Merge tag 'f2fs-for-4.17' of git://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk...
[muen/linux.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/addr.h>
44 #include <linux/jhash.h>
45 #include "xdr4.h"
46 #include "xdr4cb.h"
47 #include "vfs.h"
48 #include "current_stateid.h"
49
50 #include "netns.h"
51 #include "pnfs.h"
52
53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
54
55 #define all_ones {{~0,~0},~0}
56 static const stateid_t one_stateid = {
57         .si_generation = ~0,
58         .si_opaque = all_ones,
59 };
60 static const stateid_t zero_stateid = {
61         /* all fields zero */
62 };
63 static const stateid_t currentstateid = {
64         .si_generation = 1,
65 };
66 static const stateid_t close_stateid = {
67         .si_generation = 0xffffffffU,
68 };
69
70 static u64 current_sessionid = 1;
71
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
76
77 /* forward declarations */
78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
80
81 /* Locking: */
82
83 /*
84  * Currently used for the del_recall_lru and file hash table.  In an
85  * effort to decrease the scope of the client_mutex, this spinlock may
86  * eventually cover more:
87  */
88 static DEFINE_SPINLOCK(state_lock);
89
90 enum nfsd4_st_mutex_lock_subclass {
91         OPEN_STATEID_MUTEX = 0,
92         LOCK_STATEID_MUTEX = 1,
93 };
94
95 /*
96  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
97  * the refcount on the open stateid to drop.
98  */
99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
100
101 static struct kmem_cache *openowner_slab;
102 static struct kmem_cache *lockowner_slab;
103 static struct kmem_cache *file_slab;
104 static struct kmem_cache *stateid_slab;
105 static struct kmem_cache *deleg_slab;
106 static struct kmem_cache *odstate_slab;
107
108 static void free_session(struct nfsd4_session *);
109
110 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
111 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
112
113 static bool is_session_dead(struct nfsd4_session *ses)
114 {
115         return ses->se_flags & NFS4_SESSION_DEAD;
116 }
117
118 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
119 {
120         if (atomic_read(&ses->se_ref) > ref_held_by_me)
121                 return nfserr_jukebox;
122         ses->se_flags |= NFS4_SESSION_DEAD;
123         return nfs_ok;
124 }
125
126 static bool is_client_expired(struct nfs4_client *clp)
127 {
128         return clp->cl_time == 0;
129 }
130
131 static __be32 get_client_locked(struct nfs4_client *clp)
132 {
133         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
134
135         lockdep_assert_held(&nn->client_lock);
136
137         if (is_client_expired(clp))
138                 return nfserr_expired;
139         atomic_inc(&clp->cl_refcount);
140         return nfs_ok;
141 }
142
143 /* must be called under the client_lock */
144 static inline void
145 renew_client_locked(struct nfs4_client *clp)
146 {
147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
148
149         if (is_client_expired(clp)) {
150                 WARN_ON(1);
151                 printk("%s: client (clientid %08x/%08x) already expired\n",
152                         __func__,
153                         clp->cl_clientid.cl_boot,
154                         clp->cl_clientid.cl_id);
155                 return;
156         }
157
158         dprintk("renewing client (clientid %08x/%08x)\n",
159                         clp->cl_clientid.cl_boot,
160                         clp->cl_clientid.cl_id);
161         list_move_tail(&clp->cl_lru, &nn->client_lru);
162         clp->cl_time = get_seconds();
163 }
164
165 static void put_client_renew_locked(struct nfs4_client *clp)
166 {
167         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
168
169         lockdep_assert_held(&nn->client_lock);
170
171         if (!atomic_dec_and_test(&clp->cl_refcount))
172                 return;
173         if (!is_client_expired(clp))
174                 renew_client_locked(clp);
175 }
176
177 static void put_client_renew(struct nfs4_client *clp)
178 {
179         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
180
181         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
182                 return;
183         if (!is_client_expired(clp))
184                 renew_client_locked(clp);
185         spin_unlock(&nn->client_lock);
186 }
187
188 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
189 {
190         __be32 status;
191
192         if (is_session_dead(ses))
193                 return nfserr_badsession;
194         status = get_client_locked(ses->se_client);
195         if (status)
196                 return status;
197         atomic_inc(&ses->se_ref);
198         return nfs_ok;
199 }
200
201 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
202 {
203         struct nfs4_client *clp = ses->se_client;
204         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
205
206         lockdep_assert_held(&nn->client_lock);
207
208         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
209                 free_session(ses);
210         put_client_renew_locked(clp);
211 }
212
213 static void nfsd4_put_session(struct nfsd4_session *ses)
214 {
215         struct nfs4_client *clp = ses->se_client;
216         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
217
218         spin_lock(&nn->client_lock);
219         nfsd4_put_session_locked(ses);
220         spin_unlock(&nn->client_lock);
221 }
222
223 static struct nfsd4_blocked_lock *
224 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
225                         struct nfsd_net *nn)
226 {
227         struct nfsd4_blocked_lock *cur, *found = NULL;
228
229         spin_lock(&nn->blocked_locks_lock);
230         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
231                 if (fh_match(fh, &cur->nbl_fh)) {
232                         list_del_init(&cur->nbl_list);
233                         list_del_init(&cur->nbl_lru);
234                         found = cur;
235                         break;
236                 }
237         }
238         spin_unlock(&nn->blocked_locks_lock);
239         if (found)
240                 posix_unblock_lock(&found->nbl_lock);
241         return found;
242 }
243
244 static struct nfsd4_blocked_lock *
245 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
246                         struct nfsd_net *nn)
247 {
248         struct nfsd4_blocked_lock *nbl;
249
250         nbl = find_blocked_lock(lo, fh, nn);
251         if (!nbl) {
252                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
253                 if (nbl) {
254                         fh_copy_shallow(&nbl->nbl_fh, fh);
255                         locks_init_lock(&nbl->nbl_lock);
256                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
257                                         &nfsd4_cb_notify_lock_ops,
258                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
259                 }
260         }
261         return nbl;
262 }
263
264 static void
265 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
266 {
267         locks_release_private(&nbl->nbl_lock);
268         kfree(nbl);
269 }
270
271 static void
272 remove_blocked_locks(struct nfs4_lockowner *lo)
273 {
274         struct nfs4_client *clp = lo->lo_owner.so_client;
275         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
276         struct nfsd4_blocked_lock *nbl;
277         LIST_HEAD(reaplist);
278
279         /* Dequeue all blocked locks */
280         spin_lock(&nn->blocked_locks_lock);
281         while (!list_empty(&lo->lo_blocked)) {
282                 nbl = list_first_entry(&lo->lo_blocked,
283                                         struct nfsd4_blocked_lock,
284                                         nbl_list);
285                 list_del_init(&nbl->nbl_list);
286                 list_move(&nbl->nbl_lru, &reaplist);
287         }
288         spin_unlock(&nn->blocked_locks_lock);
289
290         /* Now free them */
291         while (!list_empty(&reaplist)) {
292                 nbl = list_first_entry(&reaplist, struct nfsd4_blocked_lock,
293                                         nbl_lru);
294                 list_del_init(&nbl->nbl_lru);
295                 posix_unblock_lock(&nbl->nbl_lock);
296                 free_blocked_lock(nbl);
297         }
298 }
299
300 static int
301 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
302 {
303         /*
304          * Since this is just an optimization, we don't try very hard if it
305          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
306          * just quit trying on anything else.
307          */
308         switch (task->tk_status) {
309         case -NFS4ERR_DELAY:
310                 rpc_delay(task, 1 * HZ);
311                 return 0;
312         default:
313                 return 1;
314         }
315 }
316
317 static void
318 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
319 {
320         struct nfsd4_blocked_lock       *nbl = container_of(cb,
321                                                 struct nfsd4_blocked_lock, nbl_cb);
322
323         free_blocked_lock(nbl);
324 }
325
326 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
327         .done           = nfsd4_cb_notify_lock_done,
328         .release        = nfsd4_cb_notify_lock_release,
329 };
330
331 static inline struct nfs4_stateowner *
332 nfs4_get_stateowner(struct nfs4_stateowner *sop)
333 {
334         atomic_inc(&sop->so_count);
335         return sop;
336 }
337
338 static int
339 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
340 {
341         return (sop->so_owner.len == owner->len) &&
342                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
343 }
344
345 static struct nfs4_openowner *
346 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
347                         struct nfs4_client *clp)
348 {
349         struct nfs4_stateowner *so;
350
351         lockdep_assert_held(&clp->cl_lock);
352
353         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
354                             so_strhash) {
355                 if (!so->so_is_open_owner)
356                         continue;
357                 if (same_owner_str(so, &open->op_owner))
358                         return openowner(nfs4_get_stateowner(so));
359         }
360         return NULL;
361 }
362
363 static struct nfs4_openowner *
364 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
365                         struct nfs4_client *clp)
366 {
367         struct nfs4_openowner *oo;
368
369         spin_lock(&clp->cl_lock);
370         oo = find_openstateowner_str_locked(hashval, open, clp);
371         spin_unlock(&clp->cl_lock);
372         return oo;
373 }
374
375 static inline u32
376 opaque_hashval(const void *ptr, int nbytes)
377 {
378         unsigned char *cptr = (unsigned char *) ptr;
379
380         u32 x = 0;
381         while (nbytes--) {
382                 x *= 37;
383                 x += *cptr++;
384         }
385         return x;
386 }
387
388 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
389 {
390         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
391
392         kmem_cache_free(file_slab, fp);
393 }
394
395 void
396 put_nfs4_file(struct nfs4_file *fi)
397 {
398         might_lock(&state_lock);
399
400         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
401                 hlist_del_rcu(&fi->fi_hash);
402                 spin_unlock(&state_lock);
403                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
404                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
405                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
406         }
407 }
408
409 static struct file *
410 __nfs4_get_fd(struct nfs4_file *f, int oflag)
411 {
412         if (f->fi_fds[oflag])
413                 return get_file(f->fi_fds[oflag]);
414         return NULL;
415 }
416
417 static struct file *
418 find_writeable_file_locked(struct nfs4_file *f)
419 {
420         struct file *ret;
421
422         lockdep_assert_held(&f->fi_lock);
423
424         ret = __nfs4_get_fd(f, O_WRONLY);
425         if (!ret)
426                 ret = __nfs4_get_fd(f, O_RDWR);
427         return ret;
428 }
429
430 static struct file *
431 find_writeable_file(struct nfs4_file *f)
432 {
433         struct file *ret;
434
435         spin_lock(&f->fi_lock);
436         ret = find_writeable_file_locked(f);
437         spin_unlock(&f->fi_lock);
438
439         return ret;
440 }
441
442 static struct file *find_readable_file_locked(struct nfs4_file *f)
443 {
444         struct file *ret;
445
446         lockdep_assert_held(&f->fi_lock);
447
448         ret = __nfs4_get_fd(f, O_RDONLY);
449         if (!ret)
450                 ret = __nfs4_get_fd(f, O_RDWR);
451         return ret;
452 }
453
454 static struct file *
455 find_readable_file(struct nfs4_file *f)
456 {
457         struct file *ret;
458
459         spin_lock(&f->fi_lock);
460         ret = find_readable_file_locked(f);
461         spin_unlock(&f->fi_lock);
462
463         return ret;
464 }
465
466 struct file *
467 find_any_file(struct nfs4_file *f)
468 {
469         struct file *ret;
470
471         spin_lock(&f->fi_lock);
472         ret = __nfs4_get_fd(f, O_RDWR);
473         if (!ret) {
474                 ret = __nfs4_get_fd(f, O_WRONLY);
475                 if (!ret)
476                         ret = __nfs4_get_fd(f, O_RDONLY);
477         }
478         spin_unlock(&f->fi_lock);
479         return ret;
480 }
481
482 static atomic_long_t num_delegations;
483 unsigned long max_delegations;
484
485 /*
486  * Open owner state (share locks)
487  */
488
489 /* hash tables for lock and open owners */
490 #define OWNER_HASH_BITS              8
491 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
492 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
493
494 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
495 {
496         unsigned int ret;
497
498         ret = opaque_hashval(ownername->data, ownername->len);
499         return ret & OWNER_HASH_MASK;
500 }
501
502 /* hash table for nfs4_file */
503 #define FILE_HASH_BITS                   8
504 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
505
506 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
507 {
508         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
509 }
510
511 static unsigned int file_hashval(struct knfsd_fh *fh)
512 {
513         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
514 }
515
516 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
517
518 static void
519 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
520 {
521         lockdep_assert_held(&fp->fi_lock);
522
523         if (access & NFS4_SHARE_ACCESS_WRITE)
524                 atomic_inc(&fp->fi_access[O_WRONLY]);
525         if (access & NFS4_SHARE_ACCESS_READ)
526                 atomic_inc(&fp->fi_access[O_RDONLY]);
527 }
528
529 static __be32
530 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
531 {
532         lockdep_assert_held(&fp->fi_lock);
533
534         /* Does this access mode make sense? */
535         if (access & ~NFS4_SHARE_ACCESS_BOTH)
536                 return nfserr_inval;
537
538         /* Does it conflict with a deny mode already set? */
539         if ((access & fp->fi_share_deny) != 0)
540                 return nfserr_share_denied;
541
542         __nfs4_file_get_access(fp, access);
543         return nfs_ok;
544 }
545
546 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
547 {
548         /* Common case is that there is no deny mode. */
549         if (deny) {
550                 /* Does this deny mode make sense? */
551                 if (deny & ~NFS4_SHARE_DENY_BOTH)
552                         return nfserr_inval;
553
554                 if ((deny & NFS4_SHARE_DENY_READ) &&
555                     atomic_read(&fp->fi_access[O_RDONLY]))
556                         return nfserr_share_denied;
557
558                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
559                     atomic_read(&fp->fi_access[O_WRONLY]))
560                         return nfserr_share_denied;
561         }
562         return nfs_ok;
563 }
564
565 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
566 {
567         might_lock(&fp->fi_lock);
568
569         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
570                 struct file *f1 = NULL;
571                 struct file *f2 = NULL;
572
573                 swap(f1, fp->fi_fds[oflag]);
574                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
575                         swap(f2, fp->fi_fds[O_RDWR]);
576                 spin_unlock(&fp->fi_lock);
577                 if (f1)
578                         fput(f1);
579                 if (f2)
580                         fput(f2);
581         }
582 }
583
584 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
585 {
586         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
587
588         if (access & NFS4_SHARE_ACCESS_WRITE)
589                 __nfs4_file_put_access(fp, O_WRONLY);
590         if (access & NFS4_SHARE_ACCESS_READ)
591                 __nfs4_file_put_access(fp, O_RDONLY);
592 }
593
594 /*
595  * Allocate a new open/delegation state counter. This is needed for
596  * pNFS for proper return on close semantics.
597  *
598  * Note that we only allocate it for pNFS-enabled exports, otherwise
599  * all pointers to struct nfs4_clnt_odstate are always NULL.
600  */
601 static struct nfs4_clnt_odstate *
602 alloc_clnt_odstate(struct nfs4_client *clp)
603 {
604         struct nfs4_clnt_odstate *co;
605
606         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
607         if (co) {
608                 co->co_client = clp;
609                 refcount_set(&co->co_odcount, 1);
610         }
611         return co;
612 }
613
614 static void
615 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
616 {
617         struct nfs4_file *fp = co->co_file;
618
619         lockdep_assert_held(&fp->fi_lock);
620         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
621 }
622
623 static inline void
624 get_clnt_odstate(struct nfs4_clnt_odstate *co)
625 {
626         if (co)
627                 refcount_inc(&co->co_odcount);
628 }
629
630 static void
631 put_clnt_odstate(struct nfs4_clnt_odstate *co)
632 {
633         struct nfs4_file *fp;
634
635         if (!co)
636                 return;
637
638         fp = co->co_file;
639         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
640                 list_del(&co->co_perfile);
641                 spin_unlock(&fp->fi_lock);
642
643                 nfsd4_return_all_file_layouts(co->co_client, fp);
644                 kmem_cache_free(odstate_slab, co);
645         }
646 }
647
648 static struct nfs4_clnt_odstate *
649 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
650 {
651         struct nfs4_clnt_odstate *co;
652         struct nfs4_client *cl;
653
654         if (!new)
655                 return NULL;
656
657         cl = new->co_client;
658
659         spin_lock(&fp->fi_lock);
660         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
661                 if (co->co_client == cl) {
662                         get_clnt_odstate(co);
663                         goto out;
664                 }
665         }
666         co = new;
667         co->co_file = fp;
668         hash_clnt_odstate_locked(new);
669 out:
670         spin_unlock(&fp->fi_lock);
671         return co;
672 }
673
674 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
675                                   void (*sc_free)(struct nfs4_stid *))
676 {
677         struct nfs4_stid *stid;
678         int new_id;
679
680         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
681         if (!stid)
682                 return NULL;
683
684         idr_preload(GFP_KERNEL);
685         spin_lock(&cl->cl_lock);
686         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
687         spin_unlock(&cl->cl_lock);
688         idr_preload_end();
689         if (new_id < 0)
690                 goto out_free;
691
692         stid->sc_free = sc_free;
693         stid->sc_client = cl;
694         stid->sc_stateid.si_opaque.so_id = new_id;
695         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
696         /* Will be incremented before return to client: */
697         refcount_set(&stid->sc_count, 1);
698         spin_lock_init(&stid->sc_lock);
699
700         /*
701          * It shouldn't be a problem to reuse an opaque stateid value.
702          * I don't think it is for 4.1.  But with 4.0 I worry that, for
703          * example, a stray write retransmission could be accepted by
704          * the server when it should have been rejected.  Therefore,
705          * adopt a trick from the sctp code to attempt to maximize the
706          * amount of time until an id is reused, by ensuring they always
707          * "increase" (mod INT_MAX):
708          */
709         return stid;
710 out_free:
711         kmem_cache_free(slab, stid);
712         return NULL;
713 }
714
715 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
716 {
717         struct nfs4_stid *stid;
718
719         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
720         if (!stid)
721                 return NULL;
722
723         return openlockstateid(stid);
724 }
725
726 static void nfs4_free_deleg(struct nfs4_stid *stid)
727 {
728         kmem_cache_free(deleg_slab, stid);
729         atomic_long_dec(&num_delegations);
730 }
731
732 /*
733  * When we recall a delegation, we should be careful not to hand it
734  * out again straight away.
735  * To ensure this we keep a pair of bloom filters ('new' and 'old')
736  * in which the filehandles of recalled delegations are "stored".
737  * If a filehandle appear in either filter, a delegation is blocked.
738  * When a delegation is recalled, the filehandle is stored in the "new"
739  * filter.
740  * Every 30 seconds we swap the filters and clear the "new" one,
741  * unless both are empty of course.
742  *
743  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
744  * low 3 bytes as hash-table indices.
745  *
746  * 'blocked_delegations_lock', which is always taken in block_delegations(),
747  * is used to manage concurrent access.  Testing does not need the lock
748  * except when swapping the two filters.
749  */
750 static DEFINE_SPINLOCK(blocked_delegations_lock);
751 static struct bloom_pair {
752         int     entries, old_entries;
753         time_t  swap_time;
754         int     new; /* index into 'set' */
755         DECLARE_BITMAP(set[2], 256);
756 } blocked_delegations;
757
758 static int delegation_blocked(struct knfsd_fh *fh)
759 {
760         u32 hash;
761         struct bloom_pair *bd = &blocked_delegations;
762
763         if (bd->entries == 0)
764                 return 0;
765         if (seconds_since_boot() - bd->swap_time > 30) {
766                 spin_lock(&blocked_delegations_lock);
767                 if (seconds_since_boot() - bd->swap_time > 30) {
768                         bd->entries -= bd->old_entries;
769                         bd->old_entries = bd->entries;
770                         memset(bd->set[bd->new], 0,
771                                sizeof(bd->set[0]));
772                         bd->new = 1-bd->new;
773                         bd->swap_time = seconds_since_boot();
774                 }
775                 spin_unlock(&blocked_delegations_lock);
776         }
777         hash = jhash(&fh->fh_base, fh->fh_size, 0);
778         if (test_bit(hash&255, bd->set[0]) &&
779             test_bit((hash>>8)&255, bd->set[0]) &&
780             test_bit((hash>>16)&255, bd->set[0]))
781                 return 1;
782
783         if (test_bit(hash&255, bd->set[1]) &&
784             test_bit((hash>>8)&255, bd->set[1]) &&
785             test_bit((hash>>16)&255, bd->set[1]))
786                 return 1;
787
788         return 0;
789 }
790
791 static void block_delegations(struct knfsd_fh *fh)
792 {
793         u32 hash;
794         struct bloom_pair *bd = &blocked_delegations;
795
796         hash = jhash(&fh->fh_base, fh->fh_size, 0);
797
798         spin_lock(&blocked_delegations_lock);
799         __set_bit(hash&255, bd->set[bd->new]);
800         __set_bit((hash>>8)&255, bd->set[bd->new]);
801         __set_bit((hash>>16)&255, bd->set[bd->new]);
802         if (bd->entries == 0)
803                 bd->swap_time = seconds_since_boot();
804         bd->entries += 1;
805         spin_unlock(&blocked_delegations_lock);
806 }
807
808 static struct nfs4_delegation *
809 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
810                  struct nfs4_clnt_odstate *odstate)
811 {
812         struct nfs4_delegation *dp;
813         long n;
814
815         dprintk("NFSD alloc_init_deleg\n");
816         n = atomic_long_inc_return(&num_delegations);
817         if (n < 0 || n > max_delegations)
818                 goto out_dec;
819         if (delegation_blocked(&current_fh->fh_handle))
820                 goto out_dec;
821         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
822         if (dp == NULL)
823                 goto out_dec;
824
825         /*
826          * delegation seqid's are never incremented.  The 4.1 special
827          * meaning of seqid 0 isn't meaningful, really, but let's avoid
828          * 0 anyway just for consistency and use 1:
829          */
830         dp->dl_stid.sc_stateid.si_generation = 1;
831         INIT_LIST_HEAD(&dp->dl_perfile);
832         INIT_LIST_HEAD(&dp->dl_perclnt);
833         INIT_LIST_HEAD(&dp->dl_recall_lru);
834         dp->dl_clnt_odstate = odstate;
835         get_clnt_odstate(odstate);
836         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
837         dp->dl_retries = 1;
838         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
839                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
840         return dp;
841 out_dec:
842         atomic_long_dec(&num_delegations);
843         return NULL;
844 }
845
846 void
847 nfs4_put_stid(struct nfs4_stid *s)
848 {
849         struct nfs4_file *fp = s->sc_file;
850         struct nfs4_client *clp = s->sc_client;
851
852         might_lock(&clp->cl_lock);
853
854         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
855                 wake_up_all(&close_wq);
856                 return;
857         }
858         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
859         spin_unlock(&clp->cl_lock);
860         s->sc_free(s);
861         if (fp)
862                 put_nfs4_file(fp);
863 }
864
865 void
866 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
867 {
868         stateid_t *src = &stid->sc_stateid;
869
870         spin_lock(&stid->sc_lock);
871         if (unlikely(++src->si_generation == 0))
872                 src->si_generation = 1;
873         memcpy(dst, src, sizeof(*dst));
874         spin_unlock(&stid->sc_lock);
875 }
876
877 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
878 {
879         struct file *filp = NULL;
880
881         spin_lock(&fp->fi_lock);
882         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
883                 swap(filp, fp->fi_deleg_file);
884         spin_unlock(&fp->fi_lock);
885
886         if (filp) {
887                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
888                 fput(filp);
889         }
890 }
891
892 void nfs4_unhash_stid(struct nfs4_stid *s)
893 {
894         s->sc_type = 0;
895 }
896
897 /**
898  * nfs4_get_existing_delegation - Discover if this delegation already exists
899  * @clp:     a pointer to the nfs4_client we're granting a delegation to
900  * @fp:      a pointer to the nfs4_file we're granting a delegation on
901  *
902  * Return:
903  *      On success: NULL if an existing delegation was not found.
904  *
905  *      On error: -EAGAIN if one was previously granted to this nfs4_client
906  *                 for this nfs4_file.
907  *
908  */
909
910 static int
911 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
912 {
913         struct nfs4_delegation *searchdp = NULL;
914         struct nfs4_client *searchclp = NULL;
915
916         lockdep_assert_held(&state_lock);
917         lockdep_assert_held(&fp->fi_lock);
918
919         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
920                 searchclp = searchdp->dl_stid.sc_client;
921                 if (clp == searchclp) {
922                         return -EAGAIN;
923                 }
924         }
925         return 0;
926 }
927
928 /**
929  * hash_delegation_locked - Add a delegation to the appropriate lists
930  * @dp:     a pointer to the nfs4_delegation we are adding.
931  * @fp:     a pointer to the nfs4_file we're granting a delegation on
932  *
933  * Return:
934  *      On success: NULL if the delegation was successfully hashed.
935  *
936  *      On error: -EAGAIN if one was previously granted to this
937  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
938  *
939  */
940
941 static int
942 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
943 {
944         int status;
945         struct nfs4_client *clp = dp->dl_stid.sc_client;
946
947         lockdep_assert_held(&state_lock);
948         lockdep_assert_held(&fp->fi_lock);
949
950         status = nfs4_get_existing_delegation(clp, fp);
951         if (status)
952                 return status;
953         ++fp->fi_delegees;
954         refcount_inc(&dp->dl_stid.sc_count);
955         dp->dl_stid.sc_type = NFS4_DELEG_STID;
956         list_add(&dp->dl_perfile, &fp->fi_delegations);
957         list_add(&dp->dl_perclnt, &clp->cl_delegations);
958         return 0;
959 }
960
961 static bool
962 unhash_delegation_locked(struct nfs4_delegation *dp)
963 {
964         struct nfs4_file *fp = dp->dl_stid.sc_file;
965
966         lockdep_assert_held(&state_lock);
967
968         if (list_empty(&dp->dl_perfile))
969                 return false;
970
971         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
972         /* Ensure that deleg break won't try to requeue it */
973         ++dp->dl_time;
974         spin_lock(&fp->fi_lock);
975         list_del_init(&dp->dl_perclnt);
976         list_del_init(&dp->dl_recall_lru);
977         list_del_init(&dp->dl_perfile);
978         spin_unlock(&fp->fi_lock);
979         return true;
980 }
981
982 static void destroy_delegation(struct nfs4_delegation *dp)
983 {
984         bool unhashed;
985
986         spin_lock(&state_lock);
987         unhashed = unhash_delegation_locked(dp);
988         spin_unlock(&state_lock);
989         if (unhashed) {
990                 put_clnt_odstate(dp->dl_clnt_odstate);
991                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
992                 nfs4_put_stid(&dp->dl_stid);
993         }
994 }
995
996 static void revoke_delegation(struct nfs4_delegation *dp)
997 {
998         struct nfs4_client *clp = dp->dl_stid.sc_client;
999
1000         WARN_ON(!list_empty(&dp->dl_recall_lru));
1001
1002         put_clnt_odstate(dp->dl_clnt_odstate);
1003         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1004
1005         if (clp->cl_minorversion == 0)
1006                 nfs4_put_stid(&dp->dl_stid);
1007         else {
1008                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
1009                 spin_lock(&clp->cl_lock);
1010                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
1011                 spin_unlock(&clp->cl_lock);
1012         }
1013 }
1014
1015 /* 
1016  * SETCLIENTID state 
1017  */
1018
1019 static unsigned int clientid_hashval(u32 id)
1020 {
1021         return id & CLIENT_HASH_MASK;
1022 }
1023
1024 static unsigned int clientstr_hashval(const char *name)
1025 {
1026         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
1027 }
1028
1029 /*
1030  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1031  * st_{access,deny}_bmap field of the stateid, in order to track not
1032  * only what share bits are currently in force, but also what
1033  * combinations of share bits previous opens have used.  This allows us
1034  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1035  * return an error if the client attempt to downgrade to a combination
1036  * of share bits not explicable by closing some of its previous opens.
1037  *
1038  * XXX: This enforcement is actually incomplete, since we don't keep
1039  * track of access/deny bit combinations; so, e.g., we allow:
1040  *
1041  *      OPEN allow read, deny write
1042  *      OPEN allow both, deny none
1043  *      DOWNGRADE allow read, deny none
1044  *
1045  * which we should reject.
1046  */
1047 static unsigned int
1048 bmap_to_share_mode(unsigned long bmap) {
1049         int i;
1050         unsigned int access = 0;
1051
1052         for (i = 1; i < 4; i++) {
1053                 if (test_bit(i, &bmap))
1054                         access |= i;
1055         }
1056         return access;
1057 }
1058
1059 /* set share access for a given stateid */
1060 static inline void
1061 set_access(u32 access, struct nfs4_ol_stateid *stp)
1062 {
1063         unsigned char mask = 1 << access;
1064
1065         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1066         stp->st_access_bmap |= mask;
1067 }
1068
1069 /* clear share access for a given stateid */
1070 static inline void
1071 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1072 {
1073         unsigned char mask = 1 << access;
1074
1075         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1076         stp->st_access_bmap &= ~mask;
1077 }
1078
1079 /* test whether a given stateid has access */
1080 static inline bool
1081 test_access(u32 access, struct nfs4_ol_stateid *stp)
1082 {
1083         unsigned char mask = 1 << access;
1084
1085         return (bool)(stp->st_access_bmap & mask);
1086 }
1087
1088 /* set share deny for a given stateid */
1089 static inline void
1090 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1091 {
1092         unsigned char mask = 1 << deny;
1093
1094         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1095         stp->st_deny_bmap |= mask;
1096 }
1097
1098 /* clear share deny for a given stateid */
1099 static inline void
1100 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1101 {
1102         unsigned char mask = 1 << deny;
1103
1104         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1105         stp->st_deny_bmap &= ~mask;
1106 }
1107
1108 /* test whether a given stateid is denying specific access */
1109 static inline bool
1110 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1111 {
1112         unsigned char mask = 1 << deny;
1113
1114         return (bool)(stp->st_deny_bmap & mask);
1115 }
1116
1117 static int nfs4_access_to_omode(u32 access)
1118 {
1119         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1120         case NFS4_SHARE_ACCESS_READ:
1121                 return O_RDONLY;
1122         case NFS4_SHARE_ACCESS_WRITE:
1123                 return O_WRONLY;
1124         case NFS4_SHARE_ACCESS_BOTH:
1125                 return O_RDWR;
1126         }
1127         WARN_ON_ONCE(1);
1128         return O_RDONLY;
1129 }
1130
1131 /*
1132  * A stateid that had a deny mode associated with it is being released
1133  * or downgraded. Recalculate the deny mode on the file.
1134  */
1135 static void
1136 recalculate_deny_mode(struct nfs4_file *fp)
1137 {
1138         struct nfs4_ol_stateid *stp;
1139
1140         spin_lock(&fp->fi_lock);
1141         fp->fi_share_deny = 0;
1142         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1143                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1144         spin_unlock(&fp->fi_lock);
1145 }
1146
1147 static void
1148 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1149 {
1150         int i;
1151         bool change = false;
1152
1153         for (i = 1; i < 4; i++) {
1154                 if ((i & deny) != i) {
1155                         change = true;
1156                         clear_deny(i, stp);
1157                 }
1158         }
1159
1160         /* Recalculate per-file deny mode if there was a change */
1161         if (change)
1162                 recalculate_deny_mode(stp->st_stid.sc_file);
1163 }
1164
1165 /* release all access and file references for a given stateid */
1166 static void
1167 release_all_access(struct nfs4_ol_stateid *stp)
1168 {
1169         int i;
1170         struct nfs4_file *fp = stp->st_stid.sc_file;
1171
1172         if (fp && stp->st_deny_bmap != 0)
1173                 recalculate_deny_mode(fp);
1174
1175         for (i = 1; i < 4; i++) {
1176                 if (test_access(i, stp))
1177                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1178                 clear_access(i, stp);
1179         }
1180 }
1181
1182 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1183 {
1184         kfree(sop->so_owner.data);
1185         sop->so_ops->so_free(sop);
1186 }
1187
1188 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1189 {
1190         struct nfs4_client *clp = sop->so_client;
1191
1192         might_lock(&clp->cl_lock);
1193
1194         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1195                 return;
1196         sop->so_ops->so_unhash(sop);
1197         spin_unlock(&clp->cl_lock);
1198         nfs4_free_stateowner(sop);
1199 }
1200
1201 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1202 {
1203         struct nfs4_file *fp = stp->st_stid.sc_file;
1204
1205         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1206
1207         if (list_empty(&stp->st_perfile))
1208                 return false;
1209
1210         spin_lock(&fp->fi_lock);
1211         list_del_init(&stp->st_perfile);
1212         spin_unlock(&fp->fi_lock);
1213         list_del(&stp->st_perstateowner);
1214         return true;
1215 }
1216
1217 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1218 {
1219         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1220
1221         put_clnt_odstate(stp->st_clnt_odstate);
1222         release_all_access(stp);
1223         if (stp->st_stateowner)
1224                 nfs4_put_stateowner(stp->st_stateowner);
1225         kmem_cache_free(stateid_slab, stid);
1226 }
1227
1228 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1229 {
1230         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1231         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1232         struct file *file;
1233
1234         file = find_any_file(stp->st_stid.sc_file);
1235         if (file)
1236                 filp_close(file, (fl_owner_t)lo);
1237         nfs4_free_ol_stateid(stid);
1238 }
1239
1240 /*
1241  * Put the persistent reference to an already unhashed generic stateid, while
1242  * holding the cl_lock. If it's the last reference, then put it onto the
1243  * reaplist for later destruction.
1244  */
1245 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1246                                        struct list_head *reaplist)
1247 {
1248         struct nfs4_stid *s = &stp->st_stid;
1249         struct nfs4_client *clp = s->sc_client;
1250
1251         lockdep_assert_held(&clp->cl_lock);
1252
1253         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1254
1255         if (!refcount_dec_and_test(&s->sc_count)) {
1256                 wake_up_all(&close_wq);
1257                 return;
1258         }
1259
1260         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1261         list_add(&stp->st_locks, reaplist);
1262 }
1263
1264 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1265 {
1266         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1267
1268         list_del_init(&stp->st_locks);
1269         nfs4_unhash_stid(&stp->st_stid);
1270         return unhash_ol_stateid(stp);
1271 }
1272
1273 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1274 {
1275         struct nfs4_client *clp = stp->st_stid.sc_client;
1276         bool unhashed;
1277
1278         spin_lock(&clp->cl_lock);
1279         unhashed = unhash_lock_stateid(stp);
1280         spin_unlock(&clp->cl_lock);
1281         if (unhashed)
1282                 nfs4_put_stid(&stp->st_stid);
1283 }
1284
1285 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1286 {
1287         struct nfs4_client *clp = lo->lo_owner.so_client;
1288
1289         lockdep_assert_held(&clp->cl_lock);
1290
1291         list_del_init(&lo->lo_owner.so_strhash);
1292 }
1293
1294 /*
1295  * Free a list of generic stateids that were collected earlier after being
1296  * fully unhashed.
1297  */
1298 static void
1299 free_ol_stateid_reaplist(struct list_head *reaplist)
1300 {
1301         struct nfs4_ol_stateid *stp;
1302         struct nfs4_file *fp;
1303
1304         might_sleep();
1305
1306         while (!list_empty(reaplist)) {
1307                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1308                                        st_locks);
1309                 list_del(&stp->st_locks);
1310                 fp = stp->st_stid.sc_file;
1311                 stp->st_stid.sc_free(&stp->st_stid);
1312                 if (fp)
1313                         put_nfs4_file(fp);
1314         }
1315 }
1316
1317 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1318                                        struct list_head *reaplist)
1319 {
1320         struct nfs4_ol_stateid *stp;
1321
1322         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1323
1324         while (!list_empty(&open_stp->st_locks)) {
1325                 stp = list_entry(open_stp->st_locks.next,
1326                                 struct nfs4_ol_stateid, st_locks);
1327                 WARN_ON(!unhash_lock_stateid(stp));
1328                 put_ol_stateid_locked(stp, reaplist);
1329         }
1330 }
1331
1332 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1333                                 struct list_head *reaplist)
1334 {
1335         bool unhashed;
1336
1337         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1338
1339         unhashed = unhash_ol_stateid(stp);
1340         release_open_stateid_locks(stp, reaplist);
1341         return unhashed;
1342 }
1343
1344 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1345 {
1346         LIST_HEAD(reaplist);
1347
1348         spin_lock(&stp->st_stid.sc_client->cl_lock);
1349         if (unhash_open_stateid(stp, &reaplist))
1350                 put_ol_stateid_locked(stp, &reaplist);
1351         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1352         free_ol_stateid_reaplist(&reaplist);
1353 }
1354
1355 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1356 {
1357         struct nfs4_client *clp = oo->oo_owner.so_client;
1358
1359         lockdep_assert_held(&clp->cl_lock);
1360
1361         list_del_init(&oo->oo_owner.so_strhash);
1362         list_del_init(&oo->oo_perclient);
1363 }
1364
1365 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1366 {
1367         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1368                                           nfsd_net_id);
1369         struct nfs4_ol_stateid *s;
1370
1371         spin_lock(&nn->client_lock);
1372         s = oo->oo_last_closed_stid;
1373         if (s) {
1374                 list_del_init(&oo->oo_close_lru);
1375                 oo->oo_last_closed_stid = NULL;
1376         }
1377         spin_unlock(&nn->client_lock);
1378         if (s)
1379                 nfs4_put_stid(&s->st_stid);
1380 }
1381
1382 static void release_openowner(struct nfs4_openowner *oo)
1383 {
1384         struct nfs4_ol_stateid *stp;
1385         struct nfs4_client *clp = oo->oo_owner.so_client;
1386         struct list_head reaplist;
1387
1388         INIT_LIST_HEAD(&reaplist);
1389
1390         spin_lock(&clp->cl_lock);
1391         unhash_openowner_locked(oo);
1392         while (!list_empty(&oo->oo_owner.so_stateids)) {
1393                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1394                                 struct nfs4_ol_stateid, st_perstateowner);
1395                 if (unhash_open_stateid(stp, &reaplist))
1396                         put_ol_stateid_locked(stp, &reaplist);
1397         }
1398         spin_unlock(&clp->cl_lock);
1399         free_ol_stateid_reaplist(&reaplist);
1400         release_last_closed_stateid(oo);
1401         nfs4_put_stateowner(&oo->oo_owner);
1402 }
1403
1404 static inline int
1405 hash_sessionid(struct nfs4_sessionid *sessionid)
1406 {
1407         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1408
1409         return sid->sequence % SESSION_HASH_SIZE;
1410 }
1411
1412 #ifdef CONFIG_SUNRPC_DEBUG
1413 static inline void
1414 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1415 {
1416         u32 *ptr = (u32 *)(&sessionid->data[0]);
1417         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1418 }
1419 #else
1420 static inline void
1421 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1422 {
1423 }
1424 #endif
1425
1426 /*
1427  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1428  * won't be used for replay.
1429  */
1430 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1431 {
1432         struct nfs4_stateowner *so = cstate->replay_owner;
1433
1434         if (nfserr == nfserr_replay_me)
1435                 return;
1436
1437         if (!seqid_mutating_err(ntohl(nfserr))) {
1438                 nfsd4_cstate_clear_replay(cstate);
1439                 return;
1440         }
1441         if (!so)
1442                 return;
1443         if (so->so_is_open_owner)
1444                 release_last_closed_stateid(openowner(so));
1445         so->so_seqid++;
1446         return;
1447 }
1448
1449 static void
1450 gen_sessionid(struct nfsd4_session *ses)
1451 {
1452         struct nfs4_client *clp = ses->se_client;
1453         struct nfsd4_sessionid *sid;
1454
1455         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1456         sid->clientid = clp->cl_clientid;
1457         sid->sequence = current_sessionid++;
1458         sid->reserved = 0;
1459 }
1460
1461 /*
1462  * The protocol defines ca_maxresponssize_cached to include the size of
1463  * the rpc header, but all we need to cache is the data starting after
1464  * the end of the initial SEQUENCE operation--the rest we regenerate
1465  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1466  * value that is the number of bytes in our cache plus a few additional
1467  * bytes.  In order to stay on the safe side, and not promise more than
1468  * we can cache, those additional bytes must be the minimum possible: 24
1469  * bytes of rpc header (xid through accept state, with AUTH_NULL
1470  * verifier), 12 for the compound header (with zero-length tag), and 44
1471  * for the SEQUENCE op response:
1472  */
1473 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1474
1475 static void
1476 free_session_slots(struct nfsd4_session *ses)
1477 {
1478         int i;
1479
1480         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1481                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1482                 kfree(ses->se_slots[i]);
1483         }
1484 }
1485
1486 /*
1487  * We don't actually need to cache the rpc and session headers, so we
1488  * can allocate a little less for each slot:
1489  */
1490 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1491 {
1492         u32 size;
1493
1494         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1495                 size = 0;
1496         else
1497                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1498         return size + sizeof(struct nfsd4_slot);
1499 }
1500
1501 /*
1502  * XXX: If we run out of reserved DRC memory we could (up to a point)
1503  * re-negotiate active sessions and reduce their slot usage to make
1504  * room for new connections. For now we just fail the create session.
1505  */
1506 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1507 {
1508         u32 slotsize = slot_bytes(ca);
1509         u32 num = ca->maxreqs;
1510         int avail;
1511
1512         spin_lock(&nfsd_drc_lock);
1513         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1514                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1515         /*
1516          * Never use more than a third of the remaining memory,
1517          * unless it's the only way to give this client a slot:
1518          */
1519         avail = clamp_t(int, avail, slotsize, avail/3);
1520         num = min_t(int, num, avail / slotsize);
1521         nfsd_drc_mem_used += num * slotsize;
1522         spin_unlock(&nfsd_drc_lock);
1523
1524         return num;
1525 }
1526
1527 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1528 {
1529         int slotsize = slot_bytes(ca);
1530
1531         spin_lock(&nfsd_drc_lock);
1532         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1533         spin_unlock(&nfsd_drc_lock);
1534 }
1535
1536 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1537                                            struct nfsd4_channel_attrs *battrs)
1538 {
1539         int numslots = fattrs->maxreqs;
1540         int slotsize = slot_bytes(fattrs);
1541         struct nfsd4_session *new;
1542         int mem, i;
1543
1544         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1545                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1546         mem = numslots * sizeof(struct nfsd4_slot *);
1547
1548         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1549         if (!new)
1550                 return NULL;
1551         /* allocate each struct nfsd4_slot and data cache in one piece */
1552         for (i = 0; i < numslots; i++) {
1553                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1554                 if (!new->se_slots[i])
1555                         goto out_free;
1556         }
1557
1558         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1559         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1560
1561         return new;
1562 out_free:
1563         while (i--)
1564                 kfree(new->se_slots[i]);
1565         kfree(new);
1566         return NULL;
1567 }
1568
1569 static void free_conn(struct nfsd4_conn *c)
1570 {
1571         svc_xprt_put(c->cn_xprt);
1572         kfree(c);
1573 }
1574
1575 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1576 {
1577         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1578         struct nfs4_client *clp = c->cn_session->se_client;
1579
1580         spin_lock(&clp->cl_lock);
1581         if (!list_empty(&c->cn_persession)) {
1582                 list_del(&c->cn_persession);
1583                 free_conn(c);
1584         }
1585         nfsd4_probe_callback(clp);
1586         spin_unlock(&clp->cl_lock);
1587 }
1588
1589 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1590 {
1591         struct nfsd4_conn *conn;
1592
1593         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1594         if (!conn)
1595                 return NULL;
1596         svc_xprt_get(rqstp->rq_xprt);
1597         conn->cn_xprt = rqstp->rq_xprt;
1598         conn->cn_flags = flags;
1599         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1600         return conn;
1601 }
1602
1603 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1604 {
1605         conn->cn_session = ses;
1606         list_add(&conn->cn_persession, &ses->se_conns);
1607 }
1608
1609 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1610 {
1611         struct nfs4_client *clp = ses->se_client;
1612
1613         spin_lock(&clp->cl_lock);
1614         __nfsd4_hash_conn(conn, ses);
1615         spin_unlock(&clp->cl_lock);
1616 }
1617
1618 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1619 {
1620         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1621         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1622 }
1623
1624 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1625 {
1626         int ret;
1627
1628         nfsd4_hash_conn(conn, ses);
1629         ret = nfsd4_register_conn(conn);
1630         if (ret)
1631                 /* oops; xprt is already down: */
1632                 nfsd4_conn_lost(&conn->cn_xpt_user);
1633         /* We may have gained or lost a callback channel: */
1634         nfsd4_probe_callback_sync(ses->se_client);
1635 }
1636
1637 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1638 {
1639         u32 dir = NFS4_CDFC4_FORE;
1640
1641         if (cses->flags & SESSION4_BACK_CHAN)
1642                 dir |= NFS4_CDFC4_BACK;
1643         return alloc_conn(rqstp, dir);
1644 }
1645
1646 /* must be called under client_lock */
1647 static void nfsd4_del_conns(struct nfsd4_session *s)
1648 {
1649         struct nfs4_client *clp = s->se_client;
1650         struct nfsd4_conn *c;
1651
1652         spin_lock(&clp->cl_lock);
1653         while (!list_empty(&s->se_conns)) {
1654                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1655                 list_del_init(&c->cn_persession);
1656                 spin_unlock(&clp->cl_lock);
1657
1658                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1659                 free_conn(c);
1660
1661                 spin_lock(&clp->cl_lock);
1662         }
1663         spin_unlock(&clp->cl_lock);
1664 }
1665
1666 static void __free_session(struct nfsd4_session *ses)
1667 {
1668         free_session_slots(ses);
1669         kfree(ses);
1670 }
1671
1672 static void free_session(struct nfsd4_session *ses)
1673 {
1674         nfsd4_del_conns(ses);
1675         nfsd4_put_drc_mem(&ses->se_fchannel);
1676         __free_session(ses);
1677 }
1678
1679 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1680 {
1681         int idx;
1682         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1683
1684         new->se_client = clp;
1685         gen_sessionid(new);
1686
1687         INIT_LIST_HEAD(&new->se_conns);
1688
1689         new->se_cb_seq_nr = 1;
1690         new->se_flags = cses->flags;
1691         new->se_cb_prog = cses->callback_prog;
1692         new->se_cb_sec = cses->cb_sec;
1693         atomic_set(&new->se_ref, 0);
1694         idx = hash_sessionid(&new->se_sessionid);
1695         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1696         spin_lock(&clp->cl_lock);
1697         list_add(&new->se_perclnt, &clp->cl_sessions);
1698         spin_unlock(&clp->cl_lock);
1699
1700         {
1701                 struct sockaddr *sa = svc_addr(rqstp);
1702                 /*
1703                  * This is a little silly; with sessions there's no real
1704                  * use for the callback address.  Use the peer address
1705                  * as a reasonable default for now, but consider fixing
1706                  * the rpc client not to require an address in the
1707                  * future:
1708                  */
1709                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1710                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1711         }
1712 }
1713
1714 /* caller must hold client_lock */
1715 static struct nfsd4_session *
1716 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1717 {
1718         struct nfsd4_session *elem;
1719         int idx;
1720         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1721
1722         lockdep_assert_held(&nn->client_lock);
1723
1724         dump_sessionid(__func__, sessionid);
1725         idx = hash_sessionid(sessionid);
1726         /* Search in the appropriate list */
1727         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1728                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1729                             NFS4_MAX_SESSIONID_LEN)) {
1730                         return elem;
1731                 }
1732         }
1733
1734         dprintk("%s: session not found\n", __func__);
1735         return NULL;
1736 }
1737
1738 static struct nfsd4_session *
1739 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1740                 __be32 *ret)
1741 {
1742         struct nfsd4_session *session;
1743         __be32 status = nfserr_badsession;
1744
1745         session = __find_in_sessionid_hashtbl(sessionid, net);
1746         if (!session)
1747                 goto out;
1748         status = nfsd4_get_session_locked(session);
1749         if (status)
1750                 session = NULL;
1751 out:
1752         *ret = status;
1753         return session;
1754 }
1755
1756 /* caller must hold client_lock */
1757 static void
1758 unhash_session(struct nfsd4_session *ses)
1759 {
1760         struct nfs4_client *clp = ses->se_client;
1761         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1762
1763         lockdep_assert_held(&nn->client_lock);
1764
1765         list_del(&ses->se_hash);
1766         spin_lock(&ses->se_client->cl_lock);
1767         list_del(&ses->se_perclnt);
1768         spin_unlock(&ses->se_client->cl_lock);
1769 }
1770
1771 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1772 static int
1773 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1774 {
1775         /*
1776          * We're assuming the clid was not given out from a boot
1777          * precisely 2^32 (about 136 years) before this one.  That seems
1778          * a safe assumption:
1779          */
1780         if (clid->cl_boot == (u32)nn->boot_time)
1781                 return 0;
1782         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1783                 clid->cl_boot, clid->cl_id, nn->boot_time);
1784         return 1;
1785 }
1786
1787 /* 
1788  * XXX Should we use a slab cache ?
1789  * This type of memory management is somewhat inefficient, but we use it
1790  * anyway since SETCLIENTID is not a common operation.
1791  */
1792 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1793 {
1794         struct nfs4_client *clp;
1795         int i;
1796
1797         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1798         if (clp == NULL)
1799                 return NULL;
1800         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1801         if (clp->cl_name.data == NULL)
1802                 goto err_no_name;
1803         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1804                         OWNER_HASH_SIZE, GFP_KERNEL);
1805         if (!clp->cl_ownerstr_hashtbl)
1806                 goto err_no_hashtbl;
1807         for (i = 0; i < OWNER_HASH_SIZE; i++)
1808                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1809         clp->cl_name.len = name.len;
1810         INIT_LIST_HEAD(&clp->cl_sessions);
1811         idr_init(&clp->cl_stateids);
1812         atomic_set(&clp->cl_refcount, 0);
1813         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1814         INIT_LIST_HEAD(&clp->cl_idhash);
1815         INIT_LIST_HEAD(&clp->cl_openowners);
1816         INIT_LIST_HEAD(&clp->cl_delegations);
1817         INIT_LIST_HEAD(&clp->cl_lru);
1818         INIT_LIST_HEAD(&clp->cl_revoked);
1819 #ifdef CONFIG_NFSD_PNFS
1820         INIT_LIST_HEAD(&clp->cl_lo_states);
1821 #endif
1822         spin_lock_init(&clp->cl_lock);
1823         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1824         return clp;
1825 err_no_hashtbl:
1826         kfree(clp->cl_name.data);
1827 err_no_name:
1828         kfree(clp);
1829         return NULL;
1830 }
1831
1832 static void
1833 free_client(struct nfs4_client *clp)
1834 {
1835         while (!list_empty(&clp->cl_sessions)) {
1836                 struct nfsd4_session *ses;
1837                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1838                                 se_perclnt);
1839                 list_del(&ses->se_perclnt);
1840                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1841                 free_session(ses);
1842         }
1843         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1844         free_svc_cred(&clp->cl_cred);
1845         kfree(clp->cl_ownerstr_hashtbl);
1846         kfree(clp->cl_name.data);
1847         idr_destroy(&clp->cl_stateids);
1848         kfree(clp);
1849 }
1850
1851 /* must be called under the client_lock */
1852 static void
1853 unhash_client_locked(struct nfs4_client *clp)
1854 {
1855         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1856         struct nfsd4_session *ses;
1857
1858         lockdep_assert_held(&nn->client_lock);
1859
1860         /* Mark the client as expired! */
1861         clp->cl_time = 0;
1862         /* Make it invisible */
1863         if (!list_empty(&clp->cl_idhash)) {
1864                 list_del_init(&clp->cl_idhash);
1865                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1866                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1867                 else
1868                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1869         }
1870         list_del_init(&clp->cl_lru);
1871         spin_lock(&clp->cl_lock);
1872         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1873                 list_del_init(&ses->se_hash);
1874         spin_unlock(&clp->cl_lock);
1875 }
1876
1877 static void
1878 unhash_client(struct nfs4_client *clp)
1879 {
1880         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1881
1882         spin_lock(&nn->client_lock);
1883         unhash_client_locked(clp);
1884         spin_unlock(&nn->client_lock);
1885 }
1886
1887 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1888 {
1889         if (atomic_read(&clp->cl_refcount))
1890                 return nfserr_jukebox;
1891         unhash_client_locked(clp);
1892         return nfs_ok;
1893 }
1894
1895 static void
1896 __destroy_client(struct nfs4_client *clp)
1897 {
1898         int i;
1899         struct nfs4_openowner *oo;
1900         struct nfs4_delegation *dp;
1901         struct list_head reaplist;
1902
1903         INIT_LIST_HEAD(&reaplist);
1904         spin_lock(&state_lock);
1905         while (!list_empty(&clp->cl_delegations)) {
1906                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1907                 WARN_ON(!unhash_delegation_locked(dp));
1908                 list_add(&dp->dl_recall_lru, &reaplist);
1909         }
1910         spin_unlock(&state_lock);
1911         while (!list_empty(&reaplist)) {
1912                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1913                 list_del_init(&dp->dl_recall_lru);
1914                 put_clnt_odstate(dp->dl_clnt_odstate);
1915                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1916                 nfs4_put_stid(&dp->dl_stid);
1917         }
1918         while (!list_empty(&clp->cl_revoked)) {
1919                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1920                 list_del_init(&dp->dl_recall_lru);
1921                 nfs4_put_stid(&dp->dl_stid);
1922         }
1923         while (!list_empty(&clp->cl_openowners)) {
1924                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1925                 nfs4_get_stateowner(&oo->oo_owner);
1926                 release_openowner(oo);
1927         }
1928         for (i = 0; i < OWNER_HASH_SIZE; i++) {
1929                 struct nfs4_stateowner *so, *tmp;
1930
1931                 list_for_each_entry_safe(so, tmp, &clp->cl_ownerstr_hashtbl[i],
1932                                          so_strhash) {
1933                         /* Should be no openowners at this point */
1934                         WARN_ON_ONCE(so->so_is_open_owner);
1935                         remove_blocked_locks(lockowner(so));
1936                 }
1937         }
1938         nfsd4_return_all_client_layouts(clp);
1939         nfsd4_shutdown_callback(clp);
1940         if (clp->cl_cb_conn.cb_xprt)
1941                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1942         free_client(clp);
1943 }
1944
1945 static void
1946 destroy_client(struct nfs4_client *clp)
1947 {
1948         unhash_client(clp);
1949         __destroy_client(clp);
1950 }
1951
1952 static void expire_client(struct nfs4_client *clp)
1953 {
1954         unhash_client(clp);
1955         nfsd4_client_record_remove(clp);
1956         __destroy_client(clp);
1957 }
1958
1959 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1960 {
1961         memcpy(target->cl_verifier.data, source->data,
1962                         sizeof(target->cl_verifier.data));
1963 }
1964
1965 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1966 {
1967         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1968         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1969 }
1970
1971 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1972 {
1973         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1974         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1975                                                                 GFP_KERNEL);
1976         if ((source->cr_principal && ! target->cr_principal) ||
1977             (source->cr_raw_principal && ! target->cr_raw_principal))
1978                 return -ENOMEM;
1979
1980         target->cr_flavor = source->cr_flavor;
1981         target->cr_uid = source->cr_uid;
1982         target->cr_gid = source->cr_gid;
1983         target->cr_group_info = source->cr_group_info;
1984         get_group_info(target->cr_group_info);
1985         target->cr_gss_mech = source->cr_gss_mech;
1986         if (source->cr_gss_mech)
1987                 gss_mech_get(source->cr_gss_mech);
1988         return 0;
1989 }
1990
1991 static int
1992 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1993 {
1994         if (o1->len < o2->len)
1995                 return -1;
1996         if (o1->len > o2->len)
1997                 return 1;
1998         return memcmp(o1->data, o2->data, o1->len);
1999 }
2000
2001 static int same_name(const char *n1, const char *n2)
2002 {
2003         return 0 == memcmp(n1, n2, HEXDIR_LEN);
2004 }
2005
2006 static int
2007 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
2008 {
2009         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
2010 }
2011
2012 static int
2013 same_clid(clientid_t *cl1, clientid_t *cl2)
2014 {
2015         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
2016 }
2017
2018 static bool groups_equal(struct group_info *g1, struct group_info *g2)
2019 {
2020         int i;
2021
2022         if (g1->ngroups != g2->ngroups)
2023                 return false;
2024         for (i=0; i<g1->ngroups; i++)
2025                 if (!gid_eq(g1->gid[i], g2->gid[i]))
2026                         return false;
2027         return true;
2028 }
2029
2030 /*
2031  * RFC 3530 language requires clid_inuse be returned when the
2032  * "principal" associated with a requests differs from that previously
2033  * used.  We use uid, gid's, and gss principal string as our best
2034  * approximation.  We also don't want to allow non-gss use of a client
2035  * established using gss: in theory cr_principal should catch that
2036  * change, but in practice cr_principal can be null even in the gss case
2037  * since gssd doesn't always pass down a principal string.
2038  */
2039 static bool is_gss_cred(struct svc_cred *cr)
2040 {
2041         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2042         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2043 }
2044
2045
2046 static bool
2047 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2048 {
2049         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2050                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2051                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2052                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2053                 return false;
2054         if (cr1->cr_principal == cr2->cr_principal)
2055                 return true;
2056         if (!cr1->cr_principal || !cr2->cr_principal)
2057                 return false;
2058         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2059 }
2060
2061 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2062 {
2063         struct svc_cred *cr = &rqstp->rq_cred;
2064         u32 service;
2065
2066         if (!cr->cr_gss_mech)
2067                 return false;
2068         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2069         return service == RPC_GSS_SVC_INTEGRITY ||
2070                service == RPC_GSS_SVC_PRIVACY;
2071 }
2072
2073 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2074 {
2075         struct svc_cred *cr = &rqstp->rq_cred;
2076
2077         if (!cl->cl_mach_cred)
2078                 return true;
2079         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2080                 return false;
2081         if (!svc_rqst_integrity_protected(rqstp))
2082                 return false;
2083         if (cl->cl_cred.cr_raw_principal)
2084                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2085                                                 cr->cr_raw_principal);
2086         if (!cr->cr_principal)
2087                 return false;
2088         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2089 }
2090
2091 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2092 {
2093         __be32 verf[2];
2094
2095         /*
2096          * This is opaque to client, so no need to byte-swap. Use
2097          * __force to keep sparse happy
2098          */
2099         verf[0] = (__force __be32)get_seconds();
2100         verf[1] = (__force __be32)nn->clverifier_counter++;
2101         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2102 }
2103
2104 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2105 {
2106         clp->cl_clientid.cl_boot = nn->boot_time;
2107         clp->cl_clientid.cl_id = nn->clientid_counter++;
2108         gen_confirm(clp, nn);
2109 }
2110
2111 static struct nfs4_stid *
2112 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2113 {
2114         struct nfs4_stid *ret;
2115
2116         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2117         if (!ret || !ret->sc_type)
2118                 return NULL;
2119         return ret;
2120 }
2121
2122 static struct nfs4_stid *
2123 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2124 {
2125         struct nfs4_stid *s;
2126
2127         spin_lock(&cl->cl_lock);
2128         s = find_stateid_locked(cl, t);
2129         if (s != NULL) {
2130                 if (typemask & s->sc_type)
2131                         refcount_inc(&s->sc_count);
2132                 else
2133                         s = NULL;
2134         }
2135         spin_unlock(&cl->cl_lock);
2136         return s;
2137 }
2138
2139 static struct nfs4_client *create_client(struct xdr_netobj name,
2140                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2141 {
2142         struct nfs4_client *clp;
2143         struct sockaddr *sa = svc_addr(rqstp);
2144         int ret;
2145         struct net *net = SVC_NET(rqstp);
2146
2147         clp = alloc_client(name);
2148         if (clp == NULL)
2149                 return NULL;
2150
2151         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2152         if (ret) {
2153                 free_client(clp);
2154                 return NULL;
2155         }
2156         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2157         clp->cl_time = get_seconds();
2158         clear_bit(0, &clp->cl_cb_slot_busy);
2159         copy_verf(clp, verf);
2160         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2161         clp->cl_cb_session = NULL;
2162         clp->net = net;
2163         return clp;
2164 }
2165
2166 static void
2167 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2168 {
2169         struct rb_node **new = &(root->rb_node), *parent = NULL;
2170         struct nfs4_client *clp;
2171
2172         while (*new) {
2173                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2174                 parent = *new;
2175
2176                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2177                         new = &((*new)->rb_left);
2178                 else
2179                         new = &((*new)->rb_right);
2180         }
2181
2182         rb_link_node(&new_clp->cl_namenode, parent, new);
2183         rb_insert_color(&new_clp->cl_namenode, root);
2184 }
2185
2186 static struct nfs4_client *
2187 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2188 {
2189         int cmp;
2190         struct rb_node *node = root->rb_node;
2191         struct nfs4_client *clp;
2192
2193         while (node) {
2194                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2195                 cmp = compare_blob(&clp->cl_name, name);
2196                 if (cmp > 0)
2197                         node = node->rb_left;
2198                 else if (cmp < 0)
2199                         node = node->rb_right;
2200                 else
2201                         return clp;
2202         }
2203         return NULL;
2204 }
2205
2206 static void
2207 add_to_unconfirmed(struct nfs4_client *clp)
2208 {
2209         unsigned int idhashval;
2210         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2211
2212         lockdep_assert_held(&nn->client_lock);
2213
2214         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2215         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2216         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2217         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2218         renew_client_locked(clp);
2219 }
2220
2221 static void
2222 move_to_confirmed(struct nfs4_client *clp)
2223 {
2224         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2225         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2226
2227         lockdep_assert_held(&nn->client_lock);
2228
2229         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2230         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2231         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2232         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2233         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2234         renew_client_locked(clp);
2235 }
2236
2237 static struct nfs4_client *
2238 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2239 {
2240         struct nfs4_client *clp;
2241         unsigned int idhashval = clientid_hashval(clid->cl_id);
2242
2243         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2244                 if (same_clid(&clp->cl_clientid, clid)) {
2245                         if ((bool)clp->cl_minorversion != sessions)
2246                                 return NULL;
2247                         renew_client_locked(clp);
2248                         return clp;
2249                 }
2250         }
2251         return NULL;
2252 }
2253
2254 static struct nfs4_client *
2255 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2256 {
2257         struct list_head *tbl = nn->conf_id_hashtbl;
2258
2259         lockdep_assert_held(&nn->client_lock);
2260         return find_client_in_id_table(tbl, clid, sessions);
2261 }
2262
2263 static struct nfs4_client *
2264 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2265 {
2266         struct list_head *tbl = nn->unconf_id_hashtbl;
2267
2268         lockdep_assert_held(&nn->client_lock);
2269         return find_client_in_id_table(tbl, clid, sessions);
2270 }
2271
2272 static bool clp_used_exchangeid(struct nfs4_client *clp)
2273 {
2274         return clp->cl_exchange_flags != 0;
2275
2276
2277 static struct nfs4_client *
2278 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2279 {
2280         lockdep_assert_held(&nn->client_lock);
2281         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2282 }
2283
2284 static struct nfs4_client *
2285 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2286 {
2287         lockdep_assert_held(&nn->client_lock);
2288         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2289 }
2290
2291 static void
2292 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2293 {
2294         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2295         struct sockaddr *sa = svc_addr(rqstp);
2296         u32 scopeid = rpc_get_scope_id(sa);
2297         unsigned short expected_family;
2298
2299         /* Currently, we only support tcp and tcp6 for the callback channel */
2300         if (se->se_callback_netid_len == 3 &&
2301             !memcmp(se->se_callback_netid_val, "tcp", 3))
2302                 expected_family = AF_INET;
2303         else if (se->se_callback_netid_len == 4 &&
2304                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2305                 expected_family = AF_INET6;
2306         else
2307                 goto out_err;
2308
2309         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2310                                             se->se_callback_addr_len,
2311                                             (struct sockaddr *)&conn->cb_addr,
2312                                             sizeof(conn->cb_addr));
2313
2314         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2315                 goto out_err;
2316
2317         if (conn->cb_addr.ss_family == AF_INET6)
2318                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2319
2320         conn->cb_prog = se->se_callback_prog;
2321         conn->cb_ident = se->se_callback_ident;
2322         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2323         return;
2324 out_err:
2325         conn->cb_addr.ss_family = AF_UNSPEC;
2326         conn->cb_addrlen = 0;
2327         dprintk("NFSD: this client (clientid %08x/%08x) "
2328                 "will not receive delegations\n",
2329                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2330
2331         return;
2332 }
2333
2334 /*
2335  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2336  */
2337 static void
2338 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2339 {
2340         struct xdr_buf *buf = resp->xdr.buf;
2341         struct nfsd4_slot *slot = resp->cstate.slot;
2342         unsigned int base;
2343
2344         dprintk("--> %s slot %p\n", __func__, slot);
2345
2346         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2347         slot->sl_opcnt = resp->opcnt;
2348         slot->sl_status = resp->cstate.status;
2349         free_svc_cred(&slot->sl_cred);
2350         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2351
2352         if (!nfsd4_cache_this(resp)) {
2353                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2354                 return;
2355         }
2356         slot->sl_flags |= NFSD4_SLOT_CACHED;
2357
2358         base = resp->cstate.data_offset;
2359         slot->sl_datalen = buf->len - base;
2360         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2361                 WARN(1, "%s: sessions DRC could not cache compound\n",
2362                      __func__);
2363         return;
2364 }
2365
2366 /*
2367  * Encode the replay sequence operation from the slot values.
2368  * If cachethis is FALSE encode the uncached rep error on the next
2369  * operation which sets resp->p and increments resp->opcnt for
2370  * nfs4svc_encode_compoundres.
2371  *
2372  */
2373 static __be32
2374 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2375                           struct nfsd4_compoundres *resp)
2376 {
2377         struct nfsd4_op *op;
2378         struct nfsd4_slot *slot = resp->cstate.slot;
2379
2380         /* Encode the replayed sequence operation */
2381         op = &args->ops[resp->opcnt - 1];
2382         nfsd4_encode_operation(resp, op);
2383
2384         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2385                 return op->status;
2386         if (args->opcnt == 1) {
2387                 /*
2388                  * The original operation wasn't a solo sequence--we
2389                  * always cache those--so this retry must not match the
2390                  * original:
2391                  */
2392                 op->status = nfserr_seq_false_retry;
2393         } else {
2394                 op = &args->ops[resp->opcnt++];
2395                 op->status = nfserr_retry_uncached_rep;
2396                 nfsd4_encode_operation(resp, op);
2397         }
2398         return op->status;
2399 }
2400
2401 /*
2402  * The sequence operation is not cached because we can use the slot and
2403  * session values.
2404  */
2405 static __be32
2406 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2407                          struct nfsd4_sequence *seq)
2408 {
2409         struct nfsd4_slot *slot = resp->cstate.slot;
2410         struct xdr_stream *xdr = &resp->xdr;
2411         __be32 *p;
2412         __be32 status;
2413
2414         dprintk("--> %s slot %p\n", __func__, slot);
2415
2416         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2417         if (status)
2418                 return status;
2419
2420         p = xdr_reserve_space(xdr, slot->sl_datalen);
2421         if (!p) {
2422                 WARN_ON_ONCE(1);
2423                 return nfserr_serverfault;
2424         }
2425         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2426         xdr_commit_encode(xdr);
2427
2428         resp->opcnt = slot->sl_opcnt;
2429         return slot->sl_status;
2430 }
2431
2432 /*
2433  * Set the exchange_id flags returned by the server.
2434  */
2435 static void
2436 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2437 {
2438 #ifdef CONFIG_NFSD_PNFS
2439         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2440 #else
2441         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2442 #endif
2443
2444         /* Referrals are supported, Migration is not. */
2445         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2446
2447         /* set the wire flags to return to client. */
2448         clid->flags = new->cl_exchange_flags;
2449 }
2450
2451 static bool client_has_openowners(struct nfs4_client *clp)
2452 {
2453         struct nfs4_openowner *oo;
2454
2455         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2456                 if (!list_empty(&oo->oo_owner.so_stateids))
2457                         return true;
2458         }
2459         return false;
2460 }
2461
2462 static bool client_has_state(struct nfs4_client *clp)
2463 {
2464         return client_has_openowners(clp)
2465 #ifdef CONFIG_NFSD_PNFS
2466                 || !list_empty(&clp->cl_lo_states)
2467 #endif
2468                 || !list_empty(&clp->cl_delegations)
2469                 || !list_empty(&clp->cl_sessions);
2470 }
2471
2472 __be32
2473 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2474                 union nfsd4_op_u *u)
2475 {
2476         struct nfsd4_exchange_id *exid = &u->exchange_id;
2477         struct nfs4_client *conf, *new;
2478         struct nfs4_client *unconf = NULL;
2479         __be32 status;
2480         char                    addr_str[INET6_ADDRSTRLEN];
2481         nfs4_verifier           verf = exid->verifier;
2482         struct sockaddr         *sa = svc_addr(rqstp);
2483         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2484         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2485
2486         rpc_ntop(sa, addr_str, sizeof(addr_str));
2487         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2488                 "ip_addr=%s flags %x, spa_how %d\n",
2489                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2490                 addr_str, exid->flags, exid->spa_how);
2491
2492         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2493                 return nfserr_inval;
2494
2495         new = create_client(exid->clname, rqstp, &verf);
2496         if (new == NULL)
2497                 return nfserr_jukebox;
2498
2499         switch (exid->spa_how) {
2500         case SP4_MACH_CRED:
2501                 exid->spo_must_enforce[0] = 0;
2502                 exid->spo_must_enforce[1] = (
2503                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2504                         1 << (OP_EXCHANGE_ID - 32) |
2505                         1 << (OP_CREATE_SESSION - 32) |
2506                         1 << (OP_DESTROY_SESSION - 32) |
2507                         1 << (OP_DESTROY_CLIENTID - 32));
2508
2509                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2510                                         1 << (OP_OPEN_DOWNGRADE) |
2511                                         1 << (OP_LOCKU) |
2512                                         1 << (OP_DELEGRETURN));
2513
2514                 exid->spo_must_allow[1] &= (
2515                                         1 << (OP_TEST_STATEID - 32) |
2516                                         1 << (OP_FREE_STATEID - 32));
2517                 if (!svc_rqst_integrity_protected(rqstp)) {
2518                         status = nfserr_inval;
2519                         goto out_nolock;
2520                 }
2521                 /*
2522                  * Sometimes userspace doesn't give us a principal.
2523                  * Which is a bug, really.  Anyway, we can't enforce
2524                  * MACH_CRED in that case, better to give up now:
2525                  */
2526                 if (!new->cl_cred.cr_principal &&
2527                                         !new->cl_cred.cr_raw_principal) {
2528                         status = nfserr_serverfault;
2529                         goto out_nolock;
2530                 }
2531                 new->cl_mach_cred = true;
2532         case SP4_NONE:
2533                 break;
2534         default:                                /* checked by xdr code */
2535                 WARN_ON_ONCE(1);
2536         case SP4_SSV:
2537                 status = nfserr_encr_alg_unsupp;
2538                 goto out_nolock;
2539         }
2540
2541         /* Cases below refer to rfc 5661 section 18.35.4: */
2542         spin_lock(&nn->client_lock);
2543         conf = find_confirmed_client_by_name(&exid->clname, nn);
2544         if (conf) {
2545                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2546                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2547
2548                 if (update) {
2549                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2550                                 status = nfserr_inval;
2551                                 goto out;
2552                         }
2553                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2554                                 status = nfserr_wrong_cred;
2555                                 goto out;
2556                         }
2557                         if (!creds_match) { /* case 9 */
2558                                 status = nfserr_perm;
2559                                 goto out;
2560                         }
2561                         if (!verfs_match) { /* case 8 */
2562                                 status = nfserr_not_same;
2563                                 goto out;
2564                         }
2565                         /* case 6 */
2566                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2567                         goto out_copy;
2568                 }
2569                 if (!creds_match) { /* case 3 */
2570                         if (client_has_state(conf)) {
2571                                 status = nfserr_clid_inuse;
2572                                 goto out;
2573                         }
2574                         goto out_new;
2575                 }
2576                 if (verfs_match) { /* case 2 */
2577                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2578                         goto out_copy;
2579                 }
2580                 /* case 5, client reboot */
2581                 conf = NULL;
2582                 goto out_new;
2583         }
2584
2585         if (update) { /* case 7 */
2586                 status = nfserr_noent;
2587                 goto out;
2588         }
2589
2590         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2591         if (unconf) /* case 4, possible retry or client restart */
2592                 unhash_client_locked(unconf);
2593
2594         /* case 1 (normal case) */
2595 out_new:
2596         if (conf) {
2597                 status = mark_client_expired_locked(conf);
2598                 if (status)
2599                         goto out;
2600         }
2601         new->cl_minorversion = cstate->minorversion;
2602         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2603         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2604
2605         gen_clid(new, nn);
2606         add_to_unconfirmed(new);
2607         swap(new, conf);
2608 out_copy:
2609         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2610         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2611
2612         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2613         nfsd4_set_ex_flags(conf, exid);
2614
2615         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2616                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2617         status = nfs_ok;
2618
2619 out:
2620         spin_unlock(&nn->client_lock);
2621 out_nolock:
2622         if (new)
2623                 expire_client(new);
2624         if (unconf)
2625                 expire_client(unconf);
2626         return status;
2627 }
2628
2629 static __be32
2630 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2631 {
2632         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2633                 slot_seqid);
2634
2635         /* The slot is in use, and no response has been sent. */
2636         if (slot_inuse) {
2637                 if (seqid == slot_seqid)
2638                         return nfserr_jukebox;
2639                 else
2640                         return nfserr_seq_misordered;
2641         }
2642         /* Note unsigned 32-bit arithmetic handles wraparound: */
2643         if (likely(seqid == slot_seqid + 1))
2644                 return nfs_ok;
2645         if (seqid == slot_seqid)
2646                 return nfserr_replay_cache;
2647         return nfserr_seq_misordered;
2648 }
2649
2650 /*
2651  * Cache the create session result into the create session single DRC
2652  * slot cache by saving the xdr structure. sl_seqid has been set.
2653  * Do this for solo or embedded create session operations.
2654  */
2655 static void
2656 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2657                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2658 {
2659         slot->sl_status = nfserr;
2660         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2661 }
2662
2663 static __be32
2664 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2665                             struct nfsd4_clid_slot *slot)
2666 {
2667         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2668         return slot->sl_status;
2669 }
2670
2671 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2672                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2673                         1 +     /* MIN tag is length with zero, only length */ \
2674                         3 +     /* version, opcount, opcode */ \
2675                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2676                                 /* seqid, slotID, slotID, cache */ \
2677                         4 ) * sizeof(__be32))
2678
2679 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2680                         2 +     /* verifier: AUTH_NULL, length 0 */\
2681                         1 +     /* status */ \
2682                         1 +     /* MIN tag is length with zero, only length */ \
2683                         3 +     /* opcount, opcode, opstatus*/ \
2684                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2685                                 /* seqid, slotID, slotID, slotID, status */ \
2686                         5 ) * sizeof(__be32))
2687
2688 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2689 {
2690         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2691
2692         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2693                 return nfserr_toosmall;
2694         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2695                 return nfserr_toosmall;
2696         ca->headerpadsz = 0;
2697         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2698         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2699         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2700         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2701                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2702         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2703         /*
2704          * Note decreasing slot size below client's request may make it
2705          * difficult for client to function correctly, whereas
2706          * decreasing the number of slots will (just?) affect
2707          * performance.  When short on memory we therefore prefer to
2708          * decrease number of slots instead of their size.  Clients that
2709          * request larger slots than they need will get poor results:
2710          */
2711         ca->maxreqs = nfsd4_get_drc_mem(ca);
2712         if (!ca->maxreqs)
2713                 return nfserr_jukebox;
2714
2715         return nfs_ok;
2716 }
2717
2718 /*
2719  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2720  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2721  */
2722 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2723         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2724
2725 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2726         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2727
2728 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2729                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2730 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2731                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2732                                  sizeof(__be32))
2733
2734 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2735 {
2736         ca->headerpadsz = 0;
2737
2738         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2739                 return nfserr_toosmall;
2740         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2741                 return nfserr_toosmall;
2742         ca->maxresp_cached = 0;
2743         if (ca->maxops < 2)
2744                 return nfserr_toosmall;
2745
2746         return nfs_ok;
2747 }
2748
2749 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2750 {
2751         switch (cbs->flavor) {
2752         case RPC_AUTH_NULL:
2753         case RPC_AUTH_UNIX:
2754                 return nfs_ok;
2755         default:
2756                 /*
2757                  * GSS case: the spec doesn't allow us to return this
2758                  * error.  But it also doesn't allow us not to support
2759                  * GSS.
2760                  * I'd rather this fail hard than return some error the
2761                  * client might think it can already handle:
2762                  */
2763                 return nfserr_encr_alg_unsupp;
2764         }
2765 }
2766
2767 __be32
2768 nfsd4_create_session(struct svc_rqst *rqstp,
2769                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2770 {
2771         struct nfsd4_create_session *cr_ses = &u->create_session;
2772         struct sockaddr *sa = svc_addr(rqstp);
2773         struct nfs4_client *conf, *unconf;
2774         struct nfs4_client *old = NULL;
2775         struct nfsd4_session *new;
2776         struct nfsd4_conn *conn;
2777         struct nfsd4_clid_slot *cs_slot = NULL;
2778         __be32 status = 0;
2779         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2780
2781         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2782                 return nfserr_inval;
2783         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2784         if (status)
2785                 return status;
2786         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2787         if (status)
2788                 return status;
2789         status = check_backchannel_attrs(&cr_ses->back_channel);
2790         if (status)
2791                 goto out_release_drc_mem;
2792         status = nfserr_jukebox;
2793         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2794         if (!new)
2795                 goto out_release_drc_mem;
2796         conn = alloc_conn_from_crses(rqstp, cr_ses);
2797         if (!conn)
2798                 goto out_free_session;
2799
2800         spin_lock(&nn->client_lock);
2801         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2802         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2803         WARN_ON_ONCE(conf && unconf);
2804
2805         if (conf) {
2806                 status = nfserr_wrong_cred;
2807                 if (!nfsd4_mach_creds_match(conf, rqstp))
2808                         goto out_free_conn;
2809                 cs_slot = &conf->cl_cs_slot;
2810                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2811                 if (status) {
2812                         if (status == nfserr_replay_cache)
2813                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2814                         goto out_free_conn;
2815                 }
2816         } else if (unconf) {
2817                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2818                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2819                         status = nfserr_clid_inuse;
2820                         goto out_free_conn;
2821                 }
2822                 status = nfserr_wrong_cred;
2823                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2824                         goto out_free_conn;
2825                 cs_slot = &unconf->cl_cs_slot;
2826                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2827                 if (status) {
2828                         /* an unconfirmed replay returns misordered */
2829                         status = nfserr_seq_misordered;
2830                         goto out_free_conn;
2831                 }
2832                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2833                 if (old) {
2834                         status = mark_client_expired_locked(old);
2835                         if (status) {
2836                                 old = NULL;
2837                                 goto out_free_conn;
2838                         }
2839                 }
2840                 move_to_confirmed(unconf);
2841                 conf = unconf;
2842         } else {
2843                 status = nfserr_stale_clientid;
2844                 goto out_free_conn;
2845         }
2846         status = nfs_ok;
2847         /* Persistent sessions are not supported */
2848         cr_ses->flags &= ~SESSION4_PERSIST;
2849         /* Upshifting from TCP to RDMA is not supported */
2850         cr_ses->flags &= ~SESSION4_RDMA;
2851
2852         init_session(rqstp, new, conf, cr_ses);
2853         nfsd4_get_session_locked(new);
2854
2855         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2856                NFS4_MAX_SESSIONID_LEN);
2857         cs_slot->sl_seqid++;
2858         cr_ses->seqid = cs_slot->sl_seqid;
2859
2860         /* cache solo and embedded create sessions under the client_lock */
2861         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2862         spin_unlock(&nn->client_lock);
2863         /* init connection and backchannel */
2864         nfsd4_init_conn(rqstp, conn, new);
2865         nfsd4_put_session(new);
2866         if (old)
2867                 expire_client(old);
2868         return status;
2869 out_free_conn:
2870         spin_unlock(&nn->client_lock);
2871         free_conn(conn);
2872         if (old)
2873                 expire_client(old);
2874 out_free_session:
2875         __free_session(new);
2876 out_release_drc_mem:
2877         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2878         return status;
2879 }
2880
2881 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2882 {
2883         switch (*dir) {
2884         case NFS4_CDFC4_FORE:
2885         case NFS4_CDFC4_BACK:
2886                 return nfs_ok;
2887         case NFS4_CDFC4_FORE_OR_BOTH:
2888         case NFS4_CDFC4_BACK_OR_BOTH:
2889                 *dir = NFS4_CDFC4_BOTH;
2890                 return nfs_ok;
2891         };
2892         return nfserr_inval;
2893 }
2894
2895 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2896                 struct nfsd4_compound_state *cstate,
2897                 union nfsd4_op_u *u)
2898 {
2899         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2900         struct nfsd4_session *session = cstate->session;
2901         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2902         __be32 status;
2903
2904         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2905         if (status)
2906                 return status;
2907         spin_lock(&nn->client_lock);
2908         session->se_cb_prog = bc->bc_cb_program;
2909         session->se_cb_sec = bc->bc_cb_sec;
2910         spin_unlock(&nn->client_lock);
2911
2912         nfsd4_probe_callback(session->se_client);
2913
2914         return nfs_ok;
2915 }
2916
2917 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2918                      struct nfsd4_compound_state *cstate,
2919                      union nfsd4_op_u *u)
2920 {
2921         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2922         __be32 status;
2923         struct nfsd4_conn *conn;
2924         struct nfsd4_session *session;
2925         struct net *net = SVC_NET(rqstp);
2926         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2927
2928         if (!nfsd4_last_compound_op(rqstp))
2929                 return nfserr_not_only_op;
2930         spin_lock(&nn->client_lock);
2931         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2932         spin_unlock(&nn->client_lock);
2933         if (!session)
2934                 goto out_no_session;
2935         status = nfserr_wrong_cred;
2936         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2937                 goto out;
2938         status = nfsd4_map_bcts_dir(&bcts->dir);
2939         if (status)
2940                 goto out;
2941         conn = alloc_conn(rqstp, bcts->dir);
2942         status = nfserr_jukebox;
2943         if (!conn)
2944                 goto out;
2945         nfsd4_init_conn(rqstp, conn, session);
2946         status = nfs_ok;
2947 out:
2948         nfsd4_put_session(session);
2949 out_no_session:
2950         return status;
2951 }
2952
2953 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2954 {
2955         if (!session)
2956                 return 0;
2957         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2958 }
2959
2960 __be32
2961 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2962                 union nfsd4_op_u *u)
2963 {
2964         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2965         struct nfsd4_session *ses;
2966         __be32 status;
2967         int ref_held_by_me = 0;
2968         struct net *net = SVC_NET(r);
2969         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2970
2971         status = nfserr_not_only_op;
2972         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2973                 if (!nfsd4_last_compound_op(r))
2974                         goto out;
2975                 ref_held_by_me++;
2976         }
2977         dump_sessionid(__func__, &sessionid->sessionid);
2978         spin_lock(&nn->client_lock);
2979         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2980         if (!ses)
2981                 goto out_client_lock;
2982         status = nfserr_wrong_cred;
2983         if (!nfsd4_mach_creds_match(ses->se_client, r))
2984                 goto out_put_session;
2985         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2986         if (status)
2987                 goto out_put_session;
2988         unhash_session(ses);
2989         spin_unlock(&nn->client_lock);
2990
2991         nfsd4_probe_callback_sync(ses->se_client);
2992
2993         spin_lock(&nn->client_lock);
2994         status = nfs_ok;
2995 out_put_session:
2996         nfsd4_put_session_locked(ses);
2997 out_client_lock:
2998         spin_unlock(&nn->client_lock);
2999 out:
3000         return status;
3001 }
3002
3003 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
3004 {
3005         struct nfsd4_conn *c;
3006
3007         list_for_each_entry(c, &s->se_conns, cn_persession) {
3008                 if (c->cn_xprt == xpt) {
3009                         return c;
3010                 }
3011         }
3012         return NULL;
3013 }
3014
3015 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
3016 {
3017         struct nfs4_client *clp = ses->se_client;
3018         struct nfsd4_conn *c;
3019         __be32 status = nfs_ok;
3020         int ret;
3021
3022         spin_lock(&clp->cl_lock);
3023         c = __nfsd4_find_conn(new->cn_xprt, ses);
3024         if (c)
3025                 goto out_free;
3026         status = nfserr_conn_not_bound_to_session;
3027         if (clp->cl_mach_cred)
3028                 goto out_free;
3029         __nfsd4_hash_conn(new, ses);
3030         spin_unlock(&clp->cl_lock);
3031         ret = nfsd4_register_conn(new);
3032         if (ret)
3033                 /* oops; xprt is already down: */
3034                 nfsd4_conn_lost(&new->cn_xpt_user);
3035         return nfs_ok;
3036 out_free:
3037         spin_unlock(&clp->cl_lock);
3038         free_conn(new);
3039         return status;
3040 }
3041
3042 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3043 {
3044         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3045
3046         return args->opcnt > session->se_fchannel.maxops;
3047 }
3048
3049 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3050                                   struct nfsd4_session *session)
3051 {
3052         struct xdr_buf *xb = &rqstp->rq_arg;
3053
3054         return xb->len > session->se_fchannel.maxreq_sz;
3055 }
3056
3057 static bool replay_matches_cache(struct svc_rqst *rqstp,
3058                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3059 {
3060         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3061
3062         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3063             (bool)seq->cachethis)
3064                 return false;
3065         /*
3066          * If there's an error than the reply can have fewer ops than
3067          * the call.  But if we cached a reply with *more* ops than the
3068          * call you're sending us now, then this new call is clearly not
3069          * really a replay of the old one:
3070          */
3071         if (slot->sl_opcnt < argp->opcnt)
3072                 return false;
3073         /* This is the only check explicitly called by spec: */
3074         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3075                 return false;
3076         /*
3077          * There may be more comparisons we could actually do, but the
3078          * spec doesn't require us to catch every case where the calls
3079          * don't match (that would require caching the call as well as
3080          * the reply), so we don't bother.
3081          */
3082         return true;
3083 }
3084
3085 __be32
3086 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3087                 union nfsd4_op_u *u)
3088 {
3089         struct nfsd4_sequence *seq = &u->sequence;
3090         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3091         struct xdr_stream *xdr = &resp->xdr;
3092         struct nfsd4_session *session;
3093         struct nfs4_client *clp;
3094         struct nfsd4_slot *slot;
3095         struct nfsd4_conn *conn;
3096         __be32 status;
3097         int buflen;
3098         struct net *net = SVC_NET(rqstp);
3099         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3100
3101         if (resp->opcnt != 1)
3102                 return nfserr_sequence_pos;
3103
3104         /*
3105          * Will be either used or freed by nfsd4_sequence_check_conn
3106          * below.
3107          */
3108         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3109         if (!conn)
3110                 return nfserr_jukebox;
3111
3112         spin_lock(&nn->client_lock);
3113         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3114         if (!session)
3115                 goto out_no_session;
3116         clp = session->se_client;
3117
3118         status = nfserr_too_many_ops;
3119         if (nfsd4_session_too_many_ops(rqstp, session))
3120                 goto out_put_session;
3121
3122         status = nfserr_req_too_big;
3123         if (nfsd4_request_too_big(rqstp, session))
3124                 goto out_put_session;
3125
3126         status = nfserr_badslot;
3127         if (seq->slotid >= session->se_fchannel.maxreqs)
3128                 goto out_put_session;
3129
3130         slot = session->se_slots[seq->slotid];
3131         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3132
3133         /* We do not negotiate the number of slots yet, so set the
3134          * maxslots to the session maxreqs which is used to encode
3135          * sr_highest_slotid and the sr_target_slot id to maxslots */
3136         seq->maxslots = session->se_fchannel.maxreqs;
3137
3138         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3139                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3140         if (status == nfserr_replay_cache) {
3141                 status = nfserr_seq_misordered;
3142                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3143                         goto out_put_session;
3144                 status = nfserr_seq_false_retry;
3145                 if (!replay_matches_cache(rqstp, seq, slot))
3146                         goto out_put_session;
3147                 cstate->slot = slot;
3148                 cstate->session = session;
3149                 cstate->clp = clp;
3150                 /* Return the cached reply status and set cstate->status
3151                  * for nfsd4_proc_compound processing */
3152                 status = nfsd4_replay_cache_entry(resp, seq);
3153                 cstate->status = nfserr_replay_cache;
3154                 goto out;
3155         }
3156         if (status)
3157                 goto out_put_session;
3158
3159         status = nfsd4_sequence_check_conn(conn, session);
3160         conn = NULL;
3161         if (status)
3162                 goto out_put_session;
3163
3164         buflen = (seq->cachethis) ?
3165                         session->se_fchannel.maxresp_cached :
3166                         session->se_fchannel.maxresp_sz;
3167         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3168                                     nfserr_rep_too_big;
3169         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3170                 goto out_put_session;
3171         svc_reserve(rqstp, buflen);
3172
3173         status = nfs_ok;
3174         /* Success! bump slot seqid */
3175         slot->sl_seqid = seq->seqid;
3176         slot->sl_flags |= NFSD4_SLOT_INUSE;
3177         if (seq->cachethis)
3178                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3179         else
3180                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3181
3182         cstate->slot = slot;
3183         cstate->session = session;
3184         cstate->clp = clp;
3185
3186 out:
3187         switch (clp->cl_cb_state) {
3188         case NFSD4_CB_DOWN:
3189                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3190                 break;
3191         case NFSD4_CB_FAULT:
3192                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3193                 break;
3194         default:
3195                 seq->status_flags = 0;
3196         }
3197         if (!list_empty(&clp->cl_revoked))
3198                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3199 out_no_session:
3200         if (conn)
3201                 free_conn(conn);
3202         spin_unlock(&nn->client_lock);
3203         return status;
3204 out_put_session:
3205         nfsd4_put_session_locked(session);
3206         goto out_no_session;
3207 }
3208
3209 void
3210 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3211 {
3212         struct nfsd4_compound_state *cs = &resp->cstate;
3213
3214         if (nfsd4_has_session(cs)) {
3215                 if (cs->status != nfserr_replay_cache) {
3216                         nfsd4_store_cache_entry(resp);
3217                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3218                 }
3219                 /* Drop session reference that was taken in nfsd4_sequence() */
3220                 nfsd4_put_session(cs->session);
3221         } else if (cs->clp)
3222                 put_client_renew(cs->clp);
3223 }
3224
3225 __be32
3226 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3227                 struct nfsd4_compound_state *cstate,
3228                 union nfsd4_op_u *u)
3229 {
3230         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3231         struct nfs4_client *conf, *unconf;
3232         struct nfs4_client *clp = NULL;
3233         __be32 status = 0;
3234         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3235
3236         spin_lock(&nn->client_lock);
3237         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3238         conf = find_confirmed_client(&dc->clientid, true, nn);
3239         WARN_ON_ONCE(conf && unconf);
3240
3241         if (conf) {
3242                 if (client_has_state(conf)) {
3243                         status = nfserr_clientid_busy;
3244                         goto out;
3245                 }
3246                 status = mark_client_expired_locked(conf);
3247                 if (status)
3248                         goto out;
3249                 clp = conf;
3250         } else if (unconf)
3251                 clp = unconf;
3252         else {
3253                 status = nfserr_stale_clientid;
3254                 goto out;
3255         }
3256         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3257                 clp = NULL;
3258                 status = nfserr_wrong_cred;
3259                 goto out;
3260         }
3261         unhash_client_locked(clp);
3262 out:
3263         spin_unlock(&nn->client_lock);
3264         if (clp)
3265                 expire_client(clp);
3266         return status;
3267 }
3268
3269 __be32
3270 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3271                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3272 {
3273         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3274         __be32 status = 0;
3275
3276         if (rc->rca_one_fs) {
3277                 if (!cstate->current_fh.fh_dentry)
3278                         return nfserr_nofilehandle;
3279                 /*
3280                  * We don't take advantage of the rca_one_fs case.
3281                  * That's OK, it's optional, we can safely ignore it.
3282                  */
3283                 return nfs_ok;
3284         }
3285
3286         status = nfserr_complete_already;
3287         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3288                              &cstate->session->se_client->cl_flags))
3289                 goto out;
3290
3291         status = nfserr_stale_clientid;
3292         if (is_client_expired(cstate->session->se_client))
3293                 /*
3294                  * The following error isn't really legal.
3295                  * But we only get here if the client just explicitly
3296                  * destroyed the client.  Surely it no longer cares what
3297                  * error it gets back on an operation for the dead
3298                  * client.
3299                  */
3300                 goto out;
3301
3302         status = nfs_ok;
3303         nfsd4_client_record_create(cstate->session->se_client);
3304 out:
3305         return status;
3306 }
3307
3308 __be32
3309 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3310                   union nfsd4_op_u *u)
3311 {
3312         struct nfsd4_setclientid *setclid = &u->setclientid;
3313         struct xdr_netobj       clname = setclid->se_name;
3314         nfs4_verifier           clverifier = setclid->se_verf;
3315         struct nfs4_client      *conf, *new;
3316         struct nfs4_client      *unconf = NULL;
3317         __be32                  status;
3318         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3319
3320         new = create_client(clname, rqstp, &clverifier);
3321         if (new == NULL)
3322                 return nfserr_jukebox;
3323         /* Cases below refer to rfc 3530 section 14.2.33: */
3324         spin_lock(&nn->client_lock);
3325         conf = find_confirmed_client_by_name(&clname, nn);
3326         if (conf && client_has_state(conf)) {
3327                 /* case 0: */
3328                 status = nfserr_clid_inuse;
3329                 if (clp_used_exchangeid(conf))
3330                         goto out;
3331                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3332                         char addr_str[INET6_ADDRSTRLEN];
3333                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3334                                  sizeof(addr_str));
3335                         dprintk("NFSD: setclientid: string in use by client "
3336                                 "at %s\n", addr_str);
3337                         goto out;
3338                 }
3339         }
3340         unconf = find_unconfirmed_client_by_name(&clname, nn);
3341         if (unconf)
3342                 unhash_client_locked(unconf);
3343         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3344                 /* case 1: probable callback update */
3345                 copy_clid(new, conf);
3346                 gen_confirm(new, nn);
3347         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3348                 gen_clid(new, nn);
3349         new->cl_minorversion = 0;
3350         gen_callback(new, setclid, rqstp);
3351         add_to_unconfirmed(new);
3352         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3353         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3354         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3355         new = NULL;
3356         status = nfs_ok;
3357 out:
3358         spin_unlock(&nn->client_lock);
3359         if (new)
3360                 free_client(new);
3361         if (unconf)
3362                 expire_client(unconf);
3363         return status;
3364 }
3365
3366
3367 __be32
3368 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3369                         struct nfsd4_compound_state *cstate,
3370                         union nfsd4_op_u *u)
3371 {
3372         struct nfsd4_setclientid_confirm *setclientid_confirm =
3373                         &u->setclientid_confirm;
3374         struct nfs4_client *conf, *unconf;
3375         struct nfs4_client *old = NULL;
3376         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
3377         clientid_t * clid = &setclientid_confirm->sc_clientid;
3378         __be32 status;
3379         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3380
3381         if (STALE_CLIENTID(clid, nn))
3382                 return nfserr_stale_clientid;
3383
3384         spin_lock(&nn->client_lock);
3385         conf = find_confirmed_client(clid, false, nn);
3386         unconf = find_unconfirmed_client(clid, false, nn);
3387         /*
3388          * We try hard to give out unique clientid's, so if we get an
3389          * attempt to confirm the same clientid with a different cred,
3390          * the client may be buggy; this should never happen.
3391          *
3392          * Nevertheless, RFC 7530 recommends INUSE for this case:
3393          */
3394         status = nfserr_clid_inuse;
3395         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3396                 goto out;
3397         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3398                 goto out;
3399         /* cases below refer to rfc 3530 section 14.2.34: */
3400         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3401                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3402                         /* case 2: probable retransmit */
3403                         status = nfs_ok;
3404                 } else /* case 4: client hasn't noticed we rebooted yet? */
3405                         status = nfserr_stale_clientid;
3406                 goto out;
3407         }
3408         status = nfs_ok;
3409         if (conf) { /* case 1: callback update */
3410                 old = unconf;
3411                 unhash_client_locked(old);
3412                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3413         } else { /* case 3: normal case; new or rebooted client */
3414                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3415                 if (old) {
3416                         status = nfserr_clid_inuse;
3417                         if (client_has_state(old)
3418                                         && !same_creds(&unconf->cl_cred,
3419                                                         &old->cl_cred))
3420                                 goto out;
3421                         status = mark_client_expired_locked(old);
3422                         if (status) {
3423                                 old = NULL;
3424                                 goto out;
3425                         }
3426                 }
3427                 move_to_confirmed(unconf);
3428                 conf = unconf;
3429         }
3430         get_client_locked(conf);
3431         spin_unlock(&nn->client_lock);
3432         nfsd4_probe_callback(conf);
3433         spin_lock(&nn->client_lock);
3434         put_client_renew_locked(conf);
3435 out:
3436         spin_unlock(&nn->client_lock);
3437         if (old)
3438                 expire_client(old);
3439         return status;
3440 }
3441
3442 static struct nfs4_file *nfsd4_alloc_file(void)
3443 {
3444         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3445 }
3446
3447 /* OPEN Share state helper functions */
3448 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3449                                 struct nfs4_file *fp)
3450 {
3451         lockdep_assert_held(&state_lock);
3452
3453         refcount_set(&fp->fi_ref, 1);
3454         spin_lock_init(&fp->fi_lock);
3455         INIT_LIST_HEAD(&fp->fi_stateids);
3456         INIT_LIST_HEAD(&fp->fi_delegations);
3457         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3458         fh_copy_shallow(&fp->fi_fhandle, fh);
3459         fp->fi_deleg_file = NULL;
3460         fp->fi_had_conflict = false;
3461         fp->fi_share_deny = 0;
3462         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3463         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3464 #ifdef CONFIG_NFSD_PNFS
3465         INIT_LIST_HEAD(&fp->fi_lo_states);
3466         atomic_set(&fp->fi_lo_recalls, 0);
3467 #endif
3468         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3469 }
3470
3471 void
3472 nfsd4_free_slabs(void)
3473 {
3474         kmem_cache_destroy(odstate_slab);
3475         kmem_cache_destroy(openowner_slab);
3476         kmem_cache_destroy(lockowner_slab);
3477         kmem_cache_destroy(file_slab);
3478         kmem_cache_destroy(stateid_slab);
3479         kmem_cache_destroy(deleg_slab);
3480 }
3481
3482 int
3483 nfsd4_init_slabs(void)
3484 {
3485         openowner_slab = kmem_cache_create("nfsd4_openowners",
3486                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3487         if (openowner_slab == NULL)
3488                 goto out;
3489         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3490                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3491         if (lockowner_slab == NULL)
3492                 goto out_free_openowner_slab;
3493         file_slab = kmem_cache_create("nfsd4_files",
3494                         sizeof(struct nfs4_file), 0, 0, NULL);
3495         if (file_slab == NULL)
3496                 goto out_free_lockowner_slab;
3497         stateid_slab = kmem_cache_create("nfsd4_stateids",
3498                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3499         if (stateid_slab == NULL)
3500                 goto out_free_file_slab;
3501         deleg_slab = kmem_cache_create("nfsd4_delegations",
3502                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3503         if (deleg_slab == NULL)
3504                 goto out_free_stateid_slab;
3505         odstate_slab = kmem_cache_create("nfsd4_odstate",
3506                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3507         if (odstate_slab == NULL)
3508                 goto out_free_deleg_slab;
3509         return 0;
3510
3511 out_free_deleg_slab:
3512         kmem_cache_destroy(deleg_slab);
3513 out_free_stateid_slab:
3514         kmem_cache_destroy(stateid_slab);
3515 out_free_file_slab:
3516         kmem_cache_destroy(file_slab);
3517 out_free_lockowner_slab:
3518         kmem_cache_destroy(lockowner_slab);
3519 out_free_openowner_slab:
3520         kmem_cache_destroy(openowner_slab);
3521 out:
3522         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3523         return -ENOMEM;
3524 }
3525
3526 static void init_nfs4_replay(struct nfs4_replay *rp)
3527 {
3528         rp->rp_status = nfserr_serverfault;
3529         rp->rp_buflen = 0;
3530         rp->rp_buf = rp->rp_ibuf;
3531         mutex_init(&rp->rp_mutex);
3532 }
3533
3534 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3535                 struct nfs4_stateowner *so)
3536 {
3537         if (!nfsd4_has_session(cstate)) {
3538                 mutex_lock(&so->so_replay.rp_mutex);
3539                 cstate->replay_owner = nfs4_get_stateowner(so);
3540         }
3541 }
3542
3543 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3544 {
3545         struct nfs4_stateowner *so = cstate->replay_owner;
3546
3547         if (so != NULL) {
3548                 cstate->replay_owner = NULL;
3549                 mutex_unlock(&so->so_replay.rp_mutex);
3550                 nfs4_put_stateowner(so);
3551         }
3552 }
3553
3554 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3555 {
3556         struct nfs4_stateowner *sop;
3557
3558         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3559         if (!sop)
3560                 return NULL;
3561
3562         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3563         if (!sop->so_owner.data) {
3564                 kmem_cache_free(slab, sop);
3565                 return NULL;
3566         }
3567         sop->so_owner.len = owner->len;
3568
3569         INIT_LIST_HEAD(&sop->so_stateids);
3570         sop->so_client = clp;
3571         init_nfs4_replay(&sop->so_replay);
3572         atomic_set(&sop->so_count, 1);
3573         return sop;
3574 }
3575
3576 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3577 {
3578         lockdep_assert_held(&clp->cl_lock);
3579
3580         list_add(&oo->oo_owner.so_strhash,
3581                  &clp->cl_ownerstr_hashtbl[strhashval]);
3582         list_add(&oo->oo_perclient, &clp->cl_openowners);
3583 }
3584
3585 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3586 {
3587         unhash_openowner_locked(openowner(so));
3588 }
3589
3590 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3591 {
3592         struct nfs4_openowner *oo = openowner(so);
3593
3594         kmem_cache_free(openowner_slab, oo);
3595 }
3596
3597 static const struct nfs4_stateowner_operations openowner_ops = {
3598         .so_unhash =    nfs4_unhash_openowner,
3599         .so_free =      nfs4_free_openowner,
3600 };
3601
3602 static struct nfs4_ol_stateid *
3603 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3604 {
3605         struct nfs4_ol_stateid *local, *ret = NULL;
3606         struct nfs4_openowner *oo = open->op_openowner;
3607
3608         lockdep_assert_held(&fp->fi_lock);
3609
3610         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3611                 /* ignore lock owners */
3612                 if (local->st_stateowner->so_is_open_owner == 0)
3613                         continue;
3614                 if (local->st_stateowner != &oo->oo_owner)
3615                         continue;
3616                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3617                         ret = local;
3618                         refcount_inc(&ret->st_stid.sc_count);
3619                         break;
3620                 }
3621         }
3622         return ret;