3e5345e01b13b92514a91f85091bcd9f06006ced
[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_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
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_ident = cb->cb_ident;
207         dp->dl_trunc = 0;
208         dp->dl_stateid.si_boot = get_seconds();
209         dp->dl_stateid.si_stateownerid = current_delegid++;
210         dp->dl_stateid.si_fileid = 0;
211         dp->dl_stateid.si_generation = 0;
212         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
213         dp->dl_time = 0;
214         atomic_set(&dp->dl_count, 1);
215         list_add(&dp->dl_perfile, &fp->fi_delegations);
216         list_add(&dp->dl_perclnt, &clp->cl_delegations);
217         return dp;
218 }
219
220 void
221 nfs4_put_delegation(struct nfs4_delegation *dp)
222 {
223         if (atomic_dec_and_test(&dp->dl_count)) {
224                 dprintk("NFSD: freeing dp %p\n",dp);
225                 put_nfs4_file(dp->dl_file);
226                 kmem_cache_free(deleg_slab, dp);
227                 num_delegations--;
228         }
229 }
230
231 /* Remove the associated file_lock first, then remove the delegation.
232  * lease_modify() is called to remove the FS_LEASE file_lock from
233  * the i_flock list, eventually calling nfsd's lock_manager
234  * fl_release_callback.
235  */
236 static void
237 nfs4_close_delegation(struct nfs4_delegation *dp)
238 {
239         struct file *filp = dp->dl_vfs_file;
240
241         dprintk("NFSD: close_delegation dp %p\n",dp);
242         dp->dl_vfs_file = NULL;
243         /* The following nfsd_close may not actually close the file,
244          * but we want to remove the lease in any case. */
245         if (dp->dl_flock)
246                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
247         nfsd_close(filp);
248 }
249
250 /* Called under the state lock. */
251 static void
252 unhash_delegation(struct nfs4_delegation *dp)
253 {
254         list_del_init(&dp->dl_perfile);
255         list_del_init(&dp->dl_perclnt);
256         spin_lock(&recall_lock);
257         list_del_init(&dp->dl_recall_lru);
258         spin_unlock(&recall_lock);
259         nfs4_close_delegation(dp);
260         nfs4_put_delegation(dp);
261 }
262
263 /* 
264  * SETCLIENTID state 
265  */
266
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS                 4
269 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
271
272 #define clientid_hashval(id) \
273         ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
276 /*
277  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278  * used in reboot/reset lease grace period processing
279  *
280  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281  * setclientid_confirmed info. 
282  *
283  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
284  * setclientid info.
285  *
286  * client_lru holds client queue ordered by nfs4_client.cl_time
287  * for lease renewal.
288  *
289  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290  * for last close replay.
291  */
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
300
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
302 {
303         list_del(&stp->st_hash);
304         list_del(&stp->st_perfile);
305         list_del(&stp->st_perstateowner);
306 }
307
308 static void free_generic_stateid(struct nfs4_stateid *stp)
309 {
310         put_nfs4_file(stp->st_file);
311         kmem_cache_free(stateid_slab, stp);
312 }
313
314 static void release_lock_stateid(struct nfs4_stateid *stp)
315 {
316         unhash_generic_stateid(stp);
317         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318         free_generic_stateid(stp);
319 }
320
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
322 {
323         struct nfs4_stateid *stp;
324
325         list_del(&sop->so_idhash);
326         list_del(&sop->so_strhash);
327         list_del(&sop->so_perstateid);
328         while (!list_empty(&sop->so_stateids)) {
329                 stp = list_first_entry(&sop->so_stateids,
330                                 struct nfs4_stateid, st_perstateowner);
331                 release_lock_stateid(stp);
332         }
333 }
334
335 static void release_lockowner(struct nfs4_stateowner *sop)
336 {
337         unhash_lockowner(sop);
338         nfs4_put_stateowner(sop);
339 }
340
341 static void
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
343 {
344         struct nfs4_stateowner *lock_sop;
345
346         while (!list_empty(&open_stp->st_lockowners)) {
347                 lock_sop = list_entry(open_stp->st_lockowners.next,
348                                 struct nfs4_stateowner, so_perstateid);
349                 /* list_del(&open_stp->st_lockowners);  */
350                 BUG_ON(lock_sop->so_is_open_owner);
351                 release_lockowner(lock_sop);
352         }
353 }
354
355 static void release_open_stateid(struct nfs4_stateid *stp)
356 {
357         unhash_generic_stateid(stp);
358         release_stateid_lockowners(stp);
359         nfsd_close(stp->st_vfs_file);
360         free_generic_stateid(stp);
361 }
362
363 static void unhash_openowner(struct nfs4_stateowner *sop)
364 {
365         struct nfs4_stateid *stp;
366
367         list_del(&sop->so_idhash);
368         list_del(&sop->so_strhash);
369         list_del(&sop->so_perclient);
370         list_del(&sop->so_perstateid); /* XXX: necessary? */
371         while (!list_empty(&sop->so_stateids)) {
372                 stp = list_first_entry(&sop->so_stateids,
373                                 struct nfs4_stateid, st_perstateowner);
374                 release_open_stateid(stp);
375         }
376 }
377
378 static void release_openowner(struct nfs4_stateowner *sop)
379 {
380         unhash_openowner(sop);
381         list_del(&sop->so_close_lru);
382         nfs4_put_stateowner(sop);
383 }
384
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE       512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
388
389 static inline int
390 hash_sessionid(struct nfs4_sessionid *sessionid)
391 {
392         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
393
394         return sid->sequence % SESSION_HASH_SIZE;
395 }
396
397 static inline void
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
399 {
400         u32 *ptr = (u32 *)(&sessionid->data[0]);
401         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
402 }
403
404 static void
405 gen_sessionid(struct nfsd4_session *ses)
406 {
407         struct nfs4_client *clp = ses->se_client;
408         struct nfsd4_sessionid *sid;
409
410         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411         sid->clientid = clp->cl_clientid;
412         sid->sequence = current_sessionid++;
413         sid->reserved = 0;
414 }
415
416 /*
417  * Give the client the number of slots it requests bound by
418  * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
419  *
420  * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
421  * should (up to a point) re-negotiate active sessions and reduce their
422  * slot usage to make rooom for new connections. For now we just fail the
423  * create session.
424  */
425 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
426 {
427         int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
428
429         spin_lock(&nfsd_serv->sv_lock);
430         if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
431                 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
432         nfsd_serv->sv_drc_pages_used += np;
433         spin_unlock(&nfsd_serv->sv_lock);
434
435         if (np <= 0) {
436                 status = nfserr_resource;
437                 fchan->maxreqs = 0;
438         } else
439                 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
440
441         return status;
442 }
443
444 /*
445  * fchan holds the client values on input, and the server values on output
446  */
447 static int init_forechannel_attrs(struct svc_rqst *rqstp,
448                                     struct nfsd4_session *session,
449                                     struct nfsd4_channel_attrs *fchan)
450 {
451         int status = 0;
452         __u32   maxcount = svc_max_payload(rqstp);
453
454         /* headerpadsz set to zero in encode routine */
455
456         /* Use the client's max request and max response size if possible */
457         if (fchan->maxreq_sz > maxcount)
458                 fchan->maxreq_sz = maxcount;
459         session->se_fmaxreq_sz = fchan->maxreq_sz;
460
461         if (fchan->maxresp_sz > maxcount)
462                 fchan->maxresp_sz = maxcount;
463         session->se_fmaxresp_sz = fchan->maxresp_sz;
464
465         /* Set the max response cached size our default which is
466          * a multiple of PAGE_SIZE and small */
467         session->se_fmaxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
468         fchan->maxresp_cached = session->se_fmaxresp_cached;
469
470         /* Use the client's maxops if possible */
471         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
472                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
473         session->se_fmaxops = fchan->maxops;
474
475         /* try to use the client requested number of slots */
476         if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
477                 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
478
479         /* FIXME: Error means no more DRC pages so the server should
480          * recover pages from existing sessions. For now fail session
481          * creation.
482          */
483         status = set_forechannel_maxreqs(fchan);
484
485         session->se_fnumslots = fchan->maxreqs;
486         return status;
487 }
488
489 static int
490 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
491                    struct nfsd4_create_session *cses)
492 {
493         struct nfsd4_session *new, tmp;
494         int idx, status = nfserr_resource, slotsize;
495
496         memset(&tmp, 0, sizeof(tmp));
497
498         /* FIXME: For now, we just accept the client back channel attributes. */
499         status = init_forechannel_attrs(rqstp, &tmp, &cses->fore_channel);
500         if (status)
501                 goto out;
502
503         /* allocate struct nfsd4_session and slot table in one piece */
504         slotsize = tmp.se_fnumslots * sizeof(struct nfsd4_slot);
505         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
506         if (!new)
507                 goto out;
508
509         memcpy(new, &tmp, sizeof(*new));
510
511         new->se_client = clp;
512         gen_sessionid(new);
513         idx = hash_sessionid(&new->se_sessionid);
514         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
515                NFS4_MAX_SESSIONID_LEN);
516
517         new->se_flags = cses->flags;
518         kref_init(&new->se_ref);
519         spin_lock(&sessionid_lock);
520         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
521         list_add(&new->se_perclnt, &clp->cl_sessions);
522         spin_unlock(&sessionid_lock);
523
524         status = nfs_ok;
525 out:
526         return status;
527 }
528
529 /* caller must hold sessionid_lock */
530 static struct nfsd4_session *
531 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
532 {
533         struct nfsd4_session *elem;
534         int idx;
535
536         dump_sessionid(__func__, sessionid);
537         idx = hash_sessionid(sessionid);
538         dprintk("%s: idx is %d\n", __func__, idx);
539         /* Search in the appropriate list */
540         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
541                 dump_sessionid("list traversal", &elem->se_sessionid);
542                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
543                             NFS4_MAX_SESSIONID_LEN)) {
544                         return elem;
545                 }
546         }
547
548         dprintk("%s: session not found\n", __func__);
549         return NULL;
550 }
551
552 /* caller must hold sessionid_lock */
553 static void
554 unhash_session(struct nfsd4_session *ses)
555 {
556         list_del(&ses->se_hash);
557         list_del(&ses->se_perclnt);
558 }
559
560 static void
561 release_session(struct nfsd4_session *ses)
562 {
563         spin_lock(&sessionid_lock);
564         unhash_session(ses);
565         spin_unlock(&sessionid_lock);
566         nfsd4_put_session(ses);
567 }
568
569 static void nfsd4_release_respages(struct page **respages, short resused);
570
571 void
572 free_session(struct kref *kref)
573 {
574         struct nfsd4_session *ses;
575         int i;
576
577         ses = container_of(kref, struct nfsd4_session, se_ref);
578         for (i = 0; i < ses->se_fnumslots; i++) {
579                 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
580                 nfsd4_release_respages(e->ce_respages, e->ce_resused);
581         }
582         kfree(ses->se_slots);
583         kfree(ses);
584 }
585
586 static inline void
587 renew_client(struct nfs4_client *clp)
588 {
589         /*
590         * Move client to the end to the LRU list.
591         */
592         dprintk("renewing client (clientid %08x/%08x)\n", 
593                         clp->cl_clientid.cl_boot, 
594                         clp->cl_clientid.cl_id);
595         list_move_tail(&clp->cl_lru, &client_lru);
596         clp->cl_time = get_seconds();
597 }
598
599 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
600 static int
601 STALE_CLIENTID(clientid_t *clid)
602 {
603         if (clid->cl_boot == boot_time)
604                 return 0;
605         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
606                 clid->cl_boot, clid->cl_id, boot_time);
607         return 1;
608 }
609
610 /* 
611  * XXX Should we use a slab cache ?
612  * This type of memory management is somewhat inefficient, but we use it
613  * anyway since SETCLIENTID is not a common operation.
614  */
615 static struct nfs4_client *alloc_client(struct xdr_netobj name)
616 {
617         struct nfs4_client *clp;
618
619         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
620         if (clp == NULL)
621                 return NULL;
622         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
623         if (clp->cl_name.data == NULL) {
624                 kfree(clp);
625                 return NULL;
626         }
627         memcpy(clp->cl_name.data, name.data, name.len);
628         clp->cl_name.len = name.len;
629         return clp;
630 }
631
632 static void
633 shutdown_callback_client(struct nfs4_client *clp)
634 {
635         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
636
637         if (clnt) {
638                 /*
639                  * Callback threads take a reference on the client, so there
640                  * should be no outstanding callbacks at this point.
641                  */
642                 clp->cl_cb_conn.cb_client = NULL;
643                 rpc_shutdown_client(clnt);
644         }
645         if (clp->cl_cb_conn.cb_cred) {
646                 put_rpccred(clp->cl_cb_conn.cb_cred);
647                 clp->cl_cb_conn.cb_cred = NULL;
648         }
649 }
650
651 static inline void
652 free_client(struct nfs4_client *clp)
653 {
654         shutdown_callback_client(clp);
655         nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
656                              clp->cl_slot.sl_cache_entry.ce_resused);
657         if (clp->cl_cred.cr_group_info)
658                 put_group_info(clp->cl_cred.cr_group_info);
659         kfree(clp->cl_principal);
660         kfree(clp->cl_name.data);
661         kfree(clp);
662 }
663
664 void
665 put_nfs4_client(struct nfs4_client *clp)
666 {
667         if (atomic_dec_and_test(&clp->cl_count))
668                 free_client(clp);
669 }
670
671 static void
672 expire_client(struct nfs4_client *clp)
673 {
674         struct nfs4_stateowner *sop;
675         struct nfs4_delegation *dp;
676         struct list_head reaplist;
677
678         dprintk("NFSD: expire_client cl_count %d\n",
679                             atomic_read(&clp->cl_count));
680
681         INIT_LIST_HEAD(&reaplist);
682         spin_lock(&recall_lock);
683         while (!list_empty(&clp->cl_delegations)) {
684                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
685                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
686                                 dp->dl_flock);
687                 list_del_init(&dp->dl_perclnt);
688                 list_move(&dp->dl_recall_lru, &reaplist);
689         }
690         spin_unlock(&recall_lock);
691         while (!list_empty(&reaplist)) {
692                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
693                 list_del_init(&dp->dl_recall_lru);
694                 unhash_delegation(dp);
695         }
696         list_del(&clp->cl_idhash);
697         list_del(&clp->cl_strhash);
698         list_del(&clp->cl_lru);
699         while (!list_empty(&clp->cl_openowners)) {
700                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
701                 release_openowner(sop);
702         }
703         while (!list_empty(&clp->cl_sessions)) {
704                 struct nfsd4_session  *ses;
705                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
706                                  se_perclnt);
707                 release_session(ses);
708         }
709         put_nfs4_client(clp);
710 }
711
712 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
713 {
714         struct nfs4_client *clp;
715
716         clp = alloc_client(name);
717         if (clp == NULL)
718                 return NULL;
719         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
720         atomic_set(&clp->cl_count, 1);
721         atomic_set(&clp->cl_cb_conn.cb_set, 0);
722         INIT_LIST_HEAD(&clp->cl_idhash);
723         INIT_LIST_HEAD(&clp->cl_strhash);
724         INIT_LIST_HEAD(&clp->cl_openowners);
725         INIT_LIST_HEAD(&clp->cl_delegations);
726         INIT_LIST_HEAD(&clp->cl_sessions);
727         INIT_LIST_HEAD(&clp->cl_lru);
728         return clp;
729 }
730
731 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
732 {
733         memcpy(target->cl_verifier.data, source->data,
734                         sizeof(target->cl_verifier.data));
735 }
736
737 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
738 {
739         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
740         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
741 }
742
743 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
744 {
745         target->cr_uid = source->cr_uid;
746         target->cr_gid = source->cr_gid;
747         target->cr_group_info = source->cr_group_info;
748         get_group_info(target->cr_group_info);
749 }
750
751 static int same_name(const char *n1, const char *n2)
752 {
753         return 0 == memcmp(n1, n2, HEXDIR_LEN);
754 }
755
756 static int
757 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
758 {
759         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
760 }
761
762 static int
763 same_clid(clientid_t *cl1, clientid_t *cl2)
764 {
765         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
766 }
767
768 /* XXX what about NGROUP */
769 static int
770 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
771 {
772         return cr1->cr_uid == cr2->cr_uid;
773 }
774
775 static void gen_clid(struct nfs4_client *clp)
776 {
777         static u32 current_clientid = 1;
778
779         clp->cl_clientid.cl_boot = boot_time;
780         clp->cl_clientid.cl_id = current_clientid++; 
781 }
782
783 static void gen_confirm(struct nfs4_client *clp)
784 {
785         static u32 i;
786         u32 *p;
787
788         p = (u32 *)clp->cl_confirm.data;
789         *p++ = get_seconds();
790         *p++ = i++;
791 }
792
793 static int check_name(struct xdr_netobj name)
794 {
795         if (name.len == 0) 
796                 return 0;
797         if (name.len > NFS4_OPAQUE_LIMIT) {
798                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
799                 return 0;
800         }
801         return 1;
802 }
803
804 static void
805 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
806 {
807         unsigned int idhashval;
808
809         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
810         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
811         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
812         list_add_tail(&clp->cl_lru, &client_lru);
813         clp->cl_time = get_seconds();
814 }
815
816 static void
817 move_to_confirmed(struct nfs4_client *clp)
818 {
819         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
820         unsigned int strhashval;
821
822         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
823         list_del_init(&clp->cl_strhash);
824         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
825         strhashval = clientstr_hashval(clp->cl_recdir);
826         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
827         renew_client(clp);
828 }
829
830 static struct nfs4_client *
831 find_confirmed_client(clientid_t *clid)
832 {
833         struct nfs4_client *clp;
834         unsigned int idhashval = clientid_hashval(clid->cl_id);
835
836         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
837                 if (same_clid(&clp->cl_clientid, clid))
838                         return clp;
839         }
840         return NULL;
841 }
842
843 static struct nfs4_client *
844 find_unconfirmed_client(clientid_t *clid)
845 {
846         struct nfs4_client *clp;
847         unsigned int idhashval = clientid_hashval(clid->cl_id);
848
849         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
850                 if (same_clid(&clp->cl_clientid, clid))
851                         return clp;
852         }
853         return NULL;
854 }
855
856 /*
857  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
858  * parameter. Matching is based on the fact the at least one of the
859  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
860  *
861  * FIXME: we need to unify the clientid namespaces for nfsv4.x
862  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
863  * and SET_CLIENTID{,_CONFIRM}
864  */
865 static inline int
866 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
867 {
868         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
869         return use_exchange_id == has_exchange_flags;
870 }
871
872 static struct nfs4_client *
873 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
874                              bool use_exchange_id)
875 {
876         struct nfs4_client *clp;
877
878         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
879                 if (same_name(clp->cl_recdir, dname) &&
880                     match_clientid_establishment(clp, use_exchange_id))
881                         return clp;
882         }
883         return NULL;
884 }
885
886 static struct nfs4_client *
887 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
888                                bool use_exchange_id)
889 {
890         struct nfs4_client *clp;
891
892         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
893                 if (same_name(clp->cl_recdir, dname) &&
894                     match_clientid_establishment(clp, use_exchange_id))
895                         return clp;
896         }
897         return NULL;
898 }
899
900 /* a helper function for parse_callback */
901 static int
902 parse_octet(unsigned int *lenp, char **addrp)
903 {
904         unsigned int len = *lenp;
905         char *p = *addrp;
906         int n = -1;
907         char c;
908
909         for (;;) {
910                 if (!len)
911                         break;
912                 len--;
913                 c = *p++;
914                 if (c == '.')
915                         break;
916                 if ((c < '0') || (c > '9')) {
917                         n = -1;
918                         break;
919                 }
920                 if (n < 0)
921                         n = 0;
922                 n = (n * 10) + (c - '0');
923                 if (n > 255) {
924                         n = -1;
925                         break;
926                 }
927         }
928         *lenp = len;
929         *addrp = p;
930         return n;
931 }
932
933 /* parse and set the setclientid ipv4 callback address */
934 static int
935 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
936 {
937         int temp = 0;
938         u32 cbaddr = 0;
939         u16 cbport = 0;
940         u32 addrlen = addr_len;
941         char *addr = addr_val;
942         int i, shift;
943
944         /* ipaddress */
945         shift = 24;
946         for(i = 4; i > 0  ; i--) {
947                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
948                         return 0;
949                 }
950                 cbaddr |= (temp << shift);
951                 if (shift > 0)
952                 shift -= 8;
953         }
954         *cbaddrp = cbaddr;
955
956         /* port */
957         shift = 8;
958         for(i = 2; i > 0  ; i--) {
959                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
960                         return 0;
961                 }
962                 cbport |= (temp << shift);
963                 if (shift > 0)
964                         shift -= 8;
965         }
966         *cbportp = cbport;
967         return 1;
968 }
969
970 static void
971 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
972 {
973         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
974
975         /* Currently, we only support tcp for the callback channel */
976         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
977                 goto out_err;
978
979         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
980                          &cb->cb_addr, &cb->cb_port)))
981                 goto out_err;
982         cb->cb_prog = se->se_callback_prog;
983         cb->cb_ident = se->se_callback_ident;
984         return;
985 out_err:
986         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
987                 "will not receive delegations\n",
988                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
989
990         return;
991 }
992
993 void
994 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
995 {
996         struct nfsd4_compoundres *resp = rqstp->rq_resp;
997
998         resp->cstate.statp = statp;
999 }
1000
1001 /*
1002  * Dereference the result pages.
1003  */
1004 static void
1005 nfsd4_release_respages(struct page **respages, short resused)
1006 {
1007         int i;
1008
1009         dprintk("--> %s\n", __func__);
1010         for (i = 0; i < resused; i++) {
1011                 if (!respages[i])
1012                         continue;
1013                 put_page(respages[i]);
1014                 respages[i] = NULL;
1015         }
1016 }
1017
1018 static void
1019 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1020 {
1021         int i;
1022
1023         for (i = 0; i < count; i++) {
1024                 topages[i] = frompages[i];
1025                 if (!topages[i])
1026                         continue;
1027                 get_page(topages[i]);
1028         }
1029 }
1030
1031 /*
1032  * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1033  * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1034  * length of the XDR response is less than se_fmaxresp_cached
1035  * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1036  * of the reply (e.g. readdir).
1037  *
1038  * Store the base and length of the rq_req.head[0] page
1039  * of the NFSv4.1 data, just past the rpc header.
1040  */
1041 void
1042 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1043 {
1044         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1045         struct svc_rqst *rqstp = resp->rqstp;
1046         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1047         struct nfsd4_op *op = &args->ops[resp->opcnt];
1048         struct kvec *resv = &rqstp->rq_res.head[0];
1049
1050         dprintk("--> %s entry %p\n", __func__, entry);
1051
1052         /* Don't cache a failed OP_SEQUENCE. */
1053         if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1054                 return;
1055
1056         nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1057         entry->ce_opcnt = resp->opcnt;
1058         entry->ce_status = resp->cstate.status;
1059
1060         /*
1061          * Don't need a page to cache just the sequence operation - the slot
1062          * does this for us!
1063          */
1064
1065         if (nfsd4_not_cached(resp)) {
1066                 entry->ce_resused = 0;
1067                 entry->ce_rpchdrlen = 0;
1068                 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1069                         resp->cstate.slot->sl_cache_entry.ce_cachethis);
1070                 return;
1071         }
1072         entry->ce_resused = rqstp->rq_resused;
1073         if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1074                 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1075         nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1076                          entry->ce_resused);
1077         entry->ce_datav.iov_base = resp->cstate.statp;
1078         entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1079                                 (char *)page_address(rqstp->rq_respages[0]));
1080         /* Current request rpc header length*/
1081         entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1082                                 (char *)page_address(rqstp->rq_respages[0]);
1083 }
1084
1085 /*
1086  * We keep the rpc header, but take the nfs reply from the replycache.
1087  */
1088 static int
1089 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1090                         struct nfsd4_cache_entry *entry)
1091 {
1092         struct svc_rqst *rqstp = resp->rqstp;
1093         struct kvec *resv = &resp->rqstp->rq_res.head[0];
1094         int len;
1095
1096         /* Current request rpc header length*/
1097         len = (char *)resp->cstate.statp -
1098                         (char *)page_address(rqstp->rq_respages[0]);
1099         if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1100                 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1101                         entry->ce_datav.iov_len);
1102                 return 0;
1103         }
1104         /* copy the cached reply nfsd data past the current rpc header */
1105         memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1106                 entry->ce_datav.iov_len);
1107         resv->iov_len = len + entry->ce_datav.iov_len;
1108         return 1;
1109 }
1110
1111 /*
1112  * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1113  * cached page.  Replace any futher replay pages from the cache.
1114  */
1115 __be32
1116 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1117                          struct nfsd4_sequence *seq)
1118 {
1119         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1120         __be32 status;
1121
1122         dprintk("--> %s entry %p\n", __func__, entry);
1123
1124         /*
1125          * If this is just the sequence operation, we did not keep
1126          * a page in the cache entry because we can just use the
1127          * slot info stored in struct nfsd4_sequence that was checked
1128          * against the slot in nfsd4_sequence().
1129          *
1130          * This occurs when seq->cachethis is FALSE, or when the client
1131          * session inactivity timer fires and a solo sequence operation
1132          * is sent (lease renewal).
1133          */
1134         if (seq && nfsd4_not_cached(resp)) {
1135                 seq->maxslots = resp->cstate.session->se_fnumslots;
1136                 return nfs_ok;
1137         }
1138
1139         if (!nfsd41_copy_replay_data(resp, entry)) {
1140                 /*
1141                  * Not enough room to use the replay rpc header, send the
1142                  * cached header. Release all the allocated result pages.
1143                  */
1144                 svc_free_res_pages(resp->rqstp);
1145                 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1146                         entry->ce_resused);
1147         } else {
1148                 /* Release all but the first allocated result page */
1149
1150                 resp->rqstp->rq_resused--;
1151                 svc_free_res_pages(resp->rqstp);
1152
1153                 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1154                                  &entry->ce_respages[1],
1155                                  entry->ce_resused - 1);
1156         }
1157
1158         resp->rqstp->rq_resused = entry->ce_resused;
1159         resp->opcnt = entry->ce_opcnt;
1160         resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1161         status = entry->ce_status;
1162
1163         return status;
1164 }
1165
1166 /*
1167  * Set the exchange_id flags returned by the server.
1168  */
1169 static void
1170 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1171 {
1172         /* pNFS is not supported */
1173         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1174
1175         /* Referrals are supported, Migration is not. */
1176         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1177
1178         /* set the wire flags to return to client. */
1179         clid->flags = new->cl_exchange_flags;
1180 }
1181
1182 __be32
1183 nfsd4_exchange_id(struct svc_rqst *rqstp,
1184                   struct nfsd4_compound_state *cstate,
1185                   struct nfsd4_exchange_id *exid)
1186 {
1187         struct nfs4_client *unconf, *conf, *new;
1188         int status;
1189         unsigned int            strhashval;
1190         char                    dname[HEXDIR_LEN];
1191         nfs4_verifier           verf = exid->verifier;
1192         u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1193
1194         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1195                 " ip_addr=%u flags %x, spa_how %d\n",
1196                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1197                 ip_addr, exid->flags, exid->spa_how);
1198
1199         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1200                 return nfserr_inval;
1201
1202         /* Currently only support SP4_NONE */
1203         switch (exid->spa_how) {
1204         case SP4_NONE:
1205                 break;
1206         case SP4_SSV:
1207                 return nfserr_encr_alg_unsupp;
1208         default:
1209                 BUG();                          /* checked by xdr code */
1210         case SP4_MACH_CRED:
1211                 return nfserr_serverfault;      /* no excuse :-/ */
1212         }
1213
1214         status = nfs4_make_rec_clidname(dname, &exid->clname);
1215
1216         if (status)
1217                 goto error;
1218
1219         strhashval = clientstr_hashval(dname);
1220
1221         nfs4_lock_state();
1222         status = nfs_ok;
1223
1224         conf = find_confirmed_client_by_str(dname, strhashval, true);
1225         if (conf) {
1226                 if (!same_verf(&verf, &conf->cl_verifier)) {
1227                         /* 18.35.4 case 8 */
1228                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1229                                 status = nfserr_not_same;
1230                                 goto out;
1231                         }
1232                         /* Client reboot: destroy old state */
1233                         expire_client(conf);
1234                         goto out_new;
1235                 }
1236                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1237                         /* 18.35.4 case 9 */
1238                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1239                                 status = nfserr_perm;
1240                                 goto out;
1241                         }
1242                         expire_client(conf);
1243                         goto out_new;
1244                 }
1245                 if (ip_addr != conf->cl_addr &&
1246                     !(exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A)) {
1247                         /* Client collision. 18.35.4 case 3 */
1248                         status = nfserr_clid_inuse;
1249                         goto out;
1250                 }
1251                 /*
1252                  * Set bit when the owner id and verifier map to an already
1253                  * confirmed client id (18.35.3).
1254                  */
1255                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1256
1257                 /*
1258                  * Falling into 18.35.4 case 2, possible router replay.
1259                  * Leave confirmed record intact and return same result.
1260                  */
1261                 copy_verf(conf, &verf);
1262                 new = conf;
1263                 goto out_copy;
1264         } else {
1265                 /* 18.35.4 case 7 */
1266                 if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1267                         status = nfserr_noent;
1268                         goto out;
1269                 }
1270         }
1271
1272         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1273         if (unconf) {
1274                 /*
1275                  * Possible retry or client restart.  Per 18.35.4 case 4,
1276                  * a new unconfirmed record should be generated regardless
1277                  * of whether any properties have changed.
1278                  */
1279                 expire_client(unconf);
1280         }
1281
1282 out_new:
1283         /* Normal case */
1284         new = create_client(exid->clname, dname);
1285         if (new == NULL) {
1286                 status = nfserr_resource;
1287                 goto out;
1288         }
1289
1290         copy_verf(new, &verf);
1291         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1292         new->cl_addr = ip_addr;
1293         gen_clid(new);
1294         gen_confirm(new);
1295         add_to_unconfirmed(new, strhashval);
1296 out_copy:
1297         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1298         exid->clientid.cl_id = new->cl_clientid.cl_id;
1299
1300         new->cl_slot.sl_seqid = 0;
1301         exid->seqid = 1;
1302         nfsd4_set_ex_flags(new, exid);
1303
1304         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1305                 new->cl_slot.sl_seqid, new->cl_exchange_flags);
1306         status = nfs_ok;
1307
1308 out:
1309         nfs4_unlock_state();
1310 error:
1311         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1312         return status;
1313 }
1314
1315 static int
1316 check_slot_seqid(u32 seqid, struct nfsd4_slot *slot)
1317 {
1318         dprintk("%s enter. seqid %d slot->sl_seqid %d\n", __func__, seqid,
1319                 slot->sl_seqid);
1320
1321         /* The slot is in use, and no response has been sent. */
1322         if (slot->sl_inuse) {
1323                 if (seqid == slot->sl_seqid)
1324                         return nfserr_jukebox;
1325                 else
1326                         return nfserr_seq_misordered;
1327         }
1328         /* Normal */
1329         if (likely(seqid == slot->sl_seqid + 1))
1330                 return nfs_ok;
1331         /* Replay */
1332         if (seqid == slot->sl_seqid)
1333                 return nfserr_replay_cache;
1334         /* Wraparound */
1335         if (seqid == 1 && (slot->sl_seqid + 1) == 0)
1336                 return nfs_ok;
1337         /* Misordered replay or misordered new request */
1338         return nfserr_seq_misordered;
1339 }
1340
1341 __be32
1342 nfsd4_create_session(struct svc_rqst *rqstp,
1343                      struct nfsd4_compound_state *cstate,
1344                      struct nfsd4_create_session *cr_ses)
1345 {
1346         u32 ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1347         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1348         struct nfs4_client *conf, *unconf;
1349         struct nfsd4_slot *slot = NULL;
1350         int status = 0;
1351
1352         nfs4_lock_state();
1353         unconf = find_unconfirmed_client(&cr_ses->clientid);
1354         conf = find_confirmed_client(&cr_ses->clientid);
1355
1356         if (conf) {
1357                 slot = &conf->cl_slot;
1358                 status = check_slot_seqid(cr_ses->seqid, slot);
1359                 if (status == nfserr_replay_cache) {
1360                         dprintk("Got a create_session replay! seqid= %d\n",
1361                                 slot->sl_seqid);
1362                         cstate->slot = slot;
1363                         cstate->status = status;
1364                         /* Return the cached reply status */
1365                         status = nfsd4_replay_cache_entry(resp, NULL);
1366                         goto out;
1367                 } else if (cr_ses->seqid != conf->cl_slot.sl_seqid + 1) {
1368                         status = nfserr_seq_misordered;
1369                         dprintk("Sequence misordered!\n");
1370                         dprintk("Expected seqid= %d but got seqid= %d\n",
1371                                 slot->sl_seqid, cr_ses->seqid);
1372                         goto out;
1373                 }
1374                 conf->cl_slot.sl_seqid++;
1375         } else if (unconf) {
1376                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1377                     (ip_addr != unconf->cl_addr)) {
1378                         status = nfserr_clid_inuse;
1379                         goto out;
1380                 }
1381
1382                 slot = &unconf->cl_slot;
1383                 status = check_slot_seqid(cr_ses->seqid, slot);
1384                 if (status) {
1385                         /* an unconfirmed replay returns misordered */
1386                         status = nfserr_seq_misordered;
1387                         goto out;
1388                 }
1389
1390                 slot->sl_seqid++; /* from 0 to 1 */
1391                 move_to_confirmed(unconf);
1392
1393                 /*
1394                  * We do not support RDMA or persistent sessions
1395                  */
1396                 cr_ses->flags &= ~SESSION4_PERSIST;
1397                 cr_ses->flags &= ~SESSION4_RDMA;
1398
1399                 conf = unconf;
1400         } else {
1401                 status = nfserr_stale_clientid;
1402                 goto out;
1403         }
1404
1405         status = alloc_init_session(rqstp, conf, cr_ses);
1406         if (status)
1407                 goto out;
1408
1409         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1410                NFS4_MAX_SESSIONID_LEN);
1411         cr_ses->seqid = slot->sl_seqid;
1412
1413         slot->sl_inuse = true;
1414         cstate->slot = slot;
1415         /* Ensure a page is used for the cache */
1416         slot->sl_cache_entry.ce_cachethis = 1;
1417 out:
1418         nfs4_unlock_state();
1419         dprintk("%s returns %d\n", __func__, ntohl(status));
1420         return status;
1421 }
1422
1423 __be32
1424 nfsd4_destroy_session(struct svc_rqst *r,
1425                       struct nfsd4_compound_state *cstate,
1426                       struct nfsd4_destroy_session *sessionid)
1427 {
1428         struct nfsd4_session *ses;
1429         u32 status = nfserr_badsession;
1430
1431         /* Notes:
1432          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1433          * - Should we return nfserr_back_chan_busy if waiting for
1434          *   callbacks on to-be-destroyed session?
1435          * - Do we need to clear any callback info from previous session?
1436          */
1437
1438         dump_sessionid(__func__, &sessionid->sessionid);
1439         spin_lock(&sessionid_lock);
1440         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1441         if (!ses) {
1442                 spin_unlock(&sessionid_lock);
1443                 goto out;
1444         }
1445
1446         unhash_session(ses);
1447         spin_unlock(&sessionid_lock);
1448
1449         /* wait for callbacks */
1450         shutdown_callback_client(ses->se_client);
1451         nfsd4_put_session(ses);
1452         status = nfs_ok;
1453 out:
1454         dprintk("%s returns %d\n", __func__, ntohl(status));
1455         return status;
1456 }
1457
1458 __be32
1459 nfsd4_sequence(struct svc_rqst *rqstp,
1460                struct nfsd4_compound_state *cstate,
1461                struct nfsd4_sequence *seq)
1462 {
1463         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1464         struct nfsd4_session *session;
1465         struct nfsd4_slot *slot;
1466         int status;
1467
1468         if (resp->opcnt != 1)
1469                 return nfserr_sequence_pos;
1470
1471         spin_lock(&sessionid_lock);
1472         status = nfserr_badsession;
1473         session = find_in_sessionid_hashtbl(&seq->sessionid);
1474         if (!session)
1475                 goto out;
1476
1477         status = nfserr_badslot;
1478         if (seq->slotid >= session->se_fnumslots)
1479                 goto out;
1480
1481         slot = &session->se_slots[seq->slotid];
1482         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1483
1484         status = check_slot_seqid(seq->seqid, slot);
1485         if (status == nfserr_replay_cache) {
1486                 cstate->slot = slot;
1487                 cstate->session = session;
1488                 /* Return the cached reply status and set cstate->status
1489                  * for nfsd4_svc_encode_compoundres processing */
1490                 status = nfsd4_replay_cache_entry(resp, seq);
1491                 cstate->status = nfserr_replay_cache;
1492                 goto replay_cache;
1493         }
1494         if (status)
1495                 goto out;
1496
1497         /* Success! bump slot seqid */
1498         slot->sl_inuse = true;
1499         slot->sl_seqid = seq->seqid;
1500         slot->sl_cache_entry.ce_cachethis = seq->cachethis;
1501         /* Always set the cache entry cachethis for solo sequence */
1502         if (nfsd4_is_solo_sequence(resp))
1503                 slot->sl_cache_entry.ce_cachethis = 1;
1504
1505         cstate->slot = slot;
1506         cstate->session = session;
1507
1508 replay_cache:
1509         /* Renew the clientid on success and on replay.
1510          * Hold a session reference until done processing the compound:
1511          * nfsd4_put_session called only if the cstate slot is set.
1512          */
1513         renew_client(session->se_client);
1514         nfsd4_get_session(session);
1515 out:
1516         spin_unlock(&sessionid_lock);
1517         dprintk("%s: return %d\n", __func__, ntohl(status));
1518         return status;
1519 }
1520
1521 __be32
1522 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1523                   struct nfsd4_setclientid *setclid)
1524 {
1525         struct sockaddr_in      *sin = svc_addr_in(rqstp);
1526         struct xdr_netobj       clname = { 
1527                 .len = setclid->se_namelen,
1528                 .data = setclid->se_name,
1529         };
1530         nfs4_verifier           clverifier = setclid->se_verf;
1531         unsigned int            strhashval;
1532         struct nfs4_client      *conf, *unconf, *new;
1533         __be32                  status;
1534         char                    *princ;
1535         char                    dname[HEXDIR_LEN];
1536         
1537         if (!check_name(clname))
1538                 return nfserr_inval;
1539
1540         status = nfs4_make_rec_clidname(dname, &clname);
1541         if (status)
1542                 return status;
1543
1544         /* 
1545          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1546          * We get here on a DRC miss.
1547          */
1548
1549         strhashval = clientstr_hashval(dname);
1550
1551         nfs4_lock_state();
1552         conf = find_confirmed_client_by_str(dname, strhashval, false);
1553         if (conf) {
1554                 /* RFC 3530 14.2.33 CASE 0: */
1555                 status = nfserr_clid_inuse;
1556                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1557                         dprintk("NFSD: setclientid: string in use by client"
1558                                 " at %pI4\n", &conf->cl_addr);
1559                         goto out;
1560                 }
1561         }
1562         /*
1563          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1564          * has a description of SETCLIENTID request processing consisting
1565          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1566          */
1567         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1568         status = nfserr_resource;
1569         if (!conf) {
1570                 /*
1571                  * RFC 3530 14.2.33 CASE 4:
1572                  * placed first, because it is the normal case
1573                  */
1574                 if (unconf)
1575                         expire_client(unconf);
1576                 new = create_client(clname, dname);
1577                 if (new == NULL)
1578                         goto out;
1579                 gen_clid(new);
1580         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1581                 /*
1582                  * RFC 3530 14.2.33 CASE 1:
1583                  * probable callback update
1584                  */
1585                 if (unconf) {
1586                         /* Note this is removing unconfirmed {*x***},
1587                          * which is stronger than RFC recommended {vxc**}.
1588                          * This has the advantage that there is at most
1589                          * one {*x***} in either list at any time.
1590                          */
1591                         expire_client(unconf);
1592                 }
1593                 new = create_client(clname, dname);
1594                 if (new == NULL)
1595                         goto out;
1596                 copy_clid(new, conf);
1597         } else if (!unconf) {
1598                 /*
1599                  * RFC 3530 14.2.33 CASE 2:
1600                  * probable client reboot; state will be removed if
1601                  * confirmed.
1602                  */
1603                 new = create_client(clname, dname);
1604                 if (new == NULL)
1605                         goto out;
1606                 gen_clid(new);
1607         } else {
1608                 /*
1609                  * RFC 3530 14.2.33 CASE 3:
1610                  * probable client reboot; state will be removed if
1611                  * confirmed.
1612                  */
1613                 expire_client(unconf);
1614                 new = create_client(clname, dname);
1615                 if (new == NULL)
1616                         goto out;
1617                 gen_clid(new);
1618         }
1619         copy_verf(new, &clverifier);
1620         new->cl_addr = sin->sin_addr.s_addr;
1621         new->cl_flavor = rqstp->rq_flavor;
1622         princ = svc_gss_principal(rqstp);
1623         if (princ) {
1624                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1625                 if (new->cl_principal == NULL) {
1626                         free_client(new);
1627                         goto out;
1628                 }
1629         }
1630         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1631         gen_confirm(new);
1632         gen_callback(new, setclid);
1633         add_to_unconfirmed(new, strhashval);
1634         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1635         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1636         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1637         status = nfs_ok;
1638 out:
1639         nfs4_unlock_state();
1640         return status;
1641 }
1642
1643
1644 /*
1645  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1646  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1647  * bullets, labeled as CASE1 - CASE4 below.
1648  */
1649 __be32
1650 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1651                          struct nfsd4_compound_state *cstate,
1652                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1653 {
1654         struct sockaddr_in *sin = svc_addr_in(rqstp);
1655         struct nfs4_client *conf, *unconf;
1656         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1657         clientid_t * clid = &setclientid_confirm->sc_clientid;
1658         __be32 status;
1659
1660         if (STALE_CLIENTID(clid))
1661                 return nfserr_stale_clientid;
1662         /* 
1663          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1664          * We get here on a DRC miss.
1665          */
1666
1667         nfs4_lock_state();
1668
1669         conf = find_confirmed_client(clid);
1670         unconf = find_unconfirmed_client(clid);
1671
1672         status = nfserr_clid_inuse;
1673         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
1674                 goto out;
1675         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
1676                 goto out;
1677
1678         /*
1679          * section 14.2.34 of RFC 3530 has a description of
1680          * SETCLIENTID_CONFIRM request processing consisting
1681          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1682          */
1683         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1684                 /*
1685                  * RFC 3530 14.2.34 CASE 1:
1686                  * callback update
1687                  */
1688                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1689                         status = nfserr_clid_inuse;
1690                 else {
1691                         /* XXX: We just turn off callbacks until we can handle
1692                           * change request correctly. */
1693                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1694                         expire_client(unconf);
1695                         status = nfs_ok;
1696
1697                 }
1698         } else if (conf && !unconf) {
1699                 /*
1700                  * RFC 3530 14.2.34 CASE 2:
1701                  * probable retransmitted request; play it safe and
1702                  * do nothing.
1703                  */
1704                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1705                         status = nfserr_clid_inuse;
1706                 else
1707                         status = nfs_ok;
1708         } else if (!conf && unconf
1709                         && same_verf(&unconf->cl_confirm, &confirm)) {
1710                 /*
1711                  * RFC 3530 14.2.34 CASE 3:
1712                  * Normal case; new or rebooted client:
1713                  */
1714                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1715                         status = nfserr_clid_inuse;
1716                 } else {
1717                         unsigned int hash =
1718                                 clientstr_hashval(unconf->cl_recdir);
1719                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1720                                                             hash, false);
1721                         if (conf) {
1722                                 nfsd4_remove_clid_dir(conf);
1723                                 expire_client(conf);
1724                         }
1725                         move_to_confirmed(unconf);
1726                         conf = unconf;
1727                         nfsd4_probe_callback(conf);
1728                         status = nfs_ok;
1729                 }
1730         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1731             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1732                                                                 &confirm)))) {
1733                 /*
1734                  * RFC 3530 14.2.34 CASE 4:
1735                  * Client probably hasn't noticed that we rebooted yet.
1736                  */
1737                 status = nfserr_stale_clientid;
1738         } else {
1739                 /* check that we have hit one of the cases...*/
1740                 status = nfserr_clid_inuse;
1741         }
1742 out:
1743         nfs4_unlock_state();
1744         return status;
1745 }
1746
1747 /* OPEN Share state helper functions */
1748 static inline struct nfs4_file *
1749 alloc_init_file(struct inode *ino)
1750 {
1751         struct nfs4_file *fp;
1752         unsigned int hashval = file_hashval(ino);
1753
1754         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1755         if (fp) {
1756                 atomic_set(&fp->fi_ref, 1);
1757                 INIT_LIST_HEAD(&fp->fi_hash);
1758                 INIT_LIST_HEAD(&fp->fi_stateids);
1759                 INIT_LIST_HEAD(&fp->fi_delegations);
1760                 spin_lock(&recall_lock);
1761                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1762                 spin_unlock(&recall_lock);
1763                 fp->fi_inode = igrab(ino);
1764                 fp->fi_id = current_fileid++;
1765                 fp->fi_had_conflict = false;
1766                 return fp;
1767         }
1768         return NULL;
1769 }
1770
1771 static void
1772 nfsd4_free_slab(struct kmem_cache **slab)
1773 {
1774         if (*slab == NULL)
1775                 return;
1776         kmem_cache_destroy(*slab);
1777         *slab = NULL;
1778 }
1779
1780 void
1781 nfsd4_free_slabs(void)
1782 {
1783         nfsd4_free_slab(&stateowner_slab);
1784         nfsd4_free_slab(&file_slab);
1785         nfsd4_free_slab(&stateid_slab);
1786         nfsd4_free_slab(&deleg_slab);
1787 }
1788
1789 static int
1790 nfsd4_init_slabs(void)
1791 {
1792         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1793                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1794         if (stateowner_slab == NULL)
1795                 goto out_nomem;
1796         file_slab = kmem_cache_create("nfsd4_files",
1797                         sizeof(struct nfs4_file), 0, 0, NULL);
1798         if (file_slab == NULL)
1799                 goto out_nomem;
1800         stateid_slab = kmem_cache_create("nfsd4_stateids",
1801                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1802         if (stateid_slab == NULL)
1803                 goto out_nomem;
1804         deleg_slab = kmem_cache_create("nfsd4_delegations",
1805                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1806         if (deleg_slab == NULL)
1807                 goto out_nomem;
1808         return 0;
1809 out_nomem:
1810         nfsd4_free_slabs();
1811         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1812         return -ENOMEM;
1813 }
1814
1815 void
1816 nfs4_free_stateowner(struct kref *kref)
1817 {
1818         struct nfs4_stateowner *sop =
1819                 container_of(kref, struct nfs4_stateowner, so_ref);
1820         kfree(sop->so_owner.data);
1821         kmem_cache_free(stateowner_slab, sop);
1822 }
1823
1824 static inline struct nfs4_stateowner *
1825 alloc_stateowner(struct xdr_netobj *owner)
1826 {
1827         struct nfs4_stateowner *sop;
1828
1829         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1830                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1831                         memcpy(sop->so_owner.data, owner->data, owner->len);
1832                         sop->so_owner.len = owner->len;
1833                         kref_init(&sop->so_ref);
1834                         return sop;
1835                 } 
1836                 kmem_cache_free(stateowner_slab, sop);
1837         }
1838         return NULL;
1839 }
1840
1841 static struct nfs4_stateowner *
1842 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1843         struct nfs4_stateowner *sop;
1844         struct nfs4_replay *rp;
1845         unsigned int idhashval;
1846
1847         if (!(sop = alloc_stateowner(&open->op_owner)))
1848                 return NULL;
1849         idhashval = ownerid_hashval(current_ownerid);
1850         INIT_LIST_HEAD(&sop->so_idhash);
1851         INIT_LIST_HEAD(&sop->so_strhash);
1852         INIT_LIST_HEAD(&sop->so_perclient);
1853         INIT_LIST_HEAD(&sop->so_stateids);
1854         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1855         INIT_LIST_HEAD(&sop->so_close_lru);
1856         sop->so_time = 0;
1857         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1858         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1859         list_add(&sop->so_perclient, &clp->cl_openowners);
1860         sop->so_is_open_owner = 1;
1861         sop->so_id = current_ownerid++;
1862         sop->so_client = clp;
1863         sop->so_seqid = open->op_seqid;
1864         sop->so_confirmed = 0;
1865         rp = &sop->so_replay;
1866         rp->rp_status = nfserr_serverfault;
1867         rp->rp_buflen = 0;
1868         rp->rp_buf = rp->rp_ibuf;
1869         return sop;
1870 }
1871
1872 static inline void
1873 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1874         struct nfs4_stateowner *sop = open->op_stateowner;
1875         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1876
1877         INIT_LIST_HEAD(&stp->st_hash);
1878         INIT_LIST_HEAD(&stp->st_perstateowner);
1879         INIT_LIST_HEAD(&stp->st_lockowners);
1880         INIT_LIST_HEAD(&stp->st_perfile);
1881         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1882         list_add(&stp->st_perstateowner, &sop->so_stateids);
1883         list_add(&stp->st_perfile, &fp->fi_stateids);
1884         stp->st_stateowner = sop;
1885         get_nfs4_file(fp);
1886         stp->st_file = fp;
1887         stp->st_stateid.si_boot = get_seconds();
1888         stp->st_stateid.si_stateownerid = sop->so_id;
1889         stp->st_stateid.si_fileid = fp->fi_id;
1890         stp->st_stateid.si_generation = 0;
1891         stp->st_access_bmap = 0;
1892         stp->st_deny_bmap = 0;
1893         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1894                   &stp->st_access_bmap);
1895         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1896         stp->st_openstp = NULL;
1897 }
1898
1899 static void
1900 move_to_close_lru(struct nfs4_stateowner *sop)
1901 {
1902         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1903
1904         list_move_tail(&sop->so_close_lru, &close_lru);
1905         sop->so_time = get_seconds();
1906 }
1907
1908 static int
1909 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1910                                                         clientid_t *clid)
1911 {
1912         return (sop->so_owner.len == owner->len) &&
1913                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1914                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1915 }
1916
1917 static struct nfs4_stateowner *
1918 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1919 {
1920         struct nfs4_stateowner *so = NULL;
1921
1922         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1923                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1924                         return so;
1925         }
1926         return NULL;
1927 }
1928
1929 /* search file_hashtbl[] for file */
1930 static struct nfs4_file *
1931 find_file(struct inode *ino)
1932 {
1933         unsigned int hashval = file_hashval(ino);
1934         struct nfs4_file *fp;
1935
1936         spin_lock(&recall_lock);
1937         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1938                 if (fp->fi_inode == ino) {
1939                         get_nfs4_file(fp);
1940                         spin_unlock(&recall_lock);
1941                         return fp;
1942                 }
1943         }
1944         spin_unlock(&recall_lock);
1945         return NULL;
1946 }
1947
1948 static inline int access_valid(u32 x, u32 minorversion)
1949 {
1950         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1951                 return 0;
1952         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1953                 return 0;
1954         x &= ~NFS4_SHARE_ACCESS_MASK;
1955         if (minorversion && x) {
1956                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1957                         return 0;
1958                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1959                         return 0;
1960                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1961         }
1962         if (x)
1963                 return 0;
1964         return 1;
1965 }
1966
1967 static inline int deny_valid(u32 x)
1968 {
1969         /* Note: unlike access bits, deny bits may be zero. */
1970         return x <= NFS4_SHARE_DENY_BOTH;
1971 }
1972
1973 /*
1974  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1975  * st_{access,deny}_bmap field of the stateid, in order to track not
1976  * only what share bits are currently in force, but also what
1977  * combinations of share bits previous opens have used.  This allows us
1978  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1979  * return an error if the client attempt to downgrade to a combination
1980  * of share bits not explicable by closing some of its previous opens.
1981  *
1982  * XXX: This enforcement is actually incomplete, since we don't keep
1983  * track of access/deny bit combinations; so, e.g., we allow:
1984  *
1985  *      OPEN allow read, deny write
1986  *      OPEN allow both, deny none
1987  *      DOWNGRADE allow read, deny none
1988  *
1989  * which we should reject.
1990  */
1991 static void
1992 set_access(unsigned int *access, unsigned long bmap) {
1993         int i;
1994
1995         *access = 0;
1996         for (i = 1; i < 4; i++) {
1997                 if (test_bit(i, &bmap))
1998                         *access |= i;
1999         }
2000 }
2001
2002 static void
2003 set_deny(unsigned int *deny, unsigned long bmap) {
2004         int i;
2005
2006         *deny = 0;
2007         for (i = 0; i < 4; i++) {
2008                 if (test_bit(i, &bmap))
2009                         *deny |= i ;
2010         }
2011 }
2012
2013 static int
2014 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
2015         unsigned int access, deny;
2016
2017         set_access(&access, stp->st_access_bmap);
2018         set_deny(&deny, stp->st_deny_bmap);
2019         if ((access & open->op_share_deny) || (deny & open->op_share_access))
2020                 return 0;
2021         return 1;
2022 }
2023
2024 /*
2025  * Called to check deny when READ with all zero stateid or
2026  * WRITE with all zero or all one stateid
2027  */
2028 static __be32
2029 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2030 {
2031         struct inode *ino = current_fh->fh_dentry->d_inode;
2032         struct nfs4_file *fp;
2033         struct nfs4_stateid *stp;
2034         __be32 ret;
2035
2036         dprintk("NFSD: nfs4_share_conflict\n");
2037
2038         fp = find_file(ino);
2039         if (!fp)
2040                 return nfs_ok;
2041         ret = nfserr_locked;
2042         /* Search for conflicting share reservations */
2043         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2044                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2045                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2046                         goto out;
2047         }
2048         ret = nfs_ok;
2049 out:
2050         put_nfs4_file(fp);
2051         return ret;
2052 }
2053
2054 static inline void
2055 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
2056 {
2057         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2058                 drop_file_write_access(filp);
2059                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2060         }
2061 }
2062
2063 /*
2064  * Recall a delegation
2065  */
2066 static int
2067 do_recall(void *__dp)
2068 {
2069         struct nfs4_delegation *dp = __dp;
2070
2071         dp->dl_file->fi_had_conflict = true;
2072         nfsd4_cb_recall(dp);
2073         return 0;
2074 }
2075
2076 /*
2077  * Spawn a thread to perform a recall on the delegation represented
2078  * by the lease (file_lock)
2079  *
2080  * Called from break_lease() with lock_kernel() held.
2081  * Note: we assume break_lease will only call this *once* for any given
2082  * lease.
2083  */
2084 static
2085 void nfsd_break_deleg_cb(struct file_lock *fl)
2086 {
2087         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
2088         struct task_struct *t;
2089
2090         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2091         if (!dp)
2092                 return;
2093
2094         /* We're assuming the state code never drops its reference
2095          * without first removing the lease.  Since we're in this lease
2096          * callback (and since the lease code is serialized by the kernel
2097          * lock) we know the server hasn't removed the lease yet, we know
2098          * it's safe to take a reference: */
2099         atomic_inc(&dp->dl_count);
2100         atomic_inc(&dp->dl_client->cl_count);
2101
2102         spin_lock(&recall_lock);
2103         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2104         spin_unlock(&recall_lock);
2105
2106         /* only place dl_time is set. protected by lock_kernel*/
2107         dp->dl_time = get_seconds();
2108
2109         /*
2110          * We don't want the locks code to timeout the lease for us;
2111          * we'll remove it ourself if the delegation isn't returned
2112          * in time.
2113          */
2114         fl->fl_break_time = 0;
2115
2116         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
2117         if (IS_ERR(t)) {
2118                 struct nfs4_client *clp = dp->dl_client;
2119
2120                 printk(KERN_INFO "NFSD: Callback thread failed for "
2121                         "for client (clientid %08x/%08x)\n",
2122                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2123                 put_nfs4_client(dp->dl_client);
2124                 nfs4_put_delegation(dp);
2125         }
2126 }
2127
2128 /*
2129  * The file_lock is being reapd.
2130  *
2131  * Called by locks_free_lock() with lock_kernel() held.
2132  */
2133 static
2134 void nfsd_release_deleg_cb(struct file_lock *fl)
2135 {
2136         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2137
2138         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2139
2140         if (!(fl->fl_flags & FL_LEASE) || !dp)
2141                 return;
2142         dp->dl_flock = NULL;
2143 }
2144
2145 /*
2146  * Set the delegation file_lock back pointer.
2147  *
2148  * Called from setlease() with lock_kernel() held.
2149  */
2150 static
2151 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2152 {
2153         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2154
2155         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2156         if (!dp)
2157                 return;
2158         dp->dl_flock = new;
2159 }
2160
2161 /*
2162  * Called from setlease() with lock_kernel() held
2163  */
2164 static
2165 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2166 {
2167         struct nfs4_delegation *onlistd =
2168                 (struct nfs4_delegation *)onlist->fl_owner;
2169         struct nfs4_delegation *tryd =
2170                 (struct nfs4_delegation *)try->fl_owner;
2171
2172         if (onlist->fl_lmops != try->fl_lmops)
2173                 return 0;
2174
2175         return onlistd->dl_client == tryd->dl_client;
2176 }
2177
2178
2179 static
2180 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2181 {
2182         if (arg & F_UNLCK)
2183                 return lease_modify(onlist, arg);
2184         else
2185                 return -EAGAIN;
2186 }
2187
2188 static struct lock_manager_operations nfsd_lease_mng_ops = {
2189         .fl_break = nfsd_break_deleg_cb,
2190         .fl_release_private = nfsd_release_deleg_cb,
2191         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2192         .fl_mylease = nfsd_same_client_deleg_cb,
2193         .fl_change = nfsd_change_deleg_cb,
2194 };
2195
2196
2197 __be32
2198 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2199                     struct nfsd4_open *open)
2200 {
2201         clientid_t *clientid = &open->op_clientid;
2202         struct nfs4_client *clp = NULL;
2203         unsigned int strhashval;
2204         struct nfs4_stateowner *sop = NULL;
2205
2206         if (!check_name(open->op_owner))
2207                 return nfserr_inval;
2208
2209         if (STALE_CLIENTID(&open->op_clientid))
2210                 return nfserr_stale_clientid;
2211
2212         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2213         sop = find_openstateowner_str(strhashval, open);
2214         open->op_stateowner = sop;
2215         if (!sop) {
2216                 /* Make sure the client's lease hasn't expired. */
2217                 clp = find_confirmed_client(clientid);
2218                 if (clp == NULL)
2219                         return nfserr_expired;
2220                 goto renew;
2221         }
2222         /* When sessions are used, skip open sequenceid processing */
2223         if (nfsd4_has_session(cstate))
2224                 goto renew;
2225         if (!sop->so_confirmed) {
2226                 /* Replace unconfirmed owners without checking for replay. */
2227                 clp = sop->so_client;
2228                 release_openowner(sop);
2229                 open->op_stateowner = NULL;
2230                 goto renew;
2231         }
2232         if (open->op_seqid == sop->so_seqid - 1) {
2233                 if (sop->so_replay.rp_buflen)
2234                         return nfserr_replay_me;
2235                 /* The original OPEN failed so spectacularly
2236                  * that we don't even have replay data saved!
2237                  * Therefore, we have no choice but to continue
2238                  * processing this OPEN; presumably, we'll
2239                  * fail again for the same reason.
2240                  */
2241                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2242                 goto renew;
2243         }
2244         if (open->op_seqid != sop->so_seqid)
2245                 return nfserr_bad_seqid;
2246 renew:
2247         if (open->op_stateowner == NULL) {
2248                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2249                 if (sop == NULL)
2250                         return nfserr_resource;
2251                 open->op_stateowner = sop;
2252         }
2253         list_del_init(&sop->so_close_lru);
2254         renew_client(sop->so_client);
2255         return nfs_ok;
2256 }
2257
2258 static inline __be32
2259 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2260 {
2261         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2262                 return nfserr_openmode;
2263         else
2264                 return nfs_ok;
2265 }
2266
2267 static struct nfs4_delegation *
2268 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2269 {
2270         struct nfs4_delegation *dp;
2271
2272         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2273                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2274                         return dp;
2275         }
2276         return NULL;
2277 }
2278
2279 static __be32
2280 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2281                 struct nfs4_delegation **dp)
2282 {
2283         int flags;
2284         __be32 status = nfserr_bad_stateid;
2285
2286         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2287         if (*dp == NULL)
2288                 goto out;
2289         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2290                                                 RD_STATE : WR_STATE;
2291         status = nfs4_check_delegmode(*dp, flags);
2292         if (status)
2293                 *dp = NULL;
2294 out:
2295         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2296                 return nfs_ok;
2297         if (status)
2298                 return status;
2299         open->op_stateowner->so_confirmed = 1;
2300         return nfs_ok;
2301 }
2302
2303 static __be32
2304 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2305 {
2306         struct nfs4_stateid *local;
2307         __be32 status = nfserr_share_denied;
2308         struct nfs4_stateowner *sop = open->op_stateowner;
2309
2310         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2311                 /* ignore lock owners */
2312                 if (local->st_stateowner->so_is_open_owner == 0)
2313                         continue;
2314                 /* remember if we have seen this open owner */
2315                 if (local->st_stateowner == sop)
2316                         *stpp = local;
2317                 /* check for conflicting share reservations */
2318                 if (!test_share(local, open))
2319                         goto out;
2320         }
2321         status = 0;
2322 out:
2323         return status;
2324 }
2325
2326 static inline struct nfs4_stateid *
2327 nfs4_alloc_stateid(void)
2328 {
2329         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2330 }
2331
2332 static __be32
2333 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2334                 struct nfs4_delegation *dp,
2335                 struct svc_fh *cur_fh, int flags)
2336 {
2337         struct nfs4_stateid *stp;
2338
2339         stp = nfs4_alloc_stateid();
2340         if (stp == NULL)
2341                 return nfserr_resource;
2342
2343         if (dp) {
2344                 get_file(dp->dl_vfs_file);
2345                 stp->st_vfs_file = dp->dl_vfs_file;
2346         } else {
2347                 __be32 status;
2348                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2349                                 &stp->st_vfs_file);
2350                 if (status) {
2351                         if (status == nfserr_dropit)
2352                                 status = nfserr_jukebox;
2353                         kmem_cache_free(stateid_slab, stp);
2354                         return status;
2355                 }
2356         }
2357         *stpp = stp;
2358         return 0;
2359 }
2360
2361 static inline __be32
2362 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2363                 struct nfsd4_open *open)
2364 {
2365         struct iattr iattr = {
2366                 .ia_valid = ATTR_SIZE,
2367                 .ia_size = 0,
2368         };
2369         if (!open->op_truncate)
2370                 return 0;
2371         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2372                 return nfserr_inval;
2373         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2374 }
2375
2376 static __be32
2377 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2378 {
2379         struct file *filp = stp->st_vfs_file;
2380         struct inode *inode = filp->f_path.dentry->d_inode;
2381         unsigned int share_access, new_writer;
2382         __be32 status;
2383
2384         set_access(&share_access, stp->st_access_bmap);
2385         new_writer = (~share_access) & open->op_share_access
2386                         & NFS4_SHARE_ACCESS_WRITE;
2387
2388         if (new_writer) {
2389                 int err = get_write_access(inode);
2390                 if (err)
2391                         return nfserrno(err);
2392                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2393                 if (err)
2394                         return nfserrno(err);
2395                 file_take_write(filp);
2396         }
2397         status = nfsd4_truncate(rqstp, cur_fh, open);
2398         if (status) {
2399                 if (new_writer)
2400                         put_write_access(inode);
2401                 return status;
2402         }
2403         /* remember the open */
2404         filp->f_mode |= open->op_share_access;
2405         __set_bit(open->op_share_access, &stp->st_access_bmap);
2406         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2407
2408         return nfs_ok;
2409 }
2410
2411
2412 static void
2413 nfs4_set_claim_prev(struct nfsd4_open *open)
2414 {
2415         open->op_stateowner->so_confirmed = 1;
2416         open->op_stateowner->so_client->cl_firststate = 1;
2417 }
2418
2419 /*
2420  * Attempt to hand out a delegation.
2421  */
2422 static void
2423 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2424 {
2425         struct nfs4_delegation *dp;
2426         struct nfs4_stateowner *sop = stp->st_stateowner;
2427         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2428         struct file_lock fl, *flp = &fl;
2429         int status, flag = 0;
2430
2431         flag = NFS4_OPEN_DELEGATE_NONE;
2432         open->op_recall = 0;
2433         switch (open->op_claim_type) {
2434                 case NFS4_OPEN_CLAIM_PREVIOUS:
2435                         if (!atomic_read(&cb->cb_set))
2436                                 open->op_recall = 1;
2437                         flag = open->op_delegate_type;
2438                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2439                                 goto out;
2440                         break;
2441                 case NFS4_OPEN_CLAIM_NULL:
2442                         /* Let's not give out any delegations till everyone's
2443                          * had the chance to reclaim theirs.... */
2444                         if (locks_in_grace())
2445                                 goto out;
2446                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2447                                 goto out;
2448                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2449                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2450                         else
2451                                 flag = NFS4_OPEN_DELEGATE_READ;
2452                         break;
2453                 default:
2454                         goto out;
2455         }
2456
2457         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2458         if (dp == NULL) {
2459                 flag = NFS4_OPEN_DELEGATE_NONE;
2460                 goto out;
2461         }
2462         locks_init_lock(&fl);
2463         fl.fl_lmops = &nfsd_lease_mng_ops;
2464         fl.fl_flags = FL_LEASE;
2465         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2466         fl.fl_end = OFFSET_MAX;
2467         fl.fl_owner =  (fl_owner_t)dp;
2468         fl.fl_file = stp->st_vfs_file;
2469         fl.fl_pid = current->tgid;
2470
2471         /* vfs_setlease checks to see if delegation should be handed out.
2472          * the lock_manager callbacks fl_mylease and fl_change are used
2473          */
2474         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2475                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2476                 unhash_delegation(dp);
2477                 flag = NFS4_OPEN_DELEGATE_NONE;
2478                 goto out;
2479         }
2480
2481         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2482
2483         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2484                      dp->dl_stateid.si_boot,
2485                      dp->dl_stateid.si_stateownerid,
2486                      dp->dl_stateid.si_fileid,
2487                      dp->dl_stateid.si_generation);
2488 out:
2489         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2490                         && flag == NFS4_OPEN_DELEGATE_NONE
2491                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2492                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2493         open->op_delegate_type = flag;
2494 }
2495
2496 /*
2497  * called with nfs4_lock_state() held.
2498  */
2499 __be32
2500 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2501 {
2502         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2503         struct nfs4_file *fp = NULL;
2504         struct inode *ino = current_fh->fh_dentry->d_inode;
2505         struct nfs4_stateid *stp = NULL;
2506         struct nfs4_delegation *dp = NULL;
2507         __be32 status;
2508
2509         status = nfserr_inval;
2510         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2511                         || !deny_valid(open->op_share_deny))
2512                 goto out;
2513         /*
2514          * Lookup file; if found, lookup stateid and check open request,
2515          * and check for delegations in the process of being recalled.
2516          * If not found, create the nfs4_file struct
2517          */
2518         fp = find_file(ino);
2519         if (fp) {
2520                 if ((status = nfs4_check_open(fp, open, &stp)))
2521                         goto out;
2522                 status = nfs4_check_deleg(fp, open, &dp);
2523                 if (status)
2524                         goto out;
2525         } else {
2526                 status = nfserr_bad_stateid;
2527                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2528                         goto out;
2529                 status = nfserr_resource;
2530                 fp = alloc_init_file(ino);
2531                 if (fp == NULL)
2532                         goto out;
2533         }
2534
2535         /*
2536          * OPEN the file, or upgrade an existing OPEN.
2537          * If truncate fails, the OPEN fails.
2538          */
2539         if (stp) {
2540                 /* Stateid was found, this is an OPEN upgrade */
2541                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2542                 if (status)
2543                         goto out;
2544                 update_stateid(&stp->st_stateid);
2545         } else {
2546                 /* Stateid was not found, this is a new OPEN */
2547                 int flags = 0;
2548                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2549                         flags |= NFSD_MAY_READ;
2550                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2551                         flags |= NFSD_MAY_WRITE;
2552                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2553                 if (status)
2554                         goto out;
2555                 init_stateid(stp, fp, open);
2556                 status = nfsd4_truncate(rqstp, current_fh, open);
2557                 if (status) {
2558                         release_open_stateid(stp);
2559                         goto out;
2560                 }
2561                 if (nfsd4_has_session(&resp->cstate))
2562                         update_stateid(&stp->st_stateid);
2563         }
2564         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2565
2566         if (nfsd4_has_session(&resp->cstate))
2567                 open->op_stateowner->so_confirmed = 1;
2568
2569         /*
2570         * Attempt to hand out a delegation. No error return, because the
2571         * OPEN succeeds even if we fail.
2572         */
2573         nfs4_open_delegation(current_fh, open, stp);
2574
2575         status = nfs_ok;
2576
2577         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2578                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2579                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2580 out:
2581         if (fp)
2582                 put_nfs4_file(fp);
2583         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2584                 nfs4_set_claim_prev(open);
2585         /*
2586         * To finish the open response, we just need to set the rflags.
2587         */
2588         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2589         if (!open->op_stateowner->so_confirmed &&
2590             !nfsd4_has_session(&resp->cstate))
2591                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2592
2593         return status;
2594 }
2595
2596 __be32
2597 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2598             clientid_t *clid)
2599 {
2600         struct nfs4_client *clp;
2601         __be32 status;
2602
2603         nfs4_lock_state();
2604         dprintk("process_renew(%08x/%08x): starting\n", 
2605                         clid->cl_boot, clid->cl_id);
2606         status = nfserr_stale_clientid;
2607         if (STALE_CLIENTID(clid))
2608                 goto out;
2609         clp = find_confirmed_client(clid);
2610         status = nfserr_expired;
2611         if (clp == NULL) {
2612                 /* We assume the client took too long to RENEW. */
2613                 dprintk("nfsd4_renew: clientid not found!\n");
2614                 goto out;
2615         }
2616         renew_client(clp);
2617         status = nfserr_cb_path_down;
2618         if (!list_empty(&clp->cl_delegations)
2619                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2620                 goto out;
2621         status = nfs_ok;
2622 out:
2623         nfs4_unlock_state();
2624         return status;
2625 }
2626
2627 struct lock_manager nfsd4_manager = {
2628 };
2629
2630 static void
2631 nfsd4_end_grace(void)
2632 {
2633         dprintk("NFSD: end of grace period\n");
2634         nfsd4_recdir_purge_old();
2635         locks_end_grace(&nfsd4_manager);
2636 }
2637
2638 static time_t
2639 nfs4_laundromat(void)
2640 {
2641         struct nfs4_client *clp;
2642         struct nfs4_stateowner *sop;
2643         struct nfs4_delegation *dp;
2644         struct list_head *pos, *next, reaplist;
2645         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2646         time_t t, clientid_val = NFSD_LEASE_TIME;
2647         time_t u, test_val = NFSD_LEASE_TIME;
2648
2649         nfs4_lock_state();
2650
2651         dprintk("NFSD: laundromat service - starting\n");
2652         if (locks_in_grace())
2653                 nfsd4_end_grace();
2654         list_for_each_safe(pos, next, &client_lru) {
2655                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2656                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2657                         t = clp->cl_time - cutoff;
2658                         if (clientid_val > t)
2659                                 clientid_val = t;
2660                         break;
2661                 }
2662                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2663                         clp->cl_clientid.cl_id);
2664                 nfsd4_remove_clid_dir(clp);
2665                 expire_client(clp);
2666         }
2667         INIT_LIST_HEAD(&reaplist);
2668         spin_lock(&recall_lock);
2669         list_for_each_safe(pos, next, &del_recall_lru) {
2670                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2671                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2672                         u = dp->dl_time - cutoff;
2673                         if (test_val > u)
2674                                 test_val = u;
2675                         break;
2676                 }
2677                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2678                                     dp, dp->dl_flock);
2679                 list_move(&dp->dl_recall_lru, &reaplist);
2680         }
2681         spin_unlock(&recall_lock);
2682         list_for_each_safe(pos, next, &reaplist) {
2683                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2684                 list_del_init(&dp->dl_recall_lru);
2685                 unhash_delegation(dp);
2686         }
2687         test_val = NFSD_LEASE_TIME;
2688         list_for_each_safe(pos, next, &close_lru) {
2689                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2690                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2691                         u = sop->so_time - cutoff;
2692                         if (test_val > u)
2693                                 test_val = u;
2694                         break;
2695                 }
2696                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2697                         sop->so_id);
2698                 release_openowner(sop);
2699         }
2700         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2701                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2702         nfs4_unlock_state();
2703         return clientid_val;
2704 }
2705
2706 static struct workqueue_struct *laundry_wq;
2707 static void laundromat_main(struct work_struct *);
2708 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2709
2710 static void
2711 laundromat_main(struct work_struct *not_used)
2712 {
2713         time_t t;
2714
2715         t = nfs4_laundromat();
2716         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2717         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2718 }
2719
2720 static struct nfs4_stateowner *
2721 search_close_lru(u32 st_id, int flags)
2722 {
2723         struct nfs4_stateowner *local = NULL;
2724
2725         if (flags & CLOSE_STATE) {
2726                 list_for_each_entry(local, &close_lru, so_close_lru) {
2727                         if (local->so_id == st_id)
2728                                 return local;
2729                 }
2730         }
2731         return NULL;
2732 }
2733
2734 static inline int
2735 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2736 {
2737         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2738 }
2739
2740 static int
2741 STALE_STATEID(stateid_t *stateid)
2742 {
2743         if (time_after((unsigned long)boot_time,
2744                         (unsigned long)stateid->si_boot)) {
2745                 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2746                         stateid->si_boot, stateid->si_stateownerid,
2747                         stateid->si_fileid, stateid->si_generation);
2748                 return 1;
2749         }
2750         return 0;
2751 }
2752
2753 static int
2754 EXPIRED_STATEID(stateid_t *stateid)
2755 {
2756         if (time_before((unsigned long)boot_time,
2757                         ((unsigned long)stateid->si_boot)) &&
2758             time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2759                 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2760                         stateid->si_boot, stateid->si_stateownerid,
2761                         stateid->si_fileid, stateid->si_generation);
2762                 return 1;
2763         }
2764         return 0;
2765 }
2766
2767 static __be32
2768 stateid_error_map(stateid_t *stateid)
2769 {
2770         if (STALE_STATEID(stateid))
2771                 return nfserr_stale_stateid;
2772         if (EXPIRED_STATEID(stateid))
2773                 return nfserr_expired;
2774
2775         dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2776                 stateid->si_boot, stateid->si_stateownerid,
2777                 stateid->si_fileid, stateid->si_generation);
2778         return nfserr_bad_stateid;
2779 }
2780
2781 static inline int
2782 access_permit_read(unsigned long access_bmap)
2783 {
2784         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2785                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2786                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2787 }
2788
2789 static inline int
2790 access_permit_write(unsigned long access_bmap)
2791 {
2792         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2793                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2794 }
2795
2796 static
2797 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2798 {
2799         __be32 status = nfserr_openmode;
2800
2801         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2802                 goto out;
2803         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2804                 goto out;
2805         status = nfs_ok;
2806 out:
2807         return status;
2808 }
2809
2810 static inline __be32
2811 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2812 {
2813         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2814                 return nfs_ok;
2815         else if (locks_in_grace()) {
2816                 /* Answer in remaining cases depends on existance of
2817                  * conflicting state; so we must wait out the grace period. */
2818                 return nfserr_grace;
2819         } else if (flags & WR_STATE)
2820                 return nfs4_share_conflict(current_fh,
2821                                 NFS4_SHARE_DENY_WRITE);
2822         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2823                 return nfs4_share_conflict(current_fh,
2824                                 NFS4_SHARE_DENY_READ);
2825 }
2826
2827 /*
2828  * Allow READ/WRITE during grace period on recovered state only for files
2829  * that are not able to provide mandatory locking.
2830  */
2831 static inline int
2832 grace_disallows_io(struct inode *inode)
2833 {
2834         return locks_in_grace() && mandatory_lock(inode);
2835 }
2836
2837 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2838 {
2839         /*
2840          * When sessions are used the stateid generation number is ignored
2841          * when it is zero.
2842          */
2843         if ((flags & HAS_SESSION) && in->si_generation == 0)
2844                 goto out;
2845
2846         /* If the client sends us a stateid from the future, it's buggy: */
2847         if (in->si_generation > ref->si_generation)
2848                 return nfserr_bad_stateid;
2849         /*
2850          * The following, however, can happen.  For example, if the
2851          * client sends an open and some IO at the same time, the open
2852          * may bump si_generation while the IO is still in flight.
2853          * Thanks to hard links and renames, the client never knows what
2854          * file an open will affect.  So it could avoid that situation
2855          * only by serializing all opens and IO from the same open
2856          * owner.  To recover from the old_stateid error, the client
2857          * will just have to retry the IO:
2858          */
2859         if (in->si_generation < ref->si_generation)
2860                 return nfserr_old_stateid;
2861 out:
2862         return nfs_ok;
2863 }
2864
2865 static int is_delegation_stateid(stateid_t *stateid)
2866 {
2867         return stateid->si_fileid == 0;
2868 }
2869
2870 /*
2871 * Checks for stateid operations
2872 */
2873 __be32
2874 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2875                            stateid_t *stateid, int flags, struct file **filpp)
2876 {
2877         struct nfs4_stateid *stp = NULL;
2878         struct nfs4_delegation *dp = NULL;
2879         struct svc_fh *current_fh = &cstate->current_fh;
2880         struct inode *ino = current_fh->fh_dentry->d_inode;
2881         __be32 status;
2882
2883         if (filpp)
2884                 *filpp = NULL;
2885
2886         if (grace_disallows_io(ino))
2887                 return nfserr_grace;
2888
2889         if (nfsd4_has_session(cstate))
2890                 flags |= HAS_SESSION;
2891
2892         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2893                 return check_special_stateids(current_fh, stateid, flags);
2894
2895         status = nfserr_stale_stateid;
2896         if (STALE_STATEID(stateid)) 
2897                 goto out;
2898
2899         status = nfserr_bad_stateid;
2900         if (is_delegation_stateid(stateid)) {
2901                 dp = find_delegation_stateid(ino, stateid);
2902                 if (!dp) {
2903                         status = stateid_error_map(stateid);
2904                         goto out;
2905                 }
2906                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2907                                                   flags);
2908                 if (status)
2909                         goto out;
2910                 status = nfs4_check_delegmode(dp, flags);
2911                 if (status)
2912                         goto out;
2913                 renew_client(dp->dl_client);
2914                 if (filpp)
2915                         *filpp = dp->dl_vfs_file;
2916         } else { /* open or lock stateid */
2917                 stp = find_stateid(stateid, flags);
2918                 if (!stp) {
2919                         status = stateid_error_map(stateid);
2920                         goto out;
2921                 }
2922                 if (nfs4_check_fh(current_fh, stp))
2923                         goto out;
2924                 if (!stp->st_stateowner->so_confirmed)
2925                         goto out;
2926                 status = check_stateid_generation(stateid, &stp->st_stateid,
2927                                                   flags);
2928                 if (status)
2929                         goto out;
2930                 status = nfs4_check_openmode(stp, flags);
2931                 if (status)
2932                         goto out;
2933                 renew_client(stp->st_stateowner->so_client);
2934                 if (filpp)
2935                         *filpp = stp->st_vfs_file;
2936         }
2937         status = nfs_ok;
2938 out:
2939         return status;
2940 }
2941
2942 static inline int
2943 setlkflg (int type)
2944 {
2945         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2946                 RD_STATE : WR_STATE;
2947 }
2948
2949 /* 
2950  * Checks for sequence id mutating operations. 
2951  */
2952 static __be32
2953 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2954                          stateid_t *stateid, int flags,
2955                          struct nfs4_stateowner **sopp,
2956                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2957 {
2958         struct nfs4_stateid *stp;
2959         struct nfs4_stateowner *sop;
2960         struct svc_fh *current_fh = &cstate->current_fh;
2961         __be32 status;
2962
2963         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2964                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2965                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2966                 stateid->si_generation);
2967
2968         *stpp = NULL;
2969         *sopp = NULL;
2970
2971         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2972                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2973                 return nfserr_bad_stateid;
2974         }
2975
2976         if (STALE_STATEID(stateid))
2977                 return nfserr_stale_stateid;
2978
2979         if (nfsd4_has_session(cstate))
2980                 flags |= HAS_SESSION;
2981
2982         /*
2983         * We return BAD_STATEID if filehandle doesn't match stateid, 
2984         * the confirmed flag is incorrecly set, or the generation 
2985         * number is incorrect.  
2986         */
2987         stp = find_stateid(stateid, flags);
2988         if (stp == NULL) {
2989                 /*
2990                  * Also, we should make sure this isn't just the result of
2991                  * a replayed close:
2992                  */
2993                 sop = search_close_lru(stateid->si_stateownerid, flags);
2994                 if (sop == NULL)
2995                         return stateid_error_map(stateid);
2996                 *sopp = sop;
2997                 goto check_replay;
2998         }
2999
3000         *stpp = stp;
3001         *sopp = sop = stp->st_stateowner;
3002
3003         if (lock) {
3004                 clientid_t *lockclid = &lock->v.new.clientid;
3005                 struct nfs4_client *clp = sop->so_client;
3006                 int lkflg = 0;
3007                 __be32 status;
3008
3009                 lkflg = setlkflg(lock->lk_type);
3010
3011                 if (lock->lk_is_new) {
3012                         if (!sop->so_is_open_owner)
3013                                 return nfserr_bad_stateid;
3014                         if (!(flags & HAS_SESSION) &&
3015                             !same_clid(&clp->cl_clientid, lockclid))
3016                                 return nfserr_bad_stateid;
3017                         /* stp is the open stateid */
3018                         status = nfs4_check_openmode(stp, lkflg);
3019                         if (status)
3020                                 return status;
3021                 } else {
3022                         /* stp is the lock stateid */
3023                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3024                         if (status)
3025                                 return status;
3026                }
3027         }
3028
3029         if (nfs4_check_fh(current_fh, stp)) {
3030                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3031                 return nfserr_bad_stateid;
3032         }
3033
3034         /*
3035         *  We now validate the seqid and stateid generation numbers.
3036         *  For the moment, we ignore the possibility of 
3037         *  generation number wraparound.
3038         */
3039         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3040                 goto check_replay;
3041
3042         if (sop->so_confirmed && flags & CONFIRM) {
3043                 dprintk("NFSD: preprocess_seqid_op: expected"
3044                                 " unconfirmed stateowner!\n");
3045                 return nfserr_bad_stateid;
3046         }
3047         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3048                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3049                                 " confirmed yet!\n");
3050                 return nfserr_bad_stateid;
3051         }
3052         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3053         if (status)
3054                 return status;
3055         renew_client(sop->so_client);
3056         return nfs_ok;
3057
3058 check_replay:
3059         if (seqid == sop->so_seqid - 1) {
3060                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3061                 /* indicate replay to calling function */
3062                 return nfserr_replay_me;
3063         }
3064         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3065                         sop->so_seqid, seqid);
3066         *sopp = NULL;
3067         return nfserr_bad_seqid;
3068 }
3069
3070 __be32
3071 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3072                    struct nfsd4_open_confirm *oc)
3073 {
3074         __be32 status;
3075         struct nfs4_stateowner *sop;
3076         struct nfs4_stateid *stp;
3077
3078         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3079                         (int)cstate->current_fh.fh_dentry->d_name.len,
3080                         cstate->current_fh.fh_dentry->d_name.name);
3081
3082         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3083         if (status)
3084                 return status;
3085
3086         nfs4_lock_state();
3087
3088         if ((status = nfs4_preprocess_seqid_op(cstate,
3089                                         oc->oc_seqid, &oc->oc_req_stateid,
3090                                         CONFIRM | OPEN_STATE,
3091                                         &oc->oc_stateowner, &stp, NULL)))
3092                 goto out; 
3093
3094         sop = oc->oc_stateowner;
3095         sop->so_confirmed = 1;
3096         update_stateid(&stp->st_stateid);
3097         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3098         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
3099                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3100                          stp->st_stateid.si_boot,
3101                          stp->st_stateid.si_stateownerid,
3102                          stp->st_stateid.si_fileid,
3103                          stp->st_stateid.si_generation);
3104
3105         nfsd4_create_clid_dir(sop->so_client);
3106 out:
3107         if (oc->oc_stateowner) {
3108                 nfs4_get_stateowner(oc->oc_stateowner);
3109                 cstate->replay_owner = oc->oc_stateowner;
3110         }
3111         nfs4_unlock_state();
3112         return status;
3113 }
3114
3115
3116 /*
3117  * unset all bits in union bitmap (bmap) that
3118  * do not exist in share (from successful OPEN_DOWNGRADE)
3119  */
3120 static void
3121 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3122 {
3123         int i;
3124         for (i = 1; i < 4; i++) {
3125                 if ((i & access) != i)
3126                         __clear_bit(i, bmap);
3127         }
3128 }
3129
3130 static void
3131 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3132 {
3133         int i;
3134         for (i = 0; i < 4; i++) {
3135                 if ((i & deny) != i)
3136                         __clear_bit(i, bmap);
3137         }
3138 }
3139
3140 __be32
3141 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3142                      struct nfsd4_compound_state *cstate,
3143                      struct nfsd4_open_downgrade *od)
3144 {
3145         __be32 status;
3146         struct nfs4_stateid *stp;
3147         unsigned int share_access;
3148
3149         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3150                         (int)cstate->current_fh.fh_dentry->d_name.len,
3151                         cstate->current_fh.fh_dentry->d_name.name);
3152
3153         if (!access_valid(od->od_share_access, cstate->minorversion)
3154                         || !deny_valid(od->od_share_deny))
3155                 return nfserr_inval;
3156
3157         nfs4_lock_state();
3158         if ((status = nfs4_preprocess_seqid_op(cstate,
3159                                         od->od_seqid,
3160                                         &od->od_stateid, 
3161                                         OPEN_STATE,
3162                                         &od->od_stateowner, &stp, NULL)))
3163                 goto out; 
3164
3165         status = nfserr_inval;
3166         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3167                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3168                         stp->st_access_bmap, od->od_share_access);
3169                 goto out;
3170         }
3171         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3172                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3173                         stp->st_deny_bmap, od->od_share_deny);
3174                 goto out;
3175         }
3176         set_access(&share_access, stp->st_access_bmap);
3177         nfs4_file_downgrade(stp->st_vfs_file,
3178                             share_access & ~od->od_share_access);
3179
3180         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3181         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3182
3183         update_stateid(&stp->st_stateid);
3184         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3185         status = nfs_ok;
3186 out:
3187         if (od->od_stateowner) {
3188                 nfs4_get_stateowner(od->od_stateowner);
3189                 cstate->replay_owner = od->od_stateowner;
3190         }
3191         nfs4_unlock_state();
3192         return status;
3193 }
3194
3195 /*
3196  * nfs4_unlock_state() called after encode
3197  */
3198 __be32
3199 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3200             struct nfsd4_close *close)
3201 {
3202         __be32 status;
3203         struct nfs4_stateid *stp;
3204
3205         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3206                         (int)cstate->current_fh.fh_dentry->d_name.len,
3207                         cstate->current_fh.fh_dentry->d_name.name);
3208
3209         nfs4_lock_state();
3210         /* check close_lru for replay */
3211         if ((status = nfs4_preprocess_seqid_op(cstate,
3212                                         close->cl_seqid,
3213                                         &close->cl_stateid, 
3214                                         OPEN_STATE | CLOSE_STATE,
3215                                         &close->cl_stateowner, &stp, NULL)))
3216                 goto out; 
3217         status = nfs_ok;
3218         update_stateid(&stp->st_stateid);
3219         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3220
3221         /* release_stateid() calls nfsd_close() if needed */
3222         release_open_stateid(stp);
3223
3224         /* place unused nfs4_stateowners on so_close_lru list to be
3225          * released by the laundromat service after the lease period
3226          * to enable us to handle CLOSE replay
3227          */
3228         if (list_empty(&close->cl_stateowner->so_stateids))
3229                 move_to_close_lru(close->cl_stateowner);
3230 out:
3231         if (close->cl_stateowner) {
3232                 nfs4_get_stateowner(close->cl_stateowner);
3233                 cstate->replay_owner = close->cl_stateowner;
3234         }
3235         nfs4_unlock_state();
3236         return status;
3237 }
3238
3239 __be32
3240 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3241                   struct nfsd4_delegreturn *dr)
3242 {
3243         struct nfs4_delegation *dp;
3244         stateid_t *stateid = &dr->dr_stateid;
3245         struct inode *inode;
3246         __be32 status;
3247         int flags = 0;
3248
3249         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3250                 return status;
3251         inode = cstate->current_fh.fh_dentry->d_inode;
3252
3253         if (nfsd4_has_session(cstate))
3254                 flags |= HAS_SESSION;
3255         nfs4_lock_state();
3256         status = nfserr_bad_stateid;
3257         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3258                 goto out;
3259         status = nfserr_stale_stateid;
3260         if (STALE_STATEID(stateid))
3261                 goto out;
3262         status = nfserr_bad_stateid;
3263         if (!is_delegation_stateid(stateid))
3264                 goto out;
3265         dp = find_delegation_stateid(inode, stateid);
3266         if (!dp) {
3267                 status = stateid_error_map(stateid);
3268                 goto out;
3269         }
3270         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3271         if (status)
3272                 goto out;
3273         renew_client(dp->dl_client);
3274
3275         unhash_delegation(dp);
3276 out:
3277         nfs4_unlock_state();
3278
3279         return status;
3280 }
3281
3282
3283 /* 
3284  * Lock owner state (byte-range locks)
3285  */
3286 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3287 #define LOCK_HASH_BITS              8
3288 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3289 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3290
3291 static inline u64
3292 end_offset(u64 start, u64 len)
3293 {
3294         u64 end;
3295
3296         end = start + len;
3297         return end >= start ? end: NFS4_MAX_UINT64;
3298 }
3299
3300 /* last octet in a range */
3301 static inline u64
3302 last_byte_offset(u64 start, u64 len)
3303 {
3304         u64 end;
3305
3306         BUG_ON(!len);
3307         end = start + len;
3308         return end > start ? end - 1: NFS4_MAX_UINT64;
3309 }
3310
3311 #define lockownerid_hashval(id) \
3312         ((id) & LOCK_HASH_MASK)
3313
3314 static inline unsigned int
3315 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3316                 struct xdr_netobj *ownername)
3317 {
3318         return (file_hashval(inode) + cl_id
3319                         + opaque_hashval(ownername->data, ownername->len))
3320                 & LOCK_HASH_MASK;
3321 }
3322
3323 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3324 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3325 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3326
3327 static struct nfs4_stateid *
3328 find_stateid(stateid_t *stid, int flags)
3329 {
3330         struct nfs4_stateid *local;
3331         u32 st_id = stid->si_stateownerid;
3332         u32 f_id = stid->si_fileid;
3333         unsigned int hashval;
3334
3335         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3336         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3337                 hashval = stateid_hashval(st_id, f_id);
3338                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3339                         if ((local->st_stateid.si_stateownerid == st_id) &&
3340                             (local->st_stateid.si_fileid == f_id))
3341                                 return local;
3342                 }
3343         } 
3344
3345         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3346                 hashval = stateid_hashval(st_id, f_id);
3347                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3348                         if ((local->st_stateid.si_stateownerid == st_id) &&
3349                             (local->st_stateid.si_fileid == f_id))
3350                                 return local;
3351                 }
3352         }
3353         return NULL;
3354 }
3355
3356 static struct nfs4_delegation *
3357 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3358 {
3359         struct nfs4_file *fp;
3360         struct nfs4_delegation *dl;
3361
3362         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3363                     stid->si_boot, stid->si_stateownerid,
3364                     stid->si_fileid, stid->si_generation);
3365
3366         fp = find_file(ino);
3367         if (!fp)
3368                 return NULL;
3369         dl = find_delegation_file(fp, stid);
3370         put_nfs4_file(fp);
3371         return dl;
3372 }
3373
3374 /*
3375  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3376  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3377  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3378  * locking, this prevents us from being completely protocol-compliant.  The
3379  * real solution to this problem is to start using unsigned file offsets in
3380  * the VFS, but this is a very deep change!
3381  */
3382 static inline void
3383 nfs4_transform_lock_offset(struct file_lock *lock)
3384 {
3385         if (lock->fl_start < 0)
3386                 lock->fl_start = OFFSET_MAX;
3387         if (lock->fl_end < 0)
3388                 lock->fl_end = OFFSET_MAX;
3389 }
3390
3391 /* Hack!: For now, we're defining this just so we can use a pointer to it
3392  * as a unique cookie to identify our (NFSv4's) posix locks. */
3393 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3394 };
3395
3396 static inline void
3397 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3398 {
3399         struct nfs4_stateowner *sop;
3400         unsigned int hval;
3401
3402         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3403                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3404                 hval = lockownerid_hashval(sop->so_id);
3405                 kref_get(&sop->so_ref);
3406                 deny->ld_sop = sop;
3407                 deny->ld_clientid = sop->so_client->cl_clientid;
3408         } else {
3409                 deny->ld_sop = NULL;
3410                 deny->ld_clientid.cl_boot = 0;
3411                 deny->ld_clientid.cl_id = 0;
3412         }
3413         deny->ld_start = fl->fl_start;
3414         deny->ld_length = NFS4_MAX_UINT64;
3415         if (fl->fl_end != NFS4_MAX_UINT64)
3416                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3417         deny->ld_type = NFS4_READ_LT;
3418         if (fl->fl_type != F_RDLCK)
3419                 deny->ld_type = NFS4_WRITE_LT;
3420 }
3421
3422 static struct nfs4_stateowner *
3423 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3424                 struct xdr_netobj *owner)
3425 {
3426         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3427         struct nfs4_stateowner *op;
3428
3429         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3430                 if (same_owner_str(op, owner, clid))
3431                         return op;
3432         }
3433         return NULL;
3434 }
3435
3436 /*
3437  * Alloc a lock owner structure.
3438  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3439  * occured. 
3440  *
3441  * strhashval = lock_ownerstr_hashval 
3442  */
3443
3444 static struct nfs4_stateowner *
3445 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3446         struct nfs4_stateowner *sop;
3447         struct nfs4_replay *rp;
3448         unsigned int idhashval;
3449
3450         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3451                 return NULL;
3452         idhashval = lockownerid_hashval(current_ownerid);
3453         INIT_LIST_HEAD(&sop->so_idhash);
3454         INIT_LIST_HEAD(&sop->so_strhash);
3455         INIT_LIST_HEAD(&sop->so_perclient);
3456         INIT_LIST_HEAD(&sop->so_stateids);
3457         INIT_LIST_HEAD(&sop->so_perstateid);
3458         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3459         sop->so_time = 0;
3460         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3461         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3462         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3463         sop->so_is_open_owner = 0;
3464         sop->so_id = current_ownerid++;
3465         sop->so_client = clp;
3466         /* It is the openowner seqid that will be incremented in encode in the
3467          * case of new lockowners; so increment the lock seqid manually: */
3468         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3469         sop->so_confirmed = 1;
3470         rp = &sop->so_replay;
3471         rp->rp_status = nfserr_serverfault;
3472         rp->rp_buflen = 0;
3473         rp->rp_buf = rp->rp_ibuf;
3474         return sop;
3475 }
3476
3477 static struct nfs4_stateid *
3478 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3479 {
3480         struct nfs4_stateid *stp;
3481         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3482
3483         stp = nfs4_alloc_stateid();
3484         if (stp == NULL)
3485                 goto out;
3486         INIT_LIST_HEAD(&stp->st_hash);
3487         INIT_LIST_HEAD(&stp->st_perfile);
3488         INIT_LIST_HEAD(&stp->st_perstateowner);
3489         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3490         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3491         list_add(&stp->st_perfile, &fp->fi_stateids);
3492         list_add(&stp->st_perstateowner, &sop->so_stateids);
3493         stp->st_stateowner = sop;
3494         get_nfs4_file(fp);
3495         stp->st_file = fp;
3496         stp->st_stateid.si_boot = get_seconds();
3497         stp->st_stateid.si_stateownerid = sop->so_id;
3498         stp->st_stateid.si_fileid = fp->fi_id;
3499         stp->st_stateid.si_generation = 0;
3500         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3501         stp->st_access_bmap = open_stp->st_access_bmap;
3502         stp->st_deny_bmap = open_stp->st_deny_bmap;
3503         stp->st_openstp = open_stp;
3504
3505 out:
3506         return stp;
3507 }
3508
3509 static int
3510 check_lock_length(u64 offset, u64 length)
3511 {
3512         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3513              LOFF_OVERFLOW(offset, length)));
3514 }
3515
3516 /*
3517  *  LOCK operation 
3518  */
3519 __be32
3520 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3521            struct nfsd4_lock *lock)
3522 {
3523         struct nfs4_stateowner *open_sop = NULL;
3524         struct nfs4_stateowner *lock_sop = NULL;
3525         struct nfs4_stateid *lock_stp;
3526         struct file *filp;
3527         struct file_lock file_lock;
3528         struct file_lock conflock;
3529         __be32 status = 0;
3530         unsigned int strhashval;
3531         unsigned int cmd;
3532         int err;
3533
3534         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3535                 (long long) lock->lk_offset,
3536                 (long long) lock->lk_length);
3537
3538         if (check_lock_length(lock->lk_offset, lock->lk_length))
3539                  return nfserr_inval;
3540
3541         if ((status = fh_verify(rqstp, &cstate->current_fh,
3542                                 S_IFREG, NFSD_MAY_LOCK))) {
3543                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3544                 return status;
3545         }
3546
3547         nfs4_lock_state();
3548
3549         if (lock->lk_is_new) {
3550                 /*
3551                  * Client indicates that this is a new lockowner.
3552                  * Use open owner and open stateid to create lock owner and
3553                  * lock stateid.
3554                  */
3555                 struct nfs4_stateid *open_stp = NULL;
3556                 struct nfs4_file *fp;
3557                 
3558                 status = nfserr_stale_clientid;
3559                 if (!nfsd4_has_session(cstate) &&
3560                     STALE_CLIENTID(&lock->lk_new_clientid))
3561                         goto out;
3562
3563                 /* validate and update open stateid and open seqid */
3564                 status = nfs4_preprocess_seqid_op(cstate,
3565                                         lock->lk_new_open_seqid,
3566                                         &lock->lk_new_open_stateid,
3567                                         OPEN_STATE,
3568                                         &lock->lk_replay_owner, &open_stp,
3569                                         lock);
3570                 if (status)
3571                         goto out;
3572                 open_sop = lock->lk_replay_owner;
3573                 /* create lockowner and lock stateid */
3574                 fp = open_stp->st_file;
3575                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3576                                 open_sop->so_client->cl_clientid.cl_id, 
3577                                 &lock->v.new.owner);
3578                 /* XXX: Do we need to check for duplicate stateowners on
3579                  * the same file, or should they just be allowed (and
3580                  * create new stateids)? */
3581                 status = nfserr_resource;
3582                 lock_sop = alloc_init_lock_stateowner(strhashval,
3583                                 open_sop->so_client, open_stp, lock);
3584                 if (lock_sop == NULL)
3585                         goto out;
3586                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3587                 if (lock_stp == NULL)
3588                         goto out;
3589         } else {
3590                 /* lock (lock owner + lock stateid) already exists */
3591                 status = nfs4_preprocess_seqid_op(cstate,
3592                                        lock->lk_old_lock_seqid, 
3593                                        &lock->lk_old_lock_stateid, 
3594                                        LOCK_STATE,
3595                                        &lock->lk_replay_owner, &lock_stp, lock);
3596                 if (status)
3597                         goto out;
3598                 lock_sop = lock->lk_replay_owner;
3599         }
3600         /* lock->lk_replay_owner and lock_stp have been created or found */
3601         filp = lock_stp->st_vfs_file;
3602
3603         status = nfserr_grace;
3604         if (locks_in_grace() && !lock->lk_reclaim)
3605                 goto out;
3606         status = nfserr_no_grace;
3607         if (!locks_in_grace() && lock->lk_reclaim)
3608                 goto out;
3609
3610         locks_init_lock(&file_lock);
3611         switch (lock->lk_type) {
3612                 case NFS4_READ_LT:
3613                 case NFS4_READW_LT:
3614                         file_lock.fl_type = F_RDLCK;
3615                         cmd = F_SETLK;
3616                 break;
3617                 case NFS4_WRITE_LT:
3618                 case NFS4_WRITEW_LT:
3619                         file_lock.fl_type = F_WRLCK;
3620                         cmd = F_SETLK;
3621                 break;
3622                 default:
3623                         status = nfserr_inval;
3624                 goto out;
3625         }
3626         file_lock.fl_owner = (fl_owner_t)lock_sop;
3627         file_lock.fl_pid = current->tgid;
3628         file_lock.fl_file = filp;
3629         file_lock.fl_flags = FL_POSIX;
3630         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3631
3632         file_lock.fl_start = lock->lk_offset;
3633         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3634         nfs4_transform_lock_offset(&file_lock);
3635
3636         /*
3637         * Try to lock the file in the VFS.
3638         * Note: locks.c uses the BKL to protect the inode's lock list.
3639         */
3640
3641         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3642         switch (-err) {
3643         case 0: /* success! */
3644                 update_stateid(&lock_stp->st_stateid);
3645                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3646                                 sizeof(stateid_t));
3647                 status = 0;
3648                 break;
3649         case (EAGAIN):          /* conflock holds conflicting lock */
3650                 status = nfserr_denied;
3651                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3652                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3653                 break;
3654         case (EDEADLK):
3655                 status = nfserr_deadlock;
3656                 break;
3657         default:        
3658                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3659                 status = nfserr_resource;
3660                 break;
3661         }
3662 out:
3663         if (status && lock->lk_is_new && lock_sop)
3664                 release_lockowner(lock_sop);
3665         if (lock->lk_replay_owner) {
3666                 nfs4_get_stateowner(lock->lk_replay_owner);
3667                 cstate->replay_owner = lock->lk_replay_owner;
3668         }
3669         nfs4_unlock_state();
3670         return status;
3671 }
3672
3673 /*
3674  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3675  * so we do a temporary open here just to get an open file to pass to
3676  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3677  * inode operation.)
3678  */
3679 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3680 {
3681         struct file *file;
3682         int err;
3683
3684         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3685         if (err)
3686                 return err;
3687         err = vfs_test_lock(file, lock);
3688         nfsd_close(file);
3689         return err;
3690 }
3691
3692 /*
3693  * LOCKT operation
3694  */
3695 __be32
3696 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3697             struct nfsd4_lockt *lockt)
3698 {
3699         struct inode *inode;
3700         struct file_lock file_lock;
3701         int error;
3702         __be32 status;
3703
3704         if (locks_in_grace())
3705                 return nfserr_grace;
3706
3707         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3708                  return nfserr_inval;
3709
3710         lockt->lt_stateowner = NULL;
3711         nfs4_lock_state();
3712
3713         status = nfserr_stale_clientid;
3714         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3715                 goto out;
3716
3717         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3718                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3719                 if (status == nfserr_symlink)
3720                         status = nfserr_inval;
3721                 goto out;
3722         }
3723
3724         inode = cstate->current_fh.fh_dentry->d_inode;
3725         locks_init_lock(&file_lock);
3726         switch (lockt->lt_type) {
3727                 case NFS4_READ_LT:
3728                 case NFS4_READW_LT:
3729                         file_lock.fl_type = F_RDLCK;
3730                 break;
3731                 case NFS4_WRITE_LT:
3732                 case NFS4_WRITEW_LT:
3733                         file_lock.fl_type = F_WRLCK;
3734                 break;
3735                 default:
3736                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3737                         status = nfserr_inval;
3738                 goto out;
3739         }
3740
3741         lockt->lt_stateowner = find_lockstateowner_str(inode,
3742                         &lockt->lt_clientid, &lockt->lt_owner);
3743         if (lockt->lt_stateowner)
3744                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3745         file_lock.fl_pid = current->tgid;
3746         file_lock.fl_flags = FL_POSIX;
3747
3748         file_lock.fl_start = lockt->lt_offset;
3749         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3750
3751         nfs4_transform_lock_offset(&file_lock);
3752
3753         status = nfs_ok;
3754         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3755         if (error) {
3756                 status = nfserrno(error);
3757                 goto out;
3758         }
3759         if (file_lock.fl_type != F_UNLCK) {
3760                 status = nfserr_denied;
3761                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3762         }
3763 out:
3764         nfs4_unlock_state();
3765         return status;
3766 }
3767
3768 __be32
3769 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3770             struct nfsd4_locku *locku)
3771 {
3772         struct nfs4_stateid *stp;
3773         struct file *filp = NULL;
3774         struct file_lock file_lock;
3775         __be32 status;
3776         int err;
3777                                                         
3778         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3779                 (long long) locku->lu_offset,
3780                 (long long) locku->lu_length);
3781
3782         if (check_lock_length(locku->lu_offset, locku->lu_length))
3783                  return nfserr_inval;
3784
3785         nfs4_lock_state();
3786                                                                                 
3787         if ((status = nfs4_preprocess_seqid_op(cstate,
3788                                         locku->lu_seqid, 
3789                                         &locku->lu_stateid, 
3790                                         LOCK_STATE,
3791                                         &locku->lu_stateowner, &stp, NULL)))
3792                 goto out;
3793
3794         filp = stp->st_vfs_file;
3795         BUG_ON(!filp);
3796         locks_init_lock(&file_lock);
3797         file_lock.fl_type = F_UNLCK;
3798         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3799         file_lock.fl_pid = current->tgid;
3800         file_lock.fl_file = filp;
3801         file_lock.fl_flags = FL_POSIX; 
3802         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3803         file_lock.fl_start = locku->lu_offset;
3804
3805         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3806         nfs4_transform_lock_offset(&file_lock);
3807
3808         /*
3809         *  Try to unlock the file in the VFS.
3810         */
3811         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3812         if (err) {
3813                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3814                 goto out_nfserr;
3815         }
3816         /*
3817         * OK, unlock succeeded; the only thing left to do is update the stateid.
3818         */
3819         update_stateid(&stp->st_stateid);
3820         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3821
3822 out:
3823         if (locku->lu_stateowner) {
3824                 nfs4_get_stateowner(locku->lu_stateowner);
3825                 cstate->replay_owner = locku->lu_stateowner;
3826         }
3827         nfs4_unlock_state();
3828         return status;
3829
3830 out_nfserr:
3831         status = nfserrno(err);
3832         goto out;
3833 }
3834
3835 /*
3836  * returns
3837  *      1: locks held by lockowner
3838  *      0: no locks held by lockowner
3839  */
3840 static int
3841 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3842 {
3843         struct file_lock **flpp;
3844         struct inode *inode = filp->f_path.dentry->d_inode;
3845         int status = 0;
3846
3847         lock_kernel();
3848         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3849                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3850                         status = 1;
3851                         goto out;
3852                 }
3853         }
3854 out:
3855         unlock_kernel();
3856         return status;
3857 }
3858
3859 __be32
3860 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3861                         struct nfsd4_compound_state *cstate,
3862                         struct nfsd4_release_lockowner *rlockowner)
3863 {
3864         clientid_t *clid = &rlockowner->rl_clientid;
3865         struct nfs4_stateowner *sop;
3866         struct nfs4_stateid *stp;
3867         struct xdr_netobj *owner = &rlockowner->rl_owner;
3868         struct list_head matches;
3869         int i;
3870         __be32 status;
3871
3872         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3873                 clid->cl_boot, clid->cl_id);
3874
3875         /* XXX check for lease expiration */
3876
3877         status = nfserr_stale_clientid;
3878         if (STALE_CLIENTID(clid))
3879                 return status;
3880
3881         nfs4_lock_state();
3882
3883         status = nfserr_locks_held;
3884         /* XXX: we're doing a linear search through all the lockowners.
3885          * Yipes!  For now we'll just hope clients aren't really using
3886          * release_lockowner much, but eventually we have to fix these
3887          * data structures. */
3888         INIT_LIST_HEAD(&matches);
3889         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3890                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3891                         if (!same_owner_str(sop, owner, clid))
3892                                 continue;
3893                         list_for_each_entry(stp, &sop->so_stateids,
3894                                         st_perstateowner) {
3895                                 if (check_for_locks(stp->st_vfs_file, sop))
3896                                         goto out;
3897                                 /* Note: so_perclient unused for lockowners,
3898                                  * so it's OK to fool with here. */
3899                                 list_add(&sop->so_perclient, &matches);
3900                         }
3901                 }
3902         }
3903         /* Clients probably won't expect us to return with some (but not all)
3904          * of the lockowner state released; so don't release any until all
3905          * have been checked. */
3906         status = nfs_ok;
3907         while (!list_empty(&matches)) {
3908                 sop = list_entry(matches.next, struct nfs4_stateowner,
3909                                                                 so_perclient);
3910                 /* unhash_stateowner deletes so_perclient only
3911                  * for openowners. */
3912                 list_del(&sop->so_perclient);
3913                 release_lockowner(sop);
3914         }
3915 out:
3916         nfs4_unlock_state();
3917         return status;
3918 }
3919
3920 static inline struct nfs4_client_reclaim *
3921 alloc_reclaim(void)
3922 {
3923         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3924 }
3925
3926 int
3927 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3928 {
3929         unsigned int strhashval = clientstr_hashval(name);
3930         struct nfs4_client *clp;
3931
3932         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3933         return clp ? 1 : 0;
3934 }
3935
3936 /*
3937  * failure => all reset bets are off, nfserr_no_grace...
3938  */
3939 int
3940 nfs4_client_to_reclaim(const char *name)
3941 {
3942         unsigned int strhashval;
3943         struct nfs4_client_reclaim *crp = NULL;
3944
3945         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3946         crp = alloc_reclaim();
3947         if (!crp)
3948                 return 0;
3949         strhashval = clientstr_hashval(name);
3950         INIT_LIST_HEAD(&crp->cr_strhash);
3951         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3952         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3953         reclaim_str_hashtbl_size++;
3954         return 1;
3955 }
3956
3957 static void
3958 nfs4_release_reclaim(void)
3959 {
3960         struct nfs4_client_reclaim *crp = NULL;
3961         int i;
3962
3963         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3964                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3965                         crp = list_entry(reclaim_str_hashtbl[i].next,
3966                                         struct nfs4_client_reclaim, cr_strhash);
3967                         list_del(&crp->cr_strhash);
3968                         kfree(crp);
3969                         reclaim_str_hashtbl_size--;
3970                 }
3971         }
3972         BUG_ON(reclaim_str_hashtbl_size);
3973 }
3974
3975 /*
3976  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3977 static struct nfs4_client_reclaim *
3978 nfs4_find_reclaim_client(clientid_t *clid)
3979 {
3980         unsigned int strhashval;
3981         struct nfs4_client *clp;
3982         struct nfs4_client_reclaim *crp = NULL;
3983
3984
3985         /* find clientid in conf_id_hashtbl */
3986         clp = find_confirmed_client(clid);
3987         if (clp == NULL)
3988                 return NULL;
3989
3990         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3991                             clp->cl_name.len, clp->cl_name.data,
3992                             clp->cl_recdir);
3993
3994         /* find clp->cl_name in reclaim_str_hashtbl */
3995         strhashval = clientstr_hashval(clp->cl_recdir);
3996         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3997                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3998                         return crp;
3999                 }
4000         }
4001         return NULL;
4002 }
4003
4004 /*
4005 * Called from OPEN. Look for clientid in reclaim list.
4006 */
4007 __be32
4008 nfs4_check_open_reclaim(clientid_t *clid)
4009 {
4010         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4011 }
4012
4013 /* initialization to perform at module load time: */
4014
4015 int
4016 nfs4_state_init(void)
4017 {
4018         int i, status;
4019
4020         status = nfsd4_init_slabs();
4021         if (status)
4022                 return status;
4023         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4024                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4025                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4026                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4027                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4028         }
4029         for (i = 0; i < SESSION_HASH_SIZE; i++)
4030                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4031         for (i = 0; i < FILE_HASH_SIZE; i++) {
4032                 INIT_LIST_HEAD(&file_hashtbl[i]);
4033         }
4034         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4035                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4036                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4037         }
4038         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4039                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4040                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4041         }
4042         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4043                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4044                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4045         }
4046         memset(&onestateid, ~0, sizeof(stateid_t));
4047         INIT_LIST_HEAD(&close_lru);
4048         INIT_LIST_HEAD(&client_lru);
4049         INIT_LIST_HEAD(&del_recall_lru);
4050         for (i = 0; i < CLIENT_HASH_SIZE; i++)
4051                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4052         reclaim_str_hashtbl_size = 0;
4053         return 0;
4054 }
4055
4056 static void
4057 nfsd4_load_reboot_recovery_data(void)
4058 {
4059         int status;
4060
4061         nfs4_lock_state();
4062         nfsd4_init_recdir(user_recovery_dirname);
4063         status = nfsd4_recdir_load();
4064         nfs4_unlock_state();
4065         if (status)
4066                 printk("NFSD: Failure reading reboot recovery data\n");
4067 }
4068
4069 unsigned long
4070 get_nfs4_grace_period(void)
4071 {
4072         return max(user_lease_time, lease_time) * HZ;