Merge branch 'for-2.6.31' of git://fieldses.org/git/linux-nfsd
[linux-2.6.git] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58
59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
60
61 /* Globals */
62 static time_t lease_time = 90;     /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
68 static u32 nfs4_init;
69 static stateid_t zerostateid;             /* bits all 0 */
70 static stateid_t onestateid;              /* bits all 1 */
71 static u64 current_sessionid = 1;
72
73 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
74 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
75
76 /* forward declarations */
77 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
78 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
81
82 /* Locking: */
83
84 /* Currently used for almost all code touching nfsv4 state: */
85 static DEFINE_MUTEX(client_mutex);
86
87 /*
88  * Currently used for the del_recall_lru and file hash table.  In an
89  * effort to decrease the scope of the client_mutex, this spinlock may
90  * eventually cover more:
91  */
92 static DEFINE_SPINLOCK(recall_lock);
93
94 static struct kmem_cache *stateowner_slab = NULL;
95 static struct kmem_cache *file_slab = NULL;
96 static struct kmem_cache *stateid_slab = NULL;
97 static struct kmem_cache *deleg_slab = NULL;
98
99 void
100 nfs4_lock_state(void)
101 {
102         mutex_lock(&client_mutex);
103 }
104
105 void
106 nfs4_unlock_state(void)
107 {
108         mutex_unlock(&client_mutex);
109 }
110
111 static inline u32
112 opaque_hashval(const void *ptr, int nbytes)
113 {
114         unsigned char *cptr = (unsigned char *) ptr;
115
116         u32 x = 0;
117         while (nbytes--) {
118                 x *= 37;
119                 x += *cptr++;
120         }
121         return x;
122 }
123
124 static struct list_head del_recall_lru;
125
126 static inline void
127 put_nfs4_file(struct nfs4_file *fi)
128 {
129         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
130                 list_del(&fi->fi_hash);
131                 spin_unlock(&recall_lock);
132                 iput(fi->fi_inode);
133                 kmem_cache_free(file_slab, fi);
134         }
135 }
136
137 static inline void
138 get_nfs4_file(struct nfs4_file *fi)
139 {
140         atomic_inc(&fi->fi_ref);
141 }
142
143 static int num_delegations;
144 unsigned int max_delegations;
145
146 /*
147  * Open owner state (share locks)
148  */
149
150 /* hash tables for nfs4_stateowner */
151 #define OWNER_HASH_BITS              8
152 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
153 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
154
155 #define ownerid_hashval(id) \
156         ((id) & OWNER_HASH_MASK)
157 #define ownerstr_hashval(clientid, ownername) \
158         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
159
160 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
161 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
162
163 /* hash table for nfs4_file */
164 #define FILE_HASH_BITS                   8
165 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
166 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
167 /* hash table for (open)nfs4_stateid */
168 #define STATEID_HASH_BITS              10
169 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
170 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
171
172 #define file_hashval(x) \
173         hash_ptr(x, FILE_HASH_BITS)
174 #define stateid_hashval(owner_id, file_id)  \
175         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
176
177 static struct list_head file_hashtbl[FILE_HASH_SIZE];
178 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
179
180 static struct nfs4_delegation *
181 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
182 {
183         struct nfs4_delegation *dp;
184         struct nfs4_file *fp = stp->st_file;
185         struct nfs4_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_stateid.si_boot = get_seconds();
208         dp->dl_stateid.si_stateownerid = current_delegid++;
209         dp->dl_stateid.si_fileid = 0;
210         dp->dl_stateid.si_generation = 0;
211         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
212         dp->dl_time = 0;
213         atomic_set(&dp->dl_count, 1);
214         list_add(&dp->dl_perfile, &fp->fi_delegations);
215         list_add(&dp->dl_perclnt, &clp->cl_delegations);
216         return dp;
217 }
218
219 void
220 nfs4_put_delegation(struct nfs4_delegation *dp)
221 {
222         if (atomic_dec_and_test(&dp->dl_count)) {
223                 dprintk("NFSD: freeing dp %p\n",dp);
224                 put_nfs4_file(dp->dl_file);
225                 kmem_cache_free(deleg_slab, dp);
226                 num_delegations--;
227         }
228 }
229
230 /* Remove the associated file_lock first, then remove the delegation.
231  * lease_modify() is called to remove the FS_LEASE file_lock from
232  * the i_flock list, eventually calling nfsd's lock_manager
233  * fl_release_callback.
234  */
235 static void
236 nfs4_close_delegation(struct nfs4_delegation *dp)
237 {
238         struct file *filp = dp->dl_vfs_file;
239
240         dprintk("NFSD: close_delegation dp %p\n",dp);
241         dp->dl_vfs_file = NULL;
242         /* The following nfsd_close may not actually close the file,
243          * but we want to remove the lease in any case. */
244         if (dp->dl_flock)
245                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
246         nfsd_close(filp);
247 }
248
249 /* Called under the state lock. */
250 static void
251 unhash_delegation(struct nfs4_delegation *dp)
252 {
253         list_del_init(&dp->dl_perfile);
254         list_del_init(&dp->dl_perclnt);
255         spin_lock(&recall_lock);
256         list_del_init(&dp->dl_recall_lru);
257         spin_unlock(&recall_lock);
258         nfs4_close_delegation(dp);
259         nfs4_put_delegation(dp);
260 }
261
262 /* 
263  * SETCLIENTID state 
264  */
265
266 /* Hash tables for nfs4_clientid state */
267 #define CLIENT_HASH_BITS                 4
268 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
269 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
270
271 #define clientid_hashval(id) \
272         ((id) & CLIENT_HASH_MASK)
273 #define clientstr_hashval(name) \
274         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
275 /*
276  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
277  * used in reboot/reset lease grace period processing
278  *
279  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
280  * setclientid_confirmed info. 
281  *
282  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
283  * setclientid info.
284  *
285  * client_lru holds client queue ordered by nfs4_client.cl_time
286  * for lease renewal.
287  *
288  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
289  * for last close replay.
290  */
291 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
292 static int reclaim_str_hashtbl_size = 0;
293 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
294 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head client_lru;
298 static struct list_head close_lru;
299
300 static void unhash_generic_stateid(struct nfs4_stateid *stp)
301 {
302         list_del(&stp->st_hash);
303         list_del(&stp->st_perfile);
304         list_del(&stp->st_perstateowner);
305 }
306
307 static void free_generic_stateid(struct nfs4_stateid *stp)
308 {
309         put_nfs4_file(stp->st_file);
310         kmem_cache_free(stateid_slab, stp);
311 }
312
313 static void release_lock_stateid(struct nfs4_stateid *stp)
314 {
315         unhash_generic_stateid(stp);
316         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
317         free_generic_stateid(stp);
318 }
319
320 static void unhash_lockowner(struct nfs4_stateowner *sop)
321 {
322         struct nfs4_stateid *stp;
323
324         list_del(&sop->so_idhash);
325         list_del(&sop->so_strhash);
326         list_del(&sop->so_perstateid);
327         while (!list_empty(&sop->so_stateids)) {
328                 stp = list_first_entry(&sop->so_stateids,
329                                 struct nfs4_stateid, st_perstateowner);
330                 release_lock_stateid(stp);
331         }
332 }
333
334 static void release_lockowner(struct nfs4_stateowner *sop)
335 {
336         unhash_lockowner(sop);
337         nfs4_put_stateowner(sop);
338 }
339
340 static void
341 release_stateid_lockowners(struct nfs4_stateid *open_stp)
342 {
343         struct nfs4_stateowner *lock_sop;
344
345         while (!list_empty(&open_stp->st_lockowners)) {
346                 lock_sop = list_entry(open_stp->st_lockowners.next,
347                                 struct nfs4_stateowner, so_perstateid);
348                 /* list_del(&open_stp->st_lockowners);  */
349                 BUG_ON(lock_sop->so_is_open_owner);
350                 release_lockowner(lock_sop);
351         }
352 }
353
354 static void release_open_stateid(struct nfs4_stateid *stp)
355 {
356         unhash_generic_stateid(stp);
357         release_stateid_lockowners(stp);
358         nfsd_close(stp->st_vfs_file);
359         free_generic_stateid(stp);
360 }
361
362 static void unhash_openowner(struct nfs4_stateowner *sop)
363 {
364         struct nfs4_stateid *stp;
365
366         list_del(&sop->so_idhash);
367         list_del(&sop->so_strhash);
368         list_del(&sop->so_perclient);
369         list_del(&sop->so_perstateid); /* XXX: necessary? */
370         while (!list_empty(&sop->so_stateids)) {
371                 stp = list_first_entry(&sop->so_stateids,
372                                 struct nfs4_stateid, st_perstateowner);
373                 release_open_stateid(stp);
374         }
375 }
376
377 static void release_openowner(struct nfs4_stateowner *sop)
378 {
379         unhash_openowner(sop);
380         list_del(&sop->so_close_lru);
381         nfs4_put_stateowner(sop);
382 }
383
384 static DEFINE_SPINLOCK(sessionid_lock);
385 #define SESSION_HASH_SIZE       512
386 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
387
388 static inline int
389 hash_sessionid(struct nfs4_sessionid *sessionid)
390 {
391         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
392
393         return sid->sequence % SESSION_HASH_SIZE;
394 }
395
396 static inline void
397 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
398 {
399         u32 *ptr = (u32 *)(&sessionid->data[0]);
400         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
401 }
402
403 static void
404 gen_sessionid(struct nfsd4_session *ses)
405 {
406         struct nfs4_client *clp = ses->se_client;
407         struct nfsd4_sessionid *sid;
408
409         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
410         sid->clientid = clp->cl_clientid;
411         sid->sequence = current_sessionid++;
412         sid->reserved = 0;
413 }
414
415 /*
416  * Give the client the number of slots it requests bound by
417  * NFSD_MAX_SLOTS_PER_SESSION and by sv_drc_max_pages.
418  *
419  * If we run out of pages (sv_drc_pages_used == sv_drc_max_pages) we
420  * should (up to a point) re-negotiate active sessions and reduce their
421  * slot usage to make rooom for new connections. For now we just fail the
422  * create session.
423  */
424 static int set_forechannel_maxreqs(struct nfsd4_channel_attrs *fchan)
425 {
426         int status = 0, np = fchan->maxreqs * NFSD_PAGES_PER_SLOT;
427
428         if (fchan->maxreqs < 1)
429                 return nfserr_inval;
430         else if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
431                 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
432
433         spin_lock(&nfsd_serv->sv_lock);
434         if (np + nfsd_serv->sv_drc_pages_used > nfsd_serv->sv_drc_max_pages)
435                 np = nfsd_serv->sv_drc_max_pages - nfsd_serv->sv_drc_pages_used;
436         nfsd_serv->sv_drc_pages_used += np;
437         spin_unlock(&nfsd_serv->sv_lock);
438
439         if (np <= 0) {
440                 status = nfserr_resource;
441                 fchan->maxreqs = 0;
442         } else
443                 fchan->maxreqs = np / NFSD_PAGES_PER_SLOT;
444
445         return status;
446 }
447
448 /*
449  * fchan holds the client values on input, and the server values on output
450  */
451 static int init_forechannel_attrs(struct svc_rqst *rqstp,
452                                   struct nfsd4_channel_attrs *session_fchan,
453                                   struct nfsd4_channel_attrs *fchan)
454 {
455         int status = 0;
456         __u32   maxcount = svc_max_payload(rqstp);
457
458         /* headerpadsz set to zero in encode routine */
459
460         /* Use the client's max request and max response size if possible */
461         if (fchan->maxreq_sz > maxcount)
462                 fchan->maxreq_sz = maxcount;
463         session_fchan->maxreq_sz = fchan->maxreq_sz;
464
465         if (fchan->maxresp_sz > maxcount)
466                 fchan->maxresp_sz = maxcount;
467         session_fchan->maxresp_sz = fchan->maxresp_sz;
468
469         /* Set the max response cached size our default which is
470          * a multiple of PAGE_SIZE and small */
471         session_fchan->maxresp_cached = NFSD_PAGES_PER_SLOT * PAGE_SIZE;
472         fchan->maxresp_cached = session_fchan->maxresp_cached;
473
474         /* Use the client's maxops if possible */
475         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
476                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
477         session_fchan->maxops = fchan->maxops;
478
479         /* try to use the client requested number of slots */
480         if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
481                 fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
482
483         /* FIXME: Error means no more DRC pages so the server should
484          * recover pages from existing sessions. For now fail session
485          * creation.
486          */
487         status = set_forechannel_maxreqs(fchan);
488
489         session_fchan->maxreqs = fchan->maxreqs;
490         return status;
491 }
492
493 static int
494 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
495                    struct nfsd4_create_session *cses)
496 {
497         struct nfsd4_session *new, tmp;
498         int idx, status = nfserr_resource, slotsize;
499
500         memset(&tmp, 0, sizeof(tmp));
501
502         /* FIXME: For now, we just accept the client back channel attributes. */
503         tmp.se_bchannel = cses->back_channel;
504         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
505                                         &cses->fore_channel);
506         if (status)
507                 goto out;
508
509         /* allocate struct nfsd4_session and slot table in one piece */
510         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot);
511         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
512         if (!new)
513                 goto out;
514
515         memcpy(new, &tmp, sizeof(*new));
516
517         new->se_client = clp;
518         gen_sessionid(new);
519         idx = hash_sessionid(&new->se_sessionid);
520         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
521                NFS4_MAX_SESSIONID_LEN);
522
523         new->se_flags = cses->flags;
524         kref_init(&new->se_ref);
525         spin_lock(&sessionid_lock);
526         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
527         list_add(&new->se_perclnt, &clp->cl_sessions);
528         spin_unlock(&sessionid_lock);
529
530         status = nfs_ok;
531 out:
532         return status;
533 }
534
535 /* caller must hold sessionid_lock */
536 static struct nfsd4_session *
537 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
538 {
539         struct nfsd4_session *elem;
540         int idx;
541
542         dump_sessionid(__func__, sessionid);
543         idx = hash_sessionid(sessionid);
544         dprintk("%s: idx is %d\n", __func__, idx);
545         /* Search in the appropriate list */
546         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
547                 dump_sessionid("list traversal", &elem->se_sessionid);
548                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
549                             NFS4_MAX_SESSIONID_LEN)) {
550                         return elem;
551                 }
552         }
553
554         dprintk("%s: session not found\n", __func__);
555         return NULL;
556 }
557
558 /* caller must hold sessionid_lock */
559 static void
560 unhash_session(struct nfsd4_session *ses)
561 {
562         list_del(&ses->se_hash);
563         list_del(&ses->se_perclnt);
564 }
565
566 static void
567 release_session(struct nfsd4_session *ses)
568 {
569         spin_lock(&sessionid_lock);
570         unhash_session(ses);
571         spin_unlock(&sessionid_lock);
572         nfsd4_put_session(ses);
573 }
574
575 static void nfsd4_release_respages(struct page **respages, short resused);
576
577 void
578 free_session(struct kref *kref)
579 {
580         struct nfsd4_session *ses;
581         int i;
582
583         ses = container_of(kref, struct nfsd4_session, se_ref);
584         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
585                 struct nfsd4_cache_entry *e = &ses->se_slots[i].sl_cache_entry;
586                 nfsd4_release_respages(e->ce_respages, e->ce_resused);
587         }
588         kfree(ses);
589 }
590
591 static inline void
592 renew_client(struct nfs4_client *clp)
593 {
594         /*
595         * Move client to the end to the LRU list.
596         */
597         dprintk("renewing client (clientid %08x/%08x)\n", 
598                         clp->cl_clientid.cl_boot, 
599                         clp->cl_clientid.cl_id);
600         list_move_tail(&clp->cl_lru, &client_lru);
601         clp->cl_time = get_seconds();
602 }
603
604 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
605 static int
606 STALE_CLIENTID(clientid_t *clid)
607 {
608         if (clid->cl_boot == boot_time)
609                 return 0;
610         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
611                 clid->cl_boot, clid->cl_id, boot_time);
612         return 1;
613 }
614
615 /* 
616  * XXX Should we use a slab cache ?
617  * This type of memory management is somewhat inefficient, but we use it
618  * anyway since SETCLIENTID is not a common operation.
619  */
620 static struct nfs4_client *alloc_client(struct xdr_netobj name)
621 {
622         struct nfs4_client *clp;
623
624         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
625         if (clp == NULL)
626                 return NULL;
627         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
628         if (clp->cl_name.data == NULL) {
629                 kfree(clp);
630                 return NULL;
631         }
632         memcpy(clp->cl_name.data, name.data, name.len);
633         clp->cl_name.len = name.len;
634         return clp;
635 }
636
637 static void
638 shutdown_callback_client(struct nfs4_client *clp)
639 {
640         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
641
642         if (clnt) {
643                 /*
644                  * Callback threads take a reference on the client, so there
645                  * should be no outstanding callbacks at this point.
646                  */
647                 clp->cl_cb_conn.cb_client = NULL;
648                 rpc_shutdown_client(clnt);
649         }
650         if (clp->cl_cb_conn.cb_cred) {
651                 put_rpccred(clp->cl_cb_conn.cb_cred);
652                 clp->cl_cb_conn.cb_cred = NULL;
653         }
654 }
655
656 static inline void
657 free_client(struct nfs4_client *clp)
658 {
659         shutdown_callback_client(clp);
660         nfsd4_release_respages(clp->cl_slot.sl_cache_entry.ce_respages,
661                              clp->cl_slot.sl_cache_entry.ce_resused);
662         if (clp->cl_cred.cr_group_info)
663                 put_group_info(clp->cl_cred.cr_group_info);
664         kfree(clp->cl_principal);
665         kfree(clp->cl_name.data);
666         kfree(clp);
667 }
668
669 void
670 put_nfs4_client(struct nfs4_client *clp)
671 {
672         if (atomic_dec_and_test(&clp->cl_count))
673                 free_client(clp);
674 }
675
676 static void
677 expire_client(struct nfs4_client *clp)
678 {
679         struct nfs4_stateowner *sop;
680         struct nfs4_delegation *dp;
681         struct list_head reaplist;
682
683         dprintk("NFSD: expire_client cl_count %d\n",
684                             atomic_read(&clp->cl_count));
685
686         INIT_LIST_HEAD(&reaplist);
687         spin_lock(&recall_lock);
688         while (!list_empty(&clp->cl_delegations)) {
689                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
690                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
691                                 dp->dl_flock);
692                 list_del_init(&dp->dl_perclnt);
693                 list_move(&dp->dl_recall_lru, &reaplist);
694         }
695         spin_unlock(&recall_lock);
696         while (!list_empty(&reaplist)) {
697                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
698                 list_del_init(&dp->dl_recall_lru);
699                 unhash_delegation(dp);
700         }
701         list_del(&clp->cl_idhash);
702         list_del(&clp->cl_strhash);
703         list_del(&clp->cl_lru);
704         while (!list_empty(&clp->cl_openowners)) {
705                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
706                 release_openowner(sop);
707         }
708         while (!list_empty(&clp->cl_sessions)) {
709                 struct nfsd4_session  *ses;
710                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
711                                  se_perclnt);
712                 release_session(ses);
713         }
714         put_nfs4_client(clp);
715 }
716
717 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
718 {
719         struct nfs4_client *clp;
720
721         clp = alloc_client(name);
722         if (clp == NULL)
723                 return NULL;
724         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
725         atomic_set(&clp->cl_count, 1);
726         atomic_set(&clp->cl_cb_conn.cb_set, 0);
727         INIT_LIST_HEAD(&clp->cl_idhash);
728         INIT_LIST_HEAD(&clp->cl_strhash);
729         INIT_LIST_HEAD(&clp->cl_openowners);
730         INIT_LIST_HEAD(&clp->cl_delegations);
731         INIT_LIST_HEAD(&clp->cl_sessions);
732         INIT_LIST_HEAD(&clp->cl_lru);
733         return clp;
734 }
735
736 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
737 {
738         memcpy(target->cl_verifier.data, source->data,
739                         sizeof(target->cl_verifier.data));
740 }
741
742 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
743 {
744         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
745         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
746 }
747
748 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
749 {
750         target->cr_uid = source->cr_uid;
751         target->cr_gid = source->cr_gid;
752         target->cr_group_info = source->cr_group_info;
753         get_group_info(target->cr_group_info);
754 }
755
756 static int same_name(const char *n1, const char *n2)
757 {
758         return 0 == memcmp(n1, n2, HEXDIR_LEN);
759 }
760
761 static int
762 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
763 {
764         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
765 }
766
767 static int
768 same_clid(clientid_t *cl1, clientid_t *cl2)
769 {
770         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
771 }
772
773 /* XXX what about NGROUP */
774 static int
775 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
776 {
777         return cr1->cr_uid == cr2->cr_uid;
778 }
779
780 static void gen_clid(struct nfs4_client *clp)
781 {
782         static u32 current_clientid = 1;
783
784         clp->cl_clientid.cl_boot = boot_time;
785         clp->cl_clientid.cl_id = current_clientid++; 
786 }
787
788 static void gen_confirm(struct nfs4_client *clp)
789 {
790         static u32 i;
791         u32 *p;
792
793         p = (u32 *)clp->cl_confirm.data;
794         *p++ = get_seconds();
795         *p++ = i++;
796 }
797
798 static int check_name(struct xdr_netobj name)
799 {
800         if (name.len == 0) 
801                 return 0;
802         if (name.len > NFS4_OPAQUE_LIMIT) {
803                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
804                 return 0;
805         }
806         return 1;
807 }
808
809 static void
810 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
811 {
812         unsigned int idhashval;
813
814         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
815         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
816         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
817         list_add_tail(&clp->cl_lru, &client_lru);
818         clp->cl_time = get_seconds();
819 }
820
821 static void
822 move_to_confirmed(struct nfs4_client *clp)
823 {
824         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
825         unsigned int strhashval;
826
827         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
828         list_del_init(&clp->cl_strhash);
829         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
830         strhashval = clientstr_hashval(clp->cl_recdir);
831         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
832         renew_client(clp);
833 }
834
835 static struct nfs4_client *
836 find_confirmed_client(clientid_t *clid)
837 {
838         struct nfs4_client *clp;
839         unsigned int idhashval = clientid_hashval(clid->cl_id);
840
841         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
842                 if (same_clid(&clp->cl_clientid, clid))
843                         return clp;
844         }
845         return NULL;
846 }
847
848 static struct nfs4_client *
849 find_unconfirmed_client(clientid_t *clid)
850 {
851         struct nfs4_client *clp;
852         unsigned int idhashval = clientid_hashval(clid->cl_id);
853
854         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
855                 if (same_clid(&clp->cl_clientid, clid))
856                         return clp;
857         }
858         return NULL;
859 }
860
861 /*
862  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
863  * parameter. Matching is based on the fact the at least one of the
864  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
865  *
866  * FIXME: we need to unify the clientid namespaces for nfsv4.x
867  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
868  * and SET_CLIENTID{,_CONFIRM}
869  */
870 static inline int
871 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
872 {
873         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
874         return use_exchange_id == has_exchange_flags;
875 }
876
877 static struct nfs4_client *
878 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
879                              bool use_exchange_id)
880 {
881         struct nfs4_client *clp;
882
883         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
884                 if (same_name(clp->cl_recdir, dname) &&
885                     match_clientid_establishment(clp, use_exchange_id))
886                         return clp;
887         }
888         return NULL;
889 }
890
891 static struct nfs4_client *
892 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
893                                bool use_exchange_id)
894 {
895         struct nfs4_client *clp;
896
897         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
898                 if (same_name(clp->cl_recdir, dname) &&
899                     match_clientid_establishment(clp, use_exchange_id))
900                         return clp;
901         }
902         return NULL;
903 }
904
905 /* a helper function for parse_callback */
906 static int
907 parse_octet(unsigned int *lenp, char **addrp)
908 {
909         unsigned int len = *lenp;
910         char *p = *addrp;
911         int n = -1;
912         char c;
913
914         for (;;) {
915                 if (!len)
916                         break;
917                 len--;
918                 c = *p++;
919                 if (c == '.')
920                         break;
921                 if ((c < '0') || (c > '9')) {
922                         n = -1;
923                         break;
924                 }
925                 if (n < 0)
926                         n = 0;
927                 n = (n * 10) + (c - '0');
928                 if (n > 255) {
929                         n = -1;
930                         break;
931                 }
932         }
933         *lenp = len;
934         *addrp = p;
935         return n;
936 }
937
938 /* parse and set the setclientid ipv4 callback address */
939 static int
940 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
941 {
942         int temp = 0;
943         u32 cbaddr = 0;
944         u16 cbport = 0;
945         u32 addrlen = addr_len;
946         char *addr = addr_val;
947         int i, shift;
948
949         /* ipaddress */
950         shift = 24;
951         for(i = 4; i > 0  ; i--) {
952                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
953                         return 0;
954                 }
955                 cbaddr |= (temp << shift);
956                 if (shift > 0)
957                 shift -= 8;
958         }
959         *cbaddrp = cbaddr;
960
961         /* port */
962         shift = 8;
963         for(i = 2; i > 0  ; i--) {
964                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
965                         return 0;
966                 }
967                 cbport |= (temp << shift);
968                 if (shift > 0)
969                         shift -= 8;
970         }
971         *cbportp = cbport;
972         return 1;
973 }
974
975 static void
976 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
977 {
978         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
979
980         /* Currently, we only support tcp for the callback channel */
981         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
982                 goto out_err;
983
984         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
985                          &cb->cb_addr, &cb->cb_port)))
986                 goto out_err;
987         cb->cb_minorversion = 0;
988         cb->cb_prog = se->se_callback_prog;
989         cb->cb_ident = se->se_callback_ident;
990         return;
991 out_err:
992         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
993                 "will not receive delegations\n",
994                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
995
996         return;
997 }
998
999 void
1000 nfsd4_set_statp(struct svc_rqst *rqstp, __be32 *statp)
1001 {
1002         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1003
1004         resp->cstate.statp = statp;
1005 }
1006
1007 /*
1008  * Dereference the result pages.
1009  */
1010 static void
1011 nfsd4_release_respages(struct page **respages, short resused)
1012 {
1013         int i;
1014
1015         dprintk("--> %s\n", __func__);
1016         for (i = 0; i < resused; i++) {
1017                 if (!respages[i])
1018                         continue;
1019                 put_page(respages[i]);
1020                 respages[i] = NULL;
1021         }
1022 }
1023
1024 static void
1025 nfsd4_copy_pages(struct page **topages, struct page **frompages, short count)
1026 {
1027         int i;
1028
1029         for (i = 0; i < count; i++) {
1030                 topages[i] = frompages[i];
1031                 if (!topages[i])
1032                         continue;
1033                 get_page(topages[i]);
1034         }
1035 }
1036
1037 /*
1038  * Cache the reply pages up to NFSD_PAGES_PER_SLOT + 1, clearing the previous
1039  * pages. We add a page to NFSD_PAGES_PER_SLOT for the case where the total
1040  * length of the XDR response is less than se_fmaxresp_cached
1041  * (NFSD_PAGES_PER_SLOT * PAGE_SIZE) but the xdr_buf pages is used for a
1042  * of the reply (e.g. readdir).
1043  *
1044  * Store the base and length of the rq_req.head[0] page
1045  * of the NFSv4.1 data, just past the rpc header.
1046  */
1047 void
1048 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1049 {
1050         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1051         struct svc_rqst *rqstp = resp->rqstp;
1052         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1053         struct nfsd4_op *op = &args->ops[resp->opcnt];
1054         struct kvec *resv = &rqstp->rq_res.head[0];
1055
1056         dprintk("--> %s entry %p\n", __func__, entry);
1057
1058         /* Don't cache a failed OP_SEQUENCE. */
1059         if (resp->opcnt == 1 && op->opnum == OP_SEQUENCE && resp->cstate.status)
1060                 return;
1061
1062         nfsd4_release_respages(entry->ce_respages, entry->ce_resused);
1063         entry->ce_opcnt = resp->opcnt;
1064         entry->ce_status = resp->cstate.status;
1065
1066         /*
1067          * Don't need a page to cache just the sequence operation - the slot
1068          * does this for us!
1069          */
1070
1071         if (nfsd4_not_cached(resp)) {
1072                 entry->ce_resused = 0;
1073                 entry->ce_rpchdrlen = 0;
1074                 dprintk("%s Just cache SEQUENCE. ce_cachethis %d\n", __func__,
1075                         resp->cstate.slot->sl_cache_entry.ce_cachethis);
1076                 return;
1077         }
1078         entry->ce_resused = rqstp->rq_resused;
1079         if (entry->ce_resused > NFSD_PAGES_PER_SLOT + 1)
1080                 entry->ce_resused = NFSD_PAGES_PER_SLOT + 1;
1081         nfsd4_copy_pages(entry->ce_respages, rqstp->rq_respages,
1082                          entry->ce_resused);
1083         entry->ce_datav.iov_base = resp->cstate.statp;
1084         entry->ce_datav.iov_len = resv->iov_len - ((char *)resp->cstate.statp -
1085                                 (char *)page_address(rqstp->rq_respages[0]));
1086         /* Current request rpc header length*/
1087         entry->ce_rpchdrlen = (char *)resp->cstate.statp -
1088                                 (char *)page_address(rqstp->rq_respages[0]);
1089 }
1090
1091 /*
1092  * We keep the rpc header, but take the nfs reply from the replycache.
1093  */
1094 static int
1095 nfsd41_copy_replay_data(struct nfsd4_compoundres *resp,
1096                         struct nfsd4_cache_entry *entry)
1097 {
1098         struct svc_rqst *rqstp = resp->rqstp;
1099         struct kvec *resv = &resp->rqstp->rq_res.head[0];
1100         int len;
1101
1102         /* Current request rpc header length*/
1103         len = (char *)resp->cstate.statp -
1104                         (char *)page_address(rqstp->rq_respages[0]);
1105         if (entry->ce_datav.iov_len + len > PAGE_SIZE) {
1106                 dprintk("%s v41 cached reply too large (%Zd).\n", __func__,
1107                         entry->ce_datav.iov_len);
1108                 return 0;
1109         }
1110         /* copy the cached reply nfsd data past the current rpc header */
1111         memcpy((char *)resv->iov_base + len, entry->ce_datav.iov_base,
1112                 entry->ce_datav.iov_len);
1113         resv->iov_len = len + entry->ce_datav.iov_len;
1114         return 1;
1115 }
1116
1117 /*
1118  * Keep the first page of the replay. Copy the NFSv4.1 data from the first
1119  * cached page.  Replace any futher replay pages from the cache.
1120  */
1121 __be32
1122 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1123                          struct nfsd4_sequence *seq)
1124 {
1125         struct nfsd4_cache_entry *entry = &resp->cstate.slot->sl_cache_entry;
1126         __be32 status;
1127
1128         dprintk("--> %s entry %p\n", __func__, entry);
1129
1130         /*
1131          * If this is just the sequence operation, we did not keep
1132          * a page in the cache entry because we can just use the
1133          * slot info stored in struct nfsd4_sequence that was checked
1134          * against the slot in nfsd4_sequence().
1135          *
1136          * This occurs when seq->cachethis is FALSE, or when the client
1137          * session inactivity timer fires and a solo sequence operation
1138          * is sent (lease renewal).
1139          */
1140         if (seq && nfsd4_not_cached(resp)) {
1141                 seq->maxslots = resp->cstate.session->se_fchannel.maxreqs;
1142                 return nfs_ok;
1143         }
1144
1145         if (!nfsd41_copy_replay_data(resp, entry)) {
1146                 /*
1147                  * Not enough room to use the replay rpc header, send the
1148                  * cached header. Release all the allocated result pages.
1149                  */
1150                 svc_free_res_pages(resp->rqstp);
1151                 nfsd4_copy_pages(resp->rqstp->rq_respages, entry->ce_respages,
1152                         entry->ce_resused);
1153         } else {
1154                 /* Release all but the first allocated result page */
1155
1156                 resp->rqstp->rq_resused--;
1157                 svc_free_res_pages(resp->rqstp);
1158
1159                 nfsd4_copy_pages(&resp->rqstp->rq_respages[1],
1160                                  &entry->ce_respages[1],
1161                                  entry->ce_resused - 1);
1162         }
1163
1164         resp->rqstp->rq_resused = entry->ce_resused;
1165         resp->opcnt = entry->ce_opcnt;
1166         resp->cstate.iovlen = entry->ce_datav.iov_len + entry->ce_rpchdrlen;
1167         status = entry->ce_status;
1168
1169         return status;
1170 }
1171
1172 /*
1173  * Set the exchange_id flags returned by the server.
1174  */
1175 static void
1176 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1177 {
1178         /* pNFS is not supported */
1179         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1180
1181         /* Referrals are supported, Migration is not. */
1182         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1183
1184         /* set the wire flags to return to client. */
1185         clid->flags = new->cl_exchange_flags;
1186 }
1187
1188 __be32
1189 nfsd4_exchange_id(struct svc_rqst *rqstp,
1190                   struct nfsd4_compound_state *cstate,
1191                   struct nfsd4_exchange_id *exid)
1192 {
1193         struct nfs4_client *unconf, *conf, *new;
1194         int status;
1195         unsigned int            strhashval;
1196         char                    dname[HEXDIR_LEN];
1197         nfs4_verifier           verf = exid->verifier;
1198         u32                     ip_addr = svc_addr_in(rqstp)->sin_addr.s_addr;
1199
1200         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1201                 " ip_addr=%u flags %x, spa_how %d\n",
1202                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1203                 ip_addr, exid->flags, exid->spa_how);
1204
1205         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1206                 return nfserr_inval;
1207
1208         /* Currently only support SP4_NONE */
1209         switch (exid->spa_how) {
1210         case SP4_NONE:
1211                 break;
1212         case SP4_SSV:
1213                 return nfserr_encr_alg_unsupp;
1214         default:
1215                 BUG();                          /* checked by xdr code */
1216         case SP4_MACH_CRED:
1217                 return nfserr_serverfault;      /* no excuse :-/ */
1218         }
1219
1220         status = nfs4_make_rec_clidname(dname, &exid->clname);
1221
1222         if (status)
1223                 goto error;
1224
1225         strhashval = clientstr_hashval(dname);
1226
1227         nfs4_lock_state();
1228         status = nfs_ok;
1229
1230         conf = find_confirmed_client_by_str(dname, strhashval, true);
1231         if (conf) {
1232                 if (!same_verf(&verf, &conf->cl_verifier)) {
1233                         /* 18.35.4 case 8 */
1234                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1235                                 status = nfserr_not_same;
1236                                 goto out;
1237                         }
1238                         /* Client reboot: destroy old state */
1239                         expire_client(conf);
1240                         goto out_new;
1241                 }
1242                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1243                         /* 18.35.4 case 9 */
1244                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1245                                 status = nfserr_perm;
1246                                 goto out;
1247                         }
1248                         expire_client(conf);
1249                         goto out_new;
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         }
1265
1266         /* 18.35.4 case 7 */
1267         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1268                 status = nfserr_noent;
1269                 goto out;
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_fchannel.maxreqs)
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  * Spawn a thread to perform a recall on the delegation represented
2065  * by the lease (file_lock)
2066  *
2067  * Called from break_lease() with lock_kernel() held.
2068  * Note: we assume break_lease will only call this *once* for any given
2069  * lease.
2070  */
2071 static
2072 void nfsd_break_deleg_cb(struct file_lock *fl)
2073 {
2074         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2075
2076         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2077         if (!dp)
2078                 return;
2079
2080         /* We're assuming the state code never drops its reference
2081          * without first removing the lease.  Since we're in this lease
2082          * callback (and since the lease code is serialized by the kernel
2083          * lock) we know the server hasn't removed the lease yet, we know
2084          * it's safe to take a reference: */
2085         atomic_inc(&dp->dl_count);
2086         atomic_inc(&dp->dl_client->cl_count);
2087
2088         spin_lock(&recall_lock);
2089         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2090         spin_unlock(&recall_lock);
2091
2092         /* only place dl_time is set. protected by lock_kernel*/
2093         dp->dl_time = get_seconds();
2094
2095         /*
2096          * We don't want the locks code to timeout the lease for us;
2097          * we'll remove it ourself if the delegation isn't returned
2098          * in time.
2099          */
2100         fl->fl_break_time = 0;
2101
2102         dp->dl_file->fi_had_conflict = true;
2103         nfsd4_cb_recall(dp);
2104 }
2105
2106 /*
2107  * The file_lock is being reapd.
2108  *
2109  * Called by locks_free_lock() with lock_kernel() held.
2110  */
2111 static
2112 void nfsd_release_deleg_cb(struct file_lock *fl)
2113 {
2114         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2115
2116         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2117
2118         if (!(fl->fl_flags & FL_LEASE) || !dp)
2119                 return;
2120         dp->dl_flock = NULL;
2121 }
2122
2123 /*
2124  * Set the delegation file_lock back pointer.
2125  *
2126  * Called from setlease() with lock_kernel() held.
2127  */
2128 static
2129 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2130 {
2131         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2132
2133         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2134         if (!dp)
2135                 return;
2136         dp->dl_flock = new;
2137 }
2138
2139 /*
2140  * Called from setlease() with lock_kernel() held
2141  */
2142 static
2143 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2144 {
2145         struct nfs4_delegation *onlistd =
2146                 (struct nfs4_delegation *)onlist->fl_owner;
2147         struct nfs4_delegation *tryd =
2148                 (struct nfs4_delegation *)try->fl_owner;
2149
2150         if (onlist->fl_lmops != try->fl_lmops)
2151                 return 0;
2152
2153         return onlistd->dl_client == tryd->dl_client;
2154 }
2155
2156
2157 static
2158 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2159 {
2160         if (arg & F_UNLCK)
2161                 return lease_modify(onlist, arg);
2162         else
2163                 return -EAGAIN;
2164 }
2165
2166 static struct lock_manager_operations nfsd_lease_mng_ops = {
2167         .fl_break = nfsd_break_deleg_cb,
2168         .fl_release_private = nfsd_release_deleg_cb,
2169         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2170         .fl_mylease = nfsd_same_client_deleg_cb,
2171         .fl_change = nfsd_change_deleg_cb,
2172 };
2173
2174
2175 __be32
2176 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2177                     struct nfsd4_open *open)
2178 {
2179         clientid_t *clientid = &open->op_clientid;
2180         struct nfs4_client *clp = NULL;
2181         unsigned int strhashval;
2182         struct nfs4_stateowner *sop = NULL;
2183
2184         if (!check_name(open->op_owner))
2185                 return nfserr_inval;
2186
2187         if (STALE_CLIENTID(&open->op_clientid))
2188                 return nfserr_stale_clientid;
2189
2190         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2191         sop = find_openstateowner_str(strhashval, open);
2192         open->op_stateowner = sop;
2193         if (!sop) {
2194                 /* Make sure the client's lease hasn't expired. */
2195                 clp = find_confirmed_client(clientid);
2196                 if (clp == NULL)
2197                         return nfserr_expired;
2198                 goto renew;
2199         }
2200         /* When sessions are used, skip open sequenceid processing */
2201         if (nfsd4_has_session(cstate))
2202                 goto renew;
2203         if (!sop->so_confirmed) {
2204                 /* Replace unconfirmed owners without checking for replay. */
2205                 clp = sop->so_client;
2206                 release_openowner(sop);
2207                 open->op_stateowner = NULL;
2208                 goto renew;
2209         }
2210         if (open->op_seqid == sop->so_seqid - 1) {
2211                 if (sop->so_replay.rp_buflen)
2212                         return nfserr_replay_me;
2213                 /* The original OPEN failed so spectacularly
2214                  * that we don't even have replay data saved!
2215                  * Therefore, we have no choice but to continue
2216                  * processing this OPEN; presumably, we'll
2217                  * fail again for the same reason.
2218                  */
2219                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2220                 goto renew;
2221         }
2222         if (open->op_seqid != sop->so_seqid)
2223                 return nfserr_bad_seqid;
2224 renew:
2225         if (open->op_stateowner == NULL) {
2226                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2227                 if (sop == NULL)
2228                         return nfserr_resource;
2229                 open->op_stateowner = sop;
2230         }
2231         list_del_init(&sop->so_close_lru);
2232         renew_client(sop->so_client);
2233         return nfs_ok;
2234 }
2235
2236 static inline __be32
2237 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2238 {
2239         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2240                 return nfserr_openmode;
2241         else
2242                 return nfs_ok;
2243 }
2244
2245 static struct nfs4_delegation *
2246 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2247 {
2248         struct nfs4_delegation *dp;
2249
2250         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2251                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2252                         return dp;
2253         }
2254         return NULL;
2255 }
2256
2257 static __be32
2258 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2259                 struct nfs4_delegation **dp)
2260 {
2261         int flags;
2262         __be32 status = nfserr_bad_stateid;
2263
2264         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2265         if (*dp == NULL)
2266                 goto out;
2267         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2268                                                 RD_STATE : WR_STATE;
2269         status = nfs4_check_delegmode(*dp, flags);
2270         if (status)
2271                 *dp = NULL;
2272 out:
2273         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2274                 return nfs_ok;
2275         if (status)
2276                 return status;
2277         open->op_stateowner->so_confirmed = 1;
2278         return nfs_ok;
2279 }
2280
2281 static __be32
2282 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2283 {
2284         struct nfs4_stateid *local;
2285         __be32 status = nfserr_share_denied;
2286         struct nfs4_stateowner *sop = open->op_stateowner;
2287
2288         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2289                 /* ignore lock owners */
2290                 if (local->st_stateowner->so_is_open_owner == 0)
2291                         continue;
2292                 /* remember if we have seen this open owner */
2293                 if (local->st_stateowner == sop)
2294                         *stpp = local;
2295                 /* check for conflicting share reservations */
2296                 if (!test_share(local, open))
2297                         goto out;
2298         }
2299         status = 0;
2300 out:
2301         return status;
2302 }
2303
2304 static inline struct nfs4_stateid *
2305 nfs4_alloc_stateid(void)
2306 {
2307         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2308 }
2309
2310 static __be32
2311 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2312                 struct nfs4_delegation *dp,
2313                 struct svc_fh *cur_fh, int flags)
2314 {
2315         struct nfs4_stateid *stp;
2316
2317         stp = nfs4_alloc_stateid();
2318         if (stp == NULL)
2319                 return nfserr_resource;
2320
2321         if (dp) {
2322                 get_file(dp->dl_vfs_file);
2323                 stp->st_vfs_file = dp->dl_vfs_file;
2324         } else {
2325                 __be32 status;
2326                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2327                                 &stp->st_vfs_file);
2328                 if (status) {
2329                         if (status == nfserr_dropit)
2330                                 status = nfserr_jukebox;
2331                         kmem_cache_free(stateid_slab, stp);
2332                         return status;
2333                 }
2334         }
2335         *stpp = stp;
2336         return 0;
2337 }
2338
2339 static inline __be32
2340 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2341                 struct nfsd4_open *open)
2342 {
2343         struct iattr iattr = {
2344                 .ia_valid = ATTR_SIZE,
2345                 .ia_size = 0,
2346         };
2347         if (!open->op_truncate)
2348                 return 0;
2349         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2350                 return nfserr_inval;
2351         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2352 }
2353
2354 static __be32
2355 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2356 {
2357         struct file *filp = stp->st_vfs_file;
2358         struct inode *inode = filp->f_path.dentry->d_inode;
2359         unsigned int share_access, new_writer;
2360         __be32 status;
2361
2362         set_access(&share_access, stp->st_access_bmap);
2363         new_writer = (~share_access) & open->op_share_access
2364                         & NFS4_SHARE_ACCESS_WRITE;
2365
2366         if (new_writer) {
2367                 int err = get_write_access(inode);
2368                 if (err)
2369                         return nfserrno(err);
2370                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2371                 if (err)
2372                         return nfserrno(err);
2373                 file_take_write(filp);
2374         }
2375         status = nfsd4_truncate(rqstp, cur_fh, open);
2376         if (status) {
2377                 if (new_writer)
2378                         put_write_access(inode);
2379                 return status;
2380         }
2381         /* remember the open */
2382         filp->f_mode |= open->op_share_access;
2383         __set_bit(open->op_share_access, &stp->st_access_bmap);
2384         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2385
2386         return nfs_ok;
2387 }
2388
2389
2390 static void
2391 nfs4_set_claim_prev(struct nfsd4_open *open)
2392 {
2393         open->op_stateowner->so_confirmed = 1;
2394         open->op_stateowner->so_client->cl_firststate = 1;
2395 }
2396
2397 /*
2398  * Attempt to hand out a delegation.
2399  */
2400 static void
2401 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2402 {
2403         struct nfs4_delegation *dp;
2404         struct nfs4_stateowner *sop = stp->st_stateowner;
2405         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2406         struct file_lock fl, *flp = &fl;
2407         int status, flag = 0;
2408
2409         flag = NFS4_OPEN_DELEGATE_NONE;
2410         open->op_recall = 0;
2411         switch (open->op_claim_type) {
2412                 case NFS4_OPEN_CLAIM_PREVIOUS:
2413                         if (!atomic_read(&cb->cb_set))
2414                                 open->op_recall = 1;
2415                         flag = open->op_delegate_type;
2416                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2417                                 goto out;
2418                         break;
2419                 case NFS4_OPEN_CLAIM_NULL:
2420                         /* Let's not give out any delegations till everyone's
2421                          * had the chance to reclaim theirs.... */
2422                         if (locks_in_grace())
2423                                 goto out;
2424                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2425                                 goto out;
2426                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2427                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2428                         else
2429                                 flag = NFS4_OPEN_DELEGATE_READ;
2430                         break;
2431                 default:
2432                         goto out;
2433         }
2434
2435         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2436         if (dp == NULL) {
2437                 flag = NFS4_OPEN_DELEGATE_NONE;
2438                 goto out;
2439         }
2440         locks_init_lock(&fl);
2441         fl.fl_lmops = &nfsd_lease_mng_ops;
2442         fl.fl_flags = FL_LEASE;
2443         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2444         fl.fl_end = OFFSET_MAX;
2445         fl.fl_owner =  (fl_owner_t)dp;
2446         fl.fl_file = stp->st_vfs_file;
2447         fl.fl_pid = current->tgid;
2448
2449         /* vfs_setlease checks to see if delegation should be handed out.
2450          * the lock_manager callbacks fl_mylease and fl_change are used
2451          */
2452         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2453                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2454                 unhash_delegation(dp);
2455                 flag = NFS4_OPEN_DELEGATE_NONE;
2456                 goto out;
2457         }
2458
2459         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2460
2461         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2462                      dp->dl_stateid.si_boot,
2463                      dp->dl_stateid.si_stateownerid,
2464                      dp->dl_stateid.si_fileid,
2465                      dp->dl_stateid.si_generation);
2466 out:
2467         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2468                         && flag == NFS4_OPEN_DELEGATE_NONE
2469                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2470                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2471         open->op_delegate_type = flag;
2472 }
2473
2474 /*
2475  * called with nfs4_lock_state() held.
2476  */
2477 __be32
2478 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2479 {
2480         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2481         struct nfs4_file *fp = NULL;
2482         struct inode *ino = current_fh->fh_dentry->d_inode;
2483         struct nfs4_stateid *stp = NULL;
2484         struct nfs4_delegation *dp = NULL;
2485         __be32 status;
2486
2487         status = nfserr_inval;
2488         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2489                         || !deny_valid(open->op_share_deny))
2490                 goto out;
2491         /*
2492          * Lookup file; if found, lookup stateid and check open request,
2493          * and check for delegations in the process of being recalled.
2494          * If not found, create the nfs4_file struct
2495          */
2496         fp = find_file(ino);
2497         if (fp) {
2498                 if ((status = nfs4_check_open(fp, open, &stp)))
2499                         goto out;
2500                 status = nfs4_check_deleg(fp, open, &dp);
2501                 if (status)
2502                         goto out;
2503         } else {
2504                 status = nfserr_bad_stateid;
2505                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2506                         goto out;
2507                 status = nfserr_resource;
2508                 fp = alloc_init_file(ino);
2509                 if (fp == NULL)
2510                         goto out;
2511         }
2512
2513         /*
2514          * OPEN the file, or upgrade an existing OPEN.
2515          * If truncate fails, the OPEN fails.
2516          */
2517         if (stp) {
2518                 /* Stateid was found, this is an OPEN upgrade */
2519                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2520                 if (status)
2521                         goto out;
2522                 update_stateid(&stp->st_stateid);
2523         } else {
2524                 /* Stateid was not found, this is a new OPEN */
2525                 int flags = 0;
2526                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2527                         flags |= NFSD_MAY_READ;
2528                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2529                         flags |= NFSD_MAY_WRITE;
2530                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2531                 if (status)
2532                         goto out;
2533                 init_stateid(stp, fp, open);
2534                 status = nfsd4_truncate(rqstp, current_fh, open);
2535                 if (status) {
2536                         release_open_stateid(stp);
2537                         goto out;
2538                 }
2539                 if (nfsd4_has_session(&resp->cstate))
2540                         update_stateid(&stp->st_stateid);
2541         }
2542         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2543
2544         if (nfsd4_has_session(&resp->cstate))
2545                 open->op_stateowner->so_confirmed = 1;
2546
2547         /*
2548         * Attempt to hand out a delegation. No error return, because the
2549         * OPEN succeeds even if we fail.
2550         */
2551         nfs4_open_delegation(current_fh, open, stp);
2552
2553         status = nfs_ok;
2554
2555         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2556                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2557                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2558 out:
2559         if (fp)
2560                 put_nfs4_file(fp);
2561         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2562                 nfs4_set_claim_prev(open);
2563         /*
2564         * To finish the open response, we just need to set the rflags.
2565         */
2566         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2567         if (!open->op_stateowner->so_confirmed &&
2568             !nfsd4_has_session(&resp->cstate))
2569                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2570
2571         return status;
2572 }
2573
2574 __be32
2575 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2576             clientid_t *clid)
2577 {
2578         struct nfs4_client *clp;
2579         __be32 status;
2580
2581         nfs4_lock_state();
2582         dprintk("process_renew(%08x/%08x): starting\n", 
2583                         clid->cl_boot, clid->cl_id);
2584         status = nfserr_stale_clientid;
2585         if (STALE_CLIENTID(clid))
2586                 goto out;
2587         clp = find_confirmed_client(clid);
2588         status = nfserr_expired;
2589         if (clp == NULL) {
2590                 /* We assume the client took too long to RENEW. */
2591                 dprintk("nfsd4_renew: clientid not found!\n");
2592                 goto out;
2593         }
2594         renew_client(clp);
2595         status = nfserr_cb_path_down;
2596         if (!list_empty(&clp->cl_delegations)
2597                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2598                 goto out;
2599         status = nfs_ok;
2600 out:
2601         nfs4_unlock_state();
2602         return status;
2603 }
2604
2605 struct lock_manager nfsd4_manager = {
2606 };
2607
2608 static void
2609 nfsd4_end_grace(void)
2610 {
2611         dprintk("NFSD: end of grace period\n");
2612         nfsd4_recdir_purge_old();
2613         locks_end_grace(&nfsd4_manager);
2614 }
2615
2616 static time_t
2617 nfs4_laundromat(void)
2618 {
2619         struct nfs4_client *clp;
2620         struct nfs4_stateowner *sop;
2621         struct nfs4_delegation *dp;
2622         struct list_head *pos, *next, reaplist;
2623         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2624         time_t t, clientid_val = NFSD_LEASE_TIME;
2625         time_t u, test_val = NFSD_LEASE_TIME;
2626
2627         nfs4_lock_state();
2628
2629         dprintk("NFSD: laundromat service - starting\n");
2630         if (locks_in_grace())
2631                 nfsd4_end_grace();
2632         list_for_each_safe(pos, next, &client_lru) {
2633                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2634                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2635                         t = clp->cl_time - cutoff;
2636                         if (clientid_val > t)
2637                                 clientid_val = t;
2638                         break;
2639                 }
2640                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2641                         clp->cl_clientid.cl_id);
2642                 nfsd4_remove_clid_dir(clp);
2643                 expire_client(clp);
2644         }
2645         INIT_LIST_HEAD(&reaplist);
2646         spin_lock(&recall_lock);
2647         list_for_each_safe(pos, next, &del_recall_lru) {
2648                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2649                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2650                         u = dp->dl_time - cutoff;
2651                         if (test_val > u)
2652                                 test_val = u;
2653                         break;
2654                 }
2655                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2656                                     dp, dp->dl_flock);
2657                 list_move(&dp->dl_recall_lru, &reaplist);
2658         }
2659         spin_unlock(&recall_lock);
2660         list_for_each_safe(pos, next, &reaplist) {
2661                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2662                 list_del_init(&dp->dl_recall_lru);
2663                 unhash_delegation(dp);
2664         }
2665         test_val = NFSD_LEASE_TIME;
2666         list_for_each_safe(pos, next, &close_lru) {
2667                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2668                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2669                         u = sop->so_time - cutoff;
2670                         if (test_val > u)
2671                                 test_val = u;
2672                         break;
2673                 }
2674                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2675                         sop->so_id);
2676                 release_openowner(sop);
2677         }
2678         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2679                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2680         nfs4_unlock_state();
2681         return clientid_val;
2682 }
2683
2684 static struct workqueue_struct *laundry_wq;
2685 static void laundromat_main(struct work_struct *);
2686 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2687
2688 static void
2689 laundromat_main(struct work_struct *not_used)
2690 {
2691         time_t t;
2692
2693         t = nfs4_laundromat();
2694         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2695         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2696 }
2697
2698 static struct nfs4_stateowner *
2699 search_close_lru(u32 st_id, int flags)
2700 {
2701         struct nfs4_stateowner *local = NULL;
2702
2703         if (flags & CLOSE_STATE) {
2704                 list_for_each_entry(local, &close_lru, so_close_lru) {
2705                         if (local->so_id == st_id)
2706                                 return local;
2707                 }
2708         }
2709         return NULL;
2710 }
2711
2712 static inline int
2713 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2714 {
2715         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2716 }
2717
2718 static int
2719 STALE_STATEID(stateid_t *stateid)
2720 {
2721         if (time_after((unsigned long)boot_time,
2722                         (unsigned long)stateid->si_boot)) {
2723                 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2724                         stateid->si_boot, stateid->si_stateownerid,
2725                         stateid->si_fileid, stateid->si_generation);
2726                 return 1;
2727         }
2728         return 0;
2729 }
2730
2731 static int
2732 EXPIRED_STATEID(stateid_t *stateid)
2733 {
2734         if (time_before((unsigned long)boot_time,
2735                         ((unsigned long)stateid->si_boot)) &&
2736             time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2737                 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2738                         stateid->si_boot, stateid->si_stateownerid,
2739                         stateid->si_fileid, stateid->si_generation);
2740                 return 1;
2741         }
2742         return 0;
2743 }
2744
2745 static __be32
2746 stateid_error_map(stateid_t *stateid)
2747 {
2748         if (STALE_STATEID(stateid))
2749                 return nfserr_stale_stateid;
2750         if (EXPIRED_STATEID(stateid))
2751                 return nfserr_expired;
2752
2753         dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2754                 stateid->si_boot, stateid->si_stateownerid,
2755                 stateid->si_fileid, stateid->si_generation);
2756         return nfserr_bad_stateid;
2757 }
2758
2759 static inline int
2760 access_permit_read(unsigned long access_bmap)
2761 {
2762         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2763                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2764                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2765 }
2766
2767 static inline int
2768 access_permit_write(unsigned long access_bmap)
2769 {
2770         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2771                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2772 }
2773
2774 static
2775 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2776 {
2777         __be32 status = nfserr_openmode;
2778
2779         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2780                 goto out;
2781         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2782                 goto out;
2783         status = nfs_ok;
2784 out:
2785         return status;
2786 }
2787
2788 static inline __be32
2789 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2790 {
2791         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2792                 return nfs_ok;
2793         else if (locks_in_grace()) {
2794                 /* Answer in remaining cases depends on existance of
2795                  * conflicting state; so we must wait out the grace period. */
2796                 return nfserr_grace;
2797         } else if (flags & WR_STATE)
2798                 return nfs4_share_conflict(current_fh,
2799                                 NFS4_SHARE_DENY_WRITE);
2800         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2801                 return nfs4_share_conflict(current_fh,
2802                                 NFS4_SHARE_DENY_READ);
2803 }
2804
2805 /*
2806  * Allow READ/WRITE during grace period on recovered state only for files
2807  * that are not able to provide mandatory locking.
2808  */
2809 static inline int
2810 grace_disallows_io(struct inode *inode)
2811 {
2812         return locks_in_grace() && mandatory_lock(inode);
2813 }
2814
2815 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2816 {
2817         /*
2818          * When sessions are used the stateid generation number is ignored
2819          * when it is zero.
2820          */
2821         if ((flags & HAS_SESSION) && in->si_generation == 0)
2822                 goto out;
2823
2824         /* If the client sends us a stateid from the future, it's buggy: */
2825         if (in->si_generation > ref->si_generation)
2826                 return nfserr_bad_stateid;
2827         /*
2828          * The following, however, can happen.  For example, if the
2829          * client sends an open and some IO at the same time, the open
2830          * may bump si_generation while the IO is still in flight.
2831          * Thanks to hard links and renames, the client never knows what
2832          * file an open will affect.  So it could avoid that situation
2833          * only by serializing all opens and IO from the same open
2834          * owner.  To recover from the old_stateid error, the client
2835          * will just have to retry the IO:
2836          */
2837         if (in->si_generation < ref->si_generation)
2838                 return nfserr_old_stateid;
2839 out:
2840         return nfs_ok;
2841 }
2842
2843 static int is_delegation_stateid(stateid_t *stateid)
2844 {
2845         return stateid->si_fileid == 0;
2846 }
2847
2848 /*
2849 * Checks for stateid operations
2850 */
2851 __be32
2852 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2853                            stateid_t *stateid, int flags, struct file **filpp)
2854 {
2855         struct nfs4_stateid *stp = NULL;
2856         struct nfs4_delegation *dp = NULL;
2857         struct svc_fh *current_fh = &cstate->current_fh;
2858         struct inode *ino = current_fh->fh_dentry->d_inode;
2859         __be32 status;
2860
2861         if (filpp)
2862                 *filpp = NULL;
2863
2864         if (grace_disallows_io(ino))
2865                 return nfserr_grace;
2866
2867         if (nfsd4_has_session(cstate))
2868                 flags |= HAS_SESSION;
2869
2870         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2871                 return check_special_stateids(current_fh, stateid, flags);
2872
2873         status = nfserr_stale_stateid;
2874         if (STALE_STATEID(stateid)) 
2875                 goto out;
2876
2877         status = nfserr_bad_stateid;
2878         if (is_delegation_stateid(stateid)) {
2879                 dp = find_delegation_stateid(ino, stateid);
2880                 if (!dp) {
2881                         status = stateid_error_map(stateid);
2882                         goto out;
2883                 }
2884                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2885                                                   flags);
2886                 if (status)
2887                         goto out;
2888                 status = nfs4_check_delegmode(dp, flags);
2889                 if (status)
2890                         goto out;
2891                 renew_client(dp->dl_client);
2892                 if (filpp)
2893                         *filpp = dp->dl_vfs_file;
2894         } else { /* open or lock stateid */
2895                 stp = find_stateid(stateid, flags);
2896                 if (!stp) {
2897                         status = stateid_error_map(stateid);
2898                         goto out;
2899                 }
2900                 if (nfs4_check_fh(current_fh, stp))
2901                         goto out;
2902                 if (!stp->st_stateowner->so_confirmed)
2903                         goto out;
2904                 status = check_stateid_generation(stateid, &stp->st_stateid,
2905                                                   flags);
2906                 if (status)
2907                         goto out;
2908                 status = nfs4_check_openmode(stp, flags);
2909                 if (status)
2910                         goto out;
2911                 renew_client(stp->st_stateowner->so_client);
2912                 if (filpp)
2913                         *filpp = stp->st_vfs_file;
2914         }
2915         status = nfs_ok;
2916 out:
2917         return status;
2918 }
2919
2920 static inline int
2921 setlkflg (int type)
2922 {
2923         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2924                 RD_STATE : WR_STATE;
2925 }
2926
2927 /* 
2928  * Checks for sequence id mutating operations. 
2929  */
2930 static __be32
2931 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2932                          stateid_t *stateid, int flags,
2933                          struct nfs4_stateowner **sopp,
2934                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2935 {
2936         struct nfs4_stateid *stp;
2937         struct nfs4_stateowner *sop;
2938         struct svc_fh *current_fh = &cstate->current_fh;
2939         __be32 status;
2940
2941         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2942                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2943                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2944                 stateid->si_generation);
2945
2946         *stpp = NULL;
2947         *sopp = NULL;
2948
2949         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2950                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2951                 return nfserr_bad_stateid;
2952         }
2953
2954         if (STALE_STATEID(stateid))
2955                 return nfserr_stale_stateid;
2956
2957         if (nfsd4_has_session(cstate))
2958                 flags |= HAS_SESSION;
2959
2960         /*
2961         * We return BAD_STATEID if filehandle doesn't match stateid, 
2962         * the confirmed flag is incorrecly set, or the generation 
2963         * number is incorrect.  
2964         */
2965         stp = find_stateid(stateid, flags);
2966         if (stp == NULL) {
2967                 /*
2968                  * Also, we should make sure this isn't just the result of
2969                  * a replayed close:
2970                  */
2971                 sop = search_close_lru(stateid->si_stateownerid, flags);
2972                 if (sop == NULL)
2973                         return stateid_error_map(stateid);
2974                 *sopp = sop;
2975                 goto check_replay;
2976         }
2977
2978         *stpp = stp;
2979         *sopp = sop = stp->st_stateowner;
2980
2981         if (lock) {
2982                 clientid_t *lockclid = &lock->v.new.clientid;
2983                 struct nfs4_client *clp = sop->so_client;
2984                 int lkflg = 0;
2985                 __be32 status;
2986
2987                 lkflg = setlkflg(lock->lk_type);
2988
2989                 if (lock->lk_is_new) {
2990                         if (!sop->so_is_open_owner)
2991                                 return nfserr_bad_stateid;
2992                         if (!(flags & HAS_SESSION) &&
2993                             !same_clid(&clp->cl_clientid, lockclid))
2994                                 return nfserr_bad_stateid;
2995                         /* stp is the open stateid */
2996                         status = nfs4_check_openmode(stp, lkflg);
2997                         if (status)
2998                                 return status;
2999                 } else {
3000                         /* stp is the lock stateid */
3001                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3002                         if (status)
3003                                 return status;
3004                }
3005         }
3006
3007         if (nfs4_check_fh(current_fh, stp)) {
3008                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3009                 return nfserr_bad_stateid;
3010         }
3011
3012         /*
3013         *  We now validate the seqid and stateid generation numbers.
3014         *  For the moment, we ignore the possibility of 
3015         *  generation number wraparound.
3016         */
3017         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3018                 goto check_replay;
3019
3020         if (sop->so_confirmed && flags & CONFIRM) {
3021                 dprintk("NFSD: preprocess_seqid_op: expected"
3022                                 " unconfirmed stateowner!\n");
3023                 return nfserr_bad_stateid;
3024         }
3025         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3026                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3027                                 " confirmed yet!\n");
3028                 return nfserr_bad_stateid;
3029         }
3030         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3031         if (status)
3032                 return status;
3033         renew_client(sop->so_client);
3034         return nfs_ok;
3035
3036 check_replay:
3037         if (seqid == sop->so_seqid - 1) {
3038                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3039                 /* indicate replay to calling function */
3040                 return nfserr_replay_me;
3041         }
3042         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3043                         sop->so_seqid, seqid);
3044         *sopp = NULL;
3045         return nfserr_bad_seqid;
3046 }
3047
3048 __be32
3049 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3050                    struct nfsd4_open_confirm *oc)
3051 {
3052         __be32 status;
3053         struct nfs4_stateowner *sop;
3054         struct nfs4_stateid *stp;
3055
3056         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3057                         (int)cstate->current_fh.fh_dentry->d_name.len,
3058                         cstate->current_fh.fh_dentry->d_name.name);
3059
3060         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3061         if (status)
3062                 return status;
3063
3064         nfs4_lock_state();
3065
3066         if ((status = nfs4_preprocess_seqid_op(cstate,
3067                                         oc->oc_seqid, &oc->oc_req_stateid,
3068                                         CONFIRM | OPEN_STATE,
3069                                         &oc->oc_stateowner, &stp, NULL)))
3070                 goto out; 
3071
3072         sop = oc->oc_stateowner;
3073         sop->so_confirmed = 1;
3074         update_stateid(&stp->st_stateid);
3075         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3076         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
3077                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3078                          stp->st_stateid.si_boot,
3079                          stp->st_stateid.si_stateownerid,
3080                          stp->st_stateid.si_fileid,
3081                          stp->st_stateid.si_generation);
3082
3083         nfsd4_create_clid_dir(sop->so_client);
3084 out:
3085         if (oc->oc_stateowner) {
3086                 nfs4_get_stateowner(oc->oc_stateowner);
3087                 cstate->replay_owner = oc->oc_stateowner;
3088         }
3089         nfs4_unlock_state();
3090         return status;
3091 }
3092
3093
3094 /*
3095  * unset all bits in union bitmap (bmap) that
3096  * do not exist in share (from successful OPEN_DOWNGRADE)
3097  */
3098 static void
3099 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3100 {
3101         int i;
3102         for (i = 1; i < 4; i++) {
3103                 if ((i & access) != i)
3104                         __clear_bit(i, bmap);
3105         }
3106 }
3107
3108 static void
3109 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3110 {
3111         int i;
3112         for (i = 0; i < 4; i++) {
3113                 if ((i & deny) != i)
3114                         __clear_bit(i, bmap);
3115         }
3116 }
3117
3118 __be32
3119 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3120                      struct nfsd4_compound_state *cstate,
3121                      struct nfsd4_open_downgrade *od)
3122 {
3123         __be32 status;
3124         struct nfs4_stateid *stp;
3125         unsigned int share_access;
3126
3127         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3128                         (int)cstate->current_fh.fh_dentry->d_name.len,
3129                         cstate->current_fh.fh_dentry->d_name.name);
3130
3131         if (!access_valid(od->od_share_access, cstate->minorversion)
3132                         || !deny_valid(od->od_share_deny))
3133                 return nfserr_inval;
3134
3135         nfs4_lock_state();
3136         if ((status = nfs4_preprocess_seqid_op(cstate,
3137                                         od->od_seqid,
3138                                         &od->od_stateid, 
3139                                         OPEN_STATE,
3140                                         &od->od_stateowner, &stp, NULL)))
3141                 goto out; 
3142
3143         status = nfserr_inval;
3144         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3145                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3146                         stp->st_access_bmap, od->od_share_access);
3147                 goto out;
3148         }
3149         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3150                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3151                         stp->st_deny_bmap, od->od_share_deny);
3152                 goto out;
3153         }
3154         set_access(&share_access, stp->st_access_bmap);
3155         nfs4_file_downgrade(stp->st_vfs_file,
3156                             share_access & ~od->od_share_access);
3157
3158         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3159         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3160
3161         update_stateid(&stp->st_stateid);
3162         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3163         status = nfs_ok;
3164 out:
3165         if (od->od_stateowner) {
3166                 nfs4_get_stateowner(od->od_stateowner);
3167                 cstate->replay_owner = od->od_stateowner;
3168         }
3169         nfs4_unlock_state();
3170         return status;
3171 }
3172
3173 /*
3174  * nfs4_unlock_state() called after encode
3175  */
3176 __be32
3177 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3178             struct nfsd4_close *close)
3179 {
3180         __be32 status;
3181         struct nfs4_stateid *stp;
3182
3183         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3184                         (int)cstate->current_fh.fh_dentry->d_name.len,
3185                         cstate->current_fh.fh_dentry->d_name.name);
3186
3187         nfs4_lock_state();
3188         /* check close_lru for replay */
3189         if ((status = nfs4_preprocess_seqid_op(cstate,
3190                                         close->cl_seqid,
3191                                         &close->cl_stateid, 
3192                                         OPEN_STATE | CLOSE_STATE,
3193                                         &close->cl_stateowner, &stp, NULL)))
3194                 goto out; 
3195         status = nfs_ok;
3196         update_stateid(&stp->st_stateid);
3197         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3198
3199         /* release_stateid() calls nfsd_close() if needed */
3200         release_open_stateid(stp);
3201
3202         /* place unused nfs4_stateowners on so_close_lru list to be
3203          * released by the laundromat service after the lease period
3204          * to enable us to handle CLOSE replay
3205          */
3206         if (list_empty(&close->cl_stateowner->so_stateids))
3207                 move_to_close_lru(close->cl_stateowner);
3208 out:
3209         if (close->cl_stateowner) {
3210                 nfs4_get_stateowner(close->cl_stateowner);
3211                 cstate->replay_owner = close->cl_stateowner;
3212         }
3213         nfs4_unlock_state();
3214         return status;
3215 }
3216
3217 __be32
3218 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3219                   struct nfsd4_delegreturn *dr)
3220 {
3221         struct nfs4_delegation *dp;
3222         stateid_t *stateid = &dr->dr_stateid;
3223         struct inode *inode;
3224         __be32 status;
3225         int flags = 0;
3226
3227         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3228                 return status;
3229         inode = cstate->current_fh.fh_dentry->d_inode;
3230
3231         if (nfsd4_has_session(cstate))
3232                 flags |= HAS_SESSION;
3233         nfs4_lock_state();
3234         status = nfserr_bad_stateid;
3235         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3236                 goto out;
3237         status = nfserr_stale_stateid;
3238         if (STALE_STATEID(stateid))
3239                 goto out;
3240         status = nfserr_bad_stateid;
3241         if (!is_delegation_stateid(stateid))
3242                 goto out;
3243         dp = find_delegation_stateid(inode, stateid);
3244         if (!dp) {
3245                 status = stateid_error_map(stateid);
3246                 goto out;
3247         }
3248         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3249         if (status)
3250                 goto out;
3251         renew_client(dp->dl_client);
3252
3253         unhash_delegation(dp);
3254 out:
3255         nfs4_unlock_state();
3256
3257         return status;
3258 }
3259
3260
3261 /* 
3262  * Lock owner state (byte-range locks)
3263  */
3264 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3265 #define LOCK_HASH_BITS              8
3266 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3267 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3268
3269 static inline u64
3270 end_offset(u64 start, u64 len)
3271 {
3272         u64 end;
3273
3274         end = start + len;
3275         return end >= start ? end: NFS4_MAX_UINT64;
3276 }
3277
3278 /* last octet in a range */
3279 static inline u64
3280 last_byte_offset(u64 start, u64 len)
3281 {
3282         u64 end;
3283
3284         BUG_ON(!len);
3285         end = start + len;
3286         return end > start ? end - 1: NFS4_MAX_UINT64;
3287 }
3288
3289 #define lockownerid_hashval(id) \
3290         ((id) & LOCK_HASH_MASK)
3291
3292 static inline unsigned int
3293 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3294                 struct xdr_netobj *ownername)
3295 {
3296         return (file_hashval(inode) + cl_id
3297                         + opaque_hashval(ownername->data, ownername->len))
3298                 & LOCK_HASH_MASK;
3299 }
3300
3301 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3302 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3303 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3304
3305 static struct nfs4_stateid *
3306 find_stateid(stateid_t *stid, int flags)
3307 {
3308         struct nfs4_stateid *local;
3309         u32 st_id = stid->si_stateownerid;
3310         u32 f_id = stid->si_fileid;
3311         unsigned int hashval;
3312
3313         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3314         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3315                 hashval = stateid_hashval(st_id, f_id);
3316                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3317                         if ((local->st_stateid.si_stateownerid == st_id) &&
3318                             (local->st_stateid.si_fileid == f_id))
3319                                 return local;
3320                 }
3321         } 
3322
3323         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3324                 hashval = stateid_hashval(st_id, f_id);
3325                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3326                         if ((local->st_stateid.si_stateownerid == st_id) &&
3327                             (local->st_stateid.si_fileid == f_id))
3328                                 return local;
3329                 }
3330         }
3331         return NULL;
3332 }
3333
3334 static struct nfs4_delegation *
3335 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3336 {
3337         struct nfs4_file *fp;
3338         struct nfs4_delegation *dl;
3339
3340         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3341                     stid->si_boot, stid->si_stateownerid,
3342                     stid->si_fileid, stid->si_generation);
3343
3344         fp = find_file(ino);
3345         if (!fp)
3346                 return NULL;
3347         dl = find_delegation_file(fp, stid);
3348         put_nfs4_file(fp);
3349         return dl;
3350 }
3351
3352 /*
3353  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3354  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3355  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3356  * locking, this prevents us from being completely protocol-compliant.  The
3357  * real solution to this problem is to start using unsigned file offsets in
3358  * the VFS, but this is a very deep change!
3359  */
3360 static inline void
3361 nfs4_transform_lock_offset(struct file_lock *lock)
3362 {
3363         if (lock->fl_start < 0)
3364                 lock->fl_start = OFFSET_MAX;
3365         if (lock->fl_end < 0)
3366                 lock->fl_end = OFFSET_MAX;
3367 }
3368
3369 /* Hack!: For now, we're defining this just so we can use a pointer to it
3370  * as a unique cookie to identify our (NFSv4's) posix locks. */
3371 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3372 };
3373
3374 static inline void
3375 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3376 {
3377         struct nfs4_stateowner *sop;
3378         unsigned int hval;
3379
3380         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3381                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3382                 hval = lockownerid_hashval(sop->so_id);
3383                 kref_get(&sop->so_ref);
3384                 deny->ld_sop = sop;
3385                 deny->ld_clientid = sop->so_client->cl_clientid;
3386         } else {
3387                 deny->ld_sop = NULL;
3388                 deny->ld_clientid.cl_boot = 0;
3389                 deny->ld_clientid.cl_id = 0;
3390         }
3391         deny->ld_start = fl->fl_start;
3392         deny->ld_length = NFS4_MAX_UINT64;
3393         if (fl->fl_end != NFS4_MAX_UINT64)
3394                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3395         deny->ld_type = NFS4_READ_LT;
3396         if (fl->fl_type != F_RDLCK)
3397                 deny->ld_type = NFS4_WRITE_LT;
3398 }
3399
3400 static struct nfs4_stateowner *
3401 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3402                 struct xdr_netobj *owner)
3403 {
3404         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3405         struct nfs4_stateowner *op;
3406
3407         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3408                 if (same_owner_str(op, owner, clid))
3409                         return op;
3410         }
3411         return NULL;
3412 }
3413
3414 /*
3415  * Alloc a lock owner structure.
3416  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3417  * occured. 
3418  *
3419  * strhashval = lock_ownerstr_hashval 
3420  */
3421
3422 static struct nfs4_stateowner *
3423 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3424         struct nfs4_stateowner *sop;
3425         struct nfs4_replay *rp;
3426         unsigned int idhashval;
3427
3428         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3429                 return NULL;
3430         idhashval = lockownerid_hashval(current_ownerid);
3431         INIT_LIST_HEAD(&sop->so_idhash);
3432         INIT_LIST_HEAD(&sop->so_strhash);
3433         INIT_LIST_HEAD(&sop->so_perclient);
3434         INIT_LIST_HEAD(&sop->so_stateids);
3435         INIT_LIST_HEAD(&sop->so_perstateid);
3436         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3437         sop->so_time = 0;
3438         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3439         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3440         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3441         sop->so_is_open_owner = 0;
3442         sop->so_id = current_ownerid++;
3443         sop->so_client = clp;
3444         /* It is the openowner seqid that will be incremented in encode in the
3445          * case of new lockowners; so increment the lock seqid manually: */
3446         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3447         sop->so_confirmed = 1;
3448         rp = &sop->so_replay;
3449         rp->rp_status = nfserr_serverfault;
3450         rp->rp_buflen = 0;
3451         rp->rp_buf = rp->rp_ibuf;
3452         return sop;
3453 }
3454
3455 static struct nfs4_stateid *
3456 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3457 {
3458         struct nfs4_stateid *stp;
3459         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3460
3461         stp = nfs4_alloc_stateid();
3462         if (stp == NULL)
3463                 goto out;
3464         INIT_LIST_HEAD(&stp->st_hash);
3465         INIT_LIST_HEAD(&stp->st_perfile);
3466         INIT_LIST_HEAD(&stp->st_perstateowner);
3467         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3468         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3469         list_add(&stp->st_perfile, &fp->fi_stateids);
3470         list_add(&stp->st_perstateowner, &sop->so_stateids);
3471         stp->st_stateowner = sop;
3472         get_nfs4_file(fp);
3473         stp->st_file = fp;
3474         stp->st_stateid.si_boot = get_seconds();
3475         stp->st_stateid.si_stateownerid = sop->so_id;
3476         stp->st_stateid.si_fileid = fp->fi_id;
3477         stp->st_stateid.si_generation = 0;
3478         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3479         stp->st_access_bmap = open_stp->st_access_bmap;
3480         stp->st_deny_bmap = open_stp->st_deny_bmap;
3481         stp->st_openstp = open_stp;
3482
3483 out:
3484         return stp;
3485 }
3486
3487 static int
3488 check_lock_length(u64 offset, u64 length)
3489 {
3490         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3491              LOFF_OVERFLOW(offset, length)));
3492 }
3493
3494 /*
3495  *  LOCK operation 
3496  */
3497 __be32
3498 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3499            struct nfsd4_lock *lock)
3500 {
3501         struct nfs4_stateowner *open_sop = NULL;
3502         struct nfs4_stateowner *lock_sop = NULL;
3503         struct nfs4_stateid *lock_stp;
3504         struct file *filp;
3505         struct file_lock file_lock;
3506         struct file_lock conflock;
3507         __be32 status = 0;
3508         unsigned int strhashval;
3509         unsigned int cmd;
3510         int err;
3511
3512         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3513                 (long long) lock->lk_offset,
3514                 (long long) lock->lk_length);
3515
3516         if (check_lock_length(lock->lk_offset, lock->lk_length))
3517                  return nfserr_inval;
3518
3519         if ((status = fh_verify(rqstp, &cstate->current_fh,
3520                                 S_IFREG, NFSD_MAY_LOCK))) {
3521                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3522                 return status;
3523         }
3524
3525         nfs4_lock_state();
3526
3527         if (lock->lk_is_new) {
3528                 /*
3529                  * Client indicates that this is a new lockowner.
3530                  * Use open owner and open stateid to create lock owner and
3531                  * lock stateid.
3532                  */
3533                 struct nfs4_stateid *open_stp = NULL;
3534                 struct nfs4_file *fp;
3535                 
3536                 status = nfserr_stale_clientid;
3537                 if (!nfsd4_has_session(cstate) &&
3538                     STALE_CLIENTID(&lock->lk_new_clientid))
3539                         goto out;
3540
3541                 /* validate and update open stateid and open seqid */
3542                 status = nfs4_preprocess_seqid_op(cstate,
3543                                         lock->lk_new_open_seqid,
3544                                         &lock->lk_new_open_stateid,
3545                                         OPEN_STATE,
3546                                         &lock->lk_replay_owner, &open_stp,
3547                                         lock);
3548                 if (status)
3549                         goto out;
3550                 open_sop = lock->lk_replay_owner;
3551                 /* create lockowner and lock stateid */
3552                 fp = open_stp->st_file;
3553                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3554                                 open_sop->so_client->cl_clientid.cl_id, 
3555                                 &lock->v.new.owner);
3556                 /* XXX: Do we need to check for duplicate stateowners on
3557                  * the same file, or should they just be allowed (and
3558                  * create new stateids)? */
3559                 status = nfserr_resource;
3560                 lock_sop = alloc_init_lock_stateowner(strhashval,
3561                                 open_sop->so_client, open_stp, lock);
3562                 if (lock_sop == NULL)
3563                         goto out;
3564                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3565                 if (lock_stp == NULL)
3566                         goto out;
3567         } else {
3568                 /* lock (lock owner + lock stateid) already exists */
3569                 status = nfs4_preprocess_seqid_op(cstate,
3570                                        lock->lk_old_lock_seqid, 
3571                                        &lock->lk_old_lock_stateid, 
3572                                        LOCK_STATE,
3573                                        &lock->lk_replay_owner, &lock_stp, lock);
3574                 if (status)
3575                         goto out;
3576                 lock_sop = lock->lk_replay_owner;
3577         }
3578         /* lock->lk_replay_owner and lock_stp have been created or found */
3579         filp = lock_stp->st_vfs_file;
3580
3581         status = nfserr_grace;
3582         if (locks_in_grace() && !lock->lk_reclaim)
3583                 goto out;
3584         status = nfserr_no_grace;
3585         if (!locks_in_grace() && lock->lk_reclaim)
3586                 goto out;
3587
3588         locks_init_lock(&file_lock);
3589         switch (lock->lk_type) {
3590                 case NFS4_READ_LT:
3591                 case NFS4_READW_LT:
3592                         file_lock.fl_type = F_RDLCK;
3593                         cmd = F_SETLK;
3594                 break;
3595                 case NFS4_WRITE_LT:
3596                 case NFS4_WRITEW_LT:
3597                         file_lock.fl_type = F_WRLCK;
3598                         cmd = F_SETLK;
3599                 break;
3600                 default:
3601                         status = nfserr_inval;
3602                 goto out;
3603         }
3604         file_lock.fl_owner = (fl_owner_t)lock_sop;
3605         file_lock.fl_pid = current->tgid;
3606         file_lock.fl_file = filp;
3607         file_lock.fl_flags = FL_POSIX;
3608         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3609
3610         file_lock.fl_start = lock->lk_offset;
3611         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3612         nfs4_transform_lock_offset(&file_lock);
3613
3614         /*
3615         * Try to lock the file in the VFS.
3616         * Note: locks.c uses the BKL to protect the inode's lock list.
3617         */
3618
3619         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3620         switch (-err) {
3621         case 0: /* success! */
3622                 update_stateid(&lock_stp->st_stateid);
3623                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3624                                 sizeof(stateid_t));
3625                 status = 0;
3626                 break;
3627         case (EAGAIN):          /* conflock holds conflicting lock */
3628                 status = nfserr_denied;
3629                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3630                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3631                 break;
3632         case (EDEADLK):
3633                 status = nfserr_deadlock;
3634                 break;
3635         default:        
3636                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3637                 status = nfserr_resource;
3638                 break;
3639         }
3640 out:
3641         if (status && lock->lk_is_new && lock_sop)
3642                 release_lockowner(lock_sop);
3643         if (lock->lk_replay_owner) {
3644                 nfs4_get_stateowner(lock->lk_replay_owner);
3645                 cstate->replay_owner = lock->lk_replay_owner;
3646         }
3647         nfs4_unlock_state();
3648         return status;
3649 }
3650
3651 /*
3652  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3653  * so we do a temporary open here just to get an open file to pass to
3654  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3655  * inode operation.)
3656  */
3657 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3658 {
3659         struct file *file;
3660         int err;
3661
3662         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3663         if (err)
3664                 return err;
3665         err = vfs_test_lock(file, lock);
3666         nfsd_close(file);
3667         return err;
3668 }
3669
3670 /*
3671  * LOCKT operation
3672  */
3673 __be32
3674 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3675             struct nfsd4_lockt *lockt)
3676 {
3677         struct inode *inode;
3678         struct file_lock file_lock;
3679         int error;
3680         __be32 status;
3681
3682         if (locks_in_grace())
3683                 return nfserr_grace;
3684
3685         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3686                  return nfserr_inval;
3687
3688         lockt->lt_stateowner = NULL;
3689         nfs4_lock_state();
3690
3691         status = nfserr_stale_clientid;
3692         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3693                 goto out;
3694
3695         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3696                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3697                 if (status == nfserr_symlink)
3698                         status = nfserr_inval;
3699                 goto out;
3700         }
3701
3702         inode = cstate->current_fh.fh_dentry->d_inode;
3703         locks_init_lock(&file_lock);
3704         switch (lockt->lt_type) {
3705                 case NFS4_READ_LT:
3706                 case NFS4_READW_LT:
3707                         file_lock.fl_type = F_RDLCK;
3708                 break;
3709                 case NFS4_WRITE_LT:
3710                 case NFS4_WRITEW_LT:
3711                         file_lock.fl_type = F_WRLCK;
3712                 break;
3713                 default:
3714                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3715                         status = nfserr_inval;
3716                 goto out;
3717         }
3718
3719         lockt->lt_stateowner = find_lockstateowner_str(inode,
3720                         &lockt->lt_clientid, &lockt->lt_owner);
3721         if (lockt->lt_stateowner)
3722                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3723         file_lock.fl_pid = current->tgid;
3724         file_lock.fl_flags = FL_POSIX;
3725
3726         file_lock.fl_start = lockt->lt_offset;
3727         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3728
3729         nfs4_transform_lock_offset(&file_lock);
3730
3731         status = nfs_ok;
3732         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3733         if (error) {
3734                 status = nfserrno(error);
3735                 goto out;
3736         }
3737         if (file_lock.fl_type != F_UNLCK) {
3738                 status = nfserr_denied;
3739                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3740         }
3741 out:
3742         nfs4_unlock_state();
3743         return status;
3744 }
3745
3746 __be32
3747 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3748             struct nfsd4_locku *locku)
3749 {
3750         struct nfs4_stateid *stp;
3751         struct file *filp = NULL;
3752         struct file_lock file_lock;
3753         __be32 status;
3754         int err;
3755                                                         
3756         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3757                 (long long) locku->lu_offset,
3758                 (long long) locku->lu_length);
3759
3760         if (check_lock_length(locku->lu_offset, locku->lu_length))
3761                  return nfserr_inval;
3762
3763         nfs4_lock_state();
3764                                                                                 
3765         if ((status = nfs4_preprocess_seqid_op(cstate,
3766                                         locku->lu_seqid, 
3767                                         &locku->lu_stateid, 
3768                                         LOCK_STATE,
3769                                         &locku->lu_stateowner, &stp, NULL)))
3770                 goto out;
3771
3772         filp = stp->st_vfs_file;
3773         BUG_ON(!filp);
3774         locks_init_lock(&file_lock);
3775         file_lock.fl_type = F_UNLCK;
3776         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3777         file_lock.fl_pid = current->tgid;
3778         file_lock.fl_file = filp;
3779         file_lock.fl_flags = FL_POSIX; 
3780         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3781         file_lock.fl_start = locku->lu_offset;
3782
3783         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3784         nfs4_transform_lock_offset(&file_lock);
3785
3786         /*
3787         *  Try to unlock the file in the VFS.
3788         */
3789         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3790         if (err) {
3791                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3792                 goto out_nfserr;
3793         }
3794         /*
3795         * OK, unlock succeeded; the only thing left to do is update the stateid.
3796         */
3797         update_stateid(&stp->st_stateid);
3798         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3799
3800 out:
3801         if (locku->lu_stateowner) {
3802                 nfs4_get_stateowner(locku->lu_stateowner);
3803                 cstate->replay_owner = locku->lu_stateowner;
3804         }
3805         nfs4_unlock_state();
3806         return status;
3807
3808 out_nfserr:
3809         status = nfserrno(err);
3810         goto out;
3811 }
3812
3813 /*
3814  * returns
3815  *      1: locks held by lockowner
3816  *      0: no locks held by lockowner
3817  */
3818 static int
3819 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3820 {
3821         struct file_lock **flpp;
3822         struct inode *inode = filp->f_path.dentry->d_inode;
3823         int status = 0;
3824
3825         lock_kernel();
3826         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3827                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3828                         status = 1;
3829                         goto out;
3830                 }
3831         }
3832 out:
3833         unlock_kernel();
3834         return status;
3835 }
3836
3837 __be32
3838 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3839                         struct nfsd4_compound_state *cstate,
3840                         struct nfsd4_release_lockowner *rlockowner)
3841 {
3842         clientid_t *clid = &rlockowner->rl_clientid;
3843         struct nfs4_stateowner *sop;
3844         struct nfs4_stateid *stp;
3845         struct xdr_netobj *owner = &rlockowner->rl_owner;
3846         struct list_head matches;
3847         int i;
3848         __be32 status;
3849
3850         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3851                 clid->cl_boot, clid->cl_id);
3852
3853         /* XXX check for lease expiration */
3854
3855         status = nfserr_stale_clientid;
3856         if (STALE_CLIENTID(clid))
3857                 return status;
3858
3859         nfs4_lock_state();
3860
3861         status = nfserr_locks_held;
3862         /* XXX: we're doing a linear search through all the lockowners.
3863          * Yipes!  For now we'll just hope clients aren't really using
3864          * release_lockowner much, but eventually we have to fix these
3865          * data structures. */
3866         INIT_LIST_HEAD(&matches);
3867         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3868                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3869                         if (!same_owner_str(sop, owner, clid))
3870                                 continue;
3871                         list_for_each_entry(stp, &sop->so_stateids,
3872                                         st_perstateowner) {
3873                                 if (check_for_locks(stp->st_vfs_file, sop))
3874                                         goto out;
3875                                 /* Note: so_perclient unused for lockowners,
3876                                  * so it's OK to fool with here. */
3877                                 list_add(&sop->so_perclient, &matches);
3878                         }
3879                 }
3880         }
3881         /* Clients probably won't expect us to return with some (but not all)
3882          * of the lockowner state released; so don't release any until all
3883          * have been checked. */
3884         status = nfs_ok;
3885         while (!list_empty(&matches)) {
3886                 sop = list_entry(matches.next, struct nfs4_stateowner,
3887                                                                 so_perclient);
3888                 /* unhash_stateowner deletes so_perclient only
3889                  * for openowners. */
3890                 list_del(&sop->so_perclient);
3891                 release_lockowner(sop);
3892         }
3893 out:
3894         nfs4_unlock_state();
3895         return status;
3896 }
3897
3898 static inline struct nfs4_client_reclaim *
3899 alloc_reclaim(void)
3900 {
3901         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3902 }
3903
3904 int
3905 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3906 {
3907         unsigned int strhashval = clientstr_hashval(name);
3908         struct nfs4_client *clp;
3909
3910         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3911         return clp ? 1 : 0;
3912 }
3913
3914 /*
3915  * failure => all reset bets are off, nfserr_no_grace...
3916  */
3917 int
3918 nfs4_client_to_reclaim(const char *name)
3919 {
3920         unsigned int strhashval;
3921         struct nfs4_client_reclaim *crp = NULL;
3922
3923         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3924         crp = alloc_reclaim();
3925         if (!crp)
3926                 return 0;
3927         strhashval = clientstr_hashval(name);
3928         INIT_LIST_HEAD(&crp->cr_strhash);
3929         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3930         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3931         reclaim_str_hashtbl_size++;
3932         return 1;
3933 }
3934
3935 static void
3936 nfs4_release_reclaim(void)
3937 {
3938         struct nfs4_client_reclaim *crp = NULL;
3939         int i;
3940
3941         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3942                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3943                         crp = list_entry(reclaim_str_hashtbl[i].next,
3944                                         struct nfs4_client_reclaim, cr_strhash);
3945                         list_del(&crp->cr_strhash);
3946                         kfree(crp);
3947                         reclaim_str_hashtbl_size--;
3948                 }
3949         }
3950         BUG_ON(reclaim_str_hashtbl_size);
3951 }
3952
3953 /*
3954  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3955 static struct nfs4_client_reclaim *
3956 nfs4_find_reclaim_client(clientid_t *clid)
3957 {
3958         unsigned int strhashval;
3959         struct nfs4_client *clp;
3960         struct nfs4_client_reclaim *crp = NULL;
3961
3962
3963         /* find clientid in conf_id_hashtbl */
3964         clp = find_confirmed_client(clid);
3965         if (clp == NULL)
3966                 return NULL;
3967
3968         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3969                             clp->cl_name.len, clp->cl_name.data,
3970                             clp->cl_recdir);
3971
3972         /* find clp->cl_name in reclaim_str_hashtbl */
3973         strhashval = clientstr_hashval(clp->cl_recdir);
3974         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3975                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3976                         return crp;
3977                 }
3978         }
3979         return NULL;
3980 }
3981
3982 /*
3983 * Called from OPEN. Look for clientid in reclaim list.
3984 */
3985 __be32
3986 nfs4_check_open_reclaim(clientid_t *clid)
3987 {
3988         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3989 }
3990
3991 /* initialization to perform at module load time: */
3992
3993 int
3994 nfs4_state_init(void)
3995 {
3996         int i, status;
3997
3998         status = nfsd4_init_slabs();
3999         if (status)
4000                 return status;
4001         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4002                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4003                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4004                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4005                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4006                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4007         }
4008         for (i = 0; i < SESSION_HASH_SIZE; i++)
4009                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4010         for (i = 0; i < FILE_HASH_SIZE; i++) {
4011                 INIT_LIST_HEAD(&file_hashtbl[i]);
4012         }
4013         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4014                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4015                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4016         }
4017         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4018                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4019                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4020         }
4021         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4022                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4023                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4024         }
4025         memset(&onestateid, ~0, sizeof(stateid_t));
4026         INIT_LIST_HEAD(&close_lru);
4027         INIT_LIST_HEAD(&client_lru);
4028         INIT_LIST_HEAD(&del_recall_lru);
4029         reclaim_str_hashtbl_size = 0;
4030         return 0;
4031 }
4032
4033 static void
4034 nfsd4_load_reboot_recovery_data(void)
4035 {
4036         int status;
4037
4038         nfs4_lock_state();
4039         nfsd4_init_recdir(user_recovery_dirname);
4040         status = nfsd4_recdir_load();
4041         nfs4_unlock_state();
4042         if (status)
4043                 printk("NFSD: Failure reading reboot recovery data\n");
4044 }
4045
4046 unsigned long
4047 get_nfs4_grace_period(void)
4048 {
4049         return max(user_lease_time, lease_time) * HZ;
4050 }
4051
4052 /*
4053  * Since the lifetime of a delegation isn't limited to that of an open, a
4054  * client may quite reasonably hang on to a delegation as long as it has
4055  * the inode cached.  This becomes an obvious problem the first time a
4056  * client's inode cache approaches the size of the server's total memory.
4057  *
4058  * For now we avoid this problem by imposing a hard limit on the number
4059  * of delegations, which varies according to the server's memory size.
4060  */
4061 static void
4062 set_max_delegations(void)
4063 {
4064         /*
4065          * Allow at most 4 delegations per megabyte of RAM.  Quick
4066          * estimates suggest that in the worst case (where every delegation
4067          * is for a different inode), a delegation could take about 1.5K,
4068          * giving a worst case usage of about 6% of memory.
4069          */
4070         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4071 }
4072
4073 /* initialization to perform when the nfsd service is started: */
4074
4075 static void
4076 __nfs4_state_start(void)
4077 {
4078         unsigned long grace_time;
4079
4080         boot_time = get_seconds();
4081         grace_time = get_nfs4_grace_period();
4082         lease_time = user_lease_time;
4083         locks_start_grace(&nfsd4_manager);
4084         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4085                grace_time/HZ);
4086         laundry_wq = create_singlethread_workqueue("nfsd4");
4087         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4088         set_max_delegations();
4089 }
4090
4091 void
4092 nfs4_state_start(void)
4093 {
4094         if (nfs4_init)
4095                 return;
4096         nfsd4_load_reboot_recovery_data();
4097         __nfs4_state_start();
4098         nfs4_init = 1;
4099         return;
4100 }
4101
4102 time_t
4103 nfs4_lease_time(void)
4104 {
4105         return lease_time;
4106 }
4107
4108 static void
4109 __nfs4_state_shutdown(void)
4110 {
4111         int i;
4112         struct nfs4_client *clp = NULL;
4113         struct nfs4_delegation *dp = NULL;
4114         struct list_head *pos, *next, reaplist;
4115
4116         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4117                 while (!list_empty(&conf_id_hashtbl[i])) {
4118                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4119                         expire_client(clp);
4120                 }
4121                 while (!list_empty(&unconf_str_hashtbl[i])) {
4122                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4123                         expire_client(clp);
4124                 }
4125         }
4126         INIT_LIST_HEAD(&reaplist);
4127         spin_lock(&recall_lock);
4128         list_for_each_safe(pos, next, &del_recall_lru) {
4129                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4130                 list_move(&dp->dl_recall_lru, &reaplist);
4131         }
4132         spin_unlock(&recall_lock);
4133         list_for_each_safe(pos, next, &reaplist) {
4134                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4135                 list_del_init(&dp->dl_recall_lru);
4136                 unhash_delegation(dp);
4137         }
4138
4139         nfsd4_shutdown_recdir();
4140         nfs4_init = 0;
4141 }
4142
4143 void
4144 nfs4_state_shutdown(void)
4145 {
4146         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4147         destroy_workqueue(laundry_wq);
4148         locks_end_grace(&nfsd4_manager);
4149         nfs4_lock_state();
4150         nfs4_release_reclaim();
4151         __nfs4_state_shutdown();
4152         nfs4_unlock_state();
4153 }
4154
4155 /*
4156  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4157  * accessed when nfsd is starting.
4158  */
4159 static void
4160 nfs4_set_recdir(char *recdir)
4161 {
4162         strcpy(user_recovery_dirname, recdir);
4163 }
4164
4165 /*
4166  * Change the NFSv4 recovery directory to recdir.
4167  */
4168 int
4169 nfs4_reset_recoverydir(char *recdir)
4170 {
4171         int status;
4172         struct path path;
4173
4174         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4175         if (status)
4176                 return status;
4177         status = -ENOTDIR;
4178         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4179                 nfs4_set_recdir(recdir);
4180                 status = 0;
4181         }
4182         path_put(&path);
4183         return status;
4184 }
4185
4186 char *
4187 nfs4_recoverydir(void)
4188 {
4189         return user_recovery_dirname;
4190 }
4191
4192 /*
4193  * Called when leasetime is changed.
4194  *
4195  * The only way the protocol gives us to handle on-the-fly lease changes is to
4196  * simulate a reboot.  Instead of doing that, we just wait till the next time
4197  * we start to register any changes in lease time.  If the administrator
4198  * really wants to change the lease time *now*, they can go ahead and bring
4199  * nfsd down and then back up again after changing the lease time.
4200  *
4201  * user_lease_time is protected by nfsd_mutex since it's only really accessed
4202  * when nfsd is starting
4203  */
4204 void
4205 nfs4_reset_lease(time_t leasetime)
4206 {
4207         user_lease_time = leasetime;
4208 }