Merge branch 'for-2.6.31' of git://git.linux-nfs.org/projects/trondmy/nfs-2.6
[linux-2.6.git] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58
59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
60
61 /* Globals */
62 static time_t lease_time = 90;     /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
68 static u32 nfs4_init;
69 static stateid_t zerostateid;             /* bits all 0 */
70 static stateid_t onestateid;              /* bits all 1 */
71 static u64 current_sessionid = 1;
72
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
75
76 /* forward declarations */
77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
81
82 /* Locking: */
83
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex);
86
87 /*
88  * Currently used for the del_recall_lru and file hash table.  In an
89  * effort to decrease the scope of the client_mutex, this spinlock may
90  * eventually cover more:
91  */
92 static DEFINE_SPINLOCK(recall_lock);
93
94 static struct kmem_cache *stateowner_slab = NULL;
95 static struct kmem_cache *file_slab = NULL;
96 static struct kmem_cache *stateid_slab = NULL;
97 static struct kmem_cache *deleg_slab = NULL;
98
99 void
100 nfs4_lock_state(void)
101 {
102         mutex_lock(&client_mutex);
103 }
104
105 void
106 nfs4_unlock_state(void)
107 {
108         mutex_unlock(&client_mutex);
109 }
110
111 static inline u32
112 opaque_hashval(const void *ptr, int nbytes)
113 {
114         unsigned char *cptr = (unsigned char *) ptr;
115
116         u32 x = 0;
117         while (nbytes--) {
118                 x *= 37;
119                 x += *cptr++;
120         }
121         return x;
122 }
123
124 static struct list_head del_recall_lru;
125
126 static inline void
127 put_nfs4_file(struct nfs4_file *fi)
128 {
129         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
130                 list_del(&fi->fi_hash);
131                 spin_unlock(&recall_lock);
132                 iput(fi->fi_inode);
133                 kmem_cache_free(file_slab, fi);
134         }
135 }
136
137 static inline void
138 get_nfs4_file(struct nfs4_file *fi)
139 {
140         atomic_inc(&fi->fi_ref);
141 }
142
143 static int num_delegations;
144 unsigned int max_delegations;
145
146 /*
147  * Open owner state (share locks)
148  */
149
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS              8
152 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
154
155 #define ownerid_hashval(id) \
156         ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
159
160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
162
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS                   8
165 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS              10
169 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
171
172 #define file_hashval(x) \
173         hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id)  \
175         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
176
177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
179
180 static struct nfs4_delegation *
181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
182 {
183         struct nfs4_delegation *dp;
184         struct nfs4_file *fp = stp->st_file;
185         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
186
187         dprintk("NFSD alloc_init_deleg\n");
188         if (fp->fi_had_conflict)
189                 return NULL;
190         if (num_delegations > max_delegations)
191                 return NULL;
192         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
193         if (dp == NULL)
194                 return dp;
195         num_delegations++;
196         INIT_LIST_HEAD(&dp->dl_perfile);
197         INIT_LIST_HEAD(&dp->dl_perclnt);
198         INIT_LIST_HEAD(&dp->dl_recall_lru);
199         dp->dl_client = clp;
200         get_nfs4_file(fp);
201         dp->dl_file = fp;
202         dp->dl_flock = NULL;
203         get_file(stp->st_vfs_file);
204         dp->dl_vfs_file = stp->st_vfs_file;
205         dp->dl_type = type;
206         dp->dl_recall.cbr_dp = NULL;
207         dp->dl_recall.cbr_ident = cb->cb_ident;
208         dp->dl_recall.cbr_trunc = 0;
209         dp->dl_stateid.si_boot = boot_time;
210         dp->dl_stateid.si_stateownerid = current_delegid++;
211         dp->dl_stateid.si_fileid = 0;
212         dp->dl_stateid.si_generation = 0;
213         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
214         dp->dl_time = 0;
215         atomic_set(&dp->dl_count, 1);
216         list_add(&dp->dl_perfile, &fp->fi_delegations);
217         list_add(&dp->dl_perclnt, &clp->cl_delegations);
218         return dp;
219 }
220
221 void
222 nfs4_put_delegation(struct nfs4_delegation *dp)
223 {
224         if (atomic_dec_and_test(&dp->dl_count)) {
225                 dprintk("NFSD: freeing dp %p\n",dp);
226                 put_nfs4_file(dp->dl_file);
227                 kmem_cache_free(deleg_slab, dp);
228                 num_delegations--;
229         }
230 }
231
232 /* Remove the associated file_lock first, then remove the delegation.
233  * lease_modify() is called to remove the FS_LEASE file_lock from
234  * the i_flock list, eventually calling nfsd's lock_manager
235  * fl_release_callback.
236  */
237 static void
238 nfs4_close_delegation(struct nfs4_delegation *dp)
239 {
240         struct file *filp = dp->dl_vfs_file;
241
242         dprintk("NFSD: close_delegation dp %p\n",dp);
243         dp->dl_vfs_file = NULL;
244         /* The following nfsd_close may not actually close the file,
245          * but we want to remove the lease in any case. */
246         if (dp->dl_flock)
247                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
248         nfsd_close(filp);
249 }
250
251 /* Called under the state lock. */
252 static void
253 unhash_delegation(struct nfs4_delegation *dp)
254 {
255         list_del_init(&dp->dl_perfile);
256         list_del_init(&dp->dl_perclnt);
257         spin_lock(&recall_lock);
258         list_del_init(&dp->dl_recall_lru);
259         spin_unlock(&recall_lock);
260         nfs4_close_delegation(dp);
261         nfs4_put_delegation(dp);
262 }
263
264 /* 
265  * SETCLIENTID state 
266  */
267
268 /* Hash tables for nfs4_clientid state */
269 #define CLIENT_HASH_BITS                 4
270 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
271 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
272
273 #define clientid_hashval(id) \
274         ((id) & CLIENT_HASH_MASK)
275 #define clientstr_hashval(name) \
276         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
277 /*
278  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
279  * used in reboot/reset lease grace period processing
280  *
281  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
282  * setclientid_confirmed info. 
283  *
284  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
285  * setclientid info.
286  *
287  * client_lru holds client queue ordered by nfs4_client.cl_time
288  * for lease renewal.
289  *
290  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
291  * for last close replay.
292  */
293 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
294 static int reclaim_str_hashtbl_size = 0;
295 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
299 static struct list_head client_lru;
300 static struct list_head close_lru;
301
302 static void unhash_generic_stateid(struct nfs4_stateid *stp)
303 {
304         list_del(&stp->st_hash);
305         list_del(&stp->st_perfile);
306         list_del(&stp->st_perstateowner);
307 }
308
309 static void free_generic_stateid(struct nfs4_stateid *stp)
310 {
311         put_nfs4_file(stp->st_file);
312         kmem_cache_free(stateid_slab, stp);
313 }
314
315 static void release_lock_stateid(struct nfs4_stateid *stp)
316 {
317         unhash_generic_stateid(stp);
318         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
319         free_generic_stateid(stp);
320 }
321
322 static void unhash_lockowner(struct nfs4_stateowner *sop)
323 {
324         struct nfs4_stateid *stp;
325
326         list_del(&sop->so_idhash);
327         list_del(&sop->so_strhash);
328         list_del(&sop->so_perstateid);
329         while (!list_empty(&sop->so_stateids)) {
330                 stp = list_first_entry(&sop->so_stateids,
331                                 struct nfs4_stateid, st_perstateowner);
332                 release_lock_stateid(stp);
333         }
334 }
335
336 static void release_lockowner(struct nfs4_stateowner *sop)
337 {
338         unhash_lockowner(sop);
339         nfs4_put_stateowner(sop);
340 }
341
342 static void
343 release_stateid_lockowners(struct nfs4_stateid *open_stp)
344 {
345         struct nfs4_stateowner *lock_sop;
346
347         while (!list_empty(&open_stp->st_lockowners)) {
348                 lock_sop = list_entry(open_stp->st_lockowners.next,
349                                 struct nfs4_stateowner, so_perstateid);
350                 /* list_del(&open_stp->st_lockowners);  */
351                 BUG_ON(lock_sop->so_is_open_owner);
352                 release_lockowner(lock_sop);
353         }
354 }
355
356 static void release_open_stateid(struct nfs4_stateid *stp)
357 {
358         unhash_generic_stateid(stp);
359         release_stateid_lockowners(stp);
360         nfsd_close(stp->st_vfs_file);
361         free_generic_stateid(stp);
362 }
363
364 static void unhash_openowner(struct nfs4_stateowner *sop)
365 {
366         struct nfs4_stateid *stp;
367
368         list_del(&sop->so_idhash);
369         list_del(&sop->so_strhash);
370         list_del(&sop->so_perclient);
371         list_del(&sop->so_perstateid); /* XXX: necessary? */
372         while (!list_empty(&sop->so_stateids)) {
373                 stp = list_first_entry(&sop->so_stateids,
374                                 struct nfs4_stateid, st_perstateowner);
375                 release_open_stateid(stp);
376         }
377 }
378
379 static void release_openowner(struct nfs4_stateowner *sop)
380 {
381         unhash_openowner(sop);
382         list_del(&sop->so_close_lru);
383         nfs4_put_stateowner(sop);
384 }
385
386 static DEFINE_SPINLOCK(sessionid_lock);
387 #define SESSION_HASH_SIZE       512
388 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
389
390 static inline int
391 hash_sessionid(struct nfs4_sessionid *sessionid)
392 {
393         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
394
395         return sid->sequence % SESSION_HASH_SIZE;
396 }
397
398 static inline void
399 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
400 {
401         u32 *ptr = (u32 *)(&sessionid->data[0]);
402         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
403 }
404
405 static void
406 gen_sessionid(struct nfsd4_session *ses)
407 {
408         struct nfs4_client *clp = ses->se_client;
409         struct nfsd4_sessionid *sid;
410
411         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
412         sid->clientid = clp->cl_clientid;
413         sid->sequence = current_sessionid++;
414         sid->reserved = 0;
415 }
416
417 /*
418  * Give the client the number of slots it requests bound by
419  * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
420  *
421  * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
422  * should (up to a point) re-negotiate active sessions and reduce their
423  * slot usage to make rooom for new connections. For now we just fail the
424  * create session.
425  */
426 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
427 {
428         int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
429
430         spin_lock(&nfsd_serv->sv_lock);
431         if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
432                 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
433         nfsd_serv->sv_drc_pages_used += np;
434         spin_unlock(&nfsd_serv->sv_lock);
435
436         if (np <= 0) {
437                 status = nfserr_resource;
438                 fchan->maxreqs = 0;
439         } else
440                 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
441
442         return status;
443 }
444
445 /*
446  * fchan holds the client values on input, and the server values on output
447  */
448 static int init_forechannel_attrs(struct svc_rqst *rqstp,
449                                     struct nfsd4_session *session,
450                                     struct nfsd4_channel_attrs *fchan)
451 {
452         int status = 0;
453         __u32   maxcount = svc_max_payload(rqstp);
454
455         /* headerpadsz set to zero in encode routine */
456
457         /* Use the client's max request and max response size if possible */
458         if (fchan->maxreq_sz > maxcount)
459                 fchan->maxreq_sz = maxcount;
460         session->se_fmaxreq_sz = fchan->maxreq_sz;
461
462         if (fchan->maxresp_sz > maxcount)
463                 fchan->maxresp_sz = maxcount;
464         session->se_fmaxresp_sz = fchan->maxresp_sz;
465
466         /* Set the max response cached size our default which is
467          * a multiple of PAGE_SIZE and small */
468         session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
469         fchan->maxresp_cached = session->se_fmaxresp_cached;
470
471         /* Use the client's maxops if possible */
472         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
473                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
474         session->se_fmaxops = fchan->maxops;
475
476         /* try to use the client requested number of slots */
477         if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
478                 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
479
480         /* FIXME: Error means no more DRC pages so the server should
481          * recover pages from existing sessions. For now fail session
482          * creation.
483          */
484         status = set_forechannel_maxreqs(fchan);
485
486         session->se_fnumslots = fchan->maxreqs;
487         return status;
488 }
489
490 static int
491 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
492                    struct nfsd4_create_session *cses)
493 {
494         struct nfsd4_session *new, tmp;
495         int idx, status = nfserr_resource, slotsize;
496
497         memset(&tmp, 0, sizeof(tmp));
498
499         /* FIXME: For now, we just accept the client back channel attributes. */
500         status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
501         if (status)
502                 goto out;
503
504         /* allocate struct nfsd4_session and slot table in one piece */
505         slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
506         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
507         if (!new)
508                 goto out;
509
510         memcpy(new, &tmp, sizeof(*new));
511
512         new->se_client = clp;
513         gen_sessionid(new);
514         idx = hash_sessionid(&new->se_sessionid);
515         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
516                NFS4_MAX_SESSIONID_LEN);
517
518         new->se_flags = cses->flags;
519         kref_init(&new->se_ref);
520         spin_lock(&sessionid_lock);
521         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
522         list_add(&new->se_perclnt, &clp->cl_sessions);
523         spin_unlock(&sessionid_lock);
524
525         status = nfs_ok;
526 out:
527         return status;
528 }
529
530 /* caller must hold sessionid_lock */
531 static struct nfsd4_session *
532 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
533 {
534         struct nfsd4_session *elem;
535         int idx;
536
537         dump_sessionid(__func__, sessionid);
538         idx = hash_sessionid(sessionid);
539         dprintk("%s: idx is %d\n", __func__, idx);
540         /* Search in the appropriate list */
541         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
542                 dump_sessionid("list traversal", &elem->se_sessionid);
543                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
544                             NFS4_MAX_SESSIONID_LEN)) {
545                         return elem;
546                 }
547         }
548
549         dprintk("%s: session not found\n", __func__);
550         return NULL;
551 }
552
553 /* caller must hold sessionid_lock */
554 static void
555 unhash_session(struct nfsd4_session *ses)
556 {
557         list_del(&ses->se_hash);
558         list_del(&ses->se_perclnt);
559 }
560
561 static void
562 release_session(struct nfsd4_session *ses)
563 {
564         spin_lock(&sessionid_lock);
565         unhash_session(ses);
566         spin_unlock(&sessionid_lock);
567         nfsd4_put_session(ses);
568 }
569
570 static void nfsd4_release_respages(struct page **respages, short resused);
571
572 void
573 free_session(struct kref *kref)
574 {
575         struct nfsd4_session *ses;
576         int i;
577
578         ses = container_of(kref, struct nfsd4_session, se_ref);
579         for (i = 0; i < ses->se_fnumslots; i++) {
580                 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
581                 nfsd4_release_respages(e->ce_respages, e->ce_resused);
582         }
583         kfree(ses);
584 }
585
586 static inline void
587 renew_client(struct nfs4_client *clp)
588 {
589         /*
590         * Move client to the end to the LRU list.
591         */
592         dprintk("renewing client (clientid %08x/%08x)\n", 
593                         clp->cl_clientid.cl_boot, 
594                         clp->cl_clientid.cl_id);
595         list_move_tail(&clp->cl_lru, &client_lru);
596         clp->cl_time = get_seconds();
597 }
598
599 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
600 static int
601 STALE_CLIENTID(clientid_t *clid)
602 {
603         if (clid->cl_boot == boot_time)
604                 return 0;
605         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
606                 clid->cl_boot, clid->cl_id, boot_time);
607         return 1;
608 }
609
610 /* 
611  * XXX Should we use a slab cache ?
612  * This type of memory management is somewhat inefficient, but we use it
613  * anyway since SETCLIENTID is not a common operation.
614  */
615 static struct nfs4_client *alloc_client(struct xdr_netobj name)
616 {
617         struct nfs4_client *clp;
618
619         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
620         if (clp == NULL)
621                 return NULL;
622         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
623         if (clp->cl_name.data == NULL) {
624                 kfree(clp);
625                 return NULL;
626         }
627         memcpy(clp->cl_name.data, name.data, name.len);
628         clp->cl_name.len = name.len;
629         return clp;
630 }
631
632 static void
633 shutdown_callback_client(struct nfs4_client *clp)
634 {
635         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
636
637         if (clnt) {
638                 /*
639                  * Callback threads take a reference on the client, so there
640                  * should be no outstanding callbacks at this point.
641                  */
642                 clp->cl_callback.cb_client = NULL;
643                 rpc_shutdown_client(clnt);
644         }
645 }
646
647 static inline void
648 free_client(struct nfs4_client *clp)
649 {
650         shutdown_callback_client(clp);
651         nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
652                              clp->cl_slot.sl_cache_entry.ce_resused);
653         if (clp->cl_cred.cr_group_info)
654                 put_group_info(clp->cl_cred.cr_group_info);
655         kfree(clp->cl_principal);
656         kfree(clp->cl_name.data);
657         kfree(clp);
658 }
659
660 void
661 put_nfs4_client(struct nfs4_client *clp)
662 {
663         if (atomic_dec_and_test(&clp->cl_count))
664                 free_client(clp);
665 }
666
667 static void
668 expire_client(struct nfs4_client *clp)
669 {
670         struct nfs4_stateowner *sop;
671         struct nfs4_delegation *dp;
672         struct list_head reaplist;
673
674         dprintk("NFSD: expire_client cl_count %d\n",
675                             atomic_read(&clp->cl_count));
676
677         INIT_LIST_HEAD(&reaplist);
678         spin_lock(&recall_lock);
679         while (!list_empty(&clp->cl_delegations)) {
680                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
681                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
682                                 dp->dl_flock);
683                 list_del_init(&dp->dl_perclnt);
684                 list_move(&dp->dl_recall_lru, &reaplist);
685         }
686         spin_unlock(&recall_lock);
687         while (!list_empty(&reaplist)) {
688                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
689                 list_del_init(&dp->dl_recall_lru);
690                 unhash_delegation(dp);
691         }
692         list_del(&clp->cl_idhash);
693         list_del(&clp->cl_strhash);
694         list_del(&clp->cl_lru);
695         while (!list_empty(&clp->cl_openowners)) {
696                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
697                 release_openowner(sop);
698         }
699         while (!list_empty(&clp->cl_sessions)) {
700                 struct nfsd4_session  *ses;
701                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
702                                  se_perclnt);
703                 release_session(ses);
704         }
705         put_nfs4_client(clp);
706 }
707
708 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
709 {
710         struct nfs4_client *clp;
711
712         clp = alloc_client(name);
713         if (clp == NULL)
714                 return NULL;
715         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
716         atomic_set(&clp->cl_count, 1);
717         atomic_set(&clp->cl_callback.cb_set, 0);
718         INIT_LIST_HEAD(&clp->cl_idhash);
719         INIT_LIST_HEAD(&clp->cl_strhash);
720         INIT_LIST_HEAD(&clp->cl_openowners);
721         INIT_LIST_HEAD(&clp->cl_delegations);
722         INIT_LIST_HEAD(&clp->cl_sessions);
723         INIT_LIST_HEAD(&clp->cl_lru);
724         return clp;
725 }
726
727 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
728 {
729         memcpy(target->cl_verifier.data, source->data,
730                         sizeof(target->cl_verifier.data));
731 }
732
733 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
734 {
735         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
736         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
737 }
738
739 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
740 {
741         target->cr_uid = source->cr_uid;
742         target->cr_gid = source->cr_gid;
743         target->cr_group_info = source->cr_group_info;
744         get_group_info(target->cr_group_info);
745 }
746
747 static int same_name(const char *n1, const char *n2)
748 {
749         return 0 == memcmp(n1, n2, HEXDIR_LEN);
750 }
751
752 static int
753 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
754 {
755         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
756 }
757
758 static int
759 same_clid(clientid_t *cl1, clientid_t *cl2)
760 {
761         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
762 }
763
764 /* XXX what about NGROUP */
765 static int
766 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
767 {
768         return cr1->cr_uid == cr2->cr_uid;
769 }
770
771 static void gen_clid(struct nfs4_client *clp)
772 {
773         static u32 current_clientid = 1;
774
775         clp->cl_clientid.cl_boot = boot_time;
776         clp->cl_clientid.cl_id = current_clientid++; 
777 }
778
779 static void gen_confirm(struct nfs4_client *clp)
780 {
781         static u32 i;
782         u32 *p;
783
784         p = (u32 *)clp->cl_confirm.data;
785         *p++ = get_seconds();
786         *p++ = i++;
787 }
788
789 static int check_name(struct xdr_netobj name)
790 {
791         if (name.len == 0) 
792                 return 0;
793         if (name.len > NFS4_OPAQUE_LIMIT) {
794                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
795                 return 0;
796         }
797         return 1;
798 }
799
800 static void
801 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
802 {
803         unsigned int idhashval;
804
805         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
806         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
807         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
808         list_add_tail(&clp->cl_lru, &client_lru);
809         clp->cl_time = get_seconds();
810 }
811
812 static void
813 move_to_confirmed(struct nfs4_client *clp)
814 {
815         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
816         unsigned int strhashval;
817
818         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
819         list_del_init(&clp->cl_strhash);
820         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
821         strhashval = clientstr_hashval(clp->cl_recdir);
822         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
823         renew_client(clp);
824 }
825
826 static struct nfs4_client *
827 find_confirmed_client(clientid_t *clid)
828 {
829         struct nfs4_client *clp;
830         unsigned int idhashval = clientid_hashval(clid->cl_id);
831
832         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
833                 if (same_clid(&clp->cl_clientid, clid))
834                         return clp;
835         }
836         return NULL;
837 }
838
839 static struct nfs4_client *
840 find_unconfirmed_client(clientid_t *clid)
841 {
842         struct nfs4_client *clp;
843         unsigned int idhashval = clientid_hashval(clid->cl_id);
844
845         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
846                 if (same_clid(&clp->cl_clientid, clid))
847                         return clp;
848         }
849         return NULL;
850 }
851
852 /*
853  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
854  * parameter. Matching is based on the fact the at least one of the
855  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
856  *
857  * FIXME: we need to unify the clientid namespaces for nfsv4.x
858  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
859  * and SET_CLIENTID{,_CONFIRM}
860  */
861 static inline int
862 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
863 {
864         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
865         return use_exchange_id == has_exchange_flags;
866 }
867
868 static struct nfs4_client *
869 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
870                              bool use_exchange_id)
871 {
872         struct nfs4_client *clp;
873
874         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
875                 if (same_name(clp->cl_recdir, dname) &&
876                     match_clientid_establishment(clp, use_exchange_id))
877                         return clp;
878         }
879         return NULL;
880 }
881
882 static struct nfs4_client *
883 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
884                                bool use_exchange_id)
885 {
886         struct nfs4_client *clp;
887
888         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
889                 if (same_name(clp->cl_recdir, dname) &&
890                     match_clientid_establishment(clp, use_exchange_id))
891                         return clp;
892         }
893         return NULL;
894 }
895
896 /* a helper function for parse_callback */
897 static int
898 parse_octet(unsigned int *lenp, char **addrp)
899 {
900         unsigned int len = *lenp;
901         char *p = *addrp;
902         int n = -1;
903         char c;
904
905         for (;;) {
906                 if (!len)
907                         break;
908                 len--;
909                 c = *p++;
910                 if (c == '.')
911                         break;
912                 if ((c < '0') || (c > '9')) {
913                         n = -1;
914                         break;
915                 }
916                 if (n < 0)
917                         n = 0;
918                 n = (n * 10) + (c - '0');
919                 if (n > 255) {
920                         n = -1;
921                         break;
922                 }
923         }
924         *lenp = len;
925         *addrp = p;
926         return n;
927 }
928
929 /* parse and set the setclientid ipv4 callback address */
930 static int
931 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
932 {
933         int temp = 0;
934         u32 cbaddr = 0;
935         u16 cbport = 0;
936         u32 addrlen = addr_len;
937         char *addr = addr_val;
938         int i, shift;
939
940         /* ipaddress */
941         shift = 24;
942         for(i = 4; i > 0  ; i--) {
943                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
944                         return 0;
945                 }
946                 cbaddr |= (temp << shift);
947                 if (shift > 0)
948                 shift -= 8;
949         }
950         *cbaddrp = cbaddr;
951
952         /* port */
953         shift = 8;
954         for(i = 2; i > 0  ; i--) {
955                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
956                         return 0;
957                 }
958                 cbport |= (temp << shift);
959                 if (shift > 0)
960                         shift -= 8;
961         }
962         *cbportp = cbport;
963         return 1;
964 }
965
966 static void
967 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
968 {
969         struct nfs4_callback *cb = &clp->cl_callback;
970
971         /* Currently, we only support tcp for the callback channel */
972         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
973                 goto out_err;
974
975         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
976                          &cb->cb_addr, &cb->cb_port)))
977                 goto out_err;
978         cb->cb_prog = se->se_callback_prog;
979         cb->cb_ident = se->se_callback_ident;
980         return;
981 out_err:
982         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
983                 "will not receive delegations\n",
984                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
985
986         return;
987 }
988
989 void
990 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
991 {
992         struct nfsd4_compoundres *resp = rqstp->rq_resp;
993
994         resp->cstate.statp = statp;
995 }
996
997 /*
998  * Dereference the result pages.
999  */
1000 static void
1001 nfsd4_release_respages(struct page **respages, short resused)
1002 {
1003         int i;
1004
1005         dprintk("--> %s\n", __func__);
1006         for (i = 0; i < resused; i++) {
1007                 if (!respages[i])
1008                         continue;
1009                 put_page(respages[i]);
1010                 respages[i] = NULL;
1011         }
1012 }
1013
1014 static void
1015 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1016 {
1017         int i;
1018
1019         for (i = 0; i < count; i++) {
1020                 topages[i] = frompages[i];
1021                 if (!topages[i])
1022                         continue;
1023                 get_page(topages[i]);
1024         }
1025 }
1026
1027 /*
1028  * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1029  * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1030  * length of the XDR response is less than se_fmaxresp_cached
1031  * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1032  * of the reply (e.g. readdir).
1033  *
1034  * Store the base and length of the rq_req.head[0] page
1035  * of the NFSv4.1 data, just past the rpc header.
1036  */
1037 void
1038 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1039 {
1040         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1041         struct svc_rqst *rqstp = resp->rqstp;
1042         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1043         struct nfsd4_op *op = &args->ops[resp->opcnt];
1044         struct kvec *resv = &rqstp->rq_res.head[0];
1045
1046         dprintk("--> %s entry %p\n", __func__, entry);
1047
1048         /* Don't cache a failed OP_SEQUENCE. */
1049         if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1050                 return;
1051
1052         nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1053         entry->ce_opcnt = resp->opcnt;
1054         entry->ce_status = resp->cstate.status;
1055
1056         /*
1057          * Don't need a page to cache just the sequence operation - the slot
1058          * does this for us!
1059          */
1060
1061         if (nfsd4_not_cached(resp)) {
1062                 entry->ce_resused = 0;
1063                 entry->ce_rpchdrlen = 0;
1064                 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1065                         resp->cstate.slot->sl_cache_entry.ce_cachethis);
1066                 return;
1067         }
1068         entry->ce_resused = rqstp->rq_resused;
1069         if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1070                 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1071         nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1072                          entry->ce_resused);
1073         entry->ce_datav.iov_base = resp->cstate.statp;
1074         entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1075                                 (char *)page_address(rqstp->rq_respages[0]));
1076         /* Current request rpc header length*/
1077         entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1078                                 (char *)page_address(rqstp->rq_respages[0]);
1079 }
1080
1081 /*
1082  * We keep the rpc header, but take the nfs reply from the replycache.
1083  */
1084 static int
1085 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1086                         struct nfsd4_cache_entry *entry)
1087 {
1088         struct svc_rqst *rqstp = resp->rqstp;
1089         struct kvec *resv = &resp->rqstp->rq_res.head[0];
1090         int len;
1091
1092         /* Current request rpc header length*/
1093         len = (char *)resp->cstate.statp -
1094                         (char *)page_address(rqstp->rq_respages[0]);
1095         if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1096                 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1097                         entry->ce_datav.iov_len);
1098                 return 0;
1099         }
1100         /* copy the cached reply nfsd data past the current rpc header */
1101         memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1102                 entry->ce_datav.iov_len);
1103         resv->iov_len = len + entry->ce_datav.iov_len;
1104         return 1;
1105 }
1106
1107 /*
1108  * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1109  * cached page.  Replace any futher replay pages from the cache.
1110  */
1111 __be32
1112 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1113                          struct nfsd4_sequence *seq)
1114 {
1115         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1116         __be32 status;
1117
1118         dprintk("--> %s entry %p\n", __func__, entry);
1119
1120         /*
1121          * If this is just the sequence operation, we did not keep
1122          * a page in the cache entry because we can just use the
1123          * slot info stored in struct nfsd4_sequence that was checked
1124          * against the slot in nfsd4_sequence().
1125          *
1126          * This occurs when seq->cachethis is FALSE, or when the client
1127          * session inactivity timer fires and a solo sequence operation
1128          * is sent (lease renewal).
1129          */
1130         if (seq && nfsd4_not_cached(resp)) {
1131                 seq->maxslots = resp->cstate.session->se_fnumslots;
1132                 return nfs_ok;
1133         }
1134
1135         if (!nfsd41_copy_replay_data(resp, entry)) {
1136                 /*
1137                  * Not enough room to use the replay rpc header, send the
1138                  * cached header. Release all the allocated result pages.
1139                  */
1140                 svc_free_res_pages(resp->rqstp);
1141                 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1142                         entry->ce_resused);
1143         } else {
1144                 /* Release all but the first allocated result page */
1145
1146                 resp->rqstp->rq_resused--;
1147                 svc_free_res_pages(resp->rqstp);
1148
1149                 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1150                                  &entry->ce_respages[1],
1151                                  entry->ce_resused - 1);
1152         }
1153
1154         resp->rqstp->rq_resused = entry->ce_resused;
1155         resp->opcnt = entry->ce_opcnt;
1156         resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1157         status = entry->ce_status;
1158
1159         return status;
1160 }
1161
1162 /*
1163  * Set the exchange_id flags returned by the server.
1164  */
1165 static void
1166 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1167 {
1168         /* pNFS is not supported */
1169         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1170
1171         /* Referrals are supported, Migration is not. */
1172         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1173
1174         /* set the wire flags to return to client. */
1175         clid->flags = new->cl_exchange_flags;
1176 }
1177
1178 __be32
1179 nfsd4_exchange_id(struct svc_rqst *rqstp,
1180                   struct nfsd4_compound_state *cstate,
1181                   struct nfsd4_exchange_id *exid)
1182 {
1183         struct nfs4_client *unconf, *conf, *new;
1184         int status;
1185         unsigned int            strhashval;
1186         char                    dname[HEXDIR_LEN];
1187         nfs4_verifier           verf = exid->verifier;
1188         u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1189
1190         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1191                 " ip_addr=%u flags %x, spa_how %d\n",
1192                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1193                 ip_addr, exid->flags, exid->spa_how);
1194
1195         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1196                 return nfserr_inval;
1197
1198         /* Currently only support SP4_NONE */
1199         switch (exid->spa_how) {
1200         case SP4_NONE:
1201                 break;
1202         case SP4_SSV:
1203                 return nfserr_encr_alg_unsupp;
1204         default:
1205                 BUG();                          /* checked by xdr code */
1206         case SP4_MACH_CRED:
1207                 return nfserr_serverfault;      /* no excuse :-/ */
1208         }
1209
1210         status = nfs4_make_rec_clidname(dname, &exid->clname);
1211
1212         if (status)
1213                 goto error;
1214
1215         strhashval = clientstr_hashval(dname);
1216
1217         nfs4_lock_state();
1218         status = nfs_ok;
1219
1220         conf = find_confirmed_client_by_str(dname, strhashval, true);
1221         if (conf) {
1222                 if (!same_verf(&verf, &conf->cl_verifier)) {
1223                         /* 18.35.4 case 8 */
1224                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1225                                 status = nfserr_not_same;
1226                                 goto out;
1227                         }
1228                         /* Client reboot: destroy old state */
1229                         expire_client(conf);
1230                         goto out_new;
1231                 }
1232                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1233                         /* 18.35.4 case 9 */
1234                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1235                                 status = nfserr_perm;
1236                                 goto out;
1237                         }
1238                         expire_client(conf);
1239                         goto out_new;
1240                 }
1241                 if (ip_addr != conf->cl_addr &&
1242                     !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1243                         /* Client collision. 18.35.4 case 3 */
1244                         status = nfserr_clid_inuse;
1245                         goto out;
1246                 }
1247                 /*
1248                  * Set bit when the owner id and verifier map to an already
1249                  * confirmed client id (18.35.3).
1250                  */
1251                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1252
1253                 /*
1254                  * Falling into 18.35.4 case 2, possible router replay.
1255                  * Leave confirmed record intact and return same result.
1256                  */
1257                 copy_verf(conf, &verf);
1258                 new = conf;
1259                 goto out_copy;
1260         } else {
1261                 /* 18.35.4 case 7 */
1262                 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1263                         status = nfserr_noent;
1264                         goto out;
1265                 }
1266         }
1267
1268         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1269         if (unconf) {
1270                 /*
1271                  * Possible retry or client restart.  Per 18.35.4 case 4,
1272                  * a new unconfirmed record should be generated regardless
1273                  * of whether any properties have changed.
1274                  */
1275                 expire_client(unconf);
1276         }
1277
1278 out_new:
1279         /* Normal case */
1280         new = create_client(exid->clname, dname);
1281         if (new == NULL) {
1282                 status = nfserr_resource;
1283                 goto out;
1284         }
1285
1286         copy_verf(new, &verf);
1287         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1288         new->cl_addr = ip_addr;
1289         gen_clid(new);
1290         gen_confirm(new);
1291         add_to_unconfirmed(new, strhashval);
1292 out_copy:
1293         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1294         exid->clientid.cl_id = new->cl_clientid.cl_id;
1295
1296         new->cl_slot.sl_seqid = 0;
1297         exid->seqid = 1;
1298         nfsd4_set_ex_flags(new, exid);
1299
1300         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1301                 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1302         status = nfs_ok;
1303
1304 out:
1305         nfs4_unlock_state();
1306 error:
1307         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1308         return status;
1309 }
1310
1311 static int
1312 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1313 {
1314         dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1315                 slot->sl_seqid);
1316
1317         /* The slot is in use, and no response has been sent. */
1318         if (slot->sl_inuse) {
1319                 if (seqid == slot->sl_seqid)
1320                         return nfserr_jukebox;
1321                 else
1322                         return nfserr_seq_misordered;
1323         }
1324         /* Normal */
1325         if (likely(seqid == slot->sl_seqid + 1))
1326                 return nfs_ok;
1327         /* Replay */
1328         if (seqid == slot->sl_seqid)
1329                 return nfserr_replay_cache;
1330         /* Wraparound */
1331         if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1332                 return nfs_ok;
1333         /* Misordered replay or misordered new request */
1334         return nfserr_seq_misordered;
1335 }
1336
1337 __be32
1338 nfsd4_create_session(struct svc_rqst *rqstp,
1339                      struct nfsd4_compound_state *cstate,
1340                      struct nfsd4_create_session *cr_ses)
1341 {
1342         u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1343         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1344         struct nfs4_client *conf, *unconf;
1345         struct nfsd4_slot *slot = NULL;
1346         int status = 0;
1347
1348         nfs4_lock_state();
1349         unconf = find_unconfirmed_client(&cr_ses->clientid);
1350         conf = find_confirmed_client(&cr_ses->clientid);
1351
1352         if (conf) {
1353                 slot = &conf->cl_slot;
1354                 status = check_slot_seqid(cr_ses->seqid, slot);
1355                 if (status == nfserr_replay_cache) {
1356                         dprintk("Got a create_session replay! seqid= %d\n",
1357                                 slot->sl_seqid);
1358                         cstate->slot = slot;
1359                         cstate->status = status;
1360                         /* Return the cached reply status */
1361                         status = nfsd4_replay_cache_entry(resp, NULL);
1362                         goto out;
1363                 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1364                         status = nfserr_seq_misordered;
1365                         dprintk("Sequence misordered!\n");
1366                         dprintk("Expected seqid= %d but got seqid= %d\n",
1367                                 slot->sl_seqid, cr_ses->seqid);
1368                         goto out;
1369                 }
1370                 conf->cl_slot.sl_seqid++;
1371         } else if (unconf) {
1372                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1373                     (ip_addr != unconf->cl_addr)) {
1374                         status = nfserr_clid_inuse;
1375                         goto out;
1376                 }
1377
1378                 slot = &unconf->cl_slot;
1379                 status = check_slot_seqid(cr_ses->seqid, slot);
1380                 if (status) {
1381                         /* an unconfirmed replay returns misordered */
1382                         status = nfserr_seq_misordered;
1383                         goto out;
1384                 }
1385
1386                 slot->sl_seqid++; /* from 0 to 1 */
1387                 move_to_confirmed(unconf);
1388
1389                 /*
1390                  * We do not support RDMA or persistent sessions
1391                  */
1392                 cr_ses->flags &= ~SESSION4_PERSIST;
1393                 cr_ses->flags &= ~SESSION4_RDMA;
1394
1395                 conf = unconf;
1396         } else {
1397                 status = nfserr_stale_clientid;
1398                 goto out;
1399         }
1400
1401         status = alloc_init_session(rqstp, conf, cr_ses);
1402         if (status)
1403                 goto out;
1404
1405         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1406                NFS4_MAX_SESSIONID_LEN);
1407         cr_ses->seqid = slot->sl_seqid;
1408
1409         slot->sl_inuse = true;
1410         cstate->slot = slot;
1411         /* Ensure a page is used for the cache */
1412         slot->sl_cache_entry.ce_cachethis = 1;
1413 out:
1414         nfs4_unlock_state();
1415         dprintk("%s returns %d\n", __func__, ntohl(status));
1416         return status;
1417 }
1418
1419 __be32
1420 nfsd4_destroy_session(struct svc_rqst *r,
1421                       struct nfsd4_compound_state *cstate,
1422                       struct nfsd4_destroy_session *sessionid)
1423 {
1424         struct nfsd4_session *ses;
1425         u32 status = nfserr_badsession;
1426
1427         /* Notes:
1428          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1429          * - Should we return nfserr_back_chan_busy if waiting for
1430          *   callbacks on to-be-destroyed session?
1431          * - Do we need to clear any callback info from previous session?
1432          */
1433
1434         dump_sessionid(__func__, &sessionid->sessionid);
1435         spin_lock(&sessionid_lock);
1436         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1437         if (!ses) {
1438                 spin_unlock(&sessionid_lock);
1439                 goto out;
1440         }
1441
1442         unhash_session(ses);
1443         spin_unlock(&sessionid_lock);
1444
1445         /* wait for callbacks */
1446         shutdown_callback_client(ses->se_client);
1447         nfsd4_put_session(ses);
1448         status = nfs_ok;
1449 out:
1450         dprintk("%s returns %d\n", __func__, ntohl(status));
1451         return status;
1452 }
1453
1454 __be32
1455 nfsd4_sequence(struct svc_rqst *rqstp,
1456                struct nfsd4_compound_state *cstate,
1457                struct nfsd4_sequence *seq)
1458 {
1459         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1460         struct nfsd4_session *session;
1461         struct nfsd4_slot *slot;
1462         int status;
1463
1464         if (resp->opcnt != 1)
1465                 return nfserr_sequence_pos;
1466
1467         spin_lock(&sessionid_lock);
1468         status = nfserr_badsession;
1469         session = find_in_sessionid_hashtbl(&seq->sessionid);
1470         if (!session)
1471                 goto out;
1472
1473         status = nfserr_badslot;
1474         if (seq->slotid >= session->se_fnumslots)
1475                 goto out;
1476
1477         slot = &session->se_slots[seq->slotid];
1478         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1479
1480         status = check_slot_seqid(seq->seqid, slot);
1481         if (status == nfserr_replay_cache) {
1482                 cstate->slot = slot;
1483                 cstate->session = session;
1484                 /* Return the cached reply status and set cstate->status
1485                  * for nfsd4_svc_encode_compoundres processing */
1486                 status = nfsd4_replay_cache_entry(resp, seq);
1487                 cstate->status = nfserr_replay_cache;
1488                 goto replay_cache;
1489         }
1490         if (status)
1491                 goto out;
1492
1493         /* Success! bump slot seqid */
1494         slot->sl_inuse = true;
1495         slot->sl_seqid = seq->seqid;
1496         slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1497         /* Always set the cache entry cachethis for solo sequence */
1498         if (nfsd4_is_solo_sequence(resp))
1499                 slot->sl_cache_entry.ce_cachethis = 1;
1500
1501         cstate->slot = slot;
1502         cstate->session = session;
1503
1504 replay_cache:
1505         /* Renew the clientid on success and on replay.
1506          * Hold a session reference until done processing the compound:
1507          * nfsd4_put_session called only if the cstate slot is set.
1508          */
1509         renew_client(session->se_client);
1510         nfsd4_get_session(session);
1511 out:
1512         spin_unlock(&sessionid_lock);
1513         dprintk("%s: return %d\n", __func__, ntohl(status));
1514         return status;
1515 }
1516
1517 __be32
1518 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1519                   struct nfsd4_setclientid *setclid)
1520 {
1521         struct sockaddr_in      *sin = svc_addr_in(rqstp);
1522         struct xdr_netobj       clname = { 
1523                 .len = setclid->se_namelen,
1524                 .data = setclid->se_name,
1525         };
1526         nfs4_verifier           clverifier = setclid->se_verf;
1527         unsigned int            strhashval;
1528         struct nfs4_client      *conf, *unconf, *new;
1529         __be32                  status;
1530         char                    *princ;
1531         char                    dname[HEXDIR_LEN];
1532         
1533         if (!check_name(clname))
1534                 return nfserr_inval;
1535
1536         status = nfs4_make_rec_clidname(dname, &clname);
1537         if (status)
1538                 return status;
1539
1540         /* 
1541          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1542          * We get here on a DRC miss.
1543          */
1544
1545         strhashval = clientstr_hashval(dname);
1546
1547         nfs4_lock_state();
1548         conf = find_confirmed_client_by_str(dname, strhashval, false);
1549         if (conf) {
1550                 /* RFC 3530 14.2.33 CASE 0: */
1551                 status = nfserr_clid_inuse;
1552                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1553                         dprintk("NFSD: setclientid: string in use by client"
1554                                 " at %pI4\n", &conf->cl_addr);
1555                         goto out;
1556                 }
1557         }
1558         /*
1559          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1560          * has a description of SETCLIENTID request processing consisting
1561          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1562          */
1563         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1564         status = nfserr_resource;
1565         if (!conf) {
1566                 /*
1567                  * RFC 3530 14.2.33 CASE 4:
1568                  * placed first, because it is the normal case
1569                  */
1570                 if (unconf)
1571                         expire_client(unconf);
1572                 new = create_client(clname, dname);
1573                 if (new == NULL)
1574                         goto out;
1575                 gen_clid(new);
1576         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1577                 /*
1578                  * RFC 3530 14.2.33 CASE 1:
1579                  * probable callback update
1580                  */
1581                 if (unconf) {
1582                         /* Note this is removing unconfirmed {*x***},
1583                          * which is stronger than RFC recommended {vxc**}.
1584                          * This has the advantage that there is at most
1585                          * one {*x***} in either list at any time.
1586                          */
1587                         expire_client(unconf);
1588                 }
1589                 new = create_client(clname, dname);
1590                 if (new == NULL)
1591                         goto out;
1592                 copy_clid(new, conf);
1593         } else if (!unconf) {
1594                 /*
1595                  * RFC 3530 14.2.33 CASE 2:
1596                  * probable client reboot; state will be removed if
1597                  * confirmed.
1598                  */
1599                 new = create_client(clname, dname);
1600                 if (new == NULL)
1601                         goto out;
1602                 gen_clid(new);
1603         } else {
1604                 /*
1605                  * RFC 3530 14.2.33 CASE 3:
1606                  * probable client reboot; state will be removed if
1607                  * confirmed.
1608                  */
1609                 expire_client(unconf);
1610                 new = create_client(clname, dname);
1611                 if (new == NULL)
1612                         goto out;
1613                 gen_clid(new);
1614         }
1615         copy_verf(new, &clverifier);
1616         new->cl_addr = sin->sin_addr.s_addr;
1617         new->cl_flavor = rqstp->rq_flavor;
1618         princ = svc_gss_principal(rqstp);
1619         if (princ) {
1620                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1621                 if (new->cl_principal == NULL) {
1622                         free_client(new);
1623                         goto out;
1624                 }
1625         }
1626         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1627         gen_confirm(new);
1628         gen_callback(new, setclid);
1629         add_to_unconfirmed(new, strhashval);
1630         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1631         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1632         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1633         status = nfs_ok;
1634 out:
1635         nfs4_unlock_state();
1636         return status;
1637 }
1638
1639
1640 /*
1641  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1642  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1643  * bullets, labeled as CASE1 - CASE4 below.
1644  */
1645 __be32
1646 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1647                          struct nfsd4_compound_state *cstate,
1648                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1649 {
1650         struct sockaddr_in *sin = svc_addr_in(rqstp);
1651         struct nfs4_client *conf, *unconf;
1652         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1653         clientid_t * clid = &setclientid_confirm->sc_clientid;
1654         __be32 status;
1655
1656         if (STALE_CLIENTID(clid))
1657                 return nfserr_stale_clientid;
1658         /* 
1659          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1660          * We get here on a DRC miss.
1661          */
1662
1663         nfs4_lock_state();
1664
1665         conf = find_confirmed_client(clid);
1666         unconf = find_unconfirmed_client(clid);
1667
1668         status = nfserr_clid_inuse;
1669         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1670                 goto out;
1671         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1672                 goto out;
1673
1674         /*
1675          * section 14.2.34 of RFC 3530 has a description of
1676          * SETCLIENTID_CONFIRM request processing consisting
1677          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1678          */
1679         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1680                 /*
1681                  * RFC 3530 14.2.34 CASE 1:
1682                  * callback update
1683                  */
1684                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1685                         status = nfserr_clid_inuse;
1686                 else {
1687                         /* XXX: We just turn off callbacks until we can handle
1688                           * change request correctly. */
1689                         atomic_set(&conf->cl_callback.cb_set, 0);
1690                         gen_confirm(conf);
1691                         nfsd4_remove_clid_dir(unconf);
1692                         expire_client(unconf);
1693                         status = nfs_ok;
1694
1695                 }
1696         } else if (conf && !unconf) {
1697                 /*
1698                  * RFC 3530 14.2.34 CASE 2:
1699                  * probable retransmitted request; play it safe and
1700                  * do nothing.
1701                  */
1702                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1703                         status = nfserr_clid_inuse;
1704                 else
1705                         status = nfs_ok;
1706         } else if (!conf && unconf
1707                         && same_verf(&unconf->cl_confirm, &confirm)) {
1708                 /*
1709                  * RFC 3530 14.2.34 CASE 3:
1710                  * Normal case; new or rebooted client:
1711                  */
1712                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1713                         status = nfserr_clid_inuse;
1714                 } else {
1715                         unsigned int hash =
1716                                 clientstr_hashval(unconf->cl_recdir);
1717                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1718                                                             hash, false);
1719                         if (conf) {
1720                                 nfsd4_remove_clid_dir(conf);
1721                                 expire_client(conf);
1722                         }
1723                         move_to_confirmed(unconf);
1724                         conf = unconf;
1725                         nfsd4_probe_callback(conf);
1726                         status = nfs_ok;
1727                 }
1728         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1729             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1730                                                                 &confirm)))) {
1731                 /*
1732                  * RFC 3530 14.2.34 CASE 4:
1733                  * Client probably hasn't noticed that we rebooted yet.
1734                  */
1735                 status = nfserr_stale_clientid;
1736         } else {
1737                 /* check that we have hit one of the cases...*/
1738                 status = nfserr_clid_inuse;
1739         }
1740 out:
1741         nfs4_unlock_state();
1742         return status;
1743 }
1744
1745 /* OPEN Share state helper functions */
1746 static inline struct nfs4_file *
1747 alloc_init_file(struct inode *ino)
1748 {
1749         struct nfs4_file *fp;
1750         unsigned int hashval = file_hashval(ino);
1751
1752         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1753         if (fp) {
1754                 atomic_set(&fp->fi_ref, 1);
1755                 INIT_LIST_HEAD(&fp->fi_hash);
1756                 INIT_LIST_HEAD(&fp->fi_stateids);
1757                 INIT_LIST_HEAD(&fp->fi_delegations);
1758                 spin_lock(&recall_lock);
1759                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1760                 spin_unlock(&recall_lock);
1761                 fp->fi_inode = igrab(ino);
1762                 fp->fi_id = current_fileid++;
1763                 fp->fi_had_conflict = false;
1764                 return fp;
1765         }
1766         return NULL;
1767 }
1768
1769 static void
1770 nfsd4_free_slab(struct kmem_cache **slab)
1771 {
1772         if (*slab == NULL)
1773                 return;
1774         kmem_cache_destroy(*slab);
1775         *slab = NULL;
1776 }
1777
1778 void
1779 nfsd4_free_slabs(void)
1780 {
1781         nfsd4_free_slab(&stateowner_slab);
1782         nfsd4_free_slab(&file_slab);
1783         nfsd4_free_slab(&stateid_slab);
1784         nfsd4_free_slab(&deleg_slab);
1785 }
1786
1787 static int
1788 nfsd4_init_slabs(void)
1789 {
1790         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1791                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1792         if (stateowner_slab == NULL)
1793                 goto out_nomem;
1794         file_slab = kmem_cache_create("nfsd4_files",
1795                         sizeof(struct nfs4_file), 0, 0, NULL);
1796         if (file_slab == NULL)
1797                 goto out_nomem;
1798         stateid_slab = kmem_cache_create("nfsd4_stateids",
1799                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1800         if (stateid_slab == NULL)
1801                 goto out_nomem;
1802         deleg_slab = kmem_cache_create("nfsd4_delegations",
1803                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1804         if (deleg_slab == NULL)
1805                 goto out_nomem;
1806         return 0;
1807 out_nomem:
1808         nfsd4_free_slabs();
1809         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1810         return -ENOMEM;
1811 }
1812
1813 void
1814 nfs4_free_stateowner(struct kref *kref)
1815 {
1816         struct nfs4_stateowner *sop =
1817                 container_of(kref, struct nfs4_stateowner, so_ref);
1818         kfree(sop->so_owner.data);
1819         kmem_cache_free(stateowner_slab, sop);
1820 }
1821
1822 static inline struct nfs4_stateowner *
1823 alloc_stateowner(struct xdr_netobj *owner)
1824 {
1825         struct nfs4_stateowner *sop;
1826
1827         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1828                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1829                         memcpy(sop->so_owner.data, owner->data, owner->len);
1830                         sop->so_owner.len = owner->len;
1831                         kref_init(&sop->so_ref);
1832                         return sop;
1833                 } 
1834                 kmem_cache_free(stateowner_slab, sop);
1835         }
1836         return NULL;
1837 }
1838
1839 static struct nfs4_stateowner *
1840 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1841         struct nfs4_stateowner *sop;
1842         struct nfs4_replay *rp;
1843         unsigned int idhashval;
1844
1845         if (!(sop = alloc_stateowner(&open->op_owner)))
1846                 return NULL;
1847         idhashval = ownerid_hashval(current_ownerid);
1848         INIT_LIST_HEAD(&sop->so_idhash);
1849         INIT_LIST_HEAD(&sop->so_strhash);
1850         INIT_LIST_HEAD(&sop->so_perclient);
1851         INIT_LIST_HEAD(&sop->so_stateids);
1852         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1853         INIT_LIST_HEAD(&sop->so_close_lru);
1854         sop->so_time = 0;
1855         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1856         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1857         list_add(&sop->so_perclient, &clp->cl_openowners);
1858         sop->so_is_open_owner = 1;
1859         sop->so_id = current_ownerid++;
1860         sop->so_client = clp;
1861         sop->so_seqid = open->op_seqid;
1862         sop->so_confirmed = 0;
1863         rp = &sop->so_replay;
1864         rp->rp_status = nfserr_serverfault;
1865         rp->rp_buflen = 0;
1866         rp->rp_buf = rp->rp_ibuf;
1867         return sop;
1868 }
1869
1870 static inline void
1871 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1872         struct nfs4_stateowner *sop = open->op_stateowner;
1873         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1874
1875         INIT_LIST_HEAD(&stp->st_hash);
1876         INIT_LIST_HEAD(&stp->st_perstateowner);
1877         INIT_LIST_HEAD(&stp->st_lockowners);
1878         INIT_LIST_HEAD(&stp->st_perfile);
1879         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1880         list_add(&stp->st_perstateowner, &sop->so_stateids);
1881         list_add(&stp->st_perfile, &fp->fi_stateids);
1882         stp->st_stateowner = sop;
1883         get_nfs4_file(fp);
1884         stp->st_file = fp;
1885         stp->st_stateid.si_boot = boot_time;
1886         stp->st_stateid.si_stateownerid = sop->so_id;
1887         stp->st_stateid.si_fileid = fp->fi_id;
1888         stp->st_stateid.si_generation = 0;
1889         stp->st_access_bmap = 0;
1890         stp->st_deny_bmap = 0;
1891         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1892                   &stp->st_access_bmap);
1893         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1894         stp->st_openstp = NULL;
1895 }
1896
1897 static void
1898 move_to_close_lru(struct nfs4_stateowner *sop)
1899 {
1900         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1901
1902         list_move_tail(&sop->so_close_lru, &close_lru);
1903         sop->so_time = get_seconds();
1904 }
1905
1906 static int
1907 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1908                                                         clientid_t *clid)
1909 {
1910         return (sop->so_owner.len == owner->len) &&
1911                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1912                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1913 }
1914
1915 static struct nfs4_stateowner *
1916 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1917 {
1918         struct nfs4_stateowner *so = NULL;
1919
1920         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1921                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1922                         return so;
1923         }
1924         return NULL;
1925 }
1926
1927 /* search file_hashtbl[] for file */
1928 static struct nfs4_file *
1929 find_file(struct inode *ino)
1930 {
1931         unsigned int hashval = file_hashval(ino);
1932         struct nfs4_file *fp;
1933
1934         spin_lock(&recall_lock);
1935         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1936                 if (fp->fi_inode == ino) {
1937                         get_nfs4_file(fp);
1938                         spin_unlock(&recall_lock);
1939                         return fp;
1940                 }
1941         }
1942         spin_unlock(&recall_lock);
1943         return NULL;
1944 }
1945
1946 static inline int access_valid(u32 x, u32 minorversion)
1947 {
1948         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1949                 return 0;
1950         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1951                 return 0;
1952         x &= ~NFS4_SHARE_ACCESS_MASK;
1953         if (minorversion && x) {
1954                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1955                         return 0;
1956                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1957                         return 0;
1958                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1959         }
1960         if (x)
1961                 return 0;
1962         return 1;
1963 }
1964
1965 static inline int deny_valid(u32 x)
1966 {
1967         /* Note: unlike access bits, deny bits may be zero. */
1968         return x <= NFS4_SHARE_DENY_BOTH;
1969 }
1970
1971 /*
1972  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1973  * st_{access,deny}_bmap field of the stateid, in order to track not
1974  * only what share bits are currently in force, but also what
1975  * combinations of share bits previous opens have used.  This allows us
1976  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1977  * return an error if the client attempt to downgrade to a combination
1978  * of share bits not explicable by closing some of its previous opens.
1979  *
1980  * XXX: This enforcement is actually incomplete, since we don't keep
1981  * track of access/deny bit combinations; so, e.g., we allow:
1982  *
1983  *      OPEN allow read, deny write
1984  *      OPEN allow both, deny none
1985  *      DOWNGRADE allow read, deny none
1986  *
1987  * which we should reject.
1988  */
1989 static void
1990 set_access(unsigned int *access, unsigned long bmap) {
1991         int i;
1992
1993         *access = 0;
1994         for (i = 1; i < 4; i++) {
1995                 if (test_bit(i, &bmap))
1996                         *access |= i;
1997         }
1998 }
1999
2000 static void
2001 set_deny(unsigned int *deny, unsigned long bmap) {
2002         int i;
2003
2004         *deny = 0;
2005         for (i = 0; i < 4; i++) {
2006                 if (test_bit(i, &bmap))
2007                         *deny |= i ;
2008         }
2009 }
2010
2011 static int
2012 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2013         unsigned int access, deny;
2014
2015         set_access(&access, stp->st_access_bmap);
2016         set_deny(&deny, stp->st_deny_bmap);
2017         if ((access & open->op_share_deny) || (deny & open->op_share_access))
2018                 return 0;
2019         return 1;
2020 }
2021
2022 /*
2023  * Called to check deny when READ with all zero stateid or
2024  * WRITE with all zero or all one stateid
2025  */
2026 static __be32
2027 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2028 {
2029         struct inode *ino = current_fh->fh_dentry->d_inode;
2030         struct nfs4_file *fp;
2031         struct nfs4_stateid *stp;
2032         __be32 ret;
2033
2034         dprintk("NFSD: nfs4_share_conflict\n");
2035
2036         fp = find_file(ino);
2037         if (!fp)
2038                 return nfs_ok;
2039         ret = nfserr_locked;
2040         /* Search for conflicting share reservations */
2041         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2042                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2043                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2044                         goto out;
2045         }
2046         ret = nfs_ok;
2047 out:
2048         put_nfs4_file(fp);
2049         return ret;
2050 }
2051
2052 static inline void
2053 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2054 {
2055         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2056                 drop_file_write_access(filp);
2057                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2058         }
2059 }
2060
2061 /*
2062  * Recall a delegation
2063  */
2064 static int
2065 do_recall(void *__dp)
2066 {
2067         struct nfs4_delegation *dp = __dp;
2068
2069         dp->dl_file->fi_had_conflict = true;
2070         nfsd4_cb_recall(dp);
2071         return 0;
2072 }
2073
2074 /*
2075  * Spawn a thread to perform a recall on the delegation represented
2076  * by the lease (file_lock)
2077  *
2078  * Called from break_lease() with lock_kernel() held.
2079  * Note: we assume break_lease will only call this *once* for any given
2080  * lease.
2081  */
2082 static
2083 void nfsd_break_deleg_cb(struct file_lock *fl)
2084 {
2085         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
2086         struct task_struct *t;
2087
2088         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2089         if (!dp)
2090                 return;
2091
2092         /* We're assuming the state code never drops its reference
2093          * without first removing the lease.  Since we're in this lease
2094          * callback (and since the lease code is serialized by the kernel
2095          * lock) we know the server hasn't removed the lease yet, we know
2096          * it's safe to take a reference: */
2097         atomic_inc(&dp->dl_count);
2098         atomic_inc(&dp->dl_client->cl_count);
2099
2100         spin_lock(&recall_lock);
2101         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2102         spin_unlock(&recall_lock);
2103
2104         /* only place dl_time is set. protected by lock_kernel*/
2105         dp->dl_time = get_seconds();
2106
2107         /*
2108          * We don't want the locks code to timeout the lease for us;
2109          * we'll remove it ourself if the delegation isn't returned
2110          * in time.
2111          */
2112         fl->fl_break_time = 0;
2113
2114         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2115         if (IS_ERR(t)) {
2116                 struct nfs4_client *clp = dp->dl_client;
2117
2118                 printk(KERN_INFO "NFSD: Callback thread failed for "
2119                         "for client (clientid %08x/%08x)\n",
2120                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2121                 put_nfs4_client(dp->dl_client);
2122                 nfs4_put_delegation(dp);
2123         }
2124 }
2125
2126 /*
2127  * The file_lock is being reapd.
2128  *
2129  * Called by locks_free_lock() with lock_kernel() held.
2130  */
2131 static
2132 void nfsd_release_deleg_cb(struct file_lock *fl)
2133 {
2134         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2135
2136         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2137
2138         if (!(fl->fl_flags & FL_LEASE) || !dp)
2139                 return;
2140         dp->dl_flock = NULL;
2141 }
2142
2143 /*
2144  * Set the delegation file_lock back pointer.
2145  *
2146  * Called from setlease() with lock_kernel() held.
2147  */
2148 static
2149 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2150 {
2151         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2152
2153         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2154         if (!dp)
2155                 return;
2156         dp->dl_flock = new;
2157 }
2158
2159 /*
2160  * Called from setlease() with lock_kernel() held
2161  */
2162 static
2163 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2164 {
2165         struct nfs4_delegation *onlistd =
2166                 (struct nfs4_delegation *)onlist->fl_owner;
2167         struct nfs4_delegation *tryd =
2168                 (struct nfs4_delegation *)try->fl_owner;
2169
2170         if (onlist->fl_lmops != try->fl_lmops)
2171                 return 0;
2172
2173         return onlistd->dl_client == tryd->dl_client;
2174 }
2175
2176
2177 static
2178 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2179 {
2180         if (arg & F_UNLCK)
2181                 return lease_modify(onlist, arg);
2182         else
2183                 return -EAGAIN;
2184 }
2185
2186 static struct lock_manager_operations nfsd_lease_mng_ops = {
2187         .fl_break = nfsd_break_deleg_cb,
2188         .fl_release_private = nfsd_release_deleg_cb,
2189         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2190         .fl_mylease = nfsd_same_client_deleg_cb,
2191         .fl_change = nfsd_change_deleg_cb,
2192 };
2193
2194
2195 __be32
2196 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2197                     struct nfsd4_open *open)
2198 {
2199         clientid_t *clientid = &open->op_clientid;
2200         struct nfs4_client *clp = NULL;
2201         unsigned int strhashval;
2202         struct nfs4_stateowner *sop = NULL;
2203
2204         if (!check_name(open->op_owner))
2205                 return nfserr_inval;
2206
2207         if (STALE_CLIENTID(&open->op_clientid))
2208                 return nfserr_stale_clientid;
2209
2210         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2211         sop = find_openstateowner_str(strhashval, open);
2212         open->op_stateowner = sop;
2213         if (!sop) {
2214                 /* Make sure the client's lease hasn't expired. */
2215                 clp = find_confirmed_client(clientid);
2216                 if (clp == NULL)
2217                         return nfserr_expired;
2218                 goto renew;
2219         }
2220         /* When sessions are used, skip open sequenceid processing */
2221         if (nfsd4_has_session(cstate))
2222                 goto renew;
2223         if (!sop->so_confirmed) {
2224                 /* Replace unconfirmed owners without checking for replay. */
2225                 clp = sop->so_client;
2226                 release_openowner(sop);
2227                 open->op_stateowner = NULL;
2228                 goto renew;
2229         }
2230         if (open->op_seqid == sop->so_seqid - 1) {
2231                 if (sop->so_replay.rp_buflen)
2232                         return nfserr_replay_me;
2233                 /* The original OPEN failed so spectacularly
2234                  * that we don't even have replay data saved!
2235                  * Therefore, we have no choice but to continue
2236                  * processing this OPEN; presumably, we'll
2237                  * fail again for the same reason.
2238                  */
2239                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2240                 goto renew;
2241         }
2242         if (open->op_seqid != sop->so_seqid)
2243                 return nfserr_bad_seqid;
2244 renew:
2245         if (open->op_stateowner == NULL) {
2246                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2247                 if (sop == NULL)
2248                         return nfserr_resource;
2249                 open->op_stateowner = sop;
2250         }
2251         list_del_init(&sop->so_close_lru);
2252         renew_client(sop->so_client);
2253         return nfs_ok;
2254 }
2255
2256 static inline __be32
2257 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2258 {
2259         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2260                 return nfserr_openmode;
2261         else
2262                 return nfs_ok;
2263 }
2264
2265 static struct nfs4_delegation *
2266 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2267 {
2268         struct nfs4_delegation *dp;
2269
2270         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2271                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2272                         return dp;
2273         }
2274         return NULL;
2275 }
2276
2277 static __be32
2278 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2279                 struct nfs4_delegation **dp)
2280 {
2281         int flags;
2282         __be32 status = nfserr_bad_stateid;
2283
2284         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2285         if (*dp == NULL)
2286                 goto out;
2287         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2288                                                 RD_STATE : WR_STATE;
2289         status = nfs4_check_delegmode(*dp, flags);
2290         if (status)
2291                 *dp = NULL;
2292 out:
2293         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2294                 return nfs_ok;
2295         if (status)
2296                 return status;
2297         open->op_stateowner->so_confirmed = 1;
2298         return nfs_ok;
2299 }
2300
2301 static __be32
2302 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2303 {
2304         struct nfs4_stateid *local;
2305         __be32 status = nfserr_share_denied;
2306         struct nfs4_stateowner *sop = open->op_stateowner;
2307
2308         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2309                 /* ignore lock owners */
2310                 if (local->st_stateowner->so_is_open_owner == 0)
2311                         continue;
2312                 /* remember if we have seen this open owner */
2313                 if (local->st_stateowner == sop)
2314                         *stpp = local;
2315                 /* check for conflicting share reservations */
2316                 if (!test_share(local, open))
2317                         goto out;
2318         }
2319         status = 0;
2320 out:
2321         return status;
2322 }
2323
2324 static inline struct nfs4_stateid *
2325 nfs4_alloc_stateid(void)
2326 {
2327         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2328 }
2329
2330 static __be32
2331 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2332                 struct nfs4_delegation *dp,
2333                 struct svc_fh *cur_fh, int flags)
2334 {
2335         struct nfs4_stateid *stp;
2336
2337         stp = nfs4_alloc_stateid();
2338         if (stp == NULL)
2339                 return nfserr_resource;
2340
2341         if (dp) {
2342                 get_file(dp->dl_vfs_file);
2343                 stp->st_vfs_file = dp->dl_vfs_file;
2344         } else {
2345                 __be32 status;
2346                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2347                                 &stp->st_vfs_file);
2348                 if (status) {
2349                         if (status == nfserr_dropit)
2350                                 status = nfserr_jukebox;
2351                         kmem_cache_free(stateid_slab, stp);
2352                         return status;
2353                 }
2354         }
2355         *stpp = stp;
2356         return 0;
2357 }
2358
2359 static inline __be32
2360 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2361                 struct nfsd4_open *open)
2362 {
2363         struct iattr iattr = {
2364                 .ia_valid = ATTR_SIZE,
2365                 .ia_size = 0,
2366         };
2367         if (!open->op_truncate)
2368                 return 0;
2369         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2370                 return nfserr_inval;
2371         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2372 }
2373
2374 static __be32
2375 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2376 {
2377         struct file *filp = stp->st_vfs_file;
2378         struct inode *inode = filp->f_path.dentry->d_inode;
2379         unsigned int share_access, new_writer;
2380         __be32 status;
2381
2382         set_access(&share_access, stp->st_access_bmap);
2383         new_writer = (~share_access) & open->op_share_access
2384                         & NFS4_SHARE_ACCESS_WRITE;
2385
2386         if (new_writer) {
2387                 int err = get_write_access(inode);
2388                 if (err)
2389                         return nfserrno(err);
2390                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2391                 if (err)
2392                         return nfserrno(err);
2393                 file_take_write(filp);
2394         }
2395         status = nfsd4_truncate(rqstp, cur_fh, open);
2396         if (status) {
2397                 if (new_writer)
2398                         put_write_access(inode);
2399                 return status;
2400         }
2401         /* remember the open */
2402         filp->f_mode |= open->op_share_access;
2403         __set_bit(open->op_share_access, &stp->st_access_bmap);
2404         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2405
2406         return nfs_ok;
2407 }
2408
2409
2410 static void
2411 nfs4_set_claim_prev(struct nfsd4_open *open)
2412 {
2413         open->op_stateowner->so_confirmed = 1;
2414         open->op_stateowner->so_client->cl_firststate = 1;
2415 }
2416
2417 /*
2418  * Attempt to hand out a delegation.
2419  */
2420 static void
2421 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2422 {
2423         struct nfs4_delegation *dp;
2424         struct nfs4_stateowner *sop = stp->st_stateowner;
2425         struct nfs4_callback *cb = &sop->so_client->cl_callback;
2426         struct file_lock fl, *flp = &fl;
2427         int status, flag = 0;
2428
2429         flag = NFS4_OPEN_DELEGATE_NONE;
2430         open->op_recall = 0;
2431         switch (open->op_claim_type) {
2432                 case NFS4_OPEN_CLAIM_PREVIOUS:
2433                         if (!atomic_read(&cb->cb_set))
2434                                 open->op_recall = 1;
2435                         flag = open->op_delegate_type;
2436                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2437                                 goto out;
2438                         break;
2439                 case NFS4_OPEN_CLAIM_NULL:
2440                         /* Let's not give out any delegations till everyone's
2441                          * had the chance to reclaim theirs.... */
2442                         if (locks_in_grace())
2443                                 goto out;
2444                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2445                                 goto out;
2446                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2447                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2448                         else
2449                                 flag = NFS4_OPEN_DELEGATE_READ;
2450                         break;
2451                 default:
2452                         goto out;
2453         }
2454
2455         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2456         if (dp == NULL) {
2457                 flag = NFS4_OPEN_DELEGATE_NONE;
2458                 goto out;
2459         }
2460         locks_init_lock(&fl);
2461         fl.fl_lmops = &nfsd_lease_mng_ops;
2462         fl.fl_flags = FL_LEASE;
2463         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2464         fl.fl_end = OFFSET_MAX;
2465         fl.fl_owner =  (fl_owner_t)dp;
2466         fl.fl_file = stp->st_vfs_file;
2467         fl.fl_pid = current->tgid;
2468
2469         /* vfs_setlease checks to see if delegation should be handed out.
2470          * the lock_manager callbacks fl_mylease and fl_change are used
2471          */
2472         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2473                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2474                 unhash_delegation(dp);
2475                 flag = NFS4_OPEN_DELEGATE_NONE;
2476                 goto out;
2477         }
2478
2479         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2480
2481         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2482                      dp->dl_stateid.si_boot,
2483                      dp->dl_stateid.si_stateownerid,
2484                      dp->dl_stateid.si_fileid,
2485                      dp->dl_stateid.si_generation);
2486 out:
2487         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2488                         && flag == NFS4_OPEN_DELEGATE_NONE
2489                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2490                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2491         open->op_delegate_type = flag;
2492 }
2493
2494 /*
2495  * called with nfs4_lock_state() held.
2496  */
2497 __be32
2498 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2499 {
2500         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2501         struct nfs4_file *fp = NULL;
2502         struct inode *ino = current_fh->fh_dentry->d_inode;
2503         struct nfs4_stateid *stp = NULL;
2504         struct nfs4_delegation *dp = NULL;
2505         __be32 status;
2506
2507         status = nfserr_inval;
2508         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2509                         || !deny_valid(open->op_share_deny))
2510                 goto out;
2511         /*
2512          * Lookup file; if found, lookup stateid and check open request,
2513          * and check for delegations in the process of being recalled.
2514          * If not found, create the nfs4_file struct
2515          */
2516         fp = find_file(ino);
2517         if (fp) {
2518                 if ((status = nfs4_check_open(fp, open, &stp)))
2519                         goto out;
2520                 status = nfs4_check_deleg(fp, open, &dp);
2521                 if (status)
2522                         goto out;
2523         } else {
2524                 status = nfserr_bad_stateid;
2525                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2526                         goto out;
2527                 status = nfserr_resource;
2528                 fp = alloc_init_file(ino);
2529                 if (fp == NULL)
2530                         goto out;
2531         }
2532
2533         /*
2534          * OPEN the file, or upgrade an existing OPEN.
2535          * If truncate fails, the OPEN fails.
2536          */
2537         if (stp) {
2538                 /* Stateid was found, this is an OPEN upgrade */
2539                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2540                 if (status)
2541                         goto out;
2542                 update_stateid(&stp->st_stateid);
2543         } else {
2544                 /* Stateid was not found, this is a new OPEN */
2545                 int flags = 0;
2546                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2547                         flags |= NFSD_MAY_READ;
2548                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2549                         flags |= NFSD_MAY_WRITE;
2550                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2551                 if (status)
2552                         goto out;
2553                 init_stateid(stp, fp, open);
2554                 status = nfsd4_truncate(rqstp, current_fh, open);
2555                 if (status) {
2556                         release_open_stateid(stp);
2557                         goto out;
2558                 }
2559                 if (nfsd4_has_session(&resp->cstate))
2560                         update_stateid(&stp->st_stateid);
2561         }
2562         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2563
2564         if (nfsd4_has_session(&resp->cstate))
2565                 open->op_stateowner->so_confirmed = 1;
2566
2567         /*
2568         * Attempt to hand out a delegation. No error return, because the
2569         * OPEN succeeds even if we fail.
2570         */
2571         nfs4_open_delegation(current_fh, open, stp);
2572
2573         status = nfs_ok;
2574
2575         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2576                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2577                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2578 out:
2579         if (fp)
2580                 put_nfs4_file(fp);
2581         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2582                 nfs4_set_claim_prev(open);
2583         /*
2584         * To finish the open response, we just need to set the rflags.
2585         */
2586         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2587         if (!open->op_stateowner->so_confirmed &&
2588             !nfsd4_has_session(&resp->cstate))
2589                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2590
2591         return status;
2592 }
2593
2594 __be32
2595 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2596             clientid_t *clid)
2597 {
2598         struct nfs4_client *clp;
2599         __be32 status;
2600
2601         nfs4_lock_state();
2602         dprintk("process_renew(%08x/%08x): starting\n", 
2603                         clid->cl_boot, clid->cl_id);
2604         status = nfserr_stale_clientid;
2605         if (STALE_CLIENTID(clid))
2606                 goto out;
2607         clp = find_confirmed_client(clid);
2608         status = nfserr_expired;
2609         if (clp == NULL) {
2610                 /* We assume the client took too long to RENEW. */
2611                 dprintk("nfsd4_renew: clientid not found!\n");
2612                 goto out;
2613         }
2614         renew_client(clp);
2615         status = nfserr_cb_path_down;
2616         if (!list_empty(&clp->cl_delegations)
2617                         && !atomic_read(&clp->cl_callback.cb_set))
2618                 goto out;
2619         status = nfs_ok;
2620 out:
2621         nfs4_unlock_state();
2622         return status;
2623 }
2624
2625 struct lock_manager nfsd4_manager = {
2626 };
2627
2628 static void
2629 nfsd4_end_grace(void)
2630 {
2631         dprintk("NFSD: end of grace period\n");
2632         nfsd4_recdir_purge_old();
2633         locks_end_grace(&nfsd4_manager);
2634 }
2635
2636 static time_t
2637 nfs4_laundromat(void)
2638 {
2639         struct nfs4_client *clp;
2640         struct nfs4_stateowner *sop;
2641         struct nfs4_delegation *dp;
2642         struct list_head *pos, *next, reaplist;
2643         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2644         time_t t, clientid_val = NFSD_LEASE_TIME;
2645         time_t u, test_val = NFSD_LEASE_TIME;
2646
2647         nfs4_lock_state();
2648
2649         dprintk("NFSD: laundromat service - starting\n");
2650         if (locks_in_grace())
2651                 nfsd4_end_grace();
2652         list_for_each_safe(pos, next, &client_lru) {
2653                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2654                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2655                         t = clp->cl_time - cutoff;
2656                         if (clientid_val > t)
2657                                 clientid_val = t;
2658                         break;
2659                 }
2660                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2661                         clp->cl_clientid.cl_id);
2662                 nfsd4_remove_clid_dir(clp);
2663                 expire_client(clp);
2664         }
2665         INIT_LIST_HEAD(&reaplist);
2666         spin_lock(&recall_lock);
2667         list_for_each_safe(pos, next, &del_recall_lru) {
2668                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2669                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2670                         u = dp->dl_time - cutoff;
2671                         if (test_val > u)
2672                                 test_val = u;
2673                         break;
2674                 }
2675                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2676                                     dp, dp->dl_flock);
2677                 list_move(&dp->dl_recall_lru, &reaplist);
2678         }
2679         spin_unlock(&recall_lock);
2680         list_for_each_safe(pos, next, &reaplist) {
2681                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2682                 list_del_init(&dp->dl_recall_lru);
2683                 unhash_delegation(dp);
2684         }
2685         test_val = NFSD_LEASE_TIME;
2686         list_for_each_safe(pos, next, &close_lru) {
2687                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2688                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2689                         u = sop->so_time - cutoff;
2690                         if (test_val > u)
2691                                 test_val = u;
2692                         break;
2693                 }
2694                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2695                         sop->so_id);
2696                 release_openowner(sop);
2697         }
2698         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2699                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2700         nfs4_unlock_state();
2701         return clientid_val;
2702 }
2703
2704 static struct workqueue_struct *laundry_wq;
2705 static void laundromat_main(struct work_struct *);
2706 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2707
2708 static void
2709 laundromat_main(struct work_struct *not_used)
2710 {
2711         time_t t;
2712
2713         t = nfs4_laundromat();
2714         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2715         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2716 }
2717
2718 static struct nfs4_stateowner *
2719 search_close_lru(u32 st_id, int flags)
2720 {
2721         struct nfs4_stateowner *local = NULL;
2722
2723         if (flags & CLOSE_STATE) {
2724                 list_for_each_entry(local, &close_lru, so_close_lru) {
2725                         if (local->so_id == st_id)
2726                                 return local;
2727                 }
2728         }
2729         return NULL;
2730 }
2731
2732 static inline int
2733 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2734 {
2735         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2736 }
2737
2738 static int
2739 STALE_STATEID(stateid_t *stateid)
2740 {
2741         if (stateid->si_boot == boot_time)
2742                 return 0;
2743         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2744                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2745                 stateid->si_generation);
2746         return 1;
2747 }
2748
2749 static inline int
2750 access_permit_read(unsigned long access_bmap)
2751 {
2752         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2753                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2754                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2755 }
2756
2757 static inline int
2758 access_permit_write(unsigned long access_bmap)
2759 {
2760         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2761                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2762 }
2763
2764 static
2765 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2766 {
2767         __be32 status = nfserr_openmode;
2768
2769         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2770                 goto out;
2771         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2772                 goto out;
2773         status = nfs_ok;
2774 out:
2775         return status;
2776 }
2777
2778 static inline __be32
2779 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2780 {
2781         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2782                 return nfs_ok;
2783         else if (locks_in_grace()) {
2784                 /* Answer in remaining cases depends on existance of
2785                  * conflicting state; so we must wait out the grace period. */
2786                 return nfserr_grace;
2787         } else if (flags & WR_STATE)
2788                 return nfs4_share_conflict(current_fh,
2789                                 NFS4_SHARE_DENY_WRITE);
2790         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2791                 return nfs4_share_conflict(current_fh,
2792                                 NFS4_SHARE_DENY_READ);
2793 }
2794
2795 /*
2796  * Allow READ/WRITE during grace period on recovered state only for files
2797  * that are not able to provide mandatory locking.
2798  */
2799 static inline int
2800 grace_disallows_io(struct inode *inode)
2801 {
2802         return locks_in_grace() && mandatory_lock(inode);
2803 }
2804
2805 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2806 {
2807         /*
2808          * When sessions are used the stateid generation number is ignored
2809          * when it is zero.
2810          */
2811         if ((flags & HAS_SESSION) && in->si_generation == 0)
2812                 goto out;
2813
2814         /* If the client sends us a stateid from the future, it's buggy: */
2815         if (in->si_generation > ref->si_generation)
2816                 return nfserr_bad_stateid;
2817         /*
2818          * The following, however, can happen.  For example, if the
2819          * client sends an open and some IO at the same time, the open
2820          * may bump si_generation while the IO is still in flight.
2821          * Thanks to hard links and renames, the client never knows what
2822          * file an open will affect.  So it could avoid that situation
2823          * only by serializing all opens and IO from the same open
2824          * owner.  To recover from the old_stateid error, the client
2825          * will just have to retry the IO:
2826          */
2827         if (in->si_generation < ref->si_generation)
2828                 return nfserr_old_stateid;
2829 out:
2830         return nfs_ok;
2831 }
2832
2833 static int is_delegation_stateid(stateid_t *stateid)
2834 {
2835         return stateid->si_fileid == 0;
2836 }
2837
2838 /*
2839 * Checks for stateid operations
2840 */
2841 __be32
2842 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2843                            stateid_t *stateid, int flags, struct file **filpp)
2844 {
2845         struct nfs4_stateid *stp = NULL;
2846         struct nfs4_delegation *dp = NULL;
2847         struct svc_fh *current_fh = &cstate->current_fh;
2848         struct inode *ino = current_fh->fh_dentry->d_inode;
2849         __be32 status;
2850
2851         if (filpp)
2852                 *filpp = NULL;
2853
2854         if (grace_disallows_io(ino))
2855                 return nfserr_grace;
2856
2857         if (nfsd4_has_session(cstate))
2858                 flags |= HAS_SESSION;
2859
2860         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2861                 return check_special_stateids(current_fh, stateid, flags);
2862
2863         status = nfserr_stale_stateid;
2864         if (STALE_STATEID(stateid)) 
2865                 goto out;
2866
2867         status = nfserr_bad_stateid;
2868         if (is_delegation_stateid(stateid)) {
2869                 dp = find_delegation_stateid(ino, stateid);
2870                 if (!dp)
2871                         goto out;
2872                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2873                                                   flags);
2874                 if (status)
2875                         goto out;
2876                 status = nfs4_check_delegmode(dp, flags);
2877                 if (status)
2878                         goto out;
2879                 renew_client(dp->dl_client);
2880                 if (filpp)
2881                         *filpp = dp->dl_vfs_file;
2882         } else { /* open or lock stateid */
2883                 stp = find_stateid(stateid, flags);
2884                 if (!stp)
2885                         goto out;
2886                 if (nfs4_check_fh(current_fh, stp))
2887                         goto out;
2888                 if (!stp->st_stateowner->so_confirmed)
2889                         goto out;
2890                 status = check_stateid_generation(stateid, &stp->st_stateid,
2891                                                   flags);
2892                 if (status)
2893                         goto out;
2894                 status = nfs4_check_openmode(stp, flags);
2895                 if (status)
2896                         goto out;
2897                 renew_client(stp->st_stateowner->so_client);
2898                 if (filpp)
2899                         *filpp = stp->st_vfs_file;
2900         }
2901         status = nfs_ok;
2902 out:
2903         return status;
2904 }
2905
2906 static inline int
2907 setlkflg (int type)
2908 {
2909         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2910                 RD_STATE : WR_STATE;
2911 }
2912
2913 /* 
2914  * Checks for sequence id mutating operations. 
2915  */
2916 static __be32
2917 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2918                          stateid_t *stateid, int flags,
2919                          struct nfs4_stateowner **sopp,
2920                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2921 {
2922         struct nfs4_stateid *stp;
2923         struct nfs4_stateowner *sop;
2924         struct svc_fh *current_fh = &cstate->current_fh;
2925         __be32 status;
2926
2927         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2928                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2929                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2930                 stateid->si_generation);
2931
2932         *stpp = NULL;
2933         *sopp = NULL;
2934
2935         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2936                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2937                 return nfserr_bad_stateid;
2938         }
2939
2940         if (STALE_STATEID(stateid))
2941                 return nfserr_stale_stateid;
2942
2943         if (nfsd4_has_session(cstate))
2944                 flags |= HAS_SESSION;
2945
2946         /*
2947         * We return BAD_STATEID if filehandle doesn't match stateid, 
2948         * the confirmed flag is incorrecly set, or the generation 
2949         * number is incorrect.  
2950         */
2951         stp = find_stateid(stateid, flags);
2952         if (stp == NULL) {
2953                 /*
2954                  * Also, we should make sure this isn't just the result of
2955                  * a replayed close:
2956                  */
2957                 sop = search_close_lru(stateid->si_stateownerid, flags);
2958                 if (sop == NULL)
2959                         return nfserr_bad_stateid;
2960                 *sopp = sop;
2961                 goto check_replay;
2962         }
2963
2964         *stpp = stp;
2965         *sopp = sop = stp->st_stateowner;
2966
2967         if (lock) {
2968                 clientid_t *lockclid = &lock->v.new.clientid;
2969                 struct nfs4_client *clp = sop->so_client;
2970                 int lkflg = 0;
2971                 __be32 status;
2972
2973                 lkflg = setlkflg(lock->lk_type);
2974
2975                 if (lock->lk_is_new) {
2976                         if (!sop->so_is_open_owner)
2977                                 return nfserr_bad_stateid;
2978                         if (!(flags & HAS_SESSION) &&
2979                             !same_clid(&clp->cl_clientid, lockclid))
2980                                 return nfserr_bad_stateid;
2981                         /* stp is the open stateid */
2982                         status = nfs4_check_openmode(stp, lkflg);
2983                         if (status)
2984                                 return status;
2985                 } else {
2986                         /* stp is the lock stateid */
2987                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2988                         if (status)
2989                                 return status;
2990                }
2991         }
2992
2993         if (nfs4_check_fh(current_fh, stp)) {
2994                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2995                 return nfserr_bad_stateid;
2996         }
2997
2998         /*
2999         *  We now validate the seqid and stateid generation numbers.
3000         *  For the moment, we ignore the possibility of 
3001         *  generation number wraparound.
3002         */
3003         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3004                 goto check_replay;
3005
3006         if (sop->so_confirmed && flags & CONFIRM) {
3007                 dprintk("NFSD: preprocess_seqid_op: expected"
3008                                 " unconfirmed stateowner!\n");
3009                 return nfserr_bad_stateid;
3010         }
3011         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3012                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3013                                 " confirmed yet!\n");
3014                 return nfserr_bad_stateid;
3015         }
3016         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3017         if (status)
3018                 return status;
3019         renew_client(sop->so_client);
3020         return nfs_ok;
3021
3022 check_replay:
3023         if (seqid == sop->so_seqid - 1) {
3024                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3025                 /* indicate replay to calling function */
3026                 return nfserr_replay_me;
3027         }
3028         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3029                         sop->so_seqid, seqid);
3030         *sopp = NULL;
3031         return nfserr_bad_seqid;
3032 }
3033
3034 __be32
3035 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3036                    struct nfsd4_open_confirm *oc)
3037 {
3038         __be32 status;
3039         struct nfs4_stateowner *sop;
3040         struct nfs4_stateid *stp;
3041
3042         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3043                         (int)cstate->current_fh.fh_dentry->d_name.len,
3044                         cstate->current_fh.fh_dentry->d_name.name);
3045
3046         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3047         if (status)
3048                 return status;
3049
3050         nfs4_lock_state();
3051
3052         if ((status = nfs4_preprocess_seqid_op(cstate,
3053                                         oc->oc_seqid, &oc->oc_req_stateid,
3054                                         CONFIRM | OPEN_STATE,
3055                                         &oc->oc_stateowner, &stp, NULL)))
3056                 goto out; 
3057
3058         sop = oc->oc_stateowner;
3059         sop->so_confirmed = 1;
3060         update_stateid(&stp->st_stateid);
3061         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3062         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
3063                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3064                          stp->st_stateid.si_boot,
3065                          stp->st_stateid.si_stateownerid,
3066                          stp->st_stateid.si_fileid,
3067                          stp->st_stateid.si_generation);
3068
3069         nfsd4_create_clid_dir(sop->so_client);
3070 out:
3071         if (oc->oc_stateowner) {
3072                 nfs4_get_stateowner(oc->oc_stateowner);
3073                 cstate->replay_owner = oc->oc_stateowner;
3074         }
3075         nfs4_unlock_state();
3076         return status;
3077 }
3078
3079
3080 /*
3081  * unset all bits in union bitmap (bmap) that
3082  * do not exist in share (from successful OPEN_DOWNGRADE)
3083  */
3084 static void
3085 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3086 {
3087         int i;
3088         for (i = 1; i < 4; i++) {
3089                 if ((i & access) != i)
3090                         __clear_bit(i, bmap);
3091         }
3092 }
3093
3094 static void
3095 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3096 {
3097         int i;
3098         for (i = 0; i < 4; i++) {
3099                 if ((i & deny) != i)
3100                         __clear_bit(i, bmap);
3101         }
3102 }
3103
3104 __be32
3105 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3106                      struct nfsd4_compound_state *cstate,
3107                      struct nfsd4_open_downgrade *od)
3108 {
3109         __be32 status;
3110         struct nfs4_stateid *stp;
3111         unsigned int share_access;
3112
3113         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3114                         (int)cstate->current_fh.fh_dentry->d_name.len,
3115                         cstate->current_fh.fh_dentry->d_name.name);
3116
3117         if (!access_valid(od->od_share_access, cstate->minorversion)
3118                         || !deny_valid(od->od_share_deny))
3119                 return nfserr_inval;
3120
3121         nfs4_lock_state();
3122         if ((status = nfs4_preprocess_seqid_op(cstate,
3123                                         od->od_seqid,
3124                                         &od->od_stateid, 
3125                                         OPEN_STATE,
3126                                         &od->od_stateowner, &stp, NULL)))
3127                 goto out; 
3128
3129         status = nfserr_inval;
3130         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3131                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3132                         stp->st_access_bmap, od->od_share_access);
3133                 goto out;
3134         }
3135         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3136                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3137                         stp->st_deny_bmap, od->od_share_deny);
3138                 goto out;
3139         }
3140         set_access(&share_access, stp->st_access_bmap);
3141         nfs4_file_downgrade(stp->st_vfs_file,
3142                             share_access & ~od->od_share_access);
3143
3144         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3145         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3146
3147         update_stateid(&stp->st_stateid);
3148         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3149         status = nfs_ok;
3150 out:
3151         if (od->od_stateowner) {
3152                 nfs4_get_stateowner(od->od_stateowner);
3153                 cstate->replay_owner = od->od_stateowner;
3154         }
3155         nfs4_unlock_state();
3156         return status;
3157 }
3158
3159 /*
3160  * nfs4_unlock_state() called after encode
3161  */
3162 __be32
3163 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3164             struct nfsd4_close *close)
3165 {
3166         __be32 status;
3167         struct nfs4_stateid *stp;
3168
3169         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3170                         (int)cstate->current_fh.fh_dentry->d_name.len,
3171                         cstate->current_fh.fh_dentry->d_name.name);
3172
3173         nfs4_lock_state();
3174         /* check close_lru for replay */
3175         if ((status = nfs4_preprocess_seqid_op(cstate,
3176                                         close->cl_seqid,
3177                                         &close->cl_stateid, 
3178                                         OPEN_STATE | CLOSE_STATE,
3179                                         &close->cl_stateowner, &stp, NULL)))
3180                 goto out; 
3181         status = nfs_ok;
3182         update_stateid(&stp->st_stateid);
3183         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3184
3185         /* release_stateid() calls nfsd_close() if needed */
3186         release_open_stateid(stp);
3187
3188         /* place unused nfs4_stateowners on so_close_lru list to be
3189          * released by the laundromat service after the lease period
3190          * to enable us to handle CLOSE replay
3191          */
3192         if (list_empty(&close->cl_stateowner->so_stateids))
3193                 move_to_close_lru(close->cl_stateowner);
3194 out:
3195         if (close->cl_stateowner) {
3196                 nfs4_get_stateowner(close->cl_stateowner);
3197                 cstate->replay_owner = close->cl_stateowner;
3198         }
3199         nfs4_unlock_state();
3200         return status;
3201 }
3202
3203 __be32
3204 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3205                   struct nfsd4_delegreturn *dr)
3206 {
3207         struct nfs4_delegation *dp;
3208         stateid_t *stateid = &dr->dr_stateid;
3209         struct inode *inode;
3210         __be32 status;
3211         int flags = 0;
3212
3213         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3214                 return status;
3215         inode = cstate->current_fh.fh_dentry->d_inode;
3216
3217         if (nfsd4_has_session(cstate))
3218                 flags |= HAS_SESSION;
3219         nfs4_lock_state();
3220         status = nfserr_bad_stateid;
3221         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3222                 goto out;
3223         status = nfserr_stale_stateid;
3224         if (STALE_STATEID(stateid))
3225                 goto out;
3226         status = nfserr_bad_stateid;
3227         if (!is_delegation_stateid(stateid))
3228                 goto out;
3229         dp = find_delegation_stateid(inode, stateid);
3230         if (!dp)
3231                 goto out;
3232         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3233         if (status)
3234                 goto out;
3235         renew_client(dp->dl_client);
3236
3237         unhash_delegation(dp);
3238 out:
3239         nfs4_unlock_state();
3240
3241         return status;
3242 }
3243
3244
3245 /* 
3246  * Lock owner state (byte-range locks)
3247  */
3248 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3249 #define LOCK_HASH_BITS              8
3250 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3251 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3252
3253 static inline u64
3254 end_offset(u64 start, u64 len)
3255 {
3256         u64 end;
3257
3258         end = start + len;
3259         return end >= start ? end: NFS4_MAX_UINT64;
3260 }
3261
3262 /* last octet in a range */
3263 static inline u64
3264 last_byte_offset(u64 start, u64 len)
3265 {
3266         u64 end;
3267
3268         BUG_ON(!len);
3269         end = start + len;
3270         return end > start ? end - 1: NFS4_MAX_UINT64;
3271 }
3272
3273 #define lockownerid_hashval(id) \
3274         ((id) & LOCK_HASH_MASK)
3275
3276 static inline unsigned int
3277 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3278                 struct xdr_netobj *ownername)
3279 {
3280         return (file_hashval(inode) + cl_id
3281                         + opaque_hashval(ownername->data, ownername->len))
3282                 & LOCK_HASH_MASK;
3283 }
3284
3285 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3286 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3287 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3288
3289 static struct nfs4_stateid *
3290 find_stateid(stateid_t *stid, int flags)
3291 {
3292         struct nfs4_stateid *local;
3293         u32 st_id = stid->si_stateownerid;
3294         u32 f_id = stid->si_fileid;
3295         unsigned int hashval;
3296
3297         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3298         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3299                 hashval = stateid_hashval(st_id, f_id);
3300                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3301                         if ((local->st_stateid.si_stateownerid == st_id) &&
3302                             (local->st_stateid.si_fileid == f_id))
3303                                 return local;
3304                 }
3305         } 
3306
3307         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3308                 hashval = stateid_hashval(st_id, f_id);
3309                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3310                         if ((local->st_stateid.si_stateownerid == st_id) &&
3311                             (local->st_stateid.si_fileid == f_id))
3312                                 return local;
3313                 }
3314         }
3315         return NULL;
3316 }
3317
3318 static struct nfs4_delegation *
3319 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3320 {
3321         struct nfs4_file *fp;
3322         struct nfs4_delegation *dl;
3323
3324         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3325                     stid->si_boot, stid->si_stateownerid,
3326                     stid->si_fileid, stid->si_generation);
3327
3328         fp = find_file(ino);
3329         if (!fp)
3330                 return NULL;
3331         dl = find_delegation_file(fp, stid);
3332         put_nfs4_file(fp);
3333         return dl;
3334 }
3335
3336 /*
3337  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3338  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3339  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3340  * locking, this prevents us from being completely protocol-compliant.  The
3341  * real solution to this problem is to start using unsigned file offsets in
3342  * the VFS, but this is a very deep change!
3343  */
3344 static inline void
3345 nfs4_transform_lock_offset(struct file_lock *lock)
3346 {
3347         if (lock->fl_start < 0)
3348                 lock->fl_start = OFFSET_MAX;
3349         if (lock->fl_end < 0)
3350                 lock->fl_end = OFFSET_MAX;
3351 }
3352
3353 /* Hack!: For now, we're defining this just so we can use a pointer to it
3354  * as a unique cookie to identify our (NFSv4's) posix locks. */
3355 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3356 };
3357
3358 static inline void
3359 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3360 {
3361         struct nfs4_stateowner *sop;
3362         unsigned int hval;
3363
3364         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3365                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3366                 hval = lockownerid_hashval(sop->so_id);
3367                 kref_get(&sop->so_ref);
3368                 deny->ld_sop = sop;
3369                 deny->ld_clientid = sop->so_client->cl_clientid;
3370         } else {
3371                 deny->ld_sop = NULL;
3372                 deny->ld_clientid.cl_boot = 0;
3373                 deny->ld_clientid.cl_id = 0;
3374         }
3375         deny->ld_start = fl->fl_start;
3376         deny->ld_length = NFS4_MAX_UINT64;
3377         if (fl->fl_end != NFS4_MAX_UINT64)
3378                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3379         deny->ld_type = NFS4_READ_LT;
3380         if (fl->fl_type != F_RDLCK)
3381                 deny->ld_type = NFS4_WRITE_LT;
3382 }
3383
3384 static struct nfs4_stateowner *
3385 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3386                 struct xdr_netobj *owner)
3387 {
3388         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3389         struct nfs4_stateowner *op;
3390
3391         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3392                 if (same_owner_str(op, owner, clid))
3393                         return op;
3394         }
3395         return NULL;
3396 }
3397
3398 /*
3399  * Alloc a lock owner structure.
3400  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3401  * occured. 
3402  *
3403  * strhashval = lock_ownerstr_hashval 
3404  */
3405
3406 static struct nfs4_stateowner *
3407 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3408         struct nfs4_stateowner *sop;
3409         struct nfs4_replay *rp;
3410         unsigned int idhashval;
3411
3412         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3413                 return NULL;
3414         idhashval = lockownerid_hashval(current_ownerid);
3415         INIT_LIST_HEAD(&sop->so_idhash);
3416         INIT_LIST_HEAD(&sop->so_strhash);
3417         INIT_LIST_HEAD(&sop->so_perclient);
3418         INIT_LIST_HEAD(&sop->so_stateids);
3419         INIT_LIST_HEAD(&sop->so_perstateid);
3420         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3421         sop->so_time = 0;
3422         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3423         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3424         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3425         sop->so_is_open_owner = 0;
3426         sop->so_id = current_ownerid++;
3427         sop->so_client = clp;
3428         /* It is the openowner seqid that will be incremented in encode in the
3429          * case of new lockowners; so increment the lock seqid manually: */
3430         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3431         sop->so_confirmed = 1;
3432         rp = &sop->so_replay;
3433         rp->rp_status = nfserr_serverfault;
3434         rp->rp_buflen = 0;
3435         rp->rp_buf = rp->rp_ibuf;
3436         return sop;
3437 }
3438
3439 static struct nfs4_stateid *
3440 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3441 {
3442         struct nfs4_stateid *stp;
3443         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3444
3445         stp = nfs4_alloc_stateid();
3446         if (stp == NULL)
3447                 goto out;
3448         INIT_LIST_HEAD(&stp->st_hash);
3449         INIT_LIST_HEAD(&stp->st_perfile);
3450         INIT_LIST_HEAD(&stp->st_perstateowner);
3451         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3452         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3453         list_add(&stp->st_perfile, &fp->fi_stateids);
3454         list_add(&stp->st_perstateowner, &sop->so_stateids);
3455         stp->st_stateowner = sop;
3456         get_nfs4_file(fp);
3457         stp->st_file = fp;
3458         stp->st_stateid.si_boot = boot_time;
3459         stp->st_stateid.si_stateownerid = sop->so_id;
3460         stp->st_stateid.si_fileid = fp->fi_id;
3461         stp->st_stateid.si_generation = 0;
3462         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3463         stp->st_access_bmap = open_stp->st_access_bmap;
3464         stp->st_deny_bmap = open_stp->st_deny_bmap;
3465         stp->st_openstp = open_stp;
3466
3467 out:
3468         return stp;
3469 }
3470
3471 static int
3472 check_lock_length(u64 offset, u64 length)
3473 {
3474         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3475              LOFF_OVERFLOW(offset, length)));
3476 }
3477
3478 /*
3479  *  LOCK operation 
3480  */
3481 __be32
3482 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3483            struct nfsd4_lock *lock)
3484 {
3485         struct nfs4_stateowner *open_sop = NULL;
3486         struct nfs4_stateowner *lock_sop = NULL;
3487         struct nfs4_stateid *lock_stp;
3488         struct file *filp;
3489         struct file_lock file_lock;
3490         struct file_lock conflock;
3491         __be32 status = 0;
3492         unsigned int strhashval;
3493         unsigned int cmd;
3494         int err;
3495
3496         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3497                 (long long) lock->lk_offset,
3498                 (long long) lock->lk_length);
3499
3500         if (check_lock_length(lock->lk_offset, lock->lk_length))
3501                  return nfserr_inval;
3502
3503         if ((status = fh_verify(rqstp, &cstate->current_fh,
3504                                 S_IFREG, NFSD_MAY_LOCK))) {
3505                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3506                 return status;
3507         }
3508
3509         nfs4_lock_state();
3510
3511         if (lock->lk_is_new) {
3512                 /*
3513                  * Client indicates that this is a new lockowner.
3514                  * Use open owner and open stateid to create lock owner and
3515                  * lock stateid.
3516                  */
3517                 struct nfs4_stateid *open_stp = NULL;
3518                 struct nfs4_file *fp;
3519                 
3520                 status = nfserr_stale_clientid;
3521                 if (!nfsd4_has_session(cstate) &&
3522                     STALE_CLIENTID(&lock->lk_new_clientid))
3523                         goto out;
3524
3525                 /* validate and update open stateid and open seqid */
3526                 status = nfs4_preprocess_seqid_op(cstate,
3527                                         lock->lk_new_open_seqid,
3528                                         &lock->lk_new_open_stateid,
3529                                         OPEN_STATE,
3530                                         &lock->lk_replay_owner, &open_stp,
3531                                         lock);
3532                 if (status)
3533                         goto out;
3534                 open_sop = lock->lk_replay_owner;
3535                 /* create lockowner and lock stateid */
3536                 fp = open_stp->st_file;
3537                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3538                                 open_sop->so_client->cl_clientid.cl_id, 
3539                                 &lock->v.new.owner);
3540                 /* XXX: Do we need to check for duplicate stateowners on
3541                  * the same file, or should they just be allowed (and
3542                  * create new stateids)? */
3543                 status = nfserr_resource;
3544                 lock_sop = alloc_init_lock_stateowner(strhashval,
3545                                 open_sop->so_client, open_stp, lock);
3546                 if (lock_sop == NULL)
3547                         goto out;
3548                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3549                 if (lock_stp == NULL)
3550                         goto out;
3551         } else {
3552                 /* lock (lock owner + lock stateid) already exists */
3553                 status = nfs4_preprocess_seqid_op(cstate,
3554                                        lock->lk_old_lock_seqid, 
3555                                        &lock->lk_old_lock_stateid, 
3556                                        LOCK_STATE,
3557                                        &lock->lk_replay_owner, &lock_stp, lock);
3558                 if (status)
3559                         goto out;
3560                 lock_sop = lock->lk_replay_owner;
3561         }
3562         /* lock->lk_replay_owner and lock_stp have been created or found */
3563         filp = lock_stp->st_vfs_file;
3564
3565         status = nfserr_grace;
3566         if (locks_in_grace() && !lock->lk_reclaim)
3567                 goto out;
3568         status = nfserr_no_grace;
3569         if (!locks_in_grace() && lock->lk_reclaim)
3570                 goto out;
3571
3572         locks_init_lock(&file_lock);
3573         switch (lock->lk_type) {
3574                 case NFS4_READ_LT:
3575                 case NFS4_READW_LT:
3576                         file_lock.fl_type = F_RDLCK;
3577                         cmd = F_SETLK;
3578                 break;
3579                 case NFS4_WRITE_LT:
3580                 case NFS4_WRITEW_LT:
3581                         file_lock.fl_type = F_WRLCK;
3582                         cmd = F_SETLK;
3583                 break;
3584                 default:
3585                         status = nfserr_inval;
3586                 goto out;
3587         }
3588         file_lock.fl_owner = (fl_owner_t)lock_sop;
3589         file_lock.fl_pid = current->tgid;
3590         file_lock.fl_file = filp;
3591         file_lock.fl_flags = FL_POSIX;
3592         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3593
3594         file_lock.fl_start = lock->lk_offset;
3595         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3596         nfs4_transform_lock_offset(&file_lock);
3597
3598         /*
3599         * Try to lock the file in the VFS.
3600         * Note: locks.c uses the BKL to protect the inode's lock list.
3601         */
3602
3603         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3604         switch (-err) {
3605         case 0: /* success! */
3606                 update_stateid(&lock_stp->st_stateid);
3607                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3608                                 sizeof(stateid_t));
3609                 status = 0;
3610                 break;
3611         case (EAGAIN):          /* conflock holds conflicting lock */
3612                 status = nfserr_denied;
3613                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3614                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3615                 break;
3616         case (EDEADLK):
3617                 status = nfserr_deadlock;
3618                 break;
3619         default:        
3620                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3621                 status = nfserr_resource;
3622                 break;
3623         }
3624 out:
3625         if (status && lock->lk_is_new && lock_sop)
3626                 release_lockowner(lock_sop);
3627         if (lock->lk_replay_owner) {
3628                 nfs4_get_stateowner(lock->lk_replay_owner);
3629                 cstate->replay_owner = lock->lk_replay_owner;
3630         }
3631         nfs4_unlock_state();
3632         return status;
3633 }
3634
3635 /*
3636  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3637  * so we do a temporary open here just to get an open file to pass to
3638  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3639  * inode operation.)
3640  */
3641 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3642 {
3643         struct file *file;
3644         int err;
3645
3646         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3647         if (err)
3648                 return err;
3649         err = vfs_test_lock(file, lock);
3650         nfsd_close(file);
3651         return err;
3652 }
3653
3654 /*
3655  * LOCKT operation
3656  */
3657 __be32
3658 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3659             struct nfsd4_lockt *lockt)
3660 {
3661         struct inode *inode;
3662         struct file_lock file_lock;
3663         int error;
3664         __be32 status;
3665
3666         if (locks_in_grace())
3667                 return nfserr_grace;
3668
3669         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3670                  return nfserr_inval;
3671
3672         lockt->lt_stateowner = NULL;
3673         nfs4_lock_state();
3674
3675         status = nfserr_stale_clientid;
3676         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3677                 goto out;
3678
3679         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3680                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3681                 if (status == nfserr_symlink)
3682                         status = nfserr_inval;
3683                 goto out;
3684         }
3685
3686         inode = cstate->current_fh.fh_dentry->d_inode;
3687         locks_init_lock(&file_lock);
3688         switch (lockt->lt_type) {
3689                 case NFS4_READ_LT:
3690                 case NFS4_READW_LT:
3691                         file_lock.fl_type = F_RDLCK;
3692                 break;
3693                 case NFS4_WRITE_LT:
3694                 case NFS4_WRITEW_LT:
3695                         file_lock.fl_type = F_WRLCK;
3696                 break;
3697                 default:
3698                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3699                         status = nfserr_inval;
3700                 goto out;
3701         }
3702
3703         lockt->lt_stateowner = find_lockstateowner_str(inode,
3704                         &lockt->lt_clientid, &lockt->lt_owner);
3705         if (lockt->lt_stateowner)
3706                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3707         file_lock.fl_pid = current->tgid;
3708         file_lock.fl_flags = FL_POSIX;
3709
3710         file_lock.fl_start = lockt->lt_offset;
3711         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3712
3713         nfs4_transform_lock_offset(&file_lock);
3714
3715         status = nfs_ok;
3716         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3717         if (error) {
3718                 status = nfserrno(error);
3719                 goto out;
3720         }
3721         if (file_lock.fl_type != F_UNLCK) {
3722                 status = nfserr_denied;
3723                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3724         }
3725 out:
3726         nfs4_unlock_state();
3727         return status;
3728 }
3729
3730 __be32
3731 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3732             struct nfsd4_locku *locku)
3733 {
3734         struct nfs4_stateid *stp;
3735         struct file *filp = NULL;
3736         struct file_lock file_lock;
3737         __be32 status;
3738         int err;
3739                                                         
3740         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3741                 (long long) locku->lu_offset,
3742                 (long long) locku->lu_length);
3743
3744         if (check_lock_length(locku->lu_offset, locku->lu_length))
3745                  return nfserr_inval;
3746
3747         nfs4_lock_state();
3748                                                                                 
3749         if ((status = nfs4_preprocess_seqid_op(cstate,
3750                                         locku->lu_seqid, 
3751                                         &locku->lu_stateid, 
3752                                         LOCK_STATE,
3753                                         &locku->lu_stateowner, &stp, NULL)))
3754                 goto out;
3755
3756         filp = stp->st_vfs_file;
3757         BUG_ON(!filp);
3758         locks_init_lock(&file_lock);
3759         file_lock.fl_type = F_UNLCK;
3760         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3761         file_lock.fl_pid = current->tgid;
3762         file_lock.fl_file = filp;
3763         file_lock.fl_flags = FL_POSIX; 
3764         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3765         file_lock.fl_start = locku->lu_offset;
3766
3767         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3768         nfs4_transform_lock_offset(&file_lock);
3769
3770         /*
3771         *  Try to unlock the file in the VFS.
3772         */
3773         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3774         if (err) {
3775                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3776                 goto out_nfserr;
3777         }
3778         /*
3779         * OK, unlock succeeded; the only thing left to do is update the stateid.
3780         */
3781         update_stateid(&stp->st_stateid);
3782         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3783
3784 out:
3785         if (locku->lu_stateowner) {
3786                 nfs4_get_stateowner(locku->lu_stateowner);
3787                 cstate->replay_owner = locku->lu_stateowner;
3788         }
3789         nfs4_unlock_state();
3790         return status;
3791
3792 out_nfserr:
3793         status = nfserrno(err);
3794         goto out;
3795 }
3796
3797 /*
3798  * returns
3799  *      1: locks held by lockowner
3800  *      0: no locks held by lockowner
3801  */
3802 static int
3803 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3804 {
3805         struct file_lock **flpp;
3806         struct inode *inode = filp->f_path.dentry->d_inode;
3807         int status = 0;
3808
3809         lock_kernel();
3810         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3811                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3812                         status = 1;
3813                         goto out;
3814                 }
3815         }
3816 out:
3817         unlock_kernel();
3818         return status;
3819 }
3820
3821 __be32
3822 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3823                         struct nfsd4_compound_state *cstate,
3824                         struct nfsd4_release_lockowner *rlockowner)
3825 {
3826         clientid_t *clid = &rlockowner->rl_clientid;
3827         struct nfs4_stateowner *sop;
3828         struct nfs4_stateid *stp;
3829         struct xdr_netobj *owner = &rlockowner->rl_owner;
3830         struct list_head matches;
3831         int i;
3832         __be32 status;
3833
3834         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3835                 clid->cl_boot, clid->cl_id);
3836
3837         /* XXX check for lease expiration */
3838
3839         status = nfserr_stale_clientid;
3840         if (STALE_CLIENTID(clid))
3841                 return status;
3842
3843         nfs4_lock_state();
3844
3845         status = nfserr_locks_held;
3846         /* XXX: we're doing a linear search through all the lockowners.
3847          * Yipes!  For now we'll just hope clients aren't really using
3848          * release_lockowner much, but eventually we have to fix these
3849          * data structures. */
3850         INIT_LIST_HEAD(&matches);
3851         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3852                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3853                         if (!same_owner_str(sop, owner, clid))
3854                                 continue;
3855                         list_for_each_entry(stp, &sop->so_stateids,
3856                                         st_perstateowner) {
3857                                 if (check_for_locks(stp->st_vfs_file, sop))
3858                                         goto out;
3859                                 /* Note: so_perclient unused for lockowners,
3860                                  * so it's OK to fool with here. */
3861                                 list_add(&sop->so_perclient, &matches);
3862                         }
3863                 }
3864         }
3865         /* Clients probably won't expect us to return with some (but not all)
3866          * of the lockowner state released; so don't release any until all
3867          * have been checked. */
3868         status = nfs_ok;
3869         while (!list_empty(&matches)) {
3870                 sop = list_entry(matches.next, struct nfs4_stateowner,
3871                                                                 so_perclient);
3872                 /* unhash_stateowner deletes so_perclient only
3873                  * for openowners. */
3874                 list_del(&sop->so_perclient);
3875                 release_lockowner(sop);
3876         }
3877 out:
3878         nfs4_unlock_state();
3879         return status;
3880 }
3881
3882 static inline struct nfs4_client_reclaim *
3883 alloc_reclaim(void)
3884 {
3885         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3886 }
3887
3888 int
3889 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3890 {
3891         unsigned int strhashval = clientstr_hashval(name);
3892         struct nfs4_client *clp;
3893
3894         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3895         return clp ? 1 : 0;
3896 }
3897
3898 /*
3899  * failure => all reset bets are off, nfserr_no_grace...
3900  */
3901 int
3902 nfs4_client_to_reclaim(const char *name)
3903 {
3904         unsigned int strhashval;
3905         struct nfs4_client_reclaim *crp = NULL;
3906
3907         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3908         crp = alloc_reclaim();
3909         if (!crp)
3910                 return 0;
3911         strhashval = clientstr_hashval(name);
3912         INIT_LIST_HEAD(&crp->cr_strhash);
3913         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3914         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3915         reclaim_str_hashtbl_size++;
3916         return 1;
3917 }
3918
3919 static void
3920 nfs4_release_reclaim(void)
3921 {
3922         struct nfs4_client_reclaim *crp = NULL;
3923         int i;
3924
3925         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3926                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3927                         crp = list_entry(reclaim_str_hashtbl[i].next,
3928                                         struct nfs4_client_reclaim, cr_strhash);
3929                         list_del(&crp->cr_strhash);
3930                         kfree(crp);
3931                         reclaim_str_hashtbl_size--;
3932                 }
3933         }
3934         BUG_ON(reclaim_str_hashtbl_size);
3935 }
3936
3937 /*
3938  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3939 static struct nfs4_client_reclaim *
3940 nfs4_find_reclaim_client(clientid_t *clid)
3941 {
3942         unsigned int strhashval;
3943         struct nfs4_client *clp;
3944         struct nfs4_client_reclaim *crp = NULL;
3945
3946
3947         /* find clientid in conf_id_hashtbl */
3948         clp = find_confirmed_client(clid);
3949         if (clp == NULL)
3950                 return NULL;
3951
3952         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3953                             clp->cl_name.len, clp->cl_name.data,
3954                             clp->cl_recdir);
3955
3956         /* find clp->cl_name in reclaim_str_hashtbl */
3957         strhashval = clientstr_hashval(clp->cl_recdir);
3958         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3959                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3960                         return crp;
3961                 }
3962         }
3963         return NULL;
3964 }
3965
3966 /*
3967 * Called from OPEN. Look for clientid in reclaim list.
3968 */
3969 __be32
3970 nfs4_check_open_reclaim(clientid_t *clid)
3971 {
3972         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3973 }
3974
3975 /* initialization to perform at module load time: */
3976
3977 int
3978 nfs4_state_init(void)
3979 {
3980         int i, status;
3981
3982         status = nfsd4_init_slabs();
3983         if (status)
3984                 return status;
3985         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3986                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3987                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3988                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3989                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3990         }
3991         for (i = 0; i < SESSION_HASH_SIZE; i++)
3992                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3993         for (i = 0; i < FILE_HASH_SIZE; i++) {
3994                 INIT_LIST_HEAD(&file_hashtbl[i]);
3995         }
3996         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3997                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3998                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3999         }
4000         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4001                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4002                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4003         }
4004         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4005                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4006                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4007         }
4008         memset(&onestateid, ~0, sizeof(stateid_t));
4009         INIT_LIST_HEAD(&close_lru);
4010         INIT_LIST_HEAD(&client_lru);
4011         INIT_LIST_HEAD(&del_recall_lru);
4012         for (i = 0; i < CLIENT_HASH_SIZE; i++)
4013                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4014         reclaim_str_hashtbl_size = 0;
4015         return 0;
4016 }
4017
4018 static void
4019 nfsd4_load_reboot_recovery_data(void)
4020 {
4021         int status;
4022
4023         nfs4_lock_state();
4024         nfsd4_init_recdir(user_recovery_dirname);
4025         status = nfsd4_recdir_load();
4026         nfs4_unlock_state();
4027         if (status)
4028                 printk("NFSD: Failure reading reboot recovery data\n");
4029 }
4030
4031 unsigned long
4032 get_nfs4_grace_period(void)
4033 {
4034         return max(user_lease_time, lease_time) * HZ;
4035 }
4036
4037 /*
4038  * Since the lifetime of a delegation isn't limited to that of an open, a
4039  * client may quite reasonably hang on to a delegation as long as it has
4040  * the inode cached.  This becomes an obvious problem the first time a
4041  * client's inode cache approaches the size of the server's total memory.
4042  *
4043  * For now we avoid this problem by imposing a hard limit on the number
4044  * of delegations, which varies according to the server's memory size.
4045  */
4046 static void
4047 set_max_delegations(void)
4048 {
4049         /*
4050          * Allow at most 4 delegations per megabyte of RAM.  Quick
4051          * estimates suggest that in the worst case (where every delegation
4052          * is for a different inode), a delegation could take about 1.5K,
4053          * giving a worst case usage of about 6% of memory.
4054          */
4055         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4056 }
4057
4058 /* initialization to perform when the nfsd service is started: */
4059
4060 static void
4061 __nfs4_state_start(void)
4062 {
4063         unsigned long grace_time;
4064
4065         boot_time = get_seconds();
4066         grace_time = get_nfs4_grace_period();
4067         lease_time = user_lease_time;
4068         locks_start_grace(&nfsd4_manager);
4069         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4070                grace_time/HZ);
4071         laundry_wq = create_singlethread_workqueue("nfsd4");
4072         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4073         set_max_delegations();
4074 }
4075
4076 void
4077 nfs4_state_start(void)
4078 {
4079         if (nfs4_init)
4080                 return;
4081         nfsd4_load_reboot_recovery_data();
4082         __nfs4_state_start();
4083         nfs4_init = 1;
4084         return;
4085 }
4086
4087 time_t
4088 nfs4_lease_time(void)
4089 {
4090         return lease_time;
4091 }
4092
4093 static void
4094 __nfs4_state_shutdown(void)
4095 {
4096         int i;
4097         struct nfs4_client *clp = NULL;
4098         struct nfs4_delegation *dp = NULL;
4099         struct list_head *pos, *next, reaplist;
4100
4101         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4102                 while (!list_empty(&conf_id_hashtbl[i])) {
4103                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4104                         expire_client(clp);
4105                 }
4106                 while (!list_empty(&unconf_str_hashtbl[i])) {
4107                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4108                         expire_client(clp);
4109                 }
4110         }
4111         INIT_LIST_HEAD(&reaplist);
4112         spin_lock(&recall_lock);
4113         list_for_each_safe(pos, next, &del_recall_lru) {
4114                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4115                 list_move(&dp->dl_recall_lru, &reaplist);
4116         }
4117         spin_unlock(&recall_lock);
4118         list_for_each_safe(pos, next, &reaplist) {
4119                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4120                 list_del_init(&dp->dl_recall_lru);
4121                 unhash_delegation(dp);
4122         }
4123
4124         nfsd4_shutdown_recdir();
4125         nfs4_init = 0;
4126 }
4127
4128 void
4129 nfs4_state_shutdown(void)
4130 {
4131         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4132         destroy_workqueue(laundry_wq);
4133         locks_end_grace(&nfsd4_manager);
4134         nfs4_lock_state();
4135         nfs4_release_reclaim();
4136         __nfs4_state_shutdown();
4137         nfs4_unlock_state();
4138 }
4139
4140 /*
4141  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4142  * accessed when nfsd is starting.
4143  */
4144 static void
4145 nfs4_set_recdir(char *recdir)
4146 {
4147         strcpy(user_recovery_dirname, recdir);
4148 }
4149
4150 /*
4151  * Change the NFSv4 recovery directory to recdir.
4152  */
4153 int
4154 nfs4_reset_recoverydir(char *recdir)
4155 {
4156         int status;
4157         struct path path;
4158
4159         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4160         if (status)
4161                 return status;
4162         status = -ENOTDIR;
4163         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4164                 nfs4_set_recdir(recdir);
4165                 status = 0;
4166         }
4167         path_put(&path);
4168         return status;
4169 }
4170
4171 char *
4172 nfs4_recoverydir(void)
4173 {
4174         return user_recovery_dirname;
4175 }
4176
4177 /*
4178  * Called when leasetime is changed.
4179  *
4180  * The only way the protocol gives us to handle on-the-fly lease changes is to
4181  * simulate a reboot.  Instead of doing that, we just wait till the next time
4182  * we start to register any changes in lease time.  If the administrator
4183  * really wants to change the lease time *now*, they can go ahead and bring
4184  * nfsd down and then back up again after changing the lease time.
4185  *
4186  * user_lease_time is protected by nfsd_mutex since it's only really accessed
4187  * when nfsd is starting
4188  */
4189 void
4190 nfs4_reset_lease(time_t leasetime)
4191 {
4192         user_lease_time = leasetime;
4193 }