp->fi_ref))
4359 return fp;
4360 }
4361 }
4362 return NULL;
4363 }
4364
4365 struct nfs4_file *
4366 find_file(struct knfsd_fh *fh)
4367 {
4368 struct nfs4_file *fp;
4369 unsigned int hashval = file_hashval(fh);
4370
4371 rcu_read_lock();
4372 fp = find_file_locked(fh, hashval);
4373 rcu_read_unlock();
4374 return fp;
4375 }
4376
4377 static struct nfs4_file *
4378 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
4379 {
4380 struct nfs4_file *fp;
4381 unsigned int hashval = file_hashval(fh);
4382
4383 rcu_read_lock();
4384 fp = find_file_locked(fh, hashval);
4385 rcu_read_unlock();
4386 if (fp)
4387 return fp;
4388
4389 spin_lock(&state_lock);
4390 fp = find_file_locked(fh, hashval);
4391 if (likely(fp == NULL)) {
4392 nfsd4_init_file(fh, hashval, new);
4393 fp = new;
4394 }
4395 spin_unlock(&state_lock);
4396
4397 return fp;
4398 }
4399
4400 /*
4401 * Called to check deny when READ with all zero stateid or
4402 * WRITE with all zero or all one stateid
4403 */
4404 static __be32
4405 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
4406 {
4407 struct nfs4_file *fp;
4408 __be32 ret = nfs_ok;
4409
4410 fp = find_file(&current_fh->fh_handle);
4411 if (!fp)
4412 return ret;
4413 /* Check for conflicting share reservations */
4414 spin_lock(&fp->fi_lock);
4415 if (fp->fi_share_deny & deny_type)
4416 ret = nfserr_locked;
4417 spin_unlock(&fp->fi_lock);
4418 put_nfs4_file(fp);
4419 return ret;
4420 }
4421
4422 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
4423 {
4424 struct nfs4_delegation *dp = cb_to_delegation(cb);
4425 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
4426 nfsd_net_id);
4427
4428 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
4429
4430 /*
4431 * We can't do this in nfsd_break_deleg_cb because it is
4432 * already holding inode->i_lock.
4433 *
4434 * If the dl_time != 0, then we know that it has already been
4435 * queued for a lease break. Don't queue it again.
4436 */
4437 spin_lock(&state_lock);
4438 if (dp->dl_time == 0) {
4439 dp->dl_time = ktime_get_boottime_seconds();
4440 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
4441 }
4442 spin_unlock(&state_lock);
4443 }
4444
4445 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
4446 struct rpc_task *task)
4447 {
4448 struct nfs4_delegation *dp = cb_to_delegation(cb);
4449
4450 if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
4451 return 1;
4452
4453 switch (task->tk_status) {
4454 case 0:
4455 return 1;
4456 case -NFS4ERR_DELAY:
4457 rpc_delay(task, 2 * HZ);
4458 return 0;
4459 case -EBADHANDLE:
4460 case -NFS4ERR_BAD_STATEID:
4461 /*
4462 * Race: client probably got cb_recall before open reply
4463 * granting delegation.
4464 */
4465 if (dp->dl_retries--) {
4466 rpc_delay(task, 2 * HZ);
4467 return 0;
4468 }
4469 /*FALLTHRU*/
4470 default:
4471 return 1;
4472 }
4473 }
4474
4475 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
4476 {
4477 struct nfs4_delegation *dp = cb_to_delegation(cb);
4478
4479 nfs4_put_stid(&dp->dl_stid);
4480 }
4481
4482 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
4483 .prepare = nfsd4_cb_recall_prepare,
4484 .done = nfsd4_cb_recall_done,
4485 .release = nfsd4_cb_recall_release,
4486 };
4487
4488 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
4489 {
4490 /*
4491 * We're assuming the state code never drops its reference
4492 * without first removing the lease. Since we're in this lease
4493 * callback (and since the lease code is serialized by the
4494 * i_lock) we know the server hasn't removed the lease yet, and
4495 * we know it's safe to take a reference.
4496 */
4497 refcount_inc(&dp->dl_stid.sc_count);
4498 nfsd4_run_cb(&dp->dl_recall);
4499 }
4500
4501 /* Called from break_lease() with i_lock held. */
4502 static bool
4503 nfsd_break_deleg_cb(struct file_lock *fl)
4504 {
4505 bool ret = false;
4506 struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
4507 struct nfs4_file *fp = dp->dl_stid.sc_file;
4508
4509 /*
4510 * We don't want the locks code to timeout the lease for us;
4511 * we'll remove it ourself if a delegation isn't returned
4512 * in time:
4513 */
4514 fl->fl_break_time = 0;
4515
4516 spin_lock(&fp->fi_lock);
4517 fp->fi_had_conflict = true;
4518 nfsd_break_one_deleg(dp);
4519 spin_unlock(&fp->fi_lock);
4520 return ret;
4521 }
4522
4523 static int
4524 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
4525 struct list_head *dispose)
4526 {
4527 if (arg & F_UNLCK)
4528 return lease_modify(onlist, arg, dispose);
4529 else
4530 return -EAGAIN;
4531 }
4532
4533 static const struct lock_manager_operations nfsd_lease_mng_ops = {
4534 .lm_break = nfsd_break_deleg_cb,
4535 .lm_change = nfsd_change_deleg_cb,
4536 };
4537
4538 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
4539 {
4540 if (nfsd4_has_session(cstate))
4541 return nfs_ok;
4542 if (seqid == so->so_seqid - 1)
4543 return nfserr_replay_me;
4544 if (seqid == so->so_seqid)
4545 return nfs_ok;
4546 return nfserr_bad_seqid;
4547 }
4548
4549 static __be32 lookup_clientid(clientid_t *clid,
4550 struct nfsd4_compound_state *cstate,
4551 struct nfsd_net *nn,
4552 bool sessions)
4553 {
4554 struct nfs4_client *found;
4555
4556 if (cstate->clp) {
4557 found = cstate->clp;
4558 if (!same_clid(&found->cl_clientid, clid))
4559 return nfserr_stale_clientid;
4560 return nfs_ok;
4561 }
4562
4563 if (STALE_CLIENTID(clid, nn))
4564 return nfserr_stale_clientid;
4565
4566 /*
4567 * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
4568 * cached already then we know this is for is for v4.0 and "sessions"
4569 * will be false.
4570 */
4571 WARN_ON_ONCE(cstate->session);
4572 spin_lock(&nn->client_lock);
4573 found = find_confirmed_client(clid, sessions, nn);
4574 if (!found) {
4575 spin_unlock(&nn->client_lock);
4576 return nfserr_expired;
4577 }
4578 atomic_inc(&found->cl_rpc_users);
4579 spin_unlock(&nn->client_lock);
4580
4581 /* Cache the nfs4_client in cstate! */
4582 cstate->clp = found;
4583 return nfs_ok;
4584 }
4585
4586 __be32
4587 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4588 struct nfsd4_open *open, struct nfsd_net *nn)
4589 {
4590 clientid_t *clientid = &open->op_clientid;
4591 struct nfs4_client *clp = NULL;
4592 unsigned int strhashval;
4593 struct nfs4_openowner *oo = NULL;
4594 __be32 status;
4595
4596 if (STALE_CLIENTID(&open->op_clientid, nn))
4597 return nfserr_stale_clientid;
4598 /*
4599 * In case we need it later, after we've already created the
4600 * file and don't want to risk a further failure:
4601 */
4602 open->op_file = nfsd4_alloc_file();
4603 if (open->op_file == NULL)
4604 return nfserr_jukebox;
4605
4606 status = lookup_clientid(clientid, cstate, nn, false);
4607 if (status)
4608 return status;
4609 clp = cstate->clp;
4610
4611 strhashval = ownerstr_hashval(&open->op_owner);
4612 oo = find_openstateowner_str(strhashval, open, clp);
4613 open->op_openowner = oo;
4614 if (!oo) {
4615 goto new_owner;
4616 }
4617 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4618 /* Replace unconfirmed owners without checking for replay. */
4619 release_openowner(oo);
4620 open->op_openowner = NULL;
4621 goto new_owner;
4622 }
4623 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4624 if (status)
4625 return status;
4626 goto alloc_stateid;
4627 new_owner:
4628 oo = alloc_init_open_stateowner(strhashval, open, cstate);
4629 if (oo == NULL)
4630 return nfserr_jukebox;
4631 open->op_openowner = oo;
4632 alloc_stateid:
4633 open->op_stp = nfs4_alloc_open_stateid(clp);
4634 if (!open->op_stp)
4635 return nfserr_jukebox;
4636
4637 if (nfsd4_has_session(cstate) &&
4638 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4639 open->op_odstate = alloc_clnt_odstate(clp);
4640 if (!open->op_odstate)
4641 return nfserr_jukebox;
4642 }
4643
4644 return nfs_ok;
4645 }
4646
4647 static inline __be32
4648 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4649 {
4650 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4651 return nfserr_openmode;
4652 else
4653 return nfs_ok;
4654 }
4655
4656 static int share_access_to_flags(u32 share_access)
4657 {
4658 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4659 }
4660
4661 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4662 {
4663 struct nfs4_stid *ret;
4664
4665 ret = find_stateid_by_type(cl, s,
4666 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4667 if (!ret)
4668 return NULL;
4669 return delegstateid(ret);
4670 }
4671
4672 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4673 {
4674 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4675 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4676 }
4677
4678 static __be32
4679 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4680 struct nfs4_delegation **dp)
4681 {
4682 int flags;
4683 __be32 status = nfserr_bad_stateid;
4684 struct nfs4_delegation *deleg;
4685
4686 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4687 if (deleg == NULL)
4688 goto out;
4689 if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4690 nfs4_put_stid(&deleg->dl_stid);
4691 if (cl->cl_minorversion)
4692 status = nfserr_deleg_revoked;
4693 goto out;
4694 }
4695 flags = share_access_to_flags(open->op_share_access);
4696 status = nfs4_check_delegmode(deleg, flags);
4697 if (status) {
4698 nfs4_put_stid(&deleg->dl_stid);
4699 goto out;
4700 }
4701 *dp = deleg;
4702 out:
4703 if (!nfsd4_is_deleg_cur(open))
4704 return nfs_ok;
4705 if (status)
4706 return status;
4707 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4708 return nfs_ok;
4709 }
4710
4711 static inline int nfs4_access_to_access(u32 nfs4_access)
4712 {
4713 int flags = 0;
4714
4715 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4716 flags |= NFSD_MAY_READ;
4717 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4718 flags |= NFSD_MAY_WRITE;
4719 return flags;
4720 }
4721
4722 static inline __be32
4723 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4724 struct nfsd4_open *open)
4725 {
4726 struct iattr iattr = {
4727 .ia_valid = ATTR_SIZE,
4728 .ia_size = 0,
4729 };
4730 if (!open->op_truncate)
4731 return 0;
4732 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4733 return nfserr_inval;
4734 return nfsd_setattr(rqstp, fh, &iattr, 0, (time64_t)0);
4735 }
4736
4737 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4738 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4739 struct nfsd4_open *open)
4740 {
4741 struct nfsd_file *nf = NULL;
4742 __be32 status;
4743 int oflag = nfs4_access_to_omode(open->op_share_access);
4744 int access = nfs4_access_to_access(open->op_share_access);
4745 unsigned char old_access_bmap, old_deny_bmap;
4746
4747 spin_lock(&fp->fi_lock);
4748
4749 /*
4750 * Are we trying to set a deny mode that would conflict with
4751 * current access?
4752 */
4753 status = nfs4_file_check_deny(fp, open->op_share_deny);
4754 if (status != nfs_ok) {
4755 spin_unlock(&fp->fi_lock);
4756 goto out;
4757 }
4758
4759 /* set access to the file */
4760 status = nfs4_file_get_access(fp, open->op_share_access);
4761 if (status != nfs_ok) {
4762 spin_unlock(&fp->fi_lock);
4763 goto out;
4764 }
4765
4766 /* Set access bits in stateid */
4767 old_access_bmap = stp->st_access_bmap;
4768 set_access(open->op_share_access, stp);
4769
4770 /* Set new deny mask */
4771 old_deny_bmap = stp->st_deny_bmap;
4772 set_deny(open->op_share_deny, stp);
4773 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4774
4775 if (!fp->fi_fds[oflag]) {
4776 spin_unlock(&fp->fi_lock);
4777 status = nfsd_file_acquire(rqstp, cur_fh, access, &nf);
4778 if (status)
4779 goto out_put_access;
4780 spin_lock(&fp->fi_lock);
4781 if (!fp->fi_fds[oflag]) {
4782 fp->fi_fds[oflag] = nf;
4783 nf = NULL;
4784 }
4785 }
4786 spin_unlock(&fp->fi_lock);
4787 if (nf)
4788 nfsd_file_put(nf);
4789
4790 status = nfsd4_truncate(rqstp, cur_fh, open);
4791 if (status)
4792 goto out_put_access;
4793 out:
4794 return status;
4795 out_put_access:
4796 stp->st_access_bmap = old_access_bmap;
4797 nfs4_file_put_access(fp, open->op_share_access);
4798 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4799 goto out;
4800 }
4801
4802 static __be32
4803 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4804 {
4805 __be32 status;
4806 unsigned char old_deny_bmap = stp->st_deny_bmap;
4807
4808 if (!test_access(open->op_share_access, stp))
4809 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4810
4811 /* test and set deny mode */
4812 spin_lock(&fp->fi_lock);
4813 status = nfs4_file_check_deny(fp, open->op_share_deny);
4814 if (status == nfs_ok) {
4815 set_deny(open->op_share_deny, stp);
4816 fp->fi_share_deny |=
4817 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4818 }
4819 spin_unlock(&fp->fi_lock);
4820
4821 if (status != nfs_ok)
4822 return status;
4823
4824 status = nfsd4_truncate(rqstp, cur_fh, open);
4825 if (status != nfs_ok)
4826 reset_union_bmap_deny(old_deny_bmap, stp);
4827 return status;
4828 }
4829
4830 /* Should we give out recallable state?: */
4831 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4832 {
4833 if (clp->cl_cb_state == NFSD4_CB_UP)
4834 return true;
4835 /*
4836 * In the sessions case, since we don't have to establish a
4837 * separate connection for callbacks, we assume it's OK
4838 * until we hear otherwise:
4839 */
4840 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4841 }
4842
4843 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
4844 int flag)
4845 {
4846 struct file_lock *fl;
4847
4848 fl = locks_alloc_lock();
4849 if (!fl)
4850 return NULL;
4851 fl->fl_lmops = &nfsd_lease_mng_ops;
4852 fl->fl_flags = FL_DELEG;
4853 fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4854 fl->fl_end = OFFSET_MAX;
4855 fl->fl_owner = (fl_owner_t)dp;
4856 fl->fl_pid = current->tgid;
4857 fl->fl_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
4858 return fl;
4859 }
4860
4861 static struct nfs4_delegation *
4862 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4863 struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4864 {
4865 int status = 0;
4866 struct nfs4_delegation *dp;
4867 struct nfsd_file *nf;
4868 struct file_lock *fl;
4869
4870 /*
4871 * The fi_had_conflict and nfs_get_existing_delegation checks
4872 * here are just optimizations; we'll need to recheck them at
4873 * the end:
4874 */
4875 if (fp->fi_had_conflict)
4876 return ERR_PTR(-EAGAIN);
4877
4878 nf = find_readable_file(fp);
4879 if (!nf) {
4880 /* We should always have a readable file here */
4881 WARN_ON_ONCE(1);
4882 return ERR_PTR(-EBADF);
4883 }
4884 spin_lock(&state_lock);
4885 spin_lock(&fp->fi_lock);
4886 if (nfs4_delegation_exists(clp, fp))
4887 status = -EAGAIN;
4888 else if (!fp->fi_deleg_file) {
4889 fp->fi_deleg_file = nf;
4890 /* increment early to prevent fi_deleg_file from being
4891 * cleared */
4892 fp->fi_delegees = 1;
4893 nf = NULL;
4894 } else
4895 fp->fi_delegees++;
4896 spin_unlock(&fp->fi_lock);
4897 spin_unlock(&state_lock);
4898 if (nf)
4899 nfsd_file_put(nf);
4900 if (status)
4901 return ERR_PTR(status);
4902
4903 status = -ENOMEM;
4904 dp = alloc_init_deleg(clp, fp, fh, odstate);
4905 if (!dp)
4906 goto out_delegees;
4907
4908 fl = nfs4_alloc_init_lease(dp, NFS4_OPEN_DELEGATE_READ);
4909 if (!fl)
4910 goto out_clnt_odstate;
4911
4912 status = vfs_setlease(fp->fi_deleg_file->nf_file, fl->fl_type, &fl, NULL);
4913 if (fl)
4914 locks_free_lock(fl);
4915 if (status)
4916 goto out_clnt_odstate;
4917
4918 spin_lock(&state_lock);
4919 spin_lock(&fp->fi_lock);
4920 if (fp->fi_had_conflict)
4921 status = -EAGAIN;
4922 else
4923 status = hash_delegation_locked(dp, fp);
4924 spin_unlock(&fp->fi_lock);
4925 spin_unlock(&state_lock);
4926
4927 if (status)
4928 goto out_unlock;
4929
4930 return dp;
4931 out_unlock:
4932 vfs_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
4933 out_clnt_odstate:
4934 put_clnt_odstate(dp->dl_clnt_odstate);
4935 nfs4_put_stid(&dp->dl_stid);
4936 out_delegees:
4937 put_deleg_file(fp);
4938 return ERR_PTR(status);
4939 }
4940
4941 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4942 {
4943 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4944 if (status == -EAGAIN)
4945 open->op_why_no_deleg = WND4_CONTENTION;
4946 else {
4947 open->op_why_no_deleg = WND4_RESOURCE;
4948 switch (open->op_deleg_want) {
4949 case NFS4_SHARE_WANT_READ_DELEG:
4950 case NFS4_SHARE_WANT_WRITE_DELEG:
4951 case NFS4_SHARE_WANT_ANY_DELEG:
4952 break;
4953 case NFS4_SHARE_WANT_CANCEL:
4954 open->op_why_no_deleg = WND4_CANCELLED;
4955 break;
4956 case NFS4_SHARE_WANT_NO_DELEG:
4957 WARN_ON_ONCE(1);
4958 }
4959 }
4960 }
4961
4962 /*
4963 * Attempt to hand out a delegation.
4964 *
4965 * Note we don't support write delegations, and won't until the vfs has
4966 * proper support for them.
4967 */
4968 static void
4969 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4970 struct nfs4_ol_stateid *stp)
4971 {
4972 struct nfs4_delegation *dp;
4973 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4974 struct nfs4_client *clp = stp->st_stid.sc_client;
4975 int cb_up;
4976 int status = 0;
4977
4978 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4979 open->op_recall = 0;
4980 switch (open->op_claim_type) {
4981 case NFS4_OPEN_CLAIM_PREVIOUS:
4982 if (!cb_up)
4983 open->op_recall = 1;
4984 if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4985 goto out_no_deleg;
4986 break;
4987 case NFS4_OPEN_CLAIM_NULL:
4988 case NFS4_OPEN_CLAIM_FH:
4989 /*
4990 * Let's not give out any delegations till everyone's
4991 * had the chance to reclaim theirs, *and* until
4992 * NLM locks have all been reclaimed:
4993 */
4994 if (locks_in_grace(clp->net))
4995 goto out_no_deleg;
4996 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4997 goto out_no_deleg;
4998 /*
4999 * Also, if the file was opened for write or
5000 * create, there's a good chance the client's
5001 * about to write to it, resulting in an
5002 * immediate recall (since we don't support
5003 * write delegations):
5004 */
5005 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
5006 goto out_no_deleg;
5007 if (open->op_create == NFS4_OPEN_CREATE)
5008 goto out_no_deleg;
5009 break;
5010 default:
5011 goto out_no_deleg;
5012 }
5013 dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
5014 if (IS_ERR(dp))
5015 goto out_no_deleg;
5016
5017 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5018
5019 dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
5020 STATEID_VAL(&dp->dl_stid.sc_stateid));
5021 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
5022 nfs4_put_stid(&dp->dl_stid);
5023 return;
5024 out_no_deleg:
5025 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
5026 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
5027 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
5028 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
5029 open->op_recall = 1;
5030 }
5031
5032 /* 4.1 client asking for a delegation? */
5033 if (open->op_deleg_want)
5034 nfsd4_open_deleg_none_ext(open, status);
5035 return;
5036 }
5037
5038 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
5039 struct nfs4_delegation *dp)
5040 {
5041 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
5042 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5043 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5044 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
5045 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
5046 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
5047 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5048 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
5049 }
5050 /* Otherwise the client must be confused wanting a delegation
5051 * it already has, therefore we don't return
5052 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
5053 */
5054 }
5055
5056 __be32
5057 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
5058 {
5059 struct nfsd4_compoundres *resp = rqstp->rq_resp;
5060 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
5061 struct nfs4_file *fp = NULL;
5062 struct nfs4_ol_stateid *stp = NULL;
5063 struct nfs4_delegation *dp = NULL;
5064 __be32 status;
5065 bool new_stp = false;
5066
5067 /*
5068 * Lookup file; if found, lookup stateid and check open request,
5069 * and check for delegations in the process of being recalled.
5070 * If not found, create the nfs4_file struct
5071 */
5072 fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
5073 if (fp != open->op_file) {
5074 status = nfs4_check_deleg(cl, open, &dp);
5075 if (status)
5076 goto out;
5077 stp = nfsd4_find_and_lock_existing_open(fp, open);
5078 } else {
5079 open->op_file = NULL;
5080 status = nfserr_bad_stateid;
5081 if (nfsd4_is_deleg_cur(open))
5082 goto out;
5083 }
5084
5085 if (!stp) {
5086 stp = init_open_stateid(fp, open);
5087 if (!open->op_stp)
5088 new_stp = true;
5089 }
5090
5091 /*
5092 * OPEN the file, or upgrade an existing OPEN.
5093 * If truncate fails, the OPEN fails.
5094 *
5095 * stp is already locked.
5096 */
5097 if (!new_stp) {
5098 /* Stateid was found, this is an OPEN upgrade */
5099 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
5100 if (status) {
5101 mutex_unlock(&stp->st_mutex);
5102 goto out;
5103 }
5104 } else {
5105 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
5106 if (status) {
5107 stp->st_stid.sc_type = NFS4_CLOSED_STID;
5108 release_open_stateid(stp);
5109 mutex_unlock(&stp->st_mutex);
5110 goto out;
5111 }
5112
5113 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
5114 open->op_odstate);
5115 if (stp->st_clnt_odstate == open->op_odstate)
5116 open->op_odstate = NULL;
5117 }
5118
5119 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
5120 mutex_unlock(&stp->st_mutex);
5121
5122 if (nfsd4_has_session(&resp->cstate)) {
5123 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
5124 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5125 open->op_why_no_deleg = WND4_NOT_WANTED;
5126 goto nodeleg;
5127 }
5128 }
5129
5130 /*
5131 * Attempt to hand out a delegation. No error return, because the
5132 * OPEN succeeds even if we fail.
5133 */
5134 nfs4_open_delegation(current_fh, open, stp);
5135 nodeleg:
5136 status = nfs_ok;
5137
5138 dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
5139 STATEID_VAL(&stp->st_stid.sc_stateid));
5140 out:
5141 /* 4.1 client trying to upgrade/downgrade delegation? */
5142 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
5143 open->op_deleg_want)
5144 nfsd4_deleg_xgrade_none_ext(open, dp);
5145
5146 if (fp)
5147 put_nfs4_file(fp);
5148 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
5149 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5150 /*
5151 * To finish the open response, we just need to set the rflags.
5152 */
5153 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
5154 if (nfsd4_has_session(&resp->cstate))
5155 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
5156 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
5157 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
5158
5159 if (dp)
5160 nfs4_put_stid(&dp->dl_stid);
5161 if (stp)
5162 nfs4_put_stid(&stp->st_stid);
5163
5164 return status;
5165 }
5166
5167 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
5168 struct nfsd4_open *open)
5169 {
5170 if (open->op_openowner) {
5171 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
5172
5173 nfsd4_cstate_assign_replay(cstate, so);
5174 nfs4_put_stateowner(so);
5175 }
5176 if (open->op_file)
5177 kmem_cache_free(file_slab, open->op_file);
5178 if (open->op_stp)
5179 nfs4_put_stid(&open->op_stp->st_stid);
5180 if (open->op_odstate)
5181 kmem_cache_free(odstate_slab, open->op_odstate);
5182 }
5183
5184 __be32
5185 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5186 union nfsd4_op_u *u)
5187 {
5188 clientid_t *clid = &u->renew;
5189 struct nfs4_client *clp;
5190 __be32 status;
5191 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5192
5193 dprintk("process_renew(%08x/%08x): starting\n",
5194 clid->cl_boot, clid->cl_id);
5195 status = lookup_clientid(clid, cstate, nn, false);
5196 if (status)
5197 goto out;
5198 clp = cstate->clp;
5199 status = nfserr_cb_path_down;
5200 if (!list_empty(&clp->cl_delegations)
5201 && clp->cl_cb_state != NFSD4_CB_UP)
5202 goto out;
5203 status = nfs_ok;
5204 out:
5205 return status;
5206 }
5207
5208 void
5209 nfsd4_end_grace(struct nfsd_net *nn)
5210 {
5211 /* do nothing if grace period already ended */
5212 if (nn->grace_ended)
5213 return;
5214
5215 nn->grace_ended = true;
5216 /*
5217 * If the server goes down again right now, an NFSv4
5218 * client will still be allowed to reclaim after it comes back up,
5219 * even if it hasn't yet had a chance to reclaim state this time.
5220 *
5221 */
5222 nfsd4_record_grace_done(nn);
5223 /*
5224 * At this point, NFSv4 clients can still reclaim. But if the
5225 * server crashes, any that have not yet reclaimed will be out
5226 * of luck on the next boot.
5227 *
5228 * (NFSv4.1+ clients are considered to have reclaimed once they
5229 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
5230 * have reclaimed after their first OPEN.)
5231 */
5232 locks_end_grace(&nn->nfsd4_manager);
5233 /*
5234 * At this point, and once lockd and/or any other containers
5235 * exit their grace period, further reclaims will fail and
5236 * regular locking can resume.
5237 */
5238 }
5239
5240 /*
5241 * If we've waited a lease period but there are still clients trying to
5242 * reclaim, wait a little longer to give them a chance to finish.
5243 */
5244 static bool clients_still_reclaiming(struct nfsd_net *nn)
5245 {
5246 time64_t double_grace_period_end = nn->boot_time +
5247 2 * nn->nfsd4_lease;
5248
5249 if (nn->track_reclaim_completes &&
5250 atomic_read(&nn->nr_reclaim_complete) ==
5251 nn->reclaim_str_hashtbl_size)
5252 return false;
5253 if (!nn->somebody_reclaimed)
5254 return false;
5255 nn->somebody_reclaimed = false;
5256 /*
5257 * If we've given them *two* lease times to reclaim, and they're
5258 * still not done, give up:
5259 */
5260 if (ktime_get_boottime_seconds() > double_grace_period_end)
5261 return false;
5262 return true;
5263 }
5264
5265 static time64_t
5266 nfs4_laundromat(struct nfsd_net *nn)
5267 {
5268 struct nfs4_client *clp;
5269 struct nfs4_openowner *oo;
5270 struct nfs4_delegation *dp;
5271 struct nfs4_ol_stateid *stp;
5272 struct nfsd4_blocked_lock *nbl;
5273 struct list_head *pos, *next, reaplist;
5274 time64_t cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease;
5275 time64_t t, new_timeo = nn->nfsd4_lease;
5276 struct nfs4_cpntf_state *cps;
5277 copy_stateid_t *cps_t;
5278 int i;
5279
5280 dprintk("NFSD: laundromat service - starting\n");
5281
5282 if (clients_still_reclaiming(nn)) {
5283 new_timeo = 0;
5284 goto out;
5285 }
5286 dprintk("NFSD: end of grace period\n");
5287 nfsd4_end_grace(nn);
5288 INIT_LIST_HEAD(&reaplist);
5289
5290 spin_lock(&nn->s2s_cp_lock);
5291 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
5292 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
5293 if (cps->cp_stateid.sc_type == NFS4_COPYNOTIFY_STID &&
5294 cps->cpntf_time > cutoff)
5295 _free_cpntf_state_locked(nn, cps);
5296 }
5297 spin_unlock(&nn->s2s_cp_lock);
5298
5299 spin_lock(&nn->client_lock);
5300 list_for_each_safe(pos, next, &nn->client_lru) {
5301 clp = list_entry(pos, struct nfs4_client, cl_lru);
5302 if (clp->cl_time > cutoff) {
5303 t = clp->cl_time - cutoff;
5304 new_timeo = min(new_timeo, t);
5305 break;
5306 }
5307 if (mark_client_expired_locked(clp)) {
5308 dprintk("NFSD: client in use (clientid %08x)\n",
5309 clp->cl_clientid.cl_id);
5310 continue;
5311 }
5312 list_add(&clp->cl_lru, &reaplist);
5313 }
5314 spin_unlock(&nn->client_lock);
5315 list_for_each_safe(pos, next, &reaplist) {
5316 clp = list_entry(pos, struct nfs4_client, cl_lru);
5317 dprintk("NFSD: purging unused client (clientid %08x)\n",
5318 clp->cl_clientid.cl_id);
5319 list_del_init(&clp->cl_lru);
5320 expire_client(clp);
5321 }
5322 spin_lock(&state_lock);
5323 list_for_each_safe(pos, next, &nn->del_recall_lru) {
5324 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5325 if (dp->dl_time > cutoff) {
5326 t = dp->dl_time - cutoff;
5327 new_timeo = min(new_timeo, t);
5328 break;
5329 }
5330 WARN_ON(!unhash_delegation_locked(dp));
5331 list_add(&dp->dl_recall_lru, &reaplist);
5332 }
5333 spin_unlock(&state_lock);
5334 while (!list_empty(&reaplist)) {
5335 dp = list_first_entry(&reaplist, struct nfs4_delegation,
5336 dl_recall_lru);
5337 list_del_init(&dp->dl_recall_lru);
5338 revoke_delegation(dp);
5339 }
5340
5341 spin_lock(&nn->client_lock);
5342 while (!list_empty(&nn->close_lru)) {
5343 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
5344 oo_close_lru);
5345 if (oo->oo_time > cutoff) {
5346 t = oo->oo_time - cutoff;
5347 new_timeo = min(new_timeo, t);
5348 break;
5349 }
5350 list_del_init(&oo->oo_close_lru);
5351 stp = oo->oo_last_closed_stid;
5352 oo->oo_last_closed_stid = NULL;
5353 spin_unlock(&nn->client_lock);
5354 nfs4_put_stid(&stp->st_stid);
5355 spin_lock(&nn->client_lock);
5356 }
5357 spin_unlock(&nn->client_lock);
5358
5359 /*
5360 * It's possible for a client to try and acquire an already held lock
5361 * that is being held for a long time, and then lose interest in it.
5362 * So, we clean out any un-revisited request after a lease period
5363 * under the assumption that the client is no longer interested.
5364 *
5365 * RFC5661, sec. 9.6 states that the client must not rely on getting
5366 * notifications and must continue to poll for locks, even when the
5367 * server supports them. Thus this shouldn't lead to clients blocking
5368 * indefinitely once the lock does become free.
5369 */
5370 BUG_ON(!list_empty(&reaplist));
5371 spin_lock(&nn->blocked_locks_lock);
5372 while (!list_empty(&nn->blocked_locks_lru)) {
5373 nbl = list_first_entry(&nn->blocked_locks_lru,
5374 struct nfsd4_blocked_lock, nbl_lru);
5375 if (nbl->nbl_time > cutoff) {
5376 t = nbl->nbl_time - cutoff;
5377 new_timeo = min(new_timeo, t);
5378 break;
5379 }
5380 list_move(&nbl->nbl_lru, &reaplist);
5381 list_del_init(&nbl->nbl_list);
5382 }
5383 spin_unlock(&nn->blocked_locks_lock);
5384
5385 while (!list_empty(&reaplist)) {
5386 nbl = list_first_entry(&reaplist,
5387 struct nfsd4_blocked_lock, nbl_lru);
5388 list_del_init(&nbl->nbl_lru);
5389 free_blocked_lock(nbl);
5390 }
5391 out:
5392 new_timeo = max_t(time64_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
5393 return new_timeo;
5394 }
5395
5396 static struct workqueue_struct *laundry_wq;
5397 static void laundromat_main(struct work_struct *);
5398
5399 static void
5400 laundromat_main(struct work_struct *laundry)
5401 {
5402 time64_t t;
5403 struct delayed_work *dwork = to_delayed_work(laundry);
5404 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
5405 laundromat_work);
5406
5407 t = nfs4_laundromat(nn);
5408 dprintk("NFSD: laundromat_main - sleeping for %lld seconds\n", t);
5409 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
5410 }
5411
5412 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
5413 {
5414 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
5415 return nfserr_bad_stateid;
5416 return nfs_ok;
5417 }
5418
5419 static inline int
5420 access_permit_read(struct nfs4_ol_stateid *stp)
5421 {
5422 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
5423 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
5424 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
5425 }
5426
5427 static inline int
5428 access_permit_write(struct nfs4_ol_stateid *stp)
5429 {
5430 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
5431 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
5432 }
5433
5434 static
5435 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
5436 {
5437 __be32 status = nfserr_openmode;
5438
5439 /* For lock stateid's, we test the parent open, not the lock: */
5440 if (stp->st_openstp)
5441 stp = stp->st_openstp;
5442 if ((flags & WR_STATE) && !access_permit_write(stp))
5443 goto out;
5444 if ((flags & RD_STATE) && !access_permit_read(stp))
5445 goto out;
5446 status = nfs_ok;
5447 out:
5448 return status;
5449 }
5450
5451 static inline __be32
5452 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
5453 {
5454 if (ONE_STATEID(stateid) && (flags & RD_STATE))
5455 return nfs_ok;
5456 else if (opens_in_grace(net)) {
5457 /* Answer in remaining cases depends on existence of
5458 * conflicting state; so we must wait out the grace period. */
5459 return nfserr_grace;
5460 } else if (flags & WR_STATE)
5461 return nfs4_share_conflict(current_fh,
5462 NFS4_SHARE_DENY_WRITE);
5463 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
5464 return nfs4_share_conflict(current_fh,
5465 NFS4_SHARE_DENY_READ);
5466 }
5467
5468 /*
5469 * Allow READ/WRITE during grace period on recovered state only for files
5470 * that are not able to provide mandatory locking.
5471 */
5472 static inline int
5473 grace_disallows_io(struct net *net, struct inode *inode)
5474 {
5475 return opens_in_grace(net) && mandatory_lock(inode);
5476 }
5477
5478 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
5479 {
5480 /*
5481 * When sessions are used the stateid generation number is ignored
5482 * when it is zero.
5483 */
5484 if (has_session && in->si_generation == 0)
5485 return nfs_ok;
5486
5487 if (in->si_generation == ref->si_generation)
5488 return nfs_ok;
5489
5490 /* If the client sends us a stateid from the future, it's buggy: */
5491 if (nfsd4_stateid_generation_after(in, ref))
5492 return nfserr_bad_stateid;
5493 /*
5494 * However, we could see a stateid from the past, even from a
5495 * non-buggy client. For example, if the client sends a lock
5496 * while some IO is outstanding, the lock may bump si_generation
5497 * while the IO is still in flight. The client could avoid that
5498 * situation by waiting for responses on all the IO requests,
5499 * but better performance may result in retrying IO that
5500 * receives an old_stateid error if requests are rarely
5501 * reordered in flight:
5502 */
5503 return nfserr_old_stateid;
5504 }
5505
5506 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
5507 {
5508 __be32 ret;
5509
5510 spin_lock(&s->sc_lock);
5511 ret = nfsd4_verify_open_stid(s);
5512 if (ret == nfs_ok)
5513 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
5514 spin_unlock(&s->sc_lock);
5515 return ret;
5516 }
5517
5518 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
5519 {
5520 if (ols->st_stateowner->so_is_open_owner &&
5521 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
5522 return nfserr_bad_stateid;
5523 return nfs_ok;
5524 }
5525
5526 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
5527 {
5528 struct nfs4_stid *s;
5529 __be32 status = nfserr_bad_stateid;
5530
5531 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5532 CLOSE_STATEID(stateid))
5533 return status;
5534 if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid))
5535 return status;
5536 spin_lock(&cl->cl_lock);
5537 s = find_stateid_locked(cl, stateid);
5538 if (!s)
5539 goto out_unlock;
5540 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
5541 if (status)
5542 goto out_unlock;
5543 switch (s->sc_type) {
5544 case NFS4_DELEG_STID:
5545 status = nfs_ok;
5546 break;
5547 case NFS4_REVOKED_DELEG_STID:
5548 status = nfserr_deleg_revoked;
5549 break;
5550 case NFS4_OPEN_STID:
5551 case NFS4_LOCK_STID:
5552 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
5553 break;
5554 default:
5555 printk("unknown stateid type %x\n", s->sc_type);
5556 /* Fallthrough */
5557 case NFS4_CLOSED_STID:
5558 case NFS4_CLOSED_DELEG_STID:
5559 status = nfserr_bad_stateid;
5560 }
5561 out_unlock:
5562 spin_unlock(&cl->cl_lock);
5563 return status;
5564 }
5565
5566 __be32
5567 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
5568 stateid_t *stateid, unsigned char typemask,
5569 struct nfs4_stid **s, struct nfsd_net *nn)
5570 {
5571 __be32 status;
5572 bool return_revoked = false;
5573
5574 /*
5575 * only return revoked delegations if explicitly asked.
5576 * otherwise we report revoked or bad_stateid status.
5577 */
5578 if (typemask & NFS4_REVOKED_DELEG_STID)
5579 return_revoked = true;
5580 else if (typemask & NFS4_DELEG_STID)
5581 typemask |= NFS4_REVOKED_DELEG_STID;
5582
5583 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5584 CLOSE_STATEID(stateid))
5585 return nfserr_bad_stateid;
5586 status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn,
5587 false);
5588 if (status == nfserr_stale_clientid) {
5589 if (cstate->session)
5590 return nfserr_bad_stateid;
5591 return nfserr_stale_stateid;
5592 }
5593 if (status)
5594 return status;
5595 *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5596 if (!*s)
5597 return nfserr_bad_stateid;
5598 if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5599 nfs4_put_stid(*s);
5600 if (cstate->minorversion)
5601 return nfserr_deleg_revoked;
5602 return nfserr_bad_stateid;
5603 }
5604 return nfs_ok;
5605 }
5606
5607 static struct nfsd_file *
5608 nfs4_find_file(struct nfs4_stid *s, int flags)
5609 {
5610 if (!s)
5611 return NULL;
5612
5613 switch (s->sc_type) {
5614 case NFS4_DELEG_STID:
5615 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5616 return NULL;
5617 return nfsd_file_get(s->sc_file->fi_deleg_file);
5618 case NFS4_OPEN_STID:
5619 case NFS4_LOCK_STID:
5620 if (flags & RD_STATE)
5621 return find_readable_file(s->sc_file);
5622 else
5623 return find_writeable_file(s->sc_file);
5624 break;
5625 }
5626
5627 return NULL;
5628 }
5629
5630 static __be32
5631 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
5632 {
5633 __be32 status;
5634
5635 status = nfsd4_check_openowner_confirmed(ols);
5636 if (status)
5637 return status;
5638 return nfs4_check_openmode(ols, flags);
5639 }
5640
5641 static __be32
5642 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5643 struct nfsd_file **nfp, int flags)
5644 {
5645 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5646 struct nfsd_file *nf;
5647 __be32 status;
5648
5649 nf = nfs4_find_file(s, flags);
5650 if (nf) {
5651 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5652 acc | NFSD_MAY_OWNER_OVERRIDE);
5653 if (status) {
5654 nfsd_file_put(nf);
5655 goto out;
5656 }
5657 } else {
5658 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
5659 if (status)
5660 return status;
5661 }
5662 *nfp = nf;
5663 out:
5664 return status;
5665 }
5666 static void
5667 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5668 {
5669 WARN_ON_ONCE(cps->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID);
5670 if (!refcount_dec_and_test(&cps->cp_stateid.sc_count))
5671 return;
5672 list_del(&cps->cp_list);
5673 idr_remove(&nn->s2s_cp_stateids,
5674 cps->cp_stateid.stid.si_opaque.so_id);
5675 kfree(cps);
5676 }
5677 /*
5678 * A READ from an inter server to server COPY will have a
5679 * copy stateid. Look up the copy notify stateid from the
5680 * idr structure and take a reference on it.
5681 */
5682 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5683 struct nfs4_client *clp,
5684 struct nfs4_cpntf_state **cps)
5685 {
5686 copy_stateid_t *cps_t;
5687 struct nfs4_cpntf_state *state = NULL;
5688
5689 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
5690 return nfserr_bad_stateid;
5691 spin_lock(&nn->s2s_cp_lock);
5692 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
5693 if (cps_t) {
5694 state = container_of(cps_t, struct nfs4_cpntf_state,
5695 cp_stateid);
5696 if (state->cp_stateid.sc_type != NFS4_COPYNOTIFY_STID) {
5697 state = NULL;
5698 goto unlock;
5699 }
5700 if (!clp)
5701 refcount_inc(&state->cp_stateid.sc_count);
5702 else
5703 _free_cpntf_state_locked(nn, state);
5704 }
5705 unlock:
5706 spin_unlock(&nn->s2s_cp_lock);
5707 if (!state)
5708 return nfserr_bad_stateid;
5709 if (!clp && state)
5710 *cps = state;
5711 return 0;
5712 }
5713
5714 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
5715 struct nfs4_stid **stid)
5716 {
5717 __be32 status;
5718 struct nfs4_cpntf_state *cps = NULL;
5719 struct nfsd4_compound_state cstate;
5720
5721 status = manage_cpntf_state(nn, st, NULL, &cps);
5722 if (status)
5723 return status;
5724
5725 cps->cpntf_time = ktime_get_boottime_seconds();
5726 memset(&cstate, 0, sizeof(cstate));
5727 status = lookup_clientid(&cps->cp_p_clid, &cstate, nn, true);
5728 if (status)
5729 goto out;
5730 status = nfsd4_lookup_stateid(&cstate, &cps->cp_p_stateid,
5731 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5732 stid, nn);
5733 put_client_renew(cstate.clp);
5734 out:
5735 nfs4_put_cpntf_state(nn, cps);
5736 return status;
5737 }
5738
5739 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
5740 {
5741 spin_lock(&nn->s2s_cp_lock);
5742 _free_cpntf_state_locked(nn, cps);
5743 spin_unlock(&nn->s2s_cp_lock);
5744 }
5745
5746 /*
5747 * Checks for stateid operations
5748 */
5749 __be32
5750 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5751 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5752 stateid_t *stateid, int flags, struct nfsd_file **nfp,
5753 struct nfs4_stid **cstid)
5754 {
5755 struct inode *ino = d_inode(fhp->fh_dentry);
5756 struct net *net = SVC_NET(rqstp);
5757 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5758 struct nfs4_stid *s = NULL;
5759 __be32 status;
5760
5761 if (nfp)
5762 *nfp = NULL;
5763
5764 if (grace_disallows_io(net, ino))
5765 return nfserr_grace;
5766
5767 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5768 status = check_special_stateids(net, fhp, stateid, flags);
5769 goto done;
5770 }
5771
5772 status = nfsd4_lookup_stateid(cstate, stateid,
5773 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5774 &s, nn);
5775 if (status == nfserr_bad_stateid)
5776 status = find_cpntf_state(nn, stateid, &s);
5777 if (status)
5778 return status;
5779 status = nfsd4_stid_check_stateid_generation(stateid, s,
5780 nfsd4_has_session(cstate));
5781 if (status)
5782 goto out;
5783
5784 switch (s->sc_type) {
5785 case NFS4_DELEG_STID:
5786 status = nfs4_check_delegmode(delegstateid(s), flags);
5787 break;
5788 case NFS4_OPEN_STID:
5789 case NFS4_LOCK_STID:
5790 status = nfs4_check_olstateid(openlockstateid(s), flags);
5791 break;
5792 default:
5793 status = nfserr_bad_stateid;
5794 break;
5795 }
5796 if (status)
5797 goto out;
5798 status = nfs4_check_fh(fhp, s);
5799
5800 done:
5801 if (status == nfs_ok && nfp)
5802 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
5803 out:
5804 if (s) {
5805 if (!status && cstid)
5806 *cstid = s;
5807 else
5808 nfs4_put_stid(s);
5809 }
5810 return status;
5811 }
5812
5813 /*
5814 * Test if the stateid is valid
5815 */
5816 __be32
5817 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5818 union nfsd4_op_u *u)
5819 {
5820 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5821 struct nfsd4_test_stateid_id *stateid;
5822 struct nfs4_client *cl = cstate->session->se_client;
5823
5824 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5825 stateid->ts_id_status =
5826 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5827
5828 return nfs_ok;
5829 }
5830
5831 static __be32
5832 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5833 {
5834 struct nfs4_ol_stateid *stp = openlockstateid(s);
5835 __be32 ret;
5836
5837 ret = nfsd4_lock_ol_stateid(stp);
5838 if (ret)
5839 goto out_put_stid;
5840
5841 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5842 if (ret)
5843 goto out;
5844
5845 ret = nfserr_locks_held;
5846 if (check_for_locks(stp->st_stid.sc_file,
5847 lockowner(stp->st_stateowner)))
5848 goto out;
5849
5850 release_lock_stateid(stp);
5851 ret = nfs_ok;
5852
5853 out:
5854 mutex_unlock(&stp->st_mutex);
5855 out_put_stid:
5856 nfs4_put_stid(s);
5857 return ret;
5858 }
5859
5860 __be32
5861 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5862 union nfsd4_op_u *u)
5863 {
5864 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5865 stateid_t *stateid = &free_stateid->fr_stateid;
5866 struct nfs4_stid *s;
5867 struct nfs4_delegation *dp;
5868 struct nfs4_client *cl = cstate->session->se_client;
5869 __be32 ret = nfserr_bad_stateid;
5870
5871 spin_lock(&cl->cl_lock);
5872 s = find_stateid_locked(cl, stateid);
5873 if (!s)
5874 goto out_unlock;
5875 spin_lock(&s->sc_lock);
5876 switch (s->sc_type) {
5877 case NFS4_DELEG_STID:
5878 ret = nfserr_locks_held;
5879 break;
5880 case NFS4_OPEN_STID:
5881 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5882 if (ret)
5883 break;
5884 ret = nfserr_locks_held;
5885 break;
5886 case NFS4_LOCK_STID:
5887 spin_unlock(&s->sc_lock);
5888 refcount_inc(&s->sc_count);
5889 spin_unlock(&cl->cl_lock);
5890 ret = nfsd4_free_lock_stateid(stateid, s);
5891 goto out;
5892 case NFS4_REVOKED_DELEG_STID:
5893 spin_unlock(&s->sc_lock);
5894 dp = delegstateid(s);
5895 list_del_init(&dp->dl_recall_lru);
5896 spin_unlock(&cl->cl_lock);
5897 nfs4_put_stid(s);
5898 ret = nfs_ok;
5899 goto out;
5900 /* Default falls through and returns nfserr_bad_stateid */
5901 }
5902 spin_unlock(&s->sc_lock);
5903 out_unlock:
5904 spin_unlock(&cl->cl_lock);
5905 out:
5906 return ret;
5907 }
5908
5909 static inline int
5910 setlkflg (int type)
5911 {
5912 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5913 RD_STATE : WR_STATE;
5914 }
5915
5916 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5917 {
5918 struct svc_fh *current_fh = &cstate->current_fh;
5919 struct nfs4_stateowner *sop = stp->st_stateowner;
5920 __be32 status;
5921
5922 status = nfsd4_check_seqid(cstate, sop, seqid);
5923 if (status)
5924 return status;
5925 status = nfsd4_lock_ol_stateid(stp);
5926 if (status != nfs_ok)
5927 return status;
5928 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5929 if (status == nfs_ok)
5930 status = nfs4_check_fh(current_fh, &stp->st_stid);
5931 if (status != nfs_ok)
5932 mutex_unlock(&stp->st_mutex);
5933 return status;
5934 }
5935
5936 /*
5937 * Checks for sequence id mutating operations.
5938 */
5939 static __be32
5940 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5941 stateid_t *stateid, char typemask,
5942 struct nfs4_ol_stateid **stpp,
5943 struct nfsd_net *nn)
5944 {
5945 __be32 status;
5946 struct nfs4_stid *s;
5947 struct nfs4_ol_stateid *stp = NULL;
5948
5949 dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5950 seqid, STATEID_VAL(stateid));
5951
5952 *stpp = NULL;
5953 status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5954 if (status)
5955 return status;
5956 stp = openlockstateid(s);
5957 nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5958
5959 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5960 if (!status)
5961 *stpp = stp;
5962 else
5963 nfs4_put_stid(&stp->st_stid);
5964 return status;
5965 }
5966
5967 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5968 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5969 {
5970 __be32 status;
5971 struct nfs4_openowner *oo;
5972 struct nfs4_ol_stateid *stp;
5973
5974 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5975 NFS4_OPEN_STID, &stp, nn);
5976 if (status)
5977 return status;
5978 oo = openowner(stp->st_stateowner);
5979 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5980 mutex_unlock(&stp->st_mutex);
5981 nfs4_put_stid(&stp->st_stid);
5982 return nfserr_bad_stateid;
5983 }
5984 *stpp = stp;
5985 return nfs_ok;
5986 }
5987
5988 __be32
5989 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5990 union nfsd4_op_u *u)
5991 {
5992 struct nfsd4_open_confirm *oc = &u->open_confirm;
5993 __be32 status;
5994 struct nfs4_openowner *oo;
5995 struct nfs4_ol_stateid *stp;
5996 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5997
5998 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5999 cstate->current_fh.fh_dentry);
6000
6001 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
6002 if (status)
6003 return status;
6004
6005 status = nfs4_preprocess_seqid_op(cstate,
6006 oc->oc_seqid, &oc->oc_req_stateid,
6007 NFS4_OPEN_STID, &stp, nn);
6008 if (status)
6009 goto out;
6010 oo = openowner(stp->st_stateowner);
6011 status = nfserr_bad_stateid;
6012 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
6013 mutex_unlock(&stp->st_mutex);
6014 goto put_stateid;
6015 }
6016 oo->oo_flags |= NFS4_OO_CONFIRMED;
6017 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
6018 mutex_unlock(&stp->st_mutex);
6019 dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
6020 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
6021
6022 nfsd4_client_record_create(oo->oo_owner.so_client);
6023 status = nfs_ok;
6024 put_stateid:
6025 nfs4_put_stid(&stp->st_stid);
6026 out:
6027 nfsd4_bump_seqid(cstate, status);
6028 return status;
6029 }
6030
6031 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
6032 {
6033 if (!test_access(access, stp))
6034 return;
6035 nfs4_file_put_access(stp->st_stid.sc_file, access);
6036 clear_access(access, stp);
6037 }
6038
6039 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
6040 {
6041 switch (to_access) {
6042 case NFS4_SHARE_ACCESS_READ:
6043 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
6044 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6045 break;
6046 case NFS4_SHARE_ACCESS_WRITE:
6047 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
6048 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
6049 break;
6050 case NFS4_SHARE_ACCESS_BOTH:
6051 break;
6052 default:
6053 WARN_ON_ONCE(1);
6054 }
6055 }
6056
6057 __be32
6058 nfsd4_open_downgrade(struct svc_rqst *rqstp,
6059 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
6060 {
6061 struct nfsd4_open_downgrade *od = &u->open_downgrade;
6062 __be32 status;
6063 struct nfs4_ol_stateid *stp;
6064 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6065
6066 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
6067 cstate->current_fh.fh_dentry);
6068
6069 /* We don't yet support WANT bits: */
6070 if (od->od_deleg_want)
6071 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
6072 od->od_deleg_want);
6073
6074 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
6075 &od->od_stateid, &stp, nn);
6076 if (status)
6077 goto out;
6078 status = nfserr_inval;
6079 if (!test_access(od->od_share_access, stp)) {
6080 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
6081 stp->st_access_bmap, od->od_share_access);
6082 goto put_stateid;
6083 }
6084 if (!test_deny(od->od_share_deny, stp)) {
6085 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
6086 stp->st_deny_bmap, od->od_share_deny);
6087 goto put_stateid;
6088 }
6089 nfs4_stateid_downgrade(stp, od->od_share_access);
6090 reset_union_bmap_deny(od->od_share_deny, stp);
6091 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
6092 status = nfs_ok;
6093 put_stateid:
6094 mutex_unlock(&stp->st_mutex);
6095 nfs4_put_stid(&stp->st_stid);
6096 out:
6097 nfsd4_bump_seqid(cstate, status);
6098 return status;
6099 }
6100
6101 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
6102 {
6103 struct nfs4_client *clp = s->st_stid.sc_client;
6104 bool unhashed;
6105 LIST_HEAD(reaplist);
6106
6107 spin_lock(&clp->cl_lock);
6108 unhashed = unhash_open_stateid(s, &reaplist);
6109
6110 if (clp->cl_minorversion) {
6111 if (unhashed)
6112 put_ol_stateid_locked(s, &reaplist);
6113 spin_unlock(&clp->cl_lock);
6114 free_ol_stateid_reaplist(&reaplist);
6115 } else {
6116 spin_unlock(&clp->cl_lock);
6117 free_ol_stateid_reaplist(&reaplist);
6118 if (unhashed)
6119 move_to_close_lru(s, clp->net);
6120 }
6121 }
6122
6123 /*
6124 * nfs4_unlock_state() called after encode
6125 */
6126 __be32
6127 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6128 union nfsd4_op_u *u)
6129 {
6130 struct nfsd4_close *close = &u->close;
6131 __be32 status;
6132 struct nfs4_ol_stateid *stp;
6133 struct net *net = SVC_NET(rqstp);
6134 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6135
6136 dprintk("NFSD: nfsd4_close on file %pd\n",
6137 cstate->current_fh.fh_dentry);
6138
6139 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
6140 &close->cl_stateid,
6141 NFS4_OPEN_STID|NFS4_CLOSED_STID,
6142 &stp, nn);
6143 nfsd4_bump_seqid(cstate, status);
6144 if (status)
6145 goto out;
6146
6147 stp->st_stid.sc_type = NFS4_CLOSED_STID;
6148
6149 /*
6150 * Technically we don't _really_ have to increment or copy it, since
6151 * it should just be gone after this operation and we clobber the
6152 * copied value below, but we continue to do so here just to ensure
6153 * that racing ops see that there was a state change.
6154 */
6155 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
6156
6157 nfsd4_close_open_stateid(stp);
6158 mutex_unlock(&stp->st_mutex);
6159
6160 /* v4.1+ suggests that we send a special stateid in here, since the
6161 * clients should just ignore this anyway. Since this is not useful
6162 * for v4.0 clients either, we set it to the special close_stateid
6163 * universally.
6164 *
6165 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
6166 */
6167 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
6168
6169 /* put reference from nfs4_preprocess_seqid_op */
6170 nfs4_put_stid(&stp->st_stid);
6171 out:
6172 return status;
6173 }
6174
6175 __be32
6176 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6177 union nfsd4_op_u *u)
6178 {
6179 struct nfsd4_delegreturn *dr = &u->delegreturn;
6180 struct nfs4_delegation *dp;
6181 stateid_t *stateid = &dr->dr_stateid;
6182 struct nfs4_stid *s;
6183 __be32 status;
6184 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6185
6186 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6187 return status;
6188
6189 status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
6190 if (status)
6191 goto out;
6192 dp = delegstateid(s);
6193 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
6194 if (status)
6195 goto put_stateid;
6196
6197 destroy_delegation(dp);
6198 put_stateid:
6199 nfs4_put_stid(&dp->dl_stid);
6200 out:
6201 return status;
6202 }
6203
6204 static inline u64
6205 end_offset(u64 start, u64 len)
6206 {
6207 u64 end;
6208
6209 end = start + len;
6210 return end >= start ? end: NFS4_MAX_UINT64;
6211 }
6212
6213 /* last octet in a range */
6214 static inline u64
6215 last_byte_offset(u64 start, u64 len)
6216 {
6217 u64 end;
6218
6219 WARN_ON_ONCE(!len);
6220 end = start + len;
6221 return end > start ? end - 1: NFS4_MAX_UINT64;
6222 }
6223
6224 /*
6225 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
6226 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
6227 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
6228 * locking, this prevents us from being completely protocol-compliant. The
6229 * real solution to this problem is to start using unsigned file offsets in
6230 * the VFS, but this is a very deep change!
6231 */
6232 static inline void
6233 nfs4_transform_lock_offset(struct file_lock *lock)
6234 {
6235 if (lock->fl_start < 0)
6236 lock->fl_start = OFFSET_MAX;
6237 if (lock->fl_end < 0)
6238 lock->fl_end = OFFSET_MAX;
6239 }
6240
6241 static fl_owner_t
6242 nfsd4_fl_get_owner(fl_owner_t owner)
6243 {
6244 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6245
6246 nfs4_get_stateowner(&lo->lo_owner);
6247 return owner;
6248 }
6249
6250 static void
6251 nfsd4_fl_put_owner(fl_owner_t owner)
6252 {
6253 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
6254
6255 if (lo)
6256 nfs4_put_stateowner(&lo->lo_owner);
6257 }
6258
6259 static void
6260 nfsd4_lm_notify(struct file_lock *fl)
6261 {
6262 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)fl->fl_owner;
6263 struct net *net = lo->lo_owner.so_client->net;
6264 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6265 struct nfsd4_blocked_lock *nbl = container_of(fl,
6266 struct nfsd4_blocked_lock, nbl_lock);
6267 bool queue = false;
6268
6269 /* An empty list means that something else is going to be using it */
6270 spin_lock(&nn->blocked_locks_lock);
6271 if (!list_empty(&nbl->nbl_list)) {
6272 list_del_init(&nbl->nbl_list);
6273 list_del_init(&nbl->nbl_lru);
6274 queue = true;
6275 }
6276 spin_unlock(&nn->blocked_locks_lock);
6277
6278 if (queue)
6279 nfsd4_run_cb(&nbl->nbl_cb);
6280 }
6281
6282 static const struct lock_manager_operations nfsd_posix_mng_ops = {
6283 .lm_notify = nfsd4_lm_notify,
6284 .lm_get_owner = nfsd4_fl_get_owner,
6285 .lm_put_owner = nfsd4_fl_put_owner,
6286 };
6287
6288 static inline void
6289 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
6290 {
6291 struct nfs4_lockowner *lo;
6292
6293 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
6294 lo = (struct nfs4_lockowner *) fl->fl_owner;
6295 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
6296 GFP_KERNEL);
6297 if (!deny->ld_owner.data)
6298 /* We just don't care that much */
6299 goto nevermind;
6300 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
6301 } else {
6302 nevermind:
6303 deny->ld_owner.len = 0;
6304 deny->ld_owner.data = NULL;
6305 deny->ld_clientid.cl_boot = 0;
6306 deny->ld_clientid.cl_id = 0;
6307 }
6308 deny->ld_start = fl->fl_start;
6309 deny->ld_length = NFS4_MAX_UINT64;
6310 if (fl->fl_end != NFS4_MAX_UINT64)
6311 deny->ld_length = fl->fl_end - fl->fl_start + 1;
6312 deny->ld_type = NFS4_READ_LT;
6313 if (fl->fl_type != F_RDLCK)
6314 deny->ld_type = NFS4_WRITE_LT;
6315 }
6316
6317 static struct nfs4_lockowner *
6318 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
6319 {
6320 unsigned int strhashval = ownerstr_hashval(owner);
6321 struct nfs4_stateowner *so;
6322
6323 lockdep_assert_held(&clp->cl_lock);
6324
6325 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
6326 so_strhash) {
6327 if (so->so_is_open_owner)
6328 continue;
6329 if (same_owner_str(so, owner))
6330 return lockowner(nfs4_get_stateowner(so));
6331 }
6332 return NULL;
6333 }
6334
6335 static struct nfs4_lockowner *
6336 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
6337 {
6338 struct nfs4_lockowner *lo;
6339
6340 spin_lock(&clp->cl_lock);
6341 lo = find_lockowner_str_locked(clp, owner);
6342 spin_unlock(&clp->cl_lock);
6343 return lo;
6344 }
6345
6346 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
6347 {
6348 unhash_lockowner_locked(lockowner(sop));
6349 }
6350
6351 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
6352 {
6353 struct nfs4_lockowner *lo = lockowner(sop);
6354
6355 kmem_cache_free(lockowner_slab, lo);
6356 }
6357
6358 static const struct nfs4_stateowner_operations lockowner_ops = {
6359 .so_unhash = nfs4_unhash_lockowner,
6360 .so_free = nfs4_free_lockowner,
6361 };
6362
6363 /*
6364 * Alloc a lock owner structure.
6365 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
6366 * occurred.
6367 *
6368 * strhashval = ownerstr_hashval
6369 */
6370 static struct nfs4_lockowner *
6371 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
6372 struct nfs4_ol_stateid *open_stp,
6373 struct nfsd4_lock *lock)
6374 {
6375 struct nfs4_lockowner *lo, *ret;
6376
6377 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
6378 if (!lo)
6379 return NULL;
6380 INIT_LIST_HEAD(&lo->lo_blocked);
6381 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
6382 lo->lo_owner.so_is_open_owner = 0;
6383 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
6384 lo->lo_owner.so_ops = &lockowner_ops;
6385 spin_lock(&clp->cl_lock);
6386 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
6387 if (ret == NULL) {
6388 list_add(&lo->lo_owner.so_strhash,
6389 &clp->cl_ownerstr_hashtbl[strhashval]);
6390 ret = lo;
6391 } else
6392 nfs4_free_stateowner(&lo->lo_owner);
6393
6394 spin_unlock(&clp->cl_lock);
6395 return ret;
6396 }
6397
6398 static struct nfs4_ol_stateid *
6399 find_lock_stateid(const struct nfs4_lockowner *lo,
6400 const struct nfs4_ol_stateid *ost)
6401 {
6402 struct nfs4_ol_stateid *lst;
6403
6404 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
6405
6406 /* If ost is not hashed, ost->st_locks will not be valid */
6407 if (!nfs4_ol_stateid_unhashed(ost))
6408 list_for_each_entry(lst, &ost->st_locks, st_locks) {
6409 if (lst->st_stateowner == &lo->lo_owner) {
6410 refcount_inc(&lst->st_stid.sc_count);
6411 return lst;
6412 }
6413 }
6414 return NULL;
6415 }
6416
6417 static struct nfs4_ol_stateid *
6418 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
6419 struct nfs4_file *fp, struct inode *inode,
6420 struct nfs4_ol_stateid *open_stp)
6421 {
6422 struct nfs4_client *clp = lo->lo_owner.so_client;
6423 struct nfs4_ol_stateid *retstp;
6424
6425 mutex_init(&stp->st_mutex);
6426 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
6427 retry:
6428 spin_lock(&clp->cl_lock);
6429 if (nfs4_ol_stateid_unhashed(open_stp))
6430 goto out_close;
6431 retstp = find_lock_stateid(lo, open_stp);
6432 if (retstp)
6433 goto out_found;
6434 refcount_inc(&stp->st_stid.sc_count);
6435 stp->st_stid.sc_type = NFS4_LOCK_STID;
6436 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
6437 get_nfs4_file(fp);
6438 stp->st_stid.sc_file = fp;
6439 stp->st_access_bmap = 0;
6440 stp->st_deny_bmap = open_stp->st_deny_bmap;
6441 stp->st_openstp = open_stp;
6442 spin_lock(&fp->fi_lock);
6443 list_add(&stp->st_locks, &open_stp->st_locks);
6444 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
6445 list_add(&stp->st_perfile, &fp->fi_stateids);
6446 spin_unlock(&fp->fi_lock);
6447 spin_unlock(&clp->cl_lock);
6448 return stp;
6449 out_found:
6450 spin_unlock(&clp->cl_lock);
6451 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
6452 nfs4_put_stid(&retstp->st_stid);
6453 goto retry;
6454 }
6455 /* To keep mutex tracking happy */
6456 mutex_unlock(&stp->st_mutex);
6457 return retstp;
6458 out_close:
6459 spin_unlock(&clp->cl_lock);
6460 mutex_unlock(&stp->st_mutex);
6461 return NULL;
6462 }
6463
6464 static struct nfs4_ol_stateid *
6465 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
6466 struct inode *inode, struct nfs4_ol_stateid *ost,
6467 bool *new)
6468 {
6469 struct nfs4_stid *ns = NULL;
6470 struct nfs4_ol_stateid *lst;
6471 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6472 struct nfs4_client *clp = oo->oo_owner.so_client;
6473
6474 *new = false;
6475 spin_lock(&clp->cl_lock);
6476 lst = find_lock_stateid(lo, ost);
6477 spin_unlock(&clp->cl_lock);
6478 if (lst != NULL) {
6479 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
6480 goto out;
6481 nfs4_put_stid(&lst->st_stid);
6482 }
6483 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
6484 if (ns == NULL)
6485 return NULL;
6486
6487 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
6488 if (lst == openlockstateid(ns))
6489 *new = true;
6490 else
6491 nfs4_put_stid(ns);
6492 out:
6493 return lst;
6494 }
6495
6496 static int
6497 check_lock_length(u64 offset, u64 length)
6498 {
6499 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
6500 (length > ~offset)));
6501 }
6502
6503 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
6504 {
6505 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
6506
6507 lockdep_assert_held(&fp->fi_lock);
6508
6509 if (test_access(access, lock_stp))
6510 return;
6511 __nfs4_file_get_access(fp, access);
6512 set_access(access, lock_stp);
6513 }
6514
6515 static __be32
6516 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
6517 struct nfs4_ol_stateid *ost,
6518 struct nfsd4_lock *lock,
6519 struct nfs4_ol_stateid **plst, bool *new)
6520 {
6521 __be32 status;
6522 struct nfs4_file *fi = ost->st_stid.sc_file;
6523 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
6524 struct nfs4_client *cl = oo->oo_owner.so_client;
6525 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
6526 struct nfs4_lockowner *lo;
6527 struct nfs4_ol_stateid *lst;
6528 unsigned int strhashval;
6529
6530 lo = find_lockowner_str(cl, &lock->lk_new_owner);
6531 if (!lo) {
6532 strhashval = ownerstr_hashval(&lock->lk_new_owner);
6533 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
6534 if (lo == NULL)
6535 return nfserr_jukebox;
6536 } else {
6537 /* with an existing lockowner, seqids must be the same */
6538 status = nfserr_bad_seqid;
6539 if (!cstate->minorversion &&
6540 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
6541 goto out;
6542 }
6543
6544 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
6545 if (lst == NULL) {
6546 status = nfserr_jukebox;
6547 goto out;
6548 }
6549
6550 status = nfs_ok;
6551 *plst = lst;
6552 out:
6553 nfs4_put_stateowner(&lo->lo_owner);
6554 return status;
6555 }
6556
6557 /*
6558 * LOCK operation
6559 */
6560 __be32
6561 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6562 union nfsd4_op_u *u)
6563 {
6564 struct nfsd4_lock *lock = &u->lock;
6565 struct nfs4_openowner *open_sop = NULL;
6566 struct nfs4_lockowner *lock_sop = NULL;
6567 struct nfs4_ol_stateid *lock_stp = NULL;
6568 struct nfs4_ol_stateid *open_stp = NULL;
6569 struct nfs4_file *fp;
6570 struct nfsd_file *nf = NULL;
6571 struct nfsd4_blocked_lock *nbl = NULL;
6572 struct file_lock *file_lock = NULL;
6573 struct file_lock *conflock = NULL;
6574 __be32 status = 0;
6575 int lkflg;
6576 int err;
6577 bool new = false;
6578 unsigned char fl_type;
6579 unsigned int fl_flags = FL_POSIX;
6580 struct net *net = SVC_NET(rqstp);
6581 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6582
6583 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
6584 (long long) lock->lk_offset,
6585 (long long) lock->lk_length);
6586
6587 if (check_lock_length(lock->lk_offset, lock->lk_length))
6588 return nfserr_inval;
6589
6590 if ((status = fh_verify(rqstp, &cstate->current_fh,
6591 S_IFREG, NFSD_MAY_LOCK))) {
6592 dprintk("NFSD: nfsd4_lock: permission denied!\n");
6593 return status;
6594 }
6595
6596 if (lock->lk_is_new) {
6597 if (nfsd4_has_session(cstate))
6598 /* See rfc 5661 18.10.3: given clientid is ignored: */
6599 memcpy(&lock->lk_new_clientid,
6600 &cstate->session->se_client->cl_clientid,
6601 sizeof(clientid_t));
6602
6603 status = nfserr_stale_clientid;
6604 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
6605 goto out;
6606
6607 /* validate and update open stateid and open seqid */
6608 status = nfs4_preprocess_confirmed_seqid_op(cstate,
6609 lock->lk_new_open_seqid,
6610 &lock->lk_new_open_stateid,
6611 &open_stp, nn);
6612 if (status)
6613 goto out;
6614 mutex_unlock(&open_stp->st_mutex);
6615 open_sop = openowner(open_stp->st_stateowner);
6616 status = nfserr_bad_stateid;
6617 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
6618 &lock->lk_new_clientid))
6619 goto out;
6620 status = lookup_or_create_lock_state(cstate, open_stp, lock,
6621 &lock_stp, &new);
6622 } else {
6623 status = nfs4_preprocess_seqid_op(cstate,
6624 lock->lk_old_lock_seqid,
6625 &lock->lk_old_lock_stateid,
6626 NFS4_LOCK_STID, &lock_stp, nn);
6627 }
6628 if (status)
6629 goto out;
6630 lock_sop = lockowner(lock_stp->st_stateowner);
6631
6632 lkflg = setlkflg(lock->lk_type);
6633 status = nfs4_check_openmode(lock_stp, lkflg);
6634 if (status)
6635 goto out;
6636
6637 status = nfserr_grace;
6638 if (locks_in_grace(net) && !lock->lk_reclaim)
6639 goto out;
6640 status = nfserr_no_grace;
6641 if (!locks_in_grace(net) && lock->lk_reclaim)
6642 goto out;
6643
6644 fp = lock_stp->st_stid.sc_file;
6645 switch (lock->lk_type) {
6646 case NFS4_READW_LT:
6647 if (nfsd4_has_session(cstate))
6648 fl_flags |= FL_SLEEP;
6649 /* Fallthrough */
6650 case NFS4_READ_LT:
6651 spin_lock(&fp->fi_lock);
6652 nf = find_readable_file_locked(fp);
6653 if (nf)
6654 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
6655 spin_unlock(&fp->fi_lock);
6656 fl_type = F_RDLCK;
6657 break;
6658 case NFS4_WRITEW_LT:
6659 if (nfsd4_has_session(cstate))
6660 fl_flags |= FL_SLEEP;
6661 /* Fallthrough */
6662 case NFS4_WRITE_LT:
6663 spin_lock(&fp->fi_lock);
6664 nf = find_writeable_file_locked(fp);
6665 if (nf)
6666 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
6667 spin_unlock(&fp->fi_lock);
6668 fl_type = F_WRLCK;
6669 break;
6670 default:
6671 status = nfserr_inval;
6672 goto out;
6673 }
6674
6675 if (!nf) {
6676 status = nfserr_openmode;
6677 goto out;
6678 }
6679
6680 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6681 if (!nbl) {
6682 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6683 status = nfserr_jukebox;
6684 goto out;
6685 }
6686
6687 file_lock = &nbl->nbl_lock;
6688 file_lock->fl_type = fl_type;
6689 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6690 file_lock->fl_pid = current->tgid;
6691 file_lock->fl_file = nf->nf_file;
6692 file_lock->fl_flags = fl_flags;
6693 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6694 file_lock->fl_start = lock->lk_offset;
6695 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6696 nfs4_transform_lock_offset(file_lock);
6697
6698 conflock = locks_alloc_lock();
6699 if (!conflock) {
6700 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6701 status = nfserr_jukebox;
6702 goto out;
6703 }
6704
6705 if (fl_flags & FL_SLEEP) {
6706 nbl->nbl_time = ktime_get_boottime_seconds();
6707 spin_lock(&nn->blocked_locks_lock);
6708 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6709 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6710 spin_unlock(&nn->blocked_locks_lock);
6711 }
6712
6713 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
6714 switch (err) {
6715 case 0: /* success! */
6716 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6717 status = 0;
6718 if (lock->lk_reclaim)
6719 nn->somebody_reclaimed = true;
6720 break;
6721 case FILE_LOCK_DEFERRED:
6722 nbl = NULL;
6723 /* Fallthrough */
6724 case -EAGAIN: /* conflock holds conflicting lock */
6725 status = nfserr_denied;
6726 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6727 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6728 break;
6729 case -EDEADLK:
6730 status = nfserr_deadlock;
6731 break;
6732 default:
6733 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6734 status = nfserrno(err);
6735 break;
6736 }
6737 out:
6738 if (nbl) {
6739 /* dequeue it if we queued it before */
6740 if (fl_flags & FL_SLEEP) {
6741 spin_lock(&nn->blocked_locks_lock);
6742 list_del_init(&nbl->nbl_list);
6743 list_del_init(&nbl->nbl_lru);
6744 spin_unlock(&nn->blocked_locks_lock);
6745 }
6746 free_blocked_lock(nbl);
6747 }
6748 if (nf)
6749 nfsd_file_put(nf);
6750 if (lock_stp) {
6751 /* Bump seqid manually if the 4.0 replay owner is openowner */
6752 if (cstate->replay_owner &&
6753 cstate->replay_owner != &lock_sop->lo_owner &&
6754 seqid_mutating_err(ntohl(status)))
6755 lock_sop->lo_owner.so_seqid++;
6756
6757 /*
6758 * If this is a new, never-before-used stateid, and we are
6759 * returning an error, then just go ahead and release it.
6760 */
6761 if (status && new)
6762 release_lock_stateid(lock_stp);
6763
6764 mutex_unlock(&lock_stp->st_mutex);
6765
6766 nfs4_put_stid(&lock_stp->st_stid);
6767 }
6768 if (open_stp)
6769 nfs4_put_stid(&open_stp->st_stid);
6770 nfsd4_bump_seqid(cstate, status);
6771 if (conflock)
6772 locks_free_lock(conflock);
6773 return status;
6774 }
6775
6776 /*
6777 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6778 * so we do a temporary open here just to get an open file to pass to
6779 * vfs_test_lock. (Arguably perhaps test_lock should be done with an
6780 * inode operation.)
6781 */
6782 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6783 {
6784 struct nfsd_file *nf;
6785 __be32 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
6786 if (!err) {
6787 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
6788 nfsd_file_put(nf);
6789 }
6790 return err;
6791 }
6792
6793 /*
6794 * LOCKT operation
6795 */
6796 __be32
6797 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6798 union nfsd4_op_u *u)
6799 {
6800 struct nfsd4_lockt *lockt = &u->lockt;
6801 struct file_lock *file_lock = NULL;
6802 struct nfs4_lockowner *lo = NULL;
6803 __be32 status;
6804 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6805
6806 if (locks_in_grace(SVC_NET(rqstp)))
6807 return nfserr_grace;
6808
6809 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6810 return nfserr_inval;
6811
6812 if (!nfsd4_has_session(cstate)) {
6813 status = lookup_clientid(&lockt->lt_clientid, cstate, nn,
6814 false);
6815 if (status)
6816 goto out;
6817 }
6818
6819 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6820 goto out;
6821
6822 file_lock = locks_alloc_lock();
6823 if (!file_lock) {
6824 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6825 status = nfserr_jukebox;
6826 goto out;
6827 }
6828
6829 switch (lockt->lt_type) {
6830 case NFS4_READ_LT:
6831 case NFS4_READW_LT:
6832 file_lock->fl_type = F_RDLCK;
6833 break;
6834 case NFS4_WRITE_LT:
6835 case NFS4_WRITEW_LT:
6836 file_lock->fl_type = F_WRLCK;
6837 break;
6838 default:
6839 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6840 status = nfserr_inval;
6841 goto out;
6842 }
6843
6844 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6845 if (lo)
6846 file_lock->fl_owner = (fl_owner_t)lo;
6847 file_lock->fl_pid = current->tgid;
6848 file_lock->fl_flags = FL_POSIX;
6849
6850 file_lock->fl_start = lockt->lt_offset;
6851 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6852
6853 nfs4_transform_lock_offset(file_lock);
6854
6855 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6856 if (status)
6857 goto out;
6858
6859 if (file_lock->fl_type != F_UNLCK) {
6860 status = nfserr_denied;
6861 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6862 }
6863 out:
6864 if (lo)
6865 nfs4_put_stateowner(&lo->lo_owner);
6866 if (file_lock)
6867 locks_free_lock(file_lock);
6868 return status;
6869 }
6870
6871 __be32
6872 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6873 union nfsd4_op_u *u)
6874 {
6875 struct nfsd4_locku *locku = &u->locku;
6876 struct nfs4_ol_stateid *stp;
6877 struct nfsd_file *nf = NULL;
6878 struct file_lock *file_lock = NULL;
6879 __be32 status;
6880 int err;
6881 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6882
6883 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6884 (long long) locku->lu_offset,
6885 (long long) locku->lu_length);
6886
6887 if (check_lock_length(locku->lu_offset, locku->lu_length))
6888 return nfserr_inval;
6889
6890 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6891 &locku->lu_stateid, NFS4_LOCK_STID,
6892 &stp, nn);
6893 if (status)
6894 goto out;
6895 nf = find_any_file(stp->st_stid.sc_file);
6896 if (!nf) {
6897 status = nfserr_lock_range;
6898 goto put_stateid;
6899 }
6900 file_lock = locks_alloc_lock();
6901 if (!file_lock) {
6902 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6903 status = nfserr_jukebox;
6904 goto put_file;
6905 }
6906
6907 file_lock->fl_type = F_UNLCK;
6908 file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6909 file_lock->fl_pid = current->tgid;
6910 file_lock->fl_file = nf->nf_file;
6911 file_lock->fl_flags = FL_POSIX;
6912 file_lock->fl_lmops = &nfsd_posix_mng_ops;
6913 file_lock->fl_start = locku->lu_offset;
6914
6915 file_lock->fl_end = last_byte_offset(locku->lu_offset,
6916 locku->lu_length);
6917 nfs4_transform_lock_offset(file_lock);
6918
6919 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
6920 if (err) {
6921 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6922 goto out_nfserr;
6923 }
6924 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6925 put_file:
6926 nfsd_file_put(nf);
6927 put_stateid:
6928 mutex_unlock(&stp->st_mutex);
6929 nfs4_put_stid(&stp->st_stid);
6930 out:
6931 nfsd4_bump_seqid(cstate, status);
6932 if (file_lock)
6933 locks_free_lock(file_lock);
6934 return status;
6935
6936 out_nfserr:
6937 status = nfserrno(err);
6938 goto put_file;
6939 }
6940
6941 /*
6942 * returns
6943 * true: locks held by lockowner
6944 * false: no locks held by lockowner
6945 */
6946 static bool
6947 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6948 {
6949 struct file_lock *fl;
6950 int status = false;
6951 struct nfsd_file *nf = find_any_file(fp);
6952 struct inode *inode;
6953 struct file_lock_context *flctx;
6954
6955 if (!nf) {
6956 /* Any valid lock stateid should have some sort of access */
6957 WARN_ON_ONCE(1);
6958 return status;
6959 }
6960
6961 inode = locks_inode(nf->nf_file);
6962 flctx = inode->i_flctx;
6963
6964 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6965 spin_lock(&flctx->flc_lock);
6966 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6967 if (fl->fl_owner == (fl_owner_t)lowner) {
6968 status = true;
6969 break;
6970 }
6971 }
6972 spin_unlock(&flctx->flc_lock);
6973 }
6974 nfsd_file_put(nf);
6975 return status;
6976 }
6977
6978 __be32
6979 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6980 struct nfsd4_compound_state *cstate,
6981 union nfsd4_op_u *u)
6982 {
6983 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6984 clientid_t *clid = &rlockowner->rl_clientid;
6985 struct nfs4_stateowner *sop;
6986 struct nfs4_lockowner *lo = NULL;
6987 struct nfs4_ol_stateid *stp;
6988 struct xdr_netobj *owner = &rlockowner->rl_owner;
6989 unsigned int hashval = ownerstr_hashval(owner);
6990 __be32 status;
6991 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6992 struct nfs4_client *clp;
6993 LIST_HEAD (reaplist);
6994
6995 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6996 clid->cl_boot, clid->cl_id);
6997
6998 status = lookup_clientid(clid, cstate, nn, false);
6999 if (status)
7000 return status;
7001
7002 clp = cstate->clp;
7003 /* Find the matching lock stateowner */
7004 spin_lock(&clp->cl_lock);
7005 list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
7006 so_strhash) {
7007
7008 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
7009 continue;
7010
7011 /* see if there are still any locks associated with it */
7012 lo = lockowner(sop);
7013 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
7014 if (check_for_locks(stp->st_stid.sc_file, lo)) {
7015 status = nfserr_locks_held;
7016 spin_unlock(&clp->cl_lock);
7017 return status;
7018 }
7019 }
7020
7021 nfs4_get_stateowner(sop);
7022 break;
7023 }
7024 if (!lo) {
7025 spin_unlock(&clp->cl_lock);
7026 return status;
7027 }
7028
7029 unhash_lockowner_locked(lo);
7030 while (!list_empty(&lo->lo_owner.so_stateids)) {
7031 stp = list_first_entry(&lo->lo_owner.so_stateids,
7032 struct nfs4_ol_stateid,
7033 st_perstateowner);
7034 WARN_ON(!unhash_lock_stateid(stp));
7035 put_ol_stateid_locked(stp, &reaplist);
7036 }
7037 spin_unlock(&clp->cl_lock);
7038 free_ol_stateid_reaplist(&reaplist);
7039 remove_blocked_locks(lo);
7040 nfs4_put_stateowner(&lo->lo_owner);
7041
7042 return status;
7043 }
7044
7045 static inline struct nfs4_client_reclaim *
7046 alloc_reclaim(void)
7047 {
7048 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
7049 }
7050
7051 bool
7052 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
7053 {
7054 struct nfs4_client_reclaim *crp;
7055
7056 crp = nfsd4_find_reclaim_client(name, nn);
7057 return (crp && crp->cr_clp);
7058 }
7059
7060 /*
7061 * failure => all reset bets are off, nfserr_no_grace...
7062 *
7063 * The caller is responsible for freeing name.data if NULL is returned (it
7064 * will be freed in nfs4_remove_reclaim_record in the normal case).
7065 */
7066 struct nfs4_client_reclaim *
7067 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
7068 struct nfsd_net *nn)
7069 {
7070 unsigned int strhashval;
7071 struct nfs4_client_reclaim *crp;
7072
7073 dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", name.len, name.data);
7074 crp = alloc_reclaim();
7075 if (crp) {
7076 strhashval = clientstr_hashval(name);
7077 INIT_LIST_HEAD(&crp->cr_strhash);
7078 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
7079 crp->cr_name.data = name.data;
7080 crp->cr_name.len = name.len;
7081 crp->cr_princhash.data = princhash.data;
7082 crp->cr_princhash.len = princhash.len;
7083 crp->cr_clp = NULL;
7084 nn->reclaim_str_hashtbl_size++;
7085 }
7086 return crp;
7087 }
7088
7089 void
7090 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
7091 {
7092 list_del(&crp->cr_strhash);
7093 kfree(crp->cr_name.data);
7094 kfree(crp->cr_princhash.data);
7095 kfree(crp);
7096 nn->reclaim_str_hashtbl_size--;
7097 }
7098
7099 void
7100 nfs4_release_reclaim(struct nfsd_net *nn)
7101 {
7102 struct nfs4_client_reclaim *crp = NULL;
7103 int i;
7104
7105 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7106 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
7107 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
7108 struct nfs4_client_reclaim, cr_strhash);
7109 nfs4_remove_reclaim_record(crp, nn);
7110 }
7111 }
7112 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
7113 }
7114
7115 /*
7116 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
7117 struct nfs4_client_reclaim *
7118 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
7119 {
7120 unsigned int strhashval;
7121 struct nfs4_client_reclaim *crp = NULL;
7122
7123 dprintk("NFSD: nfs4_find_reclaim_client for name %.*s\n", name.len, name.data);
7124
7125 strhashval = clientstr_hashval(name);
7126 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
7127 if (compare_blob(&crp->cr_name, &name) == 0) {
7128 return crp;
7129 }
7130 }
7131 return NULL;
7132 }
7133
7134 /*
7135 * Called from OPEN. Look for clientid in reclaim list.
7136 */
7137 __be32
7138 nfs4_check_open_reclaim(clientid_t *clid,
7139 struct nfsd4_compound_state *cstate,
7140 struct nfsd_net *nn)
7141 {
7142 __be32 status;
7143
7144 /* find clientid in conf_id_hashtbl */
7145 status = lookup_clientid(clid, cstate, nn, false);
7146 if (status)
7147 return nfserr_reclaim_bad;
7148
7149 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
7150 return nfserr_no_grace;
7151
7152 if (nfsd4_client_record_check(cstate->clp))
7153 return nfserr_reclaim_bad;
7154
7155 return nfs_ok;
7156 }
7157
7158 #ifdef CONFIG_NFSD_FAULT_INJECTION
7159 static inline void
7160 put_client(struct nfs4_client *clp)
7161 {
7162 atomic_dec(&clp->cl_rpc_users);
7163 }
7164
7165 static struct nfs4_client *
7166 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
7167 {
7168 struct nfs4_client *clp;
7169 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7170 nfsd_net_id);
7171
7172 if (!nfsd_netns_ready(nn))
7173 return NULL;
7174
7175 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7176 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
7177 return clp;
7178 }
7179 return NULL;
7180 }
7181
7182 u64
7183 nfsd_inject_print_clients(void)
7184 {
7185 struct nfs4_client *clp;
7186 u64 count = 0;
7187 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7188 nfsd_net_id);
7189 char buf[INET6_ADDRSTRLEN];
7190
7191 if (!nfsd_netns_ready(nn))
7192 return 0;
7193
7194 spin_lock(&nn->client_lock);
7195 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7196 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7197 pr_info("NFS Client: %s\n", buf);
7198 ++count;
7199 }
7200 spin_unlock(&nn->client_lock);
7201
7202 return count;
7203 }
7204
7205 u64
7206 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
7207 {
7208 u64 count = 0;
7209 struct nfs4_client *clp;
7210 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7211 nfsd_net_id);
7212
7213 if (!nfsd_netns_ready(nn))
7214 return count;
7215
7216 spin_lock(&nn->client_lock);
7217 clp = nfsd_find_client(addr, addr_size);
7218 if (clp) {
7219 if (mark_client_expired_locked(clp) == nfs_ok)
7220 ++count;
7221 else
7222 clp = NULL;
7223 }
7224 spin_unlock(&nn->client_lock);
7225
7226 if (clp)
7227 expire_client(clp);
7228
7229 return count;
7230 }
7231
7232 u64
7233 nfsd_inject_forget_clients(u64 max)
7234 {
7235 u64 count = 0;
7236 struct nfs4_client *clp, *next;
7237 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7238 nfsd_net_id);
7239 LIST_HEAD(reaplist);
7240
7241 if (!nfsd_netns_ready(nn))
7242 return count;
7243
7244 spin_lock(&nn->client_lock);
7245 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7246 if (mark_client_expired_locked(clp) == nfs_ok) {
7247 list_add(&clp->cl_lru, &reaplist);
7248 if (max != 0 && ++count >= max)
7249 break;
7250 }
7251 }
7252 spin_unlock(&nn->client_lock);
7253
7254 list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
7255 expire_client(clp);
7256
7257 return count;
7258 }
7259
7260 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
7261 const char *type)
7262 {
7263 char buf[INET6_ADDRSTRLEN];
7264 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
7265 printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
7266 }
7267
7268 static void
7269 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
7270 struct list_head *collect)
7271 {
7272 struct nfs4_client *clp = lst->st_stid.sc_client;
7273 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7274 nfsd_net_id);
7275
7276 if (!collect)
7277 return;
7278
7279 lockdep_assert_held(&nn->client_lock);
7280 atomic_inc(&clp->cl_rpc_users);
7281 list_add(&lst->st_locks, collect);
7282 }
7283
7284 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
7285 struct list_head *collect,
7286 bool (*func)(struct nfs4_ol_stateid *))
7287 {
7288 struct nfs4_openowner *oop;
7289 struct nfs4_ol_stateid *stp, *st_next;
7290 struct nfs4_ol_stateid *lst, *lst_next;
7291 u64 count = 0;
7292
7293 spin_lock(&clp->cl_lock);
7294 list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
7295 list_for_each_entry_safe(stp, st_next,
7296 &oop->oo_owner.so_stateids, st_perstateowner) {
7297 list_for_each_entry_safe(lst, lst_next,
7298 &stp->st_locks, st_locks) {
7299 if (func) {
7300 if (func(lst))
7301 nfsd_inject_add_lock_to_list(lst,
7302 collect);
7303 }
7304 ++count;
7305 /*
7306 * Despite the fact that these functions deal
7307 * with 64-bit integers for "count", we must
7308 * ensure that it doesn't blow up the
7309 * clp->cl_rpc_users. Throw a warning if we
7310 * start to approach INT_MAX here.
7311 */
7312 WARN_ON_ONCE(count == (INT_MAX / 2));
7313 if (count == max)
7314 goto out;
7315 }
7316 }
7317 }
7318 out:
7319 spin_unlock(&clp->cl_lock);
7320
7321 return count;
7322 }
7323
7324 static u64
7325 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
7326 u64 max)
7327 {
7328 return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
7329 }
7330
7331 static u64
7332 nfsd_print_client_locks(struct nfs4_client *clp)
7333 {
7334 u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
7335 nfsd_print_count(clp, count, "locked files");
7336 return count;
7337 }
7338
7339 u64
7340 nfsd_inject_print_locks(void)
7341 {
7342 struct nfs4_client *clp;
7343 u64 count = 0;
7344 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7345 nfsd_net_id);
7346
7347 if (!nfsd_netns_ready(nn))
7348 return 0;
7349
7350 spin_lock(&nn->client_lock);
7351 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7352 count += nfsd_print_client_locks(clp);
7353 spin_unlock(&nn->client_lock);
7354
7355 return count;
7356 }
7357
7358 static void
7359 nfsd_reap_locks(struct list_head *reaplist)
7360 {
7361 struct nfs4_client *clp;
7362 struct nfs4_ol_stateid *stp, *next;
7363
7364 list_for_each_entry_safe(stp, next, reaplist, st_locks) {
7365 list_del_init(&stp->st_locks);
7366 clp = stp->st_stid.sc_client;
7367 nfs4_put_stid(&stp->st_stid);
7368 put_client(clp);
7369 }
7370 }
7371
7372 u64
7373 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
7374 {
7375 unsigned int count = 0;
7376 struct nfs4_client *clp;
7377 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7378 nfsd_net_id);
7379 LIST_HEAD(reaplist);
7380
7381 if (!nfsd_netns_ready(nn))
7382 return count;
7383
7384 spin_lock(&nn->client_lock);
7385 clp = nfsd_find_client(addr, addr_size);
7386 if (clp)
7387 count = nfsd_collect_client_locks(clp, &reaplist, 0);
7388 spin_unlock(&nn->client_lock);
7389 nfsd_reap_locks(&reaplist);
7390 return count;
7391 }
7392
7393 u64
7394 nfsd_inject_forget_locks(u64 max)
7395 {
7396 u64 count = 0;
7397 struct nfs4_client *clp;
7398 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7399 nfsd_net_id);
7400 LIST_HEAD(reaplist);
7401
7402 if (!nfsd_netns_ready(nn))
7403 return count;
7404
7405 spin_lock(&nn->client_lock);
7406 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7407 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
7408 if (max != 0 && count >= max)
7409 break;
7410 }
7411 spin_unlock(&nn->client_lock);
7412 nfsd_reap_locks(&reaplist);
7413 return count;
7414 }
7415
7416 static u64
7417 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
7418 struct list_head *collect,
7419 void (*func)(struct nfs4_openowner *))
7420 {
7421 struct nfs4_openowner *oop, *next;
7422 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7423 nfsd_net_id);
7424 u64 count = 0;
7425
7426 lockdep_assert_held(&nn->client_lock);
7427
7428 spin_lock(&clp->cl_lock);
7429 list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
7430 if (func) {
7431 func(oop);
7432 if (collect) {
7433 atomic_inc(&clp->cl_rpc_users);
7434 list_add(&oop->oo_perclient, collect);
7435 }
7436 }
7437 ++count;
7438 /*
7439 * Despite the fact that these functions deal with
7440 * 64-bit integers for "count", we must ensure that
7441 * it doesn't blow up the clp->cl_rpc_users. Throw a
7442 * warning if we start to approach INT_MAX here.
7443 */
7444 WARN_ON_ONCE(count == (INT_MAX / 2));
7445 if (count == max)
7446 break;
7447 }
7448 spin_unlock(&clp->cl_lock);
7449
7450 return count;
7451 }
7452
7453 static u64
7454 nfsd_print_client_openowners(struct nfs4_client *clp)
7455 {
7456 u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
7457
7458 nfsd_print_count(clp, count, "openowners");
7459 return count;
7460 }
7461
7462 static u64
7463 nfsd_collect_client_openowners(struct nfs4_client *clp,
7464 struct list_head *collect, u64 max)
7465 {
7466 return nfsd_foreach_client_openowner(clp, max, collect,
7467 unhash_openowner_locked);
7468 }
7469
7470 u64
7471 nfsd_inject_print_openowners(void)
7472 {
7473 struct nfs4_client *clp;
7474 u64 count = 0;
7475 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7476 nfsd_net_id);
7477
7478 if (!nfsd_netns_ready(nn))
7479 return 0;
7480
7481 spin_lock(&nn->client_lock);
7482 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7483 count += nfsd_print_client_openowners(clp);
7484 spin_unlock(&nn->client_lock);
7485
7486 return count;
7487 }
7488
7489 static void
7490 nfsd_reap_openowners(struct list_head *reaplist)
7491 {
7492 struct nfs4_client *clp;
7493 struct nfs4_openowner *oop, *next;
7494
7495 list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
7496 list_del_init(&oop->oo_perclient);
7497 clp = oop->oo_owner.so_client;
7498 release_openowner(oop);
7499 put_client(clp);
7500 }
7501 }
7502
7503 u64
7504 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
7505 size_t addr_size)
7506 {
7507 unsigned int count = 0;
7508 struct nfs4_client *clp;
7509 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7510 nfsd_net_id);
7511 LIST_HEAD(reaplist);
7512
7513 if (!nfsd_netns_ready(nn))
7514 return count;
7515
7516 spin_lock(&nn->client_lock);
7517 clp = nfsd_find_client(addr, addr_size);
7518 if (clp)
7519 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
7520 spin_unlock(&nn->client_lock);
7521 nfsd_reap_openowners(&reaplist);
7522 return count;
7523 }
7524
7525 u64
7526 nfsd_inject_forget_openowners(u64 max)
7527 {
7528 u64 count = 0;
7529 struct nfs4_client *clp;
7530 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7531 nfsd_net_id);
7532 LIST_HEAD(reaplist);
7533
7534 if (!nfsd_netns_ready(nn))
7535 return count;
7536
7537 spin_lock(&nn->client_lock);
7538 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7539 count += nfsd_collect_client_openowners(clp, &reaplist,
7540 max - count);
7541 if (max != 0 && count >= max)
7542 break;
7543 }
7544 spin_unlock(&nn->client_lock);
7545 nfsd_reap_openowners(&reaplist);
7546 return count;
7547 }
7548
7549 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
7550 struct list_head *victims)
7551 {
7552 struct nfs4_delegation *dp, *next;
7553 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7554 nfsd_net_id);
7555 u64 count = 0;
7556
7557 lockdep_assert_held(&nn->client_lock);
7558
7559 spin_lock(&state_lock);
7560 list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
7561 if (victims) {
7562 /*
7563 * It's not safe to mess with delegations that have a
7564 * non-zero dl_time. They might have already been broken
7565 * and could be processed by the laundromat outside of
7566 * the state_lock. Just leave them be.
7567 */
7568 if (dp->dl_time != 0)
7569 continue;
7570
7571 atomic_inc(&clp->cl_rpc_users);
7572 WARN_ON(!unhash_delegation_locked(dp));
7573 list_add(&dp->dl_recall_lru, victims);
7574 }
7575 ++count;
7576 /*
7577 * Despite the fact that these functions deal with
7578 * 64-bit integers for "count", we must ensure that
7579 * it doesn't blow up the clp->cl_rpc_users. Throw a
7580 * warning if we start to approach INT_MAX here.
7581 */
7582 WARN_ON_ONCE(count == (INT_MAX / 2));
7583 if (count == max)
7584 break;
7585 }
7586 spin_unlock(&state_lock);
7587 return count;
7588 }
7589
7590 static u64
7591 nfsd_print_client_delegations(struct nfs4_client *clp)
7592 {
7593 u64 count = nfsd_find_all_delegations(clp, 0, NULL);
7594
7595 nfsd_print_count(clp, count, "delegations");
7596 return count;
7597 }
7598
7599 u64
7600 nfsd_inject_print_delegations(void)
7601 {
7602 struct nfs4_client *clp;
7603 u64 count = 0;
7604 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7605 nfsd_net_id);
7606
7607 if (!nfsd_netns_ready(nn))
7608 return 0;
7609
7610 spin_lock(&nn->client_lock);
7611 list_for_each_entry(clp, &nn->client_lru, cl_lru)
7612 count += nfsd_print_client_delegations(clp);
7613 spin_unlock(&nn->client_lock);
7614
7615 return count;
7616 }
7617
7618 static void
7619 nfsd_forget_delegations(struct list_head *reaplist)
7620 {
7621 struct nfs4_client *clp;
7622 struct nfs4_delegation *dp, *next;
7623
7624 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7625 list_del_init(&dp->dl_recall_lru);
7626 clp = dp->dl_stid.sc_client;
7627 revoke_delegation(dp);
7628 put_client(clp);
7629 }
7630 }
7631
7632 u64
7633 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
7634 size_t addr_size)
7635 {
7636 u64 count = 0;
7637 struct nfs4_client *clp;
7638 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7639 nfsd_net_id);
7640 LIST_HEAD(reaplist);
7641
7642 if (!nfsd_netns_ready(nn))
7643 return count;
7644
7645 spin_lock(&nn->client_lock);
7646 clp = nfsd_find_client(addr, addr_size);
7647 if (clp)
7648 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7649 spin_unlock(&nn->client_lock);
7650
7651 nfsd_forget_delegations(&reaplist);
7652 return count;
7653 }
7654
7655 u64
7656 nfsd_inject_forget_delegations(u64 max)
7657 {
7658 u64 count = 0;
7659 struct nfs4_client *clp;
7660 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7661 nfsd_net_id);
7662 LIST_HEAD(reaplist);
7663
7664 if (!nfsd_netns_ready(nn))
7665 return count;
7666
7667 spin_lock(&nn->client_lock);
7668 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
7669 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7670 if (max != 0 && count >= max)
7671 break;
7672 }
7673 spin_unlock(&nn->client_lock);
7674 nfsd_forget_delegations(&reaplist);
7675 return count;
7676 }
7677
7678 static void
7679 nfsd_recall_delegations(struct list_head *reaplist)
7680 {
7681 struct nfs4_client *clp;
7682 struct nfs4_delegation *dp, *next;
7683
7684 list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
7685 list_del_init(&dp->dl_recall_lru);
7686 clp = dp->dl_stid.sc_client;
7687 /*
7688 * We skipped all entries that had a zero dl_time before,
7689 * so we can now reset the dl_time back to 0. If a delegation
7690 * break comes in now, then it won't make any difference since
7691 * we're recalling it either way.
7692 */
7693 spin_lock(&state_lock);
7694 dp->dl_time = 0;
7695 spin_unlock(&state_lock);
7696 nfsd_break_one_deleg(dp);
7697 put_client(clp);
7698 }
7699 }
7700
7701 u64
7702 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7703 size_t addr_size)
7704 {
7705 u64 count = 0;
7706 struct nfs4_client *clp;
7707 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7708 nfsd_net_id);
7709 LIST_HEAD(reaplist);
7710
7711 if (!nfsd_netns_ready(nn))
7712 return count;
7713
7714 spin_lock(&nn->client_lock);
7715 clp = nfsd_find_client(addr, addr_size);
7716 if (clp)
7717 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7718 spin_unlock(&nn->client_lock);
7719
7720 nfsd_recall_delegations(&reaplist);
7721 return count;
7722 }
7723
7724 u64
7725 nfsd_inject_recall_delegations(u64 max)
7726 {
7727 u64 count = 0;
7728 struct nfs4_client *clp, *next;
7729 struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7730 nfsd_net_id);
7731 LIST_HEAD(reaplist);
7732
7733 if (!nfsd_netns_ready(nn))
7734 return count;
7735
7736 spin_lock(&nn->client_lock);
7737 list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7738 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7739 if (max != 0 && ++count >= max)
7740 break;
7741 }
7742 spin_unlock(&nn->client_lock);
7743 nfsd_recall_delegations(&reaplist);
7744 return count;
7745 }
7746 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7747
7748 /*
7749 * Since the lifetime of a delegation isn't limited to that of an open, a
7750 * client may quite reasonably hang on to a delegation as long as it has
7751 * the inode cached. This becomes an obvious problem the first time a
7752 * client's inode cache approaches the size of the server's total memory.
7753 *
7754 * For now we avoid this problem by imposing a hard limit on the number
7755 * of delegations, which varies according to the server's memory size.
7756 */
7757 static void
7758 set_max_delegations(void)
7759 {
7760 /*
7761 * Allow at most 4 delegations per megabyte of RAM. Quick
7762 * estimates suggest that in the worst case (where every delegation
7763 * is for a different inode), a delegation could take about 1.5K,
7764 * giving a worst case usage of about 6% of memory.
7765 */
7766 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7767 }
7768
7769 static int nfs4_state_create_net(struct net *net)
7770 {
7771 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7772 int i;
7773
7774 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7775 sizeof(struct list_head),
7776 GFP_KERNEL);
7777 if (!nn->conf_id_hashtbl)
7778 goto err;
7779 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
7780 sizeof(struct list_head),
7781 GFP_KERNEL);
7782 if (!nn->unconf_id_hashtbl)
7783 goto err_unconf_id;
7784 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
7785 sizeof(struct list_head),
7786 GFP_KERNEL);
7787 if (!nn->sessionid_hashtbl)
7788 goto err_sessionid;
7789
7790 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7791 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7792 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7793 }
7794 for (i = 0; i < SESSION_HASH_SIZE; i++)
7795 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7796 nn->conf_name_tree = RB_ROOT;
7797 nn->unconf_name_tree = RB_ROOT;
7798 nn->boot_time = ktime_get_real_seconds();
7799 nn->grace_ended = false;
7800 nn->nfsd4_manager.block_opens = true;
7801 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7802 INIT_LIST_HEAD(&nn->client_lru);
7803 INIT_LIST_HEAD(&nn->close_lru);
7804 INIT_LIST_HEAD(&nn->del_recall_lru);
7805 spin_lock_init(&nn->client_lock);
7806 spin_lock_init(&nn->s2s_cp_lock);
7807 idr_init(&nn->s2s_cp_stateids);
7808
7809 spin_lock_init(&nn->blocked_locks_lock);
7810 INIT_LIST_HEAD(&nn->blocked_locks_lru);
7811
7812 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7813 get_net(net);
7814
7815 return 0;
7816
7817 err_sessionid:
7818 kfree(nn->unconf_id_hashtbl);
7819 err_unconf_id:
7820 kfree(nn->conf_id_hashtbl);
7821 err:
7822 return -ENOMEM;
7823 }
7824
7825 static void
7826 nfs4_state_destroy_net(struct net *net)
7827 {
7828 int i;
7829 struct nfs4_client *clp = NULL;
7830 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7831
7832 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7833 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7834 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7835 destroy_client(clp);
7836 }
7837 }
7838
7839 WARN_ON(!list_empty(&nn->blocked_locks_lru));
7840
7841 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7842 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7843 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7844 destroy_client(clp);
7845 }
7846 }
7847
7848 kfree(nn->sessionid_hashtbl);
7849 kfree(nn->unconf_id_hashtbl);
7850 kfree(nn->conf_id_hashtbl);
7851 put_net(net);
7852 }
7853
7854 int
7855 nfs4_state_start_net(struct net *net)
7856 {
7857 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7858 int ret;
7859
7860 ret = nfs4_state_create_net(net);
7861 if (ret)
7862 return ret;
7863 locks_start_grace(net, &nn->nfsd4_manager);
7864 nfsd4_client_tracking_init(net);
7865 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
7866 goto skip_grace;
7867 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
7868 nn->nfsd4_grace, net->ns.inum);
7869 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7870 return 0;
7871
7872 skip_grace:
7873 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
7874 net->ns.inum);
7875 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
7876 nfsd4_end_grace(nn);
7877 return 0;
7878 }
7879
7880 /* initialization to perform when the nfsd service is started: */
7881
7882 int
7883 nfs4_state_start(void)
7884 {
7885 int ret;
7886
7887 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7888 if (laundry_wq == NULL) {
7889 ret = -ENOMEM;
7890 goto out;
7891 }
7892 ret = nfsd4_create_callback_queue();
7893 if (ret)
7894 goto out_free_laundry;
7895
7896 set_max_delegations();
7897 return 0;
7898
7899 out_free_laundry:
7900 destroy_workqueue(laundry_wq);
7901 out:
7902 return ret;
7903 }
7904
7905 void
7906 nfs4_state_shutdown_net(struct net *net)
7907 {
7908 struct nfs4_delegation *dp = NULL;
7909 struct list_head *pos, *next, reaplist;
7910 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7911
7912 cancel_delayed_work_sync(&nn->laundromat_work);
7913 locks_end_grace(&nn->nfsd4_manager);
7914
7915 INIT_LIST_HEAD(&reaplist);
7916 spin_lock(&state_lock);
7917 list_for_each_safe(pos, next, &nn->del_recall_lru) {
7918 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7919 WARN_ON(!unhash_delegation_locked(dp));
7920 list_add(&dp->dl_recall_lru, &reaplist);
7921 }
7922 spin_unlock(&state_lock);
7923 list_for_each_safe(pos, next, &reaplist) {
7924 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7925 list_del_init(&dp->dl_recall_lru);
7926 destroy_unhashed_deleg(dp);
7927 }
7928
7929 nfsd4_client_tracking_exit(net);
7930 nfs4_state_destroy_net(net);
7931 }
7932
7933 void
7934 nfs4_state_shutdown(void)
7935 {
7936 destroy_workqueue(laundry_wq);
7937 nfsd4_destroy_callback_queue();
7938 }
7939
7940 static void
7941 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7942 {
7943 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
7944 CURRENT_STATEID(stateid))
7945 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7946 }
7947
7948 static void
7949 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7950 {
7951 if (cstate->minorversion) {
7952 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7953 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7954 }
7955 }
7956
7957 void
7958 clear_current_stateid(struct nfsd4_compound_state *cstate)
7959 {
7960 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
7961 }
7962
7963 /*
7964 * functions to set current state id
7965 */
7966 void
7967 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7968 union nfsd4_op_u *u)
7969 {
7970 put_stateid(cstate, &u->open_downgrade.od_stateid);
7971 }
7972
7973 void
7974 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7975 union nfsd4_op_u *u)
7976 {
7977 put_stateid(cstate, &u->open.op_stateid);
7978 }
7979
7980 void
7981 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7982 union nfsd4_op_u *u)
7983 {
7984 put_stateid(cstate, &u->close.cl_stateid);
7985 }
7986
7987 void
7988 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7989 union nfsd4_op_u *u)
7990 {
7991 put_stateid(cstate, &u->lock.lk_resp_stateid);
7992 }
7993
7994 /*
7995 * functions to consume current state id
7996 */
7997
7998 void
7999 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8000 union nfsd4_op_u *u)
8001 {
8002 get_stateid(cstate, &u->open_downgrade.od_stateid);
8003 }
8004
8005 void
8006 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8007 union nfsd4_op_u *u)
8008 {
8009 get_stateid(cstate, &u->delegreturn.dr_stateid);
8010 }
8011
8012 void
8013 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8014 union nfsd4_op_u *u)
8015 {
8016 get_stateid(cstate, &u->free_stateid.fr_stateid);
8017 }
8018
8019 void
8020 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8021 union nfsd4_op_u *u)
8022 {
8023 get_stateid(cstate, &u->setattr.sa_stateid);
8024 }
8025
8026 void
8027 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8028 union nfsd4_op_u *u)
8029 {
8030 get_stateid(cstate, &u->close.cl_stateid);
8031 }
8032
8033 void
8034 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8035 union nfsd4_op_u *u)
8036 {
8037 get_stateid(cstate, &u->locku.lu_stateid);
8038 }
8039
8040 void
8041 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8042 union nfsd4_op_u *u)
8043 {
8044 get_stateid(cstate, &u->read.rd_stateid);
8045 }
8046
8047 void
8048 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8049 union nfsd4_op_u *u)
8050 {
8051 get_stateid(cstate, &u->write.wr_stateid);
8052 }