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 <linux/fsnotify.h>
47 #include <linux/rhashtable.h>
48 #include <linux/nfs_ssc.h>
49
50 #include "xdr4.h"
51 #include "xdr4cb.h"
52 #include "vfs.h"
53 #include "current_stateid.h"
54
55 #include "netns.h"
56 #include "pnfs.h"
57 #include "filecache.h"
58 #include "trace.h"
59
60 #define NFSDDBG_FACILITY NFSDDBG_PROC
61
62 #define all_ones {{ ~0, ~0}, ~0}
63 static const stateid_t one_stateid = {
64 .si_generation = ~0,
65 .si_opaque = all_ones,
66 };
67 static const stateid_t zero_stateid = {
68 /* all fields zero */
69 };
70 static const stateid_t currentstateid = {
71 .si_generation = 1,
72 };
73 static const stateid_t close_stateid = {
74 .si_generation = 0xffffffffU,
75 };
76
77 static u64 current_sessionid = 1;
78
79 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
80 #define ONE_STATEID(stateid) (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
81 #define CURRENT_STATEID(stateid) (!memcmp((stateid), ¤tstateid, sizeof(stateid_t)))
82 #define CLOSE_STATEID(stateid) (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
83
84 /* forward declarations */
85 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
86 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
87 void nfsd4_end_grace(struct nfsd_net *nn);
88 static void _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps);
89 static void nfsd4_file_hash_remove(struct nfs4_file *fi);
90 static void deleg_reaper(struct nfsd_net *nn);
91
92 /* Locking: */
93
94 /*
95 * Currently used for the del_recall_lru and file hash table. In an
96 * effort to decrease the scope of the client_mutex, this spinlock may
97 * eventually cover more:
98 */
99 static DEFINE_SPINLOCK(state_lock);
100
101 enum nfsd4_st_mutex_lock_subclass {
102 OPEN_STATEID_MUTEX = 0,
103 LOCK_STATEID_MUTEX = 1,
104 };
105
106 /*
107 * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
108 * the refcount on the open stateid to drop.
109 */
110 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
111
112 /*
113 * A waitqueue where a writer to clients/#/ctl destroying a client can
114 * wait for cl_rpc_users to drop to 0 and then for the client to be
115 * unhashed.
116 */
117 static DECLARE_WAIT_QUEUE_HEAD(expiry_wq);
118
119 static struct kmem_cache *client_slab;
120 static struct kmem_cache *openowner_slab;
121 static struct kmem_cache *lockowner_slab;
122 static struct kmem_cache *file_slab;
123 static struct kmem_cache *stateid_slab;
124 static struct kmem_cache *deleg_slab;
125 static struct kmem_cache *odstate_slab;
126
127 static void free_session(struct nfsd4_session *);
128
129 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
130 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
131 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops;
132
133 static struct workqueue_struct *laundry_wq;
134
nfsd4_create_laundry_wq(void)135 int nfsd4_create_laundry_wq(void)
136 {
137 int rc = 0;
138
139 laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
140 if (laundry_wq == NULL)
141 rc = -ENOMEM;
142 return rc;
143 }
144
nfsd4_destroy_laundry_wq(void)145 void nfsd4_destroy_laundry_wq(void)
146 {
147 destroy_workqueue(laundry_wq);
148 }
149
is_session_dead(struct nfsd4_session * ses)150 static bool is_session_dead(struct nfsd4_session *ses)
151 {
152 return ses->se_flags & NFS4_SESSION_DEAD;
153 }
154
mark_session_dead_locked(struct nfsd4_session * ses,int ref_held_by_me)155 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
156 {
157 if (atomic_read(&ses->se_ref) > ref_held_by_me)
158 return nfserr_jukebox;
159 ses->se_flags |= NFS4_SESSION_DEAD;
160 return nfs_ok;
161 }
162
is_client_expired(struct nfs4_client * clp)163 static bool is_client_expired(struct nfs4_client *clp)
164 {
165 return clp->cl_time == 0;
166 }
167
nfsd4_dec_courtesy_client_count(struct nfsd_net * nn,struct nfs4_client * clp)168 static void nfsd4_dec_courtesy_client_count(struct nfsd_net *nn,
169 struct nfs4_client *clp)
170 {
171 if (clp->cl_state != NFSD4_ACTIVE)
172 atomic_add_unless(&nn->nfsd_courtesy_clients, -1, 0);
173 }
174
get_client_locked(struct nfs4_client * clp)175 static __be32 get_client_locked(struct nfs4_client *clp)
176 {
177 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
178
179 lockdep_assert_held(&nn->client_lock);
180
181 if (is_client_expired(clp))
182 return nfserr_expired;
183 atomic_inc(&clp->cl_rpc_users);
184 nfsd4_dec_courtesy_client_count(nn, clp);
185 clp->cl_state = NFSD4_ACTIVE;
186 return nfs_ok;
187 }
188
189 /* must be called under the client_lock */
190 static inline void
renew_client_locked(struct nfs4_client * clp)191 renew_client_locked(struct nfs4_client *clp)
192 {
193 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
194
195 if (is_client_expired(clp)) {
196 WARN_ON(1);
197 printk("%s: client (clientid %08x/%08x) already expired\n",
198 __func__,
199 clp->cl_clientid.cl_boot,
200 clp->cl_clientid.cl_id);
201 return;
202 }
203
204 list_move_tail(&clp->cl_lru, &nn->client_lru);
205 clp->cl_time = ktime_get_boottime_seconds();
206 nfsd4_dec_courtesy_client_count(nn, clp);
207 clp->cl_state = NFSD4_ACTIVE;
208 }
209
put_client_renew_locked(struct nfs4_client * clp)210 static void put_client_renew_locked(struct nfs4_client *clp)
211 {
212 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
213
214 lockdep_assert_held(&nn->client_lock);
215
216 if (!atomic_dec_and_test(&clp->cl_rpc_users))
217 return;
218 if (!is_client_expired(clp))
219 renew_client_locked(clp);
220 else
221 wake_up_all(&expiry_wq);
222 }
223
put_client_renew(struct nfs4_client * clp)224 static void put_client_renew(struct nfs4_client *clp)
225 {
226 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
227
228 if (!atomic_dec_and_lock(&clp->cl_rpc_users, &nn->client_lock))
229 return;
230 if (!is_client_expired(clp))
231 renew_client_locked(clp);
232 else
233 wake_up_all(&expiry_wq);
234 spin_unlock(&nn->client_lock);
235 }
236
nfsd4_get_session_locked(struct nfsd4_session * ses)237 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
238 {
239 __be32 status;
240
241 if (is_session_dead(ses))
242 return nfserr_badsession;
243 status = get_client_locked(ses->se_client);
244 if (status)
245 return status;
246 atomic_inc(&ses->se_ref);
247 return nfs_ok;
248 }
249
nfsd4_put_session_locked(struct nfsd4_session * ses)250 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
251 {
252 struct nfs4_client *clp = ses->se_client;
253 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
254
255 lockdep_assert_held(&nn->client_lock);
256
257 if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
258 free_session(ses);
259 put_client_renew_locked(clp);
260 }
261
nfsd4_put_session(struct nfsd4_session * ses)262 static void nfsd4_put_session(struct nfsd4_session *ses)
263 {
264 struct nfs4_client *clp = ses->se_client;
265 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
266
267 spin_lock(&nn->client_lock);
268 nfsd4_put_session_locked(ses);
269 spin_unlock(&nn->client_lock);
270 }
271
272 static struct nfsd4_blocked_lock *
find_blocked_lock(struct nfs4_lockowner * lo,struct knfsd_fh * fh,struct nfsd_net * nn)273 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
274 struct nfsd_net *nn)
275 {
276 struct nfsd4_blocked_lock *cur, *found = NULL;
277
278 spin_lock(&nn->blocked_locks_lock);
279 list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
280 if (fh_match(fh, &cur->nbl_fh)) {
281 list_del_init(&cur->nbl_list);
282 WARN_ON(list_empty(&cur->nbl_lru));
283 list_del_init(&cur->nbl_lru);
284 found = cur;
285 break;
286 }
287 }
288 spin_unlock(&nn->blocked_locks_lock);
289 if (found)
290 locks_delete_block(&found->nbl_lock);
291 return found;
292 }
293
294 static struct nfsd4_blocked_lock *
find_or_allocate_block(struct nfs4_lockowner * lo,struct knfsd_fh * fh,struct nfsd_net * nn)295 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
296 struct nfsd_net *nn)
297 {
298 struct nfsd4_blocked_lock *nbl;
299
300 nbl = find_blocked_lock(lo, fh, nn);
301 if (!nbl) {
302 nbl = kmalloc(sizeof(*nbl), GFP_KERNEL);
303 if (nbl) {
304 INIT_LIST_HEAD(&nbl->nbl_list);
305 INIT_LIST_HEAD(&nbl->nbl_lru);
306 fh_copy_shallow(&nbl->nbl_fh, fh);
307 locks_init_lock(&nbl->nbl_lock);
308 kref_init(&nbl->nbl_kref);
309 nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
310 &nfsd4_cb_notify_lock_ops,
311 NFSPROC4_CLNT_CB_NOTIFY_LOCK);
312 }
313 }
314 return nbl;
315 }
316
317 static void
free_nbl(struct kref * kref)318 free_nbl(struct kref *kref)
319 {
320 struct nfsd4_blocked_lock *nbl;
321
322 nbl = container_of(kref, struct nfsd4_blocked_lock, nbl_kref);
323 locks_release_private(&nbl->nbl_lock);
324 kfree(nbl);
325 }
326
327 static void
free_blocked_lock(struct nfsd4_blocked_lock * nbl)328 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
329 {
330 locks_delete_block(&nbl->nbl_lock);
331 kref_put(&nbl->nbl_kref, free_nbl);
332 }
333
334 static void
remove_blocked_locks(struct nfs4_lockowner * lo)335 remove_blocked_locks(struct nfs4_lockowner *lo)
336 {
337 struct nfs4_client *clp = lo->lo_owner.so_client;
338 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
339 struct nfsd4_blocked_lock *nbl;
340 LIST_HEAD(reaplist);
341
342 /* Dequeue all blocked locks */
343 spin_lock(&nn->blocked_locks_lock);
344 while (!list_empty(&lo->lo_blocked)) {
345 nbl = list_first_entry(&lo->lo_blocked,
346 struct nfsd4_blocked_lock,
347 nbl_list);
348 list_del_init(&nbl->nbl_list);
349 WARN_ON(list_empty(&nbl->nbl_lru));
350 list_move(&nbl->nbl_lru, &reaplist);
351 }
352 spin_unlock(&nn->blocked_locks_lock);
353
354 /* Now free them */
355 while (!list_empty(&reaplist)) {
356 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
357 nbl_lru);
358 list_del_init(&nbl->nbl_lru);
359 free_blocked_lock(nbl);
360 }
361 }
362
363 static void
nfsd4_cb_notify_lock_prepare(struct nfsd4_callback * cb)364 nfsd4_cb_notify_lock_prepare(struct nfsd4_callback *cb)
365 {
366 struct nfsd4_blocked_lock *nbl = container_of(cb,
367 struct nfsd4_blocked_lock, nbl_cb);
368 locks_delete_block(&nbl->nbl_lock);
369 }
370
371 static int
nfsd4_cb_notify_lock_done(struct nfsd4_callback * cb,struct rpc_task * task)372 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
373 {
374 trace_nfsd_cb_notify_lock_done(&zero_stateid, task);
375
376 /*
377 * Since this is just an optimization, we don't try very hard if it
378 * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
379 * just quit trying on anything else.
380 */
381 switch (task->tk_status) {
382 case -NFS4ERR_DELAY:
383 rpc_delay(task, 1 * HZ);
384 return 0;
385 default:
386 return 1;
387 }
388 }
389
390 static void
nfsd4_cb_notify_lock_release(struct nfsd4_callback * cb)391 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
392 {
393 struct nfsd4_blocked_lock *nbl = container_of(cb,
394 struct nfsd4_blocked_lock, nbl_cb);
395
396 free_blocked_lock(nbl);
397 }
398
399 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
400 .prepare = nfsd4_cb_notify_lock_prepare,
401 .done = nfsd4_cb_notify_lock_done,
402 .release = nfsd4_cb_notify_lock_release,
403 };
404
405 /*
406 * We store the NONE, READ, WRITE, and BOTH bits separately in the
407 * st_{access,deny}_bmap field of the stateid, in order to track not
408 * only what share bits are currently in force, but also what
409 * combinations of share bits previous opens have used. This allows us
410 * to enforce the recommendation in
411 * https://datatracker.ietf.org/doc/html/rfc7530#section-16.19.4 that
412 * the server return an error if the client attempt to downgrade to a
413 * combination of share bits not explicable by closing some of its
414 * previous opens.
415 *
416 * This enforcement is arguably incomplete, since we don't keep
417 * track of access/deny bit combinations; so, e.g., we allow:
418 *
419 * OPEN allow read, deny write
420 * OPEN allow both, deny none
421 * DOWNGRADE allow read, deny none
422 *
423 * which we should reject.
424 *
425 * But you could also argue that our current code is already overkill,
426 * since it only exists to return NFS4ERR_INVAL on incorrect client
427 * behavior.
428 */
429 static unsigned int
bmap_to_share_mode(unsigned long bmap)430 bmap_to_share_mode(unsigned long bmap)
431 {
432 int i;
433 unsigned int access = 0;
434
435 for (i = 1; i < 4; i++) {
436 if (test_bit(i, &bmap))
437 access |= i;
438 }
439 return access;
440 }
441
442 /* set share access for a given stateid */
443 static inline void
set_access(u32 access,struct nfs4_ol_stateid * stp)444 set_access(u32 access, struct nfs4_ol_stateid *stp)
445 {
446 unsigned char mask = 1 << access;
447
448 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
449 stp->st_access_bmap |= mask;
450 }
451
452 /* clear share access for a given stateid */
453 static inline void
clear_access(u32 access,struct nfs4_ol_stateid * stp)454 clear_access(u32 access, struct nfs4_ol_stateid *stp)
455 {
456 unsigned char mask = 1 << access;
457
458 WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
459 stp->st_access_bmap &= ~mask;
460 }
461
462 /* test whether a given stateid has access */
463 static inline bool
test_access(u32 access,struct nfs4_ol_stateid * stp)464 test_access(u32 access, struct nfs4_ol_stateid *stp)
465 {
466 unsigned char mask = 1 << access;
467
468 return (bool)(stp->st_access_bmap & mask);
469 }
470
471 /* set share deny for a given stateid */
472 static inline void
set_deny(u32 deny,struct nfs4_ol_stateid * stp)473 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
474 {
475 unsigned char mask = 1 << deny;
476
477 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
478 stp->st_deny_bmap |= mask;
479 }
480
481 /* clear share deny for a given stateid */
482 static inline void
clear_deny(u32 deny,struct nfs4_ol_stateid * stp)483 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
484 {
485 unsigned char mask = 1 << deny;
486
487 WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
488 stp->st_deny_bmap &= ~mask;
489 }
490
491 /* test whether a given stateid is denying specific access */
492 static inline bool
test_deny(u32 deny,struct nfs4_ol_stateid * stp)493 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
494 {
495 unsigned char mask = 1 << deny;
496
497 return (bool)(stp->st_deny_bmap & mask);
498 }
499
nfs4_access_to_omode(u32 access)500 static int nfs4_access_to_omode(u32 access)
501 {
502 switch (access & NFS4_SHARE_ACCESS_BOTH) {
503 case NFS4_SHARE_ACCESS_READ:
504 return O_RDONLY;
505 case NFS4_SHARE_ACCESS_WRITE:
506 return O_WRONLY;
507 case NFS4_SHARE_ACCESS_BOTH:
508 return O_RDWR;
509 }
510 WARN_ON_ONCE(1);
511 return O_RDONLY;
512 }
513
514 static inline int
access_permit_read(struct nfs4_ol_stateid * stp)515 access_permit_read(struct nfs4_ol_stateid *stp)
516 {
517 return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
518 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
519 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
520 }
521
522 static inline int
access_permit_write(struct nfs4_ol_stateid * stp)523 access_permit_write(struct nfs4_ol_stateid *stp)
524 {
525 return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
526 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
527 }
528
529 static inline struct nfs4_stateowner *
nfs4_get_stateowner(struct nfs4_stateowner * sop)530 nfs4_get_stateowner(struct nfs4_stateowner *sop)
531 {
532 atomic_inc(&sop->so_count);
533 return sop;
534 }
535
536 static int
same_owner_str(struct nfs4_stateowner * sop,struct xdr_netobj * owner)537 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
538 {
539 return (sop->so_owner.len == owner->len) &&
540 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
541 }
542
543 static struct nfs4_openowner *
find_openstateowner_str(unsigned int hashval,struct nfsd4_open * open,struct nfs4_client * clp)544 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
545 struct nfs4_client *clp)
546 {
547 struct nfs4_stateowner *so;
548
549 lockdep_assert_held(&clp->cl_lock);
550
551 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
552 so_strhash) {
553 if (!so->so_is_open_owner)
554 continue;
555 if (same_owner_str(so, &open->op_owner))
556 return openowner(nfs4_get_stateowner(so));
557 }
558 return NULL;
559 }
560
561 static inline u32
opaque_hashval(const void * ptr,int nbytes)562 opaque_hashval(const void *ptr, int nbytes)
563 {
564 unsigned char *cptr = (unsigned char *) ptr;
565
566 u32 x = 0;
567 while (nbytes--) {
568 x *= 37;
569 x += *cptr++;
570 }
571 return x;
572 }
573
nfsd4_free_file_rcu(struct rcu_head * rcu)574 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
575 {
576 struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
577
578 kmem_cache_free(file_slab, fp);
579 }
580
581 void
put_nfs4_file(struct nfs4_file * fi)582 put_nfs4_file(struct nfs4_file *fi)
583 {
584 if (refcount_dec_and_test(&fi->fi_ref)) {
585 nfsd4_file_hash_remove(fi);
586 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
587 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
588 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
589 }
590 }
591
592 static struct nfsd_file *
find_writeable_file_locked(struct nfs4_file * f)593 find_writeable_file_locked(struct nfs4_file *f)
594 {
595 struct nfsd_file *ret;
596
597 lockdep_assert_held(&f->fi_lock);
598
599 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
600 if (!ret)
601 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
602 return ret;
603 }
604
605 static struct nfsd_file *
find_writeable_file(struct nfs4_file * f)606 find_writeable_file(struct nfs4_file *f)
607 {
608 struct nfsd_file *ret;
609
610 spin_lock(&f->fi_lock);
611 ret = find_writeable_file_locked(f);
612 spin_unlock(&f->fi_lock);
613
614 return ret;
615 }
616
617 static struct nfsd_file *
find_readable_file_locked(struct nfs4_file * f)618 find_readable_file_locked(struct nfs4_file *f)
619 {
620 struct nfsd_file *ret;
621
622 lockdep_assert_held(&f->fi_lock);
623
624 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
625 if (!ret)
626 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
627 return ret;
628 }
629
630 static struct nfsd_file *
find_readable_file(struct nfs4_file * f)631 find_readable_file(struct nfs4_file *f)
632 {
633 struct nfsd_file *ret;
634
635 spin_lock(&f->fi_lock);
636 ret = find_readable_file_locked(f);
637 spin_unlock(&f->fi_lock);
638
639 return ret;
640 }
641
642 static struct nfsd_file *
find_rw_file(struct nfs4_file * f)643 find_rw_file(struct nfs4_file *f)
644 {
645 struct nfsd_file *ret;
646
647 spin_lock(&f->fi_lock);
648 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
649 spin_unlock(&f->fi_lock);
650
651 return ret;
652 }
653
654 struct nfsd_file *
find_any_file(struct nfs4_file * f)655 find_any_file(struct nfs4_file *f)
656 {
657 struct nfsd_file *ret;
658
659 if (!f)
660 return NULL;
661 spin_lock(&f->fi_lock);
662 ret = nfsd_file_get(f->fi_fds[O_RDWR]);
663 if (!ret) {
664 ret = nfsd_file_get(f->fi_fds[O_WRONLY]);
665 if (!ret)
666 ret = nfsd_file_get(f->fi_fds[O_RDONLY]);
667 }
668 spin_unlock(&f->fi_lock);
669 return ret;
670 }
671
find_any_file_locked(struct nfs4_file * f)672 static struct nfsd_file *find_any_file_locked(struct nfs4_file *f)
673 {
674 lockdep_assert_held(&f->fi_lock);
675
676 if (f->fi_fds[O_RDWR])
677 return f->fi_fds[O_RDWR];
678 if (f->fi_fds[O_WRONLY])
679 return f->fi_fds[O_WRONLY];
680 if (f->fi_fds[O_RDONLY])
681 return f->fi_fds[O_RDONLY];
682 return NULL;
683 }
684
685 static atomic_long_t num_delegations;
686 unsigned long max_delegations;
687
688 /*
689 * Open owner state (share locks)
690 */
691
692 /* hash tables for lock and open owners */
693 #define OWNER_HASH_BITS 8
694 #define OWNER_HASH_SIZE (1 << OWNER_HASH_BITS)
695 #define OWNER_HASH_MASK (OWNER_HASH_SIZE - 1)
696
ownerstr_hashval(struct xdr_netobj * ownername)697 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
698 {
699 unsigned int ret;
700
701 ret = opaque_hashval(ownername->data, ownername->len);
702 return ret & OWNER_HASH_MASK;
703 }
704
705 static struct rhltable nfs4_file_rhltable ____cacheline_aligned_in_smp;
706
707 static const struct rhashtable_params nfs4_file_rhash_params = {
708 .key_len = sizeof_field(struct nfs4_file, fi_inode),
709 .key_offset = offsetof(struct nfs4_file, fi_inode),
710 .head_offset = offsetof(struct nfs4_file, fi_rlist),
711
712 /*
713 * Start with a single page hash table to reduce resizing churn
714 * on light workloads.
715 */
716 .min_size = 256,
717 .automatic_shrinking = true,
718 };
719
720 /*
721 * Check if courtesy clients have conflicting access and resolve it if possible
722 *
723 * access: is op_share_access if share_access is true.
724 * Check if access mode, op_share_access, would conflict with
725 * the current deny mode of the file 'fp'.
726 * access: is op_share_deny if share_access is false.
727 * Check if the deny mode, op_share_deny, would conflict with
728 * current access of the file 'fp'.
729 * stp: skip checking this entry.
730 * new_stp: normal open, not open upgrade.
731 *
732 * Function returns:
733 * false - access/deny mode conflict with normal client.
734 * true - no conflict or conflict with courtesy client(s) is resolved.
735 */
736 static bool
nfs4_resolve_deny_conflicts_locked(struct nfs4_file * fp,bool new_stp,struct nfs4_ol_stateid * stp,u32 access,bool share_access)737 nfs4_resolve_deny_conflicts_locked(struct nfs4_file *fp, bool new_stp,
738 struct nfs4_ol_stateid *stp, u32 access, bool share_access)
739 {
740 struct nfs4_ol_stateid *st;
741 bool resolvable = true;
742 unsigned char bmap;
743 struct nfsd_net *nn;
744 struct nfs4_client *clp;
745
746 lockdep_assert_held(&fp->fi_lock);
747 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
748 /* ignore lock stateid */
749 if (st->st_openstp)
750 continue;
751 if (st == stp && new_stp)
752 continue;
753 /* check file access against deny mode or vice versa */
754 bmap = share_access ? st->st_deny_bmap : st->st_access_bmap;
755 if (!(access & bmap_to_share_mode(bmap)))
756 continue;
757 clp = st->st_stid.sc_client;
758 if (try_to_expire_client(clp))
759 continue;
760 resolvable = false;
761 break;
762 }
763 if (resolvable) {
764 clp = stp->st_stid.sc_client;
765 nn = net_generic(clp->net, nfsd_net_id);
766 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
767 }
768 return resolvable;
769 }
770
771 static void
__nfs4_file_get_access(struct nfs4_file * fp,u32 access)772 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
773 {
774 lockdep_assert_held(&fp->fi_lock);
775
776 if (access & NFS4_SHARE_ACCESS_WRITE)
777 atomic_inc(&fp->fi_access[O_WRONLY]);
778 if (access & NFS4_SHARE_ACCESS_READ)
779 atomic_inc(&fp->fi_access[O_RDONLY]);
780 }
781
782 static __be32
nfs4_file_get_access(struct nfs4_file * fp,u32 access)783 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
784 {
785 lockdep_assert_held(&fp->fi_lock);
786
787 /* Does this access mode make sense? */
788 if (access & ~NFS4_SHARE_ACCESS_BOTH)
789 return nfserr_inval;
790
791 /* Does it conflict with a deny mode already set? */
792 if ((access & fp->fi_share_deny) != 0)
793 return nfserr_share_denied;
794
795 __nfs4_file_get_access(fp, access);
796 return nfs_ok;
797 }
798
nfs4_file_check_deny(struct nfs4_file * fp,u32 deny)799 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
800 {
801 /* Common case is that there is no deny mode. */
802 if (deny) {
803 /* Does this deny mode make sense? */
804 if (deny & ~NFS4_SHARE_DENY_BOTH)
805 return nfserr_inval;
806
807 if ((deny & NFS4_SHARE_DENY_READ) &&
808 atomic_read(&fp->fi_access[O_RDONLY]))
809 return nfserr_share_denied;
810
811 if ((deny & NFS4_SHARE_DENY_WRITE) &&
812 atomic_read(&fp->fi_access[O_WRONLY]))
813 return nfserr_share_denied;
814 }
815 return nfs_ok;
816 }
817
__nfs4_file_put_access(struct nfs4_file * fp,int oflag)818 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
819 {
820 might_lock(&fp->fi_lock);
821
822 if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
823 struct nfsd_file *f1 = NULL;
824 struct nfsd_file *f2 = NULL;
825
826 swap(f1, fp->fi_fds[oflag]);
827 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
828 swap(f2, fp->fi_fds[O_RDWR]);
829 spin_unlock(&fp->fi_lock);
830 if (f1)
831 nfsd_file_put(f1);
832 if (f2)
833 nfsd_file_put(f2);
834 }
835 }
836
nfs4_file_put_access(struct nfs4_file * fp,u32 access)837 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
838 {
839 WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
840
841 if (access & NFS4_SHARE_ACCESS_WRITE)
842 __nfs4_file_put_access(fp, O_WRONLY);
843 if (access & NFS4_SHARE_ACCESS_READ)
844 __nfs4_file_put_access(fp, O_RDONLY);
845 }
846
847 /*
848 * Allocate a new open/delegation state counter. This is needed for
849 * pNFS for proper return on close semantics.
850 *
851 * Note that we only allocate it for pNFS-enabled exports, otherwise
852 * all pointers to struct nfs4_clnt_odstate are always NULL.
853 */
854 static struct nfs4_clnt_odstate *
alloc_clnt_odstate(struct nfs4_client * clp)855 alloc_clnt_odstate(struct nfs4_client *clp)
856 {
857 struct nfs4_clnt_odstate *co;
858
859 co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
860 if (co) {
861 co->co_client = clp;
862 refcount_set(&co->co_odcount, 1);
863 }
864 return co;
865 }
866
867 static void
hash_clnt_odstate_locked(struct nfs4_clnt_odstate * co)868 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
869 {
870 struct nfs4_file *fp = co->co_file;
871
872 lockdep_assert_held(&fp->fi_lock);
873 list_add(&co->co_perfile, &fp->fi_clnt_odstate);
874 }
875
876 static inline void
get_clnt_odstate(struct nfs4_clnt_odstate * co)877 get_clnt_odstate(struct nfs4_clnt_odstate *co)
878 {
879 if (co)
880 refcount_inc(&co->co_odcount);
881 }
882
883 static void
put_clnt_odstate(struct nfs4_clnt_odstate * co)884 put_clnt_odstate(struct nfs4_clnt_odstate *co)
885 {
886 struct nfs4_file *fp;
887
888 if (!co)
889 return;
890
891 fp = co->co_file;
892 if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
893 list_del(&co->co_perfile);
894 spin_unlock(&fp->fi_lock);
895
896 nfsd4_return_all_file_layouts(co->co_client, fp);
897 kmem_cache_free(odstate_slab, co);
898 }
899 }
900
901 static struct nfs4_clnt_odstate *
find_or_hash_clnt_odstate(struct nfs4_file * fp,struct nfs4_clnt_odstate * new)902 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
903 {
904 struct nfs4_clnt_odstate *co;
905 struct nfs4_client *cl;
906
907 if (!new)
908 return NULL;
909
910 cl = new->co_client;
911
912 spin_lock(&fp->fi_lock);
913 list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
914 if (co->co_client == cl) {
915 get_clnt_odstate(co);
916 goto out;
917 }
918 }
919 co = new;
920 co->co_file = fp;
921 hash_clnt_odstate_locked(new);
922 out:
923 spin_unlock(&fp->fi_lock);
924 return co;
925 }
926
nfs4_alloc_stid(struct nfs4_client * cl,struct kmem_cache * slab,void (* sc_free)(struct nfs4_stid *))927 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
928 void (*sc_free)(struct nfs4_stid *))
929 {
930 struct nfs4_stid *stid;
931 int new_id;
932
933 stid = kmem_cache_zalloc(slab, GFP_KERNEL);
934 if (!stid)
935 return NULL;
936
937 idr_preload(GFP_KERNEL);
938 spin_lock(&cl->cl_lock);
939 /* Reserving 0 for start of file in nfsdfs "states" file: */
940 new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 1, 0, GFP_NOWAIT);
941 spin_unlock(&cl->cl_lock);
942 idr_preload_end();
943 if (new_id < 0)
944 goto out_free;
945
946 stid->sc_free = sc_free;
947 stid->sc_client = cl;
948 stid->sc_stateid.si_opaque.so_id = new_id;
949 stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
950 /* Will be incremented before return to client: */
951 refcount_set(&stid->sc_count, 1);
952 spin_lock_init(&stid->sc_lock);
953 INIT_LIST_HEAD(&stid->sc_cp_list);
954
955 /*
956 * It shouldn't be a problem to reuse an opaque stateid value.
957 * I don't think it is for 4.1. But with 4.0 I worry that, for
958 * example, a stray write retransmission could be accepted by
959 * the server when it should have been rejected. Therefore,
960 * adopt a trick from the sctp code to attempt to maximize the
961 * amount of time until an id is reused, by ensuring they always
962 * "increase" (mod INT_MAX):
963 */
964 return stid;
965 out_free:
966 kmem_cache_free(slab, stid);
967 return NULL;
968 }
969
970 /*
971 * Create a unique stateid_t to represent each COPY.
972 */
nfs4_init_cp_state(struct nfsd_net * nn,copy_stateid_t * stid,unsigned char cs_type)973 static int nfs4_init_cp_state(struct nfsd_net *nn, copy_stateid_t *stid,
974 unsigned char cs_type)
975 {
976 int new_id;
977
978 stid->cs_stid.si_opaque.so_clid.cl_boot = (u32)nn->boot_time;
979 stid->cs_stid.si_opaque.so_clid.cl_id = nn->s2s_cp_cl_id;
980
981 idr_preload(GFP_KERNEL);
982 spin_lock(&nn->s2s_cp_lock);
983 new_id = idr_alloc_cyclic(&nn->s2s_cp_stateids, stid, 0, 0, GFP_NOWAIT);
984 stid->cs_stid.si_opaque.so_id = new_id;
985 stid->cs_stid.si_generation = 1;
986 spin_unlock(&nn->s2s_cp_lock);
987 idr_preload_end();
988 if (new_id < 0)
989 return 0;
990 stid->cs_type = cs_type;
991 return 1;
992 }
993
nfs4_init_copy_state(struct nfsd_net * nn,struct nfsd4_copy * copy)994 int nfs4_init_copy_state(struct nfsd_net *nn, struct nfsd4_copy *copy)
995 {
996 return nfs4_init_cp_state(nn, ©->cp_stateid, NFS4_COPY_STID);
997 }
998
nfs4_alloc_init_cpntf_state(struct nfsd_net * nn,struct nfs4_stid * p_stid)999 struct nfs4_cpntf_state *nfs4_alloc_init_cpntf_state(struct nfsd_net *nn,
1000 struct nfs4_stid *p_stid)
1001 {
1002 struct nfs4_cpntf_state *cps;
1003
1004 cps = kzalloc(sizeof(struct nfs4_cpntf_state), GFP_KERNEL);
1005 if (!cps)
1006 return NULL;
1007 cps->cpntf_time = ktime_get_boottime_seconds();
1008 refcount_set(&cps->cp_stateid.cs_count, 1);
1009 if (!nfs4_init_cp_state(nn, &cps->cp_stateid, NFS4_COPYNOTIFY_STID))
1010 goto out_free;
1011 spin_lock(&nn->s2s_cp_lock);
1012 list_add(&cps->cp_list, &p_stid->sc_cp_list);
1013 spin_unlock(&nn->s2s_cp_lock);
1014 return cps;
1015 out_free:
1016 kfree(cps);
1017 return NULL;
1018 }
1019
nfs4_free_copy_state(struct nfsd4_copy * copy)1020 void nfs4_free_copy_state(struct nfsd4_copy *copy)
1021 {
1022 struct nfsd_net *nn;
1023
1024 if (copy->cp_stateid.cs_type != NFS4_COPY_STID)
1025 return;
1026 nn = net_generic(copy->cp_clp->net, nfsd_net_id);
1027 spin_lock(&nn->s2s_cp_lock);
1028 idr_remove(&nn->s2s_cp_stateids,
1029 copy->cp_stateid.cs_stid.si_opaque.so_id);
1030 spin_unlock(&nn->s2s_cp_lock);
1031 }
1032
nfs4_free_cpntf_statelist(struct net * net,struct nfs4_stid * stid)1033 static void nfs4_free_cpntf_statelist(struct net *net, struct nfs4_stid *stid)
1034 {
1035 struct nfs4_cpntf_state *cps;
1036 struct nfsd_net *nn;
1037
1038 nn = net_generic(net, nfsd_net_id);
1039 spin_lock(&nn->s2s_cp_lock);
1040 while (!list_empty(&stid->sc_cp_list)) {
1041 cps = list_first_entry(&stid->sc_cp_list,
1042 struct nfs4_cpntf_state, cp_list);
1043 _free_cpntf_state_locked(nn, cps);
1044 }
1045 spin_unlock(&nn->s2s_cp_lock);
1046 }
1047
nfs4_alloc_open_stateid(struct nfs4_client * clp)1048 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
1049 {
1050 struct nfs4_stid *stid;
1051
1052 stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
1053 if (!stid)
1054 return NULL;
1055
1056 return openlockstateid(stid);
1057 }
1058
nfs4_free_deleg(struct nfs4_stid * stid)1059 static void nfs4_free_deleg(struct nfs4_stid *stid)
1060 {
1061 struct nfs4_delegation *dp = delegstateid(stid);
1062
1063 WARN_ON_ONCE(!list_empty(&stid->sc_cp_list));
1064 WARN_ON_ONCE(!list_empty(&dp->dl_perfile));
1065 WARN_ON_ONCE(!list_empty(&dp->dl_perclnt));
1066 WARN_ON_ONCE(!list_empty(&dp->dl_recall_lru));
1067 kmem_cache_free(deleg_slab, stid);
1068 atomic_long_dec(&num_delegations);
1069 }
1070
1071 /*
1072 * When we recall a delegation, we should be careful not to hand it
1073 * out again straight away.
1074 * To ensure this we keep a pair of bloom filters ('new' and 'old')
1075 * in which the filehandles of recalled delegations are "stored".
1076 * If a filehandle appear in either filter, a delegation is blocked.
1077 * When a delegation is recalled, the filehandle is stored in the "new"
1078 * filter.
1079 * Every 30 seconds we swap the filters and clear the "new" one,
1080 * unless both are empty of course.
1081 *
1082 * Each filter is 256 bits. We hash the filehandle to 32bit and use the
1083 * low 3 bytes as hash-table indices.
1084 *
1085 * 'blocked_delegations_lock', which is always taken in block_delegations(),
1086 * is used to manage concurrent access. Testing does not need the lock
1087 * except when swapping the two filters.
1088 */
1089 static DEFINE_SPINLOCK(blocked_delegations_lock);
1090 static struct bloom_pair {
1091 int entries, old_entries;
1092 time64_t swap_time;
1093 int new; /* index into 'set' */
1094 DECLARE_BITMAP(set[2], 256);
1095 } blocked_delegations;
1096
delegation_blocked(struct knfsd_fh * fh)1097 static int delegation_blocked(struct knfsd_fh *fh)
1098 {
1099 u32 hash;
1100 struct bloom_pair *bd = &blocked_delegations;
1101
1102 if (bd->entries == 0)
1103 return 0;
1104 if (ktime_get_seconds() - bd->swap_time > 30) {
1105 spin_lock(&blocked_delegations_lock);
1106 if (ktime_get_seconds() - bd->swap_time > 30) {
1107 bd->entries -= bd->old_entries;
1108 bd->old_entries = bd->entries;
1109 memset(bd->set[bd->new], 0,
1110 sizeof(bd->set[0]));
1111 bd->new = 1-bd->new;
1112 bd->swap_time = ktime_get_seconds();
1113 }
1114 spin_unlock(&blocked_delegations_lock);
1115 }
1116 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1117 if (test_bit(hash&255, bd->set[0]) &&
1118 test_bit((hash>>8)&255, bd->set[0]) &&
1119 test_bit((hash>>16)&255, bd->set[0]))
1120 return 1;
1121
1122 if (test_bit(hash&255, bd->set[1]) &&
1123 test_bit((hash>>8)&255, bd->set[1]) &&
1124 test_bit((hash>>16)&255, bd->set[1]))
1125 return 1;
1126
1127 return 0;
1128 }
1129
block_delegations(struct knfsd_fh * fh)1130 static void block_delegations(struct knfsd_fh *fh)
1131 {
1132 u32 hash;
1133 struct bloom_pair *bd = &blocked_delegations;
1134
1135 hash = jhash(&fh->fh_raw, fh->fh_size, 0);
1136
1137 spin_lock(&blocked_delegations_lock);
1138 __set_bit(hash&255, bd->set[bd->new]);
1139 __set_bit((hash>>8)&255, bd->set[bd->new]);
1140 __set_bit((hash>>16)&255, bd->set[bd->new]);
1141 if (bd->entries == 0)
1142 bd->swap_time = ktime_get_seconds();
1143 bd->entries += 1;
1144 spin_unlock(&blocked_delegations_lock);
1145 }
1146
1147 static struct nfs4_delegation *
alloc_init_deleg(struct nfs4_client * clp,struct nfs4_file * fp,struct nfs4_clnt_odstate * odstate,u32 dl_type)1148 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
1149 struct nfs4_clnt_odstate *odstate, u32 dl_type)
1150 {
1151 struct nfs4_delegation *dp;
1152 struct nfs4_stid *stid;
1153 long n;
1154
1155 dprintk("NFSD alloc_init_deleg\n");
1156 n = atomic_long_inc_return(&num_delegations);
1157 if (n < 0 || n > max_delegations)
1158 goto out_dec;
1159 if (delegation_blocked(&fp->fi_fhandle))
1160 goto out_dec;
1161 stid = nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg);
1162 if (stid == NULL)
1163 goto out_dec;
1164 dp = delegstateid(stid);
1165
1166 /*
1167 * delegation seqid's are never incremented. The 4.1 special
1168 * meaning of seqid 0 isn't meaningful, really, but let's avoid
1169 * 0 anyway just for consistency and use 1:
1170 */
1171 dp->dl_stid.sc_stateid.si_generation = 1;
1172 INIT_LIST_HEAD(&dp->dl_perfile);
1173 INIT_LIST_HEAD(&dp->dl_perclnt);
1174 INIT_LIST_HEAD(&dp->dl_recall_lru);
1175 dp->dl_clnt_odstate = odstate;
1176 get_clnt_odstate(odstate);
1177 dp->dl_type = dl_type;
1178 dp->dl_retries = 1;
1179 dp->dl_recalled = false;
1180 nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
1181 &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
1182 nfsd4_init_cb(&dp->dl_cb_fattr.ncf_getattr, dp->dl_stid.sc_client,
1183 &nfsd4_cb_getattr_ops, NFSPROC4_CLNT_CB_GETATTR);
1184 dp->dl_cb_fattr.ncf_file_modified = false;
1185 dp->dl_cb_fattr.ncf_cb_bmap[0] = FATTR4_WORD0_CHANGE | FATTR4_WORD0_SIZE;
1186 get_nfs4_file(fp);
1187 dp->dl_stid.sc_file = fp;
1188 return dp;
1189 out_dec:
1190 atomic_long_dec(&num_delegations);
1191 return NULL;
1192 }
1193
1194 void
nfs4_put_stid(struct nfs4_stid * s)1195 nfs4_put_stid(struct nfs4_stid *s)
1196 {
1197 struct nfs4_file *fp = s->sc_file;
1198 struct nfs4_client *clp = s->sc_client;
1199
1200 might_lock(&clp->cl_lock);
1201
1202 if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
1203 wake_up_all(&close_wq);
1204 return;
1205 }
1206 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1207 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1208 atomic_dec(&s->sc_client->cl_admin_revoked);
1209 nfs4_free_cpntf_statelist(clp->net, s);
1210 spin_unlock(&clp->cl_lock);
1211 s->sc_free(s);
1212 if (fp)
1213 put_nfs4_file(fp);
1214 }
1215
1216 void
nfs4_inc_and_copy_stateid(stateid_t * dst,struct nfs4_stid * stid)1217 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
1218 {
1219 stateid_t *src = &stid->sc_stateid;
1220
1221 spin_lock(&stid->sc_lock);
1222 if (unlikely(++src->si_generation == 0))
1223 src->si_generation = 1;
1224 memcpy(dst, src, sizeof(*dst));
1225 spin_unlock(&stid->sc_lock);
1226 }
1227
put_deleg_file(struct nfs4_file * fp)1228 static void put_deleg_file(struct nfs4_file *fp)
1229 {
1230 struct nfsd_file *nf = NULL;
1231
1232 spin_lock(&fp->fi_lock);
1233 if (--fp->fi_delegees == 0)
1234 swap(nf, fp->fi_deleg_file);
1235 spin_unlock(&fp->fi_lock);
1236
1237 if (nf)
1238 nfsd_file_put(nf);
1239 }
1240
nfs4_unlock_deleg_lease(struct nfs4_delegation * dp)1241 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
1242 {
1243 struct nfs4_file *fp = dp->dl_stid.sc_file;
1244 struct nfsd_file *nf = fp->fi_deleg_file;
1245
1246 WARN_ON_ONCE(!fp->fi_delegees);
1247
1248 kernel_setlease(nf->nf_file, F_UNLCK, NULL, (void **)&dp);
1249 put_deleg_file(fp);
1250 }
1251
destroy_unhashed_deleg(struct nfs4_delegation * dp)1252 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
1253 {
1254 put_clnt_odstate(dp->dl_clnt_odstate);
1255 nfs4_unlock_deleg_lease(dp);
1256 nfs4_put_stid(&dp->dl_stid);
1257 }
1258
1259 /**
1260 * nfs4_delegation_exists - Discover if this delegation already exists
1261 * @clp: a pointer to the nfs4_client we're granting a delegation to
1262 * @fp: a pointer to the nfs4_file we're granting a delegation on
1263 *
1264 * Return:
1265 * On success: true iff an existing delegation is found
1266 */
1267
1268 static bool
nfs4_delegation_exists(struct nfs4_client * clp,struct nfs4_file * fp)1269 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
1270 {
1271 struct nfs4_delegation *searchdp = NULL;
1272 struct nfs4_client *searchclp = NULL;
1273
1274 lockdep_assert_held(&state_lock);
1275 lockdep_assert_held(&fp->fi_lock);
1276
1277 list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
1278 searchclp = searchdp->dl_stid.sc_client;
1279 if (clp == searchclp) {
1280 return true;
1281 }
1282 }
1283 return false;
1284 }
1285
1286 /**
1287 * hash_delegation_locked - Add a delegation to the appropriate lists
1288 * @dp: a pointer to the nfs4_delegation we are adding.
1289 * @fp: a pointer to the nfs4_file we're granting a delegation on
1290 *
1291 * Return:
1292 * On success: NULL if the delegation was successfully hashed.
1293 *
1294 * On error: -EAGAIN if one was previously granted to this
1295 * nfs4_client for this nfs4_file. Delegation is not hashed.
1296 *
1297 */
1298
1299 static int
hash_delegation_locked(struct nfs4_delegation * dp,struct nfs4_file * fp)1300 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
1301 {
1302 struct nfs4_client *clp = dp->dl_stid.sc_client;
1303
1304 lockdep_assert_held(&state_lock);
1305 lockdep_assert_held(&fp->fi_lock);
1306 lockdep_assert_held(&clp->cl_lock);
1307
1308 if (nfs4_delegation_exists(clp, fp))
1309 return -EAGAIN;
1310 refcount_inc(&dp->dl_stid.sc_count);
1311 dp->dl_stid.sc_type = SC_TYPE_DELEG;
1312 list_add(&dp->dl_perfile, &fp->fi_delegations);
1313 list_add(&dp->dl_perclnt, &clp->cl_delegations);
1314 return 0;
1315 }
1316
delegation_hashed(struct nfs4_delegation * dp)1317 static bool delegation_hashed(struct nfs4_delegation *dp)
1318 {
1319 return !(list_empty(&dp->dl_perfile));
1320 }
1321
1322 static bool
unhash_delegation_locked(struct nfs4_delegation * dp,unsigned short statusmask)1323 unhash_delegation_locked(struct nfs4_delegation *dp, unsigned short statusmask)
1324 {
1325 struct nfs4_file *fp = dp->dl_stid.sc_file;
1326
1327 lockdep_assert_held(&state_lock);
1328
1329 if (!delegation_hashed(dp))
1330 return false;
1331
1332 if (statusmask == SC_STATUS_REVOKED &&
1333 dp->dl_stid.sc_client->cl_minorversion == 0)
1334 statusmask = SC_STATUS_CLOSED;
1335 dp->dl_stid.sc_status |= statusmask;
1336 if (statusmask & SC_STATUS_ADMIN_REVOKED)
1337 atomic_inc(&dp->dl_stid.sc_client->cl_admin_revoked);
1338
1339 /* Ensure that deleg break won't try to requeue it */
1340 ++dp->dl_time;
1341 spin_lock(&fp->fi_lock);
1342 list_del_init(&dp->dl_perclnt);
1343 list_del_init(&dp->dl_recall_lru);
1344 list_del_init(&dp->dl_perfile);
1345 spin_unlock(&fp->fi_lock);
1346 return true;
1347 }
1348
destroy_delegation(struct nfs4_delegation * dp)1349 static void destroy_delegation(struct nfs4_delegation *dp)
1350 {
1351 bool unhashed;
1352
1353 spin_lock(&state_lock);
1354 unhashed = unhash_delegation_locked(dp, SC_STATUS_CLOSED);
1355 spin_unlock(&state_lock);
1356 if (unhashed)
1357 destroy_unhashed_deleg(dp);
1358 }
1359
revoke_delegation(struct nfs4_delegation * dp)1360 static void revoke_delegation(struct nfs4_delegation *dp)
1361 {
1362 struct nfs4_client *clp = dp->dl_stid.sc_client;
1363
1364 WARN_ON(!list_empty(&dp->dl_recall_lru));
1365
1366 trace_nfsd_stid_revoke(&dp->dl_stid);
1367
1368 if (dp->dl_stid.sc_status &
1369 (SC_STATUS_REVOKED | SC_STATUS_ADMIN_REVOKED)) {
1370 spin_lock(&clp->cl_lock);
1371 refcount_inc(&dp->dl_stid.sc_count);
1372 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1373 spin_unlock(&clp->cl_lock);
1374 }
1375 destroy_unhashed_deleg(dp);
1376 }
1377
1378 /*
1379 * SETCLIENTID state
1380 */
1381
clientid_hashval(u32 id)1382 static unsigned int clientid_hashval(u32 id)
1383 {
1384 return id & CLIENT_HASH_MASK;
1385 }
1386
clientstr_hashval(struct xdr_netobj name)1387 static unsigned int clientstr_hashval(struct xdr_netobj name)
1388 {
1389 return opaque_hashval(name.data, 8) & CLIENT_HASH_MASK;
1390 }
1391
1392 /*
1393 * A stateid that had a deny mode associated with it is being released
1394 * or downgraded. Recalculate the deny mode on the file.
1395 */
1396 static void
recalculate_deny_mode(struct nfs4_file * fp)1397 recalculate_deny_mode(struct nfs4_file *fp)
1398 {
1399 struct nfs4_ol_stateid *stp;
1400 u32 old_deny;
1401
1402 spin_lock(&fp->fi_lock);
1403 old_deny = fp->fi_share_deny;
1404 fp->fi_share_deny = 0;
1405 list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1406 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1407 if (fp->fi_share_deny == old_deny)
1408 break;
1409 }
1410 spin_unlock(&fp->fi_lock);
1411 }
1412
1413 static void
reset_union_bmap_deny(u32 deny,struct nfs4_ol_stateid * stp)1414 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1415 {
1416 int i;
1417 bool change = false;
1418
1419 for (i = 1; i < 4; i++) {
1420 if ((i & deny) != i) {
1421 change = true;
1422 clear_deny(i, stp);
1423 }
1424 }
1425
1426 /* Recalculate per-file deny mode if there was a change */
1427 if (change)
1428 recalculate_deny_mode(stp->st_stid.sc_file);
1429 }
1430
1431 /* release all access and file references for a given stateid */
1432 static void
release_all_access(struct nfs4_ol_stateid * stp)1433 release_all_access(struct nfs4_ol_stateid *stp)
1434 {
1435 int i;
1436 struct nfs4_file *fp = stp->st_stid.sc_file;
1437
1438 if (fp && stp->st_deny_bmap != 0)
1439 recalculate_deny_mode(fp);
1440
1441 for (i = 1; i < 4; i++) {
1442 if (test_access(i, stp))
1443 nfs4_file_put_access(stp->st_stid.sc_file, i);
1444 clear_access(i, stp);
1445 }
1446 }
1447
nfs4_free_stateowner(struct nfs4_stateowner * sop)1448 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1449 {
1450 kfree(sop->so_owner.data);
1451 sop->so_ops->so_free(sop);
1452 }
1453
nfs4_put_stateowner(struct nfs4_stateowner * sop)1454 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1455 {
1456 struct nfs4_client *clp = sop->so_client;
1457
1458 might_lock(&clp->cl_lock);
1459
1460 if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1461 return;
1462 sop->so_ops->so_unhash(sop);
1463 spin_unlock(&clp->cl_lock);
1464 nfs4_free_stateowner(sop);
1465 }
1466
1467 static bool
nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid * stp)1468 nfs4_ol_stateid_unhashed(const struct nfs4_ol_stateid *stp)
1469 {
1470 return list_empty(&stp->st_perfile);
1471 }
1472
unhash_ol_stateid(struct nfs4_ol_stateid * stp)1473 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1474 {
1475 struct nfs4_file *fp = stp->st_stid.sc_file;
1476
1477 lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1478
1479 if (list_empty(&stp->st_perfile))
1480 return false;
1481
1482 spin_lock(&fp->fi_lock);
1483 list_del_init(&stp->st_perfile);
1484 spin_unlock(&fp->fi_lock);
1485 list_del(&stp->st_perstateowner);
1486 return true;
1487 }
1488
nfs4_free_ol_stateid(struct nfs4_stid * stid)1489 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1490 {
1491 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1492
1493 put_clnt_odstate(stp->st_clnt_odstate);
1494 release_all_access(stp);
1495 if (stp->st_stateowner)
1496 nfs4_put_stateowner(stp->st_stateowner);
1497 WARN_ON(!list_empty(&stid->sc_cp_list));
1498 kmem_cache_free(stateid_slab, stid);
1499 }
1500
nfs4_free_lock_stateid(struct nfs4_stid * stid)1501 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1502 {
1503 struct nfs4_ol_stateid *stp = openlockstateid(stid);
1504 struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1505 struct nfsd_file *nf;
1506
1507 nf = find_any_file(stp->st_stid.sc_file);
1508 if (nf) {
1509 get_file(nf->nf_file);
1510 filp_close(nf->nf_file, (fl_owner_t)lo);
1511 nfsd_file_put(nf);
1512 }
1513 nfs4_free_ol_stateid(stid);
1514 }
1515
1516 /*
1517 * Put the persistent reference to an already unhashed generic stateid, while
1518 * holding the cl_lock. If it's the last reference, then put it onto the
1519 * reaplist for later destruction.
1520 */
put_ol_stateid_locked(struct nfs4_ol_stateid * stp,struct list_head * reaplist)1521 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1522 struct list_head *reaplist)
1523 {
1524 struct nfs4_stid *s = &stp->st_stid;
1525 struct nfs4_client *clp = s->sc_client;
1526
1527 lockdep_assert_held(&clp->cl_lock);
1528
1529 WARN_ON_ONCE(!list_empty(&stp->st_locks));
1530
1531 if (!refcount_dec_and_test(&s->sc_count)) {
1532 wake_up_all(&close_wq);
1533 return;
1534 }
1535
1536 idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1537 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
1538 atomic_dec(&s->sc_client->cl_admin_revoked);
1539 list_add(&stp->st_locks, reaplist);
1540 }
1541
unhash_lock_stateid(struct nfs4_ol_stateid * stp)1542 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1543 {
1544 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1545
1546 if (!unhash_ol_stateid(stp))
1547 return false;
1548 list_del_init(&stp->st_locks);
1549 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1550 return true;
1551 }
1552
release_lock_stateid(struct nfs4_ol_stateid * stp)1553 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1554 {
1555 struct nfs4_client *clp = stp->st_stid.sc_client;
1556 bool unhashed;
1557
1558 spin_lock(&clp->cl_lock);
1559 unhashed = unhash_lock_stateid(stp);
1560 spin_unlock(&clp->cl_lock);
1561 if (unhashed)
1562 nfs4_put_stid(&stp->st_stid);
1563 }
1564
unhash_lockowner_locked(struct nfs4_lockowner * lo)1565 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1566 {
1567 struct nfs4_client *clp = lo->lo_owner.so_client;
1568
1569 lockdep_assert_held(&clp->cl_lock);
1570
1571 list_del_init(&lo->lo_owner.so_strhash);
1572 }
1573
1574 /*
1575 * Free a list of generic stateids that were collected earlier after being
1576 * fully unhashed.
1577 */
1578 static void
free_ol_stateid_reaplist(struct list_head * reaplist)1579 free_ol_stateid_reaplist(struct list_head *reaplist)
1580 {
1581 struct nfs4_ol_stateid *stp;
1582 struct nfs4_file *fp;
1583
1584 might_sleep();
1585
1586 while (!list_empty(reaplist)) {
1587 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1588 st_locks);
1589 list_del(&stp->st_locks);
1590 fp = stp->st_stid.sc_file;
1591 stp->st_stid.sc_free(&stp->st_stid);
1592 if (fp)
1593 put_nfs4_file(fp);
1594 }
1595 }
1596
release_open_stateid_locks(struct nfs4_ol_stateid * open_stp,struct list_head * reaplist)1597 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1598 struct list_head *reaplist)
1599 {
1600 struct nfs4_ol_stateid *stp;
1601
1602 lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1603
1604 while (!list_empty(&open_stp->st_locks)) {
1605 stp = list_entry(open_stp->st_locks.next,
1606 struct nfs4_ol_stateid, st_locks);
1607 unhash_lock_stateid(stp);
1608 put_ol_stateid_locked(stp, reaplist);
1609 }
1610 }
1611
unhash_open_stateid(struct nfs4_ol_stateid * stp,struct list_head * reaplist)1612 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1613 struct list_head *reaplist)
1614 {
1615 lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1616
1617 if (!unhash_ol_stateid(stp))
1618 return false;
1619 release_open_stateid_locks(stp, reaplist);
1620 return true;
1621 }
1622
release_open_stateid(struct nfs4_ol_stateid * stp)1623 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1624 {
1625 LIST_HEAD(reaplist);
1626
1627 spin_lock(&stp->st_stid.sc_client->cl_lock);
1628 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
1629 if (unhash_open_stateid(stp, &reaplist))
1630 put_ol_stateid_locked(stp, &reaplist);
1631 spin_unlock(&stp->st_stid.sc_client->cl_lock);
1632 free_ol_stateid_reaplist(&reaplist);
1633 }
1634
unhash_openowner_locked(struct nfs4_openowner * oo)1635 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1636 {
1637 struct nfs4_client *clp = oo->oo_owner.so_client;
1638
1639 lockdep_assert_held(&clp->cl_lock);
1640
1641 list_del_init(&oo->oo_owner.so_strhash);
1642 list_del_init(&oo->oo_perclient);
1643 }
1644
release_last_closed_stateid(struct nfs4_openowner * oo)1645 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1646 {
1647 struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1648 nfsd_net_id);
1649 struct nfs4_ol_stateid *s;
1650
1651 spin_lock(&nn->client_lock);
1652 s = oo->oo_last_closed_stid;
1653 if (s) {
1654 list_del_init(&oo->oo_close_lru);
1655 oo->oo_last_closed_stid = NULL;
1656 }
1657 spin_unlock(&nn->client_lock);
1658 if (s)
1659 nfs4_put_stid(&s->st_stid);
1660 }
1661
release_openowner(struct nfs4_openowner * oo)1662 static void release_openowner(struct nfs4_openowner *oo)
1663 {
1664 struct nfs4_ol_stateid *stp;
1665 struct nfs4_client *clp = oo->oo_owner.so_client;
1666 struct list_head reaplist;
1667
1668 INIT_LIST_HEAD(&reaplist);
1669
1670 spin_lock(&clp->cl_lock);
1671 unhash_openowner_locked(oo);
1672 while (!list_empty(&oo->oo_owner.so_stateids)) {
1673 stp = list_first_entry(&oo->oo_owner.so_stateids,
1674 struct nfs4_ol_stateid, st_perstateowner);
1675 if (unhash_open_stateid(stp, &reaplist))
1676 put_ol_stateid_locked(stp, &reaplist);
1677 }
1678 spin_unlock(&clp->cl_lock);
1679 free_ol_stateid_reaplist(&reaplist);
1680 release_last_closed_stateid(oo);
1681 nfs4_put_stateowner(&oo->oo_owner);
1682 }
1683
find_one_sb_stid(struct nfs4_client * clp,struct super_block * sb,unsigned int sc_types)1684 static struct nfs4_stid *find_one_sb_stid(struct nfs4_client *clp,
1685 struct super_block *sb,
1686 unsigned int sc_types)
1687 {
1688 unsigned long id, tmp;
1689 struct nfs4_stid *stid;
1690
1691 spin_lock(&clp->cl_lock);
1692 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
1693 if ((stid->sc_type & sc_types) &&
1694 stid->sc_status == 0 &&
1695 stid->sc_file->fi_inode->i_sb == sb) {
1696 refcount_inc(&stid->sc_count);
1697 break;
1698 }
1699 spin_unlock(&clp->cl_lock);
1700 return stid;
1701 }
1702
1703 /**
1704 * nfsd4_revoke_states - revoke all nfsv4 states associated with given filesystem
1705 * @net: used to identify instance of nfsd (there is one per net namespace)
1706 * @sb: super_block used to identify target filesystem
1707 *
1708 * All nfs4 states (open, lock, delegation, layout) held by the server instance
1709 * and associated with a file on the given filesystem will be revoked resulting
1710 * in any files being closed and so all references from nfsd to the filesystem
1711 * being released. Thus nfsd will no longer prevent the filesystem from being
1712 * unmounted.
1713 *
1714 * The clients which own the states will subsequently being notified that the
1715 * states have been "admin-revoked".
1716 */
nfsd4_revoke_states(struct net * net,struct super_block * sb)1717 void nfsd4_revoke_states(struct net *net, struct super_block *sb)
1718 {
1719 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1720 unsigned int idhashval;
1721 unsigned int sc_types;
1722
1723 sc_types = SC_TYPE_OPEN | SC_TYPE_LOCK | SC_TYPE_DELEG | SC_TYPE_LAYOUT;
1724
1725 spin_lock(&nn->client_lock);
1726 for (idhashval = 0; idhashval < CLIENT_HASH_MASK; idhashval++) {
1727 struct list_head *head = &nn->conf_id_hashtbl[idhashval];
1728 struct nfs4_client *clp;
1729 retry:
1730 list_for_each_entry(clp, head, cl_idhash) {
1731 struct nfs4_stid *stid = find_one_sb_stid(clp, sb,
1732 sc_types);
1733 if (stid) {
1734 struct nfs4_ol_stateid *stp;
1735 struct nfs4_delegation *dp;
1736 struct nfs4_layout_stateid *ls;
1737
1738 spin_unlock(&nn->client_lock);
1739 switch (stid->sc_type) {
1740 case SC_TYPE_OPEN:
1741 stp = openlockstateid(stid);
1742 mutex_lock_nested(&stp->st_mutex,
1743 OPEN_STATEID_MUTEX);
1744
1745 spin_lock(&clp->cl_lock);
1746 if (stid->sc_status == 0) {
1747 stid->sc_status |=
1748 SC_STATUS_ADMIN_REVOKED;
1749 atomic_inc(&clp->cl_admin_revoked);
1750 spin_unlock(&clp->cl_lock);
1751 release_all_access(stp);
1752 } else
1753 spin_unlock(&clp->cl_lock);
1754 mutex_unlock(&stp->st_mutex);
1755 break;
1756 case SC_TYPE_LOCK:
1757 stp = openlockstateid(stid);
1758 mutex_lock_nested(&stp->st_mutex,
1759 LOCK_STATEID_MUTEX);
1760 spin_lock(&clp->cl_lock);
1761 if (stid->sc_status == 0) {
1762 struct nfs4_lockowner *lo =
1763 lockowner(stp->st_stateowner);
1764 struct nfsd_file *nf;
1765
1766 stid->sc_status |=
1767 SC_STATUS_ADMIN_REVOKED;
1768 atomic_inc(&clp->cl_admin_revoked);
1769 spin_unlock(&clp->cl_lock);
1770 nf = find_any_file(stp->st_stid.sc_file);
1771 if (nf) {
1772 get_file(nf->nf_file);
1773 filp_close(nf->nf_file,
1774 (fl_owner_t)lo);
1775 nfsd_file_put(nf);
1776 }
1777 release_all_access(stp);
1778 } else
1779 spin_unlock(&clp->cl_lock);
1780 mutex_unlock(&stp->st_mutex);
1781 break;
1782 case SC_TYPE_DELEG:
1783 dp = delegstateid(stid);
1784 spin_lock(&state_lock);
1785 if (!unhash_delegation_locked(
1786 dp, SC_STATUS_ADMIN_REVOKED))
1787 dp = NULL;
1788 spin_unlock(&state_lock);
1789 if (dp)
1790 revoke_delegation(dp);
1791 break;
1792 case SC_TYPE_LAYOUT:
1793 ls = layoutstateid(stid);
1794 nfsd4_close_layout(ls);
1795 break;
1796 }
1797 nfs4_put_stid(stid);
1798 spin_lock(&nn->client_lock);
1799 if (clp->cl_minorversion == 0)
1800 /* Allow cleanup after a lease period.
1801 * store_release ensures cleanup will
1802 * see any newly revoked states if it
1803 * sees the time updated.
1804 */
1805 nn->nfs40_last_revoke =
1806 ktime_get_boottime_seconds();
1807 goto retry;
1808 }
1809 }
1810 }
1811 spin_unlock(&nn->client_lock);
1812 }
1813
1814 static inline int
hash_sessionid(struct nfs4_sessionid * sessionid)1815 hash_sessionid(struct nfs4_sessionid *sessionid)
1816 {
1817 struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1818
1819 return sid->sequence % SESSION_HASH_SIZE;
1820 }
1821
1822 #ifdef CONFIG_SUNRPC_DEBUG
1823 static inline void
dump_sessionid(const char * fn,struct nfs4_sessionid * sessionid)1824 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1825 {
1826 u32 *ptr = (u32 *)(&sessionid->data[0]);
1827 dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1828 }
1829 #else
1830 static inline void
dump_sessionid(const char * fn,struct nfs4_sessionid * sessionid)1831 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1832 {
1833 }
1834 #endif
1835
1836 /*
1837 * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1838 * won't be used for replay.
1839 */
nfsd4_bump_seqid(struct nfsd4_compound_state * cstate,__be32 nfserr)1840 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1841 {
1842 struct nfs4_stateowner *so = cstate->replay_owner;
1843
1844 if (nfserr == nfserr_replay_me)
1845 return;
1846
1847 if (!seqid_mutating_err(ntohl(nfserr))) {
1848 nfsd4_cstate_clear_replay(cstate);
1849 return;
1850 }
1851 if (!so)
1852 return;
1853 if (so->so_is_open_owner)
1854 release_last_closed_stateid(openowner(so));
1855 so->so_seqid++;
1856 return;
1857 }
1858
1859 static void
gen_sessionid(struct nfsd4_session * ses)1860 gen_sessionid(struct nfsd4_session *ses)
1861 {
1862 struct nfs4_client *clp = ses->se_client;
1863 struct nfsd4_sessionid *sid;
1864
1865 sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1866 sid->clientid = clp->cl_clientid;
1867 sid->sequence = current_sessionid++;
1868 sid->reserved = 0;
1869 }
1870
1871 /*
1872 * The protocol defines ca_maxresponssize_cached to include the size of
1873 * the rpc header, but all we need to cache is the data starting after
1874 * the end of the initial SEQUENCE operation--the rest we regenerate
1875 * each time. Therefore we can advertise a ca_maxresponssize_cached
1876 * value that is the number of bytes in our cache plus a few additional
1877 * bytes. In order to stay on the safe side, and not promise more than
1878 * we can cache, those additional bytes must be the minimum possible: 24
1879 * bytes of rpc header (xid through accept state, with AUTH_NULL
1880 * verifier), 12 for the compound header (with zero-length tag), and 44
1881 * for the SEQUENCE op response:
1882 */
1883 #define NFSD_MIN_HDR_SEQ_SZ (24 + 12 + 44)
1884
1885 static void
free_session_slots(struct nfsd4_session * ses)1886 free_session_slots(struct nfsd4_session *ses)
1887 {
1888 int i;
1889
1890 for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1891 free_svc_cred(&ses->se_slots[i]->sl_cred);
1892 kfree(ses->se_slots[i]);
1893 }
1894 }
1895
1896 /*
1897 * We don't actually need to cache the rpc and session headers, so we
1898 * can allocate a little less for each slot:
1899 */
slot_bytes(struct nfsd4_channel_attrs * ca)1900 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1901 {
1902 u32 size;
1903
1904 if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1905 size = 0;
1906 else
1907 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1908 return size + sizeof(struct nfsd4_slot);
1909 }
1910
1911 /*
1912 * XXX: If we run out of reserved DRC memory we could (up to a point)
1913 * re-negotiate active sessions and reduce their slot usage to make
1914 * room for new connections. For now we just fail the create session.
1915 */
nfsd4_get_drc_mem(struct nfsd4_channel_attrs * ca,struct nfsd_net * nn)1916 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
1917 {
1918 u32 slotsize = slot_bytes(ca);
1919 u32 num = ca->maxreqs;
1920 unsigned long avail, total_avail;
1921 unsigned int scale_factor;
1922
1923 spin_lock(&nfsd_drc_lock);
1924 if (nfsd_drc_max_mem > nfsd_drc_mem_used)
1925 total_avail = nfsd_drc_max_mem - nfsd_drc_mem_used;
1926 else
1927 /* We have handed out more space than we chose in
1928 * set_max_drc() to allow. That isn't really a
1929 * problem as long as that doesn't make us think we
1930 * have lots more due to integer overflow.
1931 */
1932 total_avail = 0;
1933 avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION, total_avail);
1934 /*
1935 * Never use more than a fraction of the remaining memory,
1936 * unless it's the only way to give this client a slot.
1937 * The chosen fraction is either 1/8 or 1/number of threads,
1938 * whichever is smaller. This ensures there are adequate
1939 * slots to support multiple clients per thread.
1940 * Give the client one slot even if that would require
1941 * over-allocation--it is better than failure.
1942 */
1943 scale_factor = max_t(unsigned int, 8, nn->nfsd_serv->sv_nrthreads);
1944
1945 avail = clamp_t(unsigned long, avail, slotsize,
1946 total_avail/scale_factor);
1947 num = min_t(int, num, avail / slotsize);
1948 num = max_t(int, num, 1);
1949 nfsd_drc_mem_used += num * slotsize;
1950 spin_unlock(&nfsd_drc_lock);
1951
1952 return num;
1953 }
1954
nfsd4_put_drc_mem(struct nfsd4_channel_attrs * ca)1955 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1956 {
1957 int slotsize = slot_bytes(ca);
1958
1959 spin_lock(&nfsd_drc_lock);
1960 nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1961 spin_unlock(&nfsd_drc_lock);
1962 }
1963
alloc_session(struct nfsd4_channel_attrs * fattrs,struct nfsd4_channel_attrs * battrs)1964 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1965 struct nfsd4_channel_attrs *battrs)
1966 {
1967 int numslots = fattrs->maxreqs;
1968 int slotsize = slot_bytes(fattrs);
1969 struct nfsd4_session *new;
1970 int i;
1971
1972 BUILD_BUG_ON(struct_size(new, se_slots, NFSD_MAX_SLOTS_PER_SESSION)
1973 > PAGE_SIZE);
1974
1975 new = kzalloc(struct_size(new, se_slots, numslots), GFP_KERNEL);
1976 if (!new)
1977 return NULL;
1978 /* allocate each struct nfsd4_slot and data cache in one piece */
1979 for (i = 0; i < numslots; i++) {
1980 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1981 if (!new->se_slots[i])
1982 goto out_free;
1983 }
1984
1985 memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1986 memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1987
1988 return new;
1989 out_free:
1990 while (i--)
1991 kfree(new->se_slots[i]);
1992 kfree(new);
1993 return NULL;
1994 }
1995
free_conn(struct nfsd4_conn * c)1996 static void free_conn(struct nfsd4_conn *c)
1997 {
1998 svc_xprt_put(c->cn_xprt);
1999 kfree(c);
2000 }
2001
nfsd4_conn_lost(struct svc_xpt_user * u)2002 static void nfsd4_conn_lost(struct svc_xpt_user *u)
2003 {
2004 struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
2005 struct nfs4_client *clp = c->cn_session->se_client;
2006
2007 trace_nfsd_cb_lost(clp);
2008
2009 spin_lock(&clp->cl_lock);
2010 if (!list_empty(&c->cn_persession)) {
2011 list_del(&c->cn_persession);
2012 free_conn(c);
2013 }
2014 nfsd4_probe_callback(clp);
2015 spin_unlock(&clp->cl_lock);
2016 }
2017
alloc_conn(struct svc_rqst * rqstp,u32 flags)2018 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
2019 {
2020 struct nfsd4_conn *conn;
2021
2022 conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
2023 if (!conn)
2024 return NULL;
2025 svc_xprt_get(rqstp->rq_xprt);
2026 conn->cn_xprt = rqstp->rq_xprt;
2027 conn->cn_flags = flags;
2028 INIT_LIST_HEAD(&conn->cn_xpt_user.list);
2029 return conn;
2030 }
2031
__nfsd4_hash_conn(struct nfsd4_conn * conn,struct nfsd4_session * ses)2032 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2033 {
2034 conn->cn_session = ses;
2035 list_add(&conn->cn_persession, &ses->se_conns);
2036 }
2037
nfsd4_hash_conn(struct nfsd4_conn * conn,struct nfsd4_session * ses)2038 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
2039 {
2040 struct nfs4_client *clp = ses->se_client;
2041
2042 spin_lock(&clp->cl_lock);
2043 __nfsd4_hash_conn(conn, ses);
2044 spin_unlock(&clp->cl_lock);
2045 }
2046
nfsd4_register_conn(struct nfsd4_conn * conn)2047 static int nfsd4_register_conn(struct nfsd4_conn *conn)
2048 {
2049 conn->cn_xpt_user.callback = nfsd4_conn_lost;
2050 return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
2051 }
2052
nfsd4_init_conn(struct svc_rqst * rqstp,struct nfsd4_conn * conn,struct nfsd4_session * ses)2053 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
2054 {
2055 int ret;
2056
2057 nfsd4_hash_conn(conn, ses);
2058 ret = nfsd4_register_conn(conn);
2059 if (ret)
2060 /* oops; xprt is already down: */
2061 nfsd4_conn_lost(&conn->cn_xpt_user);
2062 /* We may have gained or lost a callback channel: */
2063 nfsd4_probe_callback_sync(ses->se_client);
2064 }
2065
alloc_conn_from_crses(struct svc_rqst * rqstp,struct nfsd4_create_session * cses)2066 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
2067 {
2068 u32 dir = NFS4_CDFC4_FORE;
2069
2070 if (cses->flags & SESSION4_BACK_CHAN)
2071 dir |= NFS4_CDFC4_BACK;
2072 return alloc_conn(rqstp, dir);
2073 }
2074
2075 /* must be called under client_lock */
nfsd4_del_conns(struct nfsd4_session * s)2076 static void nfsd4_del_conns(struct nfsd4_session *s)
2077 {
2078 struct nfs4_client *clp = s->se_client;
2079 struct nfsd4_conn *c;
2080
2081 spin_lock(&clp->cl_lock);
2082 while (!list_empty(&s->se_conns)) {
2083 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
2084 list_del_init(&c->cn_persession);
2085 spin_unlock(&clp->cl_lock);
2086
2087 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
2088 free_conn(c);
2089
2090 spin_lock(&clp->cl_lock);
2091 }
2092 spin_unlock(&clp->cl_lock);
2093 }
2094
__free_session(struct nfsd4_session * ses)2095 static void __free_session(struct nfsd4_session *ses)
2096 {
2097 free_session_slots(ses);
2098 kfree(ses);
2099 }
2100
free_session(struct nfsd4_session * ses)2101 static void free_session(struct nfsd4_session *ses)
2102 {
2103 nfsd4_del_conns(ses);
2104 nfsd4_put_drc_mem(&ses->se_fchannel);
2105 __free_session(ses);
2106 }
2107
init_session(struct svc_rqst * rqstp,struct nfsd4_session * new,struct nfs4_client * clp,struct nfsd4_create_session * cses)2108 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
2109 {
2110 int idx;
2111 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2112
2113 new->se_client = clp;
2114 gen_sessionid(new);
2115
2116 INIT_LIST_HEAD(&new->se_conns);
2117
2118 new->se_cb_seq_nr = 1;
2119 new->se_flags = cses->flags;
2120 new->se_cb_prog = cses->callback_prog;
2121 new->se_cb_sec = cses->cb_sec;
2122 atomic_set(&new->se_ref, 0);
2123 idx = hash_sessionid(&new->se_sessionid);
2124 list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
2125 spin_lock(&clp->cl_lock);
2126 list_add(&new->se_perclnt, &clp->cl_sessions);
2127 spin_unlock(&clp->cl_lock);
2128
2129 {
2130 struct sockaddr *sa = svc_addr(rqstp);
2131 /*
2132 * This is a little silly; with sessions there's no real
2133 * use for the callback address. Use the peer address
2134 * as a reasonable default for now, but consider fixing
2135 * the rpc client not to require an address in the
2136 * future:
2137 */
2138 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
2139 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
2140 }
2141 }
2142
2143 /* caller must hold client_lock */
2144 static struct nfsd4_session *
__find_in_sessionid_hashtbl(struct nfs4_sessionid * sessionid,struct net * net)2145 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
2146 {
2147 struct nfsd4_session *elem;
2148 int idx;
2149 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2150
2151 lockdep_assert_held(&nn->client_lock);
2152
2153 dump_sessionid(__func__, sessionid);
2154 idx = hash_sessionid(sessionid);
2155 /* Search in the appropriate list */
2156 list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
2157 if (!memcmp(elem->se_sessionid.data, sessionid->data,
2158 NFS4_MAX_SESSIONID_LEN)) {
2159 return elem;
2160 }
2161 }
2162
2163 dprintk("%s: session not found\n", __func__);
2164 return NULL;
2165 }
2166
2167 static struct nfsd4_session *
find_in_sessionid_hashtbl(struct nfs4_sessionid * sessionid,struct net * net,__be32 * ret)2168 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
2169 __be32 *ret)
2170 {
2171 struct nfsd4_session *session;
2172 __be32 status = nfserr_badsession;
2173
2174 session = __find_in_sessionid_hashtbl(sessionid, net);
2175 if (!session)
2176 goto out;
2177 status = nfsd4_get_session_locked(session);
2178 if (status)
2179 session = NULL;
2180 out:
2181 *ret = status;
2182 return session;
2183 }
2184
2185 /* caller must hold client_lock */
2186 static void
unhash_session(struct nfsd4_session * ses)2187 unhash_session(struct nfsd4_session *ses)
2188 {
2189 struct nfs4_client *clp = ses->se_client;
2190 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2191
2192 lockdep_assert_held(&nn->client_lock);
2193
2194 list_del(&ses->se_hash);
2195 spin_lock(&ses->se_client->cl_lock);
2196 list_del(&ses->se_perclnt);
2197 spin_unlock(&ses->se_client->cl_lock);
2198 }
2199
2200 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
2201 static int
STALE_CLIENTID(clientid_t * clid,struct nfsd_net * nn)2202 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
2203 {
2204 /*
2205 * We're assuming the clid was not given out from a boot
2206 * precisely 2^32 (about 136 years) before this one. That seems
2207 * a safe assumption:
2208 */
2209 if (clid->cl_boot == (u32)nn->boot_time)
2210 return 0;
2211 trace_nfsd_clid_stale(clid);
2212 return 1;
2213 }
2214
2215 /*
2216 * XXX Should we use a slab cache ?
2217 * This type of memory management is somewhat inefficient, but we use it
2218 * anyway since SETCLIENTID is not a common operation.
2219 */
alloc_client(struct xdr_netobj name,struct nfsd_net * nn)2220 static struct nfs4_client *alloc_client(struct xdr_netobj name,
2221 struct nfsd_net *nn)
2222 {
2223 struct nfs4_client *clp;
2224 int i;
2225
2226 if (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) {
2227 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
2228 return NULL;
2229 }
2230 clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
2231 if (clp == NULL)
2232 return NULL;
2233 xdr_netobj_dup(&clp->cl_name, &name, GFP_KERNEL);
2234 if (clp->cl_name.data == NULL)
2235 goto err_no_name;
2236 clp->cl_ownerstr_hashtbl = kmalloc_array(OWNER_HASH_SIZE,
2237 sizeof(struct list_head),
2238 GFP_KERNEL);
2239 if (!clp->cl_ownerstr_hashtbl)
2240 goto err_no_hashtbl;
2241 clp->cl_callback_wq = alloc_ordered_workqueue("nfsd4_callbacks", 0);
2242 if (!clp->cl_callback_wq)
2243 goto err_no_callback_wq;
2244
2245 for (i = 0; i < OWNER_HASH_SIZE; i++)
2246 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
2247 INIT_LIST_HEAD(&clp->cl_sessions);
2248 idr_init(&clp->cl_stateids);
2249 atomic_set(&clp->cl_rpc_users, 0);
2250 clp->cl_cb_state = NFSD4_CB_UNKNOWN;
2251 clp->cl_state = NFSD4_ACTIVE;
2252 atomic_inc(&nn->nfs4_client_count);
2253 atomic_set(&clp->cl_delegs_in_recall, 0);
2254 INIT_LIST_HEAD(&clp->cl_idhash);
2255 INIT_LIST_HEAD(&clp->cl_openowners);
2256 INIT_LIST_HEAD(&clp->cl_delegations);
2257 INIT_LIST_HEAD(&clp->cl_lru);
2258 INIT_LIST_HEAD(&clp->cl_revoked);
2259 #ifdef CONFIG_NFSD_PNFS
2260 INIT_LIST_HEAD(&clp->cl_lo_states);
2261 #endif
2262 INIT_LIST_HEAD(&clp->async_copies);
2263 spin_lock_init(&clp->async_lock);
2264 spin_lock_init(&clp->cl_lock);
2265 rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
2266 return clp;
2267 err_no_callback_wq:
2268 kfree(clp->cl_ownerstr_hashtbl);
2269 err_no_hashtbl:
2270 kfree(clp->cl_name.data);
2271 err_no_name:
2272 kmem_cache_free(client_slab, clp);
2273 return NULL;
2274 }
2275
__free_client(struct kref * k)2276 static void __free_client(struct kref *k)
2277 {
2278 struct nfsdfs_client *c = container_of(k, struct nfsdfs_client, cl_ref);
2279 struct nfs4_client *clp = container_of(c, struct nfs4_client, cl_nfsdfs);
2280
2281 free_svc_cred(&clp->cl_cred);
2282 destroy_workqueue(clp->cl_callback_wq);
2283 kfree(clp->cl_ownerstr_hashtbl);
2284 kfree(clp->cl_name.data);
2285 kfree(clp->cl_nii_domain.data);
2286 kfree(clp->cl_nii_name.data);
2287 idr_destroy(&clp->cl_stateids);
2288 kfree(clp->cl_ra);
2289 kmem_cache_free(client_slab, clp);
2290 }
2291
drop_client(struct nfs4_client * clp)2292 static void drop_client(struct nfs4_client *clp)
2293 {
2294 kref_put(&clp->cl_nfsdfs.cl_ref, __free_client);
2295 }
2296
2297 static void
free_client(struct nfs4_client * clp)2298 free_client(struct nfs4_client *clp)
2299 {
2300 while (!list_empty(&clp->cl_sessions)) {
2301 struct nfsd4_session *ses;
2302 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
2303 se_perclnt);
2304 list_del(&ses->se_perclnt);
2305 WARN_ON_ONCE(atomic_read(&ses->se_ref));
2306 free_session(ses);
2307 }
2308 rpc_destroy_wait_queue(&clp->cl_cb_waitq);
2309 if (clp->cl_nfsd_dentry) {
2310 nfsd_client_rmdir(clp->cl_nfsd_dentry);
2311 clp->cl_nfsd_dentry = NULL;
2312 wake_up_all(&expiry_wq);
2313 }
2314 drop_client(clp);
2315 }
2316
2317 /* must be called under the client_lock */
2318 static void
unhash_client_locked(struct nfs4_client * clp)2319 unhash_client_locked(struct nfs4_client *clp)
2320 {
2321 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2322 struct nfsd4_session *ses;
2323
2324 lockdep_assert_held(&nn->client_lock);
2325
2326 /* Mark the client as expired! */
2327 clp->cl_time = 0;
2328 /* Make it invisible */
2329 if (!list_empty(&clp->cl_idhash)) {
2330 list_del_init(&clp->cl_idhash);
2331 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2332 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
2333 else
2334 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2335 }
2336 list_del_init(&clp->cl_lru);
2337 spin_lock(&clp->cl_lock);
2338 list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
2339 list_del_init(&ses->se_hash);
2340 spin_unlock(&clp->cl_lock);
2341 }
2342
2343 static void
unhash_client(struct nfs4_client * clp)2344 unhash_client(struct nfs4_client *clp)
2345 {
2346 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2347
2348 spin_lock(&nn->client_lock);
2349 unhash_client_locked(clp);
2350 spin_unlock(&nn->client_lock);
2351 }
2352
mark_client_expired_locked(struct nfs4_client * clp)2353 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
2354 {
2355 int users = atomic_read(&clp->cl_rpc_users);
2356
2357 trace_nfsd_mark_client_expired(clp, users);
2358
2359 if (users)
2360 return nfserr_jukebox;
2361 unhash_client_locked(clp);
2362 return nfs_ok;
2363 }
2364
2365 static void
__destroy_client(struct nfs4_client * clp)2366 __destroy_client(struct nfs4_client *clp)
2367 {
2368 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2369 int i;
2370 struct nfs4_openowner *oo;
2371 struct nfs4_delegation *dp;
2372 struct list_head reaplist;
2373
2374 INIT_LIST_HEAD(&reaplist);
2375 spin_lock(&state_lock);
2376 while (!list_empty(&clp->cl_delegations)) {
2377 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
2378 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
2379 list_add(&dp->dl_recall_lru, &reaplist);
2380 }
2381 spin_unlock(&state_lock);
2382 while (!list_empty(&reaplist)) {
2383 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
2384 list_del_init(&dp->dl_recall_lru);
2385 destroy_unhashed_deleg(dp);
2386 }
2387 while (!list_empty(&clp->cl_revoked)) {
2388 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
2389 list_del_init(&dp->dl_recall_lru);
2390 nfs4_put_stid(&dp->dl_stid);
2391 }
2392 while (!list_empty(&clp->cl_openowners)) {
2393 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
2394 nfs4_get_stateowner(&oo->oo_owner);
2395 release_openowner(oo);
2396 }
2397 for (i = 0; i < OWNER_HASH_SIZE; i++) {
2398 struct nfs4_stateowner *so, *tmp;
2399
2400 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
2401 so_strhash) {
2402 /* Should be no openowners at this point */
2403 WARN_ON_ONCE(so->so_is_open_owner);
2404 remove_blocked_locks(lockowner(so));
2405 }
2406 }
2407 nfsd4_return_all_client_layouts(clp);
2408 nfsd4_shutdown_copy(clp);
2409 nfsd4_shutdown_callback(clp);
2410 if (clp->cl_cb_conn.cb_xprt)
2411 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
2412 atomic_add_unless(&nn->nfs4_client_count, -1, 0);
2413 nfsd4_dec_courtesy_client_count(nn, clp);
2414 free_client(clp);
2415 wake_up_all(&expiry_wq);
2416 }
2417
2418 static void
destroy_client(struct nfs4_client * clp)2419 destroy_client(struct nfs4_client *clp)
2420 {
2421 unhash_client(clp);
2422 __destroy_client(clp);
2423 }
2424
inc_reclaim_complete(struct nfs4_client * clp)2425 static void inc_reclaim_complete(struct nfs4_client *clp)
2426 {
2427 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2428
2429 if (!nn->track_reclaim_completes)
2430 return;
2431 if (!nfsd4_find_reclaim_client(clp->cl_name, nn))
2432 return;
2433 if (atomic_inc_return(&nn->nr_reclaim_complete) ==
2434 nn->reclaim_str_hashtbl_size) {
2435 printk(KERN_INFO "NFSD: all clients done reclaiming, ending NFSv4 grace period (net %x)\n",
2436 clp->net->ns.inum);
2437 nfsd4_end_grace(nn);
2438 }
2439 }
2440
expire_client(struct nfs4_client * clp)2441 static void expire_client(struct nfs4_client *clp)
2442 {
2443 unhash_client(clp);
2444 nfsd4_client_record_remove(clp);
2445 __destroy_client(clp);
2446 }
2447
copy_verf(struct nfs4_client * target,nfs4_verifier * source)2448 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
2449 {
2450 memcpy(target->cl_verifier.data, source->data,
2451 sizeof(target->cl_verifier.data));
2452 }
2453
copy_clid(struct nfs4_client * target,struct nfs4_client * source)2454 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
2455 {
2456 target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
2457 target->cl_clientid.cl_id = source->cl_clientid.cl_id;
2458 }
2459
copy_cred(struct svc_cred * target,struct svc_cred * source)2460 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
2461 {
2462 target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
2463 target->cr_raw_principal = kstrdup(source->cr_raw_principal,
2464 GFP_KERNEL);
2465 target->cr_targ_princ = kstrdup(source->cr_targ_princ, GFP_KERNEL);
2466 if ((source->cr_principal && !target->cr_principal) ||
2467 (source->cr_raw_principal && !target->cr_raw_principal) ||
2468 (source->cr_targ_princ && !target->cr_targ_princ))
2469 return -ENOMEM;
2470
2471 target->cr_flavor = source->cr_flavor;
2472 target->cr_uid = source->cr_uid;
2473 target->cr_gid = source->cr_gid;
2474 target->cr_group_info = source->cr_group_info;
2475 get_group_info(target->cr_group_info);
2476 target->cr_gss_mech = source->cr_gss_mech;
2477 if (source->cr_gss_mech)
2478 gss_mech_get(source->cr_gss_mech);
2479 return 0;
2480 }
2481
2482 static int
compare_blob(const struct xdr_netobj * o1,const struct xdr_netobj * o2)2483 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
2484 {
2485 if (o1->len < o2->len)
2486 return -1;
2487 if (o1->len > o2->len)
2488 return 1;
2489 return memcmp(o1->data, o2->data, o1->len);
2490 }
2491
2492 static int
same_verf(nfs4_verifier * v1,nfs4_verifier * v2)2493 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2494 {
2495 return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2496 }
2497
2498 static int
same_clid(clientid_t * cl1,clientid_t * cl2)2499 same_clid(clientid_t *cl1, clientid_t *cl2)
2500 {
2501 return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2502 }
2503
groups_equal(struct group_info * g1,struct group_info * g2)2504 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2505 {
2506 int i;
2507
2508 if (g1->ngroups != g2->ngroups)
2509 return false;
2510 for (i=0; i<g1->ngroups; i++)
2511 if (!gid_eq(g1->gid[i], g2->gid[i]))
2512 return false;
2513 return true;
2514 }
2515
2516 /*
2517 * RFC 3530 language requires clid_inuse be returned when the
2518 * "principal" associated with a requests differs from that previously
2519 * used. We use uid, gid's, and gss principal string as our best
2520 * approximation. We also don't want to allow non-gss use of a client
2521 * established using gss: in theory cr_principal should catch that
2522 * change, but in practice cr_principal can be null even in the gss case
2523 * since gssd doesn't always pass down a principal string.
2524 */
is_gss_cred(struct svc_cred * cr)2525 static bool is_gss_cred(struct svc_cred *cr)
2526 {
2527 /* Is cr_flavor one of the gss "pseudoflavors"?: */
2528 return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2529 }
2530
2531
2532 static bool
same_creds(struct svc_cred * cr1,struct svc_cred * cr2)2533 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2534 {
2535 if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2536 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2537 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2538 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2539 return false;
2540 /* XXX: check that cr_targ_princ fields match ? */
2541 if (cr1->cr_principal == cr2->cr_principal)
2542 return true;
2543 if (!cr1->cr_principal || !cr2->cr_principal)
2544 return false;
2545 return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2546 }
2547
svc_rqst_integrity_protected(struct svc_rqst * rqstp)2548 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2549 {
2550 struct svc_cred *cr = &rqstp->rq_cred;
2551 u32 service;
2552
2553 if (!cr->cr_gss_mech)
2554 return false;
2555 service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2556 return service == RPC_GSS_SVC_INTEGRITY ||
2557 service == RPC_GSS_SVC_PRIVACY;
2558 }
2559
nfsd4_mach_creds_match(struct nfs4_client * cl,struct svc_rqst * rqstp)2560 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2561 {
2562 struct svc_cred *cr = &rqstp->rq_cred;
2563
2564 if (!cl->cl_mach_cred)
2565 return true;
2566 if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2567 return false;
2568 if (!svc_rqst_integrity_protected(rqstp))
2569 return false;
2570 if (cl->cl_cred.cr_raw_principal)
2571 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2572 cr->cr_raw_principal);
2573 if (!cr->cr_principal)
2574 return false;
2575 return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2576 }
2577
gen_confirm(struct nfs4_client * clp,struct nfsd_net * nn)2578 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2579 {
2580 __be32 verf[2];
2581
2582 /*
2583 * This is opaque to client, so no need to byte-swap. Use
2584 * __force to keep sparse happy
2585 */
2586 verf[0] = (__force __be32)(u32)ktime_get_real_seconds();
2587 verf[1] = (__force __be32)nn->clverifier_counter++;
2588 memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2589 }
2590
gen_clid(struct nfs4_client * clp,struct nfsd_net * nn)2591 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2592 {
2593 clp->cl_clientid.cl_boot = (u32)nn->boot_time;
2594 clp->cl_clientid.cl_id = nn->clientid_counter++;
2595 gen_confirm(clp, nn);
2596 }
2597
2598 static struct nfs4_stid *
find_stateid_locked(struct nfs4_client * cl,stateid_t * t)2599 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2600 {
2601 struct nfs4_stid *ret;
2602
2603 ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2604 if (!ret || !ret->sc_type)
2605 return NULL;
2606 return ret;
2607 }
2608
2609 static struct nfs4_stid *
find_stateid_by_type(struct nfs4_client * cl,stateid_t * t,unsigned short typemask,unsigned short ok_states)2610 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t,
2611 unsigned short typemask, unsigned short ok_states)
2612 {
2613 struct nfs4_stid *s;
2614
2615 spin_lock(&cl->cl_lock);
2616 s = find_stateid_locked(cl, t);
2617 if (s != NULL) {
2618 if ((s->sc_status & ~ok_states) == 0 &&
2619 (typemask & s->sc_type))
2620 refcount_inc(&s->sc_count);
2621 else
2622 s = NULL;
2623 }
2624 spin_unlock(&cl->cl_lock);
2625 return s;
2626 }
2627
get_nfsdfs_clp(struct inode * inode)2628 static struct nfs4_client *get_nfsdfs_clp(struct inode *inode)
2629 {
2630 struct nfsdfs_client *nc;
2631 nc = get_nfsdfs_client(inode);
2632 if (!nc)
2633 return NULL;
2634 return container_of(nc, struct nfs4_client, cl_nfsdfs);
2635 }
2636
seq_quote_mem(struct seq_file * m,char * data,int len)2637 static void seq_quote_mem(struct seq_file *m, char *data, int len)
2638 {
2639 seq_puts(m, "\"");
2640 seq_escape_mem(m, data, len, ESCAPE_HEX | ESCAPE_NAP | ESCAPE_APPEND, "\"\\");
2641 seq_puts(m, "\"");
2642 }
2643
cb_state2str(int state)2644 static const char *cb_state2str(int state)
2645 {
2646 switch (state) {
2647 case NFSD4_CB_UP:
2648 return "UP";
2649 case NFSD4_CB_UNKNOWN:
2650 return "UNKNOWN";
2651 case NFSD4_CB_DOWN:
2652 return "DOWN";
2653 case NFSD4_CB_FAULT:
2654 return "FAULT";
2655 }
2656 return "UNDEFINED";
2657 }
2658
client_info_show(struct seq_file * m,void * v)2659 static int client_info_show(struct seq_file *m, void *v)
2660 {
2661 struct inode *inode = file_inode(m->file);
2662 struct nfs4_client *clp;
2663 u64 clid;
2664
2665 clp = get_nfsdfs_clp(inode);
2666 if (!clp)
2667 return -ENXIO;
2668 memcpy(&clid, &clp->cl_clientid, sizeof(clid));
2669 seq_printf(m, "clientid: 0x%llx\n", clid);
2670 seq_printf(m, "address: \"%pISpc\"\n", (struct sockaddr *)&clp->cl_addr);
2671
2672 if (clp->cl_state == NFSD4_COURTESY)
2673 seq_puts(m, "status: courtesy\n");
2674 else if (clp->cl_state == NFSD4_EXPIRABLE)
2675 seq_puts(m, "status: expirable\n");
2676 else if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
2677 seq_puts(m, "status: confirmed\n");
2678 else
2679 seq_puts(m, "status: unconfirmed\n");
2680 seq_printf(m, "seconds from last renew: %lld\n",
2681 ktime_get_boottime_seconds() - clp->cl_time);
2682 seq_puts(m, "name: ");
2683 seq_quote_mem(m, clp->cl_name.data, clp->cl_name.len);
2684 seq_printf(m, "\nminor version: %d\n", clp->cl_minorversion);
2685 if (clp->cl_nii_domain.data) {
2686 seq_puts(m, "Implementation domain: ");
2687 seq_quote_mem(m, clp->cl_nii_domain.data,
2688 clp->cl_nii_domain.len);
2689 seq_puts(m, "\nImplementation name: ");
2690 seq_quote_mem(m, clp->cl_nii_name.data, clp->cl_nii_name.len);
2691 seq_printf(m, "\nImplementation time: [%lld, %ld]\n",
2692 clp->cl_nii_time.tv_sec, clp->cl_nii_time.tv_nsec);
2693 }
2694 seq_printf(m, "callback state: %s\n", cb_state2str(clp->cl_cb_state));
2695 seq_printf(m, "callback address: %pISpc\n", &clp->cl_cb_conn.cb_addr);
2696 seq_printf(m, "admin-revoked states: %d\n",
2697 atomic_read(&clp->cl_admin_revoked));
2698 drop_client(clp);
2699
2700 return 0;
2701 }
2702
2703 DEFINE_SHOW_ATTRIBUTE(client_info);
2704
states_start(struct seq_file * s,loff_t * pos)2705 static void *states_start(struct seq_file *s, loff_t *pos)
2706 __acquires(&clp->cl_lock)
2707 {
2708 struct nfs4_client *clp = s->private;
2709 unsigned long id = *pos;
2710 void *ret;
2711
2712 spin_lock(&clp->cl_lock);
2713 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2714 *pos = id;
2715 return ret;
2716 }
2717
states_next(struct seq_file * s,void * v,loff_t * pos)2718 static void *states_next(struct seq_file *s, void *v, loff_t *pos)
2719 {
2720 struct nfs4_client *clp = s->private;
2721 unsigned long id = *pos;
2722 void *ret;
2723
2724 id = *pos;
2725 id++;
2726 ret = idr_get_next_ul(&clp->cl_stateids, &id);
2727 *pos = id;
2728 return ret;
2729 }
2730
states_stop(struct seq_file * s,void * v)2731 static void states_stop(struct seq_file *s, void *v)
2732 __releases(&clp->cl_lock)
2733 {
2734 struct nfs4_client *clp = s->private;
2735
2736 spin_unlock(&clp->cl_lock);
2737 }
2738
nfs4_show_fname(struct seq_file * s,struct nfsd_file * f)2739 static void nfs4_show_fname(struct seq_file *s, struct nfsd_file *f)
2740 {
2741 seq_printf(s, "filename: \"%pD2\"", f->nf_file);
2742 }
2743
nfs4_show_superblock(struct seq_file * s,struct nfsd_file * f)2744 static void nfs4_show_superblock(struct seq_file *s, struct nfsd_file *f)
2745 {
2746 struct inode *inode = file_inode(f->nf_file);
2747
2748 seq_printf(s, "superblock: \"%02x:%02x:%ld\"",
2749 MAJOR(inode->i_sb->s_dev),
2750 MINOR(inode->i_sb->s_dev),
2751 inode->i_ino);
2752 }
2753
nfs4_show_owner(struct seq_file * s,struct nfs4_stateowner * oo)2754 static void nfs4_show_owner(struct seq_file *s, struct nfs4_stateowner *oo)
2755 {
2756 seq_puts(s, "owner: ");
2757 seq_quote_mem(s, oo->so_owner.data, oo->so_owner.len);
2758 }
2759
nfs4_show_stateid(struct seq_file * s,stateid_t * stid)2760 static void nfs4_show_stateid(struct seq_file *s, stateid_t *stid)
2761 {
2762 seq_printf(s, "0x%.8x", stid->si_generation);
2763 seq_printf(s, "%12phN", &stid->si_opaque);
2764 }
2765
nfs4_show_open(struct seq_file * s,struct nfs4_stid * st)2766 static int nfs4_show_open(struct seq_file *s, struct nfs4_stid *st)
2767 {
2768 struct nfs4_ol_stateid *ols;
2769 struct nfs4_file *nf;
2770 struct nfsd_file *file;
2771 struct nfs4_stateowner *oo;
2772 unsigned int access, deny;
2773
2774 ols = openlockstateid(st);
2775 oo = ols->st_stateowner;
2776 nf = st->sc_file;
2777
2778 seq_puts(s, "- ");
2779 nfs4_show_stateid(s, &st->sc_stateid);
2780 seq_puts(s, ": { type: open, ");
2781
2782 access = bmap_to_share_mode(ols->st_access_bmap);
2783 deny = bmap_to_share_mode(ols->st_deny_bmap);
2784
2785 seq_printf(s, "access: %s%s, ",
2786 access & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2787 access & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2788 seq_printf(s, "deny: %s%s, ",
2789 deny & NFS4_SHARE_ACCESS_READ ? "r" : "-",
2790 deny & NFS4_SHARE_ACCESS_WRITE ? "w" : "-");
2791
2792 if (nf) {
2793 spin_lock(&nf->fi_lock);
2794 file = find_any_file_locked(nf);
2795 if (file) {
2796 nfs4_show_superblock(s, file);
2797 seq_puts(s, ", ");
2798 nfs4_show_fname(s, file);
2799 seq_puts(s, ", ");
2800 }
2801 spin_unlock(&nf->fi_lock);
2802 } else
2803 seq_puts(s, "closed, ");
2804 nfs4_show_owner(s, oo);
2805 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2806 seq_puts(s, ", admin-revoked");
2807 seq_puts(s, " }\n");
2808 return 0;
2809 }
2810
nfs4_show_lock(struct seq_file * s,struct nfs4_stid * st)2811 static int nfs4_show_lock(struct seq_file *s, struct nfs4_stid *st)
2812 {
2813 struct nfs4_ol_stateid *ols;
2814 struct nfs4_file *nf;
2815 struct nfsd_file *file;
2816 struct nfs4_stateowner *oo;
2817
2818 ols = openlockstateid(st);
2819 oo = ols->st_stateowner;
2820 nf = st->sc_file;
2821
2822 seq_puts(s, "- ");
2823 nfs4_show_stateid(s, &st->sc_stateid);
2824 seq_puts(s, ": { type: lock, ");
2825
2826 spin_lock(&nf->fi_lock);
2827 file = find_any_file_locked(nf);
2828 if (file) {
2829 /*
2830 * Note: a lock stateid isn't really the same thing as a lock,
2831 * it's the locking state held by one owner on a file, and there
2832 * may be multiple (or no) lock ranges associated with it.
2833 * (Same for the matter is true of open stateids.)
2834 */
2835
2836 nfs4_show_superblock(s, file);
2837 /* XXX: open stateid? */
2838 seq_puts(s, ", ");
2839 nfs4_show_fname(s, file);
2840 seq_puts(s, ", ");
2841 }
2842 nfs4_show_owner(s, oo);
2843 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2844 seq_puts(s, ", admin-revoked");
2845 seq_puts(s, " }\n");
2846 spin_unlock(&nf->fi_lock);
2847 return 0;
2848 }
2849
nfs4_show_deleg(struct seq_file * s,struct nfs4_stid * st)2850 static int nfs4_show_deleg(struct seq_file *s, struct nfs4_stid *st)
2851 {
2852 struct nfs4_delegation *ds;
2853 struct nfs4_file *nf;
2854 struct nfsd_file *file;
2855
2856 ds = delegstateid(st);
2857 nf = st->sc_file;
2858
2859 seq_puts(s, "- ");
2860 nfs4_show_stateid(s, &st->sc_stateid);
2861 seq_puts(s, ": { type: deleg, ");
2862
2863 seq_printf(s, "access: %s",
2864 ds->dl_type == NFS4_OPEN_DELEGATE_READ ? "r" : "w");
2865
2866 /* XXX: lease time, whether it's being recalled. */
2867
2868 spin_lock(&nf->fi_lock);
2869 file = nf->fi_deleg_file;
2870 if (file) {
2871 seq_puts(s, ", ");
2872 nfs4_show_superblock(s, file);
2873 seq_puts(s, ", ");
2874 nfs4_show_fname(s, file);
2875 }
2876 spin_unlock(&nf->fi_lock);
2877 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2878 seq_puts(s, ", admin-revoked");
2879 seq_puts(s, " }\n");
2880 return 0;
2881 }
2882
nfs4_show_layout(struct seq_file * s,struct nfs4_stid * st)2883 static int nfs4_show_layout(struct seq_file *s, struct nfs4_stid *st)
2884 {
2885 struct nfs4_layout_stateid *ls;
2886 struct nfsd_file *file;
2887
2888 ls = container_of(st, struct nfs4_layout_stateid, ls_stid);
2889
2890 seq_puts(s, "- ");
2891 nfs4_show_stateid(s, &st->sc_stateid);
2892 seq_puts(s, ": { type: layout");
2893
2894 /* XXX: What else would be useful? */
2895
2896 spin_lock(&ls->ls_stid.sc_file->fi_lock);
2897 file = ls->ls_file;
2898 if (file) {
2899 seq_puts(s, ", ");
2900 nfs4_show_superblock(s, file);
2901 seq_puts(s, ", ");
2902 nfs4_show_fname(s, file);
2903 }
2904 spin_unlock(&ls->ls_stid.sc_file->fi_lock);
2905 if (st->sc_status & SC_STATUS_ADMIN_REVOKED)
2906 seq_puts(s, ", admin-revoked");
2907 seq_puts(s, " }\n");
2908
2909 return 0;
2910 }
2911
states_show(struct seq_file * s,void * v)2912 static int states_show(struct seq_file *s, void *v)
2913 {
2914 struct nfs4_stid *st = v;
2915
2916 switch (st->sc_type) {
2917 case SC_TYPE_OPEN:
2918 return nfs4_show_open(s, st);
2919 case SC_TYPE_LOCK:
2920 return nfs4_show_lock(s, st);
2921 case SC_TYPE_DELEG:
2922 return nfs4_show_deleg(s, st);
2923 case SC_TYPE_LAYOUT:
2924 return nfs4_show_layout(s, st);
2925 default:
2926 return 0; /* XXX: or SEQ_SKIP? */
2927 }
2928 /* XXX: copy stateids? */
2929 }
2930
2931 static struct seq_operations states_seq_ops = {
2932 .start = states_start,
2933 .next = states_next,
2934 .stop = states_stop,
2935 .show = states_show
2936 };
2937
client_states_open(struct inode * inode,struct file * file)2938 static int client_states_open(struct inode *inode, struct file *file)
2939 {
2940 struct seq_file *s;
2941 struct nfs4_client *clp;
2942 int ret;
2943
2944 clp = get_nfsdfs_clp(inode);
2945 if (!clp)
2946 return -ENXIO;
2947
2948 ret = seq_open(file, &states_seq_ops);
2949 if (ret)
2950 return ret;
2951 s = file->private_data;
2952 s->private = clp;
2953 return 0;
2954 }
2955
client_opens_release(struct inode * inode,struct file * file)2956 static int client_opens_release(struct inode *inode, struct file *file)
2957 {
2958 struct seq_file *m = file->private_data;
2959 struct nfs4_client *clp = m->private;
2960
2961 /* XXX: alternatively, we could get/drop in seq start/stop */
2962 drop_client(clp);
2963 return seq_release(inode, file);
2964 }
2965
2966 static const struct file_operations client_states_fops = {
2967 .open = client_states_open,
2968 .read = seq_read,
2969 .llseek = seq_lseek,
2970 .release = client_opens_release,
2971 };
2972
2973 /*
2974 * Normally we refuse to destroy clients that are in use, but here the
2975 * administrator is telling us to just do it. We also want to wait
2976 * so the caller has a guarantee that the client's locks are gone by
2977 * the time the write returns:
2978 */
force_expire_client(struct nfs4_client * clp)2979 static void force_expire_client(struct nfs4_client *clp)
2980 {
2981 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2982 bool already_expired;
2983
2984 trace_nfsd_clid_admin_expired(&clp->cl_clientid);
2985
2986 spin_lock(&nn->client_lock);
2987 clp->cl_time = 0;
2988 spin_unlock(&nn->client_lock);
2989
2990 wait_event(expiry_wq, atomic_read(&clp->cl_rpc_users) == 0);
2991 spin_lock(&nn->client_lock);
2992 already_expired = list_empty(&clp->cl_lru);
2993 if (!already_expired)
2994 unhash_client_locked(clp);
2995 spin_unlock(&nn->client_lock);
2996
2997 if (!already_expired)
2998 expire_client(clp);
2999 else
3000 wait_event(expiry_wq, clp->cl_nfsd_dentry == NULL);
3001 }
3002
client_ctl_write(struct file * file,const char __user * buf,size_t size,loff_t * pos)3003 static ssize_t client_ctl_write(struct file *file, const char __user *buf,
3004 size_t size, loff_t *pos)
3005 {
3006 char *data;
3007 struct nfs4_client *clp;
3008
3009 data = simple_transaction_get(file, buf, size);
3010 if (IS_ERR(data))
3011 return PTR_ERR(data);
3012 if (size != 7 || 0 != memcmp(data, "expire\n", 7))
3013 return -EINVAL;
3014 clp = get_nfsdfs_clp(file_inode(file));
3015 if (!clp)
3016 return -ENXIO;
3017 force_expire_client(clp);
3018 drop_client(clp);
3019 return 7;
3020 }
3021
3022 static const struct file_operations client_ctl_fops = {
3023 .write = client_ctl_write,
3024 .release = simple_transaction_release,
3025 };
3026
3027 static const struct tree_descr client_files[] = {
3028 [0] = {"info", &client_info_fops, S_IRUSR},
3029 [1] = {"states", &client_states_fops, S_IRUSR},
3030 [2] = {"ctl", &client_ctl_fops, S_IWUSR},
3031 [3] = {""},
3032 };
3033
3034 static int
nfsd4_cb_recall_any_done(struct nfsd4_callback * cb,struct rpc_task * task)3035 nfsd4_cb_recall_any_done(struct nfsd4_callback *cb,
3036 struct rpc_task *task)
3037 {
3038 trace_nfsd_cb_recall_any_done(cb, task);
3039 switch (task->tk_status) {
3040 case -NFS4ERR_DELAY:
3041 rpc_delay(task, 2 * HZ);
3042 return 0;
3043 default:
3044 return 1;
3045 }
3046 }
3047
3048 static void
nfsd4_cb_recall_any_release(struct nfsd4_callback * cb)3049 nfsd4_cb_recall_any_release(struct nfsd4_callback *cb)
3050 {
3051 struct nfs4_client *clp = cb->cb_clp;
3052
3053 clear_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
3054 drop_client(clp);
3055 }
3056
3057 static int
nfsd4_cb_getattr_done(struct nfsd4_callback * cb,struct rpc_task * task)3058 nfsd4_cb_getattr_done(struct nfsd4_callback *cb, struct rpc_task *task)
3059 {
3060 struct nfs4_cb_fattr *ncf =
3061 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3062
3063 ncf->ncf_cb_status = task->tk_status;
3064 switch (task->tk_status) {
3065 case -NFS4ERR_DELAY:
3066 rpc_delay(task, 2 * HZ);
3067 return 0;
3068 default:
3069 return 1;
3070 }
3071 }
3072
3073 static void
nfsd4_cb_getattr_release(struct nfsd4_callback * cb)3074 nfsd4_cb_getattr_release(struct nfsd4_callback *cb)
3075 {
3076 struct nfs4_cb_fattr *ncf =
3077 container_of(cb, struct nfs4_cb_fattr, ncf_getattr);
3078 struct nfs4_delegation *dp =
3079 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3080
3081 clear_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags);
3082 wake_up_bit(&ncf->ncf_cb_flags, CB_GETATTR_BUSY);
3083 nfs4_put_stid(&dp->dl_stid);
3084 }
3085
3086 static const struct nfsd4_callback_ops nfsd4_cb_recall_any_ops = {
3087 .done = nfsd4_cb_recall_any_done,
3088 .release = nfsd4_cb_recall_any_release,
3089 };
3090
3091 static const struct nfsd4_callback_ops nfsd4_cb_getattr_ops = {
3092 .done = nfsd4_cb_getattr_done,
3093 .release = nfsd4_cb_getattr_release,
3094 };
3095
nfs4_cb_getattr(struct nfs4_cb_fattr * ncf)3096 static void nfs4_cb_getattr(struct nfs4_cb_fattr *ncf)
3097 {
3098 struct nfs4_delegation *dp =
3099 container_of(ncf, struct nfs4_delegation, dl_cb_fattr);
3100
3101 if (test_and_set_bit(CB_GETATTR_BUSY, &ncf->ncf_cb_flags))
3102 return;
3103 /* set to proper status when nfsd4_cb_getattr_done runs */
3104 ncf->ncf_cb_status = NFS4ERR_IO;
3105
3106 refcount_inc(&dp->dl_stid.sc_count);
3107 nfsd4_run_cb(&ncf->ncf_getattr);
3108 }
3109
create_client(struct xdr_netobj name,struct svc_rqst * rqstp,nfs4_verifier * verf)3110 static struct nfs4_client *create_client(struct xdr_netobj name,
3111 struct svc_rqst *rqstp, nfs4_verifier *verf)
3112 {
3113 struct nfs4_client *clp;
3114 struct sockaddr *sa = svc_addr(rqstp);
3115 int ret;
3116 struct net *net = SVC_NET(rqstp);
3117 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3118 struct dentry *dentries[ARRAY_SIZE(client_files)];
3119
3120 clp = alloc_client(name, nn);
3121 if (clp == NULL)
3122 return NULL;
3123
3124 ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
3125 if (ret) {
3126 free_client(clp);
3127 return NULL;
3128 }
3129 gen_clid(clp, nn);
3130 kref_init(&clp->cl_nfsdfs.cl_ref);
3131 nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
3132 clp->cl_time = ktime_get_boottime_seconds();
3133 clear_bit(0, &clp->cl_cb_slot_busy);
3134 copy_verf(clp, verf);
3135 memcpy(&clp->cl_addr, sa, sizeof(struct sockaddr_storage));
3136 clp->cl_cb_session = NULL;
3137 clp->net = net;
3138 clp->cl_nfsd_dentry = nfsd_client_mkdir(
3139 nn, &clp->cl_nfsdfs,
3140 clp->cl_clientid.cl_id - nn->clientid_base,
3141 client_files, dentries);
3142 clp->cl_nfsd_info_dentry = dentries[0];
3143 if (!clp->cl_nfsd_dentry) {
3144 free_client(clp);
3145 return NULL;
3146 }
3147 clp->cl_ra = kzalloc(sizeof(*clp->cl_ra), GFP_KERNEL);
3148 if (!clp->cl_ra) {
3149 free_client(clp);
3150 return NULL;
3151 }
3152 clp->cl_ra_time = 0;
3153 nfsd4_init_cb(&clp->cl_ra->ra_cb, clp, &nfsd4_cb_recall_any_ops,
3154 NFSPROC4_CLNT_CB_RECALL_ANY);
3155 return clp;
3156 }
3157
3158 static void
add_clp_to_name_tree(struct nfs4_client * new_clp,struct rb_root * root)3159 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
3160 {
3161 struct rb_node **new = &(root->rb_node), *parent = NULL;
3162 struct nfs4_client *clp;
3163
3164 while (*new) {
3165 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
3166 parent = *new;
3167
3168 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
3169 new = &((*new)->rb_left);
3170 else
3171 new = &((*new)->rb_right);
3172 }
3173
3174 rb_link_node(&new_clp->cl_namenode, parent, new);
3175 rb_insert_color(&new_clp->cl_namenode, root);
3176 }
3177
3178 static struct nfs4_client *
find_clp_in_name_tree(struct xdr_netobj * name,struct rb_root * root)3179 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
3180 {
3181 int cmp;
3182 struct rb_node *node = root->rb_node;
3183 struct nfs4_client *clp;
3184
3185 while (node) {
3186 clp = rb_entry(node, struct nfs4_client, cl_namenode);
3187 cmp = compare_blob(&clp->cl_name, name);
3188 if (cmp > 0)
3189 node = node->rb_left;
3190 else if (cmp < 0)
3191 node = node->rb_right;
3192 else
3193 return clp;
3194 }
3195 return NULL;
3196 }
3197
3198 static void
add_to_unconfirmed(struct nfs4_client * clp)3199 add_to_unconfirmed(struct nfs4_client *clp)
3200 {
3201 unsigned int idhashval;
3202 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3203
3204 lockdep_assert_held(&nn->client_lock);
3205
3206 clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3207 add_clp_to_name_tree(clp, &nn->unconf_name_tree);
3208 idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3209 list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
3210 renew_client_locked(clp);
3211 }
3212
3213 static void
move_to_confirmed(struct nfs4_client * clp)3214 move_to_confirmed(struct nfs4_client *clp)
3215 {
3216 unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
3217 struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
3218
3219 lockdep_assert_held(&nn->client_lock);
3220
3221 list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
3222 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
3223 add_clp_to_name_tree(clp, &nn->conf_name_tree);
3224 set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
3225 trace_nfsd_clid_confirmed(&clp->cl_clientid);
3226 renew_client_locked(clp);
3227 }
3228
3229 static struct nfs4_client *
find_client_in_id_table(struct list_head * tbl,clientid_t * clid,bool sessions)3230 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
3231 {
3232 struct nfs4_client *clp;
3233 unsigned int idhashval = clientid_hashval(clid->cl_id);
3234
3235 list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
3236 if (same_clid(&clp->cl_clientid, clid)) {
3237 if ((bool)clp->cl_minorversion != sessions)
3238 return NULL;
3239 renew_client_locked(clp);
3240 return clp;
3241 }
3242 }
3243 return NULL;
3244 }
3245
3246 static struct nfs4_client *
find_confirmed_client(clientid_t * clid,bool sessions,struct nfsd_net * nn)3247 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3248 {
3249 struct list_head *tbl = nn->conf_id_hashtbl;
3250
3251 lockdep_assert_held(&nn->client_lock);
3252 return find_client_in_id_table(tbl, clid, sessions);
3253 }
3254
3255 static struct nfs4_client *
find_unconfirmed_client(clientid_t * clid,bool sessions,struct nfsd_net * nn)3256 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
3257 {
3258 struct list_head *tbl = nn->unconf_id_hashtbl;
3259
3260 lockdep_assert_held(&nn->client_lock);
3261 return find_client_in_id_table(tbl, clid, sessions);
3262 }
3263
clp_used_exchangeid(struct nfs4_client * clp)3264 static bool clp_used_exchangeid(struct nfs4_client *clp)
3265 {
3266 return clp->cl_exchange_flags != 0;
3267 }
3268
3269 static struct nfs4_client *
find_confirmed_client_by_name(struct xdr_netobj * name,struct nfsd_net * nn)3270 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3271 {
3272 lockdep_assert_held(&nn->client_lock);
3273 return find_clp_in_name_tree(name, &nn->conf_name_tree);
3274 }
3275
3276 static struct nfs4_client *
find_unconfirmed_client_by_name(struct xdr_netobj * name,struct nfsd_net * nn)3277 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
3278 {
3279 lockdep_assert_held(&nn->client_lock);
3280 return find_clp_in_name_tree(name, &nn->unconf_name_tree);
3281 }
3282
3283 static void
gen_callback(struct nfs4_client * clp,struct nfsd4_setclientid * se,struct svc_rqst * rqstp)3284 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
3285 {
3286 struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
3287 struct sockaddr *sa = svc_addr(rqstp);
3288 u32 scopeid = rpc_get_scope_id(sa);
3289 unsigned short expected_family;
3290
3291 /* Currently, we only support tcp and tcp6 for the callback channel */
3292 if (se->se_callback_netid_len == 3 &&
3293 !memcmp(se->se_callback_netid_val, "tcp", 3))
3294 expected_family = AF_INET;
3295 else if (se->se_callback_netid_len == 4 &&
3296 !memcmp(se->se_callback_netid_val, "tcp6", 4))
3297 expected_family = AF_INET6;
3298 else
3299 goto out_err;
3300
3301 conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
3302 se->se_callback_addr_len,
3303 (struct sockaddr *)&conn->cb_addr,
3304 sizeof(conn->cb_addr));
3305
3306 if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
3307 goto out_err;
3308
3309 if (conn->cb_addr.ss_family == AF_INET6)
3310 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
3311
3312 conn->cb_prog = se->se_callback_prog;
3313 conn->cb_ident = se->se_callback_ident;
3314 memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
3315 trace_nfsd_cb_args(clp, conn);
3316 return;
3317 out_err:
3318 conn->cb_addr.ss_family = AF_UNSPEC;
3319 conn->cb_addrlen = 0;
3320 trace_nfsd_cb_nodelegs(clp);
3321 return;
3322 }
3323
3324 /*
3325 * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
3326 */
3327 static void
nfsd4_store_cache_entry(struct nfsd4_compoundres * resp)3328 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
3329 {
3330 struct xdr_buf *buf = resp->xdr->buf;
3331 struct nfsd4_slot *slot = resp->cstate.slot;
3332 unsigned int base;
3333
3334 dprintk("--> %s slot %p\n", __func__, slot);
3335
3336 slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
3337 slot->sl_opcnt = resp->opcnt;
3338 slot->sl_status = resp->cstate.status;
3339 free_svc_cred(&slot->sl_cred);
3340 copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
3341
3342 if (!nfsd4_cache_this(resp)) {
3343 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
3344 return;
3345 }
3346 slot->sl_flags |= NFSD4_SLOT_CACHED;
3347
3348 base = resp->cstate.data_offset;
3349 slot->sl_datalen = buf->len - base;
3350 if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
3351 WARN(1, "%s: sessions DRC could not cache compound\n",
3352 __func__);
3353 return;
3354 }
3355
3356 /*
3357 * Encode the replay sequence operation from the slot values.
3358 * If cachethis is FALSE encode the uncached rep error on the next
3359 * operation which sets resp->p and increments resp->opcnt for
3360 * nfs4svc_encode_compoundres.
3361 *
3362 */
3363 static __be32
nfsd4_enc_sequence_replay(struct nfsd4_compoundargs * args,struct nfsd4_compoundres * resp)3364 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
3365 struct nfsd4_compoundres *resp)
3366 {
3367 struct nfsd4_op *op;
3368 struct nfsd4_slot *slot = resp->cstate.slot;
3369
3370 /* Encode the replayed sequence operation */
3371 op = &args->ops[resp->opcnt - 1];
3372 nfsd4_encode_operation(resp, op);
3373
3374 if (slot->sl_flags & NFSD4_SLOT_CACHED)
3375 return op->status;
3376 if (args->opcnt == 1) {
3377 /*
3378 * The original operation wasn't a solo sequence--we
3379 * always cache those--so this retry must not match the
3380 * original:
3381 */
3382 op->status = nfserr_seq_false_retry;
3383 } else {
3384 op = &args->ops[resp->opcnt++];
3385 op->status = nfserr_retry_uncached_rep;
3386 nfsd4_encode_operation(resp, op);
3387 }
3388 return op->status;
3389 }
3390
3391 /*
3392 * The sequence operation is not cached because we can use the slot and
3393 * session values.
3394 */
3395 static __be32
nfsd4_replay_cache_entry(struct nfsd4_compoundres * resp,struct nfsd4_sequence * seq)3396 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
3397 struct nfsd4_sequence *seq)
3398 {
3399 struct nfsd4_slot *slot = resp->cstate.slot;
3400 struct xdr_stream *xdr = resp->xdr;
3401 __be32 *p;
3402 __be32 status;
3403
3404 dprintk("--> %s slot %p\n", __func__, slot);
3405
3406 status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
3407 if (status)
3408 return status;
3409
3410 p = xdr_reserve_space(xdr, slot->sl_datalen);
3411 if (!p) {
3412 WARN_ON_ONCE(1);
3413 return nfserr_serverfault;
3414 }
3415 xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
3416 xdr_commit_encode(xdr);
3417
3418 resp->opcnt = slot->sl_opcnt;
3419 return slot->sl_status;
3420 }
3421
3422 /*
3423 * Set the exchange_id flags returned by the server.
3424 */
3425 static void
nfsd4_set_ex_flags(struct nfs4_client * new,struct nfsd4_exchange_id * clid)3426 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
3427 {
3428 #ifdef CONFIG_NFSD_PNFS
3429 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
3430 #else
3431 new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
3432 #endif
3433
3434 /* Referrals are supported, Migration is not. */
3435 new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
3436
3437 /* set the wire flags to return to client. */
3438 clid->flags = new->cl_exchange_flags;
3439 }
3440
client_has_openowners(struct nfs4_client * clp)3441 static bool client_has_openowners(struct nfs4_client *clp)
3442 {
3443 struct nfs4_openowner *oo;
3444
3445 list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
3446 if (!list_empty(&oo->oo_owner.so_stateids))
3447 return true;
3448 }
3449 return false;
3450 }
3451
client_has_state(struct nfs4_client * clp)3452 static bool client_has_state(struct nfs4_client *clp)
3453 {
3454 return client_has_openowners(clp)
3455 #ifdef CONFIG_NFSD_PNFS
3456 || !list_empty(&clp->cl_lo_states)
3457 #endif
3458 || !list_empty(&clp->cl_delegations)
3459 || !list_empty(&clp->cl_sessions)
3460 || !list_empty(&clp->async_copies);
3461 }
3462
copy_impl_id(struct nfs4_client * clp,struct nfsd4_exchange_id * exid)3463 static __be32 copy_impl_id(struct nfs4_client *clp,
3464 struct nfsd4_exchange_id *exid)
3465 {
3466 if (!exid->nii_domain.data)
3467 return 0;
3468 xdr_netobj_dup(&clp->cl_nii_domain, &exid->nii_domain, GFP_KERNEL);
3469 if (!clp->cl_nii_domain.data)
3470 return nfserr_jukebox;
3471 xdr_netobj_dup(&clp->cl_nii_name, &exid->nii_name, GFP_KERNEL);
3472 if (!clp->cl_nii_name.data)
3473 return nfserr_jukebox;
3474 clp->cl_nii_time = exid->nii_time;
3475 return 0;
3476 }
3477
3478 __be32
nfsd4_exchange_id(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)3479 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3480 union nfsd4_op_u *u)
3481 {
3482 struct nfsd4_exchange_id *exid = &u->exchange_id;
3483 struct nfs4_client *conf, *new;
3484 struct nfs4_client *unconf = NULL;
3485 __be32 status;
3486 char addr_str[INET6_ADDRSTRLEN];
3487 nfs4_verifier verf = exid->verifier;
3488 struct sockaddr *sa = svc_addr(rqstp);
3489 bool update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
3490 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3491
3492 rpc_ntop(sa, addr_str, sizeof(addr_str));
3493 dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
3494 "ip_addr=%s flags %x, spa_how %u\n",
3495 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
3496 addr_str, exid->flags, exid->spa_how);
3497
3498 if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
3499 return nfserr_inval;
3500
3501 new = create_client(exid->clname, rqstp, &verf);
3502 if (new == NULL)
3503 return nfserr_jukebox;
3504 status = copy_impl_id(new, exid);
3505 if (status)
3506 goto out_nolock;
3507
3508 switch (exid->spa_how) {
3509 case SP4_MACH_CRED:
3510 exid->spo_must_enforce[0] = 0;
3511 exid->spo_must_enforce[1] = (
3512 1 << (OP_BIND_CONN_TO_SESSION - 32) |
3513 1 << (OP_EXCHANGE_ID - 32) |
3514 1 << (OP_CREATE_SESSION - 32) |
3515 1 << (OP_DESTROY_SESSION - 32) |
3516 1 << (OP_DESTROY_CLIENTID - 32));
3517
3518 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
3519 1 << (OP_OPEN_DOWNGRADE) |
3520 1 << (OP_LOCKU) |
3521 1 << (OP_DELEGRETURN));
3522
3523 exid->spo_must_allow[1] &= (
3524 1 << (OP_TEST_STATEID - 32) |
3525 1 << (OP_FREE_STATEID - 32));
3526 if (!svc_rqst_integrity_protected(rqstp)) {
3527 status = nfserr_inval;
3528 goto out_nolock;
3529 }
3530 /*
3531 * Sometimes userspace doesn't give us a principal.
3532 * Which is a bug, really. Anyway, we can't enforce
3533 * MACH_CRED in that case, better to give up now:
3534 */
3535 if (!new->cl_cred.cr_principal &&
3536 !new->cl_cred.cr_raw_principal) {
3537 status = nfserr_serverfault;
3538 goto out_nolock;
3539 }
3540 new->cl_mach_cred = true;
3541 break;
3542 case SP4_NONE:
3543 break;
3544 default: /* checked by xdr code */
3545 WARN_ON_ONCE(1);
3546 fallthrough;
3547 case SP4_SSV:
3548 status = nfserr_encr_alg_unsupp;
3549 goto out_nolock;
3550 }
3551
3552 /* Cases below refer to rfc 5661 section 18.35.4: */
3553 spin_lock(&nn->client_lock);
3554 conf = find_confirmed_client_by_name(&exid->clname, nn);
3555 if (conf) {
3556 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
3557 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
3558
3559 if (update) {
3560 if (!clp_used_exchangeid(conf)) { /* buggy client */
3561 status = nfserr_inval;
3562 goto out;
3563 }
3564 if (!nfsd4_mach_creds_match(conf, rqstp)) {
3565 status = nfserr_wrong_cred;
3566 goto out;
3567 }
3568 if (!creds_match) { /* case 9 */
3569 status = nfserr_perm;
3570 goto out;
3571 }
3572 if (!verfs_match) { /* case 8 */
3573 status = nfserr_not_same;
3574 goto out;
3575 }
3576 /* case 6 */
3577 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
3578 trace_nfsd_clid_confirmed_r(conf);
3579 goto out_copy;
3580 }
3581 if (!creds_match) { /* case 3 */
3582 if (client_has_state(conf)) {
3583 status = nfserr_clid_inuse;
3584 trace_nfsd_clid_cred_mismatch(conf, rqstp);
3585 goto out;
3586 }
3587 goto out_new;
3588 }
3589 if (verfs_match) { /* case 2 */
3590 conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
3591 trace_nfsd_clid_confirmed_r(conf);
3592 goto out_copy;
3593 }
3594 /* case 5, client reboot */
3595 trace_nfsd_clid_verf_mismatch(conf, rqstp, &verf);
3596 conf = NULL;
3597 goto out_new;
3598 }
3599
3600 if (update) { /* case 7 */
3601 status = nfserr_noent;
3602 goto out;
3603 }
3604
3605 unconf = find_unconfirmed_client_by_name(&exid->clname, nn);
3606 if (unconf) /* case 4, possible retry or client restart */
3607 unhash_client_locked(unconf);
3608
3609 /* case 1, new owner ID */
3610 trace_nfsd_clid_fresh(new);
3611
3612 out_new:
3613 if (conf) {
3614 status = mark_client_expired_locked(conf);
3615 if (status)
3616 goto out;
3617 trace_nfsd_clid_replaced(&conf->cl_clientid);
3618 }
3619 new->cl_minorversion = cstate->minorversion;
3620 new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
3621 new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
3622
3623 /* Contrived initial CREATE_SESSION response */
3624 new->cl_cs_slot.sl_status = nfserr_seq_misordered;
3625
3626 add_to_unconfirmed(new);
3627 swap(new, conf);
3628 out_copy:
3629 exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
3630 exid->clientid.cl_id = conf->cl_clientid.cl_id;
3631
3632 exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
3633 nfsd4_set_ex_flags(conf, exid);
3634
3635 dprintk("nfsd4_exchange_id seqid %d flags %x\n",
3636 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
3637 status = nfs_ok;
3638
3639 out:
3640 spin_unlock(&nn->client_lock);
3641 out_nolock:
3642 if (new)
3643 expire_client(new);
3644 if (unconf) {
3645 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
3646 expire_client(unconf);
3647 }
3648 return status;
3649 }
3650
check_slot_seqid(u32 seqid,u32 slot_seqid,bool slot_inuse)3651 static __be32 check_slot_seqid(u32 seqid, u32 slot_seqid, bool slot_inuse)
3652 {
3653 /* The slot is in use, and no response has been sent. */
3654 if (slot_inuse) {
3655 if (seqid == slot_seqid)
3656 return nfserr_jukebox;
3657 else
3658 return nfserr_seq_misordered;
3659 }
3660 /* Note unsigned 32-bit arithmetic handles wraparound: */
3661 if (likely(seqid == slot_seqid + 1))
3662 return nfs_ok;
3663 if (seqid == slot_seqid)
3664 return nfserr_replay_cache;
3665 return nfserr_seq_misordered;
3666 }
3667
3668 /*
3669 * Cache the create session result into the create session single DRC
3670 * slot cache by saving the xdr structure. sl_seqid has been set.
3671 * Do this for solo or embedded create session operations.
3672 */
3673 static void
nfsd4_cache_create_session(struct nfsd4_create_session * cr_ses,struct nfsd4_clid_slot * slot,__be32 nfserr)3674 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
3675 struct nfsd4_clid_slot *slot, __be32 nfserr)
3676 {
3677 slot->sl_status = nfserr;
3678 memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
3679 }
3680
3681 static __be32
nfsd4_replay_create_session(struct nfsd4_create_session * cr_ses,struct nfsd4_clid_slot * slot)3682 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
3683 struct nfsd4_clid_slot *slot)
3684 {
3685 memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
3686 return slot->sl_status;
3687 }
3688
3689 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
3690 2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
3691 1 + /* MIN tag is length with zero, only length */ \
3692 3 + /* version, opcount, opcode */ \
3693 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3694 /* seqid, slotID, slotID, cache */ \
3695 4 ) * sizeof(__be32))
3696
3697 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
3698 2 + /* verifier: AUTH_NULL, length 0 */\
3699 1 + /* status */ \
3700 1 + /* MIN tag is length with zero, only length */ \
3701 3 + /* opcount, opcode, opstatus*/ \
3702 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
3703 /* seqid, slotID, slotID, slotID, status */ \
3704 5 ) * sizeof(__be32))
3705
check_forechannel_attrs(struct nfsd4_channel_attrs * ca,struct nfsd_net * nn)3706 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
3707 {
3708 u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
3709
3710 if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
3711 return nfserr_toosmall;
3712 if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
3713 return nfserr_toosmall;
3714 ca->headerpadsz = 0;
3715 ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
3716 ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
3717 ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
3718 ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
3719 NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
3720 ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
3721 /*
3722 * Note decreasing slot size below client's request may make it
3723 * difficult for client to function correctly, whereas
3724 * decreasing the number of slots will (just?) affect
3725 * performance. When short on memory we therefore prefer to
3726 * decrease number of slots instead of their size. Clients that
3727 * request larger slots than they need will get poor results:
3728 * Note that we always allow at least one slot, because our
3729 * accounting is soft and provides no guarantees either way.
3730 */
3731 ca->maxreqs = nfsd4_get_drc_mem(ca, nn);
3732
3733 return nfs_ok;
3734 }
3735
3736 /*
3737 * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
3738 * These are based on similar macros in linux/sunrpc/msg_prot.h .
3739 */
3740 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
3741 (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
3742
3743 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
3744 (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
3745
3746 #define NFSD_CB_MAX_REQ_SZ ((NFS4_enc_cb_recall_sz + \
3747 RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
3748 #define NFSD_CB_MAX_RESP_SZ ((NFS4_dec_cb_recall_sz + \
3749 RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
3750 sizeof(__be32))
3751
check_backchannel_attrs(struct nfsd4_channel_attrs * ca)3752 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
3753 {
3754 ca->headerpadsz = 0;
3755
3756 if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
3757 return nfserr_toosmall;
3758 if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
3759 return nfserr_toosmall;
3760 ca->maxresp_cached = 0;
3761 if (ca->maxops < 2)
3762 return nfserr_toosmall;
3763
3764 return nfs_ok;
3765 }
3766
nfsd4_check_cb_sec(struct nfsd4_cb_sec * cbs)3767 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
3768 {
3769 switch (cbs->flavor) {
3770 case RPC_AUTH_NULL:
3771 case RPC_AUTH_UNIX:
3772 return nfs_ok;
3773 default:
3774 /*
3775 * GSS case: the spec doesn't allow us to return this
3776 * error. But it also doesn't allow us not to support
3777 * GSS.
3778 * I'd rather this fail hard than return some error the
3779 * client might think it can already handle:
3780 */
3781 return nfserr_encr_alg_unsupp;
3782 }
3783 }
3784
3785 __be32
nfsd4_create_session(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)3786 nfsd4_create_session(struct svc_rqst *rqstp,
3787 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3788 {
3789 struct nfsd4_create_session *cr_ses = &u->create_session;
3790 struct sockaddr *sa = svc_addr(rqstp);
3791 struct nfs4_client *conf, *unconf;
3792 struct nfsd4_clid_slot *cs_slot;
3793 struct nfs4_client *old = NULL;
3794 struct nfsd4_session *new;
3795 struct nfsd4_conn *conn;
3796 __be32 status = 0;
3797 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3798
3799 if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
3800 return nfserr_inval;
3801 status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
3802 if (status)
3803 return status;
3804 status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
3805 if (status)
3806 return status;
3807 status = check_backchannel_attrs(&cr_ses->back_channel);
3808 if (status)
3809 goto out_release_drc_mem;
3810 status = nfserr_jukebox;
3811 new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
3812 if (!new)
3813 goto out_release_drc_mem;
3814 conn = alloc_conn_from_crses(rqstp, cr_ses);
3815 if (!conn)
3816 goto out_free_session;
3817
3818 spin_lock(&nn->client_lock);
3819
3820 /* RFC 8881 Section 18.36.4 Phase 1: Client record look-up. */
3821 unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
3822 conf = find_confirmed_client(&cr_ses->clientid, true, nn);
3823 if (!conf && !unconf) {
3824 status = nfserr_stale_clientid;
3825 goto out_free_conn;
3826 }
3827
3828 /* RFC 8881 Section 18.36.4 Phase 2: Sequence ID processing. */
3829 if (conf) {
3830 cs_slot = &conf->cl_cs_slot;
3831 trace_nfsd_slot_seqid_conf(conf, cr_ses);
3832 } else {
3833 cs_slot = &unconf->cl_cs_slot;
3834 trace_nfsd_slot_seqid_unconf(unconf, cr_ses);
3835 }
3836 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
3837 switch (status) {
3838 case nfs_ok:
3839 cs_slot->sl_seqid++;
3840 cr_ses->seqid = cs_slot->sl_seqid;
3841 break;
3842 case nfserr_replay_cache:
3843 status = nfsd4_replay_create_session(cr_ses, cs_slot);
3844 fallthrough;
3845 case nfserr_jukebox:
3846 /* The server MUST NOT cache NFS4ERR_DELAY */
3847 goto out_free_conn;
3848 default:
3849 goto out_cache_error;
3850 }
3851
3852 /* RFC 8881 Section 18.36.4 Phase 3: Client ID confirmation. */
3853 if (conf) {
3854 status = nfserr_wrong_cred;
3855 if (!nfsd4_mach_creds_match(conf, rqstp))
3856 goto out_cache_error;
3857 } else {
3858 status = nfserr_clid_inuse;
3859 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
3860 !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
3861 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
3862 goto out_cache_error;
3863 }
3864 status = nfserr_wrong_cred;
3865 if (!nfsd4_mach_creds_match(unconf, rqstp))
3866 goto out_cache_error;
3867 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3868 if (old) {
3869 status = mark_client_expired_locked(old);
3870 if (status)
3871 goto out_expired_error;
3872 trace_nfsd_clid_replaced(&old->cl_clientid);
3873 }
3874 move_to_confirmed(unconf);
3875 conf = unconf;
3876 }
3877
3878 /* RFC 8881 Section 18.36.4 Phase 4: Session creation. */
3879 status = nfs_ok;
3880 /* Persistent sessions are not supported */
3881 cr_ses->flags &= ~SESSION4_PERSIST;
3882 /* Upshifting from TCP to RDMA is not supported */
3883 cr_ses->flags &= ~SESSION4_RDMA;
3884
3885 init_session(rqstp, new, conf, cr_ses);
3886 nfsd4_get_session_locked(new);
3887
3888 memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
3889 NFS4_MAX_SESSIONID_LEN);
3890
3891 /* cache solo and embedded create sessions under the client_lock */
3892 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3893 spin_unlock(&nn->client_lock);
3894 if (conf == unconf)
3895 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
3896 /* init connection and backchannel */
3897 nfsd4_init_conn(rqstp, conn, new);
3898 nfsd4_put_session(new);
3899 if (old)
3900 expire_client(old);
3901 return status;
3902
3903 out_expired_error:
3904 old = NULL;
3905 /*
3906 * Revert the slot seq_nr change so the server will process
3907 * the client's resend instead of returning a cached response.
3908 */
3909 if (status == nfserr_jukebox) {
3910 cs_slot->sl_seqid--;
3911 cr_ses->seqid = cs_slot->sl_seqid;
3912 goto out_free_conn;
3913 }
3914 out_cache_error:
3915 nfsd4_cache_create_session(cr_ses, cs_slot, status);
3916 out_free_conn:
3917 spin_unlock(&nn->client_lock);
3918 free_conn(conn);
3919 if (old)
3920 expire_client(old);
3921 out_free_session:
3922 __free_session(new);
3923 out_release_drc_mem:
3924 nfsd4_put_drc_mem(&cr_ses->fore_channel);
3925 return status;
3926 }
3927
nfsd4_map_bcts_dir(u32 * dir)3928 static __be32 nfsd4_map_bcts_dir(u32 *dir)
3929 {
3930 switch (*dir) {
3931 case NFS4_CDFC4_FORE:
3932 case NFS4_CDFC4_BACK:
3933 return nfs_ok;
3934 case NFS4_CDFC4_FORE_OR_BOTH:
3935 case NFS4_CDFC4_BACK_OR_BOTH:
3936 *dir = NFS4_CDFC4_BOTH;
3937 return nfs_ok;
3938 }
3939 return nfserr_inval;
3940 }
3941
nfsd4_backchannel_ctl(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)3942 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
3943 struct nfsd4_compound_state *cstate,
3944 union nfsd4_op_u *u)
3945 {
3946 struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
3947 struct nfsd4_session *session = cstate->session;
3948 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3949 __be32 status;
3950
3951 status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
3952 if (status)
3953 return status;
3954 spin_lock(&nn->client_lock);
3955 session->se_cb_prog = bc->bc_cb_program;
3956 session->se_cb_sec = bc->bc_cb_sec;
3957 spin_unlock(&nn->client_lock);
3958
3959 nfsd4_probe_callback(session->se_client);
3960
3961 return nfs_ok;
3962 }
3963
__nfsd4_find_conn(struct svc_xprt * xpt,struct nfsd4_session * s)3964 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3965 {
3966 struct nfsd4_conn *c;
3967
3968 list_for_each_entry(c, &s->se_conns, cn_persession) {
3969 if (c->cn_xprt == xpt) {
3970 return c;
3971 }
3972 }
3973 return NULL;
3974 }
3975
nfsd4_match_existing_connection(struct svc_rqst * rqst,struct nfsd4_session * session,u32 req,struct nfsd4_conn ** conn)3976 static __be32 nfsd4_match_existing_connection(struct svc_rqst *rqst,
3977 struct nfsd4_session *session, u32 req, struct nfsd4_conn **conn)
3978 {
3979 struct nfs4_client *clp = session->se_client;
3980 struct svc_xprt *xpt = rqst->rq_xprt;
3981 struct nfsd4_conn *c;
3982 __be32 status;
3983
3984 /* Following the last paragraph of RFC 5661 Section 18.34.3: */
3985 spin_lock(&clp->cl_lock);
3986 c = __nfsd4_find_conn(xpt, session);
3987 if (!c)
3988 status = nfserr_noent;
3989 else if (req == c->cn_flags)
3990 status = nfs_ok;
3991 else if (req == NFS4_CDFC4_FORE_OR_BOTH &&
3992 c->cn_flags != NFS4_CDFC4_BACK)
3993 status = nfs_ok;
3994 else if (req == NFS4_CDFC4_BACK_OR_BOTH &&
3995 c->cn_flags != NFS4_CDFC4_FORE)
3996 status = nfs_ok;
3997 else
3998 status = nfserr_inval;
3999 spin_unlock(&clp->cl_lock);
4000 if (status == nfs_ok && conn)
4001 *conn = c;
4002 return status;
4003 }
4004
nfsd4_bind_conn_to_session(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4005 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
4006 struct nfsd4_compound_state *cstate,
4007 union nfsd4_op_u *u)
4008 {
4009 struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
4010 __be32 status;
4011 struct nfsd4_conn *conn;
4012 struct nfsd4_session *session;
4013 struct net *net = SVC_NET(rqstp);
4014 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4015
4016 if (!nfsd4_last_compound_op(rqstp))
4017 return nfserr_not_only_op;
4018 spin_lock(&nn->client_lock);
4019 session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
4020 spin_unlock(&nn->client_lock);
4021 if (!session)
4022 goto out_no_session;
4023 status = nfserr_wrong_cred;
4024 if (!nfsd4_mach_creds_match(session->se_client, rqstp))
4025 goto out;
4026 status = nfsd4_match_existing_connection(rqstp, session,
4027 bcts->dir, &conn);
4028 if (status == nfs_ok) {
4029 if (bcts->dir == NFS4_CDFC4_FORE_OR_BOTH ||
4030 bcts->dir == NFS4_CDFC4_BACK)
4031 conn->cn_flags |= NFS4_CDFC4_BACK;
4032 nfsd4_probe_callback(session->se_client);
4033 goto out;
4034 }
4035 if (status == nfserr_inval)
4036 goto out;
4037 status = nfsd4_map_bcts_dir(&bcts->dir);
4038 if (status)
4039 goto out;
4040 conn = alloc_conn(rqstp, bcts->dir);
4041 status = nfserr_jukebox;
4042 if (!conn)
4043 goto out;
4044 nfsd4_init_conn(rqstp, conn, session);
4045 status = nfs_ok;
4046 out:
4047 nfsd4_put_session(session);
4048 out_no_session:
4049 return status;
4050 }
4051
nfsd4_compound_in_session(struct nfsd4_compound_state * cstate,struct nfs4_sessionid * sid)4052 static bool nfsd4_compound_in_session(struct nfsd4_compound_state *cstate, struct nfs4_sessionid *sid)
4053 {
4054 if (!cstate->session)
4055 return false;
4056 return !memcmp(sid, &cstate->session->se_sessionid, sizeof(*sid));
4057 }
4058
4059 __be32
nfsd4_destroy_session(struct svc_rqst * r,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4060 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
4061 union nfsd4_op_u *u)
4062 {
4063 struct nfs4_sessionid *sessionid = &u->destroy_session.sessionid;
4064 struct nfsd4_session *ses;
4065 __be32 status;
4066 int ref_held_by_me = 0;
4067 struct net *net = SVC_NET(r);
4068 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4069
4070 status = nfserr_not_only_op;
4071 if (nfsd4_compound_in_session(cstate, sessionid)) {
4072 if (!nfsd4_last_compound_op(r))
4073 goto out;
4074 ref_held_by_me++;
4075 }
4076 dump_sessionid(__func__, sessionid);
4077 spin_lock(&nn->client_lock);
4078 ses = find_in_sessionid_hashtbl(sessionid, net, &status);
4079 if (!ses)
4080 goto out_client_lock;
4081 status = nfserr_wrong_cred;
4082 if (!nfsd4_mach_creds_match(ses->se_client, r))
4083 goto out_put_session;
4084 status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
4085 if (status)
4086 goto out_put_session;
4087 unhash_session(ses);
4088 spin_unlock(&nn->client_lock);
4089
4090 nfsd4_probe_callback_sync(ses->se_client);
4091
4092 spin_lock(&nn->client_lock);
4093 status = nfs_ok;
4094 out_put_session:
4095 nfsd4_put_session_locked(ses);
4096 out_client_lock:
4097 spin_unlock(&nn->client_lock);
4098 out:
4099 return status;
4100 }
4101
nfsd4_sequence_check_conn(struct nfsd4_conn * new,struct nfsd4_session * ses)4102 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
4103 {
4104 struct nfs4_client *clp = ses->se_client;
4105 struct nfsd4_conn *c;
4106 __be32 status = nfs_ok;
4107 int ret;
4108
4109 spin_lock(&clp->cl_lock);
4110 c = __nfsd4_find_conn(new->cn_xprt, ses);
4111 if (c)
4112 goto out_free;
4113 status = nfserr_conn_not_bound_to_session;
4114 if (clp->cl_mach_cred)
4115 goto out_free;
4116 __nfsd4_hash_conn(new, ses);
4117 spin_unlock(&clp->cl_lock);
4118 ret = nfsd4_register_conn(new);
4119 if (ret)
4120 /* oops; xprt is already down: */
4121 nfsd4_conn_lost(&new->cn_xpt_user);
4122 return nfs_ok;
4123 out_free:
4124 spin_unlock(&clp->cl_lock);
4125 free_conn(new);
4126 return status;
4127 }
4128
nfsd4_session_too_many_ops(struct svc_rqst * rqstp,struct nfsd4_session * session)4129 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
4130 {
4131 struct nfsd4_compoundargs *args = rqstp->rq_argp;
4132
4133 return args->opcnt > session->se_fchannel.maxops;
4134 }
4135
nfsd4_request_too_big(struct svc_rqst * rqstp,struct nfsd4_session * session)4136 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
4137 struct nfsd4_session *session)
4138 {
4139 struct xdr_buf *xb = &rqstp->rq_arg;
4140
4141 return xb->len > session->se_fchannel.maxreq_sz;
4142 }
4143
replay_matches_cache(struct svc_rqst * rqstp,struct nfsd4_sequence * seq,struct nfsd4_slot * slot)4144 static bool replay_matches_cache(struct svc_rqst *rqstp,
4145 struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
4146 {
4147 struct nfsd4_compoundargs *argp = rqstp->rq_argp;
4148
4149 if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
4150 (bool)seq->cachethis)
4151 return false;
4152 /*
4153 * If there's an error then the reply can have fewer ops than
4154 * the call.
4155 */
4156 if (slot->sl_opcnt < argp->opcnt && !slot->sl_status)
4157 return false;
4158 /*
4159 * But if we cached a reply with *more* ops than the call you're
4160 * sending us now, then this new call is clearly not really a
4161 * replay of the old one:
4162 */
4163 if (slot->sl_opcnt > argp->opcnt)
4164 return false;
4165 /* This is the only check explicitly called by spec: */
4166 if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
4167 return false;
4168 /*
4169 * There may be more comparisons we could actually do, but the
4170 * spec doesn't require us to catch every case where the calls
4171 * don't match (that would require caching the call as well as
4172 * the reply), so we don't bother.
4173 */
4174 return true;
4175 }
4176
4177 __be32
nfsd4_sequence(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4178 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4179 union nfsd4_op_u *u)
4180 {
4181 struct nfsd4_sequence *seq = &u->sequence;
4182 struct nfsd4_compoundres *resp = rqstp->rq_resp;
4183 struct xdr_stream *xdr = resp->xdr;
4184 struct nfsd4_session *session;
4185 struct nfs4_client *clp;
4186 struct nfsd4_slot *slot;
4187 struct nfsd4_conn *conn;
4188 __be32 status;
4189 int buflen;
4190 struct net *net = SVC_NET(rqstp);
4191 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4192
4193 if (resp->opcnt != 1)
4194 return nfserr_sequence_pos;
4195
4196 /*
4197 * Will be either used or freed by nfsd4_sequence_check_conn
4198 * below.
4199 */
4200 conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
4201 if (!conn)
4202 return nfserr_jukebox;
4203
4204 spin_lock(&nn->client_lock);
4205 session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
4206 if (!session)
4207 goto out_no_session;
4208 clp = session->se_client;
4209
4210 status = nfserr_too_many_ops;
4211 if (nfsd4_session_too_many_ops(rqstp, session))
4212 goto out_put_session;
4213
4214 status = nfserr_req_too_big;
4215 if (nfsd4_request_too_big(rqstp, session))
4216 goto out_put_session;
4217
4218 status = nfserr_badslot;
4219 if (seq->slotid >= session->se_fchannel.maxreqs)
4220 goto out_put_session;
4221
4222 slot = session->se_slots[seq->slotid];
4223 dprintk("%s: slotid %d\n", __func__, seq->slotid);
4224
4225 /* We do not negotiate the number of slots yet, so set the
4226 * maxslots to the session maxreqs which is used to encode
4227 * sr_highest_slotid and the sr_target_slot id to maxslots */
4228 seq->maxslots = session->se_fchannel.maxreqs;
4229
4230 trace_nfsd_slot_seqid_sequence(clp, seq, slot);
4231 status = check_slot_seqid(seq->seqid, slot->sl_seqid,
4232 slot->sl_flags & NFSD4_SLOT_INUSE);
4233 if (status == nfserr_replay_cache) {
4234 status = nfserr_seq_misordered;
4235 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
4236 goto out_put_session;
4237 status = nfserr_seq_false_retry;
4238 if (!replay_matches_cache(rqstp, seq, slot))
4239 goto out_put_session;
4240 cstate->slot = slot;
4241 cstate->session = session;
4242 cstate->clp = clp;
4243 /* Return the cached reply status and set cstate->status
4244 * for nfsd4_proc_compound processing */
4245 status = nfsd4_replay_cache_entry(resp, seq);
4246 cstate->status = nfserr_replay_cache;
4247 goto out;
4248 }
4249 if (status)
4250 goto out_put_session;
4251
4252 status = nfsd4_sequence_check_conn(conn, session);
4253 conn = NULL;
4254 if (status)
4255 goto out_put_session;
4256
4257 buflen = (seq->cachethis) ?
4258 session->se_fchannel.maxresp_cached :
4259 session->se_fchannel.maxresp_sz;
4260 status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
4261 nfserr_rep_too_big;
4262 if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
4263 goto out_put_session;
4264 svc_reserve(rqstp, buflen);
4265
4266 status = nfs_ok;
4267 /* Success! bump slot seqid */
4268 slot->sl_seqid = seq->seqid;
4269 slot->sl_flags |= NFSD4_SLOT_INUSE;
4270 if (seq->cachethis)
4271 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
4272 else
4273 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
4274
4275 cstate->slot = slot;
4276 cstate->session = session;
4277 cstate->clp = clp;
4278
4279 out:
4280 switch (clp->cl_cb_state) {
4281 case NFSD4_CB_DOWN:
4282 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
4283 break;
4284 case NFSD4_CB_FAULT:
4285 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
4286 break;
4287 default:
4288 seq->status_flags = 0;
4289 }
4290 if (!list_empty(&clp->cl_revoked))
4291 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
4292 if (atomic_read(&clp->cl_admin_revoked))
4293 seq->status_flags |= SEQ4_STATUS_ADMIN_STATE_REVOKED;
4294 trace_nfsd_seq4_status(rqstp, seq);
4295 out_no_session:
4296 if (conn)
4297 free_conn(conn);
4298 spin_unlock(&nn->client_lock);
4299 return status;
4300 out_put_session:
4301 nfsd4_put_session_locked(session);
4302 goto out_no_session;
4303 }
4304
4305 void
nfsd4_sequence_done(struct nfsd4_compoundres * resp)4306 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
4307 {
4308 struct nfsd4_compound_state *cs = &resp->cstate;
4309
4310 if (nfsd4_has_session(cs)) {
4311 if (cs->status != nfserr_replay_cache) {
4312 nfsd4_store_cache_entry(resp);
4313 cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
4314 }
4315 /* Drop session reference that was taken in nfsd4_sequence() */
4316 nfsd4_put_session(cs->session);
4317 } else if (cs->clp)
4318 put_client_renew(cs->clp);
4319 }
4320
4321 __be32
nfsd4_destroy_clientid(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4322 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
4323 struct nfsd4_compound_state *cstate,
4324 union nfsd4_op_u *u)
4325 {
4326 struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
4327 struct nfs4_client *conf, *unconf;
4328 struct nfs4_client *clp = NULL;
4329 __be32 status = 0;
4330 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4331
4332 spin_lock(&nn->client_lock);
4333 unconf = find_unconfirmed_client(&dc->clientid, true, nn);
4334 conf = find_confirmed_client(&dc->clientid, true, nn);
4335 WARN_ON_ONCE(conf && unconf);
4336
4337 if (conf) {
4338 if (client_has_state(conf)) {
4339 status = nfserr_clientid_busy;
4340 goto out;
4341 }
4342 status = mark_client_expired_locked(conf);
4343 if (status)
4344 goto out;
4345 clp = conf;
4346 } else if (unconf)
4347 clp = unconf;
4348 else {
4349 status = nfserr_stale_clientid;
4350 goto out;
4351 }
4352 if (!nfsd4_mach_creds_match(clp, rqstp)) {
4353 clp = NULL;
4354 status = nfserr_wrong_cred;
4355 goto out;
4356 }
4357 trace_nfsd_clid_destroyed(&clp->cl_clientid);
4358 unhash_client_locked(clp);
4359 out:
4360 spin_unlock(&nn->client_lock);
4361 if (clp)
4362 expire_client(clp);
4363 return status;
4364 }
4365
4366 __be32
nfsd4_reclaim_complete(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4367 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
4368 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
4369 {
4370 struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
4371 struct nfs4_client *clp = cstate->clp;
4372 __be32 status = 0;
4373
4374 if (rc->rca_one_fs) {
4375 if (!cstate->current_fh.fh_dentry)
4376 return nfserr_nofilehandle;
4377 /*
4378 * We don't take advantage of the rca_one_fs case.
4379 * That's OK, it's optional, we can safely ignore it.
4380 */
4381 return nfs_ok;
4382 }
4383
4384 status = nfserr_complete_already;
4385 if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
4386 goto out;
4387
4388 status = nfserr_stale_clientid;
4389 if (is_client_expired(clp))
4390 /*
4391 * The following error isn't really legal.
4392 * But we only get here if the client just explicitly
4393 * destroyed the client. Surely it no longer cares what
4394 * error it gets back on an operation for the dead
4395 * client.
4396 */
4397 goto out;
4398
4399 status = nfs_ok;
4400 trace_nfsd_clid_reclaim_complete(&clp->cl_clientid);
4401 nfsd4_client_record_create(clp);
4402 inc_reclaim_complete(clp);
4403 out:
4404 return status;
4405 }
4406
4407 __be32
nfsd4_setclientid(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4408 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4409 union nfsd4_op_u *u)
4410 {
4411 struct nfsd4_setclientid *setclid = &u->setclientid;
4412 struct xdr_netobj clname = setclid->se_name;
4413 nfs4_verifier clverifier = setclid->se_verf;
4414 struct nfs4_client *conf, *new;
4415 struct nfs4_client *unconf = NULL;
4416 __be32 status;
4417 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4418
4419 new = create_client(clname, rqstp, &clverifier);
4420 if (new == NULL)
4421 return nfserr_jukebox;
4422 spin_lock(&nn->client_lock);
4423 conf = find_confirmed_client_by_name(&clname, nn);
4424 if (conf && client_has_state(conf)) {
4425 status = nfserr_clid_inuse;
4426 if (clp_used_exchangeid(conf))
4427 goto out;
4428 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4429 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4430 goto out;
4431 }
4432 }
4433 unconf = find_unconfirmed_client_by_name(&clname, nn);
4434 if (unconf)
4435 unhash_client_locked(unconf);
4436 if (conf) {
4437 if (same_verf(&conf->cl_verifier, &clverifier)) {
4438 copy_clid(new, conf);
4439 gen_confirm(new, nn);
4440 } else
4441 trace_nfsd_clid_verf_mismatch(conf, rqstp,
4442 &clverifier);
4443 } else
4444 trace_nfsd_clid_fresh(new);
4445 new->cl_minorversion = 0;
4446 gen_callback(new, setclid, rqstp);
4447 add_to_unconfirmed(new);
4448 setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
4449 setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
4450 memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
4451 new = NULL;
4452 status = nfs_ok;
4453 out:
4454 spin_unlock(&nn->client_lock);
4455 if (new)
4456 free_client(new);
4457 if (unconf) {
4458 trace_nfsd_clid_expire_unconf(&unconf->cl_clientid);
4459 expire_client(unconf);
4460 }
4461 return status;
4462 }
4463
4464 __be32
nfsd4_setclientid_confirm(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)4465 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
4466 struct nfsd4_compound_state *cstate,
4467 union nfsd4_op_u *u)
4468 {
4469 struct nfsd4_setclientid_confirm *setclientid_confirm =
4470 &u->setclientid_confirm;
4471 struct nfs4_client *conf, *unconf;
4472 struct nfs4_client *old = NULL;
4473 nfs4_verifier confirm = setclientid_confirm->sc_confirm;
4474 clientid_t * clid = &setclientid_confirm->sc_clientid;
4475 __be32 status;
4476 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4477
4478 if (STALE_CLIENTID(clid, nn))
4479 return nfserr_stale_clientid;
4480
4481 spin_lock(&nn->client_lock);
4482 conf = find_confirmed_client(clid, false, nn);
4483 unconf = find_unconfirmed_client(clid, false, nn);
4484 /*
4485 * We try hard to give out unique clientid's, so if we get an
4486 * attempt to confirm the same clientid with a different cred,
4487 * the client may be buggy; this should never happen.
4488 *
4489 * Nevertheless, RFC 7530 recommends INUSE for this case:
4490 */
4491 status = nfserr_clid_inuse;
4492 if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
4493 trace_nfsd_clid_cred_mismatch(unconf, rqstp);
4494 goto out;
4495 }
4496 if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
4497 trace_nfsd_clid_cred_mismatch(conf, rqstp);
4498 goto out;
4499 }
4500 if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
4501 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
4502 status = nfs_ok;
4503 } else
4504 status = nfserr_stale_clientid;
4505 goto out;
4506 }
4507 status = nfs_ok;
4508 if (conf) {
4509 old = unconf;
4510 unhash_client_locked(old);
4511 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
4512 } else {
4513 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
4514 if (old) {
4515 status = nfserr_clid_inuse;
4516 if (client_has_state(old)
4517 && !same_creds(&unconf->cl_cred,
4518 &old->cl_cred)) {
4519 old = NULL;
4520 goto out;
4521 }
4522 status = mark_client_expired_locked(old);
4523 if (status) {
4524 old = NULL;
4525 goto out;
4526 }
4527 trace_nfsd_clid_replaced(&old->cl_clientid);
4528 }
4529 move_to_confirmed(unconf);
4530 conf = unconf;
4531 }
4532 get_client_locked(conf);
4533 spin_unlock(&nn->client_lock);
4534 if (conf == unconf)
4535 fsnotify_dentry(conf->cl_nfsd_info_dentry, FS_MODIFY);
4536 nfsd4_probe_callback(conf);
4537 spin_lock(&nn->client_lock);
4538 put_client_renew_locked(conf);
4539 out:
4540 spin_unlock(&nn->client_lock);
4541 if (old)
4542 expire_client(old);
4543 return status;
4544 }
4545
nfsd4_alloc_file(void)4546 static struct nfs4_file *nfsd4_alloc_file(void)
4547 {
4548 return kmem_cache_alloc(file_slab, GFP_KERNEL);
4549 }
4550
4551 /* OPEN Share state helper functions */
4552
nfsd4_file_init(const struct svc_fh * fh,struct nfs4_file * fp)4553 static void nfsd4_file_init(const struct svc_fh *fh, struct nfs4_file *fp)
4554 {
4555 refcount_set(&fp->fi_ref, 1);
4556 spin_lock_init(&fp->fi_lock);
4557 INIT_LIST_HEAD(&fp->fi_stateids);
4558 INIT_LIST_HEAD(&fp->fi_delegations);
4559 INIT_LIST_HEAD(&fp->fi_clnt_odstate);
4560 fh_copy_shallow(&fp->fi_fhandle, &fh->fh_handle);
4561 fp->fi_deleg_file = NULL;
4562 fp->fi_had_conflict = false;
4563 fp->fi_share_deny = 0;
4564 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
4565 memset(fp->fi_access, 0, sizeof(fp->fi_access));
4566 fp->fi_aliased = false;
4567 fp->fi_inode = d_inode(fh->fh_dentry);
4568 #ifdef CONFIG_NFSD_PNFS
4569 INIT_LIST_HEAD(&fp->fi_lo_states);
4570 atomic_set(&fp->fi_lo_recalls, 0);
4571 #endif
4572 }
4573
4574 void
nfsd4_free_slabs(void)4575 nfsd4_free_slabs(void)
4576 {
4577 kmem_cache_destroy(client_slab);
4578 kmem_cache_destroy(openowner_slab);
4579 kmem_cache_destroy(lockowner_slab);
4580 kmem_cache_destroy(file_slab);
4581 kmem_cache_destroy(stateid_slab);
4582 kmem_cache_destroy(deleg_slab);
4583 kmem_cache_destroy(odstate_slab);
4584 }
4585
4586 int
nfsd4_init_slabs(void)4587 nfsd4_init_slabs(void)
4588 {
4589 client_slab = KMEM_CACHE(nfs4_client, 0);
4590 if (client_slab == NULL)
4591 goto out;
4592 openowner_slab = KMEM_CACHE(nfs4_openowner, 0);
4593 if (openowner_slab == NULL)
4594 goto out_free_client_slab;
4595 lockowner_slab = KMEM_CACHE(nfs4_lockowner, 0);
4596 if (lockowner_slab == NULL)
4597 goto out_free_openowner_slab;
4598 file_slab = KMEM_CACHE(nfs4_file, 0);
4599 if (file_slab == NULL)
4600 goto out_free_lockowner_slab;
4601 stateid_slab = KMEM_CACHE(nfs4_ol_stateid, 0);
4602 if (stateid_slab == NULL)
4603 goto out_free_file_slab;
4604 deleg_slab = KMEM_CACHE(nfs4_delegation, 0);
4605 if (deleg_slab == NULL)
4606 goto out_free_stateid_slab;
4607 odstate_slab = KMEM_CACHE(nfs4_clnt_odstate, 0);
4608 if (odstate_slab == NULL)
4609 goto out_free_deleg_slab;
4610 return 0;
4611
4612 out_free_deleg_slab:
4613 kmem_cache_destroy(deleg_slab);
4614 out_free_stateid_slab:
4615 kmem_cache_destroy(stateid_slab);
4616 out_free_file_slab:
4617 kmem_cache_destroy(file_slab);
4618 out_free_lockowner_slab:
4619 kmem_cache_destroy(lockowner_slab);
4620 out_free_openowner_slab:
4621 kmem_cache_destroy(openowner_slab);
4622 out_free_client_slab:
4623 kmem_cache_destroy(client_slab);
4624 out:
4625 return -ENOMEM;
4626 }
4627
4628 static unsigned long
nfsd4_state_shrinker_count(struct shrinker * shrink,struct shrink_control * sc)4629 nfsd4_state_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
4630 {
4631 int count;
4632 struct nfsd_net *nn = shrink->private_data;
4633
4634 count = atomic_read(&nn->nfsd_courtesy_clients);
4635 if (!count)
4636 count = atomic_long_read(&num_delegations);
4637 if (count)
4638 queue_work(laundry_wq, &nn->nfsd_shrinker_work);
4639 return (unsigned long)count;
4640 }
4641
4642 static unsigned long
nfsd4_state_shrinker_scan(struct shrinker * shrink,struct shrink_control * sc)4643 nfsd4_state_shrinker_scan(struct shrinker *shrink, struct shrink_control *sc)
4644 {
4645 return SHRINK_STOP;
4646 }
4647
4648 void
nfsd4_init_leases_net(struct nfsd_net * nn)4649 nfsd4_init_leases_net(struct nfsd_net *nn)
4650 {
4651 struct sysinfo si;
4652 u64 max_clients;
4653
4654 nn->nfsd4_lease = 90; /* default lease time */
4655 nn->nfsd4_grace = 90;
4656 nn->somebody_reclaimed = false;
4657 nn->track_reclaim_completes = false;
4658 nn->clverifier_counter = get_random_u32();
4659 nn->clientid_base = get_random_u32();
4660 nn->clientid_counter = nn->clientid_base + 1;
4661 nn->s2s_cp_cl_id = nn->clientid_counter++;
4662
4663 atomic_set(&nn->nfs4_client_count, 0);
4664 si_meminfo(&si);
4665 max_clients = (u64)si.totalram * si.mem_unit / (1024 * 1024 * 1024);
4666 max_clients *= NFS4_CLIENTS_PER_GB;
4667 nn->nfs4_max_clients = max_t(int, max_clients, NFS4_CLIENTS_PER_GB);
4668
4669 atomic_set(&nn->nfsd_courtesy_clients, 0);
4670 }
4671
4672 enum rp_lock {
4673 RP_UNLOCKED,
4674 RP_LOCKED,
4675 RP_UNHASHED,
4676 };
4677
init_nfs4_replay(struct nfs4_replay * rp)4678 static void init_nfs4_replay(struct nfs4_replay *rp)
4679 {
4680 rp->rp_status = nfserr_serverfault;
4681 rp->rp_buflen = 0;
4682 rp->rp_buf = rp->rp_ibuf;
4683 atomic_set(&rp->rp_locked, RP_UNLOCKED);
4684 }
4685
nfsd4_cstate_assign_replay(struct nfsd4_compound_state * cstate,struct nfs4_stateowner * so)4686 static int nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
4687 struct nfs4_stateowner *so)
4688 {
4689 if (!nfsd4_has_session(cstate)) {
4690 wait_var_event(&so->so_replay.rp_locked,
4691 atomic_cmpxchg(&so->so_replay.rp_locked,
4692 RP_UNLOCKED, RP_LOCKED) != RP_LOCKED);
4693 if (atomic_read(&so->so_replay.rp_locked) == RP_UNHASHED)
4694 return -EAGAIN;
4695 cstate->replay_owner = nfs4_get_stateowner(so);
4696 }
4697 return 0;
4698 }
4699
nfsd4_cstate_clear_replay(struct nfsd4_compound_state * cstate)4700 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
4701 {
4702 struct nfs4_stateowner *so = cstate->replay_owner;
4703
4704 if (so != NULL) {
4705 cstate->replay_owner = NULL;
4706 atomic_set(&so->so_replay.rp_locked, RP_UNLOCKED);
4707 wake_up_var(&so->so_replay.rp_locked);
4708 nfs4_put_stateowner(so);
4709 }
4710 }
4711
alloc_stateowner(struct kmem_cache * slab,struct xdr_netobj * owner,struct nfs4_client * clp)4712 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
4713 {
4714 struct nfs4_stateowner *sop;
4715
4716 sop = kmem_cache_alloc(slab, GFP_KERNEL);
4717 if (!sop)
4718 return NULL;
4719
4720 xdr_netobj_dup(&sop->so_owner, owner, GFP_KERNEL);
4721 if (!sop->so_owner.data) {
4722 kmem_cache_free(slab, sop);
4723 return NULL;
4724 }
4725
4726 INIT_LIST_HEAD(&sop->so_stateids);
4727 sop->so_client = clp;
4728 init_nfs4_replay(&sop->so_replay);
4729 atomic_set(&sop->so_count, 1);
4730 return sop;
4731 }
4732
hash_openowner(struct nfs4_openowner * oo,struct nfs4_client * clp,unsigned int strhashval)4733 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
4734 {
4735 lockdep_assert_held(&clp->cl_lock);
4736
4737 list_add(&oo->oo_owner.so_strhash,
4738 &clp->cl_ownerstr_hashtbl[strhashval]);
4739 list_add(&oo->oo_perclient, &clp->cl_openowners);
4740 }
4741
nfs4_unhash_openowner(struct nfs4_stateowner * so)4742 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
4743 {
4744 unhash_openowner_locked(openowner(so));
4745 }
4746
nfs4_free_openowner(struct nfs4_stateowner * so)4747 static void nfs4_free_openowner(struct nfs4_stateowner *so)
4748 {
4749 struct nfs4_openowner *oo = openowner(so);
4750
4751 kmem_cache_free(openowner_slab, oo);
4752 }
4753
4754 static const struct nfs4_stateowner_operations openowner_ops = {
4755 .so_unhash = nfs4_unhash_openowner,
4756 .so_free = nfs4_free_openowner,
4757 };
4758
4759 static struct nfs4_ol_stateid *
nfsd4_find_existing_open(struct nfs4_file * fp,struct nfsd4_open * open)4760 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4761 {
4762 struct nfs4_ol_stateid *local, *ret = NULL;
4763 struct nfs4_openowner *oo = open->op_openowner;
4764
4765 lockdep_assert_held(&fp->fi_lock);
4766
4767 list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
4768 /* ignore lock owners */
4769 if (local->st_stateowner->so_is_open_owner == 0)
4770 continue;
4771 if (local->st_stateowner != &oo->oo_owner)
4772 continue;
4773 if (local->st_stid.sc_type == SC_TYPE_OPEN &&
4774 !local->st_stid.sc_status) {
4775 ret = local;
4776 refcount_inc(&ret->st_stid.sc_count);
4777 break;
4778 }
4779 }
4780 return ret;
4781 }
4782
nfsd4_drop_revoked_stid(struct nfs4_stid * s)4783 static void nfsd4_drop_revoked_stid(struct nfs4_stid *s)
4784 __releases(&s->sc_client->cl_lock)
4785 {
4786 struct nfs4_client *cl = s->sc_client;
4787 LIST_HEAD(reaplist);
4788 struct nfs4_ol_stateid *stp;
4789 struct nfs4_delegation *dp;
4790 bool unhashed;
4791
4792 switch (s->sc_type) {
4793 case SC_TYPE_OPEN:
4794 stp = openlockstateid(s);
4795 if (unhash_open_stateid(stp, &reaplist))
4796 put_ol_stateid_locked(stp, &reaplist);
4797 spin_unlock(&cl->cl_lock);
4798 free_ol_stateid_reaplist(&reaplist);
4799 break;
4800 case SC_TYPE_LOCK:
4801 stp = openlockstateid(s);
4802 unhashed = unhash_lock_stateid(stp);
4803 spin_unlock(&cl->cl_lock);
4804 if (unhashed)
4805 nfs4_put_stid(s);
4806 break;
4807 case SC_TYPE_DELEG:
4808 dp = delegstateid(s);
4809 list_del_init(&dp->dl_recall_lru);
4810 spin_unlock(&cl->cl_lock);
4811 nfs4_put_stid(s);
4812 break;
4813 default:
4814 spin_unlock(&cl->cl_lock);
4815 }
4816 }
4817
nfsd40_drop_revoked_stid(struct nfs4_client * cl,stateid_t * stid)4818 static void nfsd40_drop_revoked_stid(struct nfs4_client *cl,
4819 stateid_t *stid)
4820 {
4821 /* NFSv4.0 has no way for the client to tell the server
4822 * that it can forget an admin-revoked stateid.
4823 * So we keep it around until the first time that the
4824 * client uses it, and drop it the first time
4825 * nfserr_admin_revoked is returned.
4826 * For v4.1 and later we wait until explicitly told
4827 * to free the stateid.
4828 */
4829 if (cl->cl_minorversion == 0) {
4830 struct nfs4_stid *st;
4831
4832 spin_lock(&cl->cl_lock);
4833 st = find_stateid_locked(cl, stid);
4834 if (st)
4835 nfsd4_drop_revoked_stid(st);
4836 else
4837 spin_unlock(&cl->cl_lock);
4838 }
4839 }
4840
4841 static __be32
nfsd4_verify_open_stid(struct nfs4_stid * s)4842 nfsd4_verify_open_stid(struct nfs4_stid *s)
4843 {
4844 __be32 ret = nfs_ok;
4845
4846 if (s->sc_status & SC_STATUS_ADMIN_REVOKED)
4847 ret = nfserr_admin_revoked;
4848 else if (s->sc_status & SC_STATUS_REVOKED)
4849 ret = nfserr_deleg_revoked;
4850 else if (s->sc_status & SC_STATUS_CLOSED)
4851 ret = nfserr_bad_stateid;
4852 return ret;
4853 }
4854
4855 /* Lock the stateid st_mutex, and deal with races with CLOSE */
4856 static __be32
nfsd4_lock_ol_stateid(struct nfs4_ol_stateid * stp)4857 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
4858 {
4859 __be32 ret;
4860
4861 mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
4862 ret = nfsd4_verify_open_stid(&stp->st_stid);
4863 if (ret == nfserr_admin_revoked)
4864 nfsd40_drop_revoked_stid(stp->st_stid.sc_client,
4865 &stp->st_stid.sc_stateid);
4866
4867 if (ret != nfs_ok)
4868 mutex_unlock(&stp->st_mutex);
4869 return ret;
4870 }
4871
4872 static struct nfs4_ol_stateid *
nfsd4_find_and_lock_existing_open(struct nfs4_file * fp,struct nfsd4_open * open)4873 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
4874 {
4875 struct nfs4_ol_stateid *stp;
4876 for (;;) {
4877 spin_lock(&fp->fi_lock);
4878 stp = nfsd4_find_existing_open(fp, open);
4879 spin_unlock(&fp->fi_lock);
4880 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
4881 break;
4882 nfs4_put_stid(&stp->st_stid);
4883 }
4884 return stp;
4885 }
4886
4887 static struct nfs4_openowner *
find_or_alloc_open_stateowner(unsigned int strhashval,struct nfsd4_open * open,struct nfsd4_compound_state * cstate)4888 find_or_alloc_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
4889 struct nfsd4_compound_state *cstate)
4890 {
4891 struct nfs4_client *clp = cstate->clp;
4892 struct nfs4_openowner *oo, *new = NULL;
4893
4894 retry:
4895 spin_lock(&clp->cl_lock);
4896 oo = find_openstateowner_str(strhashval, open, clp);
4897 if (!oo && new) {
4898 hash_openowner(new, clp, strhashval);
4899 spin_unlock(&clp->cl_lock);
4900 return new;
4901 }
4902 spin_unlock(&clp->cl_lock);
4903
4904 if (oo && !(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4905 /* Replace unconfirmed owners without checking for replay. */
4906 release_openowner(oo);
4907 oo = NULL;
4908 }
4909 if (oo) {
4910 if (new)
4911 nfs4_free_stateowner(&new->oo_owner);
4912 return oo;
4913 }
4914
4915 new = alloc_stateowner(openowner_slab, &open->op_owner, clp);
4916 if (!new)
4917 return NULL;
4918 new->oo_owner.so_ops = &openowner_ops;
4919 new->oo_owner.so_is_open_owner = 1;
4920 new->oo_owner.so_seqid = open->op_seqid;
4921 new->oo_flags = 0;
4922 if (nfsd4_has_session(cstate))
4923 new->oo_flags |= NFS4_OO_CONFIRMED;
4924 new->oo_time = 0;
4925 new->oo_last_closed_stid = NULL;
4926 INIT_LIST_HEAD(&new->oo_close_lru);
4927 goto retry;
4928 }
4929
4930 static struct nfs4_ol_stateid *
init_open_stateid(struct nfs4_file * fp,struct nfsd4_open * open)4931 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
4932 {
4933
4934 struct nfs4_openowner *oo = open->op_openowner;
4935 struct nfs4_ol_stateid *retstp = NULL;
4936 struct nfs4_ol_stateid *stp;
4937
4938 stp = open->op_stp;
4939 /* We are moving these outside of the spinlocks to avoid the warnings */
4940 mutex_init(&stp->st_mutex);
4941 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
4942
4943 retry:
4944 spin_lock(&oo->oo_owner.so_client->cl_lock);
4945 spin_lock(&fp->fi_lock);
4946
4947 retstp = nfsd4_find_existing_open(fp, open);
4948 if (retstp)
4949 goto out_unlock;
4950
4951 open->op_stp = NULL;
4952 refcount_inc(&stp->st_stid.sc_count);
4953 stp->st_stid.sc_type = SC_TYPE_OPEN;
4954 INIT_LIST_HEAD(&stp->st_locks);
4955 stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
4956 get_nfs4_file(fp);
4957 stp->st_stid.sc_file = fp;
4958 stp->st_access_bmap = 0;
4959 stp->st_deny_bmap = 0;
4960 stp->st_openstp = NULL;
4961 list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
4962 list_add(&stp->st_perfile, &fp->fi_stateids);
4963
4964 out_unlock:
4965 spin_unlock(&fp->fi_lock);
4966 spin_unlock(&oo->oo_owner.so_client->cl_lock);
4967 if (retstp) {
4968 /* Handle races with CLOSE */
4969 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
4970 nfs4_put_stid(&retstp->st_stid);
4971 goto retry;
4972 }
4973 /* To keep mutex tracking happy */
4974 mutex_unlock(&stp->st_mutex);
4975 stp = retstp;
4976 }
4977 return stp;
4978 }
4979
4980 /*
4981 * In the 4.0 case we need to keep the owners around a little while to handle
4982 * CLOSE replay. We still do need to release any file access that is held by
4983 * them before returning however.
4984 */
4985 static void
move_to_close_lru(struct nfs4_ol_stateid * s,struct net * net)4986 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
4987 {
4988 struct nfs4_ol_stateid *last;
4989 struct nfs4_openowner *oo = openowner(s->st_stateowner);
4990 struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
4991 nfsd_net_id);
4992
4993 dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
4994
4995 /*
4996 * We know that we hold one reference via nfsd4_close, and another
4997 * "persistent" reference for the client. If the refcount is higher
4998 * than 2, then there are still calls in progress that are using this
4999 * stateid. We can't put the sc_file reference until they are finished.
5000 * Wait for the refcount to drop to 2. Since it has been unhashed,
5001 * there should be no danger of the refcount going back up again at
5002 * this point.
5003 * Some threads with a reference might be waiting for rp_locked,
5004 * so tell them to stop waiting.
5005 */
5006 atomic_set(&oo->oo_owner.so_replay.rp_locked, RP_UNHASHED);
5007 wake_up_var(&oo->oo_owner.so_replay.rp_locked);
5008 wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
5009
5010 release_all_access(s);
5011 if (s->st_stid.sc_file) {
5012 put_nfs4_file(s->st_stid.sc_file);
5013 s->st_stid.sc_file = NULL;
5014 }
5015
5016 spin_lock(&nn->client_lock);
5017 last = oo->oo_last_closed_stid;
5018 oo->oo_last_closed_stid = s;
5019 list_move_tail(&oo->oo_close_lru, &nn->close_lru);
5020 oo->oo_time = ktime_get_boottime_seconds();
5021 spin_unlock(&nn->client_lock);
5022 if (last)
5023 nfs4_put_stid(&last->st_stid);
5024 }
5025
5026 static noinline_for_stack struct nfs4_file *
nfsd4_file_hash_lookup(const struct svc_fh * fhp)5027 nfsd4_file_hash_lookup(const struct svc_fh *fhp)
5028 {
5029 struct inode *inode = d_inode(fhp->fh_dentry);
5030 struct rhlist_head *tmp, *list;
5031 struct nfs4_file *fi;
5032
5033 rcu_read_lock();
5034 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5035 nfs4_file_rhash_params);
5036 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5037 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5038 if (refcount_inc_not_zero(&fi->fi_ref)) {
5039 rcu_read_unlock();
5040 return fi;
5041 }
5042 }
5043 }
5044 rcu_read_unlock();
5045 return NULL;
5046 }
5047
5048 /*
5049 * On hash insertion, identify entries with the same inode but
5050 * distinct filehandles. They will all be on the list returned
5051 * by rhltable_lookup().
5052 *
5053 * inode->i_lock prevents racing insertions from adding an entry
5054 * for the same inode/fhp pair twice.
5055 */
5056 static noinline_for_stack struct nfs4_file *
nfsd4_file_hash_insert(struct nfs4_file * new,const struct svc_fh * fhp)5057 nfsd4_file_hash_insert(struct nfs4_file *new, const struct svc_fh *fhp)
5058 {
5059 struct inode *inode = d_inode(fhp->fh_dentry);
5060 struct rhlist_head *tmp, *list;
5061 struct nfs4_file *ret = NULL;
5062 bool alias_found = false;
5063 struct nfs4_file *fi;
5064 int err;
5065
5066 rcu_read_lock();
5067 spin_lock(&inode->i_lock);
5068
5069 list = rhltable_lookup(&nfs4_file_rhltable, &inode,
5070 nfs4_file_rhash_params);
5071 rhl_for_each_entry_rcu(fi, tmp, list, fi_rlist) {
5072 if (fh_match(&fi->fi_fhandle, &fhp->fh_handle)) {
5073 if (refcount_inc_not_zero(&fi->fi_ref))
5074 ret = fi;
5075 } else
5076 fi->fi_aliased = alias_found = true;
5077 }
5078 if (ret)
5079 goto out_unlock;
5080
5081 nfsd4_file_init(fhp, new);
5082 err = rhltable_insert(&nfs4_file_rhltable, &new->fi_rlist,
5083 nfs4_file_rhash_params);
5084 if (err)
5085 goto out_unlock;
5086
5087 new->fi_aliased = alias_found;
5088 ret = new;
5089
5090 out_unlock:
5091 spin_unlock(&inode->i_lock);
5092 rcu_read_unlock();
5093 return ret;
5094 }
5095
nfsd4_file_hash_remove(struct nfs4_file * fi)5096 static noinline_for_stack void nfsd4_file_hash_remove(struct nfs4_file *fi)
5097 {
5098 rhltable_remove(&nfs4_file_rhltable, &fi->fi_rlist,
5099 nfs4_file_rhash_params);
5100 }
5101
5102 /*
5103 * Called to check deny when READ with all zero stateid or
5104 * WRITE with all zero or all one stateid
5105 */
5106 static __be32
nfs4_share_conflict(struct svc_fh * current_fh,unsigned int deny_type)5107 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
5108 {
5109 struct nfs4_file *fp;
5110 __be32 ret = nfs_ok;
5111
5112 fp = nfsd4_file_hash_lookup(current_fh);
5113 if (!fp)
5114 return ret;
5115
5116 /* Check for conflicting share reservations */
5117 spin_lock(&fp->fi_lock);
5118 if (fp->fi_share_deny & deny_type)
5119 ret = nfserr_locked;
5120 spin_unlock(&fp->fi_lock);
5121 put_nfs4_file(fp);
5122 return ret;
5123 }
5124
nfsd4_deleg_present(const struct inode * inode)5125 static bool nfsd4_deleg_present(const struct inode *inode)
5126 {
5127 struct file_lock_context *ctx = locks_inode_context(inode);
5128
5129 return ctx && !list_empty_careful(&ctx->flc_lease);
5130 }
5131
5132 /**
5133 * nfsd_wait_for_delegreturn - wait for delegations to be returned
5134 * @rqstp: the RPC transaction being executed
5135 * @inode: in-core inode of the file being waited for
5136 *
5137 * The timeout prevents deadlock if all nfsd threads happen to be
5138 * tied up waiting for returning delegations.
5139 *
5140 * Return values:
5141 * %true: delegation was returned
5142 * %false: timed out waiting for delegreturn
5143 */
nfsd_wait_for_delegreturn(struct svc_rqst * rqstp,struct inode * inode)5144 bool nfsd_wait_for_delegreturn(struct svc_rqst *rqstp, struct inode *inode)
5145 {
5146 long __maybe_unused timeo;
5147
5148 timeo = wait_var_event_timeout(inode, !nfsd4_deleg_present(inode),
5149 NFSD_DELEGRETURN_TIMEOUT);
5150 trace_nfsd_delegret_wakeup(rqstp, inode, timeo);
5151 return timeo > 0;
5152 }
5153
nfsd4_cb_recall_prepare(struct nfsd4_callback * cb)5154 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
5155 {
5156 struct nfs4_delegation *dp = cb_to_delegation(cb);
5157 struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
5158 nfsd_net_id);
5159
5160 block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
5161
5162 /*
5163 * We can't do this in nfsd_break_deleg_cb because it is
5164 * already holding inode->i_lock.
5165 *
5166 * If the dl_time != 0, then we know that it has already been
5167 * queued for a lease break. Don't queue it again.
5168 */
5169 spin_lock(&state_lock);
5170 if (delegation_hashed(dp) && dp->dl_time == 0) {
5171 dp->dl_time = ktime_get_boottime_seconds();
5172 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
5173 }
5174 spin_unlock(&state_lock);
5175 }
5176
nfsd4_cb_recall_done(struct nfsd4_callback * cb,struct rpc_task * task)5177 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
5178 struct rpc_task *task)
5179 {
5180 struct nfs4_delegation *dp = cb_to_delegation(cb);
5181
5182 trace_nfsd_cb_recall_done(&dp->dl_stid.sc_stateid, task);
5183
5184 if (dp->dl_stid.sc_status)
5185 /* CLOSED or REVOKED */
5186 return 1;
5187
5188 switch (task->tk_status) {
5189 case 0:
5190 return 1;
5191 case -NFS4ERR_DELAY:
5192 rpc_delay(task, 2 * HZ);
5193 return 0;
5194 case -EBADHANDLE:
5195 case -NFS4ERR_BAD_STATEID:
5196 /*
5197 * Race: client probably got cb_recall before open reply
5198 * granting delegation.
5199 */
5200 if (dp->dl_retries--) {
5201 rpc_delay(task, 2 * HZ);
5202 return 0;
5203 }
5204 fallthrough;
5205 default:
5206 return 1;
5207 }
5208 }
5209
nfsd4_cb_recall_release(struct nfsd4_callback * cb)5210 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
5211 {
5212 struct nfs4_delegation *dp = cb_to_delegation(cb);
5213
5214 nfs4_put_stid(&dp->dl_stid);
5215 }
5216
5217 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
5218 .prepare = nfsd4_cb_recall_prepare,
5219 .done = nfsd4_cb_recall_done,
5220 .release = nfsd4_cb_recall_release,
5221 };
5222
nfsd_break_one_deleg(struct nfs4_delegation * dp)5223 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
5224 {
5225 /*
5226 * We're assuming the state code never drops its reference
5227 * without first removing the lease. Since we're in this lease
5228 * callback (and since the lease code is serialized by the
5229 * flc_lock) we know the server hasn't removed the lease yet, and
5230 * we know it's safe to take a reference.
5231 */
5232 refcount_inc(&dp->dl_stid.sc_count);
5233 WARN_ON_ONCE(!nfsd4_run_cb(&dp->dl_recall));
5234 }
5235
5236 /* Called from break_lease() with flc_lock held. */
5237 static bool
nfsd_break_deleg_cb(struct file_lease * fl)5238 nfsd_break_deleg_cb(struct file_lease *fl)
5239 {
5240 struct nfs4_delegation *dp = (struct nfs4_delegation *) fl->c.flc_owner;
5241 struct nfs4_file *fp = dp->dl_stid.sc_file;
5242 struct nfs4_client *clp = dp->dl_stid.sc_client;
5243 struct nfsd_net *nn;
5244
5245 trace_nfsd_cb_recall(&dp->dl_stid);
5246
5247 dp->dl_recalled = true;
5248 atomic_inc(&clp->cl_delegs_in_recall);
5249 if (try_to_expire_client(clp)) {
5250 nn = net_generic(clp->net, nfsd_net_id);
5251 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
5252 }
5253
5254 /*
5255 * We don't want the locks code to timeout the lease for us;
5256 * we'll remove it ourself if a delegation isn't returned
5257 * in time:
5258 */
5259 fl->fl_break_time = 0;
5260
5261 fp->fi_had_conflict = true;
5262 nfsd_break_one_deleg(dp);
5263 return false;
5264 }
5265
5266 /**
5267 * nfsd_breaker_owns_lease - Check if lease conflict was resolved
5268 * @fl: Lock state to check
5269 *
5270 * Return values:
5271 * %true: Lease conflict was resolved
5272 * %false: Lease conflict was not resolved.
5273 */
nfsd_breaker_owns_lease(struct file_lease * fl)5274 static bool nfsd_breaker_owns_lease(struct file_lease *fl)
5275 {
5276 struct nfs4_delegation *dl = fl->c.flc_owner;
5277 struct svc_rqst *rqst;
5278 struct nfs4_client *clp;
5279
5280 if (!i_am_nfsd())
5281 return false;
5282 rqst = kthread_data(current);
5283 /* Note rq_prog == NFS_ACL_PROGRAM is also possible: */
5284 if (rqst->rq_prog != NFS_PROGRAM || rqst->rq_vers < 4)
5285 return false;
5286 clp = *(rqst->rq_lease_breaker);
5287 return dl->dl_stid.sc_client == clp;
5288 }
5289
5290 static int
nfsd_change_deleg_cb(struct file_lease * onlist,int arg,struct list_head * dispose)5291 nfsd_change_deleg_cb(struct file_lease *onlist, int arg,
5292 struct list_head *dispose)
5293 {
5294 struct nfs4_delegation *dp = (struct nfs4_delegation *) onlist->c.flc_owner;
5295 struct nfs4_client *clp = dp->dl_stid.sc_client;
5296
5297 if (arg & F_UNLCK) {
5298 if (dp->dl_recalled)
5299 atomic_dec(&clp->cl_delegs_in_recall);
5300 return lease_modify(onlist, arg, dispose);
5301 } else
5302 return -EAGAIN;
5303 }
5304
5305 static const struct lease_manager_operations nfsd_lease_mng_ops = {
5306 .lm_breaker_owns_lease = nfsd_breaker_owns_lease,
5307 .lm_break = nfsd_break_deleg_cb,
5308 .lm_change = nfsd_change_deleg_cb,
5309 };
5310
nfsd4_check_seqid(struct nfsd4_compound_state * cstate,struct nfs4_stateowner * so,u32 seqid)5311 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
5312 {
5313 if (nfsd4_has_session(cstate))
5314 return nfs_ok;
5315 if (seqid == so->so_seqid - 1)
5316 return nfserr_replay_me;
5317 if (seqid == so->so_seqid)
5318 return nfs_ok;
5319 return nfserr_bad_seqid;
5320 }
5321
lookup_clientid(clientid_t * clid,bool sessions,struct nfsd_net * nn)5322 static struct nfs4_client *lookup_clientid(clientid_t *clid, bool sessions,
5323 struct nfsd_net *nn)
5324 {
5325 struct nfs4_client *found;
5326
5327 spin_lock(&nn->client_lock);
5328 found = find_confirmed_client(clid, sessions, nn);
5329 if (found)
5330 atomic_inc(&found->cl_rpc_users);
5331 spin_unlock(&nn->client_lock);
5332 return found;
5333 }
5334
set_client(clientid_t * clid,struct nfsd4_compound_state * cstate,struct nfsd_net * nn)5335 static __be32 set_client(clientid_t *clid,
5336 struct nfsd4_compound_state *cstate,
5337 struct nfsd_net *nn)
5338 {
5339 if (cstate->clp) {
5340 if (!same_clid(&cstate->clp->cl_clientid, clid))
5341 return nfserr_stale_clientid;
5342 return nfs_ok;
5343 }
5344 if (STALE_CLIENTID(clid, nn))
5345 return nfserr_stale_clientid;
5346 /*
5347 * We're in the 4.0 case (otherwise the SEQUENCE op would have
5348 * set cstate->clp), so session = false:
5349 */
5350 cstate->clp = lookup_clientid(clid, false, nn);
5351 if (!cstate->clp)
5352 return nfserr_expired;
5353 return nfs_ok;
5354 }
5355
5356 __be32
nfsd4_process_open1(struct nfsd4_compound_state * cstate,struct nfsd4_open * open,struct nfsd_net * nn)5357 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
5358 struct nfsd4_open *open, struct nfsd_net *nn)
5359 {
5360 clientid_t *clientid = &open->op_clientid;
5361 struct nfs4_client *clp = NULL;
5362 unsigned int strhashval;
5363 struct nfs4_openowner *oo = NULL;
5364 __be32 status;
5365
5366 /*
5367 * In case we need it later, after we've already created the
5368 * file and don't want to risk a further failure:
5369 */
5370 open->op_file = nfsd4_alloc_file();
5371 if (open->op_file == NULL)
5372 return nfserr_jukebox;
5373
5374 status = set_client(clientid, cstate, nn);
5375 if (status)
5376 return status;
5377 clp = cstate->clp;
5378
5379 strhashval = ownerstr_hashval(&open->op_owner);
5380 retry:
5381 oo = find_or_alloc_open_stateowner(strhashval, open, cstate);
5382 open->op_openowner = oo;
5383 if (!oo)
5384 return nfserr_jukebox;
5385 if (nfsd4_cstate_assign_replay(cstate, &oo->oo_owner) == -EAGAIN) {
5386 nfs4_put_stateowner(&oo->oo_owner);
5387 goto retry;
5388 }
5389 status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
5390 if (status)
5391 return status;
5392
5393 open->op_stp = nfs4_alloc_open_stateid(clp);
5394 if (!open->op_stp)
5395 return nfserr_jukebox;
5396
5397 if (nfsd4_has_session(cstate) &&
5398 (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
5399 open->op_odstate = alloc_clnt_odstate(clp);
5400 if (!open->op_odstate)
5401 return nfserr_jukebox;
5402 }
5403
5404 return nfs_ok;
5405 }
5406
5407 static inline __be32
nfs4_check_delegmode(struct nfs4_delegation * dp,int flags)5408 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
5409 {
5410 if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
5411 return nfserr_openmode;
5412 else
5413 return nfs_ok;
5414 }
5415
share_access_to_flags(u32 share_access)5416 static int share_access_to_flags(u32 share_access)
5417 {
5418 return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
5419 }
5420
find_deleg_stateid(struct nfs4_client * cl,stateid_t * s)5421 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl,
5422 stateid_t *s)
5423 {
5424 struct nfs4_stid *ret;
5425
5426 ret = find_stateid_by_type(cl, s, SC_TYPE_DELEG, SC_STATUS_REVOKED);
5427 if (!ret)
5428 return NULL;
5429 return delegstateid(ret);
5430 }
5431
nfsd4_is_deleg_cur(struct nfsd4_open * open)5432 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
5433 {
5434 return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
5435 open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
5436 }
5437
5438 static __be32
nfs4_check_deleg(struct nfs4_client * cl,struct nfsd4_open * open,struct nfs4_delegation ** dp)5439 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
5440 struct nfs4_delegation **dp)
5441 {
5442 int flags;
5443 __be32 status = nfserr_bad_stateid;
5444 struct nfs4_delegation *deleg;
5445
5446 deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
5447 if (deleg == NULL)
5448 goto out;
5449 if (deleg->dl_stid.sc_status & SC_STATUS_ADMIN_REVOKED) {
5450 nfs4_put_stid(&deleg->dl_stid);
5451 status = nfserr_admin_revoked;
5452 goto out;
5453 }
5454 if (deleg->dl_stid.sc_status & SC_STATUS_REVOKED) {
5455 nfs4_put_stid(&deleg->dl_stid);
5456 nfsd40_drop_revoked_stid(cl, &open->op_delegate_stateid);
5457 status = nfserr_deleg_revoked;
5458 goto out;
5459 }
5460 flags = share_access_to_flags(open->op_share_access);
5461 status = nfs4_check_delegmode(deleg, flags);
5462 if (status) {
5463 nfs4_put_stid(&deleg->dl_stid);
5464 goto out;
5465 }
5466 *dp = deleg;
5467 out:
5468 if (!nfsd4_is_deleg_cur(open))
5469 return nfs_ok;
5470 if (status)
5471 return status;
5472 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
5473 return nfs_ok;
5474 }
5475
nfs4_access_to_access(u32 nfs4_access)5476 static inline int nfs4_access_to_access(u32 nfs4_access)
5477 {
5478 int flags = 0;
5479
5480 if (nfs4_access & NFS4_SHARE_ACCESS_READ)
5481 flags |= NFSD_MAY_READ;
5482 if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
5483 flags |= NFSD_MAY_WRITE;
5484 return flags;
5485 }
5486
5487 static inline __be32
nfsd4_truncate(struct svc_rqst * rqstp,struct svc_fh * fh,struct nfsd4_open * open)5488 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
5489 struct nfsd4_open *open)
5490 {
5491 struct iattr iattr = {
5492 .ia_valid = ATTR_SIZE,
5493 .ia_size = 0,
5494 };
5495 struct nfsd_attrs attrs = {
5496 .na_iattr = &iattr,
5497 };
5498 if (!open->op_truncate)
5499 return 0;
5500 if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
5501 return nfserr_inval;
5502 return nfsd_setattr(rqstp, fh, &attrs, NULL);
5503 }
5504
nfs4_get_vfs_file(struct svc_rqst * rqstp,struct nfs4_file * fp,struct svc_fh * cur_fh,struct nfs4_ol_stateid * stp,struct nfsd4_open * open,bool new_stp)5505 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
5506 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5507 struct nfsd4_open *open, bool new_stp)
5508 {
5509 struct nfsd_file *nf = NULL;
5510 __be32 status;
5511 int oflag = nfs4_access_to_omode(open->op_share_access);
5512 int access = nfs4_access_to_access(open->op_share_access);
5513 unsigned char old_access_bmap, old_deny_bmap;
5514
5515 spin_lock(&fp->fi_lock);
5516
5517 /*
5518 * Are we trying to set a deny mode that would conflict with
5519 * current access?
5520 */
5521 status = nfs4_file_check_deny(fp, open->op_share_deny);
5522 if (status != nfs_ok) {
5523 if (status != nfserr_share_denied) {
5524 spin_unlock(&fp->fi_lock);
5525 goto out;
5526 }
5527 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5528 stp, open->op_share_deny, false))
5529 status = nfserr_jukebox;
5530 spin_unlock(&fp->fi_lock);
5531 goto out;
5532 }
5533
5534 /* set access to the file */
5535 status = nfs4_file_get_access(fp, open->op_share_access);
5536 if (status != nfs_ok) {
5537 if (status != nfserr_share_denied) {
5538 spin_unlock(&fp->fi_lock);
5539 goto out;
5540 }
5541 if (nfs4_resolve_deny_conflicts_locked(fp, new_stp,
5542 stp, open->op_share_access, true))
5543 status = nfserr_jukebox;
5544 spin_unlock(&fp->fi_lock);
5545 goto out;
5546 }
5547
5548 /* Set access bits in stateid */
5549 old_access_bmap = stp->st_access_bmap;
5550 set_access(open->op_share_access, stp);
5551
5552 /* Set new deny mask */
5553 old_deny_bmap = stp->st_deny_bmap;
5554 set_deny(open->op_share_deny, stp);
5555 fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5556
5557 if (!fp->fi_fds[oflag]) {
5558 spin_unlock(&fp->fi_lock);
5559
5560 status = nfsd_file_acquire_opened(rqstp, cur_fh, access,
5561 open->op_filp, &nf);
5562 if (status != nfs_ok)
5563 goto out_put_access;
5564
5565 spin_lock(&fp->fi_lock);
5566 if (!fp->fi_fds[oflag]) {
5567 fp->fi_fds[oflag] = nf;
5568 nf = NULL;
5569 }
5570 }
5571 spin_unlock(&fp->fi_lock);
5572 if (nf)
5573 nfsd_file_put(nf);
5574
5575 status = nfserrno(nfsd_open_break_lease(cur_fh->fh_dentry->d_inode,
5576 access));
5577 if (status)
5578 goto out_put_access;
5579
5580 status = nfsd4_truncate(rqstp, cur_fh, open);
5581 if (status)
5582 goto out_put_access;
5583 out:
5584 return status;
5585 out_put_access:
5586 stp->st_access_bmap = old_access_bmap;
5587 nfs4_file_put_access(fp, open->op_share_access);
5588 reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
5589 goto out;
5590 }
5591
5592 static __be32
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)5593 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp,
5594 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
5595 struct nfsd4_open *open)
5596 {
5597 __be32 status;
5598 unsigned char old_deny_bmap = stp->st_deny_bmap;
5599
5600 if (!test_access(open->op_share_access, stp))
5601 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open, false);
5602
5603 /* test and set deny mode */
5604 spin_lock(&fp->fi_lock);
5605 status = nfs4_file_check_deny(fp, open->op_share_deny);
5606 switch (status) {
5607 case nfs_ok:
5608 set_deny(open->op_share_deny, stp);
5609 fp->fi_share_deny |=
5610 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
5611 break;
5612 case nfserr_share_denied:
5613 if (nfs4_resolve_deny_conflicts_locked(fp, false,
5614 stp, open->op_share_deny, false))
5615 status = nfserr_jukebox;
5616 break;
5617 }
5618 spin_unlock(&fp->fi_lock);
5619
5620 if (status != nfs_ok)
5621 return status;
5622
5623 status = nfsd4_truncate(rqstp, cur_fh, open);
5624 if (status != nfs_ok)
5625 reset_union_bmap_deny(old_deny_bmap, stp);
5626 return status;
5627 }
5628
5629 /* Should we give out recallable state?: */
nfsd4_cb_channel_good(struct nfs4_client * clp)5630 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
5631 {
5632 if (clp->cl_cb_state == NFSD4_CB_UP)
5633 return true;
5634 /*
5635 * In the sessions case, since we don't have to establish a
5636 * separate connection for callbacks, we assume it's OK
5637 * until we hear otherwise:
5638 */
5639 return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
5640 }
5641
nfs4_alloc_init_lease(struct nfs4_delegation * dp,int flag)5642 static struct file_lease *nfs4_alloc_init_lease(struct nfs4_delegation *dp,
5643 int flag)
5644 {
5645 struct file_lease *fl;
5646
5647 fl = locks_alloc_lease();
5648 if (!fl)
5649 return NULL;
5650 fl->fl_lmops = &nfsd_lease_mng_ops;
5651 fl->c.flc_flags = FL_DELEG;
5652 fl->c.flc_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
5653 fl->c.flc_owner = (fl_owner_t)dp;
5654 fl->c.flc_pid = current->tgid;
5655 fl->c.flc_file = dp->dl_stid.sc_file->fi_deleg_file->nf_file;
5656 return fl;
5657 }
5658
nfsd4_check_conflicting_opens(struct nfs4_client * clp,struct nfs4_file * fp)5659 static int nfsd4_check_conflicting_opens(struct nfs4_client *clp,
5660 struct nfs4_file *fp)
5661 {
5662 struct nfs4_ol_stateid *st;
5663 struct file *f = fp->fi_deleg_file->nf_file;
5664 struct inode *ino = file_inode(f);
5665 int writes;
5666
5667 writes = atomic_read(&ino->i_writecount);
5668 if (!writes)
5669 return 0;
5670 /*
5671 * There could be multiple filehandles (hence multiple
5672 * nfs4_files) referencing this file, but that's not too
5673 * common; let's just give up in that case rather than
5674 * trying to go look up all the clients using that other
5675 * nfs4_file as well:
5676 */
5677 if (fp->fi_aliased)
5678 return -EAGAIN;
5679 /*
5680 * If there's a close in progress, make sure that we see it
5681 * clear any fi_fds[] entries before we see it decrement
5682 * i_writecount:
5683 */
5684 smp_mb__after_atomic();
5685
5686 if (fp->fi_fds[O_WRONLY])
5687 writes--;
5688 if (fp->fi_fds[O_RDWR])
5689 writes--;
5690 if (writes > 0)
5691 return -EAGAIN; /* There may be non-NFSv4 writers */
5692 /*
5693 * It's possible there are non-NFSv4 write opens in progress,
5694 * but if they haven't incremented i_writecount yet then they
5695 * also haven't called break lease yet; so, they'll break this
5696 * lease soon enough. So, all that's left to check for is NFSv4
5697 * opens:
5698 */
5699 spin_lock(&fp->fi_lock);
5700 list_for_each_entry(st, &fp->fi_stateids, st_perfile) {
5701 if (st->st_openstp == NULL /* it's an open */ &&
5702 access_permit_write(st) &&
5703 st->st_stid.sc_client != clp) {
5704 spin_unlock(&fp->fi_lock);
5705 return -EAGAIN;
5706 }
5707 }
5708 spin_unlock(&fp->fi_lock);
5709 /*
5710 * There's a small chance that we could be racing with another
5711 * NFSv4 open. However, any open that hasn't added itself to
5712 * the fi_stateids list also hasn't called break_lease yet; so,
5713 * they'll break this lease soon enough.
5714 */
5715 return 0;
5716 }
5717
5718 /*
5719 * It's possible that between opening the dentry and setting the delegation,
5720 * that it has been renamed or unlinked. Redo the lookup to verify that this
5721 * hasn't happened.
5722 */
5723 static int
nfsd4_verify_deleg_dentry(struct nfsd4_open * open,struct nfs4_file * fp,struct svc_fh * parent)5724 nfsd4_verify_deleg_dentry(struct nfsd4_open *open, struct nfs4_file *fp,
5725 struct svc_fh *parent)
5726 {
5727 struct svc_export *exp;
5728 struct dentry *child;
5729 __be32 err;
5730
5731 err = nfsd_lookup_dentry(open->op_rqstp, parent,
5732 open->op_fname, open->op_fnamelen,
5733 &exp, &child);
5734
5735 if (err)
5736 return -EAGAIN;
5737
5738 exp_put(exp);
5739 dput(child);
5740 if (child != file_dentry(fp->fi_deleg_file->nf_file))
5741 return -EAGAIN;
5742
5743 return 0;
5744 }
5745
5746 /*
5747 * We avoid breaking delegations held by a client due to its own activity, but
5748 * clearing setuid/setgid bits on a write is an implicit activity and the client
5749 * may not notice and continue using the old mode. Avoid giving out a delegation
5750 * on setuid/setgid files when the client is requesting an open for write.
5751 */
5752 static int
nfsd4_verify_setuid_write(struct nfsd4_open * open,struct nfsd_file * nf)5753 nfsd4_verify_setuid_write(struct nfsd4_open *open, struct nfsd_file *nf)
5754 {
5755 struct inode *inode = file_inode(nf->nf_file);
5756
5757 if ((open->op_share_access & NFS4_SHARE_ACCESS_WRITE) &&
5758 (inode->i_mode & (S_ISUID|S_ISGID)))
5759 return -EAGAIN;
5760 return 0;
5761 }
5762
5763 static struct nfs4_delegation *
nfs4_set_delegation(struct nfsd4_open * open,struct nfs4_ol_stateid * stp,struct svc_fh * parent)5764 nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5765 struct svc_fh *parent)
5766 {
5767 int status = 0;
5768 struct nfs4_client *clp = stp->st_stid.sc_client;
5769 struct nfs4_file *fp = stp->st_stid.sc_file;
5770 struct nfs4_clnt_odstate *odstate = stp->st_clnt_odstate;
5771 struct nfs4_delegation *dp;
5772 struct nfsd_file *nf = NULL;
5773 struct file_lease *fl;
5774 u32 dl_type;
5775
5776 /*
5777 * The fi_had_conflict and nfs_get_existing_delegation checks
5778 * here are just optimizations; we'll need to recheck them at
5779 * the end:
5780 */
5781 if (fp->fi_had_conflict)
5782 return ERR_PTR(-EAGAIN);
5783
5784 /*
5785 * Try for a write delegation first. RFC8881 section 10.4 says:
5786 *
5787 * "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
5788 * on its own, all opens."
5789 *
5790 * Furthermore the client can use a write delegation for most READ
5791 * operations as well, so we require a O_RDWR file here.
5792 *
5793 * Offer a write delegation in the case of a BOTH open, and ensure
5794 * we get the O_RDWR descriptor.
5795 */
5796 if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
5797 nf = find_rw_file(fp);
5798 dl_type = NFS4_OPEN_DELEGATE_WRITE;
5799 }
5800
5801 /*
5802 * If the file is being opened O_RDONLY or we couldn't get a O_RDWR
5803 * file for some reason, then try for a read delegation instead.
5804 */
5805 if (!nf && (open->op_share_access & NFS4_SHARE_ACCESS_READ)) {
5806 nf = find_readable_file(fp);
5807 dl_type = NFS4_OPEN_DELEGATE_READ;
5808 }
5809
5810 if (!nf)
5811 return ERR_PTR(-EAGAIN);
5812
5813 spin_lock(&state_lock);
5814 spin_lock(&fp->fi_lock);
5815 if (nfs4_delegation_exists(clp, fp))
5816 status = -EAGAIN;
5817 else if (nfsd4_verify_setuid_write(open, nf))
5818 status = -EAGAIN;
5819 else if (!fp->fi_deleg_file) {
5820 fp->fi_deleg_file = nf;
5821 /* increment early to prevent fi_deleg_file from being
5822 * cleared */
5823 fp->fi_delegees = 1;
5824 nf = NULL;
5825 } else
5826 fp->fi_delegees++;
5827 spin_unlock(&fp->fi_lock);
5828 spin_unlock(&state_lock);
5829 if (nf)
5830 nfsd_file_put(nf);
5831 if (status)
5832 return ERR_PTR(status);
5833
5834 status = -ENOMEM;
5835 dp = alloc_init_deleg(clp, fp, odstate, dl_type);
5836 if (!dp)
5837 goto out_delegees;
5838
5839 fl = nfs4_alloc_init_lease(dp, dl_type);
5840 if (!fl)
5841 goto out_clnt_odstate;
5842
5843 status = kernel_setlease(fp->fi_deleg_file->nf_file,
5844 fl->c.flc_type, &fl, NULL);
5845 if (fl)
5846 locks_free_lease(fl);
5847 if (status)
5848 goto out_clnt_odstate;
5849
5850 if (parent) {
5851 status = nfsd4_verify_deleg_dentry(open, fp, parent);
5852 if (status)
5853 goto out_unlock;
5854 }
5855
5856 status = nfsd4_check_conflicting_opens(clp, fp);
5857 if (status)
5858 goto out_unlock;
5859
5860 /*
5861 * Now that the deleg is set, check again to ensure that nothing
5862 * raced in and changed the mode while we weren't lookng.
5863 */
5864 status = nfsd4_verify_setuid_write(open, fp->fi_deleg_file);
5865 if (status)
5866 goto out_unlock;
5867
5868 status = -EAGAIN;
5869 if (fp->fi_had_conflict)
5870 goto out_unlock;
5871
5872 spin_lock(&state_lock);
5873 spin_lock(&clp->cl_lock);
5874 spin_lock(&fp->fi_lock);
5875 status = hash_delegation_locked(dp, fp);
5876 spin_unlock(&fp->fi_lock);
5877 spin_unlock(&clp->cl_lock);
5878 spin_unlock(&state_lock);
5879
5880 if (status)
5881 goto out_unlock;
5882
5883 return dp;
5884 out_unlock:
5885 kernel_setlease(fp->fi_deleg_file->nf_file, F_UNLCK, NULL, (void **)&dp);
5886 out_clnt_odstate:
5887 put_clnt_odstate(dp->dl_clnt_odstate);
5888 nfs4_put_stid(&dp->dl_stid);
5889 out_delegees:
5890 put_deleg_file(fp);
5891 return ERR_PTR(status);
5892 }
5893
nfsd4_open_deleg_none_ext(struct nfsd4_open * open,int status)5894 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
5895 {
5896 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
5897 if (status == -EAGAIN)
5898 open->op_why_no_deleg = WND4_CONTENTION;
5899 else {
5900 open->op_why_no_deleg = WND4_RESOURCE;
5901 switch (open->op_deleg_want) {
5902 case NFS4_SHARE_WANT_READ_DELEG:
5903 case NFS4_SHARE_WANT_WRITE_DELEG:
5904 case NFS4_SHARE_WANT_ANY_DELEG:
5905 break;
5906 case NFS4_SHARE_WANT_CANCEL:
5907 open->op_why_no_deleg = WND4_CANCELLED;
5908 break;
5909 case NFS4_SHARE_WANT_NO_DELEG:
5910 WARN_ON_ONCE(1);
5911 }
5912 }
5913 }
5914
5915 /*
5916 * The Linux NFS server does not offer write delegations to NFSv4.0
5917 * clients in order to avoid conflicts between write delegations and
5918 * GETATTRs requesting CHANGE or SIZE attributes.
5919 *
5920 * With NFSv4.1 and later minorversions, the SEQUENCE operation that
5921 * begins each COMPOUND contains a client ID. Delegation recall can
5922 * be avoided when the server recognizes the client sending a
5923 * GETATTR also holds write delegation it conflicts with.
5924 *
5925 * However, the NFSv4.0 protocol does not enable a server to
5926 * determine that a GETATTR originated from the client holding the
5927 * conflicting delegation versus coming from some other client. Per
5928 * RFC 7530 Section 16.7.5, the server must recall or send a
5929 * CB_GETATTR even when the GETATTR originates from the client that
5930 * holds the conflicting delegation.
5931 *
5932 * An NFSv4.0 client can trigger a pathological situation if it
5933 * always sends a DELEGRETURN preceded by a conflicting GETATTR in
5934 * the same COMPOUND. COMPOUND execution will always stop at the
5935 * GETATTR and the DELEGRETURN will never get executed. The server
5936 * eventually revokes the delegation, which can result in loss of
5937 * open or lock state.
5938 */
5939 static void
nfs4_open_delegation(struct nfsd4_open * open,struct nfs4_ol_stateid * stp,struct svc_fh * currentfh)5940 nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
5941 struct svc_fh *currentfh)
5942 {
5943 struct nfs4_delegation *dp;
5944 struct nfs4_openowner *oo = openowner(stp->st_stateowner);
5945 struct nfs4_client *clp = stp->st_stid.sc_client;
5946 struct svc_fh *parent = NULL;
5947 int cb_up;
5948 int status = 0;
5949 struct kstat stat;
5950 struct path path;
5951
5952 cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
5953 open->op_recall = false;
5954 switch (open->op_claim_type) {
5955 case NFS4_OPEN_CLAIM_PREVIOUS:
5956 if (!cb_up)
5957 open->op_recall = true;
5958 break;
5959 case NFS4_OPEN_CLAIM_NULL:
5960 parent = currentfh;
5961 fallthrough;
5962 case NFS4_OPEN_CLAIM_FH:
5963 /*
5964 * Let's not give out any delegations till everyone's
5965 * had the chance to reclaim theirs, *and* until
5966 * NLM locks have all been reclaimed:
5967 */
5968 if (locks_in_grace(clp->net))
5969 goto out_no_deleg;
5970 if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
5971 goto out_no_deleg;
5972 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE &&
5973 !clp->cl_minorversion)
5974 goto out_no_deleg;
5975 break;
5976 default:
5977 goto out_no_deleg;
5978 }
5979 dp = nfs4_set_delegation(open, stp, parent);
5980 if (IS_ERR(dp))
5981 goto out_no_deleg;
5982
5983 memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
5984
5985 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
5986 open->op_delegate_type = NFS4_OPEN_DELEGATE_WRITE;
5987 trace_nfsd_deleg_write(&dp->dl_stid.sc_stateid);
5988 path.mnt = currentfh->fh_export->ex_path.mnt;
5989 path.dentry = currentfh->fh_dentry;
5990 if (vfs_getattr(&path, &stat,
5991 (STATX_SIZE | STATX_CTIME | STATX_CHANGE_COOKIE),
5992 AT_STATX_SYNC_AS_STAT)) {
5993 nfs4_put_stid(&dp->dl_stid);
5994 destroy_delegation(dp);
5995 goto out_no_deleg;
5996 }
5997 dp->dl_cb_fattr.ncf_cur_fsize = stat.size;
5998 dp->dl_cb_fattr.ncf_initial_cinfo =
5999 nfsd4_change_attribute(&stat, d_inode(currentfh->fh_dentry));
6000 } else {
6001 open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
6002 trace_nfsd_deleg_read(&dp->dl_stid.sc_stateid);
6003 }
6004 nfs4_put_stid(&dp->dl_stid);
6005 return;
6006 out_no_deleg:
6007 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
6008 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
6009 open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
6010 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
6011 open->op_recall = true;
6012 }
6013
6014 /* 4.1 client asking for a delegation? */
6015 if (open->op_deleg_want)
6016 nfsd4_open_deleg_none_ext(open, status);
6017 return;
6018 }
6019
nfsd4_deleg_xgrade_none_ext(struct nfsd4_open * open,struct nfs4_delegation * dp)6020 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
6021 struct nfs4_delegation *dp)
6022 {
6023 if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
6024 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6025 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6026 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
6027 } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
6028 dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
6029 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6030 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
6031 }
6032 /* Otherwise the client must be confused wanting a delegation
6033 * it already has, therefore we don't return
6034 * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
6035 */
6036 }
6037
6038 /**
6039 * nfsd4_process_open2 - finish open processing
6040 * @rqstp: the RPC transaction being executed
6041 * @current_fh: NFSv4 COMPOUND's current filehandle
6042 * @open: OPEN arguments
6043 *
6044 * If successful, (1) truncate the file if open->op_truncate was
6045 * set, (2) set open->op_stateid, (3) set open->op_delegation.
6046 *
6047 * Returns %nfs_ok on success; otherwise an nfs4stat value in
6048 * network byte order is returned.
6049 */
6050 __be32
nfsd4_process_open2(struct svc_rqst * rqstp,struct svc_fh * current_fh,struct nfsd4_open * open)6051 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
6052 {
6053 struct nfsd4_compoundres *resp = rqstp->rq_resp;
6054 struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
6055 struct nfs4_file *fp = NULL;
6056 struct nfs4_ol_stateid *stp = NULL;
6057 struct nfs4_delegation *dp = NULL;
6058 __be32 status;
6059 bool new_stp = false;
6060
6061 /*
6062 * Lookup file; if found, lookup stateid and check open request,
6063 * and check for delegations in the process of being recalled.
6064 * If not found, create the nfs4_file struct
6065 */
6066 fp = nfsd4_file_hash_insert(open->op_file, current_fh);
6067 if (unlikely(!fp))
6068 return nfserr_jukebox;
6069 if (fp != open->op_file) {
6070 status = nfs4_check_deleg(cl, open, &dp);
6071 if (status)
6072 goto out;
6073 stp = nfsd4_find_and_lock_existing_open(fp, open);
6074 } else {
6075 open->op_file = NULL;
6076 status = nfserr_bad_stateid;
6077 if (nfsd4_is_deleg_cur(open))
6078 goto out;
6079 }
6080
6081 if (!stp) {
6082 stp = init_open_stateid(fp, open);
6083 if (!open->op_stp)
6084 new_stp = true;
6085 }
6086
6087 /*
6088 * OPEN the file, or upgrade an existing OPEN.
6089 * If truncate fails, the OPEN fails.
6090 *
6091 * stp is already locked.
6092 */
6093 if (!new_stp) {
6094 /* Stateid was found, this is an OPEN upgrade */
6095 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
6096 if (status) {
6097 mutex_unlock(&stp->st_mutex);
6098 goto out;
6099 }
6100 } else {
6101 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open, true);
6102 if (status) {
6103 release_open_stateid(stp);
6104 mutex_unlock(&stp->st_mutex);
6105 goto out;
6106 }
6107
6108 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
6109 open->op_odstate);
6110 if (stp->st_clnt_odstate == open->op_odstate)
6111 open->op_odstate = NULL;
6112 }
6113
6114 nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
6115 mutex_unlock(&stp->st_mutex);
6116
6117 if (nfsd4_has_session(&resp->cstate)) {
6118 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
6119 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
6120 open->op_why_no_deleg = WND4_NOT_WANTED;
6121 goto nodeleg;
6122 }
6123 }
6124
6125 /*
6126 * Attempt to hand out a delegation. No error return, because the
6127 * OPEN succeeds even if we fail.
6128 */
6129 nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
6130 nodeleg:
6131 status = nfs_ok;
6132 trace_nfsd_open(&stp->st_stid.sc_stateid);
6133 out:
6134 /* 4.1 client trying to upgrade/downgrade delegation? */
6135 if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
6136 open->op_deleg_want)
6137 nfsd4_deleg_xgrade_none_ext(open, dp);
6138
6139 if (fp)
6140 put_nfs4_file(fp);
6141 if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
6142 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
6143 /*
6144 * To finish the open response, we just need to set the rflags.
6145 */
6146 open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
6147 if (nfsd4_has_session(&resp->cstate))
6148 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
6149 else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
6150 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
6151
6152 if (dp)
6153 nfs4_put_stid(&dp->dl_stid);
6154 if (stp)
6155 nfs4_put_stid(&stp->st_stid);
6156
6157 return status;
6158 }
6159
nfsd4_cleanup_open_state(struct nfsd4_compound_state * cstate,struct nfsd4_open * open)6160 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
6161 struct nfsd4_open *open)
6162 {
6163 if (open->op_openowner)
6164 nfs4_put_stateowner(&open->op_openowner->oo_owner);
6165 if (open->op_file)
6166 kmem_cache_free(file_slab, open->op_file);
6167 if (open->op_stp)
6168 nfs4_put_stid(&open->op_stp->st_stid);
6169 if (open->op_odstate)
6170 kmem_cache_free(odstate_slab, open->op_odstate);
6171 }
6172
6173 __be32
nfsd4_renew(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)6174 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6175 union nfsd4_op_u *u)
6176 {
6177 clientid_t *clid = &u->renew;
6178 struct nfs4_client *clp;
6179 __be32 status;
6180 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6181
6182 trace_nfsd_clid_renew(clid);
6183 status = set_client(clid, cstate, nn);
6184 if (status)
6185 return status;
6186 clp = cstate->clp;
6187 if (!list_empty(&clp->cl_delegations)
6188 && clp->cl_cb_state != NFSD4_CB_UP)
6189 return nfserr_cb_path_down;
6190 return nfs_ok;
6191 }
6192
6193 void
nfsd4_end_grace(struct nfsd_net * nn)6194 nfsd4_end_grace(struct nfsd_net *nn)
6195 {
6196 /* do nothing if grace period already ended */
6197 if (nn->grace_ended)
6198 return;
6199
6200 trace_nfsd_grace_complete(nn);
6201 nn->grace_ended = true;
6202 /*
6203 * If the server goes down again right now, an NFSv4
6204 * client will still be allowed to reclaim after it comes back up,
6205 * even if it hasn't yet had a chance to reclaim state this time.
6206 *
6207 */
6208 nfsd4_record_grace_done(nn);
6209 /*
6210 * At this point, NFSv4 clients can still reclaim. But if the
6211 * server crashes, any that have not yet reclaimed will be out
6212 * of luck on the next boot.
6213 *
6214 * (NFSv4.1+ clients are considered to have reclaimed once they
6215 * call RECLAIM_COMPLETE. NFSv4.0 clients are considered to
6216 * have reclaimed after their first OPEN.)
6217 */
6218 locks_end_grace(&nn->nfsd4_manager);
6219 /*
6220 * At this point, and once lockd and/or any other containers
6221 * exit their grace period, further reclaims will fail and
6222 * regular locking can resume.
6223 */
6224 }
6225
6226 /*
6227 * If we've waited a lease period but there are still clients trying to
6228 * reclaim, wait a little longer to give them a chance to finish.
6229 */
clients_still_reclaiming(struct nfsd_net * nn)6230 static bool clients_still_reclaiming(struct nfsd_net *nn)
6231 {
6232 time64_t double_grace_period_end = nn->boot_time +
6233 2 * nn->nfsd4_lease;
6234
6235 if (nn->track_reclaim_completes &&
6236 atomic_read(&nn->nr_reclaim_complete) ==
6237 nn->reclaim_str_hashtbl_size)
6238 return false;
6239 if (!nn->somebody_reclaimed)
6240 return false;
6241 nn->somebody_reclaimed = false;
6242 /*
6243 * If we've given them *two* lease times to reclaim, and they're
6244 * still not done, give up:
6245 */
6246 if (ktime_get_boottime_seconds() > double_grace_period_end)
6247 return false;
6248 return true;
6249 }
6250
6251 struct laundry_time {
6252 time64_t cutoff;
6253 time64_t new_timeo;
6254 };
6255
state_expired(struct laundry_time * lt,time64_t last_refresh)6256 static bool state_expired(struct laundry_time *lt, time64_t last_refresh)
6257 {
6258 time64_t time_remaining;
6259
6260 if (last_refresh < lt->cutoff)
6261 return true;
6262 time_remaining = last_refresh - lt->cutoff;
6263 lt->new_timeo = min(lt->new_timeo, time_remaining);
6264 return false;
6265 }
6266
6267 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
nfsd4_ssc_init_umount_work(struct nfsd_net * nn)6268 void nfsd4_ssc_init_umount_work(struct nfsd_net *nn)
6269 {
6270 spin_lock_init(&nn->nfsd_ssc_lock);
6271 INIT_LIST_HEAD(&nn->nfsd_ssc_mount_list);
6272 init_waitqueue_head(&nn->nfsd_ssc_waitq);
6273 }
6274 EXPORT_SYMBOL_GPL(nfsd4_ssc_init_umount_work);
6275
6276 /*
6277 * This is called when nfsd is being shutdown, after all inter_ssc
6278 * cleanup were done, to destroy the ssc delayed unmount list.
6279 */
nfsd4_ssc_shutdown_umount(struct nfsd_net * nn)6280 static void nfsd4_ssc_shutdown_umount(struct nfsd_net *nn)
6281 {
6282 struct nfsd4_ssc_umount_item *ni = NULL;
6283 struct nfsd4_ssc_umount_item *tmp;
6284
6285 spin_lock(&nn->nfsd_ssc_lock);
6286 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6287 list_del(&ni->nsui_list);
6288 spin_unlock(&nn->nfsd_ssc_lock);
6289 mntput(ni->nsui_vfsmount);
6290 kfree(ni);
6291 spin_lock(&nn->nfsd_ssc_lock);
6292 }
6293 spin_unlock(&nn->nfsd_ssc_lock);
6294 }
6295
nfsd4_ssc_expire_umount(struct nfsd_net * nn)6296 static void nfsd4_ssc_expire_umount(struct nfsd_net *nn)
6297 {
6298 bool do_wakeup = false;
6299 struct nfsd4_ssc_umount_item *ni = NULL;
6300 struct nfsd4_ssc_umount_item *tmp;
6301
6302 spin_lock(&nn->nfsd_ssc_lock);
6303 list_for_each_entry_safe(ni, tmp, &nn->nfsd_ssc_mount_list, nsui_list) {
6304 if (time_after(jiffies, ni->nsui_expire)) {
6305 if (refcount_read(&ni->nsui_refcnt) > 1)
6306 continue;
6307
6308 /* mark being unmount */
6309 ni->nsui_busy = true;
6310 spin_unlock(&nn->nfsd_ssc_lock);
6311 mntput(ni->nsui_vfsmount);
6312 spin_lock(&nn->nfsd_ssc_lock);
6313
6314 /* waiters need to start from begin of list */
6315 list_del(&ni->nsui_list);
6316 kfree(ni);
6317
6318 /* wakeup ssc_connect waiters */
6319 do_wakeup = true;
6320 continue;
6321 }
6322 break;
6323 }
6324 if (do_wakeup)
6325 wake_up_all(&nn->nfsd_ssc_waitq);
6326 spin_unlock(&nn->nfsd_ssc_lock);
6327 }
6328 #endif
6329
6330 /* Check if any lock belonging to this lockowner has any blockers */
6331 static bool
nfs4_lockowner_has_blockers(struct nfs4_lockowner * lo)6332 nfs4_lockowner_has_blockers(struct nfs4_lockowner *lo)
6333 {
6334 struct file_lock_context *ctx;
6335 struct nfs4_ol_stateid *stp;
6336 struct nfs4_file *nf;
6337
6338 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
6339 nf = stp->st_stid.sc_file;
6340 ctx = locks_inode_context(nf->fi_inode);
6341 if (!ctx)
6342 continue;
6343 if (locks_owner_has_blockers(ctx, lo))
6344 return true;
6345 }
6346 return false;
6347 }
6348
6349 static bool
nfs4_anylock_blockers(struct nfs4_client * clp)6350 nfs4_anylock_blockers(struct nfs4_client *clp)
6351 {
6352 int i;
6353 struct nfs4_stateowner *so;
6354 struct nfs4_lockowner *lo;
6355
6356 if (atomic_read(&clp->cl_delegs_in_recall))
6357 return true;
6358 spin_lock(&clp->cl_lock);
6359 for (i = 0; i < OWNER_HASH_SIZE; i++) {
6360 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[i],
6361 so_strhash) {
6362 if (so->so_is_open_owner)
6363 continue;
6364 lo = lockowner(so);
6365 if (nfs4_lockowner_has_blockers(lo)) {
6366 spin_unlock(&clp->cl_lock);
6367 return true;
6368 }
6369 }
6370 }
6371 spin_unlock(&clp->cl_lock);
6372 return false;
6373 }
6374
6375 static void
nfs4_get_client_reaplist(struct nfsd_net * nn,struct list_head * reaplist,struct laundry_time * lt)6376 nfs4_get_client_reaplist(struct nfsd_net *nn, struct list_head *reaplist,
6377 struct laundry_time *lt)
6378 {
6379 unsigned int maxreap, reapcnt = 0;
6380 struct list_head *pos, *next;
6381 struct nfs4_client *clp;
6382
6383 maxreap = (atomic_read(&nn->nfs4_client_count) >= nn->nfs4_max_clients) ?
6384 NFSD_CLIENT_MAX_TRIM_PER_RUN : 0;
6385 INIT_LIST_HEAD(reaplist);
6386 spin_lock(&nn->client_lock);
6387 list_for_each_safe(pos, next, &nn->client_lru) {
6388 clp = list_entry(pos, struct nfs4_client, cl_lru);
6389 if (clp->cl_state == NFSD4_EXPIRABLE)
6390 goto exp_client;
6391 if (!state_expired(lt, clp->cl_time))
6392 break;
6393 if (!atomic_read(&clp->cl_rpc_users)) {
6394 if (clp->cl_state == NFSD4_ACTIVE)
6395 atomic_inc(&nn->nfsd_courtesy_clients);
6396 clp->cl_state = NFSD4_COURTESY;
6397 }
6398 if (!client_has_state(clp))
6399 goto exp_client;
6400 if (!nfs4_anylock_blockers(clp))
6401 if (reapcnt >= maxreap)
6402 continue;
6403 exp_client:
6404 if (!mark_client_expired_locked(clp)) {
6405 list_add(&clp->cl_lru, reaplist);
6406 reapcnt++;
6407 }
6408 }
6409 spin_unlock(&nn->client_lock);
6410 }
6411
6412 static void
nfs4_get_courtesy_client_reaplist(struct nfsd_net * nn,struct list_head * reaplist)6413 nfs4_get_courtesy_client_reaplist(struct nfsd_net *nn,
6414 struct list_head *reaplist)
6415 {
6416 unsigned int maxreap = 0, reapcnt = 0;
6417 struct list_head *pos, *next;
6418 struct nfs4_client *clp;
6419
6420 maxreap = NFSD_CLIENT_MAX_TRIM_PER_RUN;
6421 INIT_LIST_HEAD(reaplist);
6422
6423 spin_lock(&nn->client_lock);
6424 list_for_each_safe(pos, next, &nn->client_lru) {
6425 clp = list_entry(pos, struct nfs4_client, cl_lru);
6426 if (clp->cl_state == NFSD4_ACTIVE)
6427 break;
6428 if (reapcnt >= maxreap)
6429 break;
6430 if (!mark_client_expired_locked(clp)) {
6431 list_add(&clp->cl_lru, reaplist);
6432 reapcnt++;
6433 }
6434 }
6435 spin_unlock(&nn->client_lock);
6436 }
6437
6438 static void
nfs4_process_client_reaplist(struct list_head * reaplist)6439 nfs4_process_client_reaplist(struct list_head *reaplist)
6440 {
6441 struct list_head *pos, *next;
6442 struct nfs4_client *clp;
6443
6444 list_for_each_safe(pos, next, reaplist) {
6445 clp = list_entry(pos, struct nfs4_client, cl_lru);
6446 trace_nfsd_clid_purged(&clp->cl_clientid);
6447 list_del_init(&clp->cl_lru);
6448 expire_client(clp);
6449 }
6450 }
6451
nfs40_clean_admin_revoked(struct nfsd_net * nn,struct laundry_time * lt)6452 static void nfs40_clean_admin_revoked(struct nfsd_net *nn,
6453 struct laundry_time *lt)
6454 {
6455 struct nfs4_client *clp;
6456
6457 spin_lock(&nn->client_lock);
6458 if (nn->nfs40_last_revoke == 0 ||
6459 nn->nfs40_last_revoke > lt->cutoff) {
6460 spin_unlock(&nn->client_lock);
6461 return;
6462 }
6463 nn->nfs40_last_revoke = 0;
6464
6465 retry:
6466 list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6467 unsigned long id, tmp;
6468 struct nfs4_stid *stid;
6469
6470 if (atomic_read(&clp->cl_admin_revoked) == 0)
6471 continue;
6472
6473 spin_lock(&clp->cl_lock);
6474 idr_for_each_entry_ul(&clp->cl_stateids, stid, tmp, id)
6475 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6476 refcount_inc(&stid->sc_count);
6477 spin_unlock(&nn->client_lock);
6478 /* this function drops ->cl_lock */
6479 nfsd4_drop_revoked_stid(stid);
6480 nfs4_put_stid(stid);
6481 spin_lock(&nn->client_lock);
6482 goto retry;
6483 }
6484 spin_unlock(&clp->cl_lock);
6485 }
6486 spin_unlock(&nn->client_lock);
6487 }
6488
6489 static time64_t
nfs4_laundromat(struct nfsd_net * nn)6490 nfs4_laundromat(struct nfsd_net *nn)
6491 {
6492 struct nfs4_openowner *oo;
6493 struct nfs4_delegation *dp;
6494 struct nfs4_ol_stateid *stp;
6495 struct nfsd4_blocked_lock *nbl;
6496 struct list_head *pos, *next, reaplist;
6497 struct laundry_time lt = {
6498 .cutoff = ktime_get_boottime_seconds() - nn->nfsd4_lease,
6499 .new_timeo = nn->nfsd4_lease
6500 };
6501 struct nfs4_cpntf_state *cps;
6502 copy_stateid_t *cps_t;
6503 int i;
6504
6505 if (clients_still_reclaiming(nn)) {
6506 lt.new_timeo = 0;
6507 goto out;
6508 }
6509 nfsd4_end_grace(nn);
6510
6511 spin_lock(&nn->s2s_cp_lock);
6512 idr_for_each_entry(&nn->s2s_cp_stateids, cps_t, i) {
6513 cps = container_of(cps_t, struct nfs4_cpntf_state, cp_stateid);
6514 if (cps->cp_stateid.cs_type == NFS4_COPYNOTIFY_STID &&
6515 state_expired(<, cps->cpntf_time))
6516 _free_cpntf_state_locked(nn, cps);
6517 }
6518 spin_unlock(&nn->s2s_cp_lock);
6519 nfs4_get_client_reaplist(nn, &reaplist, <);
6520 nfs4_process_client_reaplist(&reaplist);
6521
6522 nfs40_clean_admin_revoked(nn, <);
6523
6524 spin_lock(&state_lock);
6525 list_for_each_safe(pos, next, &nn->del_recall_lru) {
6526 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6527 if (!state_expired(<, dp->dl_time))
6528 break;
6529 unhash_delegation_locked(dp, SC_STATUS_REVOKED);
6530 list_add(&dp->dl_recall_lru, &reaplist);
6531 }
6532 spin_unlock(&state_lock);
6533 while (!list_empty(&reaplist)) {
6534 dp = list_first_entry(&reaplist, struct nfs4_delegation,
6535 dl_recall_lru);
6536 list_del_init(&dp->dl_recall_lru);
6537 revoke_delegation(dp);
6538 }
6539
6540 spin_lock(&nn->client_lock);
6541 while (!list_empty(&nn->close_lru)) {
6542 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
6543 oo_close_lru);
6544 if (!state_expired(<, oo->oo_time))
6545 break;
6546 list_del_init(&oo->oo_close_lru);
6547 stp = oo->oo_last_closed_stid;
6548 oo->oo_last_closed_stid = NULL;
6549 spin_unlock(&nn->client_lock);
6550 nfs4_put_stid(&stp->st_stid);
6551 spin_lock(&nn->client_lock);
6552 }
6553 spin_unlock(&nn->client_lock);
6554
6555 /*
6556 * It's possible for a client to try and acquire an already held lock
6557 * that is being held for a long time, and then lose interest in it.
6558 * So, we clean out any un-revisited request after a lease period
6559 * under the assumption that the client is no longer interested.
6560 *
6561 * RFC5661, sec. 9.6 states that the client must not rely on getting
6562 * notifications and must continue to poll for locks, even when the
6563 * server supports them. Thus this shouldn't lead to clients blocking
6564 * indefinitely once the lock does become free.
6565 */
6566 BUG_ON(!list_empty(&reaplist));
6567 spin_lock(&nn->blocked_locks_lock);
6568 while (!list_empty(&nn->blocked_locks_lru)) {
6569 nbl = list_first_entry(&nn->blocked_locks_lru,
6570 struct nfsd4_blocked_lock, nbl_lru);
6571 if (!state_expired(<, nbl->nbl_time))
6572 break;
6573 list_move(&nbl->nbl_lru, &reaplist);
6574 list_del_init(&nbl->nbl_list);
6575 }
6576 spin_unlock(&nn->blocked_locks_lock);
6577
6578 while (!list_empty(&reaplist)) {
6579 nbl = list_first_entry(&reaplist,
6580 struct nfsd4_blocked_lock, nbl_lru);
6581 list_del_init(&nbl->nbl_lru);
6582 free_blocked_lock(nbl);
6583 }
6584 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
6585 /* service the server-to-server copy delayed unmount list */
6586 nfsd4_ssc_expire_umount(nn);
6587 #endif
6588 if (atomic_long_read(&num_delegations) >= max_delegations)
6589 deleg_reaper(nn);
6590 out:
6591 return max_t(time64_t, lt.new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
6592 }
6593
6594 static void laundromat_main(struct work_struct *);
6595
6596 static void
laundromat_main(struct work_struct * laundry)6597 laundromat_main(struct work_struct *laundry)
6598 {
6599 time64_t t;
6600 struct delayed_work *dwork = to_delayed_work(laundry);
6601 struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
6602 laundromat_work);
6603
6604 t = nfs4_laundromat(nn);
6605 queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
6606 }
6607
6608 static void
courtesy_client_reaper(struct nfsd_net * nn)6609 courtesy_client_reaper(struct nfsd_net *nn)
6610 {
6611 struct list_head reaplist;
6612
6613 nfs4_get_courtesy_client_reaplist(nn, &reaplist);
6614 nfs4_process_client_reaplist(&reaplist);
6615 }
6616
6617 static void
deleg_reaper(struct nfsd_net * nn)6618 deleg_reaper(struct nfsd_net *nn)
6619 {
6620 struct list_head *pos, *next;
6621 struct nfs4_client *clp;
6622 struct list_head cblist;
6623
6624 INIT_LIST_HEAD(&cblist);
6625 spin_lock(&nn->client_lock);
6626 list_for_each_safe(pos, next, &nn->client_lru) {
6627 clp = list_entry(pos, struct nfs4_client, cl_lru);
6628 if (clp->cl_state != NFSD4_ACTIVE ||
6629 list_empty(&clp->cl_delegations) ||
6630 atomic_read(&clp->cl_delegs_in_recall) ||
6631 test_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags) ||
6632 (ktime_get_boottime_seconds() -
6633 clp->cl_ra_time < 5)) {
6634 continue;
6635 }
6636 list_add(&clp->cl_ra_cblist, &cblist);
6637
6638 /* release in nfsd4_cb_recall_any_release */
6639 kref_get(&clp->cl_nfsdfs.cl_ref);
6640 set_bit(NFSD4_CLIENT_CB_RECALL_ANY, &clp->cl_flags);
6641 clp->cl_ra_time = ktime_get_boottime_seconds();
6642 }
6643 spin_unlock(&nn->client_lock);
6644
6645 while (!list_empty(&cblist)) {
6646 clp = list_first_entry(&cblist, struct nfs4_client,
6647 cl_ra_cblist);
6648 list_del_init(&clp->cl_ra_cblist);
6649 clp->cl_ra->ra_keep = 0;
6650 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG);
6651 clp->cl_ra->ra_bmval[0] = BIT(RCA4_TYPE_MASK_RDATA_DLG) |
6652 BIT(RCA4_TYPE_MASK_WDATA_DLG);
6653 trace_nfsd_cb_recall_any(clp->cl_ra);
6654 nfsd4_run_cb(&clp->cl_ra->ra_cb);
6655 }
6656 }
6657
6658 static void
nfsd4_state_shrinker_worker(struct work_struct * work)6659 nfsd4_state_shrinker_worker(struct work_struct *work)
6660 {
6661 struct nfsd_net *nn = container_of(work, struct nfsd_net,
6662 nfsd_shrinker_work);
6663
6664 courtesy_client_reaper(nn);
6665 deleg_reaper(nn);
6666 }
6667
nfs4_check_fh(struct svc_fh * fhp,struct nfs4_stid * stp)6668 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
6669 {
6670 if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
6671 return nfserr_bad_stateid;
6672 return nfs_ok;
6673 }
6674
6675 static
nfs4_check_openmode(struct nfs4_ol_stateid * stp,int flags)6676 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
6677 {
6678 __be32 status = nfserr_openmode;
6679
6680 /* For lock stateid's, we test the parent open, not the lock: */
6681 if (stp->st_openstp)
6682 stp = stp->st_openstp;
6683 if ((flags & WR_STATE) && !access_permit_write(stp))
6684 goto out;
6685 if ((flags & RD_STATE) && !access_permit_read(stp))
6686 goto out;
6687 status = nfs_ok;
6688 out:
6689 return status;
6690 }
6691
6692 static inline __be32
check_special_stateids(struct net * net,svc_fh * current_fh,stateid_t * stateid,int flags)6693 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
6694 {
6695 if (ONE_STATEID(stateid) && (flags & RD_STATE))
6696 return nfs_ok;
6697 else if (opens_in_grace(net)) {
6698 /* Answer in remaining cases depends on existence of
6699 * conflicting state; so we must wait out the grace period. */
6700 return nfserr_grace;
6701 } else if (flags & WR_STATE)
6702 return nfs4_share_conflict(current_fh,
6703 NFS4_SHARE_DENY_WRITE);
6704 else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
6705 return nfs4_share_conflict(current_fh,
6706 NFS4_SHARE_DENY_READ);
6707 }
6708
check_stateid_generation(stateid_t * in,stateid_t * ref,bool has_session)6709 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
6710 {
6711 /*
6712 * When sessions are used the stateid generation number is ignored
6713 * when it is zero.
6714 */
6715 if (has_session && in->si_generation == 0)
6716 return nfs_ok;
6717
6718 if (in->si_generation == ref->si_generation)
6719 return nfs_ok;
6720
6721 /* If the client sends us a stateid from the future, it's buggy: */
6722 if (nfsd4_stateid_generation_after(in, ref))
6723 return nfserr_bad_stateid;
6724 /*
6725 * However, we could see a stateid from the past, even from a
6726 * non-buggy client. For example, if the client sends a lock
6727 * while some IO is outstanding, the lock may bump si_generation
6728 * while the IO is still in flight. The client could avoid that
6729 * situation by waiting for responses on all the IO requests,
6730 * but better performance may result in retrying IO that
6731 * receives an old_stateid error if requests are rarely
6732 * reordered in flight:
6733 */
6734 return nfserr_old_stateid;
6735 }
6736
nfsd4_stid_check_stateid_generation(stateid_t * in,struct nfs4_stid * s,bool has_session)6737 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
6738 {
6739 __be32 ret;
6740
6741 spin_lock(&s->sc_lock);
6742 ret = nfsd4_verify_open_stid(s);
6743 if (ret == nfs_ok)
6744 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
6745 spin_unlock(&s->sc_lock);
6746 if (ret == nfserr_admin_revoked)
6747 nfsd40_drop_revoked_stid(s->sc_client,
6748 &s->sc_stateid);
6749 return ret;
6750 }
6751
nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid * ols)6752 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
6753 {
6754 if (ols->st_stateowner->so_is_open_owner &&
6755 !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
6756 return nfserr_bad_stateid;
6757 return nfs_ok;
6758 }
6759
nfsd4_validate_stateid(struct nfs4_client * cl,stateid_t * stateid)6760 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
6761 {
6762 struct nfs4_stid *s;
6763 __be32 status = nfserr_bad_stateid;
6764
6765 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6766 CLOSE_STATEID(stateid))
6767 return status;
6768 spin_lock(&cl->cl_lock);
6769 s = find_stateid_locked(cl, stateid);
6770 if (!s)
6771 goto out_unlock;
6772 status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
6773 if (status)
6774 goto out_unlock;
6775 status = nfsd4_verify_open_stid(s);
6776 if (status)
6777 goto out_unlock;
6778
6779 switch (s->sc_type) {
6780 case SC_TYPE_DELEG:
6781 status = nfs_ok;
6782 break;
6783 case SC_TYPE_OPEN:
6784 case SC_TYPE_LOCK:
6785 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
6786 break;
6787 default:
6788 printk("unknown stateid type %x\n", s->sc_type);
6789 status = nfserr_bad_stateid;
6790 }
6791 out_unlock:
6792 spin_unlock(&cl->cl_lock);
6793 if (status == nfserr_admin_revoked)
6794 nfsd40_drop_revoked_stid(cl, stateid);
6795 return status;
6796 }
6797
6798 __be32
nfsd4_lookup_stateid(struct nfsd4_compound_state * cstate,stateid_t * stateid,unsigned short typemask,unsigned short statusmask,struct nfs4_stid ** s,struct nfsd_net * nn)6799 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
6800 stateid_t *stateid,
6801 unsigned short typemask, unsigned short statusmask,
6802 struct nfs4_stid **s, struct nfsd_net *nn)
6803 {
6804 __be32 status;
6805 struct nfs4_stid *stid;
6806 bool return_revoked = false;
6807
6808 /*
6809 * only return revoked delegations if explicitly asked.
6810 * otherwise we report revoked or bad_stateid status.
6811 */
6812 if (statusmask & SC_STATUS_REVOKED)
6813 return_revoked = true;
6814 if (typemask & SC_TYPE_DELEG)
6815 /* Always allow REVOKED for DELEG so we can
6816 * retturn the appropriate error.
6817 */
6818 statusmask |= SC_STATUS_REVOKED;
6819
6820 statusmask |= SC_STATUS_ADMIN_REVOKED;
6821
6822 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
6823 CLOSE_STATEID(stateid))
6824 return nfserr_bad_stateid;
6825 status = set_client(&stateid->si_opaque.so_clid, cstate, nn);
6826 if (status == nfserr_stale_clientid) {
6827 if (cstate->session)
6828 return nfserr_bad_stateid;
6829 return nfserr_stale_stateid;
6830 }
6831 if (status)
6832 return status;
6833 stid = find_stateid_by_type(cstate->clp, stateid, typemask, statusmask);
6834 if (!stid)
6835 return nfserr_bad_stateid;
6836 if ((stid->sc_status & SC_STATUS_REVOKED) && !return_revoked) {
6837 nfs4_put_stid(stid);
6838 return nfserr_deleg_revoked;
6839 }
6840 if (stid->sc_status & SC_STATUS_ADMIN_REVOKED) {
6841 nfsd40_drop_revoked_stid(cstate->clp, stateid);
6842 nfs4_put_stid(stid);
6843 return nfserr_admin_revoked;
6844 }
6845 *s = stid;
6846 return nfs_ok;
6847 }
6848
6849 static struct nfsd_file *
nfs4_find_file(struct nfs4_stid * s,int flags)6850 nfs4_find_file(struct nfs4_stid *s, int flags)
6851 {
6852 struct nfsd_file *ret = NULL;
6853
6854 if (!s || s->sc_status)
6855 return NULL;
6856
6857 switch (s->sc_type) {
6858 case SC_TYPE_DELEG:
6859 spin_lock(&s->sc_file->fi_lock);
6860 ret = nfsd_file_get(s->sc_file->fi_deleg_file);
6861 spin_unlock(&s->sc_file->fi_lock);
6862 break;
6863 case SC_TYPE_OPEN:
6864 case SC_TYPE_LOCK:
6865 if (flags & RD_STATE)
6866 ret = find_readable_file(s->sc_file);
6867 else
6868 ret = find_writeable_file(s->sc_file);
6869 }
6870
6871 return ret;
6872 }
6873
6874 static __be32
nfs4_check_olstateid(struct nfs4_ol_stateid * ols,int flags)6875 nfs4_check_olstateid(struct nfs4_ol_stateid *ols, int flags)
6876 {
6877 __be32 status;
6878
6879 status = nfsd4_check_openowner_confirmed(ols);
6880 if (status)
6881 return status;
6882 return nfs4_check_openmode(ols, flags);
6883 }
6884
6885 static __be32
nfs4_check_file(struct svc_rqst * rqstp,struct svc_fh * fhp,struct nfs4_stid * s,struct nfsd_file ** nfp,int flags)6886 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
6887 struct nfsd_file **nfp, int flags)
6888 {
6889 int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
6890 struct nfsd_file *nf;
6891 __be32 status;
6892
6893 nf = nfs4_find_file(s, flags);
6894 if (nf) {
6895 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
6896 acc | NFSD_MAY_OWNER_OVERRIDE);
6897 if (status) {
6898 nfsd_file_put(nf);
6899 goto out;
6900 }
6901 } else {
6902 status = nfsd_file_acquire(rqstp, fhp, acc, &nf);
6903 if (status)
6904 return status;
6905 }
6906 *nfp = nf;
6907 out:
6908 return status;
6909 }
6910 static void
_free_cpntf_state_locked(struct nfsd_net * nn,struct nfs4_cpntf_state * cps)6911 _free_cpntf_state_locked(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6912 {
6913 WARN_ON_ONCE(cps->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID);
6914 if (!refcount_dec_and_test(&cps->cp_stateid.cs_count))
6915 return;
6916 list_del(&cps->cp_list);
6917 idr_remove(&nn->s2s_cp_stateids,
6918 cps->cp_stateid.cs_stid.si_opaque.so_id);
6919 kfree(cps);
6920 }
6921 /*
6922 * A READ from an inter server to server COPY will have a
6923 * copy stateid. Look up the copy notify stateid from the
6924 * idr structure and take a reference on it.
6925 */
manage_cpntf_state(struct nfsd_net * nn,stateid_t * st,struct nfs4_client * clp,struct nfs4_cpntf_state ** cps)6926 __be32 manage_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6927 struct nfs4_client *clp,
6928 struct nfs4_cpntf_state **cps)
6929 {
6930 copy_stateid_t *cps_t;
6931 struct nfs4_cpntf_state *state = NULL;
6932
6933 if (st->si_opaque.so_clid.cl_id != nn->s2s_cp_cl_id)
6934 return nfserr_bad_stateid;
6935 spin_lock(&nn->s2s_cp_lock);
6936 cps_t = idr_find(&nn->s2s_cp_stateids, st->si_opaque.so_id);
6937 if (cps_t) {
6938 state = container_of(cps_t, struct nfs4_cpntf_state,
6939 cp_stateid);
6940 if (state->cp_stateid.cs_type != NFS4_COPYNOTIFY_STID) {
6941 state = NULL;
6942 goto unlock;
6943 }
6944 if (!clp)
6945 refcount_inc(&state->cp_stateid.cs_count);
6946 else
6947 _free_cpntf_state_locked(nn, state);
6948 }
6949 unlock:
6950 spin_unlock(&nn->s2s_cp_lock);
6951 if (!state)
6952 return nfserr_bad_stateid;
6953 if (!clp)
6954 *cps = state;
6955 return 0;
6956 }
6957
find_cpntf_state(struct nfsd_net * nn,stateid_t * st,struct nfs4_stid ** stid)6958 static __be32 find_cpntf_state(struct nfsd_net *nn, stateid_t *st,
6959 struct nfs4_stid **stid)
6960 {
6961 __be32 status;
6962 struct nfs4_cpntf_state *cps = NULL;
6963 struct nfs4_client *found;
6964
6965 status = manage_cpntf_state(nn, st, NULL, &cps);
6966 if (status)
6967 return status;
6968
6969 cps->cpntf_time = ktime_get_boottime_seconds();
6970
6971 status = nfserr_expired;
6972 found = lookup_clientid(&cps->cp_p_clid, true, nn);
6973 if (!found)
6974 goto out;
6975
6976 *stid = find_stateid_by_type(found, &cps->cp_p_stateid,
6977 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
6978 0);
6979 if (*stid)
6980 status = nfs_ok;
6981 else
6982 status = nfserr_bad_stateid;
6983
6984 put_client_renew(found);
6985 out:
6986 nfs4_put_cpntf_state(nn, cps);
6987 return status;
6988 }
6989
nfs4_put_cpntf_state(struct nfsd_net * nn,struct nfs4_cpntf_state * cps)6990 void nfs4_put_cpntf_state(struct nfsd_net *nn, struct nfs4_cpntf_state *cps)
6991 {
6992 spin_lock(&nn->s2s_cp_lock);
6993 _free_cpntf_state_locked(nn, cps);
6994 spin_unlock(&nn->s2s_cp_lock);
6995 }
6996
6997 /**
6998 * nfs4_preprocess_stateid_op - find and prep stateid for an operation
6999 * @rqstp: incoming request from client
7000 * @cstate: current compound state
7001 * @fhp: filehandle associated with requested stateid
7002 * @stateid: stateid (provided by client)
7003 * @flags: flags describing type of operation to be done
7004 * @nfp: optional nfsd_file return pointer (may be NULL)
7005 * @cstid: optional returned nfs4_stid pointer (may be NULL)
7006 *
7007 * Given info from the client, look up a nfs4_stid for the operation. On
7008 * success, it returns a reference to the nfs4_stid and/or the nfsd_file
7009 * associated with it.
7010 */
7011 __be32
nfs4_preprocess_stateid_op(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,struct svc_fh * fhp,stateid_t * stateid,int flags,struct nfsd_file ** nfp,struct nfs4_stid ** cstid)7012 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
7013 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
7014 stateid_t *stateid, int flags, struct nfsd_file **nfp,
7015 struct nfs4_stid **cstid)
7016 {
7017 struct net *net = SVC_NET(rqstp);
7018 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7019 struct nfs4_stid *s = NULL;
7020 __be32 status;
7021
7022 if (nfp)
7023 *nfp = NULL;
7024
7025 if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
7026 if (cstid)
7027 status = nfserr_bad_stateid;
7028 else
7029 status = check_special_stateids(net, fhp, stateid,
7030 flags);
7031 goto done;
7032 }
7033
7034 status = nfsd4_lookup_stateid(cstate, stateid,
7035 SC_TYPE_DELEG|SC_TYPE_OPEN|SC_TYPE_LOCK,
7036 0, &s, nn);
7037 if (status == nfserr_bad_stateid)
7038 status = find_cpntf_state(nn, stateid, &s);
7039 if (status)
7040 return status;
7041 status = nfsd4_stid_check_stateid_generation(stateid, s,
7042 nfsd4_has_session(cstate));
7043 if (status)
7044 goto out;
7045
7046 switch (s->sc_type) {
7047 case SC_TYPE_DELEG:
7048 status = nfs4_check_delegmode(delegstateid(s), flags);
7049 break;
7050 case SC_TYPE_OPEN:
7051 case SC_TYPE_LOCK:
7052 status = nfs4_check_olstateid(openlockstateid(s), flags);
7053 break;
7054 }
7055 if (status)
7056 goto out;
7057 status = nfs4_check_fh(fhp, s);
7058
7059 done:
7060 if (status == nfs_ok && nfp)
7061 status = nfs4_check_file(rqstp, fhp, s, nfp, flags);
7062 out:
7063 if (s) {
7064 if (!status && cstid)
7065 *cstid = s;
7066 else
7067 nfs4_put_stid(s);
7068 }
7069 return status;
7070 }
7071
7072 /*
7073 * Test if the stateid is valid
7074 */
7075 __be32
nfsd4_test_stateid(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7076 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7077 union nfsd4_op_u *u)
7078 {
7079 struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
7080 struct nfsd4_test_stateid_id *stateid;
7081 struct nfs4_client *cl = cstate->clp;
7082
7083 list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
7084 stateid->ts_id_status =
7085 nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
7086
7087 return nfs_ok;
7088 }
7089
7090 static __be32
nfsd4_free_lock_stateid(stateid_t * stateid,struct nfs4_stid * s)7091 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
7092 {
7093 struct nfs4_ol_stateid *stp = openlockstateid(s);
7094 __be32 ret;
7095
7096 ret = nfsd4_lock_ol_stateid(stp);
7097 if (ret)
7098 goto out_put_stid;
7099
7100 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7101 if (ret)
7102 goto out;
7103
7104 ret = nfserr_locks_held;
7105 if (check_for_locks(stp->st_stid.sc_file,
7106 lockowner(stp->st_stateowner)))
7107 goto out;
7108
7109 release_lock_stateid(stp);
7110 ret = nfs_ok;
7111
7112 out:
7113 mutex_unlock(&stp->st_mutex);
7114 out_put_stid:
7115 nfs4_put_stid(s);
7116 return ret;
7117 }
7118
7119 __be32
nfsd4_free_stateid(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7120 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7121 union nfsd4_op_u *u)
7122 {
7123 struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
7124 stateid_t *stateid = &free_stateid->fr_stateid;
7125 struct nfs4_stid *s;
7126 struct nfs4_delegation *dp;
7127 struct nfs4_client *cl = cstate->clp;
7128 __be32 ret = nfserr_bad_stateid;
7129
7130 spin_lock(&cl->cl_lock);
7131 s = find_stateid_locked(cl, stateid);
7132 if (!s || s->sc_status & SC_STATUS_CLOSED)
7133 goto out_unlock;
7134 if (s->sc_status & SC_STATUS_ADMIN_REVOKED) {
7135 nfsd4_drop_revoked_stid(s);
7136 ret = nfs_ok;
7137 goto out;
7138 }
7139 spin_lock(&s->sc_lock);
7140 switch (s->sc_type) {
7141 case SC_TYPE_DELEG:
7142 if (s->sc_status & SC_STATUS_REVOKED) {
7143 spin_unlock(&s->sc_lock);
7144 dp = delegstateid(s);
7145 list_del_init(&dp->dl_recall_lru);
7146 spin_unlock(&cl->cl_lock);
7147 nfs4_put_stid(s);
7148 ret = nfs_ok;
7149 goto out;
7150 }
7151 ret = nfserr_locks_held;
7152 break;
7153 case SC_TYPE_OPEN:
7154 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
7155 if (ret)
7156 break;
7157 ret = nfserr_locks_held;
7158 break;
7159 case SC_TYPE_LOCK:
7160 spin_unlock(&s->sc_lock);
7161 refcount_inc(&s->sc_count);
7162 spin_unlock(&cl->cl_lock);
7163 ret = nfsd4_free_lock_stateid(stateid, s);
7164 goto out;
7165 }
7166 spin_unlock(&s->sc_lock);
7167 out_unlock:
7168 spin_unlock(&cl->cl_lock);
7169 out:
7170 return ret;
7171 }
7172
7173 static inline int
setlkflg(int type)7174 setlkflg (int type)
7175 {
7176 return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
7177 RD_STATE : WR_STATE;
7178 }
7179
nfs4_seqid_op_checks(struct nfsd4_compound_state * cstate,stateid_t * stateid,u32 seqid,struct nfs4_ol_stateid * stp)7180 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
7181 {
7182 struct svc_fh *current_fh = &cstate->current_fh;
7183 struct nfs4_stateowner *sop = stp->st_stateowner;
7184 __be32 status;
7185
7186 status = nfsd4_check_seqid(cstate, sop, seqid);
7187 if (status)
7188 return status;
7189 status = nfsd4_lock_ol_stateid(stp);
7190 if (status != nfs_ok)
7191 return status;
7192 status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
7193 if (status == nfs_ok)
7194 status = nfs4_check_fh(current_fh, &stp->st_stid);
7195 if (status != nfs_ok)
7196 mutex_unlock(&stp->st_mutex);
7197 return status;
7198 }
7199
7200 /**
7201 * nfs4_preprocess_seqid_op - find and prep an ol_stateid for a seqid-morphing op
7202 * @cstate: compund state
7203 * @seqid: seqid (provided by client)
7204 * @stateid: stateid (provided by client)
7205 * @typemask: mask of allowable types for this operation
7206 * @statusmask: mask of allowed states: 0 or STID_CLOSED
7207 * @stpp: return pointer for the stateid found
7208 * @nn: net namespace for request
7209 *
7210 * Given a stateid+seqid from a client, look up an nfs4_ol_stateid and
7211 * return it in @stpp. On a nfs_ok return, the returned stateid will
7212 * have its st_mutex locked.
7213 */
7214 static __be32
nfs4_preprocess_seqid_op(struct nfsd4_compound_state * cstate,u32 seqid,stateid_t * stateid,unsigned short typemask,unsigned short statusmask,struct nfs4_ol_stateid ** stpp,struct nfsd_net * nn)7215 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7216 stateid_t *stateid,
7217 unsigned short typemask, unsigned short statusmask,
7218 struct nfs4_ol_stateid **stpp,
7219 struct nfsd_net *nn)
7220 {
7221 __be32 status;
7222 struct nfs4_stid *s;
7223 struct nfs4_ol_stateid *stp = NULL;
7224
7225 trace_nfsd_preprocess(seqid, stateid);
7226
7227 *stpp = NULL;
7228 retry:
7229 status = nfsd4_lookup_stateid(cstate, stateid,
7230 typemask, statusmask, &s, nn);
7231 if (status)
7232 return status;
7233 stp = openlockstateid(s);
7234 if (nfsd4_cstate_assign_replay(cstate, stp->st_stateowner) == -EAGAIN) {
7235 nfs4_put_stateowner(stp->st_stateowner);
7236 goto retry;
7237 }
7238
7239 status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
7240 if (!status)
7241 *stpp = stp;
7242 else
7243 nfs4_put_stid(&stp->st_stid);
7244 return status;
7245 }
7246
nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state * cstate,u32 seqid,stateid_t * stateid,struct nfs4_ol_stateid ** stpp,struct nfsd_net * nn)7247 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
7248 stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
7249 {
7250 __be32 status;
7251 struct nfs4_openowner *oo;
7252 struct nfs4_ol_stateid *stp;
7253
7254 status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
7255 SC_TYPE_OPEN, 0, &stp, nn);
7256 if (status)
7257 return status;
7258 oo = openowner(stp->st_stateowner);
7259 if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
7260 mutex_unlock(&stp->st_mutex);
7261 nfs4_put_stid(&stp->st_stid);
7262 return nfserr_bad_stateid;
7263 }
7264 *stpp = stp;
7265 return nfs_ok;
7266 }
7267
7268 __be32
nfsd4_open_confirm(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7269 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7270 union nfsd4_op_u *u)
7271 {
7272 struct nfsd4_open_confirm *oc = &u->open_confirm;
7273 __be32 status;
7274 struct nfs4_openowner *oo;
7275 struct nfs4_ol_stateid *stp;
7276 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7277
7278 dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
7279 cstate->current_fh.fh_dentry);
7280
7281 status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
7282 if (status)
7283 return status;
7284
7285 status = nfs4_preprocess_seqid_op(cstate,
7286 oc->oc_seqid, &oc->oc_req_stateid,
7287 SC_TYPE_OPEN, 0, &stp, nn);
7288 if (status)
7289 goto out;
7290 oo = openowner(stp->st_stateowner);
7291 status = nfserr_bad_stateid;
7292 if (oo->oo_flags & NFS4_OO_CONFIRMED) {
7293 mutex_unlock(&stp->st_mutex);
7294 goto put_stateid;
7295 }
7296 oo->oo_flags |= NFS4_OO_CONFIRMED;
7297 nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
7298 mutex_unlock(&stp->st_mutex);
7299 trace_nfsd_open_confirm(oc->oc_seqid, &stp->st_stid.sc_stateid);
7300 nfsd4_client_record_create(oo->oo_owner.so_client);
7301 status = nfs_ok;
7302 put_stateid:
7303 nfs4_put_stid(&stp->st_stid);
7304 out:
7305 nfsd4_bump_seqid(cstate, status);
7306 return status;
7307 }
7308
nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid * stp,u32 access)7309 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
7310 {
7311 if (!test_access(access, stp))
7312 return;
7313 nfs4_file_put_access(stp->st_stid.sc_file, access);
7314 clear_access(access, stp);
7315 }
7316
nfs4_stateid_downgrade(struct nfs4_ol_stateid * stp,u32 to_access)7317 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
7318 {
7319 switch (to_access) {
7320 case NFS4_SHARE_ACCESS_READ:
7321 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
7322 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7323 break;
7324 case NFS4_SHARE_ACCESS_WRITE:
7325 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
7326 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
7327 break;
7328 case NFS4_SHARE_ACCESS_BOTH:
7329 break;
7330 default:
7331 WARN_ON_ONCE(1);
7332 }
7333 }
7334
7335 __be32
nfsd4_open_downgrade(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7336 nfsd4_open_downgrade(struct svc_rqst *rqstp,
7337 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
7338 {
7339 struct nfsd4_open_downgrade *od = &u->open_downgrade;
7340 __be32 status;
7341 struct nfs4_ol_stateid *stp;
7342 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7343
7344 dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
7345 cstate->current_fh.fh_dentry);
7346
7347 /* We don't yet support WANT bits: */
7348 if (od->od_deleg_want)
7349 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
7350 od->od_deleg_want);
7351
7352 status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
7353 &od->od_stateid, &stp, nn);
7354 if (status)
7355 goto out;
7356 status = nfserr_inval;
7357 if (!test_access(od->od_share_access, stp)) {
7358 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
7359 stp->st_access_bmap, od->od_share_access);
7360 goto put_stateid;
7361 }
7362 if (!test_deny(od->od_share_deny, stp)) {
7363 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
7364 stp->st_deny_bmap, od->od_share_deny);
7365 goto put_stateid;
7366 }
7367 nfs4_stateid_downgrade(stp, od->od_share_access);
7368 reset_union_bmap_deny(od->od_share_deny, stp);
7369 nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
7370 status = nfs_ok;
7371 put_stateid:
7372 mutex_unlock(&stp->st_mutex);
7373 nfs4_put_stid(&stp->st_stid);
7374 out:
7375 nfsd4_bump_seqid(cstate, status);
7376 return status;
7377 }
7378
nfsd4_close_open_stateid(struct nfs4_ol_stateid * s)7379 static bool nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
7380 {
7381 struct nfs4_client *clp = s->st_stid.sc_client;
7382 bool unhashed;
7383 LIST_HEAD(reaplist);
7384 struct nfs4_ol_stateid *stp;
7385
7386 spin_lock(&clp->cl_lock);
7387 unhashed = unhash_open_stateid(s, &reaplist);
7388
7389 if (clp->cl_minorversion) {
7390 if (unhashed)
7391 put_ol_stateid_locked(s, &reaplist);
7392 spin_unlock(&clp->cl_lock);
7393 list_for_each_entry(stp, &reaplist, st_locks)
7394 nfs4_free_cpntf_statelist(clp->net, &stp->st_stid);
7395 free_ol_stateid_reaplist(&reaplist);
7396 return false;
7397 } else {
7398 spin_unlock(&clp->cl_lock);
7399 free_ol_stateid_reaplist(&reaplist);
7400 return unhashed;
7401 }
7402 }
7403
7404 /*
7405 * nfs4_unlock_state() called after encode
7406 */
7407 __be32
nfsd4_close(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7408 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7409 union nfsd4_op_u *u)
7410 {
7411 struct nfsd4_close *close = &u->close;
7412 __be32 status;
7413 struct nfs4_ol_stateid *stp;
7414 struct net *net = SVC_NET(rqstp);
7415 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7416 bool need_move_to_close_list;
7417
7418 dprintk("NFSD: nfsd4_close on file %pd\n",
7419 cstate->current_fh.fh_dentry);
7420
7421 status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
7422 &close->cl_stateid,
7423 SC_TYPE_OPEN, SC_STATUS_CLOSED,
7424 &stp, nn);
7425 nfsd4_bump_seqid(cstate, status);
7426 if (status)
7427 goto out;
7428
7429 spin_lock(&stp->st_stid.sc_client->cl_lock);
7430 stp->st_stid.sc_status |= SC_STATUS_CLOSED;
7431 spin_unlock(&stp->st_stid.sc_client->cl_lock);
7432
7433 /*
7434 * Technically we don't _really_ have to increment or copy it, since
7435 * it should just be gone after this operation and we clobber the
7436 * copied value below, but we continue to do so here just to ensure
7437 * that racing ops see that there was a state change.
7438 */
7439 nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
7440
7441 need_move_to_close_list = nfsd4_close_open_stateid(stp);
7442 mutex_unlock(&stp->st_mutex);
7443 if (need_move_to_close_list)
7444 move_to_close_lru(stp, net);
7445
7446 /* v4.1+ suggests that we send a special stateid in here, since the
7447 * clients should just ignore this anyway. Since this is not useful
7448 * for v4.0 clients either, we set it to the special close_stateid
7449 * universally.
7450 *
7451 * See RFC5661 section 18.2.4, and RFC7530 section 16.2.5
7452 */
7453 memcpy(&close->cl_stateid, &close_stateid, sizeof(close->cl_stateid));
7454
7455 /* put reference from nfs4_preprocess_seqid_op */
7456 nfs4_put_stid(&stp->st_stid);
7457 out:
7458 return status;
7459 }
7460
7461 __be32
nfsd4_delegreturn(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7462 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7463 union nfsd4_op_u *u)
7464 {
7465 struct nfsd4_delegreturn *dr = &u->delegreturn;
7466 struct nfs4_delegation *dp;
7467 stateid_t *stateid = &dr->dr_stateid;
7468 struct nfs4_stid *s;
7469 __be32 status;
7470 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
7471
7472 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
7473 return status;
7474
7475 status = nfsd4_lookup_stateid(cstate, stateid, SC_TYPE_DELEG, 0, &s, nn);
7476 if (status)
7477 goto out;
7478 dp = delegstateid(s);
7479 status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
7480 if (status)
7481 goto put_stateid;
7482
7483 trace_nfsd_deleg_return(stateid);
7484 wake_up_var(d_inode(cstate->current_fh.fh_dentry));
7485 destroy_delegation(dp);
7486 put_stateid:
7487 nfs4_put_stid(&dp->dl_stid);
7488 out:
7489 return status;
7490 }
7491
7492 /* last octet in a range */
7493 static inline u64
last_byte_offset(u64 start,u64 len)7494 last_byte_offset(u64 start, u64 len)
7495 {
7496 u64 end;
7497
7498 WARN_ON_ONCE(!len);
7499 end = start + len;
7500 return end > start ? end - 1: NFS4_MAX_UINT64;
7501 }
7502
7503 /*
7504 * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
7505 * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
7506 * byte, because of sign extension problems. Since NFSv4 calls for 64-bit
7507 * locking, this prevents us from being completely protocol-compliant. The
7508 * real solution to this problem is to start using unsigned file offsets in
7509 * the VFS, but this is a very deep change!
7510 */
7511 static inline void
nfs4_transform_lock_offset(struct file_lock * lock)7512 nfs4_transform_lock_offset(struct file_lock *lock)
7513 {
7514 if (lock->fl_start < 0)
7515 lock->fl_start = OFFSET_MAX;
7516 if (lock->fl_end < 0)
7517 lock->fl_end = OFFSET_MAX;
7518 }
7519
7520 static fl_owner_t
nfsd4_lm_get_owner(fl_owner_t owner)7521 nfsd4_lm_get_owner(fl_owner_t owner)
7522 {
7523 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7524
7525 nfs4_get_stateowner(&lo->lo_owner);
7526 return owner;
7527 }
7528
7529 static void
nfsd4_lm_put_owner(fl_owner_t owner)7530 nfsd4_lm_put_owner(fl_owner_t owner)
7531 {
7532 struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
7533
7534 if (lo)
7535 nfs4_put_stateowner(&lo->lo_owner);
7536 }
7537
7538 /* return pointer to struct nfs4_client if client is expirable */
7539 static bool
nfsd4_lm_lock_expirable(struct file_lock * cfl)7540 nfsd4_lm_lock_expirable(struct file_lock *cfl)
7541 {
7542 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) cfl->c.flc_owner;
7543 struct nfs4_client *clp = lo->lo_owner.so_client;
7544 struct nfsd_net *nn;
7545
7546 if (try_to_expire_client(clp)) {
7547 nn = net_generic(clp->net, nfsd_net_id);
7548 mod_delayed_work(laundry_wq, &nn->laundromat_work, 0);
7549 return true;
7550 }
7551 return false;
7552 }
7553
7554 /* schedule laundromat to run immediately and wait for it to complete */
7555 static void
nfsd4_lm_expire_lock(void)7556 nfsd4_lm_expire_lock(void)
7557 {
7558 flush_workqueue(laundry_wq);
7559 }
7560
7561 static void
nfsd4_lm_notify(struct file_lock * fl)7562 nfsd4_lm_notify(struct file_lock *fl)
7563 {
7564 struct nfs4_lockowner *lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7565 struct net *net = lo->lo_owner.so_client->net;
7566 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7567 struct nfsd4_blocked_lock *nbl = container_of(fl,
7568 struct nfsd4_blocked_lock, nbl_lock);
7569 bool queue = false;
7570
7571 /* An empty list means that something else is going to be using it */
7572 spin_lock(&nn->blocked_locks_lock);
7573 if (!list_empty(&nbl->nbl_list)) {
7574 list_del_init(&nbl->nbl_list);
7575 list_del_init(&nbl->nbl_lru);
7576 queue = true;
7577 }
7578 spin_unlock(&nn->blocked_locks_lock);
7579
7580 if (queue) {
7581 trace_nfsd_cb_notify_lock(lo, nbl);
7582 nfsd4_run_cb(&nbl->nbl_cb);
7583 }
7584 }
7585
7586 static const struct lock_manager_operations nfsd_posix_mng_ops = {
7587 .lm_mod_owner = THIS_MODULE,
7588 .lm_notify = nfsd4_lm_notify,
7589 .lm_get_owner = nfsd4_lm_get_owner,
7590 .lm_put_owner = nfsd4_lm_put_owner,
7591 .lm_lock_expirable = nfsd4_lm_lock_expirable,
7592 .lm_expire_lock = nfsd4_lm_expire_lock,
7593 };
7594
7595 static inline void
nfs4_set_lock_denied(struct file_lock * fl,struct nfsd4_lock_denied * deny)7596 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
7597 {
7598 struct nfs4_lockowner *lo;
7599
7600 if (fl->fl_lmops == &nfsd_posix_mng_ops) {
7601 lo = (struct nfs4_lockowner *) fl->c.flc_owner;
7602 xdr_netobj_dup(&deny->ld_owner, &lo->lo_owner.so_owner,
7603 GFP_KERNEL);
7604 if (!deny->ld_owner.data)
7605 /* We just don't care that much */
7606 goto nevermind;
7607 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
7608 } else {
7609 nevermind:
7610 deny->ld_owner.len = 0;
7611 deny->ld_owner.data = NULL;
7612 deny->ld_clientid.cl_boot = 0;
7613 deny->ld_clientid.cl_id = 0;
7614 }
7615 deny->ld_start = fl->fl_start;
7616 deny->ld_length = NFS4_MAX_UINT64;
7617 if (fl->fl_end != NFS4_MAX_UINT64)
7618 deny->ld_length = fl->fl_end - fl->fl_start + 1;
7619 deny->ld_type = NFS4_READ_LT;
7620 if (fl->c.flc_type != F_RDLCK)
7621 deny->ld_type = NFS4_WRITE_LT;
7622 }
7623
7624 static struct nfs4_lockowner *
find_lockowner_str_locked(struct nfs4_client * clp,struct xdr_netobj * owner)7625 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
7626 {
7627 unsigned int strhashval = ownerstr_hashval(owner);
7628 struct nfs4_stateowner *so;
7629
7630 lockdep_assert_held(&clp->cl_lock);
7631
7632 list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
7633 so_strhash) {
7634 if (so->so_is_open_owner)
7635 continue;
7636 if (same_owner_str(so, owner))
7637 return lockowner(nfs4_get_stateowner(so));
7638 }
7639 return NULL;
7640 }
7641
7642 static struct nfs4_lockowner *
find_lockowner_str(struct nfs4_client * clp,struct xdr_netobj * owner)7643 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
7644 {
7645 struct nfs4_lockowner *lo;
7646
7647 spin_lock(&clp->cl_lock);
7648 lo = find_lockowner_str_locked(clp, owner);
7649 spin_unlock(&clp->cl_lock);
7650 return lo;
7651 }
7652
nfs4_unhash_lockowner(struct nfs4_stateowner * sop)7653 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
7654 {
7655 unhash_lockowner_locked(lockowner(sop));
7656 }
7657
nfs4_free_lockowner(struct nfs4_stateowner * sop)7658 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
7659 {
7660 struct nfs4_lockowner *lo = lockowner(sop);
7661
7662 kmem_cache_free(lockowner_slab, lo);
7663 }
7664
7665 static const struct nfs4_stateowner_operations lockowner_ops = {
7666 .so_unhash = nfs4_unhash_lockowner,
7667 .so_free = nfs4_free_lockowner,
7668 };
7669
7670 /*
7671 * Alloc a lock owner structure.
7672 * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
7673 * occurred.
7674 *
7675 * strhashval = ownerstr_hashval
7676 */
7677 static struct nfs4_lockowner *
alloc_init_lock_stateowner(unsigned int strhashval,struct nfs4_client * clp,struct nfs4_ol_stateid * open_stp,struct nfsd4_lock * lock)7678 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
7679 struct nfs4_ol_stateid *open_stp,
7680 struct nfsd4_lock *lock)
7681 {
7682 struct nfs4_lockowner *lo, *ret;
7683
7684 lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
7685 if (!lo)
7686 return NULL;
7687 INIT_LIST_HEAD(&lo->lo_blocked);
7688 INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
7689 lo->lo_owner.so_is_open_owner = 0;
7690 lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
7691 lo->lo_owner.so_ops = &lockowner_ops;
7692 spin_lock(&clp->cl_lock);
7693 ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
7694 if (ret == NULL) {
7695 list_add(&lo->lo_owner.so_strhash,
7696 &clp->cl_ownerstr_hashtbl[strhashval]);
7697 ret = lo;
7698 } else
7699 nfs4_free_stateowner(&lo->lo_owner);
7700
7701 spin_unlock(&clp->cl_lock);
7702 return ret;
7703 }
7704
7705 static struct nfs4_ol_stateid *
find_lock_stateid(const struct nfs4_lockowner * lo,const struct nfs4_ol_stateid * ost)7706 find_lock_stateid(const struct nfs4_lockowner *lo,
7707 const struct nfs4_ol_stateid *ost)
7708 {
7709 struct nfs4_ol_stateid *lst;
7710
7711 lockdep_assert_held(&ost->st_stid.sc_client->cl_lock);
7712
7713 /* If ost is not hashed, ost->st_locks will not be valid */
7714 if (!nfs4_ol_stateid_unhashed(ost))
7715 list_for_each_entry(lst, &ost->st_locks, st_locks) {
7716 if (lst->st_stateowner == &lo->lo_owner) {
7717 refcount_inc(&lst->st_stid.sc_count);
7718 return lst;
7719 }
7720 }
7721 return NULL;
7722 }
7723
7724 static struct nfs4_ol_stateid *
init_lock_stateid(struct nfs4_ol_stateid * stp,struct nfs4_lockowner * lo,struct nfs4_file * fp,struct inode * inode,struct nfs4_ol_stateid * open_stp)7725 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
7726 struct nfs4_file *fp, struct inode *inode,
7727 struct nfs4_ol_stateid *open_stp)
7728 {
7729 struct nfs4_client *clp = lo->lo_owner.so_client;
7730 struct nfs4_ol_stateid *retstp;
7731
7732 mutex_init(&stp->st_mutex);
7733 mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
7734 retry:
7735 spin_lock(&clp->cl_lock);
7736 if (nfs4_ol_stateid_unhashed(open_stp))
7737 goto out_close;
7738 retstp = find_lock_stateid(lo, open_stp);
7739 if (retstp)
7740 goto out_found;
7741 refcount_inc(&stp->st_stid.sc_count);
7742 stp->st_stid.sc_type = SC_TYPE_LOCK;
7743 stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
7744 get_nfs4_file(fp);
7745 stp->st_stid.sc_file = fp;
7746 stp->st_access_bmap = 0;
7747 stp->st_deny_bmap = open_stp->st_deny_bmap;
7748 stp->st_openstp = open_stp;
7749 spin_lock(&fp->fi_lock);
7750 list_add(&stp->st_locks, &open_stp->st_locks);
7751 list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
7752 list_add(&stp->st_perfile, &fp->fi_stateids);
7753 spin_unlock(&fp->fi_lock);
7754 spin_unlock(&clp->cl_lock);
7755 return stp;
7756 out_found:
7757 spin_unlock(&clp->cl_lock);
7758 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
7759 nfs4_put_stid(&retstp->st_stid);
7760 goto retry;
7761 }
7762 /* To keep mutex tracking happy */
7763 mutex_unlock(&stp->st_mutex);
7764 return retstp;
7765 out_close:
7766 spin_unlock(&clp->cl_lock);
7767 mutex_unlock(&stp->st_mutex);
7768 return NULL;
7769 }
7770
7771 static struct nfs4_ol_stateid *
find_or_create_lock_stateid(struct nfs4_lockowner * lo,struct nfs4_file * fi,struct inode * inode,struct nfs4_ol_stateid * ost,bool * new)7772 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
7773 struct inode *inode, struct nfs4_ol_stateid *ost,
7774 bool *new)
7775 {
7776 struct nfs4_stid *ns = NULL;
7777 struct nfs4_ol_stateid *lst;
7778 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7779 struct nfs4_client *clp = oo->oo_owner.so_client;
7780
7781 *new = false;
7782 spin_lock(&clp->cl_lock);
7783 lst = find_lock_stateid(lo, ost);
7784 spin_unlock(&clp->cl_lock);
7785 if (lst != NULL) {
7786 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
7787 goto out;
7788 nfs4_put_stid(&lst->st_stid);
7789 }
7790 ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
7791 if (ns == NULL)
7792 return NULL;
7793
7794 lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
7795 if (lst == openlockstateid(ns))
7796 *new = true;
7797 else
7798 nfs4_put_stid(ns);
7799 out:
7800 return lst;
7801 }
7802
7803 static int
check_lock_length(u64 offset,u64 length)7804 check_lock_length(u64 offset, u64 length)
7805 {
7806 return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
7807 (length > ~offset)));
7808 }
7809
get_lock_access(struct nfs4_ol_stateid * lock_stp,u32 access)7810 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
7811 {
7812 struct nfs4_file *fp = lock_stp->st_stid.sc_file;
7813
7814 lockdep_assert_held(&fp->fi_lock);
7815
7816 if (test_access(access, lock_stp))
7817 return;
7818 __nfs4_file_get_access(fp, access);
7819 set_access(access, lock_stp);
7820 }
7821
7822 static __be32
lookup_or_create_lock_state(struct nfsd4_compound_state * cstate,struct nfs4_ol_stateid * ost,struct nfsd4_lock * lock,struct nfs4_ol_stateid ** plst,bool * new)7823 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
7824 struct nfs4_ol_stateid *ost,
7825 struct nfsd4_lock *lock,
7826 struct nfs4_ol_stateid **plst, bool *new)
7827 {
7828 __be32 status;
7829 struct nfs4_file *fi = ost->st_stid.sc_file;
7830 struct nfs4_openowner *oo = openowner(ost->st_stateowner);
7831 struct nfs4_client *cl = oo->oo_owner.so_client;
7832 struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
7833 struct nfs4_lockowner *lo;
7834 struct nfs4_ol_stateid *lst;
7835 unsigned int strhashval;
7836
7837 lo = find_lockowner_str(cl, &lock->lk_new_owner);
7838 if (!lo) {
7839 strhashval = ownerstr_hashval(&lock->lk_new_owner);
7840 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
7841 if (lo == NULL)
7842 return nfserr_jukebox;
7843 } else {
7844 /* with an existing lockowner, seqids must be the same */
7845 status = nfserr_bad_seqid;
7846 if (!cstate->minorversion &&
7847 lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
7848 goto out;
7849 }
7850
7851 lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
7852 if (lst == NULL) {
7853 status = nfserr_jukebox;
7854 goto out;
7855 }
7856
7857 status = nfs_ok;
7858 *plst = lst;
7859 out:
7860 nfs4_put_stateowner(&lo->lo_owner);
7861 return status;
7862 }
7863
7864 /*
7865 * LOCK operation
7866 */
7867 __be32
nfsd4_lock(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)7868 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
7869 union nfsd4_op_u *u)
7870 {
7871 struct nfsd4_lock *lock = &u->lock;
7872 struct nfs4_openowner *open_sop = NULL;
7873 struct nfs4_lockowner *lock_sop = NULL;
7874 struct nfs4_ol_stateid *lock_stp = NULL;
7875 struct nfs4_ol_stateid *open_stp = NULL;
7876 struct nfs4_file *fp;
7877 struct nfsd_file *nf = NULL;
7878 struct nfsd4_blocked_lock *nbl = NULL;
7879 struct file_lock *file_lock = NULL;
7880 struct file_lock *conflock = NULL;
7881 struct super_block *sb;
7882 __be32 status = 0;
7883 int lkflg;
7884 int err;
7885 bool new = false;
7886 unsigned char type;
7887 unsigned int flags = FL_POSIX;
7888 struct net *net = SVC_NET(rqstp);
7889 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7890
7891 dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
7892 (long long) lock->lk_offset,
7893 (long long) lock->lk_length);
7894
7895 if (check_lock_length(lock->lk_offset, lock->lk_length))
7896 return nfserr_inval;
7897
7898 if ((status = fh_verify(rqstp, &cstate->current_fh,
7899 S_IFREG, NFSD_MAY_LOCK))) {
7900 dprintk("NFSD: nfsd4_lock: permission denied!\n");
7901 return status;
7902 }
7903 sb = cstate->current_fh.fh_dentry->d_sb;
7904
7905 if (lock->lk_is_new) {
7906 if (nfsd4_has_session(cstate))
7907 /* See rfc 5661 18.10.3: given clientid is ignored: */
7908 memcpy(&lock->lk_new_clientid,
7909 &cstate->clp->cl_clientid,
7910 sizeof(clientid_t));
7911
7912 /* validate and update open stateid and open seqid */
7913 status = nfs4_preprocess_confirmed_seqid_op(cstate,
7914 lock->lk_new_open_seqid,
7915 &lock->lk_new_open_stateid,
7916 &open_stp, nn);
7917 if (status)
7918 goto out;
7919 mutex_unlock(&open_stp->st_mutex);
7920 open_sop = openowner(open_stp->st_stateowner);
7921 status = nfserr_bad_stateid;
7922 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
7923 &lock->lk_new_clientid))
7924 goto out;
7925 status = lookup_or_create_lock_state(cstate, open_stp, lock,
7926 &lock_stp, &new);
7927 } else {
7928 status = nfs4_preprocess_seqid_op(cstate,
7929 lock->lk_old_lock_seqid,
7930 &lock->lk_old_lock_stateid,
7931 SC_TYPE_LOCK, 0, &lock_stp,
7932 nn);
7933 }
7934 if (status)
7935 goto out;
7936 lock_sop = lockowner(lock_stp->st_stateowner);
7937
7938 lkflg = setlkflg(lock->lk_type);
7939 status = nfs4_check_openmode(lock_stp, lkflg);
7940 if (status)
7941 goto out;
7942
7943 status = nfserr_grace;
7944 if (locks_in_grace(net) && !lock->lk_reclaim)
7945 goto out;
7946 status = nfserr_no_grace;
7947 if (!locks_in_grace(net) && lock->lk_reclaim)
7948 goto out;
7949
7950 if (lock->lk_reclaim)
7951 flags |= FL_RECLAIM;
7952
7953 fp = lock_stp->st_stid.sc_file;
7954 switch (lock->lk_type) {
7955 case NFS4_READW_LT:
7956 if (nfsd4_has_session(cstate) ||
7957 exportfs_lock_op_is_async(sb->s_export_op))
7958 flags |= FL_SLEEP;
7959 fallthrough;
7960 case NFS4_READ_LT:
7961 spin_lock(&fp->fi_lock);
7962 nf = find_readable_file_locked(fp);
7963 if (nf)
7964 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
7965 spin_unlock(&fp->fi_lock);
7966 type = F_RDLCK;
7967 break;
7968 case NFS4_WRITEW_LT:
7969 if (nfsd4_has_session(cstate) ||
7970 exportfs_lock_op_is_async(sb->s_export_op))
7971 flags |= FL_SLEEP;
7972 fallthrough;
7973 case NFS4_WRITE_LT:
7974 spin_lock(&fp->fi_lock);
7975 nf = find_writeable_file_locked(fp);
7976 if (nf)
7977 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
7978 spin_unlock(&fp->fi_lock);
7979 type = F_WRLCK;
7980 break;
7981 default:
7982 status = nfserr_inval;
7983 goto out;
7984 }
7985
7986 if (!nf) {
7987 status = nfserr_openmode;
7988 goto out;
7989 }
7990
7991 /*
7992 * Most filesystems with their own ->lock operations will block
7993 * the nfsd thread waiting to acquire the lock. That leads to
7994 * deadlocks (we don't want every nfsd thread tied up waiting
7995 * for file locks), so don't attempt blocking lock notifications
7996 * on those filesystems:
7997 */
7998 if (!exportfs_lock_op_is_async(sb->s_export_op))
7999 flags &= ~FL_SLEEP;
8000
8001 nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
8002 if (!nbl) {
8003 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
8004 status = nfserr_jukebox;
8005 goto out;
8006 }
8007
8008 file_lock = &nbl->nbl_lock;
8009 file_lock->c.flc_type = type;
8010 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
8011 file_lock->c.flc_pid = current->tgid;
8012 file_lock->c.flc_file = nf->nf_file;
8013 file_lock->c.flc_flags = flags;
8014 file_lock->fl_lmops = &nfsd_posix_mng_ops;
8015 file_lock->fl_start = lock->lk_offset;
8016 file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
8017 nfs4_transform_lock_offset(file_lock);
8018
8019 conflock = locks_alloc_lock();
8020 if (!conflock) {
8021 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8022 status = nfserr_jukebox;
8023 goto out;
8024 }
8025
8026 if (flags & FL_SLEEP) {
8027 nbl->nbl_time = ktime_get_boottime_seconds();
8028 spin_lock(&nn->blocked_locks_lock);
8029 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
8030 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
8031 kref_get(&nbl->nbl_kref);
8032 spin_unlock(&nn->blocked_locks_lock);
8033 }
8034
8035 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, conflock);
8036 switch (err) {
8037 case 0: /* success! */
8038 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
8039 status = 0;
8040 if (lock->lk_reclaim)
8041 nn->somebody_reclaimed = true;
8042 break;
8043 case FILE_LOCK_DEFERRED:
8044 kref_put(&nbl->nbl_kref, free_nbl);
8045 nbl = NULL;
8046 fallthrough;
8047 case -EAGAIN: /* conflock holds conflicting lock */
8048 status = nfserr_denied;
8049 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
8050 nfs4_set_lock_denied(conflock, &lock->lk_denied);
8051 break;
8052 case -EDEADLK:
8053 status = nfserr_deadlock;
8054 break;
8055 default:
8056 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
8057 status = nfserrno(err);
8058 break;
8059 }
8060 out:
8061 if (nbl) {
8062 /* dequeue it if we queued it before */
8063 if (flags & FL_SLEEP) {
8064 spin_lock(&nn->blocked_locks_lock);
8065 if (!list_empty(&nbl->nbl_list) &&
8066 !list_empty(&nbl->nbl_lru)) {
8067 list_del_init(&nbl->nbl_list);
8068 list_del_init(&nbl->nbl_lru);
8069 kref_put(&nbl->nbl_kref, free_nbl);
8070 }
8071 /* nbl can use one of lists to be linked to reaplist */
8072 spin_unlock(&nn->blocked_locks_lock);
8073 }
8074 free_blocked_lock(nbl);
8075 }
8076 if (nf)
8077 nfsd_file_put(nf);
8078 if (lock_stp) {
8079 /* Bump seqid manually if the 4.0 replay owner is openowner */
8080 if (cstate->replay_owner &&
8081 cstate->replay_owner != &lock_sop->lo_owner &&
8082 seqid_mutating_err(ntohl(status)))
8083 lock_sop->lo_owner.so_seqid++;
8084
8085 /*
8086 * If this is a new, never-before-used stateid, and we are
8087 * returning an error, then just go ahead and release it.
8088 */
8089 if (status && new)
8090 release_lock_stateid(lock_stp);
8091
8092 mutex_unlock(&lock_stp->st_mutex);
8093
8094 nfs4_put_stid(&lock_stp->st_stid);
8095 }
8096 if (open_stp)
8097 nfs4_put_stid(&open_stp->st_stid);
8098 nfsd4_bump_seqid(cstate, status);
8099 if (conflock)
8100 locks_free_lock(conflock);
8101 return status;
8102 }
8103
nfsd4_lock_release(union nfsd4_op_u * u)8104 void nfsd4_lock_release(union nfsd4_op_u *u)
8105 {
8106 struct nfsd4_lock *lock = &u->lock;
8107 struct nfsd4_lock_denied *deny = &lock->lk_denied;
8108
8109 kfree(deny->ld_owner.data);
8110 }
8111
8112 /*
8113 * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
8114 * so we do a temporary open here just to get an open file to pass to
8115 * vfs_test_lock.
8116 */
nfsd_test_lock(struct svc_rqst * rqstp,struct svc_fh * fhp,struct file_lock * lock)8117 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
8118 {
8119 struct nfsd_file *nf;
8120 struct inode *inode;
8121 __be32 err;
8122
8123 err = nfsd_file_acquire(rqstp, fhp, NFSD_MAY_READ, &nf);
8124 if (err)
8125 return err;
8126 inode = fhp->fh_dentry->d_inode;
8127 inode_lock(inode); /* to block new leases till after test_lock: */
8128 err = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8129 if (err)
8130 goto out;
8131 lock->c.flc_file = nf->nf_file;
8132 err = nfserrno(vfs_test_lock(nf->nf_file, lock));
8133 lock->c.flc_file = NULL;
8134 out:
8135 inode_unlock(inode);
8136 nfsd_file_put(nf);
8137 return err;
8138 }
8139
8140 /*
8141 * LOCKT operation
8142 */
8143 __be32
nfsd4_lockt(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8144 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8145 union nfsd4_op_u *u)
8146 {
8147 struct nfsd4_lockt *lockt = &u->lockt;
8148 struct file_lock *file_lock = NULL;
8149 struct nfs4_lockowner *lo = NULL;
8150 __be32 status;
8151 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8152
8153 if (locks_in_grace(SVC_NET(rqstp)))
8154 return nfserr_grace;
8155
8156 if (check_lock_length(lockt->lt_offset, lockt->lt_length))
8157 return nfserr_inval;
8158
8159 if (!nfsd4_has_session(cstate)) {
8160 status = set_client(&lockt->lt_clientid, cstate, nn);
8161 if (status)
8162 goto out;
8163 }
8164
8165 if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
8166 goto out;
8167
8168 file_lock = locks_alloc_lock();
8169 if (!file_lock) {
8170 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8171 status = nfserr_jukebox;
8172 goto out;
8173 }
8174
8175 switch (lockt->lt_type) {
8176 case NFS4_READ_LT:
8177 case NFS4_READW_LT:
8178 file_lock->c.flc_type = F_RDLCK;
8179 break;
8180 case NFS4_WRITE_LT:
8181 case NFS4_WRITEW_LT:
8182 file_lock->c.flc_type = F_WRLCK;
8183 break;
8184 default:
8185 dprintk("NFSD: nfs4_lockt: bad lock type!\n");
8186 status = nfserr_inval;
8187 goto out;
8188 }
8189
8190 lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
8191 if (lo)
8192 file_lock->c.flc_owner = (fl_owner_t)lo;
8193 file_lock->c.flc_pid = current->tgid;
8194 file_lock->c.flc_flags = FL_POSIX;
8195
8196 file_lock->fl_start = lockt->lt_offset;
8197 file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
8198
8199 nfs4_transform_lock_offset(file_lock);
8200
8201 status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
8202 if (status)
8203 goto out;
8204
8205 if (file_lock->c.flc_type != F_UNLCK) {
8206 status = nfserr_denied;
8207 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
8208 }
8209 out:
8210 if (lo)
8211 nfs4_put_stateowner(&lo->lo_owner);
8212 if (file_lock)
8213 locks_free_lock(file_lock);
8214 return status;
8215 }
8216
nfsd4_lockt_release(union nfsd4_op_u * u)8217 void nfsd4_lockt_release(union nfsd4_op_u *u)
8218 {
8219 struct nfsd4_lockt *lockt = &u->lockt;
8220 struct nfsd4_lock_denied *deny = &lockt->lt_denied;
8221
8222 kfree(deny->ld_owner.data);
8223 }
8224
8225 __be32
nfsd4_locku(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8226 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
8227 union nfsd4_op_u *u)
8228 {
8229 struct nfsd4_locku *locku = &u->locku;
8230 struct nfs4_ol_stateid *stp;
8231 struct nfsd_file *nf = NULL;
8232 struct file_lock *file_lock = NULL;
8233 __be32 status;
8234 int err;
8235 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8236
8237 dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
8238 (long long) locku->lu_offset,
8239 (long long) locku->lu_length);
8240
8241 if (check_lock_length(locku->lu_offset, locku->lu_length))
8242 return nfserr_inval;
8243
8244 status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
8245 &locku->lu_stateid, SC_TYPE_LOCK, 0,
8246 &stp, nn);
8247 if (status)
8248 goto out;
8249 nf = find_any_file(stp->st_stid.sc_file);
8250 if (!nf) {
8251 status = nfserr_lock_range;
8252 goto put_stateid;
8253 }
8254 file_lock = locks_alloc_lock();
8255 if (!file_lock) {
8256 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
8257 status = nfserr_jukebox;
8258 goto put_file;
8259 }
8260
8261 file_lock->c.flc_type = F_UNLCK;
8262 file_lock->c.flc_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
8263 file_lock->c.flc_pid = current->tgid;
8264 file_lock->c.flc_file = nf->nf_file;
8265 file_lock->c.flc_flags = FL_POSIX;
8266 file_lock->fl_lmops = &nfsd_posix_mng_ops;
8267 file_lock->fl_start = locku->lu_offset;
8268
8269 file_lock->fl_end = last_byte_offset(locku->lu_offset,
8270 locku->lu_length);
8271 nfs4_transform_lock_offset(file_lock);
8272
8273 err = vfs_lock_file(nf->nf_file, F_SETLK, file_lock, NULL);
8274 if (err) {
8275 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
8276 goto out_nfserr;
8277 }
8278 nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
8279 put_file:
8280 nfsd_file_put(nf);
8281 put_stateid:
8282 mutex_unlock(&stp->st_mutex);
8283 nfs4_put_stid(&stp->st_stid);
8284 out:
8285 nfsd4_bump_seqid(cstate, status);
8286 if (file_lock)
8287 locks_free_lock(file_lock);
8288 return status;
8289
8290 out_nfserr:
8291 status = nfserrno(err);
8292 goto put_file;
8293 }
8294
8295 /*
8296 * returns
8297 * true: locks held by lockowner
8298 * false: no locks held by lockowner
8299 */
8300 static bool
check_for_locks(struct nfs4_file * fp,struct nfs4_lockowner * lowner)8301 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
8302 {
8303 struct file_lock *fl;
8304 int status = false;
8305 struct nfsd_file *nf;
8306 struct inode *inode;
8307 struct file_lock_context *flctx;
8308
8309 spin_lock(&fp->fi_lock);
8310 nf = find_any_file_locked(fp);
8311 if (!nf) {
8312 /* Any valid lock stateid should have some sort of access */
8313 WARN_ON_ONCE(1);
8314 goto out;
8315 }
8316
8317 inode = file_inode(nf->nf_file);
8318 flctx = locks_inode_context(inode);
8319
8320 if (flctx && !list_empty_careful(&flctx->flc_posix)) {
8321 spin_lock(&flctx->flc_lock);
8322 for_each_file_lock(fl, &flctx->flc_posix) {
8323 if (fl->c.flc_owner == (fl_owner_t)lowner) {
8324 status = true;
8325 break;
8326 }
8327 }
8328 spin_unlock(&flctx->flc_lock);
8329 }
8330 out:
8331 spin_unlock(&fp->fi_lock);
8332 return status;
8333 }
8334
8335 /**
8336 * nfsd4_release_lockowner - process NFSv4.0 RELEASE_LOCKOWNER operations
8337 * @rqstp: RPC transaction
8338 * @cstate: NFSv4 COMPOUND state
8339 * @u: RELEASE_LOCKOWNER arguments
8340 *
8341 * Check if theree are any locks still held and if not - free the lockowner
8342 * and any lock state that is owned.
8343 *
8344 * Return values:
8345 * %nfs_ok: lockowner released or not found
8346 * %nfserr_locks_held: lockowner still in use
8347 * %nfserr_stale_clientid: clientid no longer active
8348 * %nfserr_expired: clientid not recognized
8349 */
8350 __be32
nfsd4_release_lockowner(struct svc_rqst * rqstp,struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8351 nfsd4_release_lockowner(struct svc_rqst *rqstp,
8352 struct nfsd4_compound_state *cstate,
8353 union nfsd4_op_u *u)
8354 {
8355 struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
8356 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8357 clientid_t *clid = &rlockowner->rl_clientid;
8358 struct nfs4_ol_stateid *stp;
8359 struct nfs4_lockowner *lo;
8360 struct nfs4_client *clp;
8361 LIST_HEAD(reaplist);
8362 __be32 status;
8363
8364 dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
8365 clid->cl_boot, clid->cl_id);
8366
8367 status = set_client(clid, cstate, nn);
8368 if (status)
8369 return status;
8370 clp = cstate->clp;
8371
8372 spin_lock(&clp->cl_lock);
8373 lo = find_lockowner_str_locked(clp, &rlockowner->rl_owner);
8374 if (!lo) {
8375 spin_unlock(&clp->cl_lock);
8376 return nfs_ok;
8377 }
8378
8379 list_for_each_entry(stp, &lo->lo_owner.so_stateids, st_perstateowner) {
8380 if (check_for_locks(stp->st_stid.sc_file, lo)) {
8381 spin_unlock(&clp->cl_lock);
8382 nfs4_put_stateowner(&lo->lo_owner);
8383 return nfserr_locks_held;
8384 }
8385 }
8386 unhash_lockowner_locked(lo);
8387 while (!list_empty(&lo->lo_owner.so_stateids)) {
8388 stp = list_first_entry(&lo->lo_owner.so_stateids,
8389 struct nfs4_ol_stateid,
8390 st_perstateowner);
8391 unhash_lock_stateid(stp);
8392 put_ol_stateid_locked(stp, &reaplist);
8393 }
8394 spin_unlock(&clp->cl_lock);
8395
8396 free_ol_stateid_reaplist(&reaplist);
8397 remove_blocked_locks(lo);
8398 nfs4_put_stateowner(&lo->lo_owner);
8399 return nfs_ok;
8400 }
8401
8402 static inline struct nfs4_client_reclaim *
alloc_reclaim(void)8403 alloc_reclaim(void)
8404 {
8405 return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
8406 }
8407
8408 bool
nfs4_has_reclaimed_state(struct xdr_netobj name,struct nfsd_net * nn)8409 nfs4_has_reclaimed_state(struct xdr_netobj name, struct nfsd_net *nn)
8410 {
8411 struct nfs4_client_reclaim *crp;
8412
8413 crp = nfsd4_find_reclaim_client(name, nn);
8414 return (crp && crp->cr_clp);
8415 }
8416
8417 /*
8418 * failure => all reset bets are off, nfserr_no_grace...
8419 *
8420 * The caller is responsible for freeing name.data if NULL is returned (it
8421 * will be freed in nfs4_remove_reclaim_record in the normal case).
8422 */
8423 struct nfs4_client_reclaim *
nfs4_client_to_reclaim(struct xdr_netobj name,struct xdr_netobj princhash,struct nfsd_net * nn)8424 nfs4_client_to_reclaim(struct xdr_netobj name, struct xdr_netobj princhash,
8425 struct nfsd_net *nn)
8426 {
8427 unsigned int strhashval;
8428 struct nfs4_client_reclaim *crp;
8429
8430 crp = alloc_reclaim();
8431 if (crp) {
8432 strhashval = clientstr_hashval(name);
8433 INIT_LIST_HEAD(&crp->cr_strhash);
8434 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
8435 crp->cr_name.data = name.data;
8436 crp->cr_name.len = name.len;
8437 crp->cr_princhash.data = princhash.data;
8438 crp->cr_princhash.len = princhash.len;
8439 crp->cr_clp = NULL;
8440 nn->reclaim_str_hashtbl_size++;
8441 }
8442 return crp;
8443 }
8444
8445 void
nfs4_remove_reclaim_record(struct nfs4_client_reclaim * crp,struct nfsd_net * nn)8446 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
8447 {
8448 list_del(&crp->cr_strhash);
8449 kfree(crp->cr_name.data);
8450 kfree(crp->cr_princhash.data);
8451 kfree(crp);
8452 nn->reclaim_str_hashtbl_size--;
8453 }
8454
8455 void
nfs4_release_reclaim(struct nfsd_net * nn)8456 nfs4_release_reclaim(struct nfsd_net *nn)
8457 {
8458 struct nfs4_client_reclaim *crp = NULL;
8459 int i;
8460
8461 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8462 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
8463 crp = list_entry(nn->reclaim_str_hashtbl[i].next,
8464 struct nfs4_client_reclaim, cr_strhash);
8465 nfs4_remove_reclaim_record(crp, nn);
8466 }
8467 }
8468 WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
8469 }
8470
8471 /*
8472 * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
8473 struct nfs4_client_reclaim *
nfsd4_find_reclaim_client(struct xdr_netobj name,struct nfsd_net * nn)8474 nfsd4_find_reclaim_client(struct xdr_netobj name, struct nfsd_net *nn)
8475 {
8476 unsigned int strhashval;
8477 struct nfs4_client_reclaim *crp = NULL;
8478
8479 strhashval = clientstr_hashval(name);
8480 list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
8481 if (compare_blob(&crp->cr_name, &name) == 0) {
8482 return crp;
8483 }
8484 }
8485 return NULL;
8486 }
8487
8488 __be32
nfs4_check_open_reclaim(struct nfs4_client * clp)8489 nfs4_check_open_reclaim(struct nfs4_client *clp)
8490 {
8491 if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &clp->cl_flags))
8492 return nfserr_no_grace;
8493
8494 if (nfsd4_client_record_check(clp))
8495 return nfserr_reclaim_bad;
8496
8497 return nfs_ok;
8498 }
8499
8500 /*
8501 * Since the lifetime of a delegation isn't limited to that of an open, a
8502 * client may quite reasonably hang on to a delegation as long as it has
8503 * the inode cached. This becomes an obvious problem the first time a
8504 * client's inode cache approaches the size of the server's total memory.
8505 *
8506 * For now we avoid this problem by imposing a hard limit on the number
8507 * of delegations, which varies according to the server's memory size.
8508 */
8509 static void
set_max_delegations(void)8510 set_max_delegations(void)
8511 {
8512 /*
8513 * Allow at most 4 delegations per megabyte of RAM. Quick
8514 * estimates suggest that in the worst case (where every delegation
8515 * is for a different inode), a delegation could take about 1.5K,
8516 * giving a worst case usage of about 6% of memory.
8517 */
8518 max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
8519 }
8520
nfs4_state_create_net(struct net * net)8521 static int nfs4_state_create_net(struct net *net)
8522 {
8523 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8524 int i;
8525
8526 nn->conf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8527 sizeof(struct list_head),
8528 GFP_KERNEL);
8529 if (!nn->conf_id_hashtbl)
8530 goto err;
8531 nn->unconf_id_hashtbl = kmalloc_array(CLIENT_HASH_SIZE,
8532 sizeof(struct list_head),
8533 GFP_KERNEL);
8534 if (!nn->unconf_id_hashtbl)
8535 goto err_unconf_id;
8536 nn->sessionid_hashtbl = kmalloc_array(SESSION_HASH_SIZE,
8537 sizeof(struct list_head),
8538 GFP_KERNEL);
8539 if (!nn->sessionid_hashtbl)
8540 goto err_sessionid;
8541
8542 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8543 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
8544 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
8545 }
8546 for (i = 0; i < SESSION_HASH_SIZE; i++)
8547 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
8548 nn->conf_name_tree = RB_ROOT;
8549 nn->unconf_name_tree = RB_ROOT;
8550 nn->boot_time = ktime_get_real_seconds();
8551 nn->grace_ended = false;
8552 nn->nfsd4_manager.block_opens = true;
8553 INIT_LIST_HEAD(&nn->nfsd4_manager.list);
8554 INIT_LIST_HEAD(&nn->client_lru);
8555 INIT_LIST_HEAD(&nn->close_lru);
8556 INIT_LIST_HEAD(&nn->del_recall_lru);
8557 spin_lock_init(&nn->client_lock);
8558 spin_lock_init(&nn->s2s_cp_lock);
8559 idr_init(&nn->s2s_cp_stateids);
8560
8561 spin_lock_init(&nn->blocked_locks_lock);
8562 INIT_LIST_HEAD(&nn->blocked_locks_lru);
8563
8564 INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
8565 INIT_WORK(&nn->nfsd_shrinker_work, nfsd4_state_shrinker_worker);
8566 get_net(net);
8567
8568 nn->nfsd_client_shrinker = shrinker_alloc(0, "nfsd-client");
8569 if (!nn->nfsd_client_shrinker)
8570 goto err_shrinker;
8571
8572 nn->nfsd_client_shrinker->scan_objects = nfsd4_state_shrinker_scan;
8573 nn->nfsd_client_shrinker->count_objects = nfsd4_state_shrinker_count;
8574 nn->nfsd_client_shrinker->private_data = nn;
8575
8576 shrinker_register(nn->nfsd_client_shrinker);
8577
8578 return 0;
8579
8580 err_shrinker:
8581 put_net(net);
8582 kfree(nn->sessionid_hashtbl);
8583 err_sessionid:
8584 kfree(nn->unconf_id_hashtbl);
8585 err_unconf_id:
8586 kfree(nn->conf_id_hashtbl);
8587 err:
8588 return -ENOMEM;
8589 }
8590
8591 static void
nfs4_state_destroy_net(struct net * net)8592 nfs4_state_destroy_net(struct net *net)
8593 {
8594 int i;
8595 struct nfs4_client *clp = NULL;
8596 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8597
8598 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8599 while (!list_empty(&nn->conf_id_hashtbl[i])) {
8600 clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8601 destroy_client(clp);
8602 }
8603 }
8604
8605 WARN_ON(!list_empty(&nn->blocked_locks_lru));
8606
8607 for (i = 0; i < CLIENT_HASH_SIZE; i++) {
8608 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
8609 clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
8610 destroy_client(clp);
8611 }
8612 }
8613
8614 kfree(nn->sessionid_hashtbl);
8615 kfree(nn->unconf_id_hashtbl);
8616 kfree(nn->conf_id_hashtbl);
8617 put_net(net);
8618 }
8619
8620 int
nfs4_state_start_net(struct net * net)8621 nfs4_state_start_net(struct net *net)
8622 {
8623 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8624 int ret;
8625
8626 ret = nfs4_state_create_net(net);
8627 if (ret)
8628 return ret;
8629 locks_start_grace(net, &nn->nfsd4_manager);
8630 nfsd4_client_tracking_init(net);
8631 if (nn->track_reclaim_completes && nn->reclaim_str_hashtbl_size == 0)
8632 goto skip_grace;
8633 printk(KERN_INFO "NFSD: starting %lld-second grace period (net %x)\n",
8634 nn->nfsd4_grace, net->ns.inum);
8635 trace_nfsd_grace_start(nn);
8636 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
8637 return 0;
8638
8639 skip_grace:
8640 printk(KERN_INFO "NFSD: no clients to reclaim, skipping NFSv4 grace period (net %x)\n",
8641 net->ns.inum);
8642 queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_lease * HZ);
8643 nfsd4_end_grace(nn);
8644 return 0;
8645 }
8646
8647 /* initialization to perform when the nfsd service is started: */
8648
8649 int
nfs4_state_start(void)8650 nfs4_state_start(void)
8651 {
8652 int ret;
8653
8654 ret = rhltable_init(&nfs4_file_rhltable, &nfs4_file_rhash_params);
8655 if (ret)
8656 return ret;
8657
8658 set_max_delegations();
8659 return 0;
8660 }
8661
8662 void
nfs4_state_shutdown_net(struct net * net)8663 nfs4_state_shutdown_net(struct net *net)
8664 {
8665 struct nfs4_delegation *dp = NULL;
8666 struct list_head *pos, *next, reaplist;
8667 struct nfsd_net *nn = net_generic(net, nfsd_net_id);
8668
8669 shrinker_free(nn->nfsd_client_shrinker);
8670 cancel_work(&nn->nfsd_shrinker_work);
8671 cancel_delayed_work_sync(&nn->laundromat_work);
8672 locks_end_grace(&nn->nfsd4_manager);
8673
8674 INIT_LIST_HEAD(&reaplist);
8675 spin_lock(&state_lock);
8676 list_for_each_safe(pos, next, &nn->del_recall_lru) {
8677 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8678 unhash_delegation_locked(dp, SC_STATUS_CLOSED);
8679 list_add(&dp->dl_recall_lru, &reaplist);
8680 }
8681 spin_unlock(&state_lock);
8682 list_for_each_safe(pos, next, &reaplist) {
8683 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
8684 list_del_init(&dp->dl_recall_lru);
8685 destroy_unhashed_deleg(dp);
8686 }
8687
8688 nfsd4_client_tracking_exit(net);
8689 nfs4_state_destroy_net(net);
8690 #ifdef CONFIG_NFSD_V4_2_INTER_SSC
8691 nfsd4_ssc_shutdown_umount(nn);
8692 #endif
8693 }
8694
8695 void
nfs4_state_shutdown(void)8696 nfs4_state_shutdown(void)
8697 {
8698 rhltable_destroy(&nfs4_file_rhltable);
8699 }
8700
8701 static void
get_stateid(struct nfsd4_compound_state * cstate,stateid_t * stateid)8702 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8703 {
8704 if (HAS_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG) &&
8705 CURRENT_STATEID(stateid))
8706 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
8707 }
8708
8709 static void
put_stateid(struct nfsd4_compound_state * cstate,stateid_t * stateid)8710 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
8711 {
8712 if (cstate->minorversion) {
8713 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
8714 SET_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8715 }
8716 }
8717
8718 void
clear_current_stateid(struct nfsd4_compound_state * cstate)8719 clear_current_stateid(struct nfsd4_compound_state *cstate)
8720 {
8721 CLEAR_CSTATE_FLAG(cstate, CURRENT_STATE_ID_FLAG);
8722 }
8723
8724 /*
8725 * functions to set current state id
8726 */
8727 void
nfsd4_set_opendowngradestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8728 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
8729 union nfsd4_op_u *u)
8730 {
8731 put_stateid(cstate, &u->open_downgrade.od_stateid);
8732 }
8733
8734 void
nfsd4_set_openstateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8735 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
8736 union nfsd4_op_u *u)
8737 {
8738 put_stateid(cstate, &u->open.op_stateid);
8739 }
8740
8741 void
nfsd4_set_closestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8742 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
8743 union nfsd4_op_u *u)
8744 {
8745 put_stateid(cstate, &u->close.cl_stateid);
8746 }
8747
8748 void
nfsd4_set_lockstateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8749 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
8750 union nfsd4_op_u *u)
8751 {
8752 put_stateid(cstate, &u->lock.lk_resp_stateid);
8753 }
8754
8755 /*
8756 * functions to consume current state id
8757 */
8758
8759 void
nfsd4_get_opendowngradestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8760 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
8761 union nfsd4_op_u *u)
8762 {
8763 get_stateid(cstate, &u->open_downgrade.od_stateid);
8764 }
8765
8766 void
nfsd4_get_delegreturnstateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8767 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
8768 union nfsd4_op_u *u)
8769 {
8770 get_stateid(cstate, &u->delegreturn.dr_stateid);
8771 }
8772
8773 void
nfsd4_get_freestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8774 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
8775 union nfsd4_op_u *u)
8776 {
8777 get_stateid(cstate, &u->free_stateid.fr_stateid);
8778 }
8779
8780 void
nfsd4_get_setattrstateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8781 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
8782 union nfsd4_op_u *u)
8783 {
8784 get_stateid(cstate, &u->setattr.sa_stateid);
8785 }
8786
8787 void
nfsd4_get_closestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8788 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
8789 union nfsd4_op_u *u)
8790 {
8791 get_stateid(cstate, &u->close.cl_stateid);
8792 }
8793
8794 void
nfsd4_get_lockustateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8795 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
8796 union nfsd4_op_u *u)
8797 {
8798 get_stateid(cstate, &u->locku.lu_stateid);
8799 }
8800
8801 void
nfsd4_get_readstateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8802 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
8803 union nfsd4_op_u *u)
8804 {
8805 get_stateid(cstate, &u->read.rd_stateid);
8806 }
8807
8808 void
nfsd4_get_writestateid(struct nfsd4_compound_state * cstate,union nfsd4_op_u * u)8809 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
8810 union nfsd4_op_u *u)
8811 {
8812 get_stateid(cstate, &u->write.wr_stateid);
8813 }
8814
8815 /**
8816 * nfsd4_deleg_getattr_conflict - Recall if GETATTR causes conflict
8817 * @rqstp: RPC transaction context
8818 * @dentry: dentry of inode to be checked for a conflict
8819 * @modified: return true if file was modified
8820 * @size: new size of file if modified is true
8821 *
8822 * This function is called when there is a conflict between a write
8823 * delegation and a change/size GETATTR from another client. The server
8824 * must either use the CB_GETATTR to get the current values of the
8825 * attributes from the client that holds the delegation or recall the
8826 * delegation before replying to the GETATTR. See RFC 8881 section
8827 * 18.7.4.
8828 *
8829 * Returns 0 if there is no conflict; otherwise an nfs_stat
8830 * code is returned.
8831 */
8832 __be32
nfsd4_deleg_getattr_conflict(struct svc_rqst * rqstp,struct dentry * dentry,bool * modified,u64 * size)8833 nfsd4_deleg_getattr_conflict(struct svc_rqst *rqstp, struct dentry *dentry,
8834 bool *modified, u64 *size)
8835 {
8836 __be32 status;
8837 struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
8838 struct file_lock_context *ctx;
8839 struct file_lease *fl;
8840 struct iattr attrs;
8841 struct nfs4_cb_fattr *ncf;
8842 struct inode *inode = d_inode(dentry);
8843
8844 *modified = false;
8845 ctx = locks_inode_context(inode);
8846 if (!ctx)
8847 return 0;
8848 spin_lock(&ctx->flc_lock);
8849 for_each_file_lock(fl, &ctx->flc_lease) {
8850 unsigned char type = fl->c.flc_type;
8851
8852 if (fl->c.flc_flags == FL_LAYOUT)
8853 continue;
8854 if (fl->fl_lmops != &nfsd_lease_mng_ops) {
8855 /*
8856 * non-nfs lease, if it's a lease with F_RDLCK then
8857 * we are done; there isn't any write delegation
8858 * on this inode
8859 */
8860 if (type == F_RDLCK)
8861 break;
8862
8863 nfsd_stats_wdeleg_getattr_inc(nn);
8864 spin_unlock(&ctx->flc_lock);
8865
8866 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8867 if (status != nfserr_jukebox ||
8868 !nfsd_wait_for_delegreturn(rqstp, inode))
8869 return status;
8870 return 0;
8871 }
8872 if (type == F_WRLCK) {
8873 struct nfs4_delegation *dp = fl->c.flc_owner;
8874
8875 if (dp->dl_recall.cb_clp == *(rqstp->rq_lease_breaker)) {
8876 spin_unlock(&ctx->flc_lock);
8877 return 0;
8878 }
8879 nfsd_stats_wdeleg_getattr_inc(nn);
8880 dp = fl->c.flc_owner;
8881 refcount_inc(&dp->dl_stid.sc_count);
8882 ncf = &dp->dl_cb_fattr;
8883 nfs4_cb_getattr(&dp->dl_cb_fattr);
8884 spin_unlock(&ctx->flc_lock);
8885 wait_on_bit_timeout(&ncf->ncf_cb_flags, CB_GETATTR_BUSY,
8886 TASK_INTERRUPTIBLE, NFSD_CB_GETATTR_TIMEOUT);
8887 if (ncf->ncf_cb_status) {
8888 /* Recall delegation only if client didn't respond */
8889 status = nfserrno(nfsd_open_break_lease(inode, NFSD_MAY_READ));
8890 if (status != nfserr_jukebox ||
8891 !nfsd_wait_for_delegreturn(rqstp, inode)) {
8892 nfs4_put_stid(&dp->dl_stid);
8893 return status;
8894 }
8895 }
8896 if (!ncf->ncf_file_modified &&
8897 (ncf->ncf_initial_cinfo != ncf->ncf_cb_change ||
8898 ncf->ncf_cur_fsize != ncf->ncf_cb_fsize))
8899 ncf->ncf_file_modified = true;
8900 if (ncf->ncf_file_modified) {
8901 int err;
8902
8903 /*
8904 * Per section 10.4.3 of RFC 8881, the server would
8905 * not update the file's metadata with the client's
8906 * modified size
8907 */
8908 attrs.ia_mtime = attrs.ia_ctime = current_time(inode);
8909 attrs.ia_valid = ATTR_MTIME | ATTR_CTIME | ATTR_DELEG;
8910 inode_lock(inode);
8911 err = notify_change(&nop_mnt_idmap, dentry, &attrs, NULL);
8912 inode_unlock(inode);
8913 if (err) {
8914 nfs4_put_stid(&dp->dl_stid);
8915 return nfserrno(err);
8916 }
8917 ncf->ncf_cur_fsize = ncf->ncf_cb_fsize;
8918 *size = ncf->ncf_cur_fsize;
8919 *modified = true;
8920 }
8921 nfs4_put_stid(&dp->dl_stid);
8922 return 0;
8923 }
8924 break;
8925 }
8926 spin_unlock(&ctx->flc_lock);
8927 return 0;
8928 }
8929