nfsd: eliminate sending duplicate and repeated delegations
[cascardo/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
67 static u64 current_sessionid = 1;
68
69 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
70 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
71 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
72
73 /* forward declarations */
74 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
75 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
76
77 /* Locking: */
78
79 /*
80  * Currently used for the del_recall_lru and file hash table.  In an
81  * effort to decrease the scope of the client_mutex, this spinlock may
82  * eventually cover more:
83  */
84 static DEFINE_SPINLOCK(state_lock);
85
86 /*
87  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
88  * the refcount on the open stateid to drop.
89  */
90 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
91
92 static struct kmem_cache *openowner_slab;
93 static struct kmem_cache *lockowner_slab;
94 static struct kmem_cache *file_slab;
95 static struct kmem_cache *stateid_slab;
96 static struct kmem_cache *deleg_slab;
97 static struct kmem_cache *odstate_slab;
98
99 static void free_session(struct nfsd4_session *);
100
101 static struct nfsd4_callback_ops nfsd4_cb_recall_ops;
102
103 static bool is_session_dead(struct nfsd4_session *ses)
104 {
105         return ses->se_flags & NFS4_SESSION_DEAD;
106 }
107
108 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
109 {
110         if (atomic_read(&ses->se_ref) > ref_held_by_me)
111                 return nfserr_jukebox;
112         ses->se_flags |= NFS4_SESSION_DEAD;
113         return nfs_ok;
114 }
115
116 static bool is_client_expired(struct nfs4_client *clp)
117 {
118         return clp->cl_time == 0;
119 }
120
121 static __be32 get_client_locked(struct nfs4_client *clp)
122 {
123         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
124
125         lockdep_assert_held(&nn->client_lock);
126
127         if (is_client_expired(clp))
128                 return nfserr_expired;
129         atomic_inc(&clp->cl_refcount);
130         return nfs_ok;
131 }
132
133 /* must be called under the client_lock */
134 static inline void
135 renew_client_locked(struct nfs4_client *clp)
136 {
137         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
138
139         if (is_client_expired(clp)) {
140                 WARN_ON(1);
141                 printk("%s: client (clientid %08x/%08x) already expired\n",
142                         __func__,
143                         clp->cl_clientid.cl_boot,
144                         clp->cl_clientid.cl_id);
145                 return;
146         }
147
148         dprintk("renewing client (clientid %08x/%08x)\n",
149                         clp->cl_clientid.cl_boot,
150                         clp->cl_clientid.cl_id);
151         list_move_tail(&clp->cl_lru, &nn->client_lru);
152         clp->cl_time = get_seconds();
153 }
154
155 static void put_client_renew_locked(struct nfs4_client *clp)
156 {
157         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
158
159         lockdep_assert_held(&nn->client_lock);
160
161         if (!atomic_dec_and_test(&clp->cl_refcount))
162                 return;
163         if (!is_client_expired(clp))
164                 renew_client_locked(clp);
165 }
166
167 static void put_client_renew(struct nfs4_client *clp)
168 {
169         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
170
171         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
172                 return;
173         if (!is_client_expired(clp))
174                 renew_client_locked(clp);
175         spin_unlock(&nn->client_lock);
176 }
177
178 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
179 {
180         __be32 status;
181
182         if (is_session_dead(ses))
183                 return nfserr_badsession;
184         status = get_client_locked(ses->se_client);
185         if (status)
186                 return status;
187         atomic_inc(&ses->se_ref);
188         return nfs_ok;
189 }
190
191 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
192 {
193         struct nfs4_client *clp = ses->se_client;
194         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
195
196         lockdep_assert_held(&nn->client_lock);
197
198         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
199                 free_session(ses);
200         put_client_renew_locked(clp);
201 }
202
203 static void nfsd4_put_session(struct nfsd4_session *ses)
204 {
205         struct nfs4_client *clp = ses->se_client;
206         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
207
208         spin_lock(&nn->client_lock);
209         nfsd4_put_session_locked(ses);
210         spin_unlock(&nn->client_lock);
211 }
212
213 static inline struct nfs4_stateowner *
214 nfs4_get_stateowner(struct nfs4_stateowner *sop)
215 {
216         atomic_inc(&sop->so_count);
217         return sop;
218 }
219
220 static int
221 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
222 {
223         return (sop->so_owner.len == owner->len) &&
224                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
225 }
226
227 static struct nfs4_openowner *
228 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
229                         struct nfs4_client *clp)
230 {
231         struct nfs4_stateowner *so;
232
233         lockdep_assert_held(&clp->cl_lock);
234
235         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
236                             so_strhash) {
237                 if (!so->so_is_open_owner)
238                         continue;
239                 if (same_owner_str(so, &open->op_owner))
240                         return openowner(nfs4_get_stateowner(so));
241         }
242         return NULL;
243 }
244
245 static struct nfs4_openowner *
246 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
247                         struct nfs4_client *clp)
248 {
249         struct nfs4_openowner *oo;
250
251         spin_lock(&clp->cl_lock);
252         oo = find_openstateowner_str_locked(hashval, open, clp);
253         spin_unlock(&clp->cl_lock);
254         return oo;
255 }
256
257 static inline u32
258 opaque_hashval(const void *ptr, int nbytes)
259 {
260         unsigned char *cptr = (unsigned char *) ptr;
261
262         u32 x = 0;
263         while (nbytes--) {
264                 x *= 37;
265                 x += *cptr++;
266         }
267         return x;
268 }
269
270 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
271 {
272         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
273
274         kmem_cache_free(file_slab, fp);
275 }
276
277 void
278 put_nfs4_file(struct nfs4_file *fi)
279 {
280         might_lock(&state_lock);
281
282         if (atomic_dec_and_lock(&fi->fi_ref, &state_lock)) {
283                 hlist_del_rcu(&fi->fi_hash);
284                 spin_unlock(&state_lock);
285                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
286                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
287                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
288         }
289 }
290
291 static struct file *
292 __nfs4_get_fd(struct nfs4_file *f, int oflag)
293 {
294         if (f->fi_fds[oflag])
295                 return get_file(f->fi_fds[oflag]);
296         return NULL;
297 }
298
299 static struct file *
300 find_writeable_file_locked(struct nfs4_file *f)
301 {
302         struct file *ret;
303
304         lockdep_assert_held(&f->fi_lock);
305
306         ret = __nfs4_get_fd(f, O_WRONLY);
307         if (!ret)
308                 ret = __nfs4_get_fd(f, O_RDWR);
309         return ret;
310 }
311
312 static struct file *
313 find_writeable_file(struct nfs4_file *f)
314 {
315         struct file *ret;
316
317         spin_lock(&f->fi_lock);
318         ret = find_writeable_file_locked(f);
319         spin_unlock(&f->fi_lock);
320
321         return ret;
322 }
323
324 static struct file *find_readable_file_locked(struct nfs4_file *f)
325 {
326         struct file *ret;
327
328         lockdep_assert_held(&f->fi_lock);
329
330         ret = __nfs4_get_fd(f, O_RDONLY);
331         if (!ret)
332                 ret = __nfs4_get_fd(f, O_RDWR);
333         return ret;
334 }
335
336 static struct file *
337 find_readable_file(struct nfs4_file *f)
338 {
339         struct file *ret;
340
341         spin_lock(&f->fi_lock);
342         ret = find_readable_file_locked(f);
343         spin_unlock(&f->fi_lock);
344
345         return ret;
346 }
347
348 struct file *
349 find_any_file(struct nfs4_file *f)
350 {
351         struct file *ret;
352
353         spin_lock(&f->fi_lock);
354         ret = __nfs4_get_fd(f, O_RDWR);
355         if (!ret) {
356                 ret = __nfs4_get_fd(f, O_WRONLY);
357                 if (!ret)
358                         ret = __nfs4_get_fd(f, O_RDONLY);
359         }
360         spin_unlock(&f->fi_lock);
361         return ret;
362 }
363
364 static atomic_long_t num_delegations;
365 unsigned long max_delegations;
366
367 /*
368  * Open owner state (share locks)
369  */
370
371 /* hash tables for lock and open owners */
372 #define OWNER_HASH_BITS              8
373 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
374 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
375
376 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
377 {
378         unsigned int ret;
379
380         ret = opaque_hashval(ownername->data, ownername->len);
381         return ret & OWNER_HASH_MASK;
382 }
383
384 /* hash table for nfs4_file */
385 #define FILE_HASH_BITS                   8
386 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
387
388 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
389 {
390         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
391 }
392
393 static unsigned int file_hashval(struct knfsd_fh *fh)
394 {
395         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
396 }
397
398 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
399
400 static void
401 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
402 {
403         lockdep_assert_held(&fp->fi_lock);
404
405         if (access & NFS4_SHARE_ACCESS_WRITE)
406                 atomic_inc(&fp->fi_access[O_WRONLY]);
407         if (access & NFS4_SHARE_ACCESS_READ)
408                 atomic_inc(&fp->fi_access[O_RDONLY]);
409 }
410
411 static __be32
412 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
413 {
414         lockdep_assert_held(&fp->fi_lock);
415
416         /* Does this access mode make sense? */
417         if (access & ~NFS4_SHARE_ACCESS_BOTH)
418                 return nfserr_inval;
419
420         /* Does it conflict with a deny mode already set? */
421         if ((access & fp->fi_share_deny) != 0)
422                 return nfserr_share_denied;
423
424         __nfs4_file_get_access(fp, access);
425         return nfs_ok;
426 }
427
428 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
429 {
430         /* Common case is that there is no deny mode. */
431         if (deny) {
432                 /* Does this deny mode make sense? */
433                 if (deny & ~NFS4_SHARE_DENY_BOTH)
434                         return nfserr_inval;
435
436                 if ((deny & NFS4_SHARE_DENY_READ) &&
437                     atomic_read(&fp->fi_access[O_RDONLY]))
438                         return nfserr_share_denied;
439
440                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
441                     atomic_read(&fp->fi_access[O_WRONLY]))
442                         return nfserr_share_denied;
443         }
444         return nfs_ok;
445 }
446
447 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
448 {
449         might_lock(&fp->fi_lock);
450
451         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
452                 struct file *f1 = NULL;
453                 struct file *f2 = NULL;
454
455                 swap(f1, fp->fi_fds[oflag]);
456                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
457                         swap(f2, fp->fi_fds[O_RDWR]);
458                 spin_unlock(&fp->fi_lock);
459                 if (f1)
460                         fput(f1);
461                 if (f2)
462                         fput(f2);
463         }
464 }
465
466 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
467 {
468         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
469
470         if (access & NFS4_SHARE_ACCESS_WRITE)
471                 __nfs4_file_put_access(fp, O_WRONLY);
472         if (access & NFS4_SHARE_ACCESS_READ)
473                 __nfs4_file_put_access(fp, O_RDONLY);
474 }
475
476 /*
477  * Allocate a new open/delegation state counter. This is needed for
478  * pNFS for proper return on close semantics.
479  *
480  * Note that we only allocate it for pNFS-enabled exports, otherwise
481  * all pointers to struct nfs4_clnt_odstate are always NULL.
482  */
483 static struct nfs4_clnt_odstate *
484 alloc_clnt_odstate(struct nfs4_client *clp)
485 {
486         struct nfs4_clnt_odstate *co;
487
488         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
489         if (co) {
490                 co->co_client = clp;
491                 atomic_set(&co->co_odcount, 1);
492         }
493         return co;
494 }
495
496 static void
497 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
498 {
499         struct nfs4_file *fp = co->co_file;
500
501         lockdep_assert_held(&fp->fi_lock);
502         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
503 }
504
505 static inline void
506 get_clnt_odstate(struct nfs4_clnt_odstate *co)
507 {
508         if (co)
509                 atomic_inc(&co->co_odcount);
510 }
511
512 static void
513 put_clnt_odstate(struct nfs4_clnt_odstate *co)
514 {
515         struct nfs4_file *fp;
516
517         if (!co)
518                 return;
519
520         fp = co->co_file;
521         if (atomic_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
522                 list_del(&co->co_perfile);
523                 spin_unlock(&fp->fi_lock);
524
525                 nfsd4_return_all_file_layouts(co->co_client, fp);
526                 kmem_cache_free(odstate_slab, co);
527         }
528 }
529
530 static struct nfs4_clnt_odstate *
531 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
532 {
533         struct nfs4_clnt_odstate *co;
534         struct nfs4_client *cl;
535
536         if (!new)
537                 return NULL;
538
539         cl = new->co_client;
540
541         spin_lock(&fp->fi_lock);
542         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
543                 if (co->co_client == cl) {
544                         get_clnt_odstate(co);
545                         goto out;
546                 }
547         }
548         co = new;
549         co->co_file = fp;
550         hash_clnt_odstate_locked(new);
551 out:
552         spin_unlock(&fp->fi_lock);
553         return co;
554 }
555
556 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl,
557                                          struct kmem_cache *slab)
558 {
559         struct nfs4_stid *stid;
560         int new_id;
561
562         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
563         if (!stid)
564                 return NULL;
565
566         idr_preload(GFP_KERNEL);
567         spin_lock(&cl->cl_lock);
568         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
569         spin_unlock(&cl->cl_lock);
570         idr_preload_end();
571         if (new_id < 0)
572                 goto out_free;
573         stid->sc_client = cl;
574         stid->sc_stateid.si_opaque.so_id = new_id;
575         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
576         /* Will be incremented before return to client: */
577         atomic_set(&stid->sc_count, 1);
578         spin_lock_init(&stid->sc_lock);
579
580         /*
581          * It shouldn't be a problem to reuse an opaque stateid value.
582          * I don't think it is for 4.1.  But with 4.0 I worry that, for
583          * example, a stray write retransmission could be accepted by
584          * the server when it should have been rejected.  Therefore,
585          * adopt a trick from the sctp code to attempt to maximize the
586          * amount of time until an id is reused, by ensuring they always
587          * "increase" (mod INT_MAX):
588          */
589         return stid;
590 out_free:
591         kmem_cache_free(slab, stid);
592         return NULL;
593 }
594
595 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
596 {
597         struct nfs4_stid *stid;
598         struct nfs4_ol_stateid *stp;
599
600         stid = nfs4_alloc_stid(clp, stateid_slab);
601         if (!stid)
602                 return NULL;
603
604         stp = openlockstateid(stid);
605         stp->st_stid.sc_free = nfs4_free_ol_stateid;
606         return stp;
607 }
608
609 static void nfs4_free_deleg(struct nfs4_stid *stid)
610 {
611         kmem_cache_free(deleg_slab, stid);
612         atomic_long_dec(&num_delegations);
613 }
614
615 /*
616  * When we recall a delegation, we should be careful not to hand it
617  * out again straight away.
618  * To ensure this we keep a pair of bloom filters ('new' and 'old')
619  * in which the filehandles of recalled delegations are "stored".
620  * If a filehandle appear in either filter, a delegation is blocked.
621  * When a delegation is recalled, the filehandle is stored in the "new"
622  * filter.
623  * Every 30 seconds we swap the filters and clear the "new" one,
624  * unless both are empty of course.
625  *
626  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
627  * low 3 bytes as hash-table indices.
628  *
629  * 'blocked_delegations_lock', which is always taken in block_delegations(),
630  * is used to manage concurrent access.  Testing does not need the lock
631  * except when swapping the two filters.
632  */
633 static DEFINE_SPINLOCK(blocked_delegations_lock);
634 static struct bloom_pair {
635         int     entries, old_entries;
636         time_t  swap_time;
637         int     new; /* index into 'set' */
638         DECLARE_BITMAP(set[2], 256);
639 } blocked_delegations;
640
641 static int delegation_blocked(struct knfsd_fh *fh)
642 {
643         u32 hash;
644         struct bloom_pair *bd = &blocked_delegations;
645
646         if (bd->entries == 0)
647                 return 0;
648         if (seconds_since_boot() - bd->swap_time > 30) {
649                 spin_lock(&blocked_delegations_lock);
650                 if (seconds_since_boot() - bd->swap_time > 30) {
651                         bd->entries -= bd->old_entries;
652                         bd->old_entries = bd->entries;
653                         memset(bd->set[bd->new], 0,
654                                sizeof(bd->set[0]));
655                         bd->new = 1-bd->new;
656                         bd->swap_time = seconds_since_boot();
657                 }
658                 spin_unlock(&blocked_delegations_lock);
659         }
660         hash = jhash(&fh->fh_base, fh->fh_size, 0);
661         if (test_bit(hash&255, bd->set[0]) &&
662             test_bit((hash>>8)&255, bd->set[0]) &&
663             test_bit((hash>>16)&255, bd->set[0]))
664                 return 1;
665
666         if (test_bit(hash&255, bd->set[1]) &&
667             test_bit((hash>>8)&255, bd->set[1]) &&
668             test_bit((hash>>16)&255, bd->set[1]))
669                 return 1;
670
671         return 0;
672 }
673
674 static void block_delegations(struct knfsd_fh *fh)
675 {
676         u32 hash;
677         struct bloom_pair *bd = &blocked_delegations;
678
679         hash = jhash(&fh->fh_base, fh->fh_size, 0);
680
681         spin_lock(&blocked_delegations_lock);
682         __set_bit(hash&255, bd->set[bd->new]);
683         __set_bit((hash>>8)&255, bd->set[bd->new]);
684         __set_bit((hash>>16)&255, bd->set[bd->new]);
685         if (bd->entries == 0)
686                 bd->swap_time = seconds_since_boot();
687         bd->entries += 1;
688         spin_unlock(&blocked_delegations_lock);
689 }
690
691 static struct nfs4_delegation *
692 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
693                  struct nfs4_clnt_odstate *odstate)
694 {
695         struct nfs4_delegation *dp;
696         long n;
697
698         dprintk("NFSD alloc_init_deleg\n");
699         n = atomic_long_inc_return(&num_delegations);
700         if (n < 0 || n > max_delegations)
701                 goto out_dec;
702         if (delegation_blocked(&current_fh->fh_handle))
703                 goto out_dec;
704         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
705         if (dp == NULL)
706                 goto out_dec;
707
708         dp->dl_stid.sc_free = nfs4_free_deleg;
709         /*
710          * delegation seqid's are never incremented.  The 4.1 special
711          * meaning of seqid 0 isn't meaningful, really, but let's avoid
712          * 0 anyway just for consistency and use 1:
713          */
714         dp->dl_stid.sc_stateid.si_generation = 1;
715         INIT_LIST_HEAD(&dp->dl_perfile);
716         INIT_LIST_HEAD(&dp->dl_perclnt);
717         INIT_LIST_HEAD(&dp->dl_recall_lru);
718         dp->dl_clnt_odstate = odstate;
719         get_clnt_odstate(odstate);
720         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
721         dp->dl_retries = 1;
722         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
723                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
724         return dp;
725 out_dec:
726         atomic_long_dec(&num_delegations);
727         return NULL;
728 }
729
730 void
731 nfs4_put_stid(struct nfs4_stid *s)
732 {
733         struct nfs4_file *fp = s->sc_file;
734         struct nfs4_client *clp = s->sc_client;
735
736         might_lock(&clp->cl_lock);
737
738         if (!atomic_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
739                 wake_up_all(&close_wq);
740                 return;
741         }
742         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
743         spin_unlock(&clp->cl_lock);
744         s->sc_free(s);
745         if (fp)
746                 put_nfs4_file(fp);
747 }
748
749 void
750 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
751 {
752         stateid_t *src = &stid->sc_stateid;
753
754         spin_lock(&stid->sc_lock);
755         if (unlikely(++src->si_generation == 0))
756                 src->si_generation = 1;
757         memcpy(dst, src, sizeof(*dst));
758         spin_unlock(&stid->sc_lock);
759 }
760
761 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
762 {
763         struct file *filp = NULL;
764
765         spin_lock(&fp->fi_lock);
766         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
767                 swap(filp, fp->fi_deleg_file);
768         spin_unlock(&fp->fi_lock);
769
770         if (filp) {
771                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
772                 fput(filp);
773         }
774 }
775
776 void nfs4_unhash_stid(struct nfs4_stid *s)
777 {
778         s->sc_type = 0;
779 }
780
781 /**
782  * nfs4_get_existing_delegation - Discover if this delegation already exists
783  * @clp:     a pointer to the nfs4_client we're granting a delegation to
784  * @fp:      a pointer to the nfs4_file we're granting a delegation on
785  *
786  * Return:
787  *      On success: NULL if an existing delegation was not found.
788  *
789  *      On error: -EAGAIN if one was previously granted to this nfs4_client
790  *                 for this nfs4_file.
791  *
792  */
793
794 static int
795 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
796 {
797         struct nfs4_delegation *searchdp = NULL;
798         struct nfs4_client *searchclp = NULL;
799
800         lockdep_assert_held(&state_lock);
801         lockdep_assert_held(&fp->fi_lock);
802
803         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
804                 searchclp = searchdp->dl_stid.sc_client;
805                 if (clp == searchclp) {
806                         return -EAGAIN;
807                 }
808         }
809         return 0;
810 }
811
812 /**
813  * hash_delegation_locked - Add a delegation to the appropriate lists
814  * @dp:     a pointer to the nfs4_delegation we are adding.
815  * @fp:     a pointer to the nfs4_file we're granting a delegation on
816  *
817  * Return:
818  *      On success: NULL if the delegation was successfully hashed.
819  *
820  *      On error: -EAGAIN if one was previously granted to this
821  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
822  *
823  */
824
825 static int
826 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
827 {
828         int status;
829         struct nfs4_client *clp = dp->dl_stid.sc_client;
830
831         lockdep_assert_held(&state_lock);
832         lockdep_assert_held(&fp->fi_lock);
833
834         status = nfs4_get_existing_delegation(clp, fp);
835         if (status)
836                 return status;
837         ++fp->fi_delegees;
838         atomic_inc(&dp->dl_stid.sc_count);
839         dp->dl_stid.sc_type = NFS4_DELEG_STID;
840         list_add(&dp->dl_perfile, &fp->fi_delegations);
841         list_add(&dp->dl_perclnt, &clp->cl_delegations);
842         return 0;
843 }
844
845 static bool
846 unhash_delegation_locked(struct nfs4_delegation *dp)
847 {
848         struct nfs4_file *fp = dp->dl_stid.sc_file;
849
850         lockdep_assert_held(&state_lock);
851
852         if (list_empty(&dp->dl_perfile))
853                 return false;
854
855         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
856         /* Ensure that deleg break won't try to requeue it */
857         ++dp->dl_time;
858         spin_lock(&fp->fi_lock);
859         list_del_init(&dp->dl_perclnt);
860         list_del_init(&dp->dl_recall_lru);
861         list_del_init(&dp->dl_perfile);
862         spin_unlock(&fp->fi_lock);
863         return true;
864 }
865
866 static void destroy_delegation(struct nfs4_delegation *dp)
867 {
868         bool unhashed;
869
870         spin_lock(&state_lock);
871         unhashed = unhash_delegation_locked(dp);
872         spin_unlock(&state_lock);
873         if (unhashed) {
874                 put_clnt_odstate(dp->dl_clnt_odstate);
875                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
876                 nfs4_put_stid(&dp->dl_stid);
877         }
878 }
879
880 static void revoke_delegation(struct nfs4_delegation *dp)
881 {
882         struct nfs4_client *clp = dp->dl_stid.sc_client;
883
884         WARN_ON(!list_empty(&dp->dl_recall_lru));
885
886         put_clnt_odstate(dp->dl_clnt_odstate);
887         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
888
889         if (clp->cl_minorversion == 0)
890                 nfs4_put_stid(&dp->dl_stid);
891         else {
892                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
893                 spin_lock(&clp->cl_lock);
894                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
895                 spin_unlock(&clp->cl_lock);
896         }
897 }
898
899 /* 
900  * SETCLIENTID state 
901  */
902
903 static unsigned int clientid_hashval(u32 id)
904 {
905         return id & CLIENT_HASH_MASK;
906 }
907
908 static unsigned int clientstr_hashval(const char *name)
909 {
910         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
911 }
912
913 /*
914  * We store the NONE, READ, WRITE, and BOTH bits separately in the
915  * st_{access,deny}_bmap field of the stateid, in order to track not
916  * only what share bits are currently in force, but also what
917  * combinations of share bits previous opens have used.  This allows us
918  * to enforce the recommendation of rfc 3530 14.2.19 that the server
919  * return an error if the client attempt to downgrade to a combination
920  * of share bits not explicable by closing some of its previous opens.
921  *
922  * XXX: This enforcement is actually incomplete, since we don't keep
923  * track of access/deny bit combinations; so, e.g., we allow:
924  *
925  *      OPEN allow read, deny write
926  *      OPEN allow both, deny none
927  *      DOWNGRADE allow read, deny none
928  *
929  * which we should reject.
930  */
931 static unsigned int
932 bmap_to_share_mode(unsigned long bmap) {
933         int i;
934         unsigned int access = 0;
935
936         for (i = 1; i < 4; i++) {
937                 if (test_bit(i, &bmap))
938                         access |= i;
939         }
940         return access;
941 }
942
943 /* set share access for a given stateid */
944 static inline void
945 set_access(u32 access, struct nfs4_ol_stateid *stp)
946 {
947         unsigned char mask = 1 << access;
948
949         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
950         stp->st_access_bmap |= mask;
951 }
952
953 /* clear share access for a given stateid */
954 static inline void
955 clear_access(u32 access, struct nfs4_ol_stateid *stp)
956 {
957         unsigned char mask = 1 << access;
958
959         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
960         stp->st_access_bmap &= ~mask;
961 }
962
963 /* test whether a given stateid has access */
964 static inline bool
965 test_access(u32 access, struct nfs4_ol_stateid *stp)
966 {
967         unsigned char mask = 1 << access;
968
969         return (bool)(stp->st_access_bmap & mask);
970 }
971
972 /* set share deny for a given stateid */
973 static inline void
974 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
975 {
976         unsigned char mask = 1 << deny;
977
978         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
979         stp->st_deny_bmap |= mask;
980 }
981
982 /* clear share deny for a given stateid */
983 static inline void
984 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
985 {
986         unsigned char mask = 1 << deny;
987
988         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
989         stp->st_deny_bmap &= ~mask;
990 }
991
992 /* test whether a given stateid is denying specific access */
993 static inline bool
994 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
995 {
996         unsigned char mask = 1 << deny;
997
998         return (bool)(stp->st_deny_bmap & mask);
999 }
1000
1001 static int nfs4_access_to_omode(u32 access)
1002 {
1003         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1004         case NFS4_SHARE_ACCESS_READ:
1005                 return O_RDONLY;
1006         case NFS4_SHARE_ACCESS_WRITE:
1007                 return O_WRONLY;
1008         case NFS4_SHARE_ACCESS_BOTH:
1009                 return O_RDWR;
1010         }
1011         WARN_ON_ONCE(1);
1012         return O_RDONLY;
1013 }
1014
1015 /*
1016  * A stateid that had a deny mode associated with it is being released
1017  * or downgraded. Recalculate the deny mode on the file.
1018  */
1019 static void
1020 recalculate_deny_mode(struct nfs4_file *fp)
1021 {
1022         struct nfs4_ol_stateid *stp;
1023
1024         spin_lock(&fp->fi_lock);
1025         fp->fi_share_deny = 0;
1026         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1027                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1028         spin_unlock(&fp->fi_lock);
1029 }
1030
1031 static void
1032 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1033 {
1034         int i;
1035         bool change = false;
1036
1037         for (i = 1; i < 4; i++) {
1038                 if ((i & deny) != i) {
1039                         change = true;
1040                         clear_deny(i, stp);
1041                 }
1042         }
1043
1044         /* Recalculate per-file deny mode if there was a change */
1045         if (change)
1046                 recalculate_deny_mode(stp->st_stid.sc_file);
1047 }
1048
1049 /* release all access and file references for a given stateid */
1050 static void
1051 release_all_access(struct nfs4_ol_stateid *stp)
1052 {
1053         int i;
1054         struct nfs4_file *fp = stp->st_stid.sc_file;
1055
1056         if (fp && stp->st_deny_bmap != 0)
1057                 recalculate_deny_mode(fp);
1058
1059         for (i = 1; i < 4; i++) {
1060                 if (test_access(i, stp))
1061                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1062                 clear_access(i, stp);
1063         }
1064 }
1065
1066 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1067 {
1068         kfree(sop->so_owner.data);
1069         sop->so_ops->so_free(sop);
1070 }
1071
1072 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1073 {
1074         struct nfs4_client *clp = sop->so_client;
1075
1076         might_lock(&clp->cl_lock);
1077
1078         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1079                 return;
1080         sop->so_ops->so_unhash(sop);
1081         spin_unlock(&clp->cl_lock);
1082         nfs4_free_stateowner(sop);
1083 }
1084
1085 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1086 {
1087         struct nfs4_file *fp = stp->st_stid.sc_file;
1088
1089         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1090
1091         if (list_empty(&stp->st_perfile))
1092                 return false;
1093
1094         spin_lock(&fp->fi_lock);
1095         list_del_init(&stp->st_perfile);
1096         spin_unlock(&fp->fi_lock);
1097         list_del(&stp->st_perstateowner);
1098         return true;
1099 }
1100
1101 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1102 {
1103         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1104
1105         put_clnt_odstate(stp->st_clnt_odstate);
1106         release_all_access(stp);
1107         if (stp->st_stateowner)
1108                 nfs4_put_stateowner(stp->st_stateowner);
1109         kmem_cache_free(stateid_slab, stid);
1110 }
1111
1112 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1113 {
1114         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1115         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1116         struct file *file;
1117
1118         file = find_any_file(stp->st_stid.sc_file);
1119         if (file)
1120                 filp_close(file, (fl_owner_t)lo);
1121         nfs4_free_ol_stateid(stid);
1122 }
1123
1124 /*
1125  * Put the persistent reference to an already unhashed generic stateid, while
1126  * holding the cl_lock. If it's the last reference, then put it onto the
1127  * reaplist for later destruction.
1128  */
1129 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1130                                        struct list_head *reaplist)
1131 {
1132         struct nfs4_stid *s = &stp->st_stid;
1133         struct nfs4_client *clp = s->sc_client;
1134
1135         lockdep_assert_held(&clp->cl_lock);
1136
1137         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1138
1139         if (!atomic_dec_and_test(&s->sc_count)) {
1140                 wake_up_all(&close_wq);
1141                 return;
1142         }
1143
1144         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1145         list_add(&stp->st_locks, reaplist);
1146 }
1147
1148 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1149 {
1150         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1151
1152         lockdep_assert_held(&oo->oo_owner.so_client->cl_lock);
1153
1154         list_del_init(&stp->st_locks);
1155         nfs4_unhash_stid(&stp->st_stid);
1156         return unhash_ol_stateid(stp);
1157 }
1158
1159 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1160 {
1161         struct nfs4_openowner *oo = openowner(stp->st_openstp->st_stateowner);
1162         bool unhashed;
1163
1164         spin_lock(&oo->oo_owner.so_client->cl_lock);
1165         unhashed = unhash_lock_stateid(stp);
1166         spin_unlock(&oo->oo_owner.so_client->cl_lock);
1167         if (unhashed)
1168                 nfs4_put_stid(&stp->st_stid);
1169 }
1170
1171 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1172 {
1173         struct nfs4_client *clp = lo->lo_owner.so_client;
1174
1175         lockdep_assert_held(&clp->cl_lock);
1176
1177         list_del_init(&lo->lo_owner.so_strhash);
1178 }
1179
1180 /*
1181  * Free a list of generic stateids that were collected earlier after being
1182  * fully unhashed.
1183  */
1184 static void
1185 free_ol_stateid_reaplist(struct list_head *reaplist)
1186 {
1187         struct nfs4_ol_stateid *stp;
1188         struct nfs4_file *fp;
1189
1190         might_sleep();
1191
1192         while (!list_empty(reaplist)) {
1193                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1194                                        st_locks);
1195                 list_del(&stp->st_locks);
1196                 fp = stp->st_stid.sc_file;
1197                 stp->st_stid.sc_free(&stp->st_stid);
1198                 if (fp)
1199                         put_nfs4_file(fp);
1200         }
1201 }
1202
1203 static void release_lockowner(struct nfs4_lockowner *lo)
1204 {
1205         struct nfs4_client *clp = lo->lo_owner.so_client;
1206         struct nfs4_ol_stateid *stp;
1207         struct list_head reaplist;
1208
1209         INIT_LIST_HEAD(&reaplist);
1210
1211         spin_lock(&clp->cl_lock);
1212         unhash_lockowner_locked(lo);
1213         while (!list_empty(&lo->lo_owner.so_stateids)) {
1214                 stp = list_first_entry(&lo->lo_owner.so_stateids,
1215                                 struct nfs4_ol_stateid, st_perstateowner);
1216                 WARN_ON(!unhash_lock_stateid(stp));
1217                 put_ol_stateid_locked(stp, &reaplist);
1218         }
1219         spin_unlock(&clp->cl_lock);
1220         free_ol_stateid_reaplist(&reaplist);
1221         nfs4_put_stateowner(&lo->lo_owner);
1222 }
1223
1224 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1225                                        struct list_head *reaplist)
1226 {
1227         struct nfs4_ol_stateid *stp;
1228
1229         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1230
1231         while (!list_empty(&open_stp->st_locks)) {
1232                 stp = list_entry(open_stp->st_locks.next,
1233                                 struct nfs4_ol_stateid, st_locks);
1234                 WARN_ON(!unhash_lock_stateid(stp));
1235                 put_ol_stateid_locked(stp, reaplist);
1236         }
1237 }
1238
1239 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1240                                 struct list_head *reaplist)
1241 {
1242         bool unhashed;
1243
1244         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1245
1246         unhashed = unhash_ol_stateid(stp);
1247         release_open_stateid_locks(stp, reaplist);
1248         return unhashed;
1249 }
1250
1251 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1252 {
1253         LIST_HEAD(reaplist);
1254
1255         spin_lock(&stp->st_stid.sc_client->cl_lock);
1256         if (unhash_open_stateid(stp, &reaplist))
1257                 put_ol_stateid_locked(stp, &reaplist);
1258         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1259         free_ol_stateid_reaplist(&reaplist);
1260 }
1261
1262 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1263 {
1264         struct nfs4_client *clp = oo->oo_owner.so_client;
1265
1266         lockdep_assert_held(&clp->cl_lock);
1267
1268         list_del_init(&oo->oo_owner.so_strhash);
1269         list_del_init(&oo->oo_perclient);
1270 }
1271
1272 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1273 {
1274         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1275                                           nfsd_net_id);
1276         struct nfs4_ol_stateid *s;
1277
1278         spin_lock(&nn->client_lock);
1279         s = oo->oo_last_closed_stid;
1280         if (s) {
1281                 list_del_init(&oo->oo_close_lru);
1282                 oo->oo_last_closed_stid = NULL;
1283         }
1284         spin_unlock(&nn->client_lock);
1285         if (s)
1286                 nfs4_put_stid(&s->st_stid);
1287 }
1288
1289 static void release_openowner(struct nfs4_openowner *oo)
1290 {
1291         struct nfs4_ol_stateid *stp;
1292         struct nfs4_client *clp = oo->oo_owner.so_client;
1293         struct list_head reaplist;
1294
1295         INIT_LIST_HEAD(&reaplist);
1296
1297         spin_lock(&clp->cl_lock);
1298         unhash_openowner_locked(oo);
1299         while (!list_empty(&oo->oo_owner.so_stateids)) {
1300                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1301                                 struct nfs4_ol_stateid, st_perstateowner);
1302                 if (unhash_open_stateid(stp, &reaplist))
1303                         put_ol_stateid_locked(stp, &reaplist);
1304         }
1305         spin_unlock(&clp->cl_lock);
1306         free_ol_stateid_reaplist(&reaplist);
1307         release_last_closed_stateid(oo);
1308         nfs4_put_stateowner(&oo->oo_owner);
1309 }
1310
1311 static inline int
1312 hash_sessionid(struct nfs4_sessionid *sessionid)
1313 {
1314         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1315
1316         return sid->sequence % SESSION_HASH_SIZE;
1317 }
1318
1319 #ifdef CONFIG_SUNRPC_DEBUG
1320 static inline void
1321 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1322 {
1323         u32 *ptr = (u32 *)(&sessionid->data[0]);
1324         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1325 }
1326 #else
1327 static inline void
1328 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1329 {
1330 }
1331 #endif
1332
1333 /*
1334  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1335  * won't be used for replay.
1336  */
1337 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1338 {
1339         struct nfs4_stateowner *so = cstate->replay_owner;
1340
1341         if (nfserr == nfserr_replay_me)
1342                 return;
1343
1344         if (!seqid_mutating_err(ntohl(nfserr))) {
1345                 nfsd4_cstate_clear_replay(cstate);
1346                 return;
1347         }
1348         if (!so)
1349                 return;
1350         if (so->so_is_open_owner)
1351                 release_last_closed_stateid(openowner(so));
1352         so->so_seqid++;
1353         return;
1354 }
1355
1356 static void
1357 gen_sessionid(struct nfsd4_session *ses)
1358 {
1359         struct nfs4_client *clp = ses->se_client;
1360         struct nfsd4_sessionid *sid;
1361
1362         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1363         sid->clientid = clp->cl_clientid;
1364         sid->sequence = current_sessionid++;
1365         sid->reserved = 0;
1366 }
1367
1368 /*
1369  * The protocol defines ca_maxresponssize_cached to include the size of
1370  * the rpc header, but all we need to cache is the data starting after
1371  * the end of the initial SEQUENCE operation--the rest we regenerate
1372  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1373  * value that is the number of bytes in our cache plus a few additional
1374  * bytes.  In order to stay on the safe side, and not promise more than
1375  * we can cache, those additional bytes must be the minimum possible: 24
1376  * bytes of rpc header (xid through accept state, with AUTH_NULL
1377  * verifier), 12 for the compound header (with zero-length tag), and 44
1378  * for the SEQUENCE op response:
1379  */
1380 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1381
1382 static void
1383 free_session_slots(struct nfsd4_session *ses)
1384 {
1385         int i;
1386
1387         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
1388                 kfree(ses->se_slots[i]);
1389 }
1390
1391 /*
1392  * We don't actually need to cache the rpc and session headers, so we
1393  * can allocate a little less for each slot:
1394  */
1395 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1396 {
1397         u32 size;
1398
1399         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1400                 size = 0;
1401         else
1402                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1403         return size + sizeof(struct nfsd4_slot);
1404 }
1405
1406 /*
1407  * XXX: If we run out of reserved DRC memory we could (up to a point)
1408  * re-negotiate active sessions and reduce their slot usage to make
1409  * room for new connections. For now we just fail the create session.
1410  */
1411 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1412 {
1413         u32 slotsize = slot_bytes(ca);
1414         u32 num = ca->maxreqs;
1415         int avail;
1416
1417         spin_lock(&nfsd_drc_lock);
1418         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1419                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1420         num = min_t(int, num, avail / slotsize);
1421         nfsd_drc_mem_used += num * slotsize;
1422         spin_unlock(&nfsd_drc_lock);
1423
1424         return num;
1425 }
1426
1427 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1428 {
1429         int slotsize = slot_bytes(ca);
1430
1431         spin_lock(&nfsd_drc_lock);
1432         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1433         spin_unlock(&nfsd_drc_lock);
1434 }
1435
1436 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1437                                            struct nfsd4_channel_attrs *battrs)
1438 {
1439         int numslots = fattrs->maxreqs;
1440         int slotsize = slot_bytes(fattrs);
1441         struct nfsd4_session *new;
1442         int mem, i;
1443
1444         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1445                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1446         mem = numslots * sizeof(struct nfsd4_slot *);
1447
1448         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1449         if (!new)
1450                 return NULL;
1451         /* allocate each struct nfsd4_slot and data cache in one piece */
1452         for (i = 0; i < numslots; i++) {
1453                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1454                 if (!new->se_slots[i])
1455                         goto out_free;
1456         }
1457
1458         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1459         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1460
1461         return new;
1462 out_free:
1463         while (i--)
1464                 kfree(new->se_slots[i]);
1465         kfree(new);
1466         return NULL;
1467 }
1468
1469 static void free_conn(struct nfsd4_conn *c)
1470 {
1471         svc_xprt_put(c->cn_xprt);
1472         kfree(c);
1473 }
1474
1475 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1476 {
1477         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1478         struct nfs4_client *clp = c->cn_session->se_client;
1479
1480         spin_lock(&clp->cl_lock);
1481         if (!list_empty(&c->cn_persession)) {
1482                 list_del(&c->cn_persession);
1483                 free_conn(c);
1484         }
1485         nfsd4_probe_callback(clp);
1486         spin_unlock(&clp->cl_lock);
1487 }
1488
1489 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1490 {
1491         struct nfsd4_conn *conn;
1492
1493         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1494         if (!conn)
1495                 return NULL;
1496         svc_xprt_get(rqstp->rq_xprt);
1497         conn->cn_xprt = rqstp->rq_xprt;
1498         conn->cn_flags = flags;
1499         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1500         return conn;
1501 }
1502
1503 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1504 {
1505         conn->cn_session = ses;
1506         list_add(&conn->cn_persession, &ses->se_conns);
1507 }
1508
1509 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1510 {
1511         struct nfs4_client *clp = ses->se_client;
1512
1513         spin_lock(&clp->cl_lock);
1514         __nfsd4_hash_conn(conn, ses);
1515         spin_unlock(&clp->cl_lock);
1516 }
1517
1518 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1519 {
1520         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1521         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1522 }
1523
1524 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1525 {
1526         int ret;
1527
1528         nfsd4_hash_conn(conn, ses);
1529         ret = nfsd4_register_conn(conn);
1530         if (ret)
1531                 /* oops; xprt is already down: */
1532                 nfsd4_conn_lost(&conn->cn_xpt_user);
1533         /* We may have gained or lost a callback channel: */
1534         nfsd4_probe_callback_sync(ses->se_client);
1535 }
1536
1537 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1538 {
1539         u32 dir = NFS4_CDFC4_FORE;
1540
1541         if (cses->flags & SESSION4_BACK_CHAN)
1542                 dir |= NFS4_CDFC4_BACK;
1543         return alloc_conn(rqstp, dir);
1544 }
1545
1546 /* must be called under client_lock */
1547 static void nfsd4_del_conns(struct nfsd4_session *s)
1548 {
1549         struct nfs4_client *clp = s->se_client;
1550         struct nfsd4_conn *c;
1551
1552         spin_lock(&clp->cl_lock);
1553         while (!list_empty(&s->se_conns)) {
1554                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1555                 list_del_init(&c->cn_persession);
1556                 spin_unlock(&clp->cl_lock);
1557
1558                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1559                 free_conn(c);
1560
1561                 spin_lock(&clp->cl_lock);
1562         }
1563         spin_unlock(&clp->cl_lock);
1564 }
1565
1566 static void __free_session(struct nfsd4_session *ses)
1567 {
1568         free_session_slots(ses);
1569         kfree(ses);
1570 }
1571
1572 static void free_session(struct nfsd4_session *ses)
1573 {
1574         nfsd4_del_conns(ses);
1575         nfsd4_put_drc_mem(&ses->se_fchannel);
1576         __free_session(ses);
1577 }
1578
1579 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1580 {
1581         int idx;
1582         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1583
1584         new->se_client = clp;
1585         gen_sessionid(new);
1586
1587         INIT_LIST_HEAD(&new->se_conns);
1588
1589         new->se_cb_seq_nr = 1;
1590         new->se_flags = cses->flags;
1591         new->se_cb_prog = cses->callback_prog;
1592         new->se_cb_sec = cses->cb_sec;
1593         atomic_set(&new->se_ref, 0);
1594         idx = hash_sessionid(&new->se_sessionid);
1595         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1596         spin_lock(&clp->cl_lock);
1597         list_add(&new->se_perclnt, &clp->cl_sessions);
1598         spin_unlock(&clp->cl_lock);
1599
1600         {
1601                 struct sockaddr *sa = svc_addr(rqstp);
1602                 /*
1603                  * This is a little silly; with sessions there's no real
1604                  * use for the callback address.  Use the peer address
1605                  * as a reasonable default for now, but consider fixing
1606                  * the rpc client not to require an address in the
1607                  * future:
1608                  */
1609                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1610                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1611         }
1612 }
1613
1614 /* caller must hold client_lock */
1615 static struct nfsd4_session *
1616 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1617 {
1618         struct nfsd4_session *elem;
1619         int idx;
1620         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1621
1622         lockdep_assert_held(&nn->client_lock);
1623
1624         dump_sessionid(__func__, sessionid);
1625         idx = hash_sessionid(sessionid);
1626         /* Search in the appropriate list */
1627         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1628                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1629                             NFS4_MAX_SESSIONID_LEN)) {
1630                         return elem;
1631                 }
1632         }
1633
1634         dprintk("%s: session not found\n", __func__);
1635         return NULL;
1636 }
1637
1638 static struct nfsd4_session *
1639 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1640                 __be32 *ret)
1641 {
1642         struct nfsd4_session *session;
1643         __be32 status = nfserr_badsession;
1644
1645         session = __find_in_sessionid_hashtbl(sessionid, net);
1646         if (!session)
1647                 goto out;
1648         status = nfsd4_get_session_locked(session);
1649         if (status)
1650                 session = NULL;
1651 out:
1652         *ret = status;
1653         return session;
1654 }
1655
1656 /* caller must hold client_lock */
1657 static void
1658 unhash_session(struct nfsd4_session *ses)
1659 {
1660         struct nfs4_client *clp = ses->se_client;
1661         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1662
1663         lockdep_assert_held(&nn->client_lock);
1664
1665         list_del(&ses->se_hash);
1666         spin_lock(&ses->se_client->cl_lock);
1667         list_del(&ses->se_perclnt);
1668         spin_unlock(&ses->se_client->cl_lock);
1669 }
1670
1671 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1672 static int
1673 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1674 {
1675         /*
1676          * We're assuming the clid was not given out from a boot
1677          * precisely 2^32 (about 136 years) before this one.  That seems
1678          * a safe assumption:
1679          */
1680         if (clid->cl_boot == (u32)nn->boot_time)
1681                 return 0;
1682         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1683                 clid->cl_boot, clid->cl_id, nn->boot_time);
1684         return 1;
1685 }
1686
1687 /* 
1688  * XXX Should we use a slab cache ?
1689  * This type of memory management is somewhat inefficient, but we use it
1690  * anyway since SETCLIENTID is not a common operation.
1691  */
1692 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1693 {
1694         struct nfs4_client *clp;
1695         int i;
1696
1697         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1698         if (clp == NULL)
1699                 return NULL;
1700         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1701         if (clp->cl_name.data == NULL)
1702                 goto err_no_name;
1703         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1704                         OWNER_HASH_SIZE, GFP_KERNEL);
1705         if (!clp->cl_ownerstr_hashtbl)
1706                 goto err_no_hashtbl;
1707         for (i = 0; i < OWNER_HASH_SIZE; i++)
1708                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1709         clp->cl_name.len = name.len;
1710         INIT_LIST_HEAD(&clp->cl_sessions);
1711         idr_init(&clp->cl_stateids);
1712         atomic_set(&clp->cl_refcount, 0);
1713         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1714         INIT_LIST_HEAD(&clp->cl_idhash);
1715         INIT_LIST_HEAD(&clp->cl_openowners);
1716         INIT_LIST_HEAD(&clp->cl_delegations);
1717         INIT_LIST_HEAD(&clp->cl_lru);
1718         INIT_LIST_HEAD(&clp->cl_revoked);
1719 #ifdef CONFIG_NFSD_PNFS
1720         INIT_LIST_HEAD(&clp->cl_lo_states);
1721 #endif
1722         spin_lock_init(&clp->cl_lock);
1723         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1724         return clp;
1725 err_no_hashtbl:
1726         kfree(clp->cl_name.data);
1727 err_no_name:
1728         kfree(clp);
1729         return NULL;
1730 }
1731
1732 static void
1733 free_client(struct nfs4_client *clp)
1734 {
1735         while (!list_empty(&clp->cl_sessions)) {
1736                 struct nfsd4_session *ses;
1737                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1738                                 se_perclnt);
1739                 list_del(&ses->se_perclnt);
1740                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1741                 free_session(ses);
1742         }
1743         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1744         free_svc_cred(&clp->cl_cred);
1745         kfree(clp->cl_ownerstr_hashtbl);
1746         kfree(clp->cl_name.data);
1747         idr_destroy(&clp->cl_stateids);
1748         kfree(clp);
1749 }
1750
1751 /* must be called under the client_lock */
1752 static void
1753 unhash_client_locked(struct nfs4_client *clp)
1754 {
1755         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1756         struct nfsd4_session *ses;
1757
1758         lockdep_assert_held(&nn->client_lock);
1759
1760         /* Mark the client as expired! */
1761         clp->cl_time = 0;
1762         /* Make it invisible */
1763         if (!list_empty(&clp->cl_idhash)) {
1764                 list_del_init(&clp->cl_idhash);
1765                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1766                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1767                 else
1768                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1769         }
1770         list_del_init(&clp->cl_lru);
1771         spin_lock(&clp->cl_lock);
1772         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1773                 list_del_init(&ses->se_hash);
1774         spin_unlock(&clp->cl_lock);
1775 }
1776
1777 static void
1778 unhash_client(struct nfs4_client *clp)
1779 {
1780         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1781
1782         spin_lock(&nn->client_lock);
1783         unhash_client_locked(clp);
1784         spin_unlock(&nn->client_lock);
1785 }
1786
1787 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1788 {
1789         if (atomic_read(&clp->cl_refcount))
1790                 return nfserr_jukebox;
1791         unhash_client_locked(clp);
1792         return nfs_ok;
1793 }
1794
1795 static void
1796 __destroy_client(struct nfs4_client *clp)
1797 {
1798         struct nfs4_openowner *oo;
1799         struct nfs4_delegation *dp;
1800         struct list_head reaplist;
1801
1802         INIT_LIST_HEAD(&reaplist);
1803         spin_lock(&state_lock);
1804         while (!list_empty(&clp->cl_delegations)) {
1805                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1806                 WARN_ON(!unhash_delegation_locked(dp));
1807                 list_add(&dp->dl_recall_lru, &reaplist);
1808         }
1809         spin_unlock(&state_lock);
1810         while (!list_empty(&reaplist)) {
1811                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1812                 list_del_init(&dp->dl_recall_lru);
1813                 put_clnt_odstate(dp->dl_clnt_odstate);
1814                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1815                 nfs4_put_stid(&dp->dl_stid);
1816         }
1817         while (!list_empty(&clp->cl_revoked)) {
1818                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1819                 list_del_init(&dp->dl_recall_lru);
1820                 nfs4_put_stid(&dp->dl_stid);
1821         }
1822         while (!list_empty(&clp->cl_openowners)) {
1823                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1824                 nfs4_get_stateowner(&oo->oo_owner);
1825                 release_openowner(oo);
1826         }
1827         nfsd4_return_all_client_layouts(clp);
1828         nfsd4_shutdown_callback(clp);
1829         if (clp->cl_cb_conn.cb_xprt)
1830                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1831         free_client(clp);
1832 }
1833
1834 static void
1835 destroy_client(struct nfs4_client *clp)
1836 {
1837         unhash_client(clp);
1838         __destroy_client(clp);
1839 }
1840
1841 static void expire_client(struct nfs4_client *clp)
1842 {
1843         unhash_client(clp);
1844         nfsd4_client_record_remove(clp);
1845         __destroy_client(clp);
1846 }
1847
1848 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1849 {
1850         memcpy(target->cl_verifier.data, source->data,
1851                         sizeof(target->cl_verifier.data));
1852 }
1853
1854 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1855 {
1856         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1857         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1858 }
1859
1860 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1861 {
1862         if (source->cr_principal) {
1863                 target->cr_principal =
1864                                 kstrdup(source->cr_principal, GFP_KERNEL);
1865                 if (target->cr_principal == NULL)
1866                         return -ENOMEM;
1867         } else
1868                 target->cr_principal = NULL;
1869         target->cr_flavor = source->cr_flavor;
1870         target->cr_uid = source->cr_uid;
1871         target->cr_gid = source->cr_gid;
1872         target->cr_group_info = source->cr_group_info;
1873         get_group_info(target->cr_group_info);
1874         target->cr_gss_mech = source->cr_gss_mech;
1875         if (source->cr_gss_mech)
1876                 gss_mech_get(source->cr_gss_mech);
1877         return 0;
1878 }
1879
1880 static int
1881 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1882 {
1883         if (o1->len < o2->len)
1884                 return -1;
1885         if (o1->len > o2->len)
1886                 return 1;
1887         return memcmp(o1->data, o2->data, o1->len);
1888 }
1889
1890 static int same_name(const char *n1, const char *n2)
1891 {
1892         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1893 }
1894
1895 static int
1896 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1897 {
1898         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1899 }
1900
1901 static int
1902 same_clid(clientid_t *cl1, clientid_t *cl2)
1903 {
1904         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1905 }
1906
1907 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1908 {
1909         int i;
1910
1911         if (g1->ngroups != g2->ngroups)
1912                 return false;
1913         for (i=0; i<g1->ngroups; i++)
1914                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1915                         return false;
1916         return true;
1917 }
1918
1919 /*
1920  * RFC 3530 language requires clid_inuse be returned when the
1921  * "principal" associated with a requests differs from that previously
1922  * used.  We use uid, gid's, and gss principal string as our best
1923  * approximation.  We also don't want to allow non-gss use of a client
1924  * established using gss: in theory cr_principal should catch that
1925  * change, but in practice cr_principal can be null even in the gss case
1926  * since gssd doesn't always pass down a principal string.
1927  */
1928 static bool is_gss_cred(struct svc_cred *cr)
1929 {
1930         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1931         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1932 }
1933
1934
1935 static bool
1936 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1937 {
1938         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1939                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1940                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1941                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1942                 return false;
1943         if (cr1->cr_principal == cr2->cr_principal)
1944                 return true;
1945         if (!cr1->cr_principal || !cr2->cr_principal)
1946                 return false;
1947         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1948 }
1949
1950 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
1951 {
1952         struct svc_cred *cr = &rqstp->rq_cred;
1953         u32 service;
1954
1955         if (!cr->cr_gss_mech)
1956                 return false;
1957         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
1958         return service == RPC_GSS_SVC_INTEGRITY ||
1959                service == RPC_GSS_SVC_PRIVACY;
1960 }
1961
1962 static bool mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
1963 {
1964         struct svc_cred *cr = &rqstp->rq_cred;
1965
1966         if (!cl->cl_mach_cred)
1967                 return true;
1968         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
1969                 return false;
1970         if (!svc_rqst_integrity_protected(rqstp))
1971                 return false;
1972         if (!cr->cr_principal)
1973                 return false;
1974         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
1975 }
1976
1977 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
1978 {
1979         __be32 verf[2];
1980
1981         /*
1982          * This is opaque to client, so no need to byte-swap. Use
1983          * __force to keep sparse happy
1984          */
1985         verf[0] = (__force __be32)get_seconds();
1986         verf[1] = (__force __be32)nn->clverifier_counter++;
1987         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1988 }
1989
1990 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1991 {
1992         clp->cl_clientid.cl_boot = nn->boot_time;
1993         clp->cl_clientid.cl_id = nn->clientid_counter++;
1994         gen_confirm(clp, nn);
1995 }
1996
1997 static struct nfs4_stid *
1998 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
1999 {
2000         struct nfs4_stid *ret;
2001
2002         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2003         if (!ret || !ret->sc_type)
2004                 return NULL;
2005         return ret;
2006 }
2007
2008 static struct nfs4_stid *
2009 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2010 {
2011         struct nfs4_stid *s;
2012
2013         spin_lock(&cl->cl_lock);
2014         s = find_stateid_locked(cl, t);
2015         if (s != NULL) {
2016                 if (typemask & s->sc_type)
2017                         atomic_inc(&s->sc_count);
2018                 else
2019                         s = NULL;
2020         }
2021         spin_unlock(&cl->cl_lock);
2022         return s;
2023 }
2024
2025 static struct nfs4_client *create_client(struct xdr_netobj name,
2026                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2027 {
2028         struct nfs4_client *clp;
2029         struct sockaddr *sa = svc_addr(rqstp);
2030         int ret;
2031         struct net *net = SVC_NET(rqstp);
2032
2033         clp = alloc_client(name);
2034         if (clp == NULL)
2035                 return NULL;
2036
2037         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2038         if (ret) {
2039                 free_client(clp);
2040                 return NULL;
2041         }
2042         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2043         clp->cl_time = get_seconds();
2044         clear_bit(0, &clp->cl_cb_slot_busy);
2045         copy_verf(clp, verf);
2046         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2047         clp->cl_cb_session = NULL;
2048         clp->net = net;
2049         return clp;
2050 }
2051
2052 static void
2053 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2054 {
2055         struct rb_node **new = &(root->rb_node), *parent = NULL;
2056         struct nfs4_client *clp;
2057
2058         while (*new) {
2059                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2060                 parent = *new;
2061
2062                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2063                         new = &((*new)->rb_left);
2064                 else
2065                         new = &((*new)->rb_right);
2066         }
2067
2068         rb_link_node(&new_clp->cl_namenode, parent, new);
2069         rb_insert_color(&new_clp->cl_namenode, root);
2070 }
2071
2072 static struct nfs4_client *
2073 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2074 {
2075         int cmp;
2076         struct rb_node *node = root->rb_node;
2077         struct nfs4_client *clp;
2078
2079         while (node) {
2080                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2081                 cmp = compare_blob(&clp->cl_name, name);
2082                 if (cmp > 0)
2083                         node = node->rb_left;
2084                 else if (cmp < 0)
2085                         node = node->rb_right;
2086                 else
2087                         return clp;
2088         }
2089         return NULL;
2090 }
2091
2092 static void
2093 add_to_unconfirmed(struct nfs4_client *clp)
2094 {
2095         unsigned int idhashval;
2096         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2097
2098         lockdep_assert_held(&nn->client_lock);
2099
2100         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2101         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2102         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2103         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2104         renew_client_locked(clp);
2105 }
2106
2107 static void
2108 move_to_confirmed(struct nfs4_client *clp)
2109 {
2110         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2111         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2112
2113         lockdep_assert_held(&nn->client_lock);
2114
2115         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2116         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2117         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2118         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2119         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2120         renew_client_locked(clp);
2121 }
2122
2123 static struct nfs4_client *
2124 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2125 {
2126         struct nfs4_client *clp;
2127         unsigned int idhashval = clientid_hashval(clid->cl_id);
2128
2129         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2130                 if (same_clid(&clp->cl_clientid, clid)) {
2131                         if ((bool)clp->cl_minorversion != sessions)
2132                                 return NULL;
2133                         renew_client_locked(clp);
2134                         return clp;
2135                 }
2136         }
2137         return NULL;
2138 }
2139
2140 static struct nfs4_client *
2141 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2142 {
2143         struct list_head *tbl = nn->conf_id_hashtbl;
2144
2145         lockdep_assert_held(&nn->client_lock);
2146         return find_client_in_id_table(tbl, clid, sessions);
2147 }
2148
2149 static struct nfs4_client *
2150 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2151 {
2152         struct list_head *tbl = nn->unconf_id_hashtbl;
2153
2154         lockdep_assert_held(&nn->client_lock);
2155         return find_client_in_id_table(tbl, clid, sessions);
2156 }
2157
2158 static bool clp_used_exchangeid(struct nfs4_client *clp)
2159 {
2160         return clp->cl_exchange_flags != 0;
2161
2162
2163 static struct nfs4_client *
2164 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2165 {
2166         lockdep_assert_held(&nn->client_lock);
2167         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2168 }
2169
2170 static struct nfs4_client *
2171 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2172 {
2173         lockdep_assert_held(&nn->client_lock);
2174         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2175 }
2176
2177 static void
2178 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2179 {
2180         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2181         struct sockaddr *sa = svc_addr(rqstp);
2182         u32 scopeid = rpc_get_scope_id(sa);
2183         unsigned short expected_family;
2184
2185         /* Currently, we only support tcp and tcp6 for the callback channel */
2186         if (se->se_callback_netid_len == 3 &&
2187             !memcmp(se->se_callback_netid_val, "tcp", 3))
2188                 expected_family = AF_INET;
2189         else if (se->se_callback_netid_len == 4 &&
2190                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2191                 expected_family = AF_INET6;
2192         else
2193                 goto out_err;
2194
2195         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2196                                             se->se_callback_addr_len,
2197                                             (struct sockaddr *)&conn->cb_addr,
2198                                             sizeof(conn->cb_addr));
2199
2200         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2201                 goto out_err;
2202
2203         if (conn->cb_addr.ss_family == AF_INET6)
2204                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2205
2206         conn->cb_prog = se->se_callback_prog;
2207         conn->cb_ident = se->se_callback_ident;
2208         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2209         return;
2210 out_err:
2211         conn->cb_addr.ss_family = AF_UNSPEC;
2212         conn->cb_addrlen = 0;
2213         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
2214                 "will not receive delegations\n",
2215                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2216
2217         return;
2218 }
2219
2220 /*
2221  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2222  */
2223 static void
2224 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2225 {
2226         struct xdr_buf *buf = resp->xdr.buf;
2227         struct nfsd4_slot *slot = resp->cstate.slot;
2228         unsigned int base;
2229
2230         dprintk("--> %s slot %p\n", __func__, slot);
2231
2232         slot->sl_opcnt = resp->opcnt;
2233         slot->sl_status = resp->cstate.status;
2234
2235         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2236         if (nfsd4_not_cached(resp)) {
2237                 slot->sl_datalen = 0;
2238                 return;
2239         }
2240         base = resp->cstate.data_offset;
2241         slot->sl_datalen = buf->len - base;
2242         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2243                 WARN("%s: sessions DRC could not cache compound\n", __func__);
2244         return;
2245 }
2246
2247 /*
2248  * Encode the replay sequence operation from the slot values.
2249  * If cachethis is FALSE encode the uncached rep error on the next
2250  * operation which sets resp->p and increments resp->opcnt for
2251  * nfs4svc_encode_compoundres.
2252  *
2253  */
2254 static __be32
2255 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2256                           struct nfsd4_compoundres *resp)
2257 {
2258         struct nfsd4_op *op;
2259         struct nfsd4_slot *slot = resp->cstate.slot;
2260
2261         /* Encode the replayed sequence operation */
2262         op = &args->ops[resp->opcnt - 1];
2263         nfsd4_encode_operation(resp, op);
2264
2265         /* Return nfserr_retry_uncached_rep in next operation. */
2266         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
2267                 op = &args->ops[resp->opcnt++];
2268                 op->status = nfserr_retry_uncached_rep;
2269                 nfsd4_encode_operation(resp, op);
2270         }
2271         return op->status;
2272 }
2273
2274 /*
2275  * The sequence operation is not cached because we can use the slot and
2276  * session values.
2277  */
2278 static __be32
2279 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2280                          struct nfsd4_sequence *seq)
2281 {
2282         struct nfsd4_slot *slot = resp->cstate.slot;
2283         struct xdr_stream *xdr = &resp->xdr;
2284         __be32 *p;
2285         __be32 status;
2286
2287         dprintk("--> %s slot %p\n", __func__, slot);
2288
2289         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2290         if (status)
2291                 return status;
2292
2293         p = xdr_reserve_space(xdr, slot->sl_datalen);
2294         if (!p) {
2295                 WARN_ON_ONCE(1);
2296                 return nfserr_serverfault;
2297         }
2298         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2299         xdr_commit_encode(xdr);
2300
2301         resp->opcnt = slot->sl_opcnt;
2302         return slot->sl_status;
2303 }
2304
2305 /*
2306  * Set the exchange_id flags returned by the server.
2307  */
2308 static void
2309 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2310 {
2311 #ifdef CONFIG_NFSD_PNFS
2312         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2313 #else
2314         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2315 #endif
2316
2317         /* Referrals are supported, Migration is not. */
2318         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2319
2320         /* set the wire flags to return to client. */
2321         clid->flags = new->cl_exchange_flags;
2322 }
2323
2324 static bool client_has_openowners(struct nfs4_client *clp)
2325 {
2326         struct nfs4_openowner *oo;
2327
2328         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2329                 if (!list_empty(&oo->oo_owner.so_stateids))
2330                         return true;
2331         }
2332         return false;
2333 }
2334
2335 static bool client_has_state(struct nfs4_client *clp)
2336 {
2337         return client_has_openowners(clp)
2338 #ifdef CONFIG_NFSD_PNFS
2339                 || !list_empty(&clp->cl_lo_states)
2340 #endif
2341                 || !list_empty(&clp->cl_delegations)
2342                 || !list_empty(&clp->cl_sessions);
2343 }
2344
2345 __be32
2346 nfsd4_exchange_id(struct svc_rqst *rqstp,
2347                   struct nfsd4_compound_state *cstate,
2348                   struct nfsd4_exchange_id *exid)
2349 {
2350         struct nfs4_client *conf, *new;
2351         struct nfs4_client *unconf = NULL;
2352         __be32 status;
2353         char                    addr_str[INET6_ADDRSTRLEN];
2354         nfs4_verifier           verf = exid->verifier;
2355         struct sockaddr         *sa = svc_addr(rqstp);
2356         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2357         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2358
2359         rpc_ntop(sa, addr_str, sizeof(addr_str));
2360         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2361                 "ip_addr=%s flags %x, spa_how %d\n",
2362                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2363                 addr_str, exid->flags, exid->spa_how);
2364
2365         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2366                 return nfserr_inval;
2367
2368         switch (exid->spa_how) {
2369         case SP4_MACH_CRED:
2370                 if (!svc_rqst_integrity_protected(rqstp))
2371                         return nfserr_inval;
2372         case SP4_NONE:
2373                 break;
2374         default:                                /* checked by xdr code */
2375                 WARN_ON_ONCE(1);
2376         case SP4_SSV:
2377                 return nfserr_encr_alg_unsupp;
2378         }
2379
2380         new = create_client(exid->clname, rqstp, &verf);
2381         if (new == NULL)
2382                 return nfserr_jukebox;
2383
2384         /* Cases below refer to rfc 5661 section 18.35.4: */
2385         spin_lock(&nn->client_lock);
2386         conf = find_confirmed_client_by_name(&exid->clname, nn);
2387         if (conf) {
2388                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2389                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2390
2391                 if (update) {
2392                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2393                                 status = nfserr_inval;
2394                                 goto out;
2395                         }
2396                         if (!mach_creds_match(conf, rqstp)) {
2397                                 status = nfserr_wrong_cred;
2398                                 goto out;
2399                         }
2400                         if (!creds_match) { /* case 9 */
2401                                 status = nfserr_perm;
2402                                 goto out;
2403                         }
2404                         if (!verfs_match) { /* case 8 */
2405                                 status = nfserr_not_same;
2406                                 goto out;
2407                         }
2408                         /* case 6 */
2409                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2410                         goto out_copy;
2411                 }
2412                 if (!creds_match) { /* case 3 */
2413                         if (client_has_state(conf)) {
2414                                 status = nfserr_clid_inuse;
2415                                 goto out;
2416                         }
2417                         goto out_new;
2418                 }
2419                 if (verfs_match) { /* case 2 */
2420                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2421                         goto out_copy;
2422                 }
2423                 /* case 5, client reboot */
2424                 conf = NULL;
2425                 goto out_new;
2426         }
2427
2428         if (update) { /* case 7 */
2429                 status = nfserr_noent;
2430                 goto out;
2431         }
2432
2433         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2434         if (unconf) /* case 4, possible retry or client restart */
2435                 unhash_client_locked(unconf);
2436
2437         /* case 1 (normal case) */
2438 out_new:
2439         if (conf) {
2440                 status = mark_client_expired_locked(conf);
2441                 if (status)
2442                         goto out;
2443         }
2444         new->cl_minorversion = cstate->minorversion;
2445         new->cl_mach_cred = (exid->spa_how == SP4_MACH_CRED);
2446
2447         gen_clid(new, nn);
2448         add_to_unconfirmed(new);
2449         swap(new, conf);
2450 out_copy:
2451         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2452         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2453
2454         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2455         nfsd4_set_ex_flags(conf, exid);
2456
2457         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2458                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2459         status = nfs_ok;
2460
2461 out:
2462         spin_unlock(&nn->client_lock);
2463         if (new)
2464                 expire_client(new);
2465         if (unconf)
2466                 expire_client(unconf);
2467         return status;
2468 }
2469
2470 static __be32
2471 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2472 {
2473         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2474                 slot_seqid);
2475
2476         /* The slot is in use, and no response has been sent. */
2477         if (slot_inuse) {
2478                 if (seqid == slot_seqid)
2479                         return nfserr_jukebox;
2480                 else
2481                         return nfserr_seq_misordered;
2482         }
2483         /* Note unsigned 32-bit arithmetic handles wraparound: */
2484         if (likely(seqid == slot_seqid + 1))
2485                 return nfs_ok;
2486         if (seqid == slot_seqid)
2487                 return nfserr_replay_cache;
2488         return nfserr_seq_misordered;
2489 }
2490
2491 /*
2492  * Cache the create session result into the create session single DRC
2493  * slot cache by saving the xdr structure. sl_seqid has been set.
2494  * Do this for solo or embedded create session operations.
2495  */
2496 static void
2497 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2498                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2499 {
2500         slot->sl_status = nfserr;
2501         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2502 }
2503
2504 static __be32
2505 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2506                             struct nfsd4_clid_slot *slot)
2507 {
2508         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2509         return slot->sl_status;
2510 }
2511
2512 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2513                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2514                         1 +     /* MIN tag is length with zero, only length */ \
2515                         3 +     /* version, opcount, opcode */ \
2516                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2517                                 /* seqid, slotID, slotID, cache */ \
2518                         4 ) * sizeof(__be32))
2519
2520 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2521                         2 +     /* verifier: AUTH_NULL, length 0 */\
2522                         1 +     /* status */ \
2523                         1 +     /* MIN tag is length with zero, only length */ \
2524                         3 +     /* opcount, opcode, opstatus*/ \
2525                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2526                                 /* seqid, slotID, slotID, slotID, status */ \
2527                         5 ) * sizeof(__be32))
2528
2529 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2530 {
2531         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2532
2533         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2534                 return nfserr_toosmall;
2535         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2536                 return nfserr_toosmall;
2537         ca->headerpadsz = 0;
2538         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2539         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2540         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2541         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2542                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2543         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2544         /*
2545          * Note decreasing slot size below client's request may make it
2546          * difficult for client to function correctly, whereas
2547          * decreasing the number of slots will (just?) affect
2548          * performance.  When short on memory we therefore prefer to
2549          * decrease number of slots instead of their size.  Clients that
2550          * request larger slots than they need will get poor results:
2551          */
2552         ca->maxreqs = nfsd4_get_drc_mem(ca);
2553         if (!ca->maxreqs)
2554                 return nfserr_jukebox;
2555
2556         return nfs_ok;
2557 }
2558
2559 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2560                                  RPC_MAX_HEADER_WITH_AUTH) * sizeof(__be32))
2561 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2562                                  RPC_MAX_REPHEADER_WITH_AUTH) * sizeof(__be32))
2563
2564 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2565 {
2566         ca->headerpadsz = 0;
2567
2568         /*
2569          * These RPC_MAX_HEADER macros are overkill, especially since we
2570          * don't even do gss on the backchannel yet.  But this is still
2571          * less than 1k.  Tighten up this estimate in the unlikely event
2572          * it turns out to be a problem for some client:
2573          */
2574         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2575                 return nfserr_toosmall;
2576         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2577                 return nfserr_toosmall;
2578         ca->maxresp_cached = 0;
2579         if (ca->maxops < 2)
2580                 return nfserr_toosmall;
2581
2582         return nfs_ok;
2583 }
2584
2585 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2586 {
2587         switch (cbs->flavor) {
2588         case RPC_AUTH_NULL:
2589         case RPC_AUTH_UNIX:
2590                 return nfs_ok;
2591         default:
2592                 /*
2593                  * GSS case: the spec doesn't allow us to return this
2594                  * error.  But it also doesn't allow us not to support
2595                  * GSS.
2596                  * I'd rather this fail hard than return some error the
2597                  * client might think it can already handle:
2598                  */
2599                 return nfserr_encr_alg_unsupp;
2600         }
2601 }
2602
2603 __be32
2604 nfsd4_create_session(struct svc_rqst *rqstp,
2605                      struct nfsd4_compound_state *cstate,
2606                      struct nfsd4_create_session *cr_ses)
2607 {
2608         struct sockaddr *sa = svc_addr(rqstp);
2609         struct nfs4_client *conf, *unconf;
2610         struct nfs4_client *old = NULL;
2611         struct nfsd4_session *new;
2612         struct nfsd4_conn *conn;
2613         struct nfsd4_clid_slot *cs_slot = NULL;
2614         __be32 status = 0;
2615         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2616
2617         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2618                 return nfserr_inval;
2619         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2620         if (status)
2621                 return status;
2622         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2623         if (status)
2624                 return status;
2625         status = check_backchannel_attrs(&cr_ses->back_channel);
2626         if (status)
2627                 goto out_release_drc_mem;
2628         status = nfserr_jukebox;
2629         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2630         if (!new)
2631                 goto out_release_drc_mem;
2632         conn = alloc_conn_from_crses(rqstp, cr_ses);
2633         if (!conn)
2634                 goto out_free_session;
2635
2636         spin_lock(&nn->client_lock);
2637         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2638         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2639         WARN_ON_ONCE(conf && unconf);
2640
2641         if (conf) {
2642                 status = nfserr_wrong_cred;
2643                 if (!mach_creds_match(conf, rqstp))
2644                         goto out_free_conn;
2645                 cs_slot = &conf->cl_cs_slot;
2646                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2647                 if (status) {
2648                         if (status == nfserr_replay_cache)
2649                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2650                         goto out_free_conn;
2651                 }
2652         } else if (unconf) {
2653                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2654                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2655                         status = nfserr_clid_inuse;
2656                         goto out_free_conn;
2657                 }
2658                 status = nfserr_wrong_cred;
2659                 if (!mach_creds_match(unconf, rqstp))
2660                         goto out_free_conn;
2661                 cs_slot = &unconf->cl_cs_slot;
2662                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2663                 if (status) {
2664                         /* an unconfirmed replay returns misordered */
2665                         status = nfserr_seq_misordered;
2666                         goto out_free_conn;
2667                 }
2668                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2669                 if (old) {
2670                         status = mark_client_expired_locked(old);
2671                         if (status) {
2672                                 old = NULL;
2673                                 goto out_free_conn;
2674                         }
2675                 }
2676                 move_to_confirmed(unconf);
2677                 conf = unconf;
2678         } else {
2679                 status = nfserr_stale_clientid;
2680                 goto out_free_conn;
2681         }
2682         status = nfs_ok;
2683         /*
2684          * We do not support RDMA or persistent sessions
2685          */
2686         cr_ses->flags &= ~SESSION4_PERSIST;
2687         cr_ses->flags &= ~SESSION4_RDMA;
2688
2689         init_session(rqstp, new, conf, cr_ses);
2690         nfsd4_get_session_locked(new);
2691
2692         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2693                NFS4_MAX_SESSIONID_LEN);
2694         cs_slot->sl_seqid++;
2695         cr_ses->seqid = cs_slot->sl_seqid;
2696
2697         /* cache solo and embedded create sessions under the client_lock */
2698         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2699         spin_unlock(&nn->client_lock);
2700         /* init connection and backchannel */
2701         nfsd4_init_conn(rqstp, conn, new);
2702         nfsd4_put_session(new);
2703         if (old)
2704                 expire_client(old);
2705         return status;
2706 out_free_conn:
2707         spin_unlock(&nn->client_lock);
2708         free_conn(conn);
2709         if (old)
2710                 expire_client(old);
2711 out_free_session:
2712         __free_session(new);
2713 out_release_drc_mem:
2714         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2715         return status;
2716 }
2717
2718 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2719 {
2720         switch (*dir) {
2721         case NFS4_CDFC4_FORE:
2722         case NFS4_CDFC4_BACK:
2723                 return nfs_ok;
2724         case NFS4_CDFC4_FORE_OR_BOTH:
2725         case NFS4_CDFC4_BACK_OR_BOTH:
2726                 *dir = NFS4_CDFC4_BOTH;
2727                 return nfs_ok;
2728         };
2729         return nfserr_inval;
2730 }
2731
2732 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
2733 {
2734         struct nfsd4_session *session = cstate->session;
2735         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2736         __be32 status;
2737
2738         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2739         if (status)
2740                 return status;
2741         spin_lock(&nn->client_lock);
2742         session->se_cb_prog = bc->bc_cb_program;
2743         session->se_cb_sec = bc->bc_cb_sec;
2744         spin_unlock(&nn->client_lock);
2745
2746         nfsd4_probe_callback(session->se_client);
2747
2748         return nfs_ok;
2749 }
2750
2751 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2752                      struct nfsd4_compound_state *cstate,
2753                      struct nfsd4_bind_conn_to_session *bcts)
2754 {
2755         __be32 status;
2756         struct nfsd4_conn *conn;
2757         struct nfsd4_session *session;
2758         struct net *net = SVC_NET(rqstp);
2759         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2760
2761         if (!nfsd4_last_compound_op(rqstp))
2762                 return nfserr_not_only_op;
2763         spin_lock(&nn->client_lock);
2764         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2765         spin_unlock(&nn->client_lock);
2766         if (!session)
2767                 goto out_no_session;
2768         status = nfserr_wrong_cred;
2769         if (!mach_creds_match(session->se_client, rqstp))
2770                 goto out;
2771         status = nfsd4_map_bcts_dir(&bcts->dir);
2772         if (status)
2773                 goto out;
2774         conn = alloc_conn(rqstp, bcts->dir);
2775         status = nfserr_jukebox;
2776         if (!conn)
2777                 goto out;
2778         nfsd4_init_conn(rqstp, conn, session);
2779         status = nfs_ok;
2780 out:
2781         nfsd4_put_session(session);
2782 out_no_session:
2783         return status;
2784 }
2785
2786 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2787 {
2788         if (!session)
2789                 return 0;
2790         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2791 }
2792
2793 __be32
2794 nfsd4_destroy_session(struct svc_rqst *r,
2795                       struct nfsd4_compound_state *cstate,
2796                       struct nfsd4_destroy_session *sessionid)
2797 {
2798         struct nfsd4_session *ses;
2799         __be32 status;
2800         int ref_held_by_me = 0;
2801         struct net *net = SVC_NET(r);
2802         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2803
2804         status = nfserr_not_only_op;
2805         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2806                 if (!nfsd4_last_compound_op(r))
2807                         goto out;
2808                 ref_held_by_me++;
2809         }
2810         dump_sessionid(__func__, &sessionid->sessionid);
2811         spin_lock(&nn->client_lock);
2812         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2813         if (!ses)
2814                 goto out_client_lock;
2815         status = nfserr_wrong_cred;
2816         if (!mach_creds_match(ses->se_client, r))
2817                 goto out_put_session;
2818         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2819         if (status)
2820                 goto out_put_session;
2821         unhash_session(ses);
2822         spin_unlock(&nn->client_lock);
2823
2824         nfsd4_probe_callback_sync(ses->se_client);
2825
2826         spin_lock(&nn->client_lock);
2827         status = nfs_ok;
2828 out_put_session:
2829         nfsd4_put_session_locked(ses);
2830 out_client_lock:
2831         spin_unlock(&nn->client_lock);
2832 out:
2833         return status;
2834 }
2835
2836 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2837 {
2838         struct nfsd4_conn *c;
2839
2840         list_for_each_entry(c, &s->se_conns, cn_persession) {
2841                 if (c->cn_xprt == xpt) {
2842                         return c;
2843                 }
2844         }
2845         return NULL;
2846 }
2847
2848 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2849 {
2850         struct nfs4_client *clp = ses->se_client;
2851         struct nfsd4_conn *c;
2852         __be32 status = nfs_ok;
2853         int ret;
2854
2855         spin_lock(&clp->cl_lock);
2856         c = __nfsd4_find_conn(new->cn_xprt, ses);
2857         if (c)
2858                 goto out_free;
2859         status = nfserr_conn_not_bound_to_session;
2860         if (clp->cl_mach_cred)
2861                 goto out_free;
2862         __nfsd4_hash_conn(new, ses);
2863         spin_unlock(&clp->cl_lock);
2864         ret = nfsd4_register_conn(new);
2865         if (ret)
2866                 /* oops; xprt is already down: */
2867                 nfsd4_conn_lost(&new->cn_xpt_user);
2868         return nfs_ok;
2869 out_free:
2870         spin_unlock(&clp->cl_lock);
2871         free_conn(new);
2872         return status;
2873 }
2874
2875 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2876 {
2877         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2878
2879         return args->opcnt > session->se_fchannel.maxops;
2880 }
2881
2882 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2883                                   struct nfsd4_session *session)
2884 {
2885         struct xdr_buf *xb = &rqstp->rq_arg;
2886
2887         return xb->len > session->se_fchannel.maxreq_sz;
2888 }
2889
2890 __be32
2891 nfsd4_sequence(struct svc_rqst *rqstp,
2892                struct nfsd4_compound_state *cstate,
2893                struct nfsd4_sequence *seq)
2894 {
2895         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2896         struct xdr_stream *xdr = &resp->xdr;
2897         struct nfsd4_session *session;
2898         struct nfs4_client *clp;
2899         struct nfsd4_slot *slot;
2900         struct nfsd4_conn *conn;
2901         __be32 status;
2902         int buflen;
2903         struct net *net = SVC_NET(rqstp);
2904         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2905
2906         if (resp->opcnt != 1)
2907                 return nfserr_sequence_pos;
2908
2909         /*
2910          * Will be either used or freed by nfsd4_sequence_check_conn
2911          * below.
2912          */
2913         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2914         if (!conn)
2915                 return nfserr_jukebox;
2916
2917         spin_lock(&nn->client_lock);
2918         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
2919         if (!session)
2920                 goto out_no_session;
2921         clp = session->se_client;
2922
2923         status = nfserr_too_many_ops;
2924         if (nfsd4_session_too_many_ops(rqstp, session))
2925                 goto out_put_session;
2926
2927         status = nfserr_req_too_big;
2928         if (nfsd4_request_too_big(rqstp, session))
2929                 goto out_put_session;
2930
2931         status = nfserr_badslot;
2932         if (seq->slotid >= session->se_fchannel.maxreqs)
2933                 goto out_put_session;
2934
2935         slot = session->se_slots[seq->slotid];
2936         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2937
2938         /* We do not negotiate the number of slots yet, so set the
2939          * maxslots to the session maxreqs which is used to encode
2940          * sr_highest_slotid and the sr_target_slot id to maxslots */
2941         seq->maxslots = session->se_fchannel.maxreqs;
2942
2943         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2944                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2945         if (status == nfserr_replay_cache) {
2946                 status = nfserr_seq_misordered;
2947                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2948                         goto out_put_session;
2949                 cstate->slot = slot;
2950                 cstate->session = session;
2951                 cstate->clp = clp;
2952                 /* Return the cached reply status and set cstate->status
2953                  * for nfsd4_proc_compound processing */
2954                 status = nfsd4_replay_cache_entry(resp, seq);
2955                 cstate->status = nfserr_replay_cache;
2956                 goto out;
2957         }
2958         if (status)
2959                 goto out_put_session;
2960
2961         status = nfsd4_sequence_check_conn(conn, session);
2962         conn = NULL;
2963         if (status)
2964                 goto out_put_session;
2965
2966         buflen = (seq->cachethis) ?
2967                         session->se_fchannel.maxresp_cached :
2968                         session->se_fchannel.maxresp_sz;
2969         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
2970                                     nfserr_rep_too_big;
2971         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
2972                 goto out_put_session;
2973         svc_reserve(rqstp, buflen);
2974
2975         status = nfs_ok;
2976         /* Success! bump slot seqid */
2977         slot->sl_seqid = seq->seqid;
2978         slot->sl_flags |= NFSD4_SLOT_INUSE;
2979         if (seq->cachethis)
2980                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2981         else
2982                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2983
2984         cstate->slot = slot;
2985         cstate->session = session;
2986         cstate->clp = clp;
2987
2988 out:
2989         switch (clp->cl_cb_state) {
2990         case NFSD4_CB_DOWN:
2991                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2992                 break;
2993         case NFSD4_CB_FAULT:
2994                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2995                 break;
2996         default:
2997                 seq->status_flags = 0;
2998         }
2999         if (!list_empty(&clp->cl_revoked))
3000                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3001 out_no_session:
3002         if (conn)
3003                 free_conn(conn);
3004         spin_unlock(&nn->client_lock);
3005         return status;
3006 out_put_session:
3007         nfsd4_put_session_locked(session);
3008         goto out_no_session;
3009 }
3010
3011 void
3012 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3013 {
3014         struct nfsd4_compound_state *cs = &resp->cstate;
3015
3016         if (nfsd4_has_session(cs)) {
3017                 if (cs->status != nfserr_replay_cache) {
3018                         nfsd4_store_cache_entry(resp);
3019                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3020                 }
3021                 /* Drop session reference that was taken in nfsd4_sequence() */
3022                 nfsd4_put_session(cs->session);
3023         } else if (cs->clp)
3024                 put_client_renew(cs->clp);
3025 }
3026
3027 __be32
3028 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
3029 {
3030         struct nfs4_client *conf, *unconf;
3031         struct nfs4_client *clp = NULL;
3032         __be32 status = 0;
3033         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3034
3035         spin_lock(&nn->client_lock);
3036         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3037         conf = find_confirmed_client(&dc->clientid, true, nn);
3038         WARN_ON_ONCE(conf && unconf);
3039
3040         if (conf) {
3041                 if (client_has_state(conf)) {
3042                         status = nfserr_clientid_busy;
3043                         goto out;
3044                 }
3045                 status = mark_client_expired_locked(conf);
3046                 if (status)
3047                         goto out;
3048                 clp = conf;
3049         } else if (unconf)
3050                 clp = unconf;
3051         else {
3052                 status = nfserr_stale_clientid;
3053                 goto out;
3054         }
3055         if (!mach_creds_match(clp, rqstp)) {
3056                 clp = NULL;
3057                 status = nfserr_wrong_cred;
3058                 goto out;
3059         }
3060         unhash_client_locked(clp);
3061 out:
3062         spin_unlock(&nn->client_lock);
3063         if (clp)
3064                 expire_client(clp);
3065         return status;
3066 }
3067
3068 __be32
3069 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
3070 {
3071         __be32 status = 0;
3072
3073         if (rc->rca_one_fs) {
3074                 if (!cstate->current_fh.fh_dentry)
3075                         return nfserr_nofilehandle;
3076                 /*
3077                  * We don't take advantage of the rca_one_fs case.
3078                  * That's OK, it's optional, we can safely ignore it.
3079                  */
3080                  return nfs_ok;
3081         }
3082
3083         status = nfserr_complete_already;
3084         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3085                              &cstate->session->se_client->cl_flags))
3086                 goto out;
3087
3088         status = nfserr_stale_clientid;
3089         if (is_client_expired(cstate->session->se_client))
3090                 /*
3091                  * The following error isn't really legal.
3092                  * But we only get here if the client just explicitly
3093                  * destroyed the client.  Surely it no longer cares what
3094                  * error it gets back on an operation for the dead
3095                  * client.
3096                  */
3097                 goto out;
3098
3099         status = nfs_ok;
3100         nfsd4_client_record_create(cstate->session->se_client);
3101 out:
3102         return status;
3103 }
3104
3105 __be32
3106 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3107                   struct nfsd4_setclientid *setclid)
3108 {
3109         struct xdr_netobj       clname = setclid->se_name;
3110         nfs4_verifier           clverifier = setclid->se_verf;
3111         struct nfs4_client      *conf, *new;
3112         struct nfs4_client      *unconf = NULL;
3113         __be32                  status;
3114         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3115
3116         new = create_client(clname, rqstp, &clverifier);
3117         if (new == NULL)
3118                 return nfserr_jukebox;
3119         /* Cases below refer to rfc 3530 section 14.2.33: */
3120         spin_lock(&nn->client_lock);
3121         conf = find_confirmed_client_by_name(&clname, nn);
3122         if (conf && client_has_state(conf)) {
3123                 /* case 0: */
3124                 status = nfserr_clid_inuse;
3125                 if (clp_used_exchangeid(conf))
3126                         goto out;
3127                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3128                         char addr_str[INET6_ADDRSTRLEN];
3129                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3130                                  sizeof(addr_str));
3131                         dprintk("NFSD: setclientid: string in use by client "
3132                                 "at %s\n", addr_str);
3133                         goto out;
3134                 }
3135         }
3136         unconf = find_unconfirmed_client_by_name(&clname, nn);
3137         if (unconf)
3138                 unhash_client_locked(unconf);
3139         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3140                 /* case 1: probable callback update */
3141                 copy_clid(new, conf);
3142                 gen_confirm(new, nn);
3143         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3144                 gen_clid(new, nn);
3145         new->cl_minorversion = 0;
3146         gen_callback(new, setclid, rqstp);
3147         add_to_unconfirmed(new);
3148         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3149         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3150         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3151         new = NULL;
3152         status = nfs_ok;
3153 out:
3154         spin_unlock(&nn->client_lock);
3155         if (new)
3156                 free_client(new);
3157         if (unconf)
3158                 expire_client(unconf);
3159         return status;
3160 }
3161
3162
3163 __be32
3164 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3165                          struct nfsd4_compound_state *cstate,
3166                          struct nfsd4_setclientid_confirm *setclientid_confirm)
3167 {
3168         struct nfs4_client *conf, *unconf;
3169         struct nfs4_client *old = NULL;
3170         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
3171         clientid_t * clid = &setclientid_confirm->sc_clientid;
3172         __be32 status;
3173         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3174
3175         if (STALE_CLIENTID(clid, nn))
3176                 return nfserr_stale_clientid;
3177
3178         spin_lock(&nn->client_lock);
3179         conf = find_confirmed_client(clid, false, nn);
3180         unconf = find_unconfirmed_client(clid, false, nn);
3181         /*
3182          * We try hard to give out unique clientid's, so if we get an
3183          * attempt to confirm the same clientid with a different cred,
3184          * the client may be buggy; this should never happen.
3185          *
3186          * Nevertheless, RFC 7530 recommends INUSE for this case:
3187          */
3188         status = nfserr_clid_inuse;
3189         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3190                 goto out;
3191         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3192                 goto out;
3193         /* cases below refer to rfc 3530 section 14.2.34: */
3194         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3195                 if (conf && !unconf) /* case 2: probable retransmit */
3196                         status = nfs_ok;
3197                 else /* case 4: client hasn't noticed we rebooted yet? */
3198                         status = nfserr_stale_clientid;
3199                 goto out;
3200         }
3201         status = nfs_ok;
3202         if (conf) { /* case 1: callback update */
3203                 old = unconf;
3204                 unhash_client_locked(old);
3205                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3206         } else { /* case 3: normal case; new or rebooted client */
3207                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3208                 if (old) {
3209                         status = nfserr_clid_inuse;
3210                         if (client_has_state(old)
3211                                         && !same_creds(&unconf->cl_cred,
3212                                                         &old->cl_cred))
3213                                 goto out;
3214                         status = mark_client_expired_locked(old);
3215                         if (status) {
3216                                 old = NULL;
3217                                 goto out;
3218                         }
3219                 }
3220                 move_to_confirmed(unconf);
3221                 conf = unconf;
3222         }
3223         get_client_locked(conf);
3224         spin_unlock(&nn->client_lock);
3225         nfsd4_probe_callback(conf);
3226         spin_lock(&nn->client_lock);
3227         put_client_renew_locked(conf);
3228 out:
3229         spin_unlock(&nn->client_lock);
3230         if (old)
3231                 expire_client(old);
3232         return status;
3233 }
3234
3235 static struct nfs4_file *nfsd4_alloc_file(void)
3236 {
3237         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3238 }
3239
3240 /* OPEN Share state helper functions */
3241 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3242                                 struct nfs4_file *fp)
3243 {
3244         lockdep_assert_held(&state_lock);
3245
3246         atomic_set(&fp->fi_ref, 1);
3247         spin_lock_init(&fp->fi_lock);
3248         INIT_LIST_HEAD(&fp->fi_stateids);
3249         INIT_LIST_HEAD(&fp->fi_delegations);
3250         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3251         fh_copy_shallow(&fp->fi_fhandle, fh);
3252         fp->fi_deleg_file = NULL;
3253         fp->fi_had_conflict = false;
3254         fp->fi_share_deny = 0;
3255         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3256         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3257 #ifdef CONFIG_NFSD_PNFS
3258         INIT_LIST_HEAD(&fp->fi_lo_states);
3259         atomic_set(&fp->fi_lo_recalls, 0);
3260 #endif
3261         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3262 }
3263
3264 void
3265 nfsd4_free_slabs(void)
3266 {
3267         kmem_cache_destroy(odstate_slab);
3268         kmem_cache_destroy(openowner_slab);
3269         kmem_cache_destroy(lockowner_slab);
3270         kmem_cache_destroy(file_slab);
3271         kmem_cache_destroy(stateid_slab);
3272         kmem_cache_destroy(deleg_slab);
3273 }
3274
3275 int
3276 nfsd4_init_slabs(void)
3277 {
3278         openowner_slab = kmem_cache_create("nfsd4_openowners",
3279                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3280         if (openowner_slab == NULL)
3281                 goto out;
3282         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3283                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3284         if (lockowner_slab == NULL)
3285                 goto out_free_openowner_slab;
3286         file_slab = kmem_cache_create("nfsd4_files",
3287                         sizeof(struct nfs4_file), 0, 0, NULL);
3288         if (file_slab == NULL)
3289                 goto out_free_lockowner_slab;
3290         stateid_slab = kmem_cache_create("nfsd4_stateids",
3291                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3292         if (stateid_slab == NULL)
3293                 goto out_free_file_slab;
3294         deleg_slab = kmem_cache_create("nfsd4_delegations",
3295                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3296         if (deleg_slab == NULL)
3297                 goto out_free_stateid_slab;
3298         odstate_slab = kmem_cache_create("nfsd4_odstate",
3299                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3300         if (odstate_slab == NULL)
3301                 goto out_free_deleg_slab;
3302         return 0;
3303
3304 out_free_deleg_slab:
3305         kmem_cache_destroy(deleg_slab);
3306 out_free_stateid_slab:
3307         kmem_cache_destroy(stateid_slab);
3308 out_free_file_slab:
3309         kmem_cache_destroy(file_slab);
3310 out_free_lockowner_slab:
3311         kmem_cache_destroy(lockowner_slab);
3312 out_free_openowner_slab:
3313         kmem_cache_destroy(openowner_slab);
3314 out:
3315         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3316         return -ENOMEM;
3317 }
3318
3319 static void init_nfs4_replay(struct nfs4_replay *rp)
3320 {
3321         rp->rp_status = nfserr_serverfault;
3322         rp->rp_buflen = 0;
3323         rp->rp_buf = rp->rp_ibuf;
3324         mutex_init(&rp->rp_mutex);
3325 }
3326
3327 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3328                 struct nfs4_stateowner *so)
3329 {
3330         if (!nfsd4_has_session(cstate)) {
3331                 mutex_lock(&so->so_replay.rp_mutex);
3332                 cstate->replay_owner = nfs4_get_stateowner(so);
3333         }
3334 }
3335
3336 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3337 {
3338         struct nfs4_stateowner *so = cstate->replay_owner;
3339
3340         if (so != NULL) {
3341                 cstate->replay_owner = NULL;
3342                 mutex_unlock(&so->so_replay.rp_mutex);
3343                 nfs4_put_stateowner(so);
3344         }
3345 }
3346
3347 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3348 {
3349         struct nfs4_stateowner *sop;
3350
3351         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3352         if (!sop)
3353                 return NULL;
3354
3355         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3356         if (!sop->so_owner.data) {
3357                 kmem_cache_free(slab, sop);
3358                 return NULL;
3359         }
3360         sop->so_owner.len = owner->len;
3361
3362         INIT_LIST_HEAD(&sop->so_stateids);
3363         sop->so_client = clp;
3364         init_nfs4_replay(&sop->so_replay);
3365         atomic_set(&sop->so_count, 1);
3366         return sop;
3367 }
3368
3369 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3370 {
3371         lockdep_assert_held(&clp->cl_lock);
3372
3373         list_add(&oo->oo_owner.so_strhash,
3374                  &clp->cl_ownerstr_hashtbl[strhashval]);
3375         list_add(&oo->oo_perclient, &clp->cl_openowners);
3376 }
3377
3378 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3379 {
3380         unhash_openowner_locked(openowner(so));
3381 }
3382
3383 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3384 {
3385         struct nfs4_openowner *oo = openowner(so);
3386
3387         kmem_cache_free(openowner_slab, oo);
3388 }
3389
3390 static const struct nfs4_stateowner_operations openowner_ops = {
3391         .so_unhash =    nfs4_unhash_openowner,
3392         .so_free =      nfs4_free_openowner,
3393 };
3394
3395 static struct nfs4_openowner *
3396 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3397                            struct nfsd4_compound_state *cstate)
3398 {
3399         struct nfs4_client *clp = cstate->clp;
3400         struct nfs4_openowner *oo, *ret;
3401
3402         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3403         if (!oo)
3404                 return NULL;
3405         oo->oo_owner.so_ops = &openowner_ops;
3406         oo->oo_owner.so_is_open_owner = 1;
3407         oo->oo_owner.so_seqid = open->op_seqid;
3408         oo->oo_flags = 0;
3409         if (nfsd4_has_session(cstate))
3410                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3411         oo->oo_time = 0;
3412         oo->oo_last_closed_stid = NULL;
3413         INIT_LIST_HEAD(&oo->oo_close_lru);
3414         spin_lock(&clp->cl_lock);
3415         ret = find_openstateowner_str_locked(strhashval, open, clp);
3416         if (ret == NULL) {
3417                 hash_openowner(oo, clp, strhashval);
3418                 ret = oo;
3419         } else
3420                 nfs4_free_stateowner(&oo->oo_owner);
3421
3422         spin_unlock(&clp->cl_lock);
3423         return ret;
3424 }
3425
3426 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
3427         struct nfs4_openowner *oo = open->op_openowner;
3428
3429         atomic_inc(&stp->st_stid.sc_count);
3430         stp->st_stid.sc_type = NFS4_OPEN_STID;
3431         INIT_LIST_HEAD(&stp->st_locks);
3432         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3433         get_nfs4_file(fp);
3434         stp->st_stid.sc_file = fp;
3435         stp->st_access_bmap = 0;
3436         stp->st_deny_bmap = 0;
3437         stp->st_openstp = NULL;
3438         init_rwsem(&stp->st_rwsem);
3439         spin_lock(&oo->oo_owner.so_client->cl_lock);
3440         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3441         spin_lock(&fp->fi_lock);
3442         list_add(&stp->st_perfile, &fp->fi_stateids);
3443         spin_unlock(&fp->fi_lock);
3444         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3445 }
3446
3447 /*
3448  * In the 4.0 case we need to keep the owners around a little while to handle
3449  * CLOSE replay. We still do need to release any file access that is held by
3450  * them before returning however.
3451  */
3452 static void
3453 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3454 {
3455         struct nfs4_ol_stateid *last;
3456         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3457         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3458                                                 nfsd_net_id);
3459
3460         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3461
3462         /*
3463          * We know that we hold one reference via nfsd4_close, and another
3464          * "persistent" reference for the client. If the refcount is higher
3465          * than 2, then there are still calls in progress that are using this
3466          * stateid. We can't put the sc_file reference until they are finished.
3467          * Wait for the refcount to drop to 2. Since it has been unhashed,
3468          * there should be no danger of the refcount going back up again at
3469          * this point.
3470          */
3471         wait_event(close_wq, atomic_read(&s->st_stid.sc_count) == 2);
3472
3473         release_all_access(s);
3474         if (s->st_stid.sc_file) {
3475                 put_nfs4_file(s->st_stid.sc_file);
3476                 s->st_stid.sc_file = NULL;
3477         }
3478
3479         spin_lock(&nn->client_lock);
3480         last = oo->oo_last_closed_stid;
3481         oo->oo_last_closed_stid = s;
3482         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3483         oo->oo_time = get_seconds();
3484         spin_unlock(&nn->client_lock);
3485         if (last)
3486                 nfs4_put_stid(&last->st_stid);
3487 }
3488
3489 /* search file_hashtbl[] for file */
3490 static struct nfs4_file *
3491 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3492 {
3493         struct nfs4_file *fp;
3494
3495         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3496                 if (fh_match(&fp->fi_fhandle, fh)) {
3497                         if (atomic_inc_not_zero(&fp->fi_ref))
3498                                 return fp;
3499                 }
3500         }
3501         return NULL;
3502 }
3503
3504 struct nfs4_file *
3505 find_file(struct knfsd_fh *fh)
3506 {
3507         struct nfs4_file *fp;
3508         unsigned int hashval = file_hashval(fh);
3509
3510         rcu_read_lock();
3511         fp = find_file_locked(fh, hashval);
3512         rcu_read_unlock();
3513         return fp;
3514 }
3515
3516 static struct nfs4_file *
3517 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3518 {
3519         struct nfs4_file *fp;
3520         unsigned int hashval = file_hashval(fh);
3521
3522         rcu_read_lock();
3523         fp = find_file_locked(fh, hashval);
3524         rcu_read_unlock();
3525         if (fp)
3526                 return fp;
3527
3528         spin_lock(&state_lock);
3529         fp = find_file_locked(fh, hashval);
3530         if (likely(fp == NULL)) {
3531                 nfsd4_init_file(fh, hashval, new);
3532                 fp = new;
3533         }
3534         spin_unlock(&state_lock);
3535
3536         return fp;
3537 }
3538
3539 /*
3540  * Called to check deny when READ with all zero stateid or
3541  * WRITE with all zero or all one stateid
3542  */
3543 static __be32
3544 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3545 {
3546         struct nfs4_file *fp;
3547         __be32 ret = nfs_ok;
3548
3549         fp = find_file(&current_fh->fh_handle);
3550         if (!fp)
3551                 return ret;
3552         /* Check for conflicting share reservations */
3553         spin_lock(&fp->fi_lock);
3554         if (fp->fi_share_deny & deny_type)
3555                 ret = nfserr_locked;
3556         spin_unlock(&fp->fi_lock);
3557         put_nfs4_file(fp);
3558         return ret;
3559 }
3560
3561 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3562 {
3563         struct nfs4_delegation *dp = cb_to_delegation(cb);
3564         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3565                                           nfsd_net_id);
3566
3567         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3568
3569         /*
3570          * We can't do this in nfsd_break_deleg_cb because it is
3571          * already holding inode->i_lock.
3572          *
3573          * If the dl_time != 0, then we know that it has already been
3574          * queued for a lease break. Don't queue it again.
3575          */
3576         spin_lock(&state_lock);
3577         if (dp->dl_time == 0) {
3578                 dp->dl_time = get_seconds();
3579                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3580         }
3581         spin_unlock(&state_lock);
3582 }
3583
3584 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3585                 struct rpc_task *task)
3586 {
3587         struct nfs4_delegation *dp = cb_to_delegation(cb);
3588
3589         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3590                 return 1;
3591
3592         switch (task->tk_status) {
3593         case 0:
3594                 return 1;
3595         case -EBADHANDLE:
3596         case -NFS4ERR_BAD_STATEID:
3597                 /*
3598                  * Race: client probably got cb_recall before open reply
3599                  * granting delegation.
3600                  */
3601                 if (dp->dl_retries--) {
3602                         rpc_delay(task, 2 * HZ);
3603                         return 0;
3604                 }
3605                 /*FALLTHRU*/
3606         default:
3607                 return -1;
3608         }
3609 }
3610
3611 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3612 {
3613         struct nfs4_delegation *dp = cb_to_delegation(cb);
3614
3615         nfs4_put_stid(&dp->dl_stid);
3616 }
3617
3618 static struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3619         .prepare        = nfsd4_cb_recall_prepare,
3620         .done           = nfsd4_cb_recall_done,
3621         .release        = nfsd4_cb_recall_release,
3622 };
3623
3624 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3625 {
3626         /*
3627          * We're assuming the state code never drops its reference
3628          * without first removing the lease.  Since we're in this lease
3629          * callback (and since the lease code is serialized by the kernel
3630          * lock) we know the server hasn't removed the lease yet, we know
3631          * it's safe to take a reference.
3632          */
3633         atomic_inc(&dp->dl_stid.sc_count);
3634         nfsd4_run_cb(&dp->dl_recall);
3635 }
3636
3637 /* Called from break_lease() with i_lock held. */
3638 static bool
3639 nfsd_break_deleg_cb(struct file_lock *fl)
3640 {
3641         bool ret = false;
3642         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3643         struct nfs4_delegation *dp;
3644
3645         if (!fp) {
3646                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3647                 return ret;
3648         }
3649         if (fp->fi_had_conflict) {
3650                 WARN(1, "duplicate break on %p\n", fp);
3651                 return ret;
3652         }
3653         /*
3654          * We don't want the locks code to timeout the lease for us;
3655          * we'll remove it ourself if a delegation isn't returned
3656          * in time:
3657          */
3658         fl->fl_break_time = 0;
3659
3660         spin_lock(&fp->fi_lock);
3661         fp->fi_had_conflict = true;
3662         /*
3663          * If there are no delegations on the list, then return true
3664          * so that the lease code will go ahead and delete it.
3665          */
3666         if (list_empty(&fp->fi_delegations))
3667                 ret = true;
3668         else
3669                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3670                         nfsd_break_one_deleg(dp);
3671         spin_unlock(&fp->fi_lock);
3672         return ret;
3673 }
3674
3675 static int
3676 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3677                      struct list_head *dispose)
3678 {
3679         if (arg & F_UNLCK)
3680                 return lease_modify(onlist, arg, dispose);
3681         else
3682                 return -EAGAIN;
3683 }
3684
3685 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3686         .lm_break = nfsd_break_deleg_cb,
3687         .lm_change = nfsd_change_deleg_cb,
3688 };
3689
3690 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3691 {
3692         if (nfsd4_has_session(cstate))
3693                 return nfs_ok;
3694         if (seqid == so->so_seqid - 1)
3695                 return nfserr_replay_me;
3696         if (seqid == so->so_seqid)
3697                 return nfs_ok;
3698         return nfserr_bad_seqid;
3699 }
3700
3701 static __be32 lookup_clientid(clientid_t *clid,
3702                 struct nfsd4_compound_state *cstate,
3703                 struct nfsd_net *nn)
3704 {
3705         struct nfs4_client *found;
3706
3707         if (cstate->clp) {
3708                 found = cstate->clp;
3709                 if (!same_clid(&found->cl_clientid, clid))
3710                         return nfserr_stale_clientid;
3711                 return nfs_ok;
3712         }
3713
3714         if (STALE_CLIENTID(clid, nn))
3715                 return nfserr_stale_clientid;
3716
3717         /*
3718          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3719          * cached already then we know this is for is for v4.0 and "sessions"
3720          * will be false.
3721          */
3722         WARN_ON_ONCE(cstate->session);
3723         spin_lock(&nn->client_lock);
3724         found = find_confirmed_client(clid, false, nn);
3725         if (!found) {
3726                 spin_unlock(&nn->client_lock);
3727                 return nfserr_expired;
3728         }
3729         atomic_inc(&found->cl_refcount);
3730         spin_unlock(&nn->client_lock);
3731
3732         /* Cache the nfs4_client in cstate! */
3733         cstate->clp = found;
3734         return nfs_ok;
3735 }
3736
3737 __be32
3738 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
3739                     struct nfsd4_open *open, struct nfsd_net *nn)
3740 {
3741         clientid_t *clientid = &open->op_clientid;
3742         struct nfs4_client *clp = NULL;
3743         unsigned int strhashval;
3744         struct nfs4_openowner *oo = NULL;
3745         __be32 status;
3746
3747         if (STALE_CLIENTID(&open->op_clientid, nn))
3748                 return nfserr_stale_clientid;
3749         /*
3750          * In case we need it later, after we've already created the
3751          * file and don't want to risk a further failure:
3752          */
3753         open->op_file = nfsd4_alloc_file();
3754         if (open->op_file == NULL)
3755                 return nfserr_jukebox;
3756
3757         status = lookup_clientid(clientid, cstate, nn);
3758         if (status)
3759                 return status;
3760         clp = cstate->clp;
3761
3762         strhashval = ownerstr_hashval(&open->op_owner);
3763         oo = find_openstateowner_str(strhashval, open, clp);
3764         open->op_openowner = oo;
3765         if (!oo) {
3766                 goto new_owner;
3767         }
3768         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
3769                 /* Replace unconfirmed owners without checking for replay. */
3770                 release_openowner(oo);
3771                 open->op_openowner = NULL;
3772                 goto new_owner;
3773         }
3774         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
3775         if (status)
3776                 return status;
3777         goto alloc_stateid;
3778 new_owner:
3779         oo = alloc_init_open_stateowner(strhashval, open, cstate);
3780         if (oo == NULL)
3781                 return nfserr_jukebox;
3782         open->op_openowner = oo;
3783 alloc_stateid:
3784         open->op_stp = nfs4_alloc_open_stateid(clp);
3785         if (!open->op_stp)
3786                 return nfserr_jukebox;
3787
3788         if (nfsd4_has_session(cstate) &&
3789             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
3790                 open->op_odstate = alloc_clnt_odstate(clp);
3791                 if (!open->op_odstate)
3792                         return nfserr_jukebox;
3793         }
3794
3795         return nfs_ok;
3796 }
3797
3798 static inline __be32
3799 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
3800 {
3801         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
3802                 return nfserr_openmode;
3803         else
3804                 return nfs_ok;
3805 }
3806
3807 static int share_access_to_flags(u32 share_access)
3808 {
3809         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
3810 }
3811
3812 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
3813 {
3814         struct nfs4_stid *ret;
3815
3816         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
3817         if (!ret)
3818                 return NULL;
3819         return delegstateid(ret);
3820 }
3821
3822 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
3823 {
3824         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
3825                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
3826 }
3827
3828 static __be32
3829 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
3830                 struct nfs4_delegation **dp)
3831 {
3832         int flags;
3833         __be32 status = nfserr_bad_stateid;
3834         struct nfs4_delegation *deleg;
3835
3836         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
3837         if (deleg == NULL)
3838                 goto out;
3839         flags = share_access_to_flags(open->op_share_access);
3840         status = nfs4_check_delegmode(deleg, flags);
3841         if (status) {
3842                 nfs4_put_stid(&deleg->dl_stid);
3843                 goto out;
3844         }
3845         *dp = deleg;
3846 out:
3847         if (!nfsd4_is_deleg_cur(open))
3848                 return nfs_ok;
3849         if (status)
3850                 return status;
3851         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3852         return nfs_ok;
3853 }
3854
3855 static struct nfs4_ol_stateid *
3856 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3857 {
3858         struct nfs4_ol_stateid *local, *ret = NULL;
3859         struct nfs4_openowner *oo = open->op_openowner;
3860
3861         spin_lock(&fp->fi_lock);
3862         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3863                 /* ignore lock owners */
3864                 if (local->st_stateowner->so_is_open_owner == 0)
3865                         continue;
3866                 if (local->st_stateowner == &oo->oo_owner) {
3867                         ret = local;
3868                         atomic_inc(&ret->st_stid.sc_count);
3869                         break;
3870                 }
3871         }
3872         spin_unlock(&fp->fi_lock);
3873         return ret;
3874 }
3875
3876 static inline int nfs4_access_to_access(u32 nfs4_access)
3877 {
3878         int flags = 0;
3879
3880         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
3881                 flags |= NFSD_MAY_READ;
3882         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
3883                 flags |= NFSD_MAY_WRITE;
3884         return flags;
3885 }
3886
3887 static inline __be32
3888 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
3889                 struct nfsd4_open *open)
3890 {
3891         struct iattr iattr = {
3892                 .ia_valid = ATTR_SIZE,
3893                 .ia_size = 0,
3894         };
3895         if (!open->op_truncate)
3896                 return 0;
3897         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
3898                 return nfserr_inval;
3899         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
3900 }
3901
3902 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
3903                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
3904                 struct nfsd4_open *open)
3905 {
3906         struct file *filp = NULL;
3907         __be32 status;
3908         int oflag = nfs4_access_to_omode(open->op_share_access);
3909         int access = nfs4_access_to_access(open->op_share_access);
3910         unsigned char old_access_bmap, old_deny_bmap;
3911
3912         spin_lock(&fp->fi_lock);
3913
3914         /*
3915          * Are we trying to set a deny mode that would conflict with
3916          * current access?
3917          */
3918         status = nfs4_file_check_deny(fp, open->op_share_deny);
3919         if (status != nfs_ok) {
3920                 spin_unlock(&fp->fi_lock);
3921                 goto out;
3922         }
3923
3924         /* set access to the file */
3925         status = nfs4_file_get_access(fp, open->op_share_access);
3926         if (status != nfs_ok) {
3927                 spin_unlock(&fp->fi_lock);
3928                 goto out;
3929         }
3930
3931         /* Set access bits in stateid */
3932         old_access_bmap = stp->st_access_bmap;
3933         set_access(open->op_share_access, stp);
3934
3935         /* Set new deny mask */
3936         old_deny_bmap = stp->st_deny_bmap;
3937         set_deny(open->op_share_deny, stp);
3938         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3939
3940         if (!fp->fi_fds[oflag]) {
3941                 spin_unlock(&fp->fi_lock);
3942                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
3943                 if (status)
3944                         goto out_put_access;
3945                 spin_lock(&fp->fi_lock);
3946                 if (!fp->fi_fds[oflag]) {
3947                         fp->fi_fds[oflag] = filp;
3948                         filp = NULL;
3949                 }
3950         }
3951         spin_unlock(&fp->fi_lock);
3952         if (filp)
3953                 fput(filp);
3954
3955         status = nfsd4_truncate(rqstp, cur_fh, open);
3956         if (status)
3957                 goto out_put_access;
3958 out:
3959         return status;
3960 out_put_access:
3961         stp->st_access_bmap = old_access_bmap;
3962         nfs4_file_put_access(fp, open->op_share_access);
3963         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
3964         goto out;
3965 }
3966
3967 static __be32
3968 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
3969 {
3970         __be32 status;
3971         unsigned char old_deny_bmap = stp->st_deny_bmap;
3972
3973         if (!test_access(open->op_share_access, stp))
3974                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
3975
3976         /* test and set deny mode */
3977         spin_lock(&fp->fi_lock);
3978         status = nfs4_file_check_deny(fp, open->op_share_deny);
3979         if (status == nfs_ok) {
3980                 set_deny(open->op_share_deny, stp);
3981                 fp->fi_share_deny |=
3982                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
3983         }
3984         spin_unlock(&fp->fi_lock);
3985
3986         if (status != nfs_ok)
3987                 return status;
3988
3989         status = nfsd4_truncate(rqstp, cur_fh, open);
3990         if (status != nfs_ok)
3991                 reset_union_bmap_deny(old_deny_bmap, stp);
3992         return status;
3993 }
3994
3995 /* Should we give out recallable state?: */
3996 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
3997 {
3998         if (clp->cl_cb_state == NFSD4_CB_UP)
3999                 return true;
4000         /*
4001          * In the sessions case, since we don't have to establish a
4002          * separate connection for callbacks, we assume it's OK
4003          * until we hear otherwise:
4004          */
4005         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4006 }
4007
4008 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4009 {
4010         struct file_lock *fl;
4011
4012         fl = locks_alloc_lock();
4013         if (!fl)
4014                 return NULL;
4015         fl->fl_lmops = &nfsd_lease_mng_ops;
4016         fl->fl_flags = FL_DELEG;
4017         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4018         fl->fl_end = OFFSET_MAX;
4019         fl->fl_owner = (fl_owner_t)fp;
4020         fl->fl_pid = current->tgid;
4021         return fl;
4022 }
4023
4024 /**
4025  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4026  * @dp:   a pointer to the nfs4_delegation we're adding.
4027  *
4028  * Return:
4029  *      On success: Return code will be 0 on success.
4030  *
4031  *      On error: -EAGAIN if there was an existing delegation.
4032  *                 nonzero if there is an error in other cases.
4033  *
4034  */
4035
4036 static int nfs4_setlease(struct nfs4_delegation *dp)
4037 {
4038         struct nfs4_file *fp = dp->dl_stid.sc_file;
4039         struct file_lock *fl;
4040         struct file *filp;
4041         int status = 0;
4042
4043         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4044         if (!fl)
4045                 return -ENOMEM;
4046         filp = find_readable_file(fp);
4047         if (!filp) {
4048                 /* We should always have a readable file here */
4049                 WARN_ON_ONCE(1);
4050                 locks_free_lock(fl);
4051                 return -EBADF;
4052         }
4053         fl->fl_file = filp;
4054         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4055         if (fl)
4056                 locks_free_lock(fl);
4057         if (status)
4058                 goto out_fput;
4059         spin_lock(&state_lock);
4060         spin_lock(&fp->fi_lock);
4061         /* Did the lease get broken before we took the lock? */
4062         status = -EAGAIN;
4063         if (fp->fi_had_conflict)
4064                 goto out_unlock;
4065         /* Race breaker */
4066         if (fp->fi_deleg_file) {
4067                 status = hash_delegation_locked(dp, fp);
4068                 goto out_unlock;
4069         }
4070         fp->fi_deleg_file = filp;
4071         fp->fi_delegees = 0;
4072         status = hash_delegation_locked(dp, fp);
4073         spin_unlock(&fp->fi_lock);
4074         spin_unlock(&state_lock);
4075         if (status) {
4076                 /* Should never happen, this is a new fi_deleg_file  */
4077                 WARN_ON_ONCE(1);
4078                 goto out_fput;
4079         }
4080         return 0;
4081 out_unlock:
4082         spin_unlock(&fp->fi_lock);
4083         spin_unlock(&state_lock);
4084 out_fput:
4085         fput(filp);
4086         return status;
4087 }
4088
4089 static struct nfs4_delegation *
4090 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4091                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4092 {
4093         int status;
4094         struct nfs4_delegation *dp;
4095
4096         if (fp->fi_had_conflict)
4097                 return ERR_PTR(-EAGAIN);
4098
4099         spin_lock(&state_lock);
4100         spin_lock(&fp->fi_lock);
4101         status = nfs4_get_existing_delegation(clp, fp);
4102         spin_unlock(&fp->fi_lock);
4103         spin_unlock(&state_lock);
4104
4105         if (status)
4106                 return ERR_PTR(status);
4107
4108         dp = alloc_init_deleg(clp, fh, odstate);
4109         if (!dp)
4110                 return ERR_PTR(-ENOMEM);
4111
4112         get_nfs4_file(fp);
4113         spin_lock(&state_lock);
4114         spin_lock(&fp->fi_lock);
4115         dp->dl_stid.sc_file = fp;
4116         if (!fp->fi_deleg_file) {
4117                 spin_unlock(&fp->fi_lock);
4118                 spin_unlock(&state_lock);
4119                 status = nfs4_setlease(dp);
4120                 goto out;
4121         }
4122         if (fp->fi_had_conflict) {
4123                 status = -EAGAIN;
4124                 goto out_unlock;
4125         }
4126         status = hash_delegation_locked(dp, fp);
4127 out_unlock:
4128         spin_unlock(&fp->fi_lock);
4129         spin_unlock(&state_lock);
4130 out:
4131         if (status) {
4132                 put_clnt_odstate(dp->dl_clnt_odstate);
4133                 nfs4_put_stid(&dp->dl_stid);
4134                 return ERR_PTR(status);
4135         }
4136         return dp;
4137 }
4138
4139 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4140 {
4141         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4142         if (status == -EAGAIN)
4143                 open->op_why_no_deleg = WND4_CONTENTION;
4144         else {
4145                 open->op_why_no_deleg = WND4_RESOURCE;
4146                 switch (open->op_deleg_want) {
4147                 case NFS4_SHARE_WANT_READ_DELEG:
4148                 case NFS4_SHARE_WANT_WRITE_DELEG:
4149                 case NFS4_SHARE_WANT_ANY_DELEG:
4150                         break;
4151                 case NFS4_SHARE_WANT_CANCEL:
4152                         open->op_why_no_deleg = WND4_CANCELLED;
4153                         break;
4154                 case NFS4_SHARE_WANT_NO_DELEG:
4155                         WARN_ON_ONCE(1);
4156                 }
4157         }
4158 }
4159
4160 /*
4161  * Attempt to hand out a delegation.
4162  *
4163  * Note we don't support write delegations, and won't until the vfs has
4164  * proper support for them.
4165  */
4166 static void
4167 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4168                         struct nfs4_ol_stateid *stp)
4169 {
4170         struct nfs4_delegation *dp;
4171         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4172         struct nfs4_client *clp = stp->st_stid.sc_client;
4173         int cb_up;
4174         int status = 0;
4175
4176         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4177         open->op_recall = 0;
4178         switch (open->op_claim_type) {
4179                 case NFS4_OPEN_CLAIM_PREVIOUS:
4180                         if (!cb_up)
4181                                 open->op_recall = 1;
4182                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4183                                 goto out_no_deleg;
4184                         break;
4185                 case NFS4_OPEN_CLAIM_NULL:
4186                 case NFS4_OPEN_CLAIM_FH:
4187                         /*
4188                          * Let's not give out any delegations till everyone's
4189                          * had the chance to reclaim theirs, *and* until
4190                          * NLM locks have all been reclaimed:
4191                          */
4192                         if (locks_in_grace(clp->net))
4193                                 goto out_no_deleg;
4194                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4195                                 goto out_no_deleg;
4196                         /*
4197                          * Also, if the file was opened for write or
4198                          * create, there's a good chance the client's
4199                          * about to write to it, resulting in an
4200                          * immediate recall (since we don't support
4201                          * write delegations):
4202                          */
4203                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4204                                 goto out_no_deleg;
4205                         if (open->op_create == NFS4_OPEN_CREATE)
4206                                 goto out_no_deleg;
4207                         break;
4208                 default:
4209                         goto out_no_deleg;
4210         }
4211         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4212         if (IS_ERR(dp))
4213                 goto out_no_deleg;
4214
4215         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4216
4217         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4218                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4219         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4220         nfs4_put_stid(&dp->dl_stid);
4221         return;
4222 out_no_deleg:
4223         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4224         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4225             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4226                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4227                 open->op_recall = 1;
4228         }
4229
4230         /* 4.1 client asking for a delegation? */
4231         if (open->op_deleg_want)
4232                 nfsd4_open_deleg_none_ext(open, status);
4233         return;
4234 }
4235
4236 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4237                                         struct nfs4_delegation *dp)
4238 {
4239         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4240             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4241                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4242                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4243         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4244                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4245                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4246                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4247         }
4248         /* Otherwise the client must be confused wanting a delegation
4249          * it already has, therefore we don't return
4250          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4251          */
4252 }
4253
4254 __be32
4255 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4256 {
4257         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4258         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4259         struct nfs4_file *fp = NULL;
4260         struct nfs4_ol_stateid *stp = NULL;
4261         struct nfs4_delegation *dp = NULL;
4262         __be32 status;
4263
4264         /*
4265          * Lookup file; if found, lookup stateid and check open request,
4266          * and check for delegations in the process of being recalled.
4267          * If not found, create the nfs4_file struct
4268          */
4269         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4270         if (fp != open->op_file) {
4271                 status = nfs4_check_deleg(cl, open, &dp);
4272                 if (status)
4273                         goto out;
4274                 stp = nfsd4_find_existing_open(fp, open);
4275         } else {
4276                 open->op_file = NULL;
4277                 status = nfserr_bad_stateid;
4278                 if (nfsd4_is_deleg_cur(open))
4279                         goto out;
4280         }
4281
4282         /*
4283          * OPEN the file, or upgrade an existing OPEN.
4284          * If truncate fails, the OPEN fails.
4285          */
4286         if (stp) {
4287                 /* Stateid was found, this is an OPEN upgrade */
4288                 down_read(&stp->st_rwsem);
4289                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4290                 if (status) {
4291                         up_read(&stp->st_rwsem);
4292                         goto out;
4293                 }
4294         } else {
4295                 stp = open->op_stp;
4296                 open->op_stp = NULL;
4297                 init_open_stateid(stp, fp, open);
4298                 down_read(&stp->st_rwsem);
4299                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4300                 if (status) {
4301                         up_read(&stp->st_rwsem);
4302                         release_open_stateid(stp);
4303                         goto out;
4304                 }
4305
4306                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4307                                                         open->op_odstate);
4308                 if (stp->st_clnt_odstate == open->op_odstate)
4309                         open->op_odstate = NULL;
4310         }
4311         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4312         up_read(&stp->st_rwsem);
4313
4314         if (nfsd4_has_session(&resp->cstate)) {
4315                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4316                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4317                         open->op_why_no_deleg = WND4_NOT_WANTED;
4318                         goto nodeleg;
4319                 }
4320         }
4321
4322         /*
4323         * Attempt to hand out a delegation. No error return, because the
4324         * OPEN succeeds even if we fail.
4325         */
4326         nfs4_open_delegation(current_fh, open, stp);
4327 nodeleg:
4328         status = nfs_ok;
4329
4330         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4331                 STATEID_VAL(&stp->st_stid.sc_stateid));
4332 out:
4333         /* 4.1 client trying to upgrade/downgrade delegation? */
4334         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4335             open->op_deleg_want)
4336                 nfsd4_deleg_xgrade_none_ext(open, dp);
4337
4338         if (fp)
4339                 put_nfs4_file(fp);
4340         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4341                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4342         /*
4343         * To finish the open response, we just need to set the rflags.
4344         */
4345         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4346         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
4347             !nfsd4_has_session(&resp->cstate))
4348                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4349         if (dp)
4350                 nfs4_put_stid(&dp->dl_stid);
4351         if (stp)
4352                 nfs4_put_stid(&stp->st_stid);
4353
4354         return status;
4355 }
4356
4357 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4358                               struct nfsd4_open *open)
4359 {
4360         if (open->op_openowner) {
4361                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4362
4363                 nfsd4_cstate_assign_replay(cstate, so);
4364                 nfs4_put_stateowner(so);
4365         }
4366         if (open->op_file)
4367                 kmem_cache_free(file_slab, open->op_file);
4368         if (open->op_stp)
4369                 nfs4_put_stid(&open->op_stp->st_stid);
4370         if (open->op_odstate)
4371                 kmem_cache_free(odstate_slab, open->op_odstate);
4372 }
4373
4374 __be32
4375 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4376             clientid_t *clid)
4377 {
4378         struct nfs4_client *clp;
4379         __be32 status;
4380         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4381
4382         dprintk("process_renew(%08x/%08x): starting\n", 
4383                         clid->cl_boot, clid->cl_id);
4384         status = lookup_clientid(clid, cstate, nn);
4385         if (status)
4386                 goto out;
4387         clp = cstate->clp;
4388         status = nfserr_cb_path_down;
4389         if (!list_empty(&clp->cl_delegations)
4390                         && clp->cl_cb_state != NFSD4_CB_UP)
4391                 goto out;
4392         status = nfs_ok;
4393 out:
4394         return status;
4395 }
4396
4397 void
4398 nfsd4_end_grace(struct nfsd_net *nn)
4399 {
4400         /* do nothing if grace period already ended */
4401         if (nn->grace_ended)
4402                 return;
4403
4404         dprintk("NFSD: end of grace period\n");
4405         nn->grace_ended = true;
4406         /*
4407          * If the server goes down again right now, an NFSv4
4408          * client will still be allowed to reclaim after it comes back up,
4409          * even if it hasn't yet had a chance to reclaim state this time.
4410          *
4411          */
4412         nfsd4_record_grace_done(nn);
4413         /*
4414          * At this point, NFSv4 clients can still reclaim.  But if the
4415          * server crashes, any that have not yet reclaimed will be out
4416          * of luck on the next boot.
4417          *
4418          * (NFSv4.1+ clients are considered to have reclaimed once they
4419          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4420          * have reclaimed after their first OPEN.)
4421          */
4422         locks_end_grace(&nn->nfsd4_manager);
4423         /*
4424          * At this point, and once lockd and/or any other containers
4425          * exit their grace period, further reclaims will fail and
4426          * regular locking can resume.
4427          */
4428 }
4429
4430 static time_t
4431 nfs4_laundromat(struct nfsd_net *nn)
4432 {
4433         struct nfs4_client *clp;
4434         struct nfs4_openowner *oo;
4435         struct nfs4_delegation *dp;
4436         struct nfs4_ol_stateid *stp;
4437         struct list_head *pos, *next, reaplist;
4438         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4439         time_t t, new_timeo = nn->nfsd4_lease;
4440
4441         dprintk("NFSD: laundromat service - starting\n");
4442         nfsd4_end_grace(nn);
4443         INIT_LIST_HEAD(&reaplist);
4444         spin_lock(&nn->client_lock);
4445         list_for_each_safe(pos, next, &nn->client_lru) {
4446                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4447                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4448                         t = clp->cl_time - cutoff;
4449                         new_timeo = min(new_timeo, t);
4450                         break;
4451                 }
4452                 if (mark_client_expired_locked(clp)) {
4453                         dprintk("NFSD: client in use (clientid %08x)\n",
4454                                 clp->cl_clientid.cl_id);
4455                         continue;
4456                 }
4457                 list_add(&clp->cl_lru, &reaplist);
4458         }
4459         spin_unlock(&nn->client_lock);
4460         list_for_each_safe(pos, next, &reaplist) {
4461                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4462                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4463                         clp->cl_clientid.cl_id);
4464                 list_del_init(&clp->cl_lru);
4465                 expire_client(clp);
4466         }
4467         spin_lock(&state_lock);
4468         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4469                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4470                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4471                         t = dp->dl_time - cutoff;
4472                         new_timeo = min(new_timeo, t);
4473                         break;
4474                 }
4475                 WARN_ON(!unhash_delegation_locked(dp));
4476                 list_add(&dp->dl_recall_lru, &reaplist);
4477         }
4478         spin_unlock(&state_lock);
4479         while (!list_empty(&reaplist)) {
4480                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4481                                         dl_recall_lru);
4482                 list_del_init(&dp->dl_recall_lru);
4483                 revoke_delegation(dp);
4484         }
4485
4486         spin_lock(&nn->client_lock);
4487         while (!list_empty(&nn->close_lru)) {
4488                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4489                                         oo_close_lru);
4490                 if (time_after((unsigned long)oo->oo_time,
4491                                (unsigned long)cutoff)) {
4492                         t = oo->oo_time - cutoff;
4493                         new_timeo = min(new_timeo, t);
4494                         break;
4495                 }
4496                 list_del_init(&oo->oo_close_lru);
4497                 stp = oo->oo_last_closed_stid;
4498                 oo->oo_last_closed_stid = NULL;
4499                 spin_unlock(&nn->client_lock);
4500                 nfs4_put_stid(&stp->st_stid);
4501                 spin_lock(&nn->client_lock);
4502         }
4503         spin_unlock(&nn->client_lock);
4504
4505         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4506         return new_timeo;
4507 }
4508
4509 static struct workqueue_struct *laundry_wq;
4510 static void laundromat_main(struct work_struct *);
4511
4512 static void
4513 laundromat_main(struct work_struct *laundry)
4514 {
4515         time_t t;
4516         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
4517                                                   work);
4518         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4519                                            laundromat_work);
4520
4521         t = nfs4_laundromat(nn);
4522         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4523         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4524 }
4525
4526 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4527 {
4528         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4529                 return nfserr_bad_stateid;
4530         return nfs_ok;
4531 }
4532
4533 static inline int
4534 access_permit_read(struct nfs4_ol_stateid *stp)
4535 {
4536         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4537                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4538                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4539 }
4540
4541 static inline int
4542 access_permit_write(struct nfs4_ol_stateid *stp)
4543 {
4544         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4545                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4546 }
4547
4548 static
4549 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4550 {
4551         __be32 status = nfserr_openmode;
4552
4553         /* For lock stateid's, we test the parent open, not the lock: */
4554         if (stp->st_openstp)
4555                 stp = stp->st_openstp;
4556         if ((flags & WR_STATE) && !access_permit_write(stp))
4557                 goto out;
4558         if ((flags & RD_STATE) && !access_permit_read(stp))
4559                 goto out;
4560         status = nfs_ok;
4561 out:
4562         return status;
4563 }
4564
4565 static inline __be32
4566 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4567 {
4568         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4569                 return nfs_ok;
4570         else if (opens_in_grace(net)) {
4571                 /* Answer in remaining cases depends on existence of
4572                  * conflicting state; so we must wait out the grace period. */
4573                 return nfserr_grace;
4574         } else if (flags & WR_STATE)
4575                 return nfs4_share_conflict(current_fh,
4576                                 NFS4_SHARE_DENY_WRITE);
4577         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4578                 return nfs4_share_conflict(current_fh,
4579                                 NFS4_SHARE_DENY_READ);
4580 }
4581
4582 /*
4583  * Allow READ/WRITE during grace period on recovered state only for files
4584  * that are not able to provide mandatory locking.
4585  */
4586 static inline int
4587 grace_disallows_io(struct net *net, struct inode *inode)
4588 {
4589         return opens_in_grace(net) && mandatory_lock(inode);
4590 }
4591
4592 /* Returns true iff a is later than b: */
4593 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
4594 {
4595         return (s32)(a->si_generation - b->si_generation) > 0;
4596 }
4597
4598 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4599 {
4600         /*
4601          * When sessions are used the stateid generation number is ignored
4602          * when it is zero.
4603          */
4604         if (has_session && in->si_generation == 0)
4605                 return nfs_ok;
4606
4607         if (in->si_generation == ref->si_generation)
4608                 return nfs_ok;
4609
4610         /* If the client sends us a stateid from the future, it's buggy: */
4611         if (stateid_generation_after(in, ref))
4612                 return nfserr_bad_stateid;
4613         /*
4614          * However, we could see a stateid from the past, even from a
4615          * non-buggy client.  For example, if the client sends a lock
4616          * while some IO is outstanding, the lock may bump si_generation
4617          * while the IO is still in flight.  The client could avoid that
4618          * situation by waiting for responses on all the IO requests,
4619          * but better performance may result in retrying IO that
4620          * receives an old_stateid error if requests are rarely
4621          * reordered in flight:
4622          */
4623         return nfserr_old_stateid;
4624 }
4625
4626 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4627 {
4628         if (ols->st_stateowner->so_is_open_owner &&
4629             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4630                 return nfserr_bad_stateid;
4631         return nfs_ok;
4632 }
4633
4634 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4635 {
4636         struct nfs4_stid *s;
4637         __be32 status = nfserr_bad_stateid;
4638
4639         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4640                 return status;
4641         /* Client debugging aid. */
4642         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4643                 char addr_str[INET6_ADDRSTRLEN];
4644                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4645                                  sizeof(addr_str));
4646                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4647                                         "with incorrect client ID\n", addr_str);
4648                 return status;
4649         }
4650         spin_lock(&cl->cl_lock);
4651         s = find_stateid_locked(cl, stateid);
4652         if (!s)
4653                 goto out_unlock;
4654         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
4655         if (status)
4656                 goto out_unlock;
4657         switch (s->sc_type) {
4658         case NFS4_DELEG_STID:
4659                 status = nfs_ok;
4660                 break;
4661         case NFS4_REVOKED_DELEG_STID:
4662                 status = nfserr_deleg_revoked;
4663                 break;
4664         case NFS4_OPEN_STID:
4665         case NFS4_LOCK_STID:
4666                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4667                 break;
4668         default:
4669                 printk("unknown stateid type %x\n", s->sc_type);
4670                 /* Fallthrough */
4671         case NFS4_CLOSED_STID:
4672         case NFS4_CLOSED_DELEG_STID:
4673                 status = nfserr_bad_stateid;
4674         }
4675 out_unlock:
4676         spin_unlock(&cl->cl_lock);
4677         return status;
4678 }
4679
4680 __be32
4681 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4682                      stateid_t *stateid, unsigned char typemask,
4683                      struct nfs4_stid **s, struct nfsd_net *nn)
4684 {
4685         __be32 status;
4686
4687         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
4688                 return nfserr_bad_stateid;
4689         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
4690         if (status == nfserr_stale_clientid) {
4691                 if (cstate->session)
4692                         return nfserr_bad_stateid;
4693                 return nfserr_stale_stateid;
4694         }
4695         if (status)
4696                 return status;
4697         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
4698         if (!*s)
4699                 return nfserr_bad_stateid;
4700         return nfs_ok;
4701 }
4702
4703 static struct file *
4704 nfs4_find_file(struct nfs4_stid *s, int flags)
4705 {
4706         if (!s)
4707                 return NULL;
4708
4709         switch (s->sc_type) {
4710         case NFS4_DELEG_STID:
4711                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
4712                         return NULL;
4713                 return get_file(s->sc_file->fi_deleg_file);
4714         case NFS4_OPEN_STID:
4715         case NFS4_LOCK_STID:
4716                 if (flags & RD_STATE)
4717                         return find_readable_file(s->sc_file);
4718                 else
4719                         return find_writeable_file(s->sc_file);
4720                 break;
4721         }
4722
4723         return NULL;
4724 }
4725
4726 static __be32
4727 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
4728 {
4729         __be32 status;
4730
4731         status = nfsd4_check_openowner_confirmed(ols);
4732         if (status)
4733                 return status;
4734         return nfs4_check_openmode(ols, flags);
4735 }
4736
4737 static __be32
4738 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
4739                 struct file **filpp, bool *tmp_file, int flags)
4740 {
4741         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
4742         struct file *file;
4743         __be32 status;
4744
4745         file = nfs4_find_file(s, flags);
4746         if (file) {
4747                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
4748                                 acc | NFSD_MAY_OWNER_OVERRIDE);
4749                 if (status) {
4750                         fput(file);
4751                         return status;
4752                 }
4753
4754                 *filpp = file;
4755         } else {
4756                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
4757                 if (status)
4758                         return status;
4759
4760                 if (tmp_file)
4761                         *tmp_file = true;
4762         }
4763
4764         return 0;
4765 }
4766
4767 /*
4768  * Checks for stateid operations
4769  */
4770 __be32
4771 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
4772                 struct nfsd4_compound_state *cstate, stateid_t *stateid,
4773                 int flags, struct file **filpp, bool *tmp_file)
4774 {
4775         struct svc_fh *fhp = &cstate->current_fh;
4776         struct inode *ino = d_inode(fhp->fh_dentry);
4777         struct net *net = SVC_NET(rqstp);
4778         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4779         struct nfs4_stid *s = NULL;
4780         __be32 status;
4781
4782         if (filpp)
4783                 *filpp = NULL;
4784         if (tmp_file)
4785                 *tmp_file = false;
4786
4787         if (grace_disallows_io(net, ino))
4788                 return nfserr_grace;
4789
4790         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
4791                 status = check_special_stateids(net, fhp, stateid, flags);
4792                 goto done;
4793         }
4794
4795         status = nfsd4_lookup_stateid(cstate, stateid,
4796                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
4797                                 &s, nn);
4798         if (status)
4799                 return status;
4800         status = check_stateid_generation(stateid, &s->sc_stateid,
4801                         nfsd4_has_session(cstate));
4802         if (status)
4803                 goto out;
4804
4805         switch (s->sc_type) {
4806         case NFS4_DELEG_STID:
4807                 status = nfs4_check_delegmode(delegstateid(s), flags);
4808                 break;
4809         case NFS4_OPEN_STID:
4810         case NFS4_LOCK_STID:
4811                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
4812                 break;
4813         default:
4814                 status = nfserr_bad_stateid;
4815                 break;
4816         }
4817         if (status)
4818                 goto out;
4819         status = nfs4_check_fh(fhp, s);
4820
4821 done:
4822         if (!status && filpp)
4823                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
4824 out:
4825         if (s)
4826                 nfs4_put_stid(s);
4827         return status;
4828 }
4829
4830 /*
4831  * Test if the stateid is valid
4832  */
4833 __be32
4834 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4835                    struct nfsd4_test_stateid *test_stateid)
4836 {
4837         struct nfsd4_test_stateid_id *stateid;
4838         struct nfs4_client *cl = cstate->session->se_client;
4839
4840         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
4841                 stateid->ts_id_status =
4842                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
4843
4844         return nfs_ok;
4845 }
4846
4847 __be32
4848 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4849                    struct nfsd4_free_stateid *free_stateid)
4850 {
4851         stateid_t *stateid = &free_stateid->fr_stateid;
4852         struct nfs4_stid *s;
4853         struct nfs4_delegation *dp;
4854         struct nfs4_ol_stateid *stp;
4855         struct nfs4_client *cl = cstate->session->se_client;
4856         __be32 ret = nfserr_bad_stateid;
4857
4858         spin_lock(&cl->cl_lock);
4859         s = find_stateid_locked(cl, stateid);
4860         if (!s)
4861                 goto out_unlock;
4862         switch (s->sc_type) {
4863         case NFS4_DELEG_STID:
4864                 ret = nfserr_locks_held;
4865                 break;
4866         case NFS4_OPEN_STID:
4867                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4868                 if (ret)
4869                         break;
4870                 ret = nfserr_locks_held;
4871                 break;
4872         case NFS4_LOCK_STID:
4873                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
4874                 if (ret)
4875                         break;
4876                 stp = openlockstateid(s);
4877                 ret = nfserr_locks_held;
4878                 if (check_for_locks(stp->st_stid.sc_file,
4879                                     lockowner(stp->st_stateowner)))
4880                         break;
4881                 WARN_ON(!unhash_lock_stateid(stp));
4882                 spin_unlock(&cl->cl_lock);
4883                 nfs4_put_stid(s);
4884                 ret = nfs_ok;
4885                 goto out;
4886         case NFS4_REVOKED_DELEG_STID:
4887                 dp = delegstateid(s);
4888                 list_del_init(&dp->dl_recall_lru);
4889                 spin_unlock(&cl->cl_lock);
4890                 nfs4_put_stid(s);
4891                 ret = nfs_ok;
4892                 goto out;
4893         /* Default falls through and returns nfserr_bad_stateid */
4894         }
4895 out_unlock:
4896         spin_unlock(&cl->cl_lock);
4897 out:
4898         return ret;
4899 }
4900
4901 static inline int
4902 setlkflg (int type)
4903 {
4904         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
4905                 RD_STATE : WR_STATE;
4906 }
4907
4908 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
4909 {
4910         struct svc_fh *current_fh = &cstate->current_fh;
4911         struct nfs4_stateowner *sop = stp->st_stateowner;
4912         __be32 status;
4913
4914         status = nfsd4_check_seqid(cstate, sop, seqid);
4915         if (status)
4916                 return status;
4917         if (stp->st_stid.sc_type == NFS4_CLOSED_STID
4918                 || stp->st_stid.sc_type == NFS4_REVOKED_DELEG_STID)
4919                 /*
4920                  * "Closed" stateid's exist *only* to return
4921                  * nfserr_replay_me from the previous step, and
4922                  * revoked delegations are kept only for free_stateid.
4923                  */
4924                 return nfserr_bad_stateid;
4925         down_write(&stp->st_rwsem);
4926         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
4927         if (status == nfs_ok)
4928                 status = nfs4_check_fh(current_fh, &stp->st_stid);
4929         if (status != nfs_ok)
4930                 up_write(&stp->st_rwsem);
4931         return status;
4932 }
4933
4934 /* 
4935  * Checks for sequence id mutating operations. 
4936  */
4937 static __be32
4938 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4939                          stateid_t *stateid, char typemask,
4940                          struct nfs4_ol_stateid **stpp,
4941                          struct nfsd_net *nn)
4942 {
4943         __be32 status;
4944         struct nfs4_stid *s;
4945         struct nfs4_ol_stateid *stp = NULL;
4946
4947         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
4948                 seqid, STATEID_VAL(stateid));
4949
4950         *stpp = NULL;
4951         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
4952         if (status)
4953                 return status;
4954         stp = openlockstateid(s);
4955         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
4956
4957         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
4958         if (!status)
4959                 *stpp = stp;
4960         else
4961                 nfs4_put_stid(&stp->st_stid);
4962         return status;
4963 }
4964
4965 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
4966                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
4967 {
4968         __be32 status;
4969         struct nfs4_openowner *oo;
4970         struct nfs4_ol_stateid *stp;
4971
4972         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
4973                                                 NFS4_OPEN_STID, &stp, nn);
4974         if (status)
4975                 return status;
4976         oo = openowner(stp->st_stateowner);
4977         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4978                 up_write(&stp->st_rwsem);
4979                 nfs4_put_stid(&stp->st_stid);
4980                 return nfserr_bad_stateid;
4981         }
4982         *stpp = stp;
4983         return nfs_ok;
4984 }
4985
4986 __be32
4987 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4988                    struct nfsd4_open_confirm *oc)
4989 {
4990         __be32 status;
4991         struct nfs4_openowner *oo;
4992         struct nfs4_ol_stateid *stp;
4993         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4994
4995         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
4996                         cstate->current_fh.fh_dentry);
4997
4998         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
4999         if (status)
5000                 return status;
5001
5002         status = nfs4_preprocess_seqid_op(cstate,
5003                                         oc->oc_seqid, &oc->oc_req_stateid,
5004                                         NFS4_OPEN_STID, &stp, nn);
5005         if (status)
5006                 goto out;
5007         oo = openowner(stp->st_stateowner);
5008         status = nfserr_bad_stateid;
5009         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5010                 up_write(&stp->st_rwsem);
5011                 goto put_stateid;
5012         }
5013         oo->oo_flags |= NFS4_OO_CONFIRMED;
5014         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5015         up_write(&stp->st_rwsem);
5016         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5017                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5018
5019         nfsd4_client_record_create(oo->oo_owner.so_client);
5020         status = nfs_ok;
5021 put_stateid:
5022         nfs4_put_stid(&stp->st_stid);
5023 out:
5024         nfsd4_bump_seqid(cstate, status);
5025         return status;
5026 }
5027
5028 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5029 {
5030         if (!test_access(access, stp))
5031                 return;
5032         nfs4_file_put_access(stp->st_stid.sc_file, access);
5033         clear_access(access, stp);
5034 }
5035
5036 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5037 {
5038         switch (to_access) {
5039         case NFS4_SHARE_ACCESS_READ:
5040                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5041                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5042                 break;
5043         case NFS4_SHARE_ACCESS_WRITE:
5044                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5045                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5046                 break;
5047         case NFS4_SHARE_ACCESS_BOTH:
5048                 break;
5049         default:
5050                 WARN_ON_ONCE(1);
5051         }
5052 }
5053
5054 __be32
5055 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5056                      struct nfsd4_compound_state *cstate,
5057                      struct nfsd4_open_downgrade *od)
5058 {
5059         __be32 status;
5060         struct nfs4_ol_stateid *stp;
5061         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5062
5063         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n", 
5064                         cstate->current_fh.fh_dentry);
5065
5066         /* We don't yet support WANT bits: */
5067         if (od->od_deleg_want)
5068                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5069                         od->od_deleg_want);
5070
5071         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5072                                         &od->od_stateid, &stp, nn);
5073         if (status)
5074                 goto out; 
5075         status = nfserr_inval;
5076         if (!test_access(od->od_share_access, stp)) {
5077                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5078                         stp->st_access_bmap, od->od_share_access);
5079                 goto put_stateid;
5080         }
5081         if (!test_deny(od->od_share_deny, stp)) {
5082                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5083                         stp->st_deny_bmap, od->od_share_deny);
5084                 goto put_stateid;
5085         }
5086         nfs4_stateid_downgrade(stp, od->od_share_access);
5087         reset_union_bmap_deny(od->od_share_deny, stp);
5088         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5089         status = nfs_ok;
5090 put_stateid:
5091         up_write(&stp->st_rwsem);
5092         nfs4_put_stid(&stp->st_stid);
5093 out:
5094         nfsd4_bump_seqid(cstate, status);
5095         return status;
5096 }
5097
5098 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5099 {
5100         struct nfs4_client *clp = s->st_stid.sc_client;
5101         bool unhashed;
5102         LIST_HEAD(reaplist);
5103
5104         s->st_stid.sc_type = NFS4_CLOSED_STID;
5105         spin_lock(&clp->cl_lock);
5106         unhashed = unhash_open_stateid(s, &reaplist);
5107
5108         if (clp->cl_minorversion) {
5109                 if (unhashed)
5110                         put_ol_stateid_locked(s, &reaplist);
5111                 spin_unlock(&clp->cl_lock);
5112                 free_ol_stateid_reaplist(&reaplist);
5113         } else {
5114                 spin_unlock(&clp->cl_lock);
5115                 free_ol_stateid_reaplist(&reaplist);
5116                 if (unhashed)
5117                         move_to_close_lru(s, clp->net);
5118         }
5119 }
5120
5121 /*
5122  * nfs4_unlock_state() called after encode
5123  */
5124 __be32
5125 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5126             struct nfsd4_close *close)
5127 {
5128         __be32 status;
5129         struct nfs4_ol_stateid *stp;
5130         struct net *net = SVC_NET(rqstp);
5131         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5132
5133         dprintk("NFSD: nfsd4_close on file %pd\n", 
5134                         cstate->current_fh.fh_dentry);
5135
5136         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5137                                         &close->cl_stateid,
5138                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5139                                         &stp, nn);
5140         nfsd4_bump_seqid(cstate, status);
5141         if (status)
5142                 goto out; 
5143         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5144         up_write(&stp->st_rwsem);
5145
5146         nfsd4_close_open_stateid(stp);
5147
5148         /* put reference from nfs4_preprocess_seqid_op */
5149         nfs4_put_stid(&stp->st_stid);
5150 out:
5151         return status;
5152 }
5153
5154 __be32
5155 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5156                   struct nfsd4_delegreturn *dr)
5157 {
5158         struct nfs4_delegation *dp;
5159         stateid_t *stateid = &dr->dr_stateid;
5160         struct nfs4_stid *s;
5161         __be32 status;
5162         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5163
5164         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5165                 return status;
5166
5167         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5168         if (status)
5169                 goto out;
5170         dp = delegstateid(s);
5171         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
5172         if (status)
5173                 goto put_stateid;
5174
5175         destroy_delegation(dp);
5176 put_stateid:
5177         nfs4_put_stid(&dp->dl_stid);
5178 out:
5179         return status;
5180 }
5181
5182 static inline u64
5183 end_offset(u64 start, u64 len)
5184 {
5185         u64 end;
5186
5187         end = start + len;
5188         return end >= start ? end: NFS4_MAX_UINT64;
5189 }
5190
5191 /* last octet in a range */
5192 static inline u64
5193 last_byte_offset(u64 start, u64 len)
5194 {
5195         u64 end;
5196
5197         WARN_ON_ONCE(!len);
5198         end = start + len;
5199         return end > start ? end - 1: NFS4_MAX_UINT64;
5200 }
5201
5202 /*
5203  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5204  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5205  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5206  * locking, this prevents us from being completely protocol-compliant.  The
5207  * real solution to this problem is to start using unsigned file offsets in
5208  * the VFS, but this is a very deep change!
5209  */
5210 static inline void
5211 nfs4_transform_lock_offset(struct file_lock *lock)
5212 {
5213         if (lock->fl_start < 0)
5214                 lock->fl_start = OFFSET_MAX;
5215         if (lock->fl_end < 0)
5216                 lock->fl_end = OFFSET_MAX;
5217 }
5218
5219 static fl_owner_t
5220 nfsd4_fl_get_owner(fl_owner_t owner)
5221 {
5222         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5223
5224         nfs4_get_stateowner(&lo->lo_owner);
5225         return owner;
5226 }
5227
5228 static void
5229 nfsd4_fl_put_owner(fl_owner_t owner)
5230 {
5231         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5232
5233         if (lo)
5234                 nfs4_put_stateowner(&lo->lo_owner);
5235 }
5236
5237 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5238         .lm_get_owner = nfsd4_fl_get_owner,
5239         .lm_put_owner = nfsd4_fl_put_owner,
5240 };
5241
5242 static inline void
5243 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5244 {
5245         struct nfs4_lockowner *lo;
5246
5247         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5248                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5249                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5250                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5251                 if (!deny->ld_owner.data)
5252                         /* We just don't care that much */
5253                         goto nevermind;
5254                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5255                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5256         } else {
5257 nevermind:
5258                 deny->ld_owner.len = 0;
5259                 deny->ld_owner.data = NULL;
5260                 deny->ld_clientid.cl_boot = 0;
5261                 deny->ld_clientid.cl_id = 0;
5262         }
5263         deny->ld_start = fl->fl_start;
5264         deny->ld_length = NFS4_MAX_UINT64;
5265         if (fl->fl_end != NFS4_MAX_UINT64)
5266                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
5267         deny->ld_type = NFS4_READ_LT;
5268         if (fl->fl_type != F_RDLCK)
5269                 deny->ld_type = NFS4_WRITE_LT;
5270 }
5271
5272 static struct nfs4_lockowner *
5273 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5274 {
5275         unsigned int strhashval = ownerstr_hashval(owner);
5276         struct nfs4_stateowner *so;
5277
5278         lockdep_assert_held(&clp->cl_lock);
5279
5280         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5281                             so_strhash) {
5282                 if (so->so_is_open_owner)
5283                         continue;
5284                 if (same_owner_str(so, owner))
5285                         return lockowner(nfs4_get_stateowner(so));
5286         }
5287         return NULL;
5288 }
5289
5290 static struct nfs4_lockowner *
5291 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5292 {
5293         struct nfs4_lockowner *lo;
5294
5295         spin_lock(&clp->cl_lock);
5296         lo = find_lockowner_str_locked(clp, owner);
5297         spin_unlock(&clp->cl_lock);
5298         return lo;
5299 }
5300
5301 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5302 {
5303         unhash_lockowner_locked(lockowner(sop));
5304 }
5305
5306 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5307 {
5308         struct nfs4_lockowner *lo = lockowner(sop);
5309
5310         kmem_cache_free(lockowner_slab, lo);
5311 }
5312
5313 static const struct nfs4_stateowner_operations lockowner_ops = {
5314         .so_unhash =    nfs4_unhash_lockowner,
5315         .so_free =      nfs4_free_lockowner,
5316 };
5317
5318 /*
5319  * Alloc a lock owner structure.
5320  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
5321  * occurred. 
5322  *
5323  * strhashval = ownerstr_hashval
5324  */
5325 static struct nfs4_lockowner *
5326 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5327                            struct nfs4_ol_stateid *open_stp,
5328                            struct nfsd4_lock *lock)
5329 {
5330         struct nfs4_lockowner *lo, *ret;
5331
5332         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5333         if (!lo)
5334                 return NULL;
5335         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5336         lo->lo_owner.so_is_open_owner = 0;
5337         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5338         lo->lo_owner.so_ops = &lockowner_ops;
5339         spin_lock(&clp->cl_lock);
5340         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5341         if (ret == NULL) {
5342                 list_add(&lo->lo_owner.so_strhash,
5343                          &clp->cl_ownerstr_hashtbl[strhashval]);
5344                 ret = lo;
5345         } else
5346                 nfs4_free_stateowner(&lo->lo_owner);
5347
5348         spin_unlock(&clp->cl_lock);
5349         return ret;
5350 }
5351
5352 static void
5353 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5354                   struct nfs4_file *fp, struct inode *inode,
5355                   struct nfs4_ol_stateid *open_stp)
5356 {
5357         struct nfs4_client *clp = lo->lo_owner.so_client;
5358
5359         lockdep_assert_held(&clp->cl_lock);
5360
5361         atomic_inc(&stp->st_stid.sc_count);
5362         stp->st_stid.sc_type = NFS4_LOCK_STID;
5363         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5364         get_nfs4_file(fp);
5365         stp->st_stid.sc_file = fp;
5366         stp->st_stid.sc_free = nfs4_free_lock_stateid;
5367         stp->st_access_bmap = 0;
5368         stp->st_deny_bmap = open_stp->st_deny_bmap;
5369         stp->st_openstp = open_stp;
5370         init_rwsem(&stp->st_rwsem);
5371         list_add(&stp->st_locks, &open_stp->st_locks);
5372         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5373         spin_lock(&fp->fi_lock);
5374         list_add(&stp->st_perfile, &fp->fi_stateids);
5375         spin_unlock(&fp->fi_lock);
5376 }
5377
5378 static struct nfs4_ol_stateid *
5379 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5380 {
5381         struct nfs4_ol_stateid *lst;
5382         struct nfs4_client *clp = lo->lo_owner.so_client;
5383
5384         lockdep_assert_held(&clp->cl_lock);
5385
5386         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5387                 if (lst->st_stid.sc_file == fp) {
5388                         atomic_inc(&lst->st_stid.sc_count);
5389                         return lst;
5390                 }
5391         }
5392         return NULL;
5393 }
5394
5395 static struct nfs4_ol_stateid *
5396 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5397                             struct inode *inode, struct nfs4_ol_stateid *ost,
5398                             bool *new)
5399 {
5400         struct nfs4_stid *ns = NULL;
5401         struct nfs4_ol_stateid *lst;
5402         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5403         struct nfs4_client *clp = oo->oo_owner.so_client;
5404
5405         spin_lock(&clp->cl_lock);
5406         lst = find_lock_stateid(lo, fi);
5407         if (lst == NULL) {
5408                 spin_unlock(&clp->cl_lock);
5409                 ns = nfs4_alloc_stid(clp, stateid_slab);
5410                 if (ns == NULL)
5411                         return NULL;
5412
5413                 spin_lock(&clp->cl_lock);
5414                 lst = find_lock_stateid(lo, fi);
5415                 if (likely(!lst)) {
5416                         lst = openlockstateid(ns);
5417                         init_lock_stateid(lst, lo, fi, inode, ost);
5418                         ns = NULL;
5419                         *new = true;
5420                 }
5421         }
5422         spin_unlock(&clp->cl_lock);
5423         if (ns)
5424                 nfs4_put_stid(ns);
5425         return lst;
5426 }
5427
5428 static int
5429 check_lock_length(u64 offset, u64 length)
5430 {
5431         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5432                 (length > ~offset)));
5433 }
5434
5435 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5436 {
5437         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5438
5439         lockdep_assert_held(&fp->fi_lock);
5440
5441         if (test_access(access, lock_stp))
5442                 return;
5443         __nfs4_file_get_access(fp, access);
5444         set_access(access, lock_stp);
5445 }
5446
5447 static __be32
5448 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5449                             struct nfs4_ol_stateid *ost,
5450                             struct nfsd4_lock *lock,
5451                             struct nfs4_ol_stateid **lst, bool *new)
5452 {
5453         __be32 status;
5454         struct nfs4_file *fi = ost->st_stid.sc_file;
5455         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5456         struct nfs4_client *cl = oo->oo_owner.so_client;
5457         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5458         struct nfs4_lockowner *lo;
5459         unsigned int strhashval;
5460
5461         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5462         if (!lo) {
5463                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5464                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5465                 if (lo == NULL)
5466                         return nfserr_jukebox;
5467         } else {
5468                 /* with an existing lockowner, seqids must be the same */
5469                 status = nfserr_bad_seqid;
5470                 if (!cstate->minorversion &&
5471                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5472                         goto out;
5473         }
5474
5475         *lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5476         if (*lst == NULL) {
5477                 status = nfserr_jukebox;
5478                 goto out;
5479         }
5480         status = nfs_ok;
5481 out:
5482         nfs4_put_stateowner(&lo->lo_owner);
5483         return status;
5484 }
5485
5486 /*
5487  *  LOCK operation 
5488  */
5489 __be32
5490 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5491            struct nfsd4_lock *lock)
5492 {
5493         struct nfs4_openowner *open_sop = NULL;
5494         struct nfs4_lockowner *lock_sop = NULL;
5495         struct nfs4_ol_stateid *lock_stp = NULL;
5496         struct nfs4_ol_stateid *open_stp = NULL;
5497         struct nfs4_file *fp;
5498         struct file *filp = NULL;
5499         struct file_lock *file_lock = NULL;
5500         struct file_lock *conflock = NULL;
5501         __be32 status = 0;
5502         int lkflg;
5503         int err;
5504         bool new = false;
5505         struct net *net = SVC_NET(rqstp);
5506         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5507
5508         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5509                 (long long) lock->lk_offset,
5510                 (long long) lock->lk_length);
5511
5512         if (check_lock_length(lock->lk_offset, lock->lk_length))
5513                  return nfserr_inval;
5514
5515         if ((status = fh_verify(rqstp, &cstate->current_fh,
5516                                 S_IFREG, NFSD_MAY_LOCK))) {
5517                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5518                 return status;
5519         }
5520
5521         if (lock->lk_is_new) {
5522                 if (nfsd4_has_session(cstate))
5523                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5524                         memcpy(&lock->lk_new_clientid,
5525                                 &cstate->session->se_client->cl_clientid,
5526                                 sizeof(clientid_t));
5527
5528                 status = nfserr_stale_clientid;
5529                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5530                         goto out;
5531
5532                 /* validate and update open stateid and open seqid */
5533                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5534                                         lock->lk_new_open_seqid,
5535                                         &lock->lk_new_open_stateid,
5536                                         &open_stp, nn);
5537                 if (status)
5538                         goto out;
5539                 up_write(&open_stp->st_rwsem);
5540                 open_sop = openowner(open_stp->st_stateowner);
5541                 status = nfserr_bad_stateid;
5542                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5543                                                 &lock->lk_new_clientid))
5544                         goto out;
5545                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5546                                                         &lock_stp, &new);
5547                 if (status == nfs_ok)
5548                         down_write(&lock_stp->st_rwsem);
5549         } else {
5550                 status = nfs4_preprocess_seqid_op(cstate,
5551                                        lock->lk_old_lock_seqid,
5552                                        &lock->lk_old_lock_stateid,
5553                                        NFS4_LOCK_STID, &lock_stp, nn);
5554         }
5555         if (status)
5556                 goto out;
5557         lock_sop = lockowner(lock_stp->st_stateowner);
5558
5559         lkflg = setlkflg(lock->lk_type);
5560         status = nfs4_check_openmode(lock_stp, lkflg);
5561         if (status)
5562                 goto out;
5563
5564         status = nfserr_grace;
5565         if (locks_in_grace(net) && !lock->lk_reclaim)
5566                 goto out;
5567         status = nfserr_no_grace;
5568         if (!locks_in_grace(net) && lock->lk_reclaim)
5569                 goto out;
5570
5571         file_lock = locks_alloc_lock();
5572         if (!file_lock) {
5573                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5574                 status = nfserr_jukebox;
5575                 goto out;
5576         }
5577
5578         fp = lock_stp->st_stid.sc_file;
5579         switch (lock->lk_type) {
5580                 case NFS4_READ_LT:
5581                 case NFS4_READW_LT:
5582                         spin_lock(&fp->fi_lock);
5583                         filp = find_readable_file_locked(fp);
5584                         if (filp)
5585                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5586                         spin_unlock(&fp->fi_lock);
5587                         file_lock->fl_type = F_RDLCK;
5588                         break;
5589                 case NFS4_WRITE_LT:
5590                 case NFS4_WRITEW_LT:
5591                         spin_lock(&fp->fi_lock);
5592                         filp = find_writeable_file_locked(fp);
5593                         if (filp)
5594                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5595                         spin_unlock(&fp->fi_lock);
5596                         file_lock->fl_type = F_WRLCK;
5597                         break;
5598                 default:
5599                         status = nfserr_inval;
5600                 goto out;
5601         }
5602         if (!filp) {
5603                 status = nfserr_openmode;
5604                 goto out;
5605         }
5606
5607         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
5608         file_lock->fl_pid = current->tgid;
5609         file_lock->fl_file = filp;
5610         file_lock->fl_flags = FL_POSIX;
5611         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5612         file_lock->fl_start = lock->lk_offset;
5613         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
5614         nfs4_transform_lock_offset(file_lock);
5615
5616         conflock = locks_alloc_lock();
5617         if (!conflock) {
5618                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5619                 status = nfserr_jukebox;
5620                 goto out;
5621         }
5622
5623         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
5624         switch (-err) {
5625         case 0: /* success! */
5626                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
5627                 status = 0;
5628                 break;
5629         case (EAGAIN):          /* conflock holds conflicting lock */
5630                 status = nfserr_denied;
5631                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
5632                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
5633                 break;
5634         case (EDEADLK):
5635                 status = nfserr_deadlock;
5636                 break;
5637         default:
5638                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
5639                 status = nfserrno(err);
5640                 break;
5641         }
5642 out:
5643         if (filp)
5644                 fput(filp);
5645         if (lock_stp) {
5646                 /* Bump seqid manually if the 4.0 replay owner is openowner */
5647                 if (cstate->replay_owner &&
5648                     cstate->replay_owner != &lock_sop->lo_owner &&
5649                     seqid_mutating_err(ntohl(status)))
5650                         lock_sop->lo_owner.so_seqid++;
5651
5652                 up_write(&lock_stp->st_rwsem);
5653
5654                 /*
5655                  * If this is a new, never-before-used stateid, and we are
5656                  * returning an error, then just go ahead and release it.
5657                  */
5658                 if (status && new)
5659                         release_lock_stateid(lock_stp);
5660
5661                 nfs4_put_stid(&lock_stp->st_stid);
5662         }
5663         if (open_stp)
5664                 nfs4_put_stid(&open_stp->st_stid);
5665         nfsd4_bump_seqid(cstate, status);
5666         if (file_lock)
5667                 locks_free_lock(file_lock);
5668         if (conflock)
5669                 locks_free_lock(conflock);
5670         return status;
5671 }
5672
5673 /*
5674  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
5675  * so we do a temporary open here just to get an open file to pass to
5676  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
5677  * inode operation.)
5678  */
5679 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
5680 {
5681         struct file *file;
5682         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
5683         if (!err) {
5684                 err = nfserrno(vfs_test_lock(file, lock));
5685                 fput(file);
5686         }
5687         return err;
5688 }
5689
5690 /*
5691  * LOCKT operation
5692  */
5693 __be32
5694 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5695             struct nfsd4_lockt *lockt)
5696 {
5697         struct file_lock *file_lock = NULL;
5698         struct nfs4_lockowner *lo = NULL;
5699         __be32 status;
5700         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5701
5702         if (locks_in_grace(SVC_NET(rqstp)))
5703                 return nfserr_grace;
5704
5705         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
5706                  return nfserr_inval;
5707
5708         if (!nfsd4_has_session(cstate)) {
5709                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
5710                 if (status)
5711                         goto out;
5712         }
5713
5714         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5715                 goto out;
5716
5717         file_lock = locks_alloc_lock();
5718         if (!file_lock) {
5719                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5720                 status = nfserr_jukebox;
5721                 goto out;
5722         }
5723
5724         switch (lockt->lt_type) {
5725                 case NFS4_READ_LT:
5726                 case NFS4_READW_LT:
5727                         file_lock->fl_type = F_RDLCK;
5728                 break;
5729                 case NFS4_WRITE_LT:
5730                 case NFS4_WRITEW_LT:
5731                         file_lock->fl_type = F_WRLCK;
5732                 break;
5733                 default:
5734                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
5735                         status = nfserr_inval;
5736                 goto out;
5737         }
5738
5739         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
5740         if (lo)
5741                 file_lock->fl_owner = (fl_owner_t)lo;
5742         file_lock->fl_pid = current->tgid;
5743         file_lock->fl_flags = FL_POSIX;
5744
5745         file_lock->fl_start = lockt->lt_offset;
5746         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
5747
5748         nfs4_transform_lock_offset(file_lock);
5749
5750         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
5751         if (status)
5752                 goto out;
5753
5754         if (file_lock->fl_type != F_UNLCK) {
5755                 status = nfserr_denied;
5756                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
5757         }
5758 out:
5759         if (lo)
5760                 nfs4_put_stateowner(&lo->lo_owner);
5761         if (file_lock)
5762                 locks_free_lock(file_lock);
5763         return status;
5764 }
5765
5766 __be32
5767 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5768             struct nfsd4_locku *locku)
5769 {
5770         struct nfs4_ol_stateid *stp;
5771         struct file *filp = NULL;
5772         struct file_lock *file_lock = NULL;
5773         __be32 status;
5774         int err;
5775         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5776
5777         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
5778                 (long long) locku->lu_offset,
5779                 (long long) locku->lu_length);
5780
5781         if (check_lock_length(locku->lu_offset, locku->lu_length))
5782                  return nfserr_inval;
5783
5784         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
5785                                         &locku->lu_stateid, NFS4_LOCK_STID,
5786                                         &stp, nn);
5787         if (status)
5788                 goto out;
5789         filp = find_any_file(stp->st_stid.sc_file);
5790         if (!filp) {
5791                 status = nfserr_lock_range;
5792                 goto put_stateid;
5793         }
5794         file_lock = locks_alloc_lock();
5795         if (!file_lock) {
5796                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
5797                 status = nfserr_jukebox;
5798                 goto fput;
5799         }
5800
5801         file_lock->fl_type = F_UNLCK;
5802         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
5803         file_lock->fl_pid = current->tgid;
5804         file_lock->fl_file = filp;
5805         file_lock->fl_flags = FL_POSIX;
5806         file_lock->fl_lmops = &nfsd_posix_mng_ops;
5807         file_lock->fl_start = locku->lu_offset;
5808
5809         file_lock->fl_end = last_byte_offset(locku->lu_offset,
5810                                                 locku->lu_length);
5811         nfs4_transform_lock_offset(file_lock);
5812
5813         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
5814         if (err) {
5815                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
5816                 goto out_nfserr;
5817         }
5818         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
5819 fput:
5820         fput(filp);
5821 put_stateid:
5822         up_write(&stp->st_rwsem);
5823         nfs4_put_stid(&stp->st_stid);
5824 out:
5825         nfsd4_bump_seqid(cstate, status);
5826         if (file_lock)
5827                 locks_free_lock(file_lock);
5828         return status;
5829
5830 out_nfserr:
5831         status = nfserrno(err);
5832         goto fput;
5833 }
5834
5835 /*
5836  * returns
5837  *      true:  locks held by lockowner
5838  *      false: no locks held by lockowner
5839  */
5840 static bool
5841 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
5842 {
5843         struct file_lock *fl;
5844         int status = false;
5845         struct file *filp = find_any_file(fp);
5846         struct inode *inode;
5847         struct file_lock_context *flctx;
5848
5849         if (!filp) {
5850                 /* Any valid lock stateid should have some sort of access */
5851                 WARN_ON_ONCE(1);
5852                 return status;
5853         }
5854
5855         inode = file_inode(filp);
5856         flctx = inode->i_flctx;
5857
5858         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
5859                 spin_lock(&flctx->flc_lock);
5860                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
5861                         if (fl->fl_owner == (fl_owner_t)lowner) {
5862                                 status = true;
5863                                 break;
5864                         }
5865                 }
5866                 spin_unlock(&flctx->flc_lock);
5867         }
5868         fput(filp);
5869         return status;
5870 }
5871
5872 __be32
5873 nfsd4_release_lockowner(struct svc_rqst *rqstp,
5874                         struct nfsd4_compound_state *cstate,
5875                         struct nfsd4_release_lockowner *rlockowner)
5876 {
5877         clientid_t *clid = &rlockowner->rl_clientid;
5878         struct nfs4_stateowner *sop;
5879         struct nfs4_lockowner *lo = NULL;
5880         struct nfs4_ol_stateid *stp;
5881         struct xdr_netobj *owner = &rlockowner->rl_owner;
5882         unsigned int hashval = ownerstr_hashval(owner);
5883         __be32 status;
5884         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5885         struct nfs4_client *clp;
5886
5887         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
5888                 clid->cl_boot, clid->cl_id);
5889
5890         status = lookup_clientid(clid, cstate, nn);
5891         if (status)
5892                 return status;
5893
5894         clp = cstate->clp;
5895         /* Find the matching lock stateowner */
5896         spin_lock(&clp->cl_lock);
5897         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
5898                             so_strhash) {
5899
5900                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
5901                         continue;
5902
5903                 /* see if there are still any locks associated with it */
5904                 lo = lockowner(sop);
5905                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
5906                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
5907                                 status = nfserr_locks_held;
5908                                 spin_unlock(&clp->cl_lock);
5909                                 return status;
5910                         }
5911                 }
5912
5913                 nfs4_get_stateowner(sop);
5914                 break;
5915         }
5916         spin_unlock(&clp->cl_lock);
5917         if (lo)
5918                 release_lockowner(lo);
5919         return status;
5920 }
5921
5922 static inline struct nfs4_client_reclaim *
5923 alloc_reclaim(void)
5924 {
5925         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
5926 }
5927
5928 bool
5929 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
5930 {
5931         struct nfs4_client_reclaim *crp;
5932
5933         crp = nfsd4_find_reclaim_client(name, nn);
5934         return (crp && crp->cr_clp);
5935 }
5936
5937 /*
5938  * failure => all reset bets are off, nfserr_no_grace...
5939  */
5940 struct nfs4_client_reclaim *
5941 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
5942 {
5943         unsigned int strhashval;
5944         struct nfs4_client_reclaim *crp;
5945
5946         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
5947         crp = alloc_reclaim();
5948         if (crp) {
5949                 strhashval = clientstr_hashval(name);
5950                 INIT_LIST_HEAD(&crp->cr_strhash);
5951                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
5952                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
5953                 crp->cr_clp = NULL;
5954                 nn->reclaim_str_hashtbl_size++;
5955         }
5956         return crp;
5957 }
5958
5959 void
5960 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
5961 {
5962         list_del(&crp->cr_strhash);
5963         kfree(crp);
5964         nn->reclaim_str_hashtbl_size--;
5965 }
5966
5967 void
5968 nfs4_release_reclaim(struct nfsd_net *nn)
5969 {
5970         struct nfs4_client_reclaim *crp = NULL;
5971         int i;
5972
5973         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
5974                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
5975                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
5976                                         struct nfs4_client_reclaim, cr_strhash);
5977                         nfs4_remove_reclaim_record(crp, nn);
5978                 }
5979         }
5980         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
5981 }
5982
5983 /*
5984  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
5985 struct nfs4_client_reclaim *
5986 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
5987 {
5988         unsigned int strhashval;
5989         struct nfs4_client_reclaim *crp = NULL;
5990
5991         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
5992
5993         strhashval = clientstr_hashval(recdir);
5994         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
5995                 if (same_name(crp->cr_recdir, recdir)) {
5996                         return crp;
5997                 }
5998         }
5999         return NULL;
6000 }
6001
6002 /*
6003 * Called from OPEN. Look for clientid in reclaim list.
6004 */
6005 __be32
6006 nfs4_check_open_reclaim(clientid_t *clid,
6007                 struct nfsd4_compound_state *cstate,
6008                 struct nfsd_net *nn)
6009 {
6010         __be32 status;
6011
6012         /* find clientid in conf_id_hashtbl */
6013         status = lookup_clientid(clid, cstate, nn);
6014         if (status)
6015                 return nfserr_reclaim_bad;
6016
6017         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6018                 return nfserr_no_grace;
6019
6020         if (nfsd4_client_record_check(cstate->clp))
6021                 return nfserr_reclaim_bad;
6022
6023         return nfs_ok;
6024 }
6025
6026 #ifdef CONFIG_NFSD_FAULT_INJECTION
6027 static inline void
6028 put_client(struct nfs4_client *clp)
6029 {
6030         atomic_dec(&clp->cl_refcount);
6031 }
6032
6033 static struct nfs4_client *
6034 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6035 {
6036         struct nfs4_client *clp;
6037         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6038                                           nfsd_net_id);
6039
6040         if (!nfsd_netns_ready(nn))
6041                 return NULL;
6042
6043         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6044                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6045                         return clp;
6046         }
6047         return NULL;
6048 }
6049
6050 u64
6051 nfsd_inject_print_clients(void)
6052 {
6053         struct nfs4_client *clp;
6054         u64 count = 0;
6055         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6056                                           nfsd_net_id);
6057         char buf[INET6_ADDRSTRLEN];
6058
6059         if (!nfsd_netns_ready(nn))
6060                 return 0;
6061
6062         spin_lock(&nn->client_lock);
6063         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6064                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6065                 pr_info("NFS Client: %s\n", buf);
6066                 ++count;
6067         }
6068         spin_unlock(&nn->client_lock);
6069
6070         return count;
6071 }
6072
6073 u64
6074 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6075 {
6076         u64 count = 0;
6077         struct nfs4_client *clp;
6078         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6079                                           nfsd_net_id);
6080
6081         if (!nfsd_netns_ready(nn))
6082                 return count;
6083
6084         spin_lock(&nn->client_lock);
6085         clp = nfsd_find_client(addr, addr_size);
6086         if (clp) {
6087                 if (mark_client_expired_locked(clp) == nfs_ok)
6088                         ++count;
6089                 else
6090                         clp = NULL;
6091         }
6092         spin_unlock(&nn->client_lock);
6093
6094         if (clp)
6095                 expire_client(clp);
6096
6097         return count;
6098 }
6099
6100 u64
6101 nfsd_inject_forget_clients(u64 max)
6102 {
6103         u64 count = 0;
6104         struct nfs4_client *clp, *next;
6105         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6106                                                 nfsd_net_id);
6107         LIST_HEAD(reaplist);
6108
6109         if (!nfsd_netns_ready(nn))
6110                 return count;
6111
6112         spin_lock(&nn->client_lock);
6113         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6114                 if (mark_client_expired_locked(clp) == nfs_ok) {
6115                         list_add(&clp->cl_lru, &reaplist);
6116                         if (max != 0 && ++count >= max)
6117                                 break;
6118                 }
6119         }
6120         spin_unlock(&nn->client_lock);
6121
6122         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6123                 expire_client(clp);
6124
6125         return count;
6126 }
6127
6128 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6129                              const char *type)
6130 {
6131         char buf[INET6_ADDRSTRLEN];
6132         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6133         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6134 }
6135
6136 static void
6137 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6138                              struct list_head *collect)
6139 {
6140         struct nfs4_client *clp = lst->st_stid.sc_client;
6141         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6142                                           nfsd_net_id);
6143
6144         if (!collect)
6145                 return;
6146
6147         lockdep_assert_held(&nn->client_lock);
6148         atomic_inc(&clp->cl_refcount);
6149         list_add(&lst->st_locks, collect);
6150 }
6151
6152 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6153                                     struct list_head *collect,
6154                                     bool (*func)(struct nfs4_ol_stateid *))
6155 {
6156         struct nfs4_openowner *oop;
6157         struct nfs4_ol_stateid *stp, *st_next;
6158         struct nfs4_ol_stateid *lst, *lst_next;
6159         u64 count = 0;
6160
6161         spin_lock(&clp->cl_lock);
6162         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6163                 list_for_each_entry_safe(stp, st_next,
6164                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6165                         list_for_each_entry_safe(lst, lst_next,
6166                                         &stp->st_locks, st_locks) {
6167                                 if (func) {
6168                                         if (func(lst))
6169                                                 nfsd_inject_add_lock_to_list(lst,
6170                                                                         collect);
6171                                 }
6172                                 ++count;
6173                                 /*
6174                                  * Despite the fact that these functions deal
6175                                  * with 64-bit integers for "count", we must
6176                                  * ensure that it doesn't blow up the
6177                                  * clp->cl_refcount. Throw a warning if we
6178                                  * start to approach INT_MAX here.
6179                                  */
6180                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6181                                 if (count == max)
6182                                         goto out;
6183                         }
6184                 }
6185         }
6186 out:
6187         spin_unlock(&clp->cl_lock);
6188
6189         return count;
6190 }
6191
6192 static u64
6193 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6194                           u64 max)
6195 {
6196         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6197 }
6198
6199 static u64
6200 nfsd_print_client_locks(struct nfs4_client *clp)
6201 {
6202         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6203         nfsd_print_count(clp, count, "locked files");
6204         return count;
6205 }
6206
6207 u64
6208 nfsd_inject_print_locks(void)
6209 {
6210         struct nfs4_client *clp;
6211         u64 count = 0;
6212         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6213                                                 nfsd_net_id);
6214
6215         if (!nfsd_netns_ready(nn))
6216                 return 0;
6217
6218         spin_lock(&nn->client_lock);
6219         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6220                 count += nfsd_print_client_locks(clp);
6221         spin_unlock(&nn->client_lock);
6222
6223         return count;
6224 }
6225
6226 static void
6227 nfsd_reap_locks(struct list_head *reaplist)
6228 {
6229         struct nfs4_client *clp;
6230         struct nfs4_ol_stateid *stp, *next;
6231
6232         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6233                 list_del_init(&stp->st_locks);
6234                 clp = stp->st_stid.sc_client;
6235                 nfs4_put_stid(&stp->st_stid);
6236                 put_client(clp);
6237         }
6238 }
6239
6240 u64
6241 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6242 {
6243         unsigned int count = 0;
6244         struct nfs4_client *clp;
6245         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6246                                                 nfsd_net_id);
6247         LIST_HEAD(reaplist);
6248
6249         if (!nfsd_netns_ready(nn))
6250                 return count;
6251
6252         spin_lock(&nn->client_lock);
6253         clp = nfsd_find_client(addr, addr_size);
6254         if (clp)
6255                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6256         spin_unlock(&nn->client_lock);
6257         nfsd_reap_locks(&reaplist);
6258         return count;
6259 }
6260
6261 u64
6262 nfsd_inject_forget_locks(u64 max)
6263 {
6264         u64 count = 0;
6265         struct nfs4_client *clp;
6266         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6267                                                 nfsd_net_id);
6268         LIST_HEAD(reaplist);
6269
6270         if (!nfsd_netns_ready(nn))
6271                 return count;
6272
6273         spin_lock(&nn->client_lock);
6274         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6275                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6276                 if (max != 0 && count >= max)
6277                         break;
6278         }
6279         spin_unlock(&nn->client_lock);
6280         nfsd_reap_locks(&reaplist);
6281         return count;
6282 }
6283
6284 static u64
6285 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6286                               struct list_head *collect,
6287                               void (*func)(struct nfs4_openowner *))
6288 {
6289         struct nfs4_openowner *oop, *next;
6290         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6291                                                 nfsd_net_id);
6292         u64 count = 0;
6293
6294         lockdep_assert_held(&nn->client_lock);
6295
6296         spin_lock(&clp->cl_lock);
6297         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6298                 if (func) {
6299                         func(oop);
6300                         if (collect) {
6301                                 atomic_inc(&clp->cl_refcount);
6302                                 list_add(&oop->oo_perclient, collect);
6303                         }
6304                 }
6305                 ++count;
6306                 /*
6307                  * Despite the fact that these functions deal with
6308                  * 64-bit integers for "count", we must ensure that
6309                  * it doesn't blow up the clp->cl_refcount. Throw a
6310                  * warning if we start to approach INT_MAX here.
6311                  */
6312                 WARN_ON_ONCE(count == (INT_MAX / 2));
6313                 if (count == max)
6314                         break;
6315         }
6316         spin_unlock(&clp->cl_lock);
6317
6318         return count;
6319 }
6320
6321 static u64
6322 nfsd_print_client_openowners(struct nfs4_client *clp)
6323 {
6324         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6325
6326         nfsd_print_count(clp, count, "openowners");
6327         return count;
6328 }
6329
6330 static u64
6331 nfsd_collect_client_openowners(struct nfs4_client *clp,
6332                                struct list_head *collect, u64 max)
6333 {
6334         return nfsd_foreach_client_openowner(clp, max, collect,
6335                                                 unhash_openowner_locked);
6336 }
6337
6338 u64
6339 nfsd_inject_print_openowners(void)
6340 {
6341         struct nfs4_client *clp;
6342         u64 count = 0;
6343         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6344                                                 nfsd_net_id);
6345
6346         if (!nfsd_netns_ready(nn))
6347                 return 0;
6348
6349         spin_lock(&nn->client_lock);
6350         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6351                 count += nfsd_print_client_openowners(clp);
6352         spin_unlock(&nn->client_lock);
6353
6354         return count;
6355 }
6356
6357 static void
6358 nfsd_reap_openowners(struct list_head *reaplist)
6359 {
6360         struct nfs4_client *clp;
6361         struct nfs4_openowner *oop, *next;
6362
6363         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6364                 list_del_init(&oop->oo_perclient);
6365                 clp = oop->oo_owner.so_client;
6366                 release_openowner(oop);
6367                 put_client(clp);
6368         }
6369 }
6370
6371 u64
6372 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6373                                      size_t addr_size)
6374 {
6375         unsigned int count = 0;
6376         struct nfs4_client *clp;
6377         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6378                                                 nfsd_net_id);
6379         LIST_HEAD(reaplist);
6380
6381         if (!nfsd_netns_ready(nn))
6382                 return count;
6383
6384         spin_lock(&nn->client_lock);
6385         clp = nfsd_find_client(addr, addr_size);
6386         if (clp)
6387                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6388         spin_unlock(&nn->client_lock);
6389         nfsd_reap_openowners(&reaplist);
6390         return count;
6391 }
6392
6393 u64
6394 nfsd_inject_forget_openowners(u64 max)
6395 {
6396         u64 count = 0;
6397         struct nfs4_client *clp;
6398         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6399                                                 nfsd_net_id);
6400         LIST_HEAD(reaplist);
6401
6402         if (!nfsd_netns_ready(nn))
6403                 return count;
6404
6405         spin_lock(&nn->client_lock);
6406         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6407                 count += nfsd_collect_client_openowners(clp, &reaplist,
6408                                                         max - count);
6409                 if (max != 0 && count >= max)
6410                         break;
6411         }
6412         spin_unlock(&nn->client_lock);
6413         nfsd_reap_openowners(&reaplist);
6414         return count;
6415 }
6416
6417 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6418                                      struct list_head *victims)
6419 {
6420         struct nfs4_delegation *dp, *next;
6421         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6422                                                 nfsd_net_id);
6423         u64 count = 0;
6424
6425         lockdep_assert_held(&nn->client_lock);
6426
6427         spin_lock(&state_lock);
6428         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6429                 if (victims) {
6430                         /*
6431                          * It's not safe to mess with delegations that have a
6432                          * non-zero dl_time. They might have already been broken
6433                          * and could be processed by the laundromat outside of
6434                          * the state_lock. Just leave them be.
6435                          */
6436                         if (dp->dl_time != 0)
6437                                 continue;
6438
6439                         atomic_inc(&clp->cl_refcount);
6440                         WARN_ON(!unhash_delegation_locked(dp));
6441                         list_add(&dp->dl_recall_lru, victims);
6442                 }
6443                 ++count;
6444                 /*
6445                  * Despite the fact that these functions deal with
6446                  * 64-bit integers for "count", we must ensure that
6447                  * it doesn't blow up the clp->cl_refcount. Throw a
6448                  * warning if we start to approach INT_MAX here.
6449                  */
6450                 WARN_ON_ONCE(count == (INT_MAX / 2));
6451                 if (count == max)
6452                         break;
6453         }
6454         spin_unlock(&state_lock);
6455         return count;
6456 }
6457
6458 static u64
6459 nfsd_print_client_delegations(struct nfs4_client *clp)
6460 {
6461         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6462
6463         nfsd_print_count(clp, count, "delegations");
6464         return count;
6465 }
6466
6467 u64
6468 nfsd_inject_print_delegations(void)
6469 {
6470         struct nfs4_client *clp;
6471         u64 count = 0;
6472         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6473                                                 nfsd_net_id);
6474
6475         if (!nfsd_netns_ready(nn))
6476                 return 0;
6477
6478         spin_lock(&nn->client_lock);
6479         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6480                 count += nfsd_print_client_delegations(clp);
6481         spin_unlock(&nn->client_lock);
6482
6483         return count;
6484 }
6485
6486 static void
6487 nfsd_forget_delegations(struct list_head *reaplist)
6488 {
6489         struct nfs4_client *clp;
6490         struct nfs4_delegation *dp, *next;
6491
6492         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6493                 list_del_init(&dp->dl_recall_lru);
6494                 clp = dp->dl_stid.sc_client;
6495                 revoke_delegation(dp);
6496                 put_client(clp);
6497         }
6498 }
6499
6500 u64
6501 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6502                                       size_t addr_size)
6503 {
6504         u64 count = 0;
6505         struct nfs4_client *clp;
6506         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6507                                                 nfsd_net_id);
6508         LIST_HEAD(reaplist);
6509
6510         if (!nfsd_netns_ready(nn))
6511                 return count;
6512
6513         spin_lock(&nn->client_lock);
6514         clp = nfsd_find_client(addr, addr_size);
6515         if (clp)
6516                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6517         spin_unlock(&nn->client_lock);
6518
6519         nfsd_forget_delegations(&reaplist);
6520         return count;
6521 }
6522
6523 u64
6524 nfsd_inject_forget_delegations(u64 max)
6525 {
6526         u64 count = 0;
6527         struct nfs4_client *clp;
6528         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6529                                                 nfsd_net_id);
6530         LIST_HEAD(reaplist);
6531
6532         if (!nfsd_netns_ready(nn))
6533                 return count;
6534
6535         spin_lock(&nn->client_lock);
6536         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6537                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6538                 if (max != 0 && count >= max)
6539                         break;
6540         }
6541         spin_unlock(&nn->client_lock);
6542         nfsd_forget_delegations(&reaplist);
6543         return count;
6544 }
6545
6546 static void
6547 nfsd_recall_delegations(struct list_head *reaplist)
6548 {
6549         struct nfs4_client *clp;
6550         struct nfs4_delegation *dp, *next;
6551
6552         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6553                 list_del_init(&dp->dl_recall_lru);
6554                 clp = dp->dl_stid.sc_client;
6555                 /*
6556                  * We skipped all entries that had a zero dl_time before,
6557                  * so we can now reset the dl_time back to 0. If a delegation
6558                  * break comes in now, then it won't make any difference since
6559                  * we're recalling it either way.
6560                  */
6561                 spin_lock(&state_lock);
6562                 dp->dl_time = 0;
6563                 spin_unlock(&state_lock);
6564                 nfsd_break_one_deleg(dp);
6565                 put_client(clp);
6566         }
6567 }
6568
6569 u64
6570 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
6571                                       size_t addr_size)
6572 {
6573         u64 count = 0;
6574         struct nfs4_client *clp;
6575         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6576                                                 nfsd_net_id);
6577         LIST_HEAD(reaplist);
6578
6579         if (!nfsd_netns_ready(nn))
6580                 return count;
6581
6582         spin_lock(&nn->client_lock);
6583         clp = nfsd_find_client(addr, addr_size);
6584         if (clp)
6585                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6586         spin_unlock(&nn->client_lock);
6587
6588         nfsd_recall_delegations(&reaplist);
6589         return count;
6590 }
6591
6592 u64
6593 nfsd_inject_recall_delegations(u64 max)
6594 {
6595         u64 count = 0;
6596         struct nfs4_client *clp, *next;
6597         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6598                                                 nfsd_net_id);
6599         LIST_HEAD(reaplist);
6600
6601         if (!nfsd_netns_ready(nn))
6602                 return count;
6603
6604         spin_lock(&nn->client_lock);
6605         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6606                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6607                 if (max != 0 && ++count >= max)
6608                         break;
6609         }
6610         spin_unlock(&nn->client_lock);
6611         nfsd_recall_delegations(&reaplist);
6612         return count;
6613 }
6614 #endif /* CONFIG_NFSD_FAULT_INJECTION */
6615
6616 /*
6617  * Since the lifetime of a delegation isn't limited to that of an open, a
6618  * client may quite reasonably hang on to a delegation as long as it has
6619  * the inode cached.  This becomes an obvious problem the first time a
6620  * client's inode cache approaches the size of the server's total memory.
6621  *
6622  * For now we avoid this problem by imposing a hard limit on the number
6623  * of delegations, which varies according to the server's memory size.
6624  */
6625 static void
6626 set_max_delegations(void)
6627 {
6628         /*
6629          * Allow at most 4 delegations per megabyte of RAM.  Quick
6630          * estimates suggest that in the worst case (where every delegation
6631          * is for a different inode), a delegation could take about 1.5K,
6632          * giving a worst case usage of about 6% of memory.
6633          */
6634         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
6635 }
6636
6637 static int nfs4_state_create_net(struct net *net)
6638 {
6639         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6640         int i;
6641
6642         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6643                         CLIENT_HASH_SIZE, GFP_KERNEL);
6644         if (!nn->conf_id_hashtbl)
6645                 goto err;
6646         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
6647                         CLIENT_HASH_SIZE, GFP_KERNEL);
6648         if (!nn->unconf_id_hashtbl)
6649                 goto err_unconf_id;
6650         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
6651                         SESSION_HASH_SIZE, GFP_KERNEL);
6652         if (!nn->sessionid_hashtbl)
6653                 goto err_sessionid;
6654
6655         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6656                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
6657                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
6658         }
6659         for (i = 0; i < SESSION_HASH_SIZE; i++)
6660                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
6661         nn->conf_name_tree = RB_ROOT;
6662         nn->unconf_name_tree = RB_ROOT;
6663         INIT_LIST_HEAD(&nn->client_lru);
6664         INIT_LIST_HEAD(&nn->close_lru);
6665         INIT_LIST_HEAD(&nn->del_recall_lru);
6666         spin_lock_init(&nn->client_lock);
6667
6668         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
6669         get_net(net);
6670
6671         return 0;
6672
6673 err_sessionid:
6674         kfree(nn->unconf_id_hashtbl);
6675 err_unconf_id:
6676         kfree(nn->conf_id_hashtbl);
6677 err:
6678         return -ENOMEM;
6679 }
6680
6681 static void
6682 nfs4_state_destroy_net(struct net *net)
6683 {
6684         int i;
6685         struct nfs4_client *clp = NULL;
6686         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6687
6688         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6689                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
6690                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6691                         destroy_client(clp);
6692                 }
6693         }
6694
6695         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6696                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
6697                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
6698                         destroy_client(clp);
6699                 }
6700         }
6701
6702         kfree(nn->sessionid_hashtbl);
6703         kfree(nn->unconf_id_hashtbl);
6704         kfree(nn->conf_id_hashtbl);
6705         put_net(net);
6706 }
6707
6708 int
6709 nfs4_state_start_net(struct net *net)
6710 {
6711         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6712         int ret;
6713
6714         ret = nfs4_state_create_net(net);
6715         if (ret)
6716                 return ret;
6717         nn->boot_time = get_seconds();
6718         nn->grace_ended = false;
6719         nn->nfsd4_manager.block_opens = true;
6720         locks_start_grace(net, &nn->nfsd4_manager);
6721         nfsd4_client_tracking_init(net);
6722         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
6723                nn->nfsd4_grace, net);
6724         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
6725         return 0;
6726 }
6727
6728 /* initialization to perform when the nfsd service is started: */
6729
6730 int
6731 nfs4_state_start(void)
6732 {
6733         int ret;
6734
6735         ret = set_callback_cred();
6736         if (ret)
6737                 return -ENOMEM;
6738         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
6739         if (laundry_wq == NULL) {
6740                 ret = -ENOMEM;
6741                 goto out_recovery;
6742         }
6743         ret = nfsd4_create_callback_queue();
6744         if (ret)
6745                 goto out_free_laundry;
6746
6747         set_max_delegations();
6748
6749         return 0;
6750
6751 out_free_laundry:
6752         destroy_workqueue(laundry_wq);
6753 out_recovery:
6754         return ret;
6755 }
6756
6757 void
6758 nfs4_state_shutdown_net(struct net *net)
6759 {
6760         struct nfs4_delegation *dp = NULL;
6761         struct list_head *pos, *next, reaplist;
6762         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
6763
6764         cancel_delayed_work_sync(&nn->laundromat_work);
6765         locks_end_grace(&nn->nfsd4_manager);
6766
6767         INIT_LIST_HEAD(&reaplist);
6768         spin_lock(&state_lock);
6769         list_for_each_safe(pos, next, &nn->del_recall_lru) {
6770                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6771                 WARN_ON(!unhash_delegation_locked(dp));
6772                 list_add(&dp->dl_recall_lru, &reaplist);
6773         }
6774         spin_unlock(&state_lock);
6775         list_for_each_safe(pos, next, &reaplist) {
6776                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
6777                 list_del_init(&dp->dl_recall_lru);
6778                 put_clnt_odstate(dp->dl_clnt_odstate);
6779                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
6780                 nfs4_put_stid(&dp->dl_stid);
6781         }
6782
6783         nfsd4_client_tracking_exit(net);
6784         nfs4_state_destroy_net(net);
6785 }
6786
6787 void
6788 nfs4_state_shutdown(void)
6789 {
6790         destroy_workqueue(laundry_wq);
6791         nfsd4_destroy_callback_queue();
6792 }
6793
6794 static void
6795 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6796 {
6797         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
6798                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
6799 }
6800
6801 static void
6802 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
6803 {
6804         if (cstate->minorversion) {
6805                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
6806                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6807         }
6808 }
6809
6810 void
6811 clear_current_stateid(struct nfsd4_compound_state *cstate)
6812 {
6813         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
6814 }
6815
6816 /*
6817  * functions to set current state id
6818  */
6819 void
6820 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6821 {
6822         put_stateid(cstate, &odp->od_stateid);
6823 }
6824
6825 void
6826 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
6827 {
6828         put_stateid(cstate, &open->op_stateid);
6829 }
6830
6831 void
6832 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6833 {
6834         put_stateid(cstate, &close->cl_stateid);
6835 }
6836
6837 void
6838 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
6839 {
6840         put_stateid(cstate, &lock->lk_resp_stateid);
6841 }
6842
6843 /*
6844  * functions to consume current state id
6845  */
6846
6847 void
6848 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
6849 {
6850         get_stateid(cstate, &odp->od_stateid);
6851 }
6852
6853 void
6854 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
6855 {
6856         get_stateid(cstate, &drp->dr_stateid);
6857 }
6858
6859 void
6860 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
6861 {
6862         get_stateid(cstate, &fsp->fr_stateid);
6863 }
6864
6865 void
6866 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
6867 {
6868         get_stateid(cstate, &setattr->sa_stateid);
6869 }
6870
6871 void
6872 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
6873 {
6874         get_stateid(cstate, &close->cl_stateid);
6875 }
6876
6877 void
6878 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
6879 {
6880         get_stateid(cstate, &locku->lu_stateid);
6881 }
6882
6883 void
6884 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
6885 {
6886         get_stateid(cstate, &read->rd_stateid);
6887 }
6888
6889 void
6890 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
6891 {
6892         get_stateid(cstate, &write->wr_stateid);
6893 }