nfsd: support ext4 i_version
[safe/jmp/linux-2.6] / 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 = get_seconds();
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->se_slots);
584         kfree(ses);
585 }
586
587 static inline void
588 renew_client(struct nfs4_client *clp)
589 {
590         /*
591         * Move client to the end to the LRU list.
592         */
593         dprintk("renewing client (clientid %08x/%08x)\n", 
594                         clp->cl_clientid.cl_boot, 
595                         clp->cl_clientid.cl_id);
596         list_move_tail(&clp->cl_lru, &client_lru);
597         clp->cl_time = get_seconds();
598 }
599
600 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
601 static int
602 STALE_CLIENTID(clientid_t *clid)
603 {
604         if (clid->cl_boot == boot_time)
605                 return 0;
606         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
607                 clid->cl_boot, clid->cl_id, boot_time);
608         return 1;
609 }
610
611 /* 
612  * XXX Should we use a slab cache ?
613  * This type of memory management is somewhat inefficient, but we use it
614  * anyway since SETCLIENTID is not a common operation.
615  */
616 static struct nfs4_client *alloc_client(struct xdr_netobj name)
617 {
618         struct nfs4_client *clp;
619
620         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
621         if (clp == NULL)
622                 return NULL;
623         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
624         if (clp->cl_name.data == NULL) {
625                 kfree(clp);
626                 return NULL;
627         }
628         memcpy(clp->cl_name.data, name.data, name.len);
629         clp->cl_name.len = name.len;
630         return clp;
631 }
632
633 static void
634 shutdown_callback_client(struct nfs4_client *clp)
635 {
636         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
637
638         if (clnt) {
639                 /*
640                  * Callback threads take a reference on the client, so there
641                  * should be no outstanding callbacks at this point.
642                  */
643                 clp->cl_callback.cb_client = NULL;
644                 rpc_shutdown_client(clnt);
645         }
646 }
647
648 static inline void
649 free_client(struct nfs4_client *clp)
650 {
651         shutdown_callback_client(clp);
652         nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
653                              clp->cl_slot.sl_cache_entry.ce_resused);
654         if (clp->cl_cred.cr_group_info)
655                 put_group_info(clp->cl_cred.cr_group_info);
656         kfree(clp->cl_principal);
657         kfree(clp->cl_name.data);
658         kfree(clp);
659 }
660
661 void
662 put_nfs4_client(struct nfs4_client *clp)
663 {
664         if (atomic_dec_and_test(&clp->cl_count))
665                 free_client(clp);
666 }
667
668 static void
669 expire_client(struct nfs4_client *clp)
670 {
671         struct nfs4_stateowner *sop;
672         struct nfs4_delegation *dp;
673         struct list_head reaplist;
674
675         dprintk("NFSD: expire_client cl_count %d\n",
676                             atomic_read(&clp->cl_count));
677
678         INIT_LIST_HEAD(&reaplist);
679         spin_lock(&recall_lock);
680         while (!list_empty(&clp->cl_delegations)) {
681                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
682                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
683                                 dp->dl_flock);
684                 list_del_init(&dp->dl_perclnt);
685                 list_move(&dp->dl_recall_lru, &reaplist);
686         }
687         spin_unlock(&recall_lock);
688         while (!list_empty(&reaplist)) {
689                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
690                 list_del_init(&dp->dl_recall_lru);
691                 unhash_delegation(dp);
692         }
693         list_del(&clp->cl_idhash);
694         list_del(&clp->cl_strhash);
695         list_del(&clp->cl_lru);
696         while (!list_empty(&clp->cl_openowners)) {
697                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
698                 release_openowner(sop);
699         }
700         while (!list_empty(&clp->cl_sessions)) {
701                 struct nfsd4_session  *ses;
702                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
703                                  se_perclnt);
704                 release_session(ses);
705         }
706         put_nfs4_client(clp);
707 }
708
709 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
710 {
711         struct nfs4_client *clp;
712
713         clp = alloc_client(name);
714         if (clp == NULL)
715                 return NULL;
716         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
717         atomic_set(&clp->cl_count, 1);
718         atomic_set(&clp->cl_callback.cb_set, 0);
719         INIT_LIST_HEAD(&clp->cl_idhash);
720         INIT_LIST_HEAD(&clp->cl_strhash);
721         INIT_LIST_HEAD(&clp->cl_openowners);
722         INIT_LIST_HEAD(&clp->cl_delegations);
723         INIT_LIST_HEAD(&clp->cl_sessions);
724         INIT_LIST_HEAD(&clp->cl_lru);
725         return clp;
726 }
727
728 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
729 {
730         memcpy(target->cl_verifier.data, source->data,
731                         sizeof(target->cl_verifier.data));
732 }
733
734 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
735 {
736         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
737         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
738 }
739
740 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
741 {
742         target->cr_uid = source->cr_uid;
743         target->cr_gid = source->cr_gid;
744         target->cr_group_info = source->cr_group_info;
745         get_group_info(target->cr_group_info);
746 }
747
748 static int same_name(const char *n1, const char *n2)
749 {
750         return 0 == memcmp(n1, n2, HEXDIR_LEN);
751 }
752
753 static int
754 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
755 {
756         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
757 }
758
759 static int
760 same_clid(clientid_t *cl1, clientid_t *cl2)
761 {
762         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
763 }
764
765 /* XXX what about NGROUP */
766 static int
767 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
768 {
769         return cr1->cr_uid == cr2->cr_uid;
770 }
771
772 static void gen_clid(struct nfs4_client *clp)
773 {
774         static u32 current_clientid = 1;
775
776         clp->cl_clientid.cl_boot = boot_time;
777         clp->cl_clientid.cl_id = current_clientid++; 
778 }
779
780 static void gen_confirm(struct nfs4_client *clp)
781 {
782         static u32 i;
783         u32 *p;
784
785         p = (u32 *)clp->cl_confirm.data;
786         *p++ = get_seconds();
787         *p++ = i++;
788 }
789
790 static int check_name(struct xdr_netobj name)
791 {
792         if (name.len == 0) 
793                 return 0;
794         if (name.len > NFS4_OPAQUE_LIMIT) {
795                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
796                 return 0;
797         }
798         return 1;
799 }
800
801 static void
802 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
803 {
804         unsigned int idhashval;
805
806         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
807         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
808         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
809         list_add_tail(&clp->cl_lru, &client_lru);
810         clp->cl_time = get_seconds();
811 }
812
813 static void
814 move_to_confirmed(struct nfs4_client *clp)
815 {
816         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
817         unsigned int strhashval;
818
819         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
820         list_del_init(&clp->cl_strhash);
821         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
822         strhashval = clientstr_hashval(clp->cl_recdir);
823         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
824         renew_client(clp);
825 }
826
827 static struct nfs4_client *
828 find_confirmed_client(clientid_t *clid)
829 {
830         struct nfs4_client *clp;
831         unsigned int idhashval = clientid_hashval(clid->cl_id);
832
833         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
834                 if (same_clid(&clp->cl_clientid, clid))
835                         return clp;
836         }
837         return NULL;
838 }
839
840 static struct nfs4_client *
841 find_unconfirmed_client(clientid_t *clid)
842 {
843         struct nfs4_client *clp;
844         unsigned int idhashval = clientid_hashval(clid->cl_id);
845
846         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
847                 if (same_clid(&clp->cl_clientid, clid))
848                         return clp;
849         }
850         return NULL;
851 }
852
853 /*
854  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
855  * parameter. Matching is based on the fact the at least one of the
856  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
857  *
858  * FIXME: we need to unify the clientid namespaces for nfsv4.x
859  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
860  * and SET_CLIENTID{,_CONFIRM}
861  */
862 static inline int
863 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
864 {
865         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
866         return use_exchange_id == has_exchange_flags;
867 }
868
869 static struct nfs4_client *
870 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
871                              bool use_exchange_id)
872 {
873         struct nfs4_client *clp;
874
875         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
876                 if (same_name(clp->cl_recdir, dname) &&
877                     match_clientid_establishment(clp, use_exchange_id))
878                         return clp;
879         }
880         return NULL;
881 }
882
883 static struct nfs4_client *
884 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
885                                bool use_exchange_id)
886 {
887         struct nfs4_client *clp;
888
889         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
890                 if (same_name(clp->cl_recdir, dname) &&
891                     match_clientid_establishment(clp, use_exchange_id))
892                         return clp;
893         }
894         return NULL;
895 }
896
897 /* a helper function for parse_callback */
898 static int
899 parse_octet(unsigned int *lenp, char **addrp)
900 {
901         unsigned int len = *lenp;
902         char *p = *addrp;
903         int n = -1;
904         char c;
905
906         for (;;) {
907                 if (!len)
908                         break;
909                 len--;
910                 c = *p++;
911                 if (c == '.')
912                         break;
913                 if ((c < '0') || (c > '9')) {
914                         n = -1;
915                         break;
916                 }
917                 if (n < 0)
918                         n = 0;
919                 n = (n * 10) + (c - '0');
920                 if (n > 255) {
921                         n = -1;
922                         break;
923                 }
924         }
925         *lenp = len;
926         *addrp = p;
927         return n;
928 }
929
930 /* parse and set the setclientid ipv4 callback address */
931 static int
932 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
933 {
934         int temp = 0;
935         u32 cbaddr = 0;
936         u16 cbport = 0;
937         u32 addrlen = addr_len;
938         char *addr = addr_val;
939         int i, shift;
940
941         /* ipaddress */
942         shift = 24;
943         for(i = 4; i > 0  ; i--) {
944                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
945                         return 0;
946                 }
947                 cbaddr |= (temp << shift);
948                 if (shift > 0)
949                 shift -= 8;
950         }
951         *cbaddrp = cbaddr;
952
953         /* port */
954         shift = 8;
955         for(i = 2; i > 0  ; i--) {
956                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
957                         return 0;
958                 }
959                 cbport |= (temp << shift);
960                 if (shift > 0)
961                         shift -= 8;
962         }
963         *cbportp = cbport;
964         return 1;
965 }
966
967 static void
968 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
969 {
970         struct nfs4_callback *cb = &clp->cl_callback;
971
972         /* Currently, we only support tcp for the callback channel */
973         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
974                 goto out_err;
975
976         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
977                          &cb->cb_addr, &cb->cb_port)))
978                 goto out_err;
979         cb->cb_prog = se->se_callback_prog;
980         cb->cb_ident = se->se_callback_ident;
981         return;
982 out_err:
983         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
984                 "will not receive delegations\n",
985                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
986
987         return;
988 }
989
990 void
991 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
992 {
993         struct nfsd4_compoundres *resp = rqstp->rq_resp;
994
995         resp->cstate.statp = statp;
996 }
997
998 /*
999  * Dereference the result pages.
1000  */
1001 static void
1002 nfsd4_release_respages(struct page **respages, short resused)
1003 {
1004         int i;
1005
1006         dprintk("--> %s\n", __func__);
1007         for (i = 0; i < resused; i++) {
1008                 if (!respages[i])
1009                         continue;
1010                 put_page(respages[i]);
1011                 respages[i] = NULL;
1012         }
1013 }
1014
1015 static void
1016 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1017 {
1018         int i;
1019
1020         for (i = 0; i < count; i++) {
1021                 topages[i] = frompages[i];
1022                 if (!topages[i])
1023                         continue;
1024                 get_page(topages[i]);
1025         }
1026 }
1027
1028 /*
1029  * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1030  * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1031  * length of the XDR response is less than se_fmaxresp_cached
1032  * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1033  * of the reply (e.g. readdir).
1034  *
1035  * Store the base and length of the rq_req.head[0] page
1036  * of the NFSv4.1 data, just past the rpc header.
1037  */
1038 void
1039 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1040 {
1041         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1042         struct svc_rqst *rqstp = resp->rqstp;
1043         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1044         struct nfsd4_op *op = &args->ops[resp->opcnt];
1045         struct kvec *resv = &rqstp->rq_res.head[0];
1046
1047         dprintk("--> %s entry %p\n", __func__, entry);
1048
1049         /* Don't cache a failed OP_SEQUENCE. */
1050         if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1051                 return;
1052
1053         nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1054         entry->ce_opcnt = resp->opcnt;
1055         entry->ce_status = resp->cstate.status;
1056
1057         /*
1058          * Don't need a page to cache just the sequence operation - the slot
1059          * does this for us!
1060          */
1061
1062         if (nfsd4_not_cached(resp)) {
1063                 entry->ce_resused = 0;
1064                 entry->ce_rpchdrlen = 0;
1065                 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1066                         resp->cstate.slot->sl_cache_entry.ce_cachethis);
1067                 return;
1068         }
1069         entry->ce_resused = rqstp->rq_resused;
1070         if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1071                 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1072         nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1073                          entry->ce_resused);
1074         entry->ce_datav.iov_base = resp->cstate.statp;
1075         entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1076                                 (char *)page_address(rqstp->rq_respages[0]));
1077         /* Current request rpc header length*/
1078         entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1079                                 (char *)page_address(rqstp->rq_respages[0]);
1080 }
1081
1082 /*
1083  * We keep the rpc header, but take the nfs reply from the replycache.
1084  */
1085 static int
1086 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1087                         struct nfsd4_cache_entry *entry)
1088 {
1089         struct svc_rqst *rqstp = resp->rqstp;
1090         struct kvec *resv = &resp->rqstp->rq_res.head[0];
1091         int len;
1092
1093         /* Current request rpc header length*/
1094         len = (char *)resp->cstate.statp -
1095                         (char *)page_address(rqstp->rq_respages[0]);
1096         if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1097                 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1098                         entry->ce_datav.iov_len);
1099                 return 0;
1100         }
1101         /* copy the cached reply nfsd data past the current rpc header */
1102         memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1103                 entry->ce_datav.iov_len);
1104         resv->iov_len = len + entry->ce_datav.iov_len;
1105         return 1;
1106 }
1107
1108 /*
1109  * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1110  * cached page.  Replace any futher replay pages from the cache.
1111  */
1112 __be32
1113 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1114                          struct nfsd4_sequence *seq)
1115 {
1116         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1117         __be32 status;
1118
1119         dprintk("--> %s entry %p\n", __func__, entry);
1120
1121         /*
1122          * If this is just the sequence operation, we did not keep
1123          * a page in the cache entry because we can just use the
1124          * slot info stored in struct nfsd4_sequence that was checked
1125          * against the slot in nfsd4_sequence().
1126          *
1127          * This occurs when seq->cachethis is FALSE, or when the client
1128          * session inactivity timer fires and a solo sequence operation
1129          * is sent (lease renewal).
1130          */
1131         if (seq && nfsd4_not_cached(resp)) {
1132                 seq->maxslots = resp->cstate.session->se_fnumslots;
1133                 return nfs_ok;
1134         }
1135
1136         if (!nfsd41_copy_replay_data(resp, entry)) {
1137                 /*
1138                  * Not enough room to use the replay rpc header, send the
1139                  * cached header. Release all the allocated result pages.
1140                  */
1141                 svc_free_res_pages(resp->rqstp);
1142                 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1143                         entry->ce_resused);
1144         } else {
1145                 /* Release all but the first allocated result page */
1146
1147                 resp->rqstp->rq_resused--;
1148                 svc_free_res_pages(resp->rqstp);
1149
1150                 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1151                                  &entry->ce_respages[1],
1152                                  entry->ce_resused - 1);
1153         }
1154
1155         resp->rqstp->rq_resused = entry->ce_resused;
1156         resp->opcnt = entry->ce_opcnt;
1157         resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1158         status = entry->ce_status;
1159
1160         return status;
1161 }
1162
1163 /*
1164  * Set the exchange_id flags returned by the server.
1165  */
1166 static void
1167 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1168 {
1169         /* pNFS is not supported */
1170         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1171
1172         /* Referrals are supported, Migration is not. */
1173         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1174
1175         /* set the wire flags to return to client. */
1176         clid->flags = new->cl_exchange_flags;
1177 }
1178
1179 __be32
1180 nfsd4_exchange_id(struct svc_rqst *rqstp,
1181                   struct nfsd4_compound_state *cstate,
1182                   struct nfsd4_exchange_id *exid)
1183 {
1184         struct nfs4_client *unconf, *conf, *new;
1185         int status;
1186         unsigned int            strhashval;
1187         char                    dname[HEXDIR_LEN];
1188         nfs4_verifier           verf = exid->verifier;
1189         u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1190
1191         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1192                 " ip_addr=%u flags %x, spa_how %d\n",
1193                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1194                 ip_addr, exid->flags, exid->spa_how);
1195
1196         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1197                 return nfserr_inval;
1198
1199         /* Currently only support SP4_NONE */
1200         switch (exid->spa_how) {
1201         case SP4_NONE:
1202                 break;
1203         case SP4_SSV:
1204                 return nfserr_encr_alg_unsupp;
1205         default:
1206                 BUG();                          /* checked by xdr code */
1207         case SP4_MACH_CRED:
1208                 return nfserr_serverfault;      /* no excuse :-/ */
1209         }
1210
1211         status = nfs4_make_rec_clidname(dname, &exid->clname);
1212
1213         if (status)
1214                 goto error;
1215
1216         strhashval = clientstr_hashval(dname);
1217
1218         nfs4_lock_state();
1219         status = nfs_ok;
1220
1221         conf = find_confirmed_client_by_str(dname, strhashval, true);
1222         if (conf) {
1223                 if (!same_verf(&verf, &conf->cl_verifier)) {
1224                         /* 18.35.4 case 8 */
1225                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1226                                 status = nfserr_not_same;
1227                                 goto out;
1228                         }
1229                         /* Client reboot: destroy old state */
1230                         expire_client(conf);
1231                         goto out_new;
1232                 }
1233                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1234                         /* 18.35.4 case 9 */
1235                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1236                                 status = nfserr_perm;
1237                                 goto out;
1238                         }
1239                         expire_client(conf);
1240                         goto out_new;
1241                 }
1242                 if (ip_addr != conf->cl_addr &&
1243                     !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1244                         /* Client collision. 18.35.4 case 3 */
1245                         status = nfserr_clid_inuse;
1246                         goto out;
1247                 }
1248                 /*
1249                  * Set bit when the owner id and verifier map to an already
1250                  * confirmed client id (18.35.3).
1251                  */
1252                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1253
1254                 /*
1255                  * Falling into 18.35.4 case 2, possible router replay.
1256                  * Leave confirmed record intact and return same result.
1257                  */
1258                 copy_verf(conf, &verf);
1259                 new = conf;
1260                 goto out_copy;
1261         } else {
1262                 /* 18.35.4 case 7 */
1263                 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1264                         status = nfserr_noent;
1265                         goto out;
1266                 }
1267         }
1268
1269         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1270         if (unconf) {
1271                 /*
1272                  * Possible retry or client restart.  Per 18.35.4 case 4,
1273                  * a new unconfirmed record should be generated regardless
1274                  * of whether any properties have changed.
1275                  */
1276                 expire_client(unconf);
1277         }
1278
1279 out_new:
1280         /* Normal case */
1281         new = create_client(exid->clname, dname);
1282         if (new == NULL) {
1283                 status = nfserr_resource;
1284                 goto out;
1285         }
1286
1287         copy_verf(new, &verf);
1288         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1289         new->cl_addr = ip_addr;
1290         gen_clid(new);
1291         gen_confirm(new);
1292         add_to_unconfirmed(new, strhashval);
1293 out_copy:
1294         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1295         exid->clientid.cl_id = new->cl_clientid.cl_id;
1296
1297         new->cl_slot.sl_seqid = 0;
1298         exid->seqid = 1;
1299         nfsd4_set_ex_flags(new, exid);
1300
1301         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1302                 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1303         status = nfs_ok;
1304
1305 out:
1306         nfs4_unlock_state();
1307 error:
1308         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1309         return status;
1310 }
1311
1312 static int
1313 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1314 {
1315         dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1316                 slot->sl_seqid);
1317
1318         /* The slot is in use, and no response has been sent. */
1319         if (slot->sl_inuse) {
1320                 if (seqid == slot->sl_seqid)
1321                         return nfserr_jukebox;
1322                 else
1323                         return nfserr_seq_misordered;
1324         }
1325         /* Normal */
1326         if (likely(seqid == slot->sl_seqid + 1))
1327                 return nfs_ok;
1328         /* Replay */
1329         if (seqid == slot->sl_seqid)
1330                 return nfserr_replay_cache;
1331         /* Wraparound */
1332         if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1333                 return nfs_ok;
1334         /* Misordered replay or misordered new request */
1335         return nfserr_seq_misordered;
1336 }
1337
1338 __be32
1339 nfsd4_create_session(struct svc_rqst *rqstp,
1340                      struct nfsd4_compound_state *cstate,
1341                      struct nfsd4_create_session *cr_ses)
1342 {
1343         u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1344         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1345         struct nfs4_client *conf, *unconf;
1346         struct nfsd4_slot *slot = NULL;
1347         int status = 0;
1348
1349         nfs4_lock_state();
1350         unconf = find_unconfirmed_client(&cr_ses->clientid);
1351         conf = find_confirmed_client(&cr_ses->clientid);
1352
1353         if (conf) {
1354                 slot = &conf->cl_slot;
1355                 status = check_slot_seqid(cr_ses->seqid, slot);
1356                 if (status == nfserr_replay_cache) {
1357                         dprintk("Got a create_session replay! seqid= %d\n",
1358                                 slot->sl_seqid);
1359                         cstate->slot = slot;
1360                         cstate->status = status;
1361                         /* Return the cached reply status */
1362                         status = nfsd4_replay_cache_entry(resp, NULL);
1363                         goto out;
1364                 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1365                         status = nfserr_seq_misordered;
1366                         dprintk("Sequence misordered!\n");
1367                         dprintk("Expected seqid= %d but got seqid= %d\n",
1368                                 slot->sl_seqid, cr_ses->seqid);
1369                         goto out;
1370                 }
1371                 conf->cl_slot.sl_seqid++;
1372         } else if (unconf) {
1373                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1374                     (ip_addr != unconf->cl_addr)) {
1375                         status = nfserr_clid_inuse;
1376                         goto out;
1377                 }
1378
1379                 slot = &unconf->cl_slot;
1380                 status = check_slot_seqid(cr_ses->seqid, slot);
1381                 if (status) {
1382                         /* an unconfirmed replay returns misordered */
1383                         status = nfserr_seq_misordered;
1384                         goto out;
1385                 }
1386
1387                 slot->sl_seqid++; /* from 0 to 1 */
1388                 move_to_confirmed(unconf);
1389
1390                 /*
1391                  * We do not support RDMA or persistent sessions
1392                  */
1393                 cr_ses->flags &= ~SESSION4_PERSIST;
1394                 cr_ses->flags &= ~SESSION4_RDMA;
1395
1396                 conf = unconf;
1397         } else {
1398                 status = nfserr_stale_clientid;
1399                 goto out;
1400         }
1401
1402         status = alloc_init_session(rqstp, conf, cr_ses);
1403         if (status)
1404                 goto out;
1405
1406         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1407                NFS4_MAX_SESSIONID_LEN);
1408         cr_ses->seqid = slot->sl_seqid;
1409
1410         slot->sl_inuse = true;
1411         cstate->slot = slot;
1412         /* Ensure a page is used for the cache */
1413         slot->sl_cache_entry.ce_cachethis = 1;
1414 out:
1415         nfs4_unlock_state();
1416         dprintk("%s returns %d\n", __func__, ntohl(status));
1417         return status;
1418 }
1419
1420 __be32
1421 nfsd4_destroy_session(struct svc_rqst *r,
1422                       struct nfsd4_compound_state *cstate,
1423                       struct nfsd4_destroy_session *sessionid)
1424 {
1425         struct nfsd4_session *ses;
1426         u32 status = nfserr_badsession;
1427
1428         /* Notes:
1429          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1430          * - Should we return nfserr_back_chan_busy if waiting for
1431          *   callbacks on to-be-destroyed session?
1432          * - Do we need to clear any callback info from previous session?
1433          */
1434
1435         dump_sessionid(__func__, &sessionid->sessionid);
1436         spin_lock(&sessionid_lock);
1437         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1438         if (!ses) {
1439                 spin_unlock(&sessionid_lock);
1440                 goto out;
1441         }
1442
1443         unhash_session(ses);
1444         spin_unlock(&sessionid_lock);
1445
1446         /* wait for callbacks */
1447         shutdown_callback_client(ses->se_client);
1448         nfsd4_put_session(ses);
1449         status = nfs_ok;
1450 out:
1451         dprintk("%s returns %d\n", __func__, ntohl(status));
1452         return status;
1453 }
1454
1455 __be32
1456 nfsd4_sequence(struct svc_rqst *rqstp,
1457                struct nfsd4_compound_state *cstate,
1458                struct nfsd4_sequence *seq)
1459 {
1460         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1461         struct nfsd4_session *session;
1462         struct nfsd4_slot *slot;
1463         int status;
1464
1465         if (resp->opcnt != 1)
1466                 return nfserr_sequence_pos;
1467
1468         spin_lock(&sessionid_lock);
1469         status = nfserr_badsession;
1470         session = find_in_sessionid_hashtbl(&seq->sessionid);
1471         if (!session)
1472                 goto out;
1473
1474         status = nfserr_badslot;
1475         if (seq->slotid >= session->se_fnumslots)
1476                 goto out;
1477
1478         slot = &session->se_slots[seq->slotid];
1479         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1480
1481         status = check_slot_seqid(seq->seqid, slot);
1482         if (status == nfserr_replay_cache) {
1483                 cstate->slot = slot;
1484                 cstate->session = session;
1485                 /* Return the cached reply status and set cstate->status
1486                  * for nfsd4_svc_encode_compoundres processing */
1487                 status = nfsd4_replay_cache_entry(resp, seq);
1488                 cstate->status = nfserr_replay_cache;
1489                 goto replay_cache;
1490         }
1491         if (status)
1492                 goto out;
1493
1494         /* Success! bump slot seqid */
1495         slot->sl_inuse = true;
1496         slot->sl_seqid = seq->seqid;
1497         slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1498         /* Always set the cache entry cachethis for solo sequence */
1499         if (nfsd4_is_solo_sequence(resp))
1500                 slot->sl_cache_entry.ce_cachethis = 1;
1501
1502         cstate->slot = slot;
1503         cstate->session = session;
1504
1505 replay_cache:
1506         /* Renew the clientid on success and on replay.
1507          * Hold a session reference until done processing the compound:
1508          * nfsd4_put_session called only if the cstate slot is set.
1509          */
1510         renew_client(session->se_client);
1511         nfsd4_get_session(session);
1512 out:
1513         spin_unlock(&sessionid_lock);
1514         dprintk("%s: return %d\n", __func__, ntohl(status));
1515         return status;
1516 }
1517
1518 __be32
1519 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1520                   struct nfsd4_setclientid *setclid)
1521 {
1522         struct sockaddr_in      *sin = svc_addr_in(rqstp);
1523         struct xdr_netobj       clname = { 
1524                 .len = setclid->se_namelen,
1525                 .data = setclid->se_name,
1526         };
1527         nfs4_verifier           clverifier = setclid->se_verf;
1528         unsigned int            strhashval;
1529         struct nfs4_client      *conf, *unconf, *new;
1530         __be32                  status;
1531         char                    *princ;
1532         char                    dname[HEXDIR_LEN];
1533         
1534         if (!check_name(clname))
1535                 return nfserr_inval;
1536
1537         status = nfs4_make_rec_clidname(dname, &clname);
1538         if (status)
1539                 return status;
1540
1541         /* 
1542          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1543          * We get here on a DRC miss.
1544          */
1545
1546         strhashval = clientstr_hashval(dname);
1547
1548         nfs4_lock_state();
1549         conf = find_confirmed_client_by_str(dname, strhashval, false);
1550         if (conf) {
1551                 /* RFC 3530 14.2.33 CASE 0: */
1552                 status = nfserr_clid_inuse;
1553                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1554                         dprintk("NFSD: setclientid: string in use by client"
1555                                 " at %pI4\n", &conf->cl_addr);
1556                         goto out;
1557                 }
1558         }
1559         /*
1560          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1561          * has a description of SETCLIENTID request processing consisting
1562          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1563          */
1564         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1565         status = nfserr_resource;
1566         if (!conf) {
1567                 /*
1568                  * RFC 3530 14.2.33 CASE 4:
1569                  * placed first, because it is the normal case
1570                  */
1571                 if (unconf)
1572                         expire_client(unconf);
1573                 new = create_client(clname, dname);
1574                 if (new == NULL)
1575                         goto out;
1576                 gen_clid(new);
1577         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1578                 /*
1579                  * RFC 3530 14.2.33 CASE 1:
1580                  * probable callback update
1581                  */
1582                 if (unconf) {
1583                         /* Note this is removing unconfirmed {*x***},
1584                          * which is stronger than RFC recommended {vxc**}.
1585                          * This has the advantage that there is at most
1586                          * one {*x***} in either list at any time.
1587                          */
1588                         expire_client(unconf);
1589                 }
1590                 new = create_client(clname, dname);
1591                 if (new == NULL)
1592                         goto out;
1593                 copy_clid(new, conf);
1594         } else if (!unconf) {
1595                 /*
1596                  * RFC 3530 14.2.33 CASE 2:
1597                  * probable client reboot; state will be removed if
1598                  * confirmed.
1599                  */
1600                 new = create_client(clname, dname);
1601                 if (new == NULL)
1602                         goto out;
1603                 gen_clid(new);
1604         } else {
1605                 /*
1606                  * RFC 3530 14.2.33 CASE 3:
1607                  * probable client reboot; state will be removed if
1608                  * confirmed.
1609                  */
1610                 expire_client(unconf);
1611                 new = create_client(clname, dname);
1612                 if (new == NULL)
1613                         goto out;
1614                 gen_clid(new);
1615         }
1616         copy_verf(new, &clverifier);
1617         new->cl_addr = sin->sin_addr.s_addr;
1618         new->cl_flavor = rqstp->rq_flavor;
1619         princ = svc_gss_principal(rqstp);
1620         if (princ) {
1621                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1622                 if (new->cl_principal == NULL) {
1623                         free_client(new);
1624                         goto out;
1625                 }
1626         }
1627         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1628         gen_confirm(new);
1629         gen_callback(new, setclid);
1630         add_to_unconfirmed(new, strhashval);
1631         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1632         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1633         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1634         status = nfs_ok;
1635 out:
1636         nfs4_unlock_state();
1637         return status;
1638 }
1639
1640
1641 /*
1642  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1643  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1644  * bullets, labeled as CASE1 - CASE4 below.
1645  */
1646 __be32
1647 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1648                          struct nfsd4_compound_state *cstate,
1649                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1650 {
1651         struct sockaddr_in *sin = svc_addr_in(rqstp);
1652         struct nfs4_client *conf, *unconf;
1653         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1654         clientid_t * clid = &setclientid_confirm->sc_clientid;
1655         __be32 status;
1656
1657         if (STALE_CLIENTID(clid))
1658                 return nfserr_stale_clientid;
1659         /* 
1660          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1661          * We get here on a DRC miss.
1662          */
1663
1664         nfs4_lock_state();
1665
1666         conf = find_confirmed_client(clid);
1667         unconf = find_unconfirmed_client(clid);
1668
1669         status = nfserr_clid_inuse;
1670         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1671                 goto out;
1672         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1673                 goto out;
1674
1675         /*
1676          * section 14.2.34 of RFC 3530 has a description of
1677          * SETCLIENTID_CONFIRM request processing consisting
1678          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1679          */
1680         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1681                 /*
1682                  * RFC 3530 14.2.34 CASE 1:
1683                  * callback update
1684                  */
1685                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1686                         status = nfserr_clid_inuse;
1687                 else {
1688                         /* XXX: We just turn off callbacks until we can handle
1689                           * change request correctly. */
1690                         atomic_set(&conf->cl_callback.cb_set, 0);
1691                         gen_confirm(conf);
1692                         nfsd4_remove_clid_dir(unconf);
1693                         expire_client(unconf);
1694                         status = nfs_ok;
1695
1696                 }
1697         } else if (conf && !unconf) {
1698                 /*
1699                  * RFC 3530 14.2.34 CASE 2:
1700                  * probable retransmitted request; play it safe and
1701                  * do nothing.
1702                  */
1703                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1704                         status = nfserr_clid_inuse;
1705                 else
1706                         status = nfs_ok;
1707         } else if (!conf && unconf
1708                         && same_verf(&unconf->cl_confirm, &confirm)) {
1709                 /*
1710                  * RFC 3530 14.2.34 CASE 3:
1711                  * Normal case; new or rebooted client:
1712                  */
1713                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1714                         status = nfserr_clid_inuse;
1715                 } else {
1716                         unsigned int hash =
1717                                 clientstr_hashval(unconf->cl_recdir);
1718                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1719                                                             hash, false);
1720                         if (conf) {
1721                                 nfsd4_remove_clid_dir(conf);
1722                                 expire_client(conf);
1723                         }
1724                         move_to_confirmed(unconf);
1725                         conf = unconf;
1726                         nfsd4_probe_callback(conf);
1727                         status = nfs_ok;
1728                 }
1729         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1730             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1731                                                                 &confirm)))) {
1732                 /*
1733                  * RFC 3530 14.2.34 CASE 4:
1734                  * Client probably hasn't noticed that we rebooted yet.
1735                  */
1736                 status = nfserr_stale_clientid;
1737         } else {
1738                 /* check that we have hit one of the cases...*/
1739                 status = nfserr_clid_inuse;
1740         }
1741 out:
1742         nfs4_unlock_state();
1743         return status;
1744 }
1745
1746 /* OPEN Share state helper functions */
1747 static inline struct nfs4_file *
1748 alloc_init_file(struct inode *ino)
1749 {
1750         struct nfs4_file *fp;
1751         unsigned int hashval = file_hashval(ino);
1752
1753         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1754         if (fp) {
1755                 atomic_set(&fp->fi_ref, 1);
1756                 INIT_LIST_HEAD(&fp->fi_hash);
1757                 INIT_LIST_HEAD(&fp->fi_stateids);
1758                 INIT_LIST_HEAD(&fp->fi_delegations);
1759                 spin_lock(&recall_lock);
1760                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1761                 spin_unlock(&recall_lock);
1762                 fp->fi_inode = igrab(ino);
1763                 fp->fi_id = current_fileid++;
1764                 fp->fi_had_conflict = false;
1765                 return fp;
1766         }
1767         return NULL;
1768 }
1769
1770 static void
1771 nfsd4_free_slab(struct kmem_cache **slab)
1772 {
1773         if (*slab == NULL)
1774                 return;
1775         kmem_cache_destroy(*slab);
1776         *slab = NULL;
1777 }
1778
1779 void
1780 nfsd4_free_slabs(void)
1781 {
1782         nfsd4_free_slab(&stateowner_slab);
1783         nfsd4_free_slab(&file_slab);
1784         nfsd4_free_slab(&stateid_slab);
1785         nfsd4_free_slab(&deleg_slab);
1786 }
1787
1788 static int
1789 nfsd4_init_slabs(void)
1790 {
1791         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1792                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1793         if (stateowner_slab == NULL)
1794                 goto out_nomem;
1795         file_slab = kmem_cache_create("nfsd4_files",
1796                         sizeof(struct nfs4_file), 0, 0, NULL);
1797         if (file_slab == NULL)
1798                 goto out_nomem;
1799         stateid_slab = kmem_cache_create("nfsd4_stateids",
1800                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1801         if (stateid_slab == NULL)
1802                 goto out_nomem;
1803         deleg_slab = kmem_cache_create("nfsd4_delegations",
1804                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1805         if (deleg_slab == NULL)
1806                 goto out_nomem;
1807         return 0;
1808 out_nomem:
1809         nfsd4_free_slabs();
1810         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1811         return -ENOMEM;
1812 }
1813
1814 void
1815 nfs4_free_stateowner(struct kref *kref)
1816 {
1817         struct nfs4_stateowner *sop =
1818                 container_of(kref, struct nfs4_stateowner, so_ref);
1819         kfree(sop->so_owner.data);
1820         kmem_cache_free(stateowner_slab, sop);
1821 }
1822
1823 static inline struct nfs4_stateowner *
1824 alloc_stateowner(struct xdr_netobj *owner)
1825 {
1826         struct nfs4_stateowner *sop;
1827
1828         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1829                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1830                         memcpy(sop->so_owner.data, owner->data, owner->len);
1831                         sop->so_owner.len = owner->len;
1832                         kref_init(&sop->so_ref);
1833                         return sop;
1834                 } 
1835                 kmem_cache_free(stateowner_slab, sop);
1836         }
1837         return NULL;
1838 }
1839
1840 static struct nfs4_stateowner *
1841 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1842         struct nfs4_stateowner *sop;
1843         struct nfs4_replay *rp;
1844         unsigned int idhashval;
1845
1846         if (!(sop = alloc_stateowner(&open->op_owner)))
1847                 return NULL;
1848         idhashval = ownerid_hashval(current_ownerid);
1849         INIT_LIST_HEAD(&sop->so_idhash);
1850         INIT_LIST_HEAD(&sop->so_strhash);
1851         INIT_LIST_HEAD(&sop->so_perclient);
1852         INIT_LIST_HEAD(&sop->so_stateids);
1853         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1854         INIT_LIST_HEAD(&sop->so_close_lru);
1855         sop->so_time = 0;
1856         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1857         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1858         list_add(&sop->so_perclient, &clp->cl_openowners);
1859         sop->so_is_open_owner = 1;
1860         sop->so_id = current_ownerid++;
1861         sop->so_client = clp;
1862         sop->so_seqid = open->op_seqid;
1863         sop->so_confirmed = 0;
1864         rp = &sop->so_replay;
1865         rp->rp_status = nfserr_serverfault;
1866         rp->rp_buflen = 0;
1867         rp->rp_buf = rp->rp_ibuf;
1868         return sop;
1869 }
1870
1871 static inline void
1872 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1873         struct nfs4_stateowner *sop = open->op_stateowner;
1874         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1875
1876         INIT_LIST_HEAD(&stp->st_hash);
1877         INIT_LIST_HEAD(&stp->st_perstateowner);
1878         INIT_LIST_HEAD(&stp->st_lockowners);
1879         INIT_LIST_HEAD(&stp->st_perfile);
1880         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1881         list_add(&stp->st_perstateowner, &sop->so_stateids);
1882         list_add(&stp->st_perfile, &fp->fi_stateids);
1883         stp->st_stateowner = sop;
1884         get_nfs4_file(fp);
1885         stp->st_file = fp;
1886         stp->st_stateid.si_boot = get_seconds();
1887         stp->st_stateid.si_stateownerid = sop->so_id;
1888         stp->st_stateid.si_fileid = fp->fi_id;
1889         stp->st_stateid.si_generation = 0;
1890         stp->st_access_bmap = 0;
1891         stp->st_deny_bmap = 0;
1892         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1893                   &stp->st_access_bmap);
1894         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1895         stp->st_openstp = NULL;
1896 }
1897
1898 static void
1899 move_to_close_lru(struct nfs4_stateowner *sop)
1900 {
1901         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1902
1903         list_move_tail(&sop->so_close_lru, &close_lru);
1904         sop->so_time = get_seconds();
1905 }
1906
1907 static int
1908 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1909                                                         clientid_t *clid)
1910 {
1911         return (sop->so_owner.len == owner->len) &&
1912                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1913                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1914 }
1915
1916 static struct nfs4_stateowner *
1917 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1918 {
1919         struct nfs4_stateowner *so = NULL;
1920
1921         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1922                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1923                         return so;
1924         }
1925         return NULL;
1926 }
1927
1928 /* search file_hashtbl[] for file */
1929 static struct nfs4_file *
1930 find_file(struct inode *ino)
1931 {
1932         unsigned int hashval = file_hashval(ino);
1933         struct nfs4_file *fp;
1934
1935         spin_lock(&recall_lock);
1936         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1937                 if (fp->fi_inode == ino) {
1938                         get_nfs4_file(fp);
1939                         spin_unlock(&recall_lock);
1940                         return fp;
1941                 }
1942         }
1943         spin_unlock(&recall_lock);
1944         return NULL;
1945 }
1946
1947 static inline int access_valid(u32 x, u32 minorversion)
1948 {
1949         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1950                 return 0;
1951         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1952                 return 0;
1953         x &= ~NFS4_SHARE_ACCESS_MASK;
1954         if (minorversion && x) {
1955                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1956                         return 0;
1957                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1958                         return 0;
1959                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1960         }
1961         if (x)
1962                 return 0;
1963         return 1;
1964 }
1965
1966 static inline int deny_valid(u32 x)
1967 {
1968         /* Note: unlike access bits, deny bits may be zero. */
1969         return x <= NFS4_SHARE_DENY_BOTH;
1970 }
1971
1972 /*
1973  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1974  * st_{access,deny}_bmap field of the stateid, in order to track not
1975  * only what share bits are currently in force, but also what
1976  * combinations of share bits previous opens have used.  This allows us
1977  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1978  * return an error if the client attempt to downgrade to a combination
1979  * of share bits not explicable by closing some of its previous opens.
1980  *
1981  * XXX: This enforcement is actually incomplete, since we don't keep
1982  * track of access/deny bit combinations; so, e.g., we allow:
1983  *
1984  *      OPEN allow read, deny write
1985  *      OPEN allow both, deny none
1986  *      DOWNGRADE allow read, deny none
1987  *
1988  * which we should reject.
1989  */
1990 static void
1991 set_access(unsigned int *access, unsigned long bmap) {
1992         int i;
1993
1994         *access = 0;
1995         for (i = 1; i < 4; i++) {
1996                 if (test_bit(i, &bmap))
1997                         *access |= i;
1998         }
1999 }
2000
2001 static void
2002 set_deny(unsigned int *deny, unsigned long bmap) {
2003         int i;
2004
2005         *deny = 0;
2006         for (i = 0; i < 4; i++) {
2007                 if (test_bit(i, &bmap))
2008                         *deny |= i ;
2009         }
2010 }
2011
2012 static int
2013 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2014         unsigned int access, deny;
2015
2016         set_access(&access, stp->st_access_bmap);
2017         set_deny(&deny, stp->st_deny_bmap);
2018         if ((access & open->op_share_deny) || (deny & open->op_share_access))
2019                 return 0;
2020         return 1;
2021 }
2022
2023 /*
2024  * Called to check deny when READ with all zero stateid or
2025  * WRITE with all zero or all one stateid
2026  */
2027 static __be32
2028 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2029 {
2030         struct inode *ino = current_fh->fh_dentry->d_inode;
2031         struct nfs4_file *fp;
2032         struct nfs4_stateid *stp;
2033         __be32 ret;
2034
2035         dprintk("NFSD: nfs4_share_conflict\n");
2036
2037         fp = find_file(ino);
2038         if (!fp)
2039                 return nfs_ok;
2040         ret = nfserr_locked;
2041         /* Search for conflicting share reservations */
2042         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2043                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2044                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2045                         goto out;
2046         }
2047         ret = nfs_ok;
2048 out:
2049         put_nfs4_file(fp);
2050         return ret;
2051 }
2052
2053 static inline void
2054 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2055 {
2056         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2057                 drop_file_write_access(filp);
2058                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2059         }
2060 }
2061
2062 /*
2063  * Recall a delegation
2064  */
2065 static int
2066 do_recall(void *__dp)
2067 {
2068         struct nfs4_delegation *dp = __dp;
2069
2070         dp->dl_file->fi_had_conflict = true;
2071         nfsd4_cb_recall(dp);
2072         return 0;
2073 }
2074
2075 /*
2076  * Spawn a thread to perform a recall on the delegation represented
2077  * by the lease (file_lock)
2078  *
2079  * Called from break_lease() with lock_kernel() held.
2080  * Note: we assume break_lease will only call this *once* for any given
2081  * lease.
2082  */
2083 static
2084 void nfsd_break_deleg_cb(struct file_lock *fl)
2085 {
2086         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
2087         struct task_struct *t;
2088
2089         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2090         if (!dp)
2091                 return;
2092
2093         /* We're assuming the state code never drops its reference
2094          * without first removing the lease.  Since we're in this lease
2095          * callback (and since the lease code is serialized by the kernel
2096          * lock) we know the server hasn't removed the lease yet, we know
2097          * it's safe to take a reference: */
2098         atomic_inc(&dp->dl_count);
2099         atomic_inc(&dp->dl_client->cl_count);
2100
2101         spin_lock(&recall_lock);
2102         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2103         spin_unlock(&recall_lock);
2104
2105         /* only place dl_time is set. protected by lock_kernel*/
2106         dp->dl_time = get_seconds();
2107
2108         /*
2109          * We don't want the locks code to timeout the lease for us;
2110          * we'll remove it ourself if the delegation isn't returned
2111          * in time.
2112          */
2113         fl->fl_break_time = 0;
2114
2115         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2116         if (IS_ERR(t)) {
2117                 struct nfs4_client *clp = dp->dl_client;
2118
2119                 printk(KERN_INFO "NFSD: Callback thread failed for "
2120                         "for client (clientid %08x/%08x)\n",
2121                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2122                 put_nfs4_client(dp->dl_client);
2123                 nfs4_put_delegation(dp);
2124         }
2125 }
2126
2127 /*
2128  * The file_lock is being reapd.
2129  *
2130  * Called by locks_free_lock() with lock_kernel() held.
2131  */
2132 static
2133 void nfsd_release_deleg_cb(struct file_lock *fl)
2134 {
2135         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2136
2137         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2138
2139         if (!(fl->fl_flags & FL_LEASE) || !dp)
2140                 return;
2141         dp->dl_flock = NULL;
2142 }
2143
2144 /*
2145  * Set the delegation file_lock back pointer.
2146  *
2147  * Called from setlease() with lock_kernel() held.
2148  */
2149 static
2150 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2151 {
2152         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2153
2154         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2155         if (!dp)
2156                 return;
2157         dp->dl_flock = new;
2158 }
2159
2160 /*
2161  * Called from setlease() with lock_kernel() held
2162  */
2163 static
2164 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2165 {
2166         struct nfs4_delegation *onlistd =
2167                 (struct nfs4_delegation *)onlist->fl_owner;
2168         struct nfs4_delegation *tryd =
2169                 (struct nfs4_delegation *)try->fl_owner;
2170
2171         if (onlist->fl_lmops != try->fl_lmops)
2172                 return 0;
2173
2174         return onlistd->dl_client == tryd->dl_client;
2175 }
2176
2177
2178 static
2179 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2180 {
2181         if (arg & F_UNLCK)
2182                 return lease_modify(onlist, arg);
2183         else
2184                 return -EAGAIN;
2185 }
2186
2187 static struct lock_manager_operations nfsd_lease_mng_ops = {
2188         .fl_break = nfsd_break_deleg_cb,
2189         .fl_release_private = nfsd_release_deleg_cb,
2190         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2191         .fl_mylease = nfsd_same_client_deleg_cb,
2192         .fl_change = nfsd_change_deleg_cb,
2193 };
2194
2195
2196 __be32
2197 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2198                     struct nfsd4_open *open)
2199 {
2200         clientid_t *clientid = &open->op_clientid;
2201         struct nfs4_client *clp = NULL;
2202         unsigned int strhashval;
2203         struct nfs4_stateowner *sop = NULL;
2204
2205         if (!check_name(open->op_owner))
2206                 return nfserr_inval;
2207
2208         if (STALE_CLIENTID(&open->op_clientid))
2209                 return nfserr_stale_clientid;
2210
2211         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2212         sop = find_openstateowner_str(strhashval, open);
2213         open->op_stateowner = sop;
2214         if (!sop) {
2215                 /* Make sure the client's lease hasn't expired. */
2216                 clp = find_confirmed_client(clientid);
2217                 if (clp == NULL)
2218                         return nfserr_expired;
2219                 goto renew;
2220         }
2221         /* When sessions are used, skip open sequenceid processing */
2222         if (nfsd4_has_session(cstate))
2223                 goto renew;
2224         if (!sop->so_confirmed) {
2225                 /* Replace unconfirmed owners without checking for replay. */
2226                 clp = sop->so_client;
2227                 release_openowner(sop);
2228                 open->op_stateowner = NULL;
2229                 goto renew;
2230         }
2231         if (open->op_seqid == sop->so_seqid - 1) {
2232                 if (sop->so_replay.rp_buflen)
2233                         return nfserr_replay_me;
2234                 /* The original OPEN failed so spectacularly
2235                  * that we don't even have replay data saved!
2236                  * Therefore, we have no choice but to continue
2237                  * processing this OPEN; presumably, we'll
2238                  * fail again for the same reason.
2239                  */
2240                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2241                 goto renew;
2242         }
2243         if (open->op_seqid != sop->so_seqid)
2244                 return nfserr_bad_seqid;
2245 renew:
2246         if (open->op_stateowner == NULL) {
2247                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2248                 if (sop == NULL)
2249                         return nfserr_resource;
2250                 open->op_stateowner = sop;
2251         }
2252         list_del_init(&sop->so_close_lru);
2253         renew_client(sop->so_client);
2254         return nfs_ok;
2255 }
2256
2257 static inline __be32
2258 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2259 {
2260         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2261                 return nfserr_openmode;
2262         else
2263                 return nfs_ok;
2264 }
2265
2266 static struct nfs4_delegation *
2267 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2268 {
2269         struct nfs4_delegation *dp;
2270
2271         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2272                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2273                         return dp;
2274         }
2275         return NULL;
2276 }
2277
2278 static __be32
2279 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2280                 struct nfs4_delegation **dp)
2281 {
2282         int flags;
2283         __be32 status = nfserr_bad_stateid;
2284
2285         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2286         if (*dp == NULL)
2287                 goto out;
2288         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2289                                                 RD_STATE : WR_STATE;
2290         status = nfs4_check_delegmode(*dp, flags);
2291         if (status)
2292                 *dp = NULL;
2293 out:
2294         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2295                 return nfs_ok;
2296         if (status)
2297                 return status;
2298         open->op_stateowner->so_confirmed = 1;
2299         return nfs_ok;
2300 }
2301
2302 static __be32
2303 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2304 {
2305         struct nfs4_stateid *local;
2306         __be32 status = nfserr_share_denied;
2307         struct nfs4_stateowner *sop = open->op_stateowner;
2308
2309         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2310                 /* ignore lock owners */
2311                 if (local->st_stateowner->so_is_open_owner == 0)
2312                         continue;
2313                 /* remember if we have seen this open owner */
2314                 if (local->st_stateowner == sop)
2315                         *stpp = local;
2316                 /* check for conflicting share reservations */
2317                 if (!test_share(local, open))
2318                         goto out;
2319         }
2320         status = 0;
2321 out:
2322         return status;
2323 }
2324
2325 static inline struct nfs4_stateid *
2326 nfs4_alloc_stateid(void)
2327 {
2328         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2329 }
2330
2331 static __be32
2332 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2333                 struct nfs4_delegation *dp,
2334                 struct svc_fh *cur_fh, int flags)
2335 {
2336         struct nfs4_stateid *stp;
2337
2338         stp = nfs4_alloc_stateid();
2339         if (stp == NULL)
2340                 return nfserr_resource;
2341
2342         if (dp) {
2343                 get_file(dp->dl_vfs_file);
2344                 stp->st_vfs_file = dp->dl_vfs_file;
2345         } else {
2346                 __be32 status;
2347                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2348                                 &stp->st_vfs_file);
2349                 if (status) {
2350                         if (status == nfserr_dropit)
2351                                 status = nfserr_jukebox;
2352                         kmem_cache_free(stateid_slab, stp);
2353                         return status;
2354                 }
2355         }
2356         *stpp = stp;
2357         return 0;
2358 }
2359
2360 static inline __be32
2361 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2362                 struct nfsd4_open *open)
2363 {
2364         struct iattr iattr = {
2365                 .ia_valid = ATTR_SIZE,
2366                 .ia_size = 0,
2367         };
2368         if (!open->op_truncate)
2369                 return 0;
2370         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2371                 return nfserr_inval;
2372         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2373 }
2374
2375 static __be32
2376 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2377 {
2378         struct file *filp = stp->st_vfs_file;
2379         struct inode *inode = filp->f_path.dentry->d_inode;
2380         unsigned int share_access, new_writer;
2381         __be32 status;
2382
2383         set_access(&share_access, stp->st_access_bmap);
2384         new_writer = (~share_access) & open->op_share_access
2385                         & NFS4_SHARE_ACCESS_WRITE;
2386
2387         if (new_writer) {
2388                 int err = get_write_access(inode);
2389                 if (err)
2390                         return nfserrno(err);
2391                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2392                 if (err)
2393                         return nfserrno(err);
2394                 file_take_write(filp);
2395         }
2396         status = nfsd4_truncate(rqstp, cur_fh, open);
2397         if (status) {
2398                 if (new_writer)
2399                         put_write_access(inode);
2400                 return status;
2401         }
2402         /* remember the open */
2403         filp->f_mode |= open->op_share_access;
2404         __set_bit(open->op_share_access, &stp->st_access_bmap);
2405         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2406
2407         return nfs_ok;
2408 }
2409
2410
2411 static void
2412 nfs4_set_claim_prev(struct nfsd4_open *open)
2413 {
2414         open->op_stateowner->so_confirmed = 1;
2415         open->op_stateowner->so_client->cl_firststate = 1;
2416 }
2417
2418 /*
2419  * Attempt to hand out a delegation.
2420  */
2421 static void
2422 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2423 {
2424         struct nfs4_delegation *dp;
2425         struct nfs4_stateowner *sop = stp->st_stateowner;
2426         struct nfs4_callback *cb = &sop->so_client->cl_callback;
2427         struct file_lock fl, *flp = &fl;
2428         int status, flag = 0;
2429
2430         flag = NFS4_OPEN_DELEGATE_NONE;
2431         open->op_recall = 0;
2432         switch (open->op_claim_type) {
2433                 case NFS4_OPEN_CLAIM_PREVIOUS:
2434                         if (!atomic_read(&cb->cb_set))
2435                                 open->op_recall = 1;
2436                         flag = open->op_delegate_type;
2437                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2438                                 goto out;
2439                         break;
2440                 case NFS4_OPEN_CLAIM_NULL:
2441                         /* Let's not give out any delegations till everyone's
2442                          * had the chance to reclaim theirs.... */
2443                         if (locks_in_grace())
2444                                 goto out;
2445                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2446                                 goto out;
2447                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2448                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2449                         else
2450                                 flag = NFS4_OPEN_DELEGATE_READ;
2451                         break;
2452                 default:
2453                         goto out;
2454         }
2455
2456         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2457         if (dp == NULL) {
2458                 flag = NFS4_OPEN_DELEGATE_NONE;
2459                 goto out;
2460         }
2461         locks_init_lock(&fl);
2462         fl.fl_lmops = &nfsd_lease_mng_ops;
2463         fl.fl_flags = FL_LEASE;
2464         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2465         fl.fl_end = OFFSET_MAX;
2466         fl.fl_owner =  (fl_owner_t)dp;
2467         fl.fl_file = stp->st_vfs_file;
2468         fl.fl_pid = current->tgid;
2469
2470         /* vfs_setlease checks to see if delegation should be handed out.
2471          * the lock_manager callbacks fl_mylease and fl_change are used
2472          */
2473         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2474                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2475                 unhash_delegation(dp);
2476                 flag = NFS4_OPEN_DELEGATE_NONE;
2477                 goto out;
2478         }
2479
2480         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2481
2482         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2483                      dp->dl_stateid.si_boot,
2484                      dp->dl_stateid.si_stateownerid,
2485                      dp->dl_stateid.si_fileid,
2486                      dp->dl_stateid.si_generation);
2487 out:
2488         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2489                         && flag == NFS4_OPEN_DELEGATE_NONE
2490                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2491                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2492         open->op_delegate_type = flag;
2493 }
2494
2495 /*
2496  * called with nfs4_lock_state() held.
2497  */
2498 __be32
2499 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2500 {
2501         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2502         struct nfs4_file *fp = NULL;
2503         struct inode *ino = current_fh->fh_dentry->d_inode;
2504         struct nfs4_stateid *stp = NULL;
2505         struct nfs4_delegation *dp = NULL;
2506         __be32 status;
2507
2508         status = nfserr_inval;
2509         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2510                         || !deny_valid(open->op_share_deny))
2511                 goto out;
2512         /*
2513          * Lookup file; if found, lookup stateid and check open request,
2514          * and check for delegations in the process of being recalled.
2515          * If not found, create the nfs4_file struct
2516          */
2517         fp = find_file(ino);
2518         if (fp) {
2519                 if ((status = nfs4_check_open(fp, open, &stp)))
2520                         goto out;
2521                 status = nfs4_check_deleg(fp, open, &dp);
2522                 if (status)
2523                         goto out;
2524         } else {
2525                 status = nfserr_bad_stateid;
2526                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2527                         goto out;
2528                 status = nfserr_resource;
2529                 fp = alloc_init_file(ino);
2530                 if (fp == NULL)
2531                         goto out;
2532         }
2533
2534         /*
2535          * OPEN the file, or upgrade an existing OPEN.
2536          * If truncate fails, the OPEN fails.
2537          */
2538         if (stp) {
2539                 /* Stateid was found, this is an OPEN upgrade */
2540                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2541                 if (status)
2542                         goto out;
2543                 update_stateid(&stp->st_stateid);
2544         } else {
2545                 /* Stateid was not found, this is a new OPEN */
2546                 int flags = 0;
2547                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2548                         flags |= NFSD_MAY_READ;
2549                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2550                         flags |= NFSD_MAY_WRITE;
2551                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2552                 if (status)
2553                         goto out;
2554                 init_stateid(stp, fp, open);
2555                 status = nfsd4_truncate(rqstp, current_fh, open);
2556                 if (status) {
2557                         release_open_stateid(stp);
2558                         goto out;
2559                 }
2560                 if (nfsd4_has_session(&resp->cstate))
2561                         update_stateid(&stp->st_stateid);
2562         }
2563         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2564
2565         if (nfsd4_has_session(&resp->cstate))
2566                 open->op_stateowner->so_confirmed = 1;
2567
2568         /*
2569         * Attempt to hand out a delegation. No error return, because the
2570         * OPEN succeeds even if we fail.
2571         */
2572         nfs4_open_delegation(current_fh, open, stp);
2573
2574         status = nfs_ok;
2575
2576         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2577                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2578                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2579 out:
2580         if (fp)
2581                 put_nfs4_file(fp);
2582         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2583                 nfs4_set_claim_prev(open);
2584         /*
2585         * To finish the open response, we just need to set the rflags.
2586         */
2587         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2588         if (!open->op_stateowner->so_confirmed &&
2589             !nfsd4_has_session(&resp->cstate))
2590                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2591
2592         return status;
2593 }
2594
2595 __be32
2596 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2597             clientid_t *clid)
2598 {
2599         struct nfs4_client *clp;
2600         __be32 status;
2601
2602         nfs4_lock_state();
2603         dprintk("process_renew(%08x/%08x): starting\n", 
2604                         clid->cl_boot, clid->cl_id);
2605         status = nfserr_stale_clientid;
2606         if (STALE_CLIENTID(clid))
2607                 goto out;
2608         clp = find_confirmed_client(clid);
2609         status = nfserr_expired;
2610         if (clp == NULL) {
2611                 /* We assume the client took too long to RENEW. */
2612                 dprintk("nfsd4_renew: clientid not found!\n");
2613                 goto out;
2614         }
2615         renew_client(clp);
2616         status = nfserr_cb_path_down;
2617         if (!list_empty(&clp->cl_delegations)
2618                         && !atomic_read(&clp->cl_callback.cb_set))
2619                 goto out;
2620         status = nfs_ok;
2621 out:
2622         nfs4_unlock_state();
2623         return status;
2624 }
2625
2626 struct lock_manager nfsd4_manager = {
2627 };
2628
2629 static void
2630 nfsd4_end_grace(void)
2631 {
2632         dprintk("NFSD: end of grace period\n");
2633         nfsd4_recdir_purge_old();
2634         locks_end_grace(&nfsd4_manager);
2635 }
2636
2637 static time_t
2638 nfs4_laundromat(void)
2639 {
2640         struct nfs4_client *clp;
2641         struct nfs4_stateowner *sop;
2642         struct nfs4_delegation *dp;
2643         struct list_head *pos, *next, reaplist;
2644         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2645         time_t t, clientid_val = NFSD_LEASE_TIME;
2646         time_t u, test_val = NFSD_LEASE_TIME;
2647
2648         nfs4_lock_state();
2649
2650         dprintk("NFSD: laundromat service - starting\n");
2651         if (locks_in_grace())
2652                 nfsd4_end_grace();
2653         list_for_each_safe(pos, next, &client_lru) {
2654                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2655                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2656                         t = clp->cl_time - cutoff;
2657                         if (clientid_val > t)
2658                                 clientid_val = t;
2659                         break;
2660                 }
2661                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2662                         clp->cl_clientid.cl_id);
2663                 nfsd4_remove_clid_dir(clp);
2664                 expire_client(clp);
2665         }
2666         INIT_LIST_HEAD(&reaplist);
2667         spin_lock(&recall_lock);
2668         list_for_each_safe(pos, next, &del_recall_lru) {
2669                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2670                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2671                         u = dp->dl_time - cutoff;
2672                         if (test_val > u)
2673                                 test_val = u;
2674                         break;
2675                 }
2676                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2677                                     dp, dp->dl_flock);
2678                 list_move(&dp->dl_recall_lru, &reaplist);
2679         }
2680         spin_unlock(&recall_lock);
2681         list_for_each_safe(pos, next, &reaplist) {
2682                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2683                 list_del_init(&dp->dl_recall_lru);
2684                 unhash_delegation(dp);
2685         }
2686         test_val = NFSD_LEASE_TIME;
2687         list_for_each_safe(pos, next, &close_lru) {
2688                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2689                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2690                         u = sop->so_time - cutoff;
2691                         if (test_val > u)
2692                                 test_val = u;
2693                         break;
2694                 }
2695                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2696                         sop->so_id);
2697                 release_openowner(sop);
2698         }
2699         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2700                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2701         nfs4_unlock_state();
2702         return clientid_val;
2703 }
2704
2705 static struct workqueue_struct *laundry_wq;
2706 static void laundromat_main(struct work_struct *);
2707 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2708
2709 static void
2710 laundromat_main(struct work_struct *not_used)
2711 {
2712         time_t t;
2713
2714         t = nfs4_laundromat();
2715         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2716         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2717 }
2718
2719 static struct nfs4_stateowner *
2720 search_close_lru(u32 st_id, int flags)
2721 {
2722         struct nfs4_stateowner *local = NULL;
2723
2724         if (flags & CLOSE_STATE) {
2725                 list_for_each_entry(local, &close_lru, so_close_lru) {
2726                         if (local->so_id == st_id)
2727                                 return local;
2728                 }
2729         }
2730         return NULL;
2731 }
2732
2733 static inline int
2734 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2735 {
2736         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2737 }
2738
2739 static int
2740 STALE_STATEID(stateid_t *stateid)
2741 {
2742         if (time_after((unsigned long)boot_time,
2743                         (unsigned long)stateid->si_boot)) {
2744                 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2745                         stateid->si_boot, stateid->si_stateownerid,
2746                         stateid->si_fileid, stateid->si_generation);
2747                 return 1;
2748         }
2749         return 0;
2750 }
2751
2752 static int
2753 EXPIRED_STATEID(stateid_t *stateid)
2754 {
2755         if (time_before((unsigned long)boot_time,
2756                         ((unsigned long)stateid->si_boot)) &&
2757             time_before((stateid->si_boot + lease_time), get_seconds())) {
2758                 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2759                         stateid->si_boot, stateid->si_stateownerid,
2760                         stateid->si_fileid, stateid->si_generation);
2761                 return 1;
2762         }
2763         return 0;
2764 }
2765
2766 static __be32
2767 stateid_error_map(stateid_t *stateid)
2768 {
2769         if (STALE_STATEID(stateid))
2770                 return nfserr_stale_stateid;
2771         if (EXPIRED_STATEID(stateid))
2772                 return nfserr_expired;
2773
2774         dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2775                 stateid->si_boot, stateid->si_stateownerid,
2776                 stateid->si_fileid, stateid->si_generation);
2777         return nfserr_bad_stateid;
2778 }
2779
2780 static inline int
2781 access_permit_read(unsigned long access_bmap)
2782 {
2783         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2784                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2785                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2786 }
2787
2788 static inline int
2789 access_permit_write(unsigned long access_bmap)
2790 {
2791         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2792                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2793 }
2794
2795 static
2796 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2797 {
2798         __be32 status = nfserr_openmode;
2799
2800         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2801                 goto out;
2802         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2803                 goto out;
2804         status = nfs_ok;
2805 out:
2806         return status;
2807 }
2808
2809 static inline __be32
2810 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2811 {
2812         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2813                 return nfs_ok;
2814         else if (locks_in_grace()) {
2815                 /* Answer in remaining cases depends on existance of
2816                  * conflicting state; so we must wait out the grace period. */
2817                 return nfserr_grace;
2818         } else if (flags & WR_STATE)
2819                 return nfs4_share_conflict(current_fh,
2820                                 NFS4_SHARE_DENY_WRITE);
2821         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2822                 return nfs4_share_conflict(current_fh,
2823                                 NFS4_SHARE_DENY_READ);
2824 }
2825
2826 /*
2827  * Allow READ/WRITE during grace period on recovered state only for files
2828  * that are not able to provide mandatory locking.
2829  */
2830 static inline int
2831 grace_disallows_io(struct inode *inode)
2832 {
2833         return locks_in_grace() && mandatory_lock(inode);
2834 }
2835
2836 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2837 {
2838         /*
2839          * When sessions are used the stateid generation number is ignored
2840          * when it is zero.
2841          */
2842         if ((flags & HAS_SESSION) && in->si_generation == 0)
2843                 goto out;
2844
2845         /* If the client sends us a stateid from the future, it's buggy: */
2846         if (in->si_generation > ref->si_generation)
2847                 return nfserr_bad_stateid;
2848         /*
2849          * The following, however, can happen.  For example, if the
2850          * client sends an open and some IO at the same time, the open
2851          * may bump si_generation while the IO is still in flight.
2852          * Thanks to hard links and renames, the client never knows what
2853          * file an open will affect.  So it could avoid that situation
2854          * only by serializing all opens and IO from the same open
2855          * owner.  To recover from the old_stateid error, the client
2856          * will just have to retry the IO:
2857          */
2858         if (in->si_generation < ref->si_generation)
2859                 return nfserr_old_stateid;
2860 out:
2861         return nfs_ok;
2862 }
2863
2864 static int is_delegation_stateid(stateid_t *stateid)
2865 {
2866         return stateid->si_fileid == 0;
2867 }
2868
2869 /*
2870 * Checks for stateid operations
2871 */
2872 __be32
2873 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2874                            stateid_t *stateid, int flags, struct file **filpp)
2875 {
2876         struct nfs4_stateid *stp = NULL;
2877         struct nfs4_delegation *dp = NULL;
2878         struct svc_fh *current_fh = &cstate->current_fh;
2879         struct inode *ino = current_fh->fh_dentry->d_inode;
2880         __be32 status;
2881
2882         if (filpp)
2883                 *filpp = NULL;
2884
2885         if (grace_disallows_io(ino))
2886                 return nfserr_grace;
2887
2888         if (nfsd4_has_session(cstate))
2889                 flags |= HAS_SESSION;
2890
2891         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2892                 return check_special_stateids(current_fh, stateid, flags);
2893
2894         status = nfserr_stale_stateid;
2895         if (STALE_STATEID(stateid)) 
2896                 goto out;
2897
2898         status = nfserr_bad_stateid;
2899         if (is_delegation_stateid(stateid)) {
2900                 dp = find_delegation_stateid(ino, stateid);
2901                 if (!dp) {
2902                         status = stateid_error_map(stateid);
2903                         goto out;
2904                 }
2905                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2906                                                   flags);
2907                 if (status)
2908                         goto out;
2909                 status = nfs4_check_delegmode(dp, flags);
2910                 if (status)
2911                         goto out;
2912                 renew_client(dp->dl_client);
2913                 if (filpp)
2914                         *filpp = dp->dl_vfs_file;
2915         } else { /* open or lock stateid */
2916                 stp = find_stateid(stateid, flags);
2917                 if (!stp) {
2918                         status = stateid_error_map(stateid);
2919                         goto out;
2920                 }
2921                 if (nfs4_check_fh(current_fh, stp))
2922                         goto out;
2923                 if (!stp->st_stateowner->so_confirmed)
2924                         goto out;
2925                 status = check_stateid_generation(stateid, &stp->st_stateid,
2926                                                   flags);
2927                 if (status)
2928                         goto out;
2929                 status = nfs4_check_openmode(stp, flags);
2930                 if (status)
2931                         goto out;
2932                 renew_client(stp->st_stateowner->so_client);
2933                 if (filpp)
2934                         *filpp = stp->st_vfs_file;
2935         }
2936         status = nfs_ok;
2937 out:
2938         return status;
2939 }
2940
2941 static inline int
2942 setlkflg (int type)
2943 {
2944         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2945                 RD_STATE : WR_STATE;
2946 }
2947
2948 /* 
2949  * Checks for sequence id mutating operations. 
2950  */
2951 static __be32
2952 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2953                          stateid_t *stateid, int flags,
2954                          struct nfs4_stateowner **sopp,
2955                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2956 {
2957         struct nfs4_stateid *stp;
2958         struct nfs4_stateowner *sop;
2959         struct svc_fh *current_fh = &cstate->current_fh;
2960         __be32 status;
2961
2962         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2963                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2964                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2965                 stateid->si_generation);
2966
2967         *stpp = NULL;
2968         *sopp = NULL;
2969
2970         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2971                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2972                 return nfserr_bad_stateid;
2973         }
2974
2975         if (STALE_STATEID(stateid))
2976                 return nfserr_stale_stateid;
2977
2978         if (nfsd4_has_session(cstate))
2979                 flags |= HAS_SESSION;
2980
2981         /*
2982         * We return BAD_STATEID if filehandle doesn't match stateid, 
2983         * the confirmed flag is incorrecly set, or the generation 
2984         * number is incorrect.  
2985         */
2986         stp = find_stateid(stateid, flags);
2987         if (stp == NULL) {
2988                 /*
2989                  * Also, we should make sure this isn't just the result of
2990                  * a replayed close:
2991                  */
2992                 sop = search_close_lru(stateid->si_stateownerid, flags);
2993                 if (sop == NULL)
2994                         return stateid_error_map(stateid);
2995                 *sopp = sop;
2996                 goto check_replay;
2997         }
2998
2999         *stpp = stp;
3000         *sopp = sop = stp->st_stateowner;
3001
3002         if (lock) {
3003                 clientid_t *lockclid = &lock->v.new.clientid;
3004                 struct nfs4_client *clp = sop->so_client;
3005                 int lkflg = 0;
3006                 __be32 status;
3007
3008                 lkflg = setlkflg(lock->lk_type);
3009
3010                 if (lock->lk_is_new) {
3011                         if (!sop->so_is_open_owner)
3012                                 return nfserr_bad_stateid;
3013                         if (!(flags & HAS_SESSION) &&
3014                             !same_clid(&clp->cl_clientid, lockclid))
3015                                 return nfserr_bad_stateid;
3016                         /* stp is the open stateid */
3017                         status = nfs4_check_openmode(stp, lkflg);
3018                         if (status)
3019                                 return status;
3020                 } else {
3021                         /* stp is the lock stateid */
3022                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3023                         if (status)
3024                                 return status;
3025                }
3026         }
3027
3028         if (nfs4_check_fh(current_fh, stp)) {
3029                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3030                 return nfserr_bad_stateid;
3031         }
3032
3033         /*
3034         *  We now validate the seqid and stateid generation numbers.
3035         *  For the moment, we ignore the possibility of 
3036         *  generation number wraparound.
3037         */
3038         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3039                 goto check_replay;
3040
3041         if (sop->so_confirmed && flags & CONFIRM) {
3042                 dprintk("NFSD: preprocess_seqid_op: expected"
3043                                 " unconfirmed stateowner!\n");
3044                 return nfserr_bad_stateid;
3045         }
3046         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3047                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3048                                 " confirmed yet!\n");
3049                 return nfserr_bad_stateid;
3050         }
3051         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3052         if (status)
3053                 return status;
3054         renew_client(sop->so_client);
3055         return nfs_ok;
3056
3057 check_replay:
3058         if (seqid == sop->so_seqid - 1) {
3059                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3060                 /* indicate replay to calling function */
3061                 return nfserr_replay_me;
3062         }
3063         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3064                         sop->so_seqid, seqid);
3065         *sopp = NULL;
3066         return nfserr_bad_seqid;
3067 }
3068
3069 __be32
3070 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3071                    struct nfsd4_open_confirm *oc)
3072 {
3073         __be32 status;
3074         struct nfs4_stateowner *sop;
3075         struct nfs4_stateid *stp;
3076
3077         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3078                         (int)cstate->current_fh.fh_dentry->d_name.len,
3079                         cstate->current_fh.fh_dentry->d_name.name);
3080
3081         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3082         if (status)
3083                 return status;
3084
3085         nfs4_lock_state();
3086
3087         if ((status = nfs4_preprocess_seqid_op(cstate,
3088                                         oc->oc_seqid, &oc->oc_req_stateid,
3089                                         CONFIRM | OPEN_STATE,
3090                                         &oc->oc_stateowner, &stp, NULL)))
3091                 goto out; 
3092
3093         sop = oc->oc_stateowner;
3094         sop->so_confirmed = 1;
3095         update_stateid(&stp->st_stateid);
3096         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3097         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
3098                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3099                          stp->st_stateid.si_boot,
3100                          stp->st_stateid.si_stateownerid,
3101                          stp->st_stateid.si_fileid,
3102                          stp->st_stateid.si_generation);
3103
3104         nfsd4_create_clid_dir(sop->so_client);
3105 out:
3106         if (oc->oc_stateowner) {
3107                 nfs4_get_stateowner(oc->oc_stateowner);
3108                 cstate->replay_owner = oc->oc_stateowner;
3109         }
3110         nfs4_unlock_state();
3111         return status;
3112 }
3113
3114
3115 /*
3116  * unset all bits in union bitmap (bmap) that
3117  * do not exist in share (from successful OPEN_DOWNGRADE)
3118  */
3119 static void
3120 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3121 {
3122         int i;
3123         for (i = 1; i < 4; i++) {
3124                 if ((i & access) != i)
3125                         __clear_bit(i, bmap);
3126         }
3127 }
3128
3129 static void
3130 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3131 {
3132         int i;
3133         for (i = 0; i < 4; i++) {
3134                 if ((i & deny) != i)
3135                         __clear_bit(i, bmap);
3136         }
3137 }
3138
3139 __be32
3140 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3141                      struct nfsd4_compound_state *cstate,
3142                      struct nfsd4_open_downgrade *od)
3143 {
3144         __be32 status;
3145         struct nfs4_stateid *stp;
3146         unsigned int share_access;
3147
3148         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3149                         (int)cstate->current_fh.fh_dentry->d_name.len,
3150                         cstate->current_fh.fh_dentry->d_name.name);
3151
3152         if (!access_valid(od->od_share_access, cstate->minorversion)
3153                         || !deny_valid(od->od_share_deny))
3154                 return nfserr_inval;
3155
3156         nfs4_lock_state();
3157         if ((status = nfs4_preprocess_seqid_op(cstate,
3158                                         od->od_seqid,
3159                                         &od->od_stateid, 
3160                                         OPEN_STATE,
3161                                         &od->od_stateowner, &stp, NULL)))
3162                 goto out; 
3163
3164         status = nfserr_inval;
3165         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3166                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3167                         stp->st_access_bmap, od->od_share_access);
3168                 goto out;
3169         }
3170         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3171                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3172                         stp->st_deny_bmap, od->od_share_deny);
3173                 goto out;
3174         }
3175         set_access(&share_access, stp->st_access_bmap);
3176         nfs4_file_downgrade(stp->st_vfs_file,
3177                             share_access & ~od->od_share_access);
3178
3179         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3180         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3181
3182         update_stateid(&stp->st_stateid);
3183         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3184         status = nfs_ok;
3185 out:
3186         if (od->od_stateowner) {
3187                 nfs4_get_stateowner(od->od_stateowner);
3188                 cstate->replay_owner = od->od_stateowner;
3189         }
3190         nfs4_unlock_state();
3191         return status;
3192 }
3193
3194 /*
3195  * nfs4_unlock_state() called after encode
3196  */
3197 __be32
3198 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3199             struct nfsd4_close *close)
3200 {
3201         __be32 status;
3202         struct nfs4_stateid *stp;
3203
3204         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3205                         (int)cstate->current_fh.fh_dentry->d_name.len,
3206                         cstate->current_fh.fh_dentry->d_name.name);
3207
3208         nfs4_lock_state();
3209         /* check close_lru for replay */
3210         if ((status = nfs4_preprocess_seqid_op(cstate,
3211                                         close->cl_seqid,
3212                                         &close->cl_stateid, 
3213                                         OPEN_STATE | CLOSE_STATE,
3214                                         &close->cl_stateowner, &stp, NULL)))
3215                 goto out; 
3216         status = nfs_ok;
3217         update_stateid(&stp->st_stateid);
3218         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3219
3220         /* release_stateid() calls nfsd_close() if needed */
3221         release_open_stateid(stp);
3222
3223         /* place unused nfs4_stateowners on so_close_lru list to be
3224          * released by the laundromat service after the lease period
3225          * to enable us to handle CLOSE replay
3226          */
3227         if (list_empty(&close->cl_stateowner->so_stateids))
3228                 move_to_close_lru(close->cl_stateowner);
3229 out:
3230         if (close->cl_stateowner) {
3231                 nfs4_get_stateowner(close->cl_stateowner);
3232                 cstate->replay_owner = close->cl_stateowner;
3233         }
3234         nfs4_unlock_state();
3235         return status;
3236 }
3237
3238 __be32
3239 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3240                   struct nfsd4_delegreturn *dr)
3241 {
3242         struct nfs4_delegation *dp;
3243         stateid_t *stateid = &dr->dr_stateid;
3244         struct inode *inode;
3245         __be32 status;
3246         int flags = 0;
3247
3248         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3249                 return status;
3250         inode = cstate->current_fh.fh_dentry->d_inode;
3251
3252         if (nfsd4_has_session(cstate))
3253                 flags |= HAS_SESSION;
3254         nfs4_lock_state();
3255         status = nfserr_bad_stateid;
3256         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3257                 goto out;
3258         status = nfserr_stale_stateid;
3259         if (STALE_STATEID(stateid))
3260                 goto out;
3261         status = nfserr_bad_stateid;
3262         if (!is_delegation_stateid(stateid))
3263                 goto out;
3264         dp = find_delegation_stateid(inode, stateid);
3265         if (!dp) {
3266                 status = stateid_error_map(stateid);
3267                 goto out;
3268         }
3269         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3270         if (status)
3271                 goto out;
3272         renew_client(dp->dl_client);
3273
3274         unhash_delegation(dp);
3275 out:
3276         nfs4_unlock_state();
3277
3278         return status;
3279 }
3280
3281
3282 /* 
3283  * Lock owner state (byte-range locks)
3284  */
3285 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3286 #define LOCK_HASH_BITS              8
3287 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3288 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3289
3290 static inline u64
3291 end_offset(u64 start, u64 len)
3292 {
3293         u64 end;
3294
3295         end = start + len;
3296         return end >= start ? end: NFS4_MAX_UINT64;
3297 }
3298
3299 /* last octet in a range */
3300 static inline u64
3301 last_byte_offset(u64 start, u64 len)
3302 {
3303         u64 end;
3304
3305         BUG_ON(!len);
3306         end = start + len;
3307         return end > start ? end - 1: NFS4_MAX_UINT64;
3308 }
3309
3310 #define lockownerid_hashval(id) \
3311         ((id) & LOCK_HASH_MASK)
3312
3313 static inline unsigned int
3314 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3315                 struct xdr_netobj *ownername)
3316 {
3317         return (file_hashval(inode) + cl_id
3318                         + opaque_hashval(ownername->data, ownername->len))
3319                 & LOCK_HASH_MASK;
3320 }
3321
3322 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3323 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3324 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3325
3326 static struct nfs4_stateid *
3327 find_stateid(stateid_t *stid, int flags)
3328 {
3329         struct nfs4_stateid *local;
3330         u32 st_id = stid->si_stateownerid;
3331         u32 f_id = stid->si_fileid;
3332         unsigned int hashval;
3333
3334         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3335         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3336                 hashval = stateid_hashval(st_id, f_id);
3337                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3338                         if ((local->st_stateid.si_stateownerid == st_id) &&
3339                             (local->st_stateid.si_fileid == f_id))
3340                                 return local;
3341                 }
3342         } 
3343
3344         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3345                 hashval = stateid_hashval(st_id, f_id);
3346                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3347                         if ((local->st_stateid.si_stateownerid == st_id) &&
3348                             (local->st_stateid.si_fileid == f_id))
3349                                 return local;
3350                 }
3351         }
3352         return NULL;
3353 }
3354
3355 static struct nfs4_delegation *
3356 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3357 {
3358         struct nfs4_file *fp;
3359         struct nfs4_delegation *dl;
3360
3361         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3362                     stid->si_boot, stid->si_stateownerid,
3363                     stid->si_fileid, stid->si_generation);
3364
3365         fp = find_file(ino);
3366         if (!fp)
3367                 return NULL;
3368         dl = find_delegation_file(fp, stid);
3369         put_nfs4_file(fp);
3370         return dl;
3371 }
3372
3373 /*
3374  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3375  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3376  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3377  * locking, this prevents us from being completely protocol-compliant.  The
3378  * real solution to this problem is to start using unsigned file offsets in
3379  * the VFS, but this is a very deep change!
3380  */
3381 static inline void
3382 nfs4_transform_lock_offset(struct file_lock *lock)
3383 {
3384         if (lock->fl_start < 0)
3385                 lock->fl_start = OFFSET_MAX;
3386         if (lock->fl_end < 0)
3387                 lock->fl_end = OFFSET_MAX;
3388 }
3389
3390 /* Hack!: For now, we're defining this just so we can use a pointer to it
3391  * as a unique cookie to identify our (NFSv4's) posix locks. */
3392 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3393 };
3394
3395 static inline void
3396 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3397 {
3398         struct nfs4_stateowner *sop;
3399         unsigned int hval;
3400
3401         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3402                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3403                 hval = lockownerid_hashval(sop->so_id);
3404                 kref_get(&sop->so_ref);
3405                 deny->ld_sop = sop;
3406                 deny->ld_clientid = sop->so_client->cl_clientid;
3407         } else {
3408                 deny->ld_sop = NULL;
3409                 deny->ld_clientid.cl_boot = 0;
3410                 deny->ld_clientid.cl_id = 0;
3411         }
3412         deny->ld_start = fl->fl_start;
3413         deny->ld_length = NFS4_MAX_UINT64;
3414         if (fl->fl_end != NFS4_MAX_UINT64)
3415                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3416         deny->ld_type = NFS4_READ_LT;
3417         if (fl->fl_type != F_RDLCK)
3418                 deny->ld_type = NFS4_WRITE_LT;
3419 }
3420
3421 static struct nfs4_stateowner *
3422 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3423                 struct xdr_netobj *owner)
3424 {
3425         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3426         struct nfs4_stateowner *op;
3427
3428         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3429                 if (same_owner_str(op, owner, clid))
3430                         return op;
3431         }
3432         return NULL;
3433 }
3434
3435 /*
3436  * Alloc a lock owner structure.
3437  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3438  * occured. 
3439  *
3440  * strhashval = lock_ownerstr_hashval 
3441  */
3442
3443 static struct nfs4_stateowner *
3444 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3445         struct nfs4_stateowner *sop;
3446         struct nfs4_replay *rp;
3447         unsigned int idhashval;
3448
3449         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3450                 return NULL;
3451         idhashval = lockownerid_hashval(current_ownerid);
3452         INIT_LIST_HEAD(&sop->so_idhash);
3453         INIT_LIST_HEAD(&sop->so_strhash);
3454         INIT_LIST_HEAD(&sop->so_perclient);
3455         INIT_LIST_HEAD(&sop->so_stateids);
3456         INIT_LIST_HEAD(&sop->so_perstateid);
3457         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3458         sop->so_time = 0;
3459         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3460         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3461         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3462         sop->so_is_open_owner = 0;
3463         sop->so_id = current_ownerid++;
3464         sop->so_client = clp;
3465         /* It is the openowner seqid that will be incremented in encode in the
3466          * case of new lockowners; so increment the lock seqid manually: */
3467         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3468         sop->so_confirmed = 1;
3469         rp = &sop->so_replay;
3470         rp->rp_status = nfserr_serverfault;
3471         rp->rp_buflen = 0;
3472         rp->rp_buf = rp->rp_ibuf;
3473         return sop;
3474 }
3475
3476 static struct nfs4_stateid *
3477 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3478 {
3479         struct nfs4_stateid *stp;
3480         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3481
3482         stp = nfs4_alloc_stateid();
3483         if (stp == NULL)
3484                 goto out;
3485         INIT_LIST_HEAD(&stp->st_hash);
3486         INIT_LIST_HEAD(&stp->st_perfile);