Merge tag 'nfsd-4.17' of git://linux-nfs.org/~bfields/linux
[muen/linux.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
45 #include "xdr4.h"
46 #include "xdr4cb.h"
47 #include "vfs.h"
48 #include "current_stateid.h"
49
50 #include "netns.h"
51 #include "pnfs.h"
52
53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
54
55 #define all_ones {{~0,~0},~0}
56 static const stateid_t one_stateid = {
57         .si_generation = ~0,
58         .si_opaque = all_ones,
59 };
60 static const stateid_t zero_stateid = {
61         /* all fields zero */
62 };
63 static const stateid_t currentstateid = {
64         .si_generation = 1,
65 };
66 static const stateid_t close_stateid = {
67         .si_generation = 0xffffffffU,
68 };
69
70 static u64 current_sessionid = 1;
71
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
76
77 /* forward declarations */
78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
80
81 /* Locking: */
82
83 /*
84  * Currently used for the del_recall_lru and file hash table.  In an
85  * effort to decrease the scope of the client_mutex, this spinlock may
86  * eventually cover more:
87  */
88 static DEFINE_SPINLOCK(state_lock);
89
90 enum nfsd4_st_mutex_lock_subclass {
91         OPEN_STATEID_MUTEX = 0,
92         LOCK_STATEID_MUTEX = 1,
93 };
94
95 /*
96  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
97  * the refcount on the open stateid to drop.
98  */
99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
100
101 static struct kmem_cache *client_slab;
102 static struct kmem_cache *openowner_slab;
103 static struct kmem_cache *lockowner_slab;
104 static struct kmem_cache *file_slab;
105 static struct kmem_cache *stateid_slab;
106 static struct kmem_cache *deleg_slab;
107 static struct kmem_cache *odstate_slab;
108
109 static void free_session(struct nfsd4_session *);
110
111 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
112 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
113
114 static bool is_session_dead(struct nfsd4_session *ses)
115 {
116         return ses->se_flags & NFS4_SESSION_DEAD;
117 }
118
119 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
120 {
121         if (atomic_read(&ses->se_ref) > ref_held_by_me)
122                 return nfserr_jukebox;
123         ses->se_flags |= NFS4_SESSION_DEAD;
124         return nfs_ok;
125 }
126
127 static bool is_client_expired(struct nfs4_client *clp)
128 {
129         return clp->cl_time == 0;
130 }
131
132 static __be32 get_client_locked(struct nfs4_client *clp)
133 {
134         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
135
136         lockdep_assert_held(&nn->client_lock);
137
138         if (is_client_expired(clp))
139                 return nfserr_expired;
140         atomic_inc(&clp->cl_refcount);
141         return nfs_ok;
142 }
143
144 /* must be called under the client_lock */
145 static inline void
146 renew_client_locked(struct nfs4_client *clp)
147 {
148         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
149
150         if (is_client_expired(clp)) {
151                 WARN_ON(1);
152                 printk("%s: client (clientid %08x/%08x) already expired\n",
153                         __func__,
154                         clp->cl_clientid.cl_boot,
155                         clp->cl_clientid.cl_id);
156                 return;
157         }
158
159         dprintk("renewing client (clientid %08x/%08x)\n",
160                         clp->cl_clientid.cl_boot,
161                         clp->cl_clientid.cl_id);
162         list_move_tail(&clp->cl_lru, &nn->client_lru);
163         clp->cl_time = get_seconds();
164 }
165
166 static void put_client_renew_locked(struct nfs4_client *clp)
167 {
168         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
169
170         lockdep_assert_held(&nn->client_lock);
171
172         if (!atomic_dec_and_test(&clp->cl_refcount))
173                 return;
174         if (!is_client_expired(clp))
175                 renew_client_locked(clp);
176 }
177
178 static void put_client_renew(struct nfs4_client *clp)
179 {
180         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
181
182         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
183                 return;
184         if (!is_client_expired(clp))
185                 renew_client_locked(clp);
186         spin_unlock(&nn->client_lock);
187 }
188
189 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
190 {
191         __be32 status;
192
193         if (is_session_dead(ses))
194                 return nfserr_badsession;
195         status = get_client_locked(ses->se_client);
196         if (status)
197                 return status;
198         atomic_inc(&ses->se_ref);
199         return nfs_ok;
200 }
201
202 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
203 {
204         struct nfs4_client *clp = ses->se_client;
205         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
206
207         lockdep_assert_held(&nn->client_lock);
208
209         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
210                 free_session(ses);
211         put_client_renew_locked(clp);
212 }
213
214 static void nfsd4_put_session(struct nfsd4_session *ses)
215 {
216         struct nfs4_client *clp = ses->se_client;
217         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
218
219         spin_lock(&nn->client_lock);
220         nfsd4_put_session_locked(ses);
221         spin_unlock(&nn->client_lock);
222 }
223
224 static struct nfsd4_blocked_lock *
225 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
226                         struct nfsd_net *nn)
227 {
228         struct nfsd4_blocked_lock *cur, *found = NULL;
229
230         spin_lock(&nn->blocked_locks_lock);
231         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
232                 if (fh_match(fh, &cur->nbl_fh)) {
233                         list_del_init(&cur->nbl_list);
234                         list_del_init(&cur->nbl_lru);
235                         found = cur;
236                         break;
237                 }
238         }
239         spin_unlock(&nn->blocked_locks_lock);
240         if (found)
241                 posix_unblock_lock(&found->nbl_lock);
242         return found;
243 }
244
245 static struct nfsd4_blocked_lock *
246 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
247                         struct nfsd_net *nn)
248 {
249         struct nfsd4_blocked_lock *nbl;
250
251         nbl = find_blocked_lock(lo, fh, nn);
252         if (!nbl) {
253                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
254                 if (nbl) {
255                         fh_copy_shallow(&nbl->nbl_fh, fh);
256                         locks_init_lock(&nbl->nbl_lock);
257                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
258                                         &nfsd4_cb_notify_lock_ops,
259                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
260                 }
261         }
262         return nbl;
263 }
264
265 static void
266 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
267 {
268         locks_release_private(&nbl->nbl_lock);
269         kfree(nbl);
270 }
271
272 static void
273 remove_blocked_locks(struct nfs4_lockowner *lo)
274 {
275         struct nfs4_client *clp = lo->lo_owner.so_client;
276         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
277         struct nfsd4_blocked_lock *nbl;
278         LIST_HEAD(reaplist);
279
280         /* Dequeue all blocked locks */
281         spin_lock(&nn->blocked_locks_lock);
282         while (!list_empty(&lo->lo_blocked)) {
283                 nbl = list_first_entry(&lo->lo_blocked,
284                                         struct nfsd4_blocked_lock,
285                                         nbl_list);
286                 list_del_init(&nbl->nbl_list);
287                 list_move(&nbl->nbl_lru, &reaplist);
288         }
289         spin_unlock(&nn->blocked_locks_lock);
290
291         /* Now free them */
292         while (!list_empty(&reaplist)) {
293                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
294                                         nbl_lru);
295                 list_del_init(&nbl->nbl_lru);
296                 posix_unblock_lock(&nbl->nbl_lock);
297                 free_blocked_lock(nbl);
298         }
299 }
300
301 static int
302 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
303 {
304         /*
305          * Since this is just an optimization, we don't try very hard if it
306          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
307          * just quit trying on anything else.
308          */
309         switch (task->tk_status) {
310         case -NFS4ERR_DELAY:
311                 rpc_delay(task, 1 * HZ);
312                 return 0;
313         default:
314                 return 1;
315         }
316 }
317
318 static void
319 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
320 {
321         struct nfsd4_blocked_lock       *nbl = container_of(cb,
322                                                 struct nfsd4_blocked_lock, nbl_cb);
323
324         free_blocked_lock(nbl);
325 }
326
327 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
328         .done           = nfsd4_cb_notify_lock_done,
329         .release        = nfsd4_cb_notify_lock_release,
330 };
331
332 static inline struct nfs4_stateowner *
333 nfs4_get_stateowner(struct nfs4_stateowner *sop)
334 {
335         atomic_inc(&sop->so_count);
336         return sop;
337 }
338
339 static int
340 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
341 {
342         return (sop->so_owner.len == owner->len) &&
343                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
344 }
345
346 static struct nfs4_openowner *
347 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
348                         struct nfs4_client *clp)
349 {
350         struct nfs4_stateowner *so;
351
352         lockdep_assert_held(&clp->cl_lock);
353
354         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
355                             so_strhash) {
356                 if (!so->so_is_open_owner)
357                         continue;
358                 if (same_owner_str(so, &open->op_owner))
359                         return openowner(nfs4_get_stateowner(so));
360         }
361         return NULL;
362 }
363
364 static struct nfs4_openowner *
365 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
366                         struct nfs4_client *clp)
367 {
368         struct nfs4_openowner *oo;
369
370         spin_lock(&clp->cl_lock);
371         oo = find_openstateowner_str_locked(hashval, open, clp);
372         spin_unlock(&clp->cl_lock);
373         return oo;
374 }
375
376 static inline u32
377 opaque_hashval(const void *ptr, int nbytes)
378 {
379         unsigned char *cptr = (unsigned char *) ptr;
380
381         u32 x = 0;
382         while (nbytes--) {
383                 x *= 37;
384                 x += *cptr++;
385         }
386         return x;
387 }
388
389 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
390 {
391         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
392
393         kmem_cache_free(file_slab, fp);
394 }
395
396 void
397 put_nfs4_file(struct nfs4_file *fi)
398 {
399         might_lock(&state_lock);
400
401         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
402                 hlist_del_rcu(&fi->fi_hash);
403                 spin_unlock(&state_lock);
404                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
405                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
406                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
407         }
408 }
409
410 static struct file *
411 __nfs4_get_fd(struct nfs4_file *f, int oflag)
412 {
413         if (f->fi_fds[oflag])
414                 return get_file(f->fi_fds[oflag]);
415         return NULL;
416 }
417
418 static struct file *
419 find_writeable_file_locked(struct nfs4_file *f)
420 {
421         struct file *ret;
422
423         lockdep_assert_held(&f->fi_lock);
424
425         ret = __nfs4_get_fd(f, O_WRONLY);
426         if (!ret)
427                 ret = __nfs4_get_fd(f, O_RDWR);
428         return ret;
429 }
430
431 static struct file *
432 find_writeable_file(struct nfs4_file *f)
433 {
434         struct file *ret;
435
436         spin_lock(&f->fi_lock);
437         ret = find_writeable_file_locked(f);
438         spin_unlock(&f->fi_lock);
439
440         return ret;
441 }
442
443 static struct file *find_readable_file_locked(struct nfs4_file *f)
444 {
445         struct file *ret;
446
447         lockdep_assert_held(&f->fi_lock);
448
449         ret = __nfs4_get_fd(f, O_RDONLY);
450         if (!ret)
451                 ret = __nfs4_get_fd(f, O_RDWR);
452         return ret;
453 }
454
455 static struct file *
456 find_readable_file(struct nfs4_file *f)
457 {
458         struct file *ret;
459
460         spin_lock(&f->fi_lock);
461         ret = find_readable_file_locked(f);
462         spin_unlock(&f->fi_lock);
463
464         return ret;
465 }
466
467 struct file *
468 find_any_file(struct nfs4_file *f)
469 {
470         struct file *ret;
471
472         spin_lock(&f->fi_lock);
473         ret = __nfs4_get_fd(f, O_RDWR);
474         if (!ret) {
475                 ret = __nfs4_get_fd(f, O_WRONLY);
476                 if (!ret)
477                         ret = __nfs4_get_fd(f, O_RDONLY);
478         }
479         spin_unlock(&f->fi_lock);
480         return ret;
481 }
482
483 static atomic_long_t num_delegations;
484 unsigned long max_delegations;
485
486 /*
487  * Open owner state (share locks)
488  */
489
490 /* hash tables for lock and open owners */
491 #define OWNER_HASH_BITS              8
492 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
493 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
494
495 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
496 {
497         unsigned int ret;
498
499         ret = opaque_hashval(ownername->data, ownername->len);
500         return ret & OWNER_HASH_MASK;
501 }
502
503 /* hash table for nfs4_file */
504 #define FILE_HASH_BITS                   8
505 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
506
507 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
508 {
509         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
510 }
511
512 static unsigned int file_hashval(struct knfsd_fh *fh)
513 {
514         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
515 }
516
517 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
518
519 static void
520 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
521 {
522         lockdep_assert_held(&fp->fi_lock);
523
524         if (access & NFS4_SHARE_ACCESS_WRITE)
525                 atomic_inc(&fp->fi_access[O_WRONLY]);
526         if (access & NFS4_SHARE_ACCESS_READ)
527                 atomic_inc(&fp->fi_access[O_RDONLY]);
528 }
529
530 static __be32
531 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
532 {
533         lockdep_assert_held(&fp->fi_lock);
534
535         /* Does this access mode make sense? */
536         if (access & ~NFS4_SHARE_ACCESS_BOTH)
537                 return nfserr_inval;
538
539         /* Does it conflict with a deny mode already set? */
540         if ((access & fp->fi_share_deny) != 0)
541                 return nfserr_share_denied;
542
543         __nfs4_file_get_access(fp, access);
544         return nfs_ok;
545 }
546
547 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
548 {
549         /* Common case is that there is no deny mode. */
550         if (deny) {
551                 /* Does this deny mode make sense? */
552                 if (deny & ~NFS4_SHARE_DENY_BOTH)
553                         return nfserr_inval;
554
555                 if ((deny & NFS4_SHARE_DENY_READ) &&
556                     atomic_read(&fp->fi_access[O_RDONLY]))
557                         return nfserr_share_denied;
558
559                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
560                     atomic_read(&fp->fi_access[O_WRONLY]))
561                         return nfserr_share_denied;
562         }
563         return nfs_ok;
564 }
565
566 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
567 {
568         might_lock(&fp->fi_lock);
569
570         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
571                 struct file *f1 = NULL;
572                 struct file *f2 = NULL;
573
574                 swap(f1, fp->fi_fds[oflag]);
575                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
576                         swap(f2, fp->fi_fds[O_RDWR]);
577                 spin_unlock(&fp->fi_lock);
578                 if (f1)
579                         fput(f1);
580                 if (f2)
581                         fput(f2);
582         }
583 }
584
585 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
586 {
587         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
588
589         if (access & NFS4_SHARE_ACCESS_WRITE)
590                 __nfs4_file_put_access(fp, O_WRONLY);
591         if (access & NFS4_SHARE_ACCESS_READ)
592                 __nfs4_file_put_access(fp, O_RDONLY);
593 }
594
595 /*
596  * Allocate a new open/delegation state counter. This is needed for
597  * pNFS for proper return on close semantics.
598  *
599  * Note that we only allocate it for pNFS-enabled exports, otherwise
600  * all pointers to struct nfs4_clnt_odstate are always NULL.
601  */
602 static struct nfs4_clnt_odstate *
603 alloc_clnt_odstate(struct nfs4_client *clp)
604 {
605         struct nfs4_clnt_odstate *co;
606
607         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
608         if (co) {
609                 co->co_client = clp;
610                 refcount_set(&co->co_odcount, 1);
611         }
612         return co;
613 }
614
615 static void
616 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
617 {
618         struct nfs4_file *fp = co->co_file;
619
620         lockdep_assert_held(&fp->fi_lock);
621         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
622 }
623
624 static inline void
625 get_clnt_odstate(struct nfs4_clnt_odstate *co)
626 {
627         if (co)
628                 refcount_inc(&co->co_odcount);
629 }
630
631 static void
632 put_clnt_odstate(struct nfs4_clnt_odstate *co)
633 {
634         struct nfs4_file *fp;
635
636         if (!co)
637                 return;
638
639         fp = co->co_file;
640         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
641                 list_del(&co->co_perfile);
642                 spin_unlock(&fp->fi_lock);
643
644                 nfsd4_return_all_file_layouts(co->co_client, fp);
645                 kmem_cache_free(odstate_slab, co);
646         }
647 }
648
649 static struct nfs4_clnt_odstate *
650 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
651 {
652         struct nfs4_clnt_odstate *co;
653         struct nfs4_client *cl;
654
655         if (!new)
656                 return NULL;
657
658         cl = new->co_client;
659
660         spin_lock(&fp->fi_lock);
661         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
662                 if (co->co_client == cl) {
663                         get_clnt_odstate(co);
664                         goto out;
665                 }
666         }
667         co = new;
668         co->co_file = fp;
669         hash_clnt_odstate_locked(new);
670 out:
671         spin_unlock(&fp->fi_lock);
672         return co;
673 }
674
675 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
676                                   void (*sc_free)(struct nfs4_stid *))
677 {
678         struct nfs4_stid *stid;
679         int new_id;
680
681         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
682         if (!stid)
683                 return NULL;
684
685         idr_preload(GFP_KERNEL);
686         spin_lock(&cl->cl_lock);
687         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
688         spin_unlock(&cl->cl_lock);
689         idr_preload_end();
690         if (new_id < 0)
691                 goto out_free;
692
693         stid->sc_free = sc_free;
694         stid->sc_client = cl;
695         stid->sc_stateid.si_opaque.so_id = new_id;
696         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
697         /* Will be incremented before return to client: */
698         refcount_set(&stid->sc_count, 1);
699         spin_lock_init(&stid->sc_lock);
700
701         /*
702          * It shouldn't be a problem to reuse an opaque stateid value.
703          * I don't think it is for 4.1.  But with 4.0 I worry that, for
704          * example, a stray write retransmission could be accepted by
705          * the server when it should have been rejected.  Therefore,
706          * adopt a trick from the sctp code to attempt to maximize the
707          * amount of time until an id is reused, by ensuring they always
708          * "increase" (mod INT_MAX):
709          */
710         return stid;
711 out_free:
712         kmem_cache_free(slab, stid);
713         return NULL;
714 }
715
716 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
717 {
718         struct nfs4_stid *stid;
719
720         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
721         if (!stid)
722                 return NULL;
723
724         return openlockstateid(stid);
725 }
726
727 static void nfs4_free_deleg(struct nfs4_stid *stid)
728 {
729         kmem_cache_free(deleg_slab, stid);
730         atomic_long_dec(&num_delegations);
731 }
732
733 /*
734  * When we recall a delegation, we should be careful not to hand it
735  * out again straight away.
736  * To ensure this we keep a pair of bloom filters ('new' and 'old')
737  * in which the filehandles of recalled delegations are "stored".
738  * If a filehandle appear in either filter, a delegation is blocked.
739  * When a delegation is recalled, the filehandle is stored in the "new"
740  * filter.
741  * Every 30 seconds we swap the filters and clear the "new" one,
742  * unless both are empty of course.
743  *
744  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
745  * low 3 bytes as hash-table indices.
746  *
747  * 'blocked_delegations_lock', which is always taken in block_delegations(),
748  * is used to manage concurrent access.  Testing does not need the lock
749  * except when swapping the two filters.
750  */
751 static DEFINE_SPINLOCK(blocked_delegations_lock);
752 static struct bloom_pair {
753         int     entries, old_entries;
754         time_t  swap_time;
755         int     new; /* index into 'set' */
756         DECLARE_BITMAP(set[2], 256);
757 } blocked_delegations;
758
759 static int delegation_blocked(struct knfsd_fh *fh)
760 {
761         u32 hash;
762         struct bloom_pair *bd = &blocked_delegations;
763
764         if (bd->entries == 0)
765                 return 0;
766         if (seconds_since_boot() - bd->swap_time > 30) {
767                 spin_lock(&blocked_delegations_lock);
768                 if (seconds_since_boot() - bd->swap_time > 30) {
769                         bd->entries -= bd->old_entries;
770                         bd->old_entries = bd->entries;
771                         memset(bd->set[bd->new], 0,
772                                sizeof(bd->set[0]));
773                         bd->new = 1-bd->new;
774                         bd->swap_time = seconds_since_boot();
775                 }
776                 spin_unlock(&blocked_delegations_lock);
777         }
778         hash = jhash(&fh->fh_base, fh->fh_size, 0);
779         if (test_bit(hash&255, bd->set[0]) &&
780             test_bit((hash>>8)&255, bd->set[0]) &&
781             test_bit((hash>>16)&255, bd->set[0]))
782                 return 1;
783
784         if (test_bit(hash&255, bd->set[1]) &&
785             test_bit((hash>>8)&255, bd->set[1]) &&
786             test_bit((hash>>16)&255, bd->set[1]))
787                 return 1;
788
789         return 0;
790 }
791
792 static void block_delegations(struct knfsd_fh *fh)
793 {
794         u32 hash;
795         struct bloom_pair *bd = &blocked_delegations;
796
797         hash = jhash(&fh->fh_base, fh->fh_size, 0);
798
799         spin_lock(&blocked_delegations_lock);
800         __set_bit(hash&255, bd->set[bd->new]);
801         __set_bit((hash>>8)&255, bd->set[bd->new]);
802         __set_bit((hash>>16)&255, bd->set[bd->new]);
803         if (bd->entries == 0)
804                 bd->swap_time = seconds_since_boot();
805         bd->entries += 1;
806         spin_unlock(&blocked_delegations_lock);
807 }
808
809 static struct nfs4_delegation *
810 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_file *fp,
811                  struct svc_fh *current_fh,
812                  struct nfs4_clnt_odstate *odstate)
813 {
814         struct nfs4_delegation *dp;
815         long n;
816
817         dprintk("NFSD alloc_init_deleg\n");
818         n = atomic_long_inc_return(&num_delegations);
819         if (n < 0 || n > max_delegations)
820                 goto out_dec;
821         if (delegation_blocked(&current_fh->fh_handle))
822                 goto out_dec;
823         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
824         if (dp == NULL)
825                 goto out_dec;
826
827         /*
828          * delegation seqid's are never incremented.  The 4.1 special
829          * meaning of seqid 0 isn't meaningful, really, but let's avoid
830          * 0 anyway just for consistency and use 1:
831          */
832         dp->dl_stid.sc_stateid.si_generation = 1;
833         INIT_LIST_HEAD(&dp->dl_perfile);
834         INIT_LIST_HEAD(&dp->dl_perclnt);
835         INIT_LIST_HEAD(&dp->dl_recall_lru);
836         dp->dl_clnt_odstate = odstate;
837         get_clnt_odstate(odstate);
838         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
839         dp->dl_retries = 1;
840         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
841                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
842         get_nfs4_file(fp);
843         dp->dl_stid.sc_file = fp;
844         return dp;
845 out_dec:
846         atomic_long_dec(&num_delegations);
847         return NULL;
848 }
849
850 void
851 nfs4_put_stid(struct nfs4_stid *s)
852 {
853         struct nfs4_file *fp = s->sc_file;
854         struct nfs4_client *clp = s->sc_client;
855
856         might_lock(&clp->cl_lock);
857
858         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
859                 wake_up_all(&close_wq);
860                 return;
861         }
862         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
863         spin_unlock(&clp->cl_lock);
864         s->sc_free(s);
865         if (fp)
866                 put_nfs4_file(fp);
867 }
868
869 void
870 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
871 {
872         stateid_t *src = &stid->sc_stateid;
873
874         spin_lock(&stid->sc_lock);
875         if (unlikely(++src->si_generation == 0))
876                 src->si_generation = 1;
877         memcpy(dst, src, sizeof(*dst));
878         spin_unlock(&stid->sc_lock);
879 }
880
881 static void put_deleg_file(struct nfs4_file *fp)
882 {
883         struct file *filp = NULL;
884
885         spin_lock(&fp->fi_lock);
886         if (--fp->fi_delegees == 0)
887                 swap(filp, fp->fi_deleg_file);
888         spin_unlock(&fp->fi_lock);
889
890         if (filp)
891                 fput(filp);
892 }
893
894 static void nfs4_unlock_deleg_lease(struct nfs4_delegation *dp)
895 {
896         struct nfs4_file *fp = dp->dl_stid.sc_file;
897         struct file *filp = fp->fi_deleg_file;
898
899         WARN_ON_ONCE(!fp->fi_delegees);
900
901         vfs_setlease(filp, F_UNLCK, NULL, (void **)&dp);
902         put_deleg_file(fp);
903 }
904
905 static void destroy_unhashed_deleg(struct nfs4_delegation *dp)
906 {
907         put_clnt_odstate(dp->dl_clnt_odstate);
908         nfs4_unlock_deleg_lease(dp);
909         nfs4_put_stid(&dp->dl_stid);
910 }
911
912 void nfs4_unhash_stid(struct nfs4_stid *s)
913 {
914         s->sc_type = 0;
915 }
916
917 /**
918  * nfs4_delegation_exists - Discover if this delegation already exists
919  * @clp:     a pointer to the nfs4_client we're granting a delegation to
920  * @fp:      a pointer to the nfs4_file we're granting a delegation on
921  *
922  * Return:
923  *      On success: true iff an existing delegation is found
924  */
925
926 static bool
927 nfs4_delegation_exists(struct nfs4_client *clp, struct nfs4_file *fp)
928 {
929         struct nfs4_delegation *searchdp = NULL;
930         struct nfs4_client *searchclp = NULL;
931
932         lockdep_assert_held(&state_lock);
933         lockdep_assert_held(&fp->fi_lock);
934
935         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
936                 searchclp = searchdp->dl_stid.sc_client;
937                 if (clp == searchclp) {
938                         return true;
939                 }
940         }
941         return false;
942 }
943
944 /**
945  * hash_delegation_locked - Add a delegation to the appropriate lists
946  * @dp:     a pointer to the nfs4_delegation we are adding.
947  * @fp:     a pointer to the nfs4_file we're granting a delegation on
948  *
949  * Return:
950  *      On success: NULL if the delegation was successfully hashed.
951  *
952  *      On error: -EAGAIN if one was previously granted to this
953  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
954  *
955  */
956
957 static int
958 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
959 {
960         struct nfs4_client *clp = dp->dl_stid.sc_client;
961
962         lockdep_assert_held(&state_lock);
963         lockdep_assert_held(&fp->fi_lock);
964
965         if (nfs4_delegation_exists(clp, fp))
966                 return -EAGAIN;
967         refcount_inc(&dp->dl_stid.sc_count);
968         dp->dl_stid.sc_type = NFS4_DELEG_STID;
969         list_add(&dp->dl_perfile, &fp->fi_delegations);
970         list_add(&dp->dl_perclnt, &clp->cl_delegations);
971         return 0;
972 }
973
974 static bool
975 unhash_delegation_locked(struct nfs4_delegation *dp)
976 {
977         struct nfs4_file *fp = dp->dl_stid.sc_file;
978
979         lockdep_assert_held(&state_lock);
980
981         if (list_empty(&dp->dl_perfile))
982                 return false;
983
984         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
985         /* Ensure that deleg break won't try to requeue it */
986         ++dp->dl_time;
987         spin_lock(&fp->fi_lock);
988         list_del_init(&dp->dl_perclnt);
989         list_del_init(&dp->dl_recall_lru);
990         list_del_init(&dp->dl_perfile);
991         spin_unlock(&fp->fi_lock);
992         return true;
993 }
994
995 static void destroy_delegation(struct nfs4_delegation *dp)
996 {
997         bool unhashed;
998
999         spin_lock(&state_lock);
1000         unhashed = unhash_delegation_locked(dp);
1001         spin_unlock(&state_lock);
1002         if (unhashed)
1003                 destroy_unhashed_deleg(dp);
1004 }
1005
1006 static void revoke_delegation(struct nfs4_delegation *dp)
1007 {
1008         struct nfs4_client *clp = dp->dl_stid.sc_client;
1009
1010         WARN_ON(!list_empty(&dp->dl_recall_lru));
1011
1012         if (clp->cl_minorversion) {
1013                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1014                 refcount_inc(&dp->dl_stid.sc_count);
1015                 spin_lock(&clp->cl_lock);
1016                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1017                 spin_unlock(&clp->cl_lock);
1018         }
1019         destroy_unhashed_deleg(dp);
1020 }
1021
1022 /* 
1023  * SETCLIENTID state 
1024  */
1025
1026 static unsigned int clientid_hashval(u32 id)
1027 {
1028         return id & CLIENT_HASH_MASK;
1029 }
1030
1031 static unsigned int clientstr_hashval(const char *name)
1032 {
1033         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1034 }
1035
1036 /*
1037  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1038  * st_{access,deny}_bmap field of the stateid, in order to track not
1039  * only what share bits are currently in force, but also what
1040  * combinations of share bits previous opens have used.  This allows us
1041  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1042  * return an error if the client attempt to downgrade to a combination
1043  * of share bits not explicable by closing some of its previous opens.
1044  *
1045  * XXX: This enforcement is actually incomplete, since we don't keep
1046  * track of access/deny bit combinations; so, e.g., we allow:
1047  *
1048  *      OPEN allow read, deny write
1049  *      OPEN allow both, deny none
1050  *      DOWNGRADE allow read, deny none
1051  *
1052  * which we should reject.
1053  */
1054 static unsigned int
1055 bmap_to_share_mode(unsigned long bmap) {
1056         int i;
1057         unsigned int access = 0;
1058
1059         for (i = 1; i < 4; i++) {
1060                 if (test_bit(i, &bmap))
1061                         access |= i;
1062         }
1063         return access;
1064 }
1065
1066 /* set share access for a given stateid */
1067 static inline void
1068 set_access(u32 access, struct nfs4_ol_stateid *stp)
1069 {
1070         unsigned char mask = 1 << access;
1071
1072         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1073         stp->st_access_bmap |= mask;
1074 }
1075
1076 /* clear share access for a given stateid */
1077 static inline void
1078 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1079 {
1080         unsigned char mask = 1 << access;
1081
1082         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1083         stp->st_access_bmap &= ~mask;
1084 }
1085
1086 /* test whether a given stateid has access */
1087 static inline bool
1088 test_access(u32 access, struct nfs4_ol_stateid *stp)
1089 {
1090         unsigned char mask = 1 << access;
1091
1092         return (bool)(stp->st_access_bmap & mask);
1093 }
1094
1095 /* set share deny for a given stateid */
1096 static inline void
1097 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1098 {
1099         unsigned char mask = 1 << deny;
1100
1101         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1102         stp->st_deny_bmap |= mask;
1103 }
1104
1105 /* clear share deny for a given stateid */
1106 static inline void
1107 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1108 {
1109         unsigned char mask = 1 << deny;
1110
1111         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1112         stp->st_deny_bmap &= ~mask;
1113 }
1114
1115 /* test whether a given stateid is denying specific access */
1116 static inline bool
1117 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1118 {
1119         unsigned char mask = 1 << deny;
1120
1121         return (bool)(stp->st_deny_bmap & mask);
1122 }
1123
1124 static int nfs4_access_to_omode(u32 access)
1125 {
1126         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1127         case NFS4_SHARE_ACCESS_READ:
1128                 return O_RDONLY;
1129         case NFS4_SHARE_ACCESS_WRITE:
1130                 return O_WRONLY;
1131         case NFS4_SHARE_ACCESS_BOTH:
1132                 return O_RDWR;
1133         }
1134         WARN_ON_ONCE(1);
1135         return O_RDONLY;
1136 }
1137
1138 /*
1139  * A stateid that had a deny mode associated with it is being released
1140  * or downgraded. Recalculate the deny mode on the file.
1141  */
1142 static void
1143 recalculate_deny_mode(struct nfs4_file *fp)
1144 {
1145         struct nfs4_ol_stateid *stp;
1146
1147         spin_lock(&fp->fi_lock);
1148         fp->fi_share_deny = 0;
1149         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1150                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1151         spin_unlock(&fp->fi_lock);
1152 }
1153
1154 static void
1155 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1156 {
1157         int i;
1158         bool change = false;
1159
1160         for (i = 1; i < 4; i++) {
1161                 if ((i & deny) != i) {
1162                         change = true;
1163                         clear_deny(i, stp);
1164                 }
1165         }
1166
1167         /* Recalculate per-file deny mode if there was a change */
1168         if (change)
1169                 recalculate_deny_mode(stp->st_stid.sc_file);
1170 }
1171
1172 /* release all access and file references for a given stateid */
1173 static void
1174 release_all_access(struct nfs4_ol_stateid *stp)
1175 {
1176         int i;
1177         struct nfs4_file *fp = stp->st_stid.sc_file;
1178
1179         if (fp && stp->st_deny_bmap != 0)
1180                 recalculate_deny_mode(fp);
1181
1182         for (i = 1; i < 4; i++) {
1183                 if (test_access(i, stp))
1184                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1185                 clear_access(i, stp);
1186         }
1187 }
1188
1189 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1190 {
1191         kfree(sop->so_owner.data);
1192         sop->so_ops->so_free(sop);
1193 }
1194
1195 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1196 {
1197         struct nfs4_client *clp = sop->so_client;
1198
1199         might_lock(&clp->cl_lock);
1200
1201         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1202                 return;
1203         sop->so_ops->so_unhash(sop);
1204         spin_unlock(&clp->cl_lock);
1205         nfs4_free_stateowner(sop);
1206 }
1207
1208 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1209 {
1210         struct nfs4_file *fp = stp->st_stid.sc_file;
1211
1212         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1213
1214         if (list_empty(&stp->st_perfile))
1215                 return false;
1216
1217         spin_lock(&fp->fi_lock);
1218         list_del_init(&stp->st_perfile);
1219         spin_unlock(&fp->fi_lock);
1220         list_del(&stp->st_perstateowner);
1221         return true;
1222 }
1223
1224 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1225 {
1226         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1227
1228         put_clnt_odstate(stp->st_clnt_odstate);
1229         release_all_access(stp);
1230         if (stp->st_stateowner)
1231                 nfs4_put_stateowner(stp->st_stateowner);
1232         kmem_cache_free(stateid_slab, stid);
1233 }
1234
1235 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1236 {
1237         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1238         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1239         struct file *file;
1240
1241         file = find_any_file(stp->st_stid.sc_file);
1242         if (file)
1243                 filp_close(file, (fl_owner_t)lo);
1244         nfs4_free_ol_stateid(stid);
1245 }
1246
1247 /*
1248  * Put the persistent reference to an already unhashed generic stateid, while
1249  * holding the cl_lock. If it's the last reference, then put it onto the
1250  * reaplist for later destruction.
1251  */
1252 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1253                                        struct list_head *reaplist)
1254 {
1255         struct nfs4_stid *s = &stp->st_stid;
1256         struct nfs4_client *clp = s->sc_client;
1257
1258         lockdep_assert_held(&clp->cl_lock);
1259
1260         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1261
1262         if (!refcount_dec_and_test(&s->sc_count)) {
1263                 wake_up_all(&close_wq);
1264                 return;
1265         }
1266
1267         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1268         list_add(&stp->st_locks, reaplist);
1269 }
1270
1271 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1272 {
1273         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1274
1275         list_del_init(&stp->st_locks);
1276         nfs4_unhash_stid(&stp->st_stid);
1277         return unhash_ol_stateid(stp);
1278 }
1279
1280 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1281 {
1282         struct nfs4_client *clp = stp->st_stid.sc_client;
1283         bool unhashed;
1284
1285         spin_lock(&clp->cl_lock);
1286         unhashed = unhash_lock_stateid(stp);
1287         spin_unlock(&clp->cl_lock);
1288         if (unhashed)
1289                 nfs4_put_stid(&stp->st_stid);
1290 }
1291
1292 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1293 {
1294         struct nfs4_client *clp = lo->lo_owner.so_client;
1295
1296         lockdep_assert_held(&clp->cl_lock);
1297
1298         list_del_init(&lo->lo_owner.so_strhash);
1299 }
1300
1301 /*
1302  * Free a list of generic stateids that were collected earlier after being
1303  * fully unhashed.
1304  */
1305 static void
1306 free_ol_stateid_reaplist(struct list_head *reaplist)
1307 {
1308         struct nfs4_ol_stateid *stp;
1309         struct nfs4_file *fp;
1310
1311         might_sleep();
1312
1313         while (!list_empty(reaplist)) {
1314                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1315                                        st_locks);
1316                 list_del(&stp->st_locks);
1317                 fp = stp->st_stid.sc_file;
1318                 stp->st_stid.sc_free(&stp->st_stid);
1319                 if (fp)
1320                         put_nfs4_file(fp);
1321         }
1322 }
1323
1324 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1325                                        struct list_head *reaplist)
1326 {
1327         struct nfs4_ol_stateid *stp;
1328
1329         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1330
1331         while (!list_empty(&open_stp->st_locks)) {
1332                 stp = list_entry(open_stp->st_locks.next,
1333                                 struct nfs4_ol_stateid, st_locks);
1334                 WARN_ON(!unhash_lock_stateid(stp));
1335                 put_ol_stateid_locked(stp, reaplist);
1336         }
1337 }
1338
1339 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1340                                 struct list_head *reaplist)
1341 {
1342         bool unhashed;
1343
1344         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1345
1346         unhashed = unhash_ol_stateid(stp);
1347         release_open_stateid_locks(stp, reaplist);
1348         return unhashed;
1349 }
1350
1351 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1352 {
1353         LIST_HEAD(reaplist);
1354
1355         spin_lock(&stp->st_stid.sc_client->cl_lock);
1356         if (unhash_open_stateid(stp, &reaplist))
1357                 put_ol_stateid_locked(stp, &reaplist);
1358         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1359         free_ol_stateid_reaplist(&reaplist);
1360 }
1361
1362 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1363 {
1364         struct nfs4_client *clp = oo->oo_owner.so_client;
1365
1366         lockdep_assert_held(&clp->cl_lock);
1367
1368         list_del_init(&oo->oo_owner.so_strhash);
1369         list_del_init(&oo->oo_perclient);
1370 }
1371
1372 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1373 {
1374         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1375                                           nfsd_net_id);
1376         struct nfs4_ol_stateid *s;
1377
1378         spin_lock(&nn->client_lock);
1379         s = oo->oo_last_closed_stid;
1380         if (s) {
1381                 list_del_init(&oo->oo_close_lru);
1382                 oo->oo_last_closed_stid = NULL;
1383         }
1384         spin_unlock(&nn->client_lock);
1385         if (s)
1386                 nfs4_put_stid(&s->st_stid);
1387 }
1388
1389 static void release_openowner(struct nfs4_openowner *oo)
1390 {
1391         struct nfs4_ol_stateid *stp;
1392         struct nfs4_client *clp = oo->oo_owner.so_client;
1393         struct list_head reaplist;
1394
1395         INIT_LIST_HEAD(&reaplist);
1396
1397         spin_lock(&clp->cl_lock);
1398         unhash_openowner_locked(oo);
1399         while (!list_empty(&oo->oo_owner.so_stateids)) {
1400                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1401                                 struct nfs4_ol_stateid, st_perstateowner);
1402                 if (unhash_open_stateid(stp, &reaplist))
1403                         put_ol_stateid_locked(stp, &reaplist);
1404         }
1405         spin_unlock(&clp->cl_lock);
1406         free_ol_stateid_reaplist(&reaplist);
1407         release_last_closed_stateid(oo);
1408         nfs4_put_stateowner(&oo->oo_owner);
1409 }
1410
1411 static inline int
1412 hash_sessionid(struct nfs4_sessionid *sessionid)
1413 {
1414         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1415
1416         return sid->sequence % SESSION_HASH_SIZE;
1417 }
1418
1419 #ifdef CONFIG_SUNRPC_DEBUG
1420 static inline void
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1422 {
1423         u32 *ptr = (u32 *)(&sessionid->data[0]);
1424         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1425 }
1426 #else
1427 static inline void
1428 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1429 {
1430 }
1431 #endif
1432
1433 /*
1434  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1435  * won't be used for replay.
1436  */
1437 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1438 {
1439         struct nfs4_stateowner *so = cstate->replay_owner;
1440
1441         if (nfserr == nfserr_replay_me)
1442                 return;
1443
1444         if (!seqid_mutating_err(ntohl(nfserr))) {
1445                 nfsd4_cstate_clear_replay(cstate);
1446                 return;
1447         }
1448         if (!so)
1449                 return;
1450         if (so->so_is_open_owner)
1451                 release_last_closed_stateid(openowner(so));
1452         so->so_seqid++;
1453         return;
1454 }
1455
1456 static void
1457 gen_sessionid(struct nfsd4_session *ses)
1458 {
1459         struct nfs4_client *clp = ses->se_client;
1460         struct nfsd4_sessionid *sid;
1461
1462         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1463         sid->clientid = clp->cl_clientid;
1464         sid->sequence = current_sessionid++;
1465         sid->reserved = 0;
1466 }
1467
1468 /*
1469  * The protocol defines ca_maxresponssize_cached to include the size of
1470  * the rpc header, but all we need to cache is the data starting after
1471  * the end of the initial SEQUENCE operation--the rest we regenerate
1472  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1473  * value that is the number of bytes in our cache plus a few additional
1474  * bytes.  In order to stay on the safe side, and not promise more than
1475  * we can cache, those additional bytes must be the minimum possible: 24
1476  * bytes of rpc header (xid through accept state, with AUTH_NULL
1477  * verifier), 12 for the compound header (with zero-length tag), and 44
1478  * for the SEQUENCE op response:
1479  */
1480 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1481
1482 static void
1483 free_session_slots(struct nfsd4_session *ses)
1484 {
1485         int i;
1486
1487         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1488                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1489                 kfree(ses->se_slots[i]);
1490         }
1491 }
1492
1493 /*
1494  * We don't actually need to cache the rpc and session headers, so we
1495  * can allocate a little less for each slot:
1496  */
1497 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1498 {
1499         u32 size;
1500
1501         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1502                 size = 0;
1503         else
1504                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1505         return size + sizeof(struct nfsd4_slot);
1506 }
1507
1508 /*
1509  * XXX: If we run out of reserved DRC memory we could (up to a point)
1510  * re-negotiate active sessions and reduce their slot usage to make
1511  * room for new connections. For now we just fail the create session.
1512  */
1513 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1514 {
1515         u32 slotsize = slot_bytes(ca);
1516         u32 num = ca->maxreqs;
1517         int avail;
1518
1519         spin_lock(&nfsd_drc_lock);
1520         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1521                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1522         /*
1523          * Never use more than a third of the remaining memory,
1524          * unless it's the only way to give this client a slot:
1525          */
1526         avail = clamp_t(int, avail, slotsize, avail/3);
1527         num = min_t(int, num, avail / slotsize);
1528         nfsd_drc_mem_used += num * slotsize;
1529         spin_unlock(&nfsd_drc_lock);
1530
1531         return num;
1532 }
1533
1534 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1535 {
1536         int slotsize = slot_bytes(ca);
1537
1538         spin_lock(&nfsd_drc_lock);
1539         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1540         spin_unlock(&nfsd_drc_lock);
1541 }
1542
1543 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1544                                            struct nfsd4_channel_attrs *battrs)
1545 {
1546         int numslots = fattrs->maxreqs;
1547         int slotsize = slot_bytes(fattrs);
1548         struct nfsd4_session *new;
1549         int mem, i;
1550
1551         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1552                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1553         mem = numslots * sizeof(struct nfsd4_slot *);
1554
1555         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1556         if (!new)
1557                 return NULL;
1558         /* allocate each struct nfsd4_slot and data cache in one piece */
1559         for (i = 0; i < numslots; i++) {
1560                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1561                 if (!new->se_slots[i])
1562                         goto out_free;
1563         }
1564
1565         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1566         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1567
1568         return new;
1569 out_free:
1570         while (i--)
1571                 kfree(new->se_slots[i]);
1572         kfree(new);
1573         return NULL;
1574 }
1575
1576 static void free_conn(struct nfsd4_conn *c)
1577 {
1578         svc_xprt_put(c->cn_xprt);
1579         kfree(c);
1580 }
1581
1582 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1583 {
1584         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1585         struct nfs4_client *clp = c->cn_session->se_client;
1586
1587         spin_lock(&clp->cl_lock);
1588         if (!list_empty(&c->cn_persession)) {
1589                 list_del(&c->cn_persession);
1590                 free_conn(c);
1591         }
1592         nfsd4_probe_callback(clp);
1593         spin_unlock(&clp->cl_lock);
1594 }
1595
1596 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1597 {
1598         struct nfsd4_conn *conn;
1599
1600         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1601         if (!conn)
1602                 return NULL;
1603         svc_xprt_get(rqstp->rq_xprt);
1604         conn->cn_xprt = rqstp->rq_xprt;
1605         conn->cn_flags = flags;
1606         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1607         return conn;
1608 }
1609
1610 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1611 {
1612         conn->cn_session = ses;
1613         list_add(&conn->cn_persession, &ses->se_conns);
1614 }
1615
1616 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1617 {
1618         struct nfs4_client *clp = ses->se_client;
1619
1620         spin_lock(&clp->cl_lock);
1621         __nfsd4_hash_conn(conn, ses);
1622         spin_unlock(&clp->cl_lock);
1623 }
1624
1625 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1626 {
1627         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1628         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1629 }
1630
1631 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1632 {
1633         int ret;
1634
1635         nfsd4_hash_conn(conn, ses);
1636         ret = nfsd4_register_conn(conn);
1637         if (ret)
1638                 /* oops; xprt is already down: */
1639                 nfsd4_conn_lost(&conn->cn_xpt_user);
1640         /* We may have gained or lost a callback channel: */
1641         nfsd4_probe_callback_sync(ses->se_client);
1642 }
1643
1644 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1645 {
1646         u32 dir = NFS4_CDFC4_FORE;
1647
1648         if (cses->flags & SESSION4_BACK_CHAN)
1649                 dir |= NFS4_CDFC4_BACK;
1650         return alloc_conn(rqstp, dir);
1651 }
1652
1653 /* must be called under client_lock */
1654 static void nfsd4_del_conns(struct nfsd4_session *s)
1655 {
1656         struct nfs4_client *clp = s->se_client;
1657         struct nfsd4_conn *c;
1658
1659         spin_lock(&clp->cl_lock);
1660         while (!list_empty(&s->se_conns)) {
1661                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1662                 list_del_init(&c->cn_persession);
1663                 spin_unlock(&clp->cl_lock);
1664
1665                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1666                 free_conn(c);
1667
1668                 spin_lock(&clp->cl_lock);
1669         }
1670         spin_unlock(&clp->cl_lock);
1671 }
1672
1673 static void __free_session(struct nfsd4_session *ses)
1674 {
1675         free_session_slots(ses);
1676         kfree(ses);
1677 }
1678
1679 static void free_session(struct nfsd4_session *ses)
1680 {
1681         nfsd4_del_conns(ses);
1682         nfsd4_put_drc_mem(&ses->se_fchannel);
1683         __free_session(ses);
1684 }
1685
1686 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1687 {
1688         int idx;
1689         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1690
1691         new->se_client = clp;
1692         gen_sessionid(new);
1693
1694         INIT_LIST_HEAD(&new->se_conns);
1695
1696         new->se_cb_seq_nr = 1;
1697         new->se_flags = cses->flags;
1698         new->se_cb_prog = cses->callback_prog;
1699         new->se_cb_sec = cses->cb_sec;
1700         atomic_set(&new->se_ref, 0);
1701         idx = hash_sessionid(&new->se_sessionid);
1702         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1703         spin_lock(&clp->cl_lock);
1704         list_add(&new->se_perclnt, &clp->cl_sessions);
1705         spin_unlock(&clp->cl_lock);
1706
1707         {
1708                 struct sockaddr *sa = svc_addr(rqstp);
1709                 /*
1710                  * This is a little silly; with sessions there's no real
1711                  * use for the callback address.  Use the peer address
1712                  * as a reasonable default for now, but consider fixing
1713                  * the rpc client not to require an address in the
1714                  * future:
1715                  */
1716                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1717                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1718         }
1719 }
1720
1721 /* caller must hold client_lock */
1722 static struct nfsd4_session *
1723 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1724 {
1725         struct nfsd4_session *elem;
1726         int idx;
1727         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1728
1729         lockdep_assert_held(&nn->client_lock);
1730
1731         dump_sessionid(__func__, sessionid);
1732         idx = hash_sessionid(sessionid);
1733         /* Search in the appropriate list */
1734         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1735                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1736                             NFS4_MAX_SESSIONID_LEN)) {
1737                         return elem;
1738                 }
1739         }
1740
1741         dprintk("%s: session not found\n", __func__);
1742         return NULL;
1743 }
1744
1745 static struct nfsd4_session *
1746 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1747                 __be32 *ret)
1748 {
1749         struct nfsd4_session *session;
1750         __be32 status = nfserr_badsession;
1751
1752         session = __find_in_sessionid_hashtbl(sessionid, net);
1753         if (!session)
1754                 goto out;
1755         status = nfsd4_get_session_locked(session);
1756         if (status)
1757                 session = NULL;
1758 out:
1759         *ret = status;
1760         return session;
1761 }
1762
1763 /* caller must hold client_lock */
1764 static void
1765 unhash_session(struct nfsd4_session *ses)
1766 {
1767         struct nfs4_client *clp = ses->se_client;
1768         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1769
1770         lockdep_assert_held(&nn->client_lock);
1771
1772         list_del(&ses->se_hash);
1773         spin_lock(&ses->se_client->cl_lock);
1774         list_del(&ses->se_perclnt);
1775         spin_unlock(&ses->se_client->cl_lock);
1776 }
1777
1778 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1779 static int
1780 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1781 {
1782         /*
1783          * We're assuming the clid was not given out from a boot
1784          * precisely 2^32 (about 136 years) before this one.  That seems
1785          * a safe assumption:
1786          */
1787         if (clid->cl_boot == (u32)nn->boot_time)
1788                 return 0;
1789         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1790                 clid->cl_boot, clid->cl_id, nn->boot_time);
1791         return 1;
1792 }
1793
1794 /* 
1795  * XXX Should we use a slab cache ?
1796  * This type of memory management is somewhat inefficient, but we use it
1797  * anyway since SETCLIENTID is not a common operation.
1798  */
1799 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1800 {
1801         struct nfs4_client *clp;
1802         int i;
1803
1804         clp = kmem_cache_zalloc(client_slab, GFP_KERNEL);
1805         if (clp == NULL)
1806                 return NULL;
1807         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1808         if (clp->cl_name.data == NULL)
1809                 goto err_no_name;
1810         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1811                         OWNER_HASH_SIZE, GFP_KERNEL);
1812         if (!clp->cl_ownerstr_hashtbl)
1813                 goto err_no_hashtbl;
1814         for (i = 0; i < OWNER_HASH_SIZE; i++)
1815                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1816         clp->cl_name.len = name.len;
1817         INIT_LIST_HEAD(&clp->cl_sessions);
1818         idr_init(&clp->cl_stateids);
1819         atomic_set(&clp->cl_refcount, 0);
1820         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1821         INIT_LIST_HEAD(&clp->cl_idhash);
1822         INIT_LIST_HEAD(&clp->cl_openowners);
1823         INIT_LIST_HEAD(&clp->cl_delegations);
1824         INIT_LIST_HEAD(&clp->cl_lru);
1825         INIT_LIST_HEAD(&clp->cl_revoked);
1826 #ifdef CONFIG_NFSD_PNFS
1827         INIT_LIST_HEAD(&clp->cl_lo_states);
1828 #endif
1829         spin_lock_init(&clp->cl_lock);
1830         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1831         return clp;
1832 err_no_hashtbl:
1833         kfree(clp->cl_name.data);
1834 err_no_name:
1835         kmem_cache_free(client_slab, clp);
1836         return NULL;
1837 }
1838
1839 static void
1840 free_client(struct nfs4_client *clp)
1841 {
1842         while (!list_empty(&clp->cl_sessions)) {
1843                 struct nfsd4_session *ses;
1844                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1845                                 se_perclnt);
1846                 list_del(&ses->se_perclnt);
1847                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1848                 free_session(ses);
1849         }
1850         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1851         free_svc_cred(&clp->cl_cred);
1852         kfree(clp->cl_ownerstr_hashtbl);
1853         kfree(clp->cl_name.data);
1854         idr_destroy(&clp->cl_stateids);
1855         kmem_cache_free(client_slab, clp);
1856 }
1857
1858 /* must be called under the client_lock */
1859 static void
1860 unhash_client_locked(struct nfs4_client *clp)
1861 {
1862         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1863         struct nfsd4_session *ses;
1864
1865         lockdep_assert_held(&nn->client_lock);
1866
1867         /* Mark the client as expired! */
1868         clp->cl_time = 0;
1869         /* Make it invisible */
1870         if (!list_empty(&clp->cl_idhash)) {
1871                 list_del_init(&clp->cl_idhash);
1872                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1873                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1874                 else
1875                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1876         }
1877         list_del_init(&clp->cl_lru);
1878         spin_lock(&clp->cl_lock);
1879         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1880                 list_del_init(&ses->se_hash);
1881         spin_unlock(&clp->cl_lock);
1882 }
1883
1884 static void
1885 unhash_client(struct nfs4_client *clp)
1886 {
1887         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1888
1889         spin_lock(&nn->client_lock);
1890         unhash_client_locked(clp);
1891         spin_unlock(&nn->client_lock);
1892 }
1893
1894 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1895 {
1896         if (atomic_read(&clp->cl_refcount))
1897                 return nfserr_jukebox;
1898         unhash_client_locked(clp);
1899         return nfs_ok;
1900 }
1901
1902 static void
1903 __destroy_client(struct nfs4_client *clp)
1904 {
1905         int i;
1906         struct nfs4_openowner *oo;
1907         struct nfs4_delegation *dp;
1908         struct list_head reaplist;
1909
1910         INIT_LIST_HEAD(&reaplist);
1911         spin_lock(&state_lock);
1912         while (!list_empty(&clp->cl_delegations)) {
1913                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1914                 WARN_ON(!unhash_delegation_locked(dp));
1915                 list_add(&dp->dl_recall_lru, &reaplist);
1916         }
1917         spin_unlock(&state_lock);
1918         while (!list_empty(&reaplist)) {
1919                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1920                 list_del_init(&dp->dl_recall_lru);
1921                 destroy_unhashed_deleg(dp);
1922         }
1923         while (!list_empty(&clp->cl_revoked)) {
1924                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1925                 list_del_init(&dp->dl_recall_lru);
1926                 nfs4_put_stid(&dp->dl_stid);
1927         }
1928         while (!list_empty(&clp->cl_openowners)) {
1929                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1930                 nfs4_get_stateowner(&oo->oo_owner);
1931                 release_openowner(oo);
1932         }
1933         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1934                 struct nfs4_stateowner *so, *tmp;
1935
1936                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1937                                          so_strhash) {
1938                         /* Should be no openowners at this point */
1939                         WARN_ON_ONCE(so->so_is_open_owner);
1940                         remove_blocked_locks(lockowner(so));
1941                 }
1942         }
1943         nfsd4_return_all_client_layouts(clp);
1944         nfsd4_shutdown_callback(clp);
1945         if (clp->cl_cb_conn.cb_xprt)
1946                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1947         free_client(clp);
1948 }
1949
1950 static void
1951 destroy_client(struct nfs4_client *clp)
1952 {
1953         unhash_client(clp);
1954         __destroy_client(clp);
1955 }
1956
1957 static void expire_client(struct nfs4_client *clp)
1958 {
1959         unhash_client(clp);
1960         nfsd4_client_record_remove(clp);
1961         __destroy_client(clp);
1962 }
1963
1964 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1965 {
1966         memcpy(target->cl_verifier.data, source->data,
1967                         sizeof(target->cl_verifier.data));
1968 }
1969
1970 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1971 {
1972         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1973         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1974 }
1975
1976 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1977 {
1978         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1979         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1980                                                                 GFP_KERNEL);
1981         if ((source->cr_principal && ! target->cr_principal) ||
1982             (source->cr_raw_principal && ! target->cr_raw_principal))
1983                 return -ENOMEM;
1984
1985         target->cr_flavor = source->cr_flavor;
1986         target->cr_uid = source->cr_uid;
1987         target->cr_gid = source->cr_gid;
1988         target->cr_group_info = source->cr_group_info;
1989         get_group_info(target->cr_group_info);
1990         target->cr_gss_mech = source->cr_gss_mech;
1991         if (source->cr_gss_mech)
1992                 gss_mech_get(source->cr_gss_mech);
1993         return 0;
1994 }
1995
1996 static int
1997 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1998 {
1999         if (o1->len < o2->len)
2000                 return -1;
2001         if (o1->len > o2->len)
2002                 return 1;
2003         return memcmp(o1->data, o2->data, o1->len);
2004 }
2005
2006 static int same_name(const char *n1, const char *n2)
2007 {
2008         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2009 }
2010
2011 static int
2012 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2013 {
2014         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2015 }
2016
2017 static int
2018 same_clid(clientid_t *cl1, clientid_t *cl2)
2019 {
2020         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2021 }
2022
2023 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2024 {
2025         int i;
2026
2027         if (g1->ngroups != g2->ngroups)
2028                 return false;
2029         for (i=0; i<g1->ngroups; i++)
2030                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2031                         return false;
2032         return true;
2033 }
2034
2035 /*
2036  * RFC 3530 language requires clid_inuse be returned when the
2037  * "principal" associated with a requests differs from that previously
2038  * used.  We use uid, gid's, and gss principal string as our best
2039  * approximation.  We also don't want to allow non-gss use of a client
2040  * established using gss: in theory cr_principal should catch that
2041  * change, but in practice cr_principal can be null even in the gss case
2042  * since gssd doesn't always pass down a principal string.
2043  */
2044 static bool is_gss_cred(struct svc_cred *cr)
2045 {
2046         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2047         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2048 }
2049
2050
2051 static bool
2052 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2053 {
2054         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2055                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2056                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2057                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2058                 return false;
2059         if (cr1->cr_principal == cr2->cr_principal)
2060                 return true;
2061         if (!cr1->cr_principal || !cr2->cr_principal)
2062                 return false;
2063         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2064 }
2065
2066 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2067 {
2068         struct svc_cred *cr = &rqstp->rq_cred;
2069         u32 service;
2070
2071         if (!cr->cr_gss_mech)
2072                 return false;
2073         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2074         return service == RPC_GSS_SVC_INTEGRITY ||
2075                service == RPC_GSS_SVC_PRIVACY;
2076 }
2077
2078 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2079 {
2080         struct svc_cred *cr = &rqstp->rq_cred;
2081
2082         if (!cl->cl_mach_cred)
2083                 return true;
2084         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2085                 return false;
2086         if (!svc_rqst_integrity_protected(rqstp))
2087                 return false;
2088         if (cl->cl_cred.cr_raw_principal)
2089                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2090                                                 cr->cr_raw_principal);
2091         if (!cr->cr_principal)
2092                 return false;
2093         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2094 }
2095
2096 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2097 {
2098         __be32 verf[2];
2099
2100         /*
2101          * This is opaque to client, so no need to byte-swap. Use
2102          * __force to keep sparse happy
2103          */
2104         verf[0] = (__force __be32)get_seconds();
2105         verf[1] = (__force __be32)nn->clverifier_counter++;
2106         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2107 }
2108
2109 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2110 {
2111         clp->cl_clientid.cl_boot = nn->boot_time;
2112         clp->cl_clientid.cl_id = nn->clientid_counter++;
2113         gen_confirm(clp, nn);
2114 }
2115
2116 static struct nfs4_stid *
2117 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2118 {
2119         struct nfs4_stid *ret;
2120
2121         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2122         if (!ret || !ret->sc_type)
2123                 return NULL;
2124         return ret;
2125 }
2126
2127 static struct nfs4_stid *
2128 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2129 {
2130         struct nfs4_stid *s;
2131
2132         spin_lock(&cl->cl_lock);
2133         s = find_stateid_locked(cl, t);
2134         if (s != NULL) {
2135                 if (typemask & s->sc_type)
2136                         refcount_inc(&s->sc_count);
2137                 else
2138                         s = NULL;
2139         }
2140         spin_unlock(&cl->cl_lock);
2141         return s;
2142 }
2143
2144 static struct nfs4_client *create_client(struct xdr_netobj name,
2145                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2146 {
2147         struct nfs4_client *clp;
2148         struct sockaddr *sa = svc_addr(rqstp);
2149         int ret;
2150         struct net *net = SVC_NET(rqstp);
2151
2152         clp = alloc_client(name);
2153         if (clp == NULL)
2154                 return NULL;
2155
2156         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2157         if (ret) {
2158                 free_client(clp);
2159                 return NULL;
2160         }
2161         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2162         clp->cl_time = get_seconds();
2163         clear_bit(0, &clp->cl_cb_slot_busy);
2164         copy_verf(clp, verf);
2165         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2166         clp->cl_cb_session = NULL;
2167         clp->net = net;
2168         return clp;
2169 }
2170
2171 static void
2172 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2173 {
2174         struct rb_node **new = &(root->rb_node), *parent = NULL;
2175         struct nfs4_client *clp;
2176
2177         while (*new) {
2178                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2179                 parent = *new;
2180
2181                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2182                         new = &((*new)->rb_left);
2183                 else
2184                         new = &((*new)->rb_right);
2185         }
2186
2187         rb_link_node(&new_clp->cl_namenode, parent, new);
2188         rb_insert_color(&new_clp->cl_namenode, root);
2189 }
2190
2191 static struct nfs4_client *
2192 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2193 {
2194         int cmp;
2195         struct rb_node *node = root->rb_node;
2196         struct nfs4_client *clp;
2197
2198         while (node) {
2199                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2200                 cmp = compare_blob(&clp->cl_name, name);
2201                 if (cmp > 0)
2202                         node = node->rb_left;
2203                 else if (cmp < 0)
2204                         node = node->rb_right;
2205                 else
2206                         return clp;
2207         }
2208         return NULL;
2209 }
2210
2211 static void
2212 add_to_unconfirmed(struct nfs4_client *clp)
2213 {
2214         unsigned int idhashval;
2215         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2216
2217         lockdep_assert_held(&nn->client_lock);
2218
2219         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2220         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2221         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2222         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2223         renew_client_locked(clp);
2224 }
2225
2226 static void
2227 move_to_confirmed(struct nfs4_client *clp)
2228 {
2229         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2230         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2231
2232         lockdep_assert_held(&nn->client_lock);
2233
2234         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2235         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2236         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2237         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2238         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2239         renew_client_locked(clp);
2240 }
2241
2242 static struct nfs4_client *
2243 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2244 {
2245         struct nfs4_client *clp;
2246         unsigned int idhashval = clientid_hashval(clid->cl_id);
2247
2248         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2249                 if (same_clid(&clp->cl_clientid, clid)) {
2250                         if ((bool)clp->cl_minorversion != sessions)
2251                                 return NULL;
2252                         renew_client_locked(clp);
2253                         return clp;
2254                 }
2255         }
2256         return NULL;
2257 }
2258
2259 static struct nfs4_client *
2260 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2261 {
2262         struct list_head *tbl = nn->conf_id_hashtbl;
2263
2264         lockdep_assert_held(&nn->client_lock);
2265         return find_client_in_id_table(tbl, clid, sessions);
2266 }
2267
2268 static struct nfs4_client *
2269 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2270 {
2271         struct list_head *tbl = nn->unconf_id_hashtbl;
2272
2273         lockdep_assert_held(&nn->client_lock);
2274         return find_client_in_id_table(tbl, clid, sessions);
2275 }
2276
2277 static bool clp_used_exchangeid(struct nfs4_client *clp)
2278 {
2279         return clp->cl_exchange_flags != 0;
2280
2281
2282 static struct nfs4_client *
2283 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2284 {
2285         lockdep_assert_held(&nn->client_lock);
2286         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2287 }
2288
2289 static struct nfs4_client *
2290 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2291 {
2292         lockdep_assert_held(&nn->client_lock);
2293         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2294 }
2295
2296 static void
2297 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2298 {
2299         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2300         struct sockaddr *sa = svc_addr(rqstp);
2301         u32 scopeid = rpc_get_scope_id(sa);
2302         unsigned short expected_family;
2303
2304         /* Currently, we only support tcp and tcp6 for the callback channel */
2305         if (se->se_callback_netid_len == 3 &&
2306             !memcmp(se->se_callback_netid_val, "tcp", 3))
2307                 expected_family = AF_INET;
2308         else if (se->se_callback_netid_len == 4 &&
2309                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2310                 expected_family = AF_INET6;
2311         else
2312                 goto out_err;
2313
2314         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2315                                             se->se_callback_addr_len,
2316                                             (struct sockaddr *)&conn->cb_addr,
2317                                             sizeof(conn->cb_addr));
2318
2319         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2320                 goto out_err;
2321
2322         if (conn->cb_addr.ss_family == AF_INET6)
2323                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2324
2325         conn->cb_prog = se->se_callback_prog;
2326         conn->cb_ident = se->se_callback_ident;
2327         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2328         return;
2329 out_err:
2330         conn->cb_addr.ss_family = AF_UNSPEC;
2331         conn->cb_addrlen = 0;
2332         dprintk("NFSD: this client (clientid %08x/%08x) "
2333                 "will not receive delegations\n",
2334                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2335
2336         return;
2337 }
2338
2339 /*
2340  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2341  */
2342 static void
2343 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2344 {
2345         struct xdr_buf *buf = resp->xdr.buf;
2346         struct nfsd4_slot *slot = resp->cstate.slot;
2347         unsigned int base;
2348
2349         dprintk("--> %s slot %p\n", __func__, slot);
2350
2351         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2352         slot->sl_opcnt = resp->opcnt;
2353         slot->sl_status = resp->cstate.status;
2354         free_svc_cred(&slot->sl_cred);
2355         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2356
2357         if (!nfsd4_cache_this(resp)) {
2358                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2359                 return;
2360         }
2361         slot->sl_flags |= NFSD4_SLOT_CACHED;
2362
2363         base = resp->cstate.data_offset;
2364         slot->sl_datalen = buf->len - base;
2365         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2366                 WARN(1, "%s: sessions DRC could not cache compound\n",
2367                      __func__);
2368         return;
2369 }
2370
2371 /*
2372  * Encode the replay sequence operation from the slot values.
2373  * If cachethis is FALSE encode the uncached rep error on the next
2374  * operation which sets resp->p and increments resp->opcnt for
2375  * nfs4svc_encode_compoundres.
2376  *
2377  */
2378 static __be32
2379 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2380                           struct nfsd4_compoundres *resp)
2381 {
2382         struct nfsd4_op *op;
2383         struct nfsd4_slot *slot = resp->cstate.slot;
2384
2385         /* Encode the replayed sequence operation */
2386         op = &args->ops[resp->opcnt - 1];
2387         nfsd4_encode_operation(resp, op);
2388
2389         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2390                 return op->status;
2391         if (args->opcnt == 1) {
2392                 /*
2393                  * The original operation wasn't a solo sequence--we
2394                  * always cache those--so this retry must not match the
2395                  * original:
2396                  */
2397                 op->status = nfserr_seq_false_retry;
2398         } else {
2399                 op = &args->ops[resp->opcnt++];
2400                 op->status = nfserr_retry_uncached_rep;
2401                 nfsd4_encode_operation(resp, op);
2402         }
2403         return op->status;
2404 }
2405
2406 /*
2407  * The sequence operation is not cached because we can use the slot and
2408  * session values.
2409  */
2410 static __be32
2411 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2412                          struct nfsd4_sequence *seq)
2413 {
2414         struct nfsd4_slot *slot = resp->cstate.slot;
2415         struct xdr_stream *xdr = &resp->xdr;
2416         __be32 *p;
2417         __be32 status;
2418
2419         dprintk("--> %s slot %p\n", __func__, slot);
2420
2421         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2422         if (status)
2423                 return status;
2424
2425         p = xdr_reserve_space(xdr, slot->sl_datalen);
2426         if (!p) {
2427                 WARN_ON_ONCE(1);
2428                 return nfserr_serverfault;
2429         }
2430         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2431         xdr_commit_encode(xdr);
2432
2433         resp->opcnt = slot->sl_opcnt;
2434         return slot->sl_status;
2435 }
2436
2437 /*
2438  * Set the exchange_id flags returned by the server.
2439  */
2440 static void
2441 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2442 {
2443 #ifdef CONFIG_NFSD_PNFS
2444         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2445 #else
2446         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2447 #endif
2448
2449         /* Referrals are supported, Migration is not. */
2450         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2451
2452         /* set the wire flags to return to client. */
2453         clid->flags = new->cl_exchange_flags;
2454 }
2455
2456 static bool client_has_openowners(struct nfs4_client *clp)
2457 {
2458         struct nfs4_openowner *oo;
2459
2460         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2461                 if (!list_empty(&oo->oo_owner.so_stateids))
2462                         return true;
2463         }
2464         return false;
2465 }
2466
2467 static bool client_has_state(struct nfs4_client *clp)
2468 {
2469         return client_has_openowners(clp)
2470 #ifdef CONFIG_NFSD_PNFS
2471                 || !list_empty(&clp->cl_lo_states)
2472 #endif
2473                 || !list_empty(&clp->cl_delegations)
2474                 || !list_empty(&clp->cl_sessions);
2475 }
2476
2477 __be32
2478 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2479                 union nfsd4_op_u *u)
2480 {
2481         struct nfsd4_exchange_id *exid = &u->exchange_id;
2482         struct nfs4_client *conf, *new;
2483         struct nfs4_client *unconf = NULL;
2484         __be32 status;
2485         char                    addr_str[INET6_ADDRSTRLEN];
2486         nfs4_verifier           verf = exid->verifier;
2487         struct sockaddr         *sa = svc_addr(rqstp);
2488         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2489         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2490
2491         rpc_ntop(sa, addr_str, sizeof(addr_str));
2492         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2493                 "ip_addr=%s flags %x, spa_how %d\n",
2494                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2495                 addr_str, exid->flags, exid->spa_how);
2496
2497         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2498                 return nfserr_inval;
2499
2500         new = create_client(exid->clname, rqstp, &verf);
2501         if (new == NULL)
2502                 return nfserr_jukebox;
2503
2504         switch (exid->spa_how) {
2505         case SP4_MACH_CRED:
2506                 exid->spo_must_enforce[0] = 0;
2507                 exid->spo_must_enforce[1] = (
2508                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2509                         1 << (OP_EXCHANGE_ID - 32) |
2510                         1 << (OP_CREATE_SESSION - 32) |
2511                         1 << (OP_DESTROY_SESSION - 32) |
2512                         1 << (OP_DESTROY_CLIENTID - 32));
2513
2514                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2515                                         1 << (OP_OPEN_DOWNGRADE) |
2516                                         1 << (OP_LOCKU) |
2517                                         1 << (OP_DELEGRETURN));
2518
2519                 exid->spo_must_allow[1] &= (
2520                                         1 << (OP_TEST_STATEID - 32) |
2521                                         1 << (OP_FREE_STATEID - 32));
2522                 if (!svc_rqst_integrity_protected(rqstp)) {
2523                         status = nfserr_inval;
2524                         goto out_nolock;
2525                 }
2526                 /*
2527                  * Sometimes userspace doesn't give us a principal.
2528                  * Which is a bug, really.  Anyway, we can't enforce
2529                  * MACH_CRED in that case, better to give up now:
2530                  */
2531                 if (!new->cl_cred.cr_principal &&
2532                                         !new->cl_cred.cr_raw_principal) {
2533                         status = nfserr_serverfault;
2534                         goto out_nolock;
2535                 }
2536                 new->cl_mach_cred = true;
2537         case SP4_NONE:
2538                 break;
2539         default:                                /* checked by xdr code */
2540                 WARN_ON_ONCE(1);
2541         case SP4_SSV:
2542                 status = nfserr_encr_alg_unsupp;
2543                 goto out_nolock;
2544         }
2545
2546         /* Cases below refer to rfc 5661 section 18.35.4: */
2547         spin_lock(&nn->client_lock);
2548         conf = find_confirmed_client_by_name(&exid->clname, nn);
2549         if (conf) {
2550                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2551                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2552
2553                 if (update) {
2554                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2555                                 status = nfserr_inval;
2556                                 goto out;
2557                         }
2558                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2559                                 status = nfserr_wrong_cred;
2560                                 goto out;
2561                         }
2562                         if (!creds_match) { /* case 9 */
2563                                 status = nfserr_perm;
2564                                 goto out;
2565                         }
2566                         if (!verfs_match) { /* case 8 */
2567                                 status = nfserr_not_same;
2568                                 goto out;
2569                         }
2570                         /* case 6 */
2571                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2572                         goto out_copy;
2573                 }
2574                 if (!creds_match) { /* case 3 */
2575                         if (client_has_state(conf)) {
2576                                 status = nfserr_clid_inuse;
2577                                 goto out;
2578                         }
2579                         goto out_new;
2580                 }
2581                 if (verfs_match) { /* case 2 */
2582                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2583                         goto out_copy;
2584                 }
2585                 /* case 5, client reboot */
2586                 conf = NULL;
2587                 goto out_new;
2588         }
2589
2590         if (update) { /* case 7 */
2591                 status = nfserr_noent;
2592                 goto out;
2593         }
2594
2595         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2596         if (unconf) /* case 4, possible retry or client restart */
2597                 unhash_client_locked(unconf);
2598
2599         /* case 1 (normal case) */
2600 out_new:
2601         if (conf) {
2602                 status = mark_client_expired_locked(conf);
2603                 if (status)
2604                         goto out;
2605         }
2606         new->cl_minorversion = cstate->minorversion;
2607         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2608         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2609
2610         gen_clid(new, nn);
2611         add_to_unconfirmed(new);
2612         swap(new, conf);
2613 out_copy:
2614         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2615         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2616
2617         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2618         nfsd4_set_ex_flags(conf, exid);
2619
2620         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2621                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2622         status = nfs_ok;
2623
2624 out:
2625         spin_unlock(&nn->client_lock);
2626 out_nolock:
2627         if (new)
2628                 expire_client(new);
2629         if (unconf)
2630                 expire_client(unconf);
2631         return status;
2632 }
2633
2634 static __be32
2635 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2636 {
2637         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2638                 slot_seqid);
2639
2640         /* The slot is in use, and no response has been sent. */
2641         if (slot_inuse) {
2642                 if (seqid == slot_seqid)
2643                         return nfserr_jukebox;
2644                 else
2645                         return nfserr_seq_misordered;
2646         }
2647         /* Note unsigned 32-bit arithmetic handles wraparound: */
2648         if (likely(seqid == slot_seqid + 1))
2649                 return nfs_ok;
2650         if (seqid == slot_seqid)
2651                 return nfserr_replay_cache;
2652         return nfserr_seq_misordered;
2653 }
2654
2655 /*
2656  * Cache the create session result into the create session single DRC
2657  * slot cache by saving the xdr structure. sl_seqid has been set.
2658  * Do this for solo or embedded create session operations.
2659  */
2660 static void
2661 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2662                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2663 {
2664         slot->sl_status = nfserr;
2665         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2666 }
2667
2668 static __be32
2669 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2670                             struct nfsd4_clid_slot *slot)
2671 {
2672         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2673         return slot->sl_status;
2674 }
2675
2676 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2677                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2678                         1 +     /* MIN tag is length with zero, only length */ \
2679                         3 +     /* version, opcount, opcode */ \
2680                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2681                                 /* seqid, slotID, slotID, cache */ \
2682                         4 ) * sizeof(__be32))
2683
2684 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2685                         2 +     /* verifier: AUTH_NULL, length 0 */\
2686                         1 +     /* status */ \
2687                         1 +     /* MIN tag is length with zero, only length */ \
2688                         3 +     /* opcount, opcode, opstatus*/ \
2689                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2690                                 /* seqid, slotID, slotID, slotID, status */ \
2691                         5 ) * sizeof(__be32))
2692
2693 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2694 {
2695         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2696
2697         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2698                 return nfserr_toosmall;
2699         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2700                 return nfserr_toosmall;
2701         ca->headerpadsz = 0;
2702         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2703         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2704         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2705         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2706                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2707         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2708         /*
2709          * Note decreasing slot size below client's request may make it
2710          * difficult for client to function correctly, whereas
2711          * decreasing the number of slots will (just?) affect
2712          * performance.  When short on memory we therefore prefer to
2713          * decrease number of slots instead of their size.  Clients that
2714          * request larger slots than they need will get poor results:
2715          */
2716         ca->maxreqs = nfsd4_get_drc_mem(ca);
2717         if (!ca->maxreqs)
2718                 return nfserr_jukebox;
2719
2720         return nfs_ok;
2721 }
2722
2723 /*
2724  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2725  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2726  */
2727 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2728         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2729
2730 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2731         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2732
2733 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2734                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2735 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2736                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2737                                  sizeof(__be32))
2738
2739 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2740 {
2741         ca->headerpadsz = 0;
2742
2743         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2744                 return nfserr_toosmall;
2745         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2746                 return nfserr_toosmall;
2747         ca->maxresp_cached = 0;
2748         if (ca->maxops < 2)
2749                 return nfserr_toosmall;
2750
2751         return nfs_ok;
2752 }
2753
2754 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2755 {
2756         switch (cbs->flavor) {
2757         case RPC_AUTH_NULL:
2758         case RPC_AUTH_UNIX:
2759                 return nfs_ok;
2760         default:
2761                 /*
2762                  * GSS case: the spec doesn't allow us to return this
2763                  * error.  But it also doesn't allow us not to support
2764                  * GSS.
2765                  * I'd rather this fail hard than return some error the
2766                  * client might think it can already handle:
2767                  */
2768                 return nfserr_encr_alg_unsupp;
2769         }
2770 }
2771
2772 __be32
2773 nfsd4_create_session(struct svc_rqst *rqstp,
2774                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2775 {
2776         struct nfsd4_create_session *cr_ses = &u->create_session;
2777         struct sockaddr *sa = svc_addr(rqstp);
2778         struct nfs4_client *conf, *unconf;
2779         struct nfs4_client *old = NULL;
2780         struct nfsd4_session *new;
2781         struct nfsd4_conn *conn;
2782         struct nfsd4_clid_slot *cs_slot = NULL;
2783         __be32 status = 0;
2784         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2785
2786         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2787                 return nfserr_inval;
2788         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2789         if (status)
2790                 return status;
2791         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2792         if (status)
2793                 return status;
2794         status = check_backchannel_attrs(&cr_ses->back_channel);
2795         if (status)
2796                 goto out_release_drc_mem;
2797         status = nfserr_jukebox;
2798         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2799         if (!new)
2800                 goto out_release_drc_mem;
2801         conn = alloc_conn_from_crses(rqstp, cr_ses);
2802         if (!conn)
2803                 goto out_free_session;
2804
2805         spin_lock(&nn->client_lock);
2806         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2807         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2808         WARN_ON_ONCE(conf && unconf);
2809
2810         if (conf) {
2811                 status = nfserr_wrong_cred;
2812                 if (!nfsd4_mach_creds_match(conf, rqstp))
2813                         goto out_free_conn;
2814                 cs_slot = &conf->cl_cs_slot;
2815                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2816                 if (status) {
2817                         if (status == nfserr_replay_cache)
2818                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2819                         goto out_free_conn;
2820                 }
2821         } else if (unconf) {
2822                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2823                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2824                         status = nfserr_clid_inuse;
2825                         goto out_free_conn;
2826                 }
2827                 status = nfserr_wrong_cred;
2828                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2829                         goto out_free_conn;
2830                 cs_slot = &unconf->cl_cs_slot;
2831                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2832                 if (status) {
2833                         /* an unconfirmed replay returns misordered */
2834                         status = nfserr_seq_misordered;
2835                         goto out_free_conn;
2836                 }
2837                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2838                 if (old) {
2839                         status = mark_client_expired_locked(old);
2840                         if (status) {
2841                                 old = NULL;
2842                                 goto out_free_conn;
2843                         }
2844                 }
2845                 move_to_confirmed(unconf);
2846                 conf = unconf;
2847         } else {
2848                 status = nfserr_stale_clientid;
2849                 goto out_free_conn;
2850         }
2851         status = nfs_ok;
2852         /* Persistent sessions are not supported */
2853         cr_ses->flags &= ~SESSION4_PERSIST;
2854         /* Upshifting from TCP to RDMA is not supported */
2855         cr_ses->flags &= ~SESSION4_RDMA;
2856
2857         init_session(rqstp, new, conf, cr_ses);
2858         nfsd4_get_session_locked(new);
2859
2860         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2861                NFS4_MAX_SESSIONID_LEN);
2862         cs_slot->sl_seqid++;
2863         cr_ses->seqid = cs_slot->sl_seqid;
2864
2865         /* cache solo and embedded create sessions under the client_lock */
2866         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2867         spin_unlock(&nn->client_lock);
2868         /* init connection and backchannel */
2869         nfsd4_init_conn(rqstp, conn, new);
2870         nfsd4_put_session(new);
2871         if (old)
2872                 expire_client(old);
2873         return status;
2874 out_free_conn:
2875         spin_unlock(&nn->client_lock);
2876         free_conn(conn);
2877         if (old)
2878                 expire_client(old);
2879 out_free_session:
2880         __free_session(new);
2881 out_release_drc_mem:
2882         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2883         return status;
2884 }
2885
2886 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2887 {
2888         switch (*dir) {
2889         case NFS4_CDFC4_FORE:
2890         case NFS4_CDFC4_BACK:
2891                 return nfs_ok;
2892         case NFS4_CDFC4_FORE_OR_BOTH:
2893         case NFS4_CDFC4_BACK_OR_BOTH:
2894                 *dir = NFS4_CDFC4_BOTH;
2895                 return nfs_ok;
2896         };
2897         return nfserr_inval;
2898 }
2899
2900 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2901                 struct nfsd4_compound_state *cstate,
2902                 union nfsd4_op_u *u)
2903 {
2904         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2905         struct nfsd4_session *session = cstate->session;
2906         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2907         __be32 status;
2908
2909         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2910         if (status)
2911                 return status;
2912         spin_lock(&nn->client_lock);
2913         session->se_cb_prog = bc->bc_cb_program;
2914         session->se_cb_sec = bc->bc_cb_sec;
2915         spin_unlock(&nn->client_lock);
2916
2917         nfsd4_probe_callback(session->se_client);
2918
2919         return nfs_ok;
2920 }
2921
2922 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2923                      struct nfsd4_compound_state *cstate,
2924                      union nfsd4_op_u *u)
2925 {
2926         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2927         __be32 status;
2928         struct nfsd4_conn *conn;
2929         struct nfsd4_session *session;
2930         struct net *net = SVC_NET(rqstp);
2931         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2932
2933         if (!nfsd4_last_compound_op(rqstp))
2934                 return nfserr_not_only_op;
2935         spin_lock(&nn->client_lock);
2936         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2937         spin_unlock(&nn->client_lock);
2938         if (!session)
2939                 goto out_no_session;
2940         status = nfserr_wrong_cred;
2941         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2942                 goto out;
2943         status = nfsd4_map_bcts_dir(&bcts->dir);
2944         if (status)
2945                 goto out;
2946         conn = alloc_conn(rqstp, bcts->dir);
2947         status = nfserr_jukebox;
2948         if (!conn)
2949                 goto out;
2950         nfsd4_init_conn(rqstp, conn, session);
2951         status = nfs_ok;
2952 out:
2953         nfsd4_put_session(session);
2954 out_no_session:
2955         return status;
2956 }
2957
2958 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2959 {
2960         if (!session)
2961                 return false;
2962         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2963 }
2964
2965 __be32
2966 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2967                 union nfsd4_op_u *u)
2968 {
2969         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2970         struct nfsd4_session *ses;
2971         __be32 status;
2972         int ref_held_by_me = 0;
2973         struct net *net = SVC_NET(r);
2974         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2975
2976         status = nfserr_not_only_op;
2977         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2978                 if (!nfsd4_last_compound_op(r))
2979                         goto out;
2980                 ref_held_by_me++;
2981         }
2982         dump_sessionid(__func__, &sessionid->sessionid);
2983         spin_lock(&nn->client_lock);
2984         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2985         if (!ses)
2986                 goto out_client_lock;
2987         status = nfserr_wrong_cred;
2988         if (!nfsd4_mach_creds_match(ses->se_client, r))
2989                 goto out_put_session;
2990         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2991         if (status)
2992                 goto out_put_session;
2993         unhash_session(ses);
2994         spin_unlock(&nn->client_lock);
2995
2996         nfsd4_probe_callback_sync(ses->se_client);
2997
2998         spin_lock(&nn->client_lock);
2999         status = nfs_ok;
3000 out_put_session:
3001         nfsd4_put_session_locked(ses);
3002 out_client_lock:
3003         spin_unlock(&nn->client_lock);
3004 out:
3005         return status;
3006 }
3007
3008 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3009 {
3010         struct nfsd4_conn *c;
3011
3012         list_for_each_entry(c, &s->se_conns, cn_persession) {
3013                 if (c->cn_xprt == xpt) {
3014                         return c;
3015                 }
3016         }
3017         return NULL;
3018 }
3019
3020 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3021 {
3022         struct nfs4_client *clp = ses->se_client;
3023         struct nfsd4_conn *c;
3024         __be32 status = nfs_ok;
3025         int ret;
3026
3027         spin_lock(&clp->cl_lock);
3028         c = __nfsd4_find_conn(new->cn_xprt, ses);
3029         if (c)
3030                 goto out_free;
3031         status = nfserr_conn_not_bound_to_session;
3032         if (clp->cl_mach_cred)
3033                 goto out_free;
3034         __nfsd4_hash_conn(new, ses);
3035         spin_unlock(&clp->cl_lock);
3036         ret = nfsd4_register_conn(new);
3037         if (ret)
3038                 /* oops; xprt is already down: */
3039                 nfsd4_conn_lost(&new->cn_xpt_user);
3040         return nfs_ok;
3041 out_free:
3042         spin_unlock(&clp->cl_lock);
3043         free_conn(new);
3044         return status;
3045 }
3046
3047 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3048 {
3049         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3050
3051         return args->opcnt > session->se_fchannel.maxops;
3052 }
3053
3054 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3055                                   struct nfsd4_session *session)
3056 {
3057         struct xdr_buf *xb = &rqstp->rq_arg;
3058
3059         return xb->len > session->se_fchannel.maxreq_sz;
3060 }
3061
3062 static bool replay_matches_cache(struct svc_rqst *rqstp,
3063                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3064 {
3065         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3066
3067         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3068             (bool)seq->cachethis)
3069                 return false;
3070         /*
3071          * If there's an error than the reply can have fewer ops than
3072          * the call.  But if we cached a reply with *more* ops than the
3073          * call you're sending us now, then this new call is clearly not
3074          * really a replay of the old one:
3075          */
3076         if (slot->sl_opcnt < argp->opcnt)
3077                 return false;
3078         /* This is the only check explicitly called by spec: */
3079         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3080                 return false;
3081         /*
3082          * There may be more comparisons we could actually do, but the
3083          * spec doesn't require us to catch every case where the calls
3084          * don't match (that would require caching the call as well as
3085          * the reply), so we don't bother.
3086          */
3087         return true;
3088 }
3089
3090 __be32
3091 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3092                 union nfsd4_op_u *u)
3093 {
3094         struct nfsd4_sequence *seq = &u->sequence;
3095         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3096         struct xdr_stream *xdr = &resp->xdr;
3097         struct nfsd4_session *session;
3098         struct nfs4_client *clp;
3099         struct nfsd4_slot *slot;
3100         struct nfsd4_conn *conn;
3101         __be32 status;
3102         int buflen;
3103         struct net *net = SVC_NET(rqstp);
3104         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3105
3106         if (resp->opcnt != 1)
3107                 return nfserr_sequence_pos;
3108
3109         /*
3110          * Will be either used or freed by nfsd4_sequence_check_conn
3111          * below.
3112          */
3113         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3114         if (!conn)
3115                 return nfserr_jukebox;
3116
3117         spin_lock(&nn->client_lock);
3118         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3119         if (!session)
3120                 goto out_no_session;
3121         clp = session->se_client;
3122
3123         status = nfserr_too_many_ops;
3124         if (nfsd4_session_too_many_ops(rqstp, session))
3125                 goto out_put_session;
3126
3127         status = nfserr_req_too_big;
3128         if (nfsd4_request_too_big(rqstp, session))
3129                 goto out_put_session;
3130
3131         status = nfserr_badslot;
3132         if (seq->slotid >= session->se_fchannel.maxreqs)
3133                 goto out_put_session;
3134
3135         slot = session->se_slots[seq->slotid];
3136         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3137
3138         /* We do not negotiate the number of slots yet, so set the
3139          * maxslots to the session maxreqs which is used to encode
3140          * sr_highest_slotid and the sr_target_slot id to maxslots */
3141         seq->maxslots = session->se_fchannel.maxreqs;
3142
3143         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3144                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3145         if (status == nfserr_replay_cache) {
3146                 status = nfserr_seq_misordered;
3147                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3148                         goto out_put_session;
3149                 status = nfserr_seq_false_retry;
3150                 if (!replay_matches_cache(rqstp, seq, slot))
3151                         goto out_put_session;
3152                 cstate->slot = slot;
3153                 cstate->session = session;
3154                 cstate->clp = clp;
3155                 /* Return the cached reply status and set cstate->status
3156                  * for nfsd4_proc_compound processing */
3157                 status = nfsd4_replay_cache_entry(resp, seq);
3158                 cstate->status = nfserr_replay_cache;
3159                 goto out;
3160         }
3161         if (status)
3162                 goto out_put_session;
3163
3164         status = nfsd4_sequence_check_conn(conn, session);
3165         conn = NULL;
3166         if (status)
3167                 goto out_put_session;
3168
3169         buflen = (seq->cachethis) ?
3170                         session->se_fchannel.maxresp_cached :
3171                         session->se_fchannel.maxresp_sz;
3172         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3173                                     nfserr_rep_too_big;
3174         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3175                 goto out_put_session;
3176         svc_reserve(rqstp, buflen);
3177
3178         status = nfs_ok;
3179         /* Success! bump slot seqid */
3180         slot->sl_seqid = seq->seqid;
3181         slot->sl_flags |= NFSD4_SLOT_INUSE;
3182         if (seq->cachethis)
3183                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3184         else
3185                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3186
3187         cstate->slot = slot;
3188         cstate->session = session;
3189         cstate->clp = clp;
3190
3191 out:
3192         switch (clp->cl_cb_state) {
3193         case NFSD4_CB_DOWN:
3194                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3195                 break;
3196         case NFSD4_CB_FAULT:
3197                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3198                 break;
3199         default:
3200                 seq->status_flags = 0;
3201         }
3202         if (!list_empty(&clp->cl_revoked))
3203                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3204 out_no_session:
3205         if (conn)
3206                 free_conn(conn);
3207         spin_unlock(&nn->client_lock);
3208         return status;
3209 out_put_session:
3210         nfsd4_put_session_locked(session);
3211         goto out_no_session;
3212 }
3213
3214 void
3215 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3216 {
3217         struct nfsd4_compound_state *cs = &resp->cstate;
3218
3219         if (nfsd4_has_session(cs)) {
3220                 if (cs->status != nfserr_replay_cache) {
3221                         nfsd4_store_cache_entry(resp);
3222                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3223                 }
3224                 /* Drop session reference that was taken in nfsd4_sequence() */
3225                 nfsd4_put_session(cs->session);
3226         } else if (cs->clp)
3227                 put_client_renew(cs->clp);
3228 }
3229
3230 __be32
3231 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3232                 struct nfsd4_compound_state *cstate,
3233                 union nfsd4_op_u *u)
3234 {
3235         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3236         struct nfs4_client *conf, *unconf;
3237         struct nfs4_client *clp = NULL;
3238         __be32 status = 0;
3239         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3240
3241         spin_lock(&nn->client_lock);
3242         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3243         conf = find_confirmed_client(&dc->clientid, true, nn);
3244         WARN_ON_ONCE(conf && unconf);
3245
3246         if (conf) {
3247                 if (client_has_state(conf)) {
3248                         status = nfserr_clientid_busy;
3249                         goto out;
3250                 }
3251                 status = mark_client_expired_locked(conf);
3252                 if (status)
3253                         goto out;
3254                 clp = conf;
3255         } else if (unconf)
3256                 clp = unconf;
3257         else {
3258                 status = nfserr_stale_clientid;
3259                 goto out;
3260         }
3261         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3262                 clp = NULL;
3263                 status = nfserr_wrong_cred;
3264                 goto out;
3265         }
3266         unhash_client_locked(clp);
3267 out:
3268         spin_unlock(&nn->client_lock);
3269         if (clp)
3270                 expire_client(clp);
3271         return status;
3272 }
3273
3274 __be32
3275 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3276                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3277 {
3278         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3279         __be32 status = 0;
3280
3281         if (rc->rca_one_fs) {
3282                 if (!cstate->current_fh.fh_dentry)
3283                         return nfserr_nofilehandle;
3284                 /*
3285                  * We don't take advantage of the rca_one_fs case.
3286                  * That's OK, it's optional, we can safely ignore it.
3287                  */
3288                 return nfs_ok;
3289         }
3290
3291         status = nfserr_complete_already;
3292         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3293                              &cstate->session->se_client->cl_flags))
3294                 goto out;
3295
3296         status = nfserr_stale_clientid;
3297         if (is_client_expired(cstate->session->se_client))
3298                 /*
3299                  * The following error isn't really legal.
3300                  * But we only get here if the client just explicitly
3301                  * destroyed the client.  Surely it no longer cares what
3302                  * error it gets back on an operation for the dead
3303                  * client.
3304                  */
3305                 goto out;
3306
3307         status = nfs_ok;
3308         nfsd4_client_record_create(cstate->session->se_client);
3309 out:
3310         return status;
3311 }
3312
3313 __be32
3314 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3315                   union nfsd4_op_u *u)
3316 {
3317         struct nfsd4_setclientid *setclid = &u->setclientid;
3318         struct xdr_netobj       clname = setclid->se_name;
3319         nfs4_verifier           clverifier = setclid->se_verf;
3320         struct nfs4_client      *conf, *new;
3321         struct nfs4_client      *unconf = NULL;
3322         __be32                  status;
3323         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3324
3325         new = create_client(clname, rqstp, &clverifier);
3326         if (new == NULL)
3327                 return nfserr_jukebox;
3328         /* Cases below refer to rfc 3530 section 14.2.33: */
3329         spin_lock(&nn->client_lock);
3330         conf = find_confirmed_client_by_name(&clname, nn);
3331         if (conf && client_has_state(conf)) {
3332                 /* case 0: */
3333                 status = nfserr_clid_inuse;
3334                 if (clp_used_exchangeid(conf))
3335                         goto out;
3336                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3337                         char addr_str[INET6_ADDRSTRLEN];
3338                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3339                                  sizeof(addr_str));
3340                         dprintk("NFSD: setclientid: string in use by client "
3341                                 "at %s\n", addr_str);
3342                         goto out;
3343                 }
3344         }
3345         unconf = find_unconfirmed_client_by_name(&clname, nn);
3346         if (unconf)
3347                 unhash_client_locked(unconf);
3348         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3349                 /* case 1: probable callback update */
3350                 copy_clid(new, conf);
3351                 gen_confirm(new, nn);
3352         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3353                 gen_clid(new, nn);
3354         new->cl_minorversion = 0;
3355         gen_callback(new, setclid, rqstp);
3356         add_to_unconfirmed(new);
3357         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3358         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3359         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3360         new = NULL;
3361         status = nfs_ok;
3362 out:
3363         spin_unlock(&nn->client_lock);
3364         if (new)
3365                 free_client(new);
3366         if (unconf)
3367                 expire_client(unconf);
3368         return status;
3369 }
3370
3371
3372 __be32
3373 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3374                         struct nfsd4_compound_state *cstate,
3375                         union nfsd4_op_u *u)
3376 {
3377         struct nfsd4_setclientid_confirm *setclientid_confirm =
3378                         &u->setclientid_confirm;
3379         struct nfs4_client *conf, *unconf;
3380         struct nfs4_client *old = NULL;
3381         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
3382         clientid_t * clid = &setclientid_confirm->sc_clientid;
3383         __be32 status;
3384         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3385
3386         if (STALE_CLIENTID(clid, nn))
3387                 return nfserr_stale_clientid;
3388
3389         spin_lock(&nn->client_lock);
3390         conf = find_confirmed_client(clid, false, nn);
3391         unconf = find_unconfirmed_client(clid, false, nn);
3392         /*
3393          * We try hard to give out unique clientid's, so if we get an
3394          * attempt to confirm the same clientid with a different cred,
3395          * the client may be buggy; this should never happen.
3396          *
3397          * Nevertheless, RFC 7530 recommends INUSE for this case:
3398          */
3399         status = nfserr_clid_inuse;
3400         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3401                 goto out;
3402         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3403                 goto out;
3404         /* cases below refer to rfc 3530 section 14.2.34: */
3405         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3406                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3407                         /* case 2: probable retransmit */
3408                         status = nfs_ok;
3409                 } else /* case 4: client hasn't noticed we rebooted yet? */
3410                         status = nfserr_stale_clientid;
3411                 goto out;
3412         }
3413         status = nfs_ok;
3414         if (conf) { /* case 1: callback update */
3415                 old = unconf;
3416                 unhash_client_locked(old);
3417                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3418         } else { /* case 3: normal case; new or rebooted client */
3419                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3420                 if (old) {
3421                         status = nfserr_clid_inuse;
3422                         if (client_has_state(old)
3423                                         && !same_creds(&unconf->cl_cred,
3424                                                         &old->cl_cred))
3425                                 goto out;
3426                         status = mark_client_expired_locked(old);
3427                         if (status) {
3428                                 old = NULL;
3429                                 goto out;
3430                         }
3431                 }
3432                 move_to_confirmed(unconf);
3433                 conf = unconf;
3434         }
3435         get_client_locked(conf);
3436         spin_unlock(&nn->client_lock);
3437         nfsd4_probe_callback(conf);
3438         spin_lock(&nn->client_lock);
3439         put_client_renew_locked(conf);
3440 out:
3441         spin_unlock(&nn->client_lock);
3442         if (old)
3443                 expire_client(old);
3444         return status;
3445 }
3446
3447 static struct nfs4_file *nfsd4_alloc_file(void)
3448 {
3449         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3450 }
3451
3452 /* OPEN Share state helper functions */
3453 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3454                                 struct nfs4_file *fp)
3455 {
3456         lockdep_assert_held(&state_lock);
3457
3458         refcount_set(&fp->fi_ref, 1);
3459         spin_lock_init(&fp->fi_lock);
3460         INIT_LIST_HEAD(&fp->fi_stateids);
3461         INIT_LIST_HEAD(&fp->fi_delegations);
3462         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3463         fh_copy_shallow(&fp->fi_fhandle, fh);
3464         fp->fi_deleg_file = NULL;
3465         fp->fi_had_conflict = false;
3466         fp->fi_share_deny = 0;
3467         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3468         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3469 #ifdef CONFIG_NFSD_PNFS
3470         INIT_LIST_HEAD(&fp->fi_lo_states);
3471         atomic_set(&fp->fi_lo_recalls, 0);
3472 #endif
3473         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3474 }
3475
3476 void
3477 nfsd4_free_slabs(void)
3478 {
3479         kmem_cache_destroy(client_slab);
3480         kmem_cache_destroy(openowner_slab);
3481         kmem_cache_destroy(lockowner_slab);
3482         kmem_cache_destroy(file_slab);
3483         kmem_cache_destroy(stateid_slab);
3484         kmem_cache_destroy(deleg_slab);
3485         kmem_cache_destroy(odstate_slab);
3486 }
3487
3488 int
3489 nfsd4_init_slabs(void)
3490 {
3491         client_slab = kmem_cache_create("nfsd4_clients",
3492                         sizeof(struct nfs4_client), 0, 0, NULL);
3493         if (client_slab == NULL)
3494                 goto out;
3495         openowner_slab = kmem_cache_create("nfsd4_openowners",
3496                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3497         if (openowner_slab == NULL)
3498                 goto out_free_client_slab;
3499         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3500                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3501         if (lockowner_slab == NULL)
3502                 goto out_free_openowner_slab;
3503         file_slab = kmem_cache_create("nfsd4_files",
3504                         sizeof(struct nfs4_file), 0, 0, NULL);
3505         if (file_slab == NULL)
3506                 goto out_free_lockowner_slab;
3507         stateid_slab = kmem_cache_create("nfsd4_stateids",
3508                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3509         if (stateid_slab == NULL)
3510                 goto out_free_file_slab;
3511         deleg_slab = kmem_cache_create("nfsd4_delegations",
3512                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3513         if (deleg_slab == NULL)
3514                 goto out_free_stateid_slab;
3515         odstate_slab = kmem_cache_create("nfsd4_odstate",
3516                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3517         if (odstate_slab == NULL)
3518                 goto out_free_deleg_slab;
3519         return 0;
3520
3521 out_free_deleg_slab:
3522         kmem_cache_destroy(deleg_slab);
3523 out_free_stateid_slab:
3524         kmem_cache_destroy(stateid_slab);
3525 out_free_file_slab:
3526         kmem_cache_destroy(file_slab);
3527 out_free_lockowner_slab:
3528         kmem_cache_destroy(lockowner_slab);
3529 out_free_openowner_slab:
3530         kmem_cache_destroy(openowner_slab);
3531 out_free_client_slab:
3532         kmem_cache_destroy(client_slab);
3533 out:
3534         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3535         return -ENOMEM;
3536 }
3537
3538 static void init_nfs4_replay(struct nfs4_replay *rp)
3539 {
3540         rp->rp_status = nfserr_serverfault;
3541         rp->rp_buflen = 0;
3542         rp->rp_buf = rp->rp_ibuf;
3543         mutex_init(&rp->rp_mutex);
3544 }
3545
3546 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3547                 struct nfs4_stateowner *so)
3548 {
3549         if (!nfsd4_has_session(cstate)) {
3550                 mutex_lock(&so->so_replay.rp_mutex);
3551                 cstate->replay_owner = nfs4_get_stateowner(so);
3552         }
3553 }
3554
3555 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3556 {
3557         struct nfs4_stateowner *so = cstate->replay_owner;
3558
3559         if (so != NULL) {
3560                 cstate->replay_owner = NULL;
3561                 mutex_unlock(&so->so_replay.rp_mutex);
3562                 nfs4_put_stateowner(so);
3563         }
3564 }
3565
3566 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3567 {
3568         struct nfs4_stateowner *sop;
3569
3570         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3571         if (!sop)
3572                 return NULL;
3573
3574         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3575         if (!sop->so_owner.data) {
3576                 kmem_cache_free(slab, sop);
3577                 return NULL;
3578         }
3579         sop->so_owner.len = owner->len;
3580
3581         INIT_LIST_HEAD(&sop->so_stateids);
3582         sop->so_client = clp;
3583         init_nfs4_replay(&sop->so_replay);
3584         atomic_set(&sop->so_count, 1);
3585         return sop;
3586 }
3587
3588 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3589 {
3590         lockdep_assert_held(&clp->cl_lock);
3591
3592         list_add(&oo->oo_owner.so_strhash,
3593                  &clp->cl_ownerstr_hashtbl[strhashval]);
3594         list_add(&oo->oo_perclient, &clp->cl_openowners);
3595 }
3596
3597 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3598 {
3599         unhash_openowner_locked(openowner(so));
3600 }
3601
3602 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3603 {
3604         struct nfs4_openowner *oo = openowner(so);
3605
3606         kmem_cache_free(openowner_slab, oo);
3607 }
3608
3609 static const struct nfs4_stateowner_operations openowner_ops = {
3610         .so_unhash =    nfs4_unhash_openowner,
3611         .so_free =      nfs4_free_openowner,
3612 };
3613
3614 static struct nfs4_ol_stateid *
3615 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3616 {
3617         struct nfs4_ol_stateid *local, *ret = NULL;
3618         struct nfs4_openowner *oo = open->op_openowner;
3619
3620         lockdep_assert_held(&fp->fi_lock);
3621
3622         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3623                 /* ignore lock owners */
3624                 if (local->st_stateowner->so_is_open_owner == 0)
3625                         continue;
3626