54651aa4579016c6e20e7de34e1118a6bb488504
[safe/jmp/linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58
59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
60
61 /* Globals */
62 static time_t lease_time = 90;     /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
68 static u32 nfs4_init;
69 static stateid_t zerostateid;             /* bits all 0 */
70 static stateid_t onestateid;              /* bits all 1 */
71
72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
74
75 /* forward declarations */
76 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
77 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
78 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
79 static void nfs4_set_recdir(char *recdir);
80
81 /* Locking: */
82
83 /* Currently used for almost all code touching nfsv4 state: */
84 static DEFINE_MUTEX(client_mutex);
85
86 /*
87  * Currently used for the del_recall_lru and file hash table.  In an
88  * effort to decrease the scope of the client_mutex, this spinlock may
89  * eventually cover more:
90  */
91 static DEFINE_SPINLOCK(recall_lock);
92
93 static struct kmem_cache *stateowner_slab = NULL;
94 static struct kmem_cache *file_slab = NULL;
95 static struct kmem_cache *stateid_slab = NULL;
96 static struct kmem_cache *deleg_slab = NULL;
97
98 void
99 nfs4_lock_state(void)
100 {
101         mutex_lock(&client_mutex);
102 }
103
104 void
105 nfs4_unlock_state(void)
106 {
107         mutex_unlock(&client_mutex);
108 }
109
110 static inline u32
111 opaque_hashval(const void *ptr, int nbytes)
112 {
113         unsigned char *cptr = (unsigned char *) ptr;
114
115         u32 x = 0;
116         while (nbytes--) {
117                 x *= 37;
118                 x += *cptr++;
119         }
120         return x;
121 }
122
123 static struct list_head del_recall_lru;
124
125 static inline void
126 put_nfs4_file(struct nfs4_file *fi)
127 {
128         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
129                 list_del(&fi->fi_hash);
130                 spin_unlock(&recall_lock);
131                 iput(fi->fi_inode);
132                 kmem_cache_free(file_slab, fi);
133         }
134 }
135
136 static inline void
137 get_nfs4_file(struct nfs4_file *fi)
138 {
139         atomic_inc(&fi->fi_ref);
140 }
141
142 static int num_delegations;
143 unsigned int max_delegations;
144
145 /*
146  * Open owner state (share locks)
147  */
148
149 /* hash tables for nfs4_stateowner */
150 #define OWNER_HASH_BITS              8
151 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
152 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
153
154 #define ownerid_hashval(id) \
155         ((id) & OWNER_HASH_MASK)
156 #define ownerstr_hashval(clientid, ownername) \
157         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
158
159 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
160 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
161
162 /* hash table for nfs4_file */
163 #define FILE_HASH_BITS                   8
164 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
165 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
166 /* hash table for (open)nfs4_stateid */
167 #define STATEID_HASH_BITS              10
168 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
169 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
170
171 #define file_hashval(x) \
172         hash_ptr(x, FILE_HASH_BITS)
173 #define stateid_hashval(owner_id, file_id)  \
174         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
175
176 static struct list_head file_hashtbl[FILE_HASH_SIZE];
177 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
178
179 static struct nfs4_delegation *
180 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
181 {
182         struct nfs4_delegation *dp;
183         struct nfs4_file *fp = stp->st_file;
184         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
185
186         dprintk("NFSD alloc_init_deleg\n");
187         if (fp->fi_had_conflict)
188                 return NULL;
189         if (num_delegations > max_delegations)
190                 return NULL;
191         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
192         if (dp == NULL)
193                 return dp;
194         num_delegations++;
195         INIT_LIST_HEAD(&dp->dl_perfile);
196         INIT_LIST_HEAD(&dp->dl_perclnt);
197         INIT_LIST_HEAD(&dp->dl_recall_lru);
198         dp->dl_client = clp;
199         get_nfs4_file(fp);
200         dp->dl_file = fp;
201         dp->dl_flock = NULL;
202         get_file(stp->st_vfs_file);
203         dp->dl_vfs_file = stp->st_vfs_file;
204         dp->dl_type = type;
205         dp->dl_recall.cbr_dp = NULL;
206         dp->dl_recall.cbr_ident = cb->cb_ident;
207         dp->dl_recall.cbr_trunc = 0;
208         dp->dl_stateid.si_boot = boot_time;
209         dp->dl_stateid.si_stateownerid = current_delegid++;
210         dp->dl_stateid.si_fileid = 0;
211         dp->dl_stateid.si_generation = 0;
212         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
213         dp->dl_time = 0;
214         atomic_set(&dp->dl_count, 1);
215         list_add(&dp->dl_perfile, &fp->fi_delegations);
216         list_add(&dp->dl_perclnt, &clp->cl_delegations);
217         return dp;
218 }
219
220 void
221 nfs4_put_delegation(struct nfs4_delegation *dp)
222 {
223         if (atomic_dec_and_test(&dp->dl_count)) {
224                 dprintk("NFSD: freeing dp %p\n",dp);
225                 put_nfs4_file(dp->dl_file);
226                 kmem_cache_free(deleg_slab, dp);
227                 num_delegations--;
228         }
229 }
230
231 /* Remove the associated file_lock first, then remove the delegation.
232  * lease_modify() is called to remove the FS_LEASE file_lock from
233  * the i_flock list, eventually calling nfsd's lock_manager
234  * fl_release_callback.
235  */
236 static void
237 nfs4_close_delegation(struct nfs4_delegation *dp)
238 {
239         struct file *filp = dp->dl_vfs_file;
240
241         dprintk("NFSD: close_delegation dp %p\n",dp);
242         dp->dl_vfs_file = NULL;
243         /* The following nfsd_close may not actually close the file,
244          * but we want to remove the lease in any case. */
245         if (dp->dl_flock)
246                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
247         nfsd_close(filp);
248 }
249
250 /* Called under the state lock. */
251 static void
252 unhash_delegation(struct nfs4_delegation *dp)
253 {
254         list_del_init(&dp->dl_perfile);
255         list_del_init(&dp->dl_perclnt);
256         spin_lock(&recall_lock);
257         list_del_init(&dp->dl_recall_lru);
258         spin_unlock(&recall_lock);
259         nfs4_close_delegation(dp);
260         nfs4_put_delegation(dp);
261 }
262
263 /* 
264  * SETCLIENTID state 
265  */
266
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS                 4
269 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
271
272 #define clientid_hashval(id) \
273         ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
276 /*
277  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278  * used in reboot/reset lease grace period processing
279  *
280  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281  * setclientid_confirmed info. 
282  *
283  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
284  * setclientid info.
285  *
286  * client_lru holds client queue ordered by nfs4_client.cl_time
287  * for lease renewal.
288  *
289  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290  * for last close replay.
291  */
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
300
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
302 {
303         list_del(&stp->st_hash);
304         list_del(&stp->st_perfile);
305         list_del(&stp->st_perstateowner);
306 }
307
308 static void free_generic_stateid(struct nfs4_stateid *stp)
309 {
310         put_nfs4_file(stp->st_file);
311         kmem_cache_free(stateid_slab, stp);
312 }
313
314 static void release_lock_stateid(struct nfs4_stateid *stp)
315 {
316         unhash_generic_stateid(stp);
317         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318         free_generic_stateid(stp);
319 }
320
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
322 {
323         struct nfs4_stateid *stp;
324
325         list_del(&sop->so_idhash);
326         list_del(&sop->so_strhash);
327         list_del(&sop->so_perstateid);
328         while (!list_empty(&sop->so_stateids)) {
329                 stp = list_first_entry(&sop->so_stateids,
330                                 struct nfs4_stateid, st_perstateowner);
331                 release_lock_stateid(stp);
332         }
333 }
334
335 static void release_lockowner(struct nfs4_stateowner *sop)
336 {
337         unhash_lockowner(sop);
338         nfs4_put_stateowner(sop);
339 }
340
341 static void
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
343 {
344         struct nfs4_stateowner *lock_sop;
345
346         while (!list_empty(&open_stp->st_lockowners)) {
347                 lock_sop = list_entry(open_stp->st_lockowners.next,
348                                 struct nfs4_stateowner, so_perstateid);
349                 /* list_del(&open_stp->st_lockowners);  */
350                 BUG_ON(lock_sop->so_is_open_owner);
351                 release_lockowner(lock_sop);
352         }
353 }
354
355 static void release_open_stateid(struct nfs4_stateid *stp)
356 {
357         unhash_generic_stateid(stp);
358         release_stateid_lockowners(stp);
359         nfsd_close(stp->st_vfs_file);
360         free_generic_stateid(stp);
361 }
362
363 static void unhash_openowner(struct nfs4_stateowner *sop)
364 {
365         struct nfs4_stateid *stp;
366
367         list_del(&sop->so_idhash);
368         list_del(&sop->so_strhash);
369         list_del(&sop->so_perclient);
370         list_del(&sop->so_perstateid); /* XXX: necessary? */
371         while (!list_empty(&sop->so_stateids)) {
372                 stp = list_first_entry(&sop->so_stateids,
373                                 struct nfs4_stateid, st_perstateowner);
374                 release_open_stateid(stp);
375         }
376 }
377
378 static void release_openowner(struct nfs4_stateowner *sop)
379 {
380         unhash_openowner(sop);
381         list_del(&sop->so_close_lru);
382         nfs4_put_stateowner(sop);
383 }
384
385 static inline void
386 renew_client(struct nfs4_client *clp)
387 {
388         /*
389         * Move client to the end to the LRU list.
390         */
391         dprintk("renewing client (clientid %08x/%08x)\n", 
392                         clp->cl_clientid.cl_boot, 
393                         clp->cl_clientid.cl_id);
394         list_move_tail(&clp->cl_lru, &client_lru);
395         clp->cl_time = get_seconds();
396 }
397
398 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
399 static int
400 STALE_CLIENTID(clientid_t *clid)
401 {
402         if (clid->cl_boot == boot_time)
403                 return 0;
404         dprintk("NFSD stale clientid (%08x/%08x)\n", 
405                         clid->cl_boot, clid->cl_id);
406         return 1;
407 }
408
409 /* 
410  * XXX Should we use a slab cache ?
411  * This type of memory management is somewhat inefficient, but we use it
412  * anyway since SETCLIENTID is not a common operation.
413  */
414 static struct nfs4_client *alloc_client(struct xdr_netobj name)
415 {
416         struct nfs4_client *clp;
417
418         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
419         if (clp == NULL)
420                 return NULL;
421         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
422         if (clp->cl_name.data == NULL) {
423                 kfree(clp);
424                 return NULL;
425         }
426         memcpy(clp->cl_name.data, name.data, name.len);
427         clp->cl_name.len = name.len;
428         return clp;
429 }
430
431 static void
432 shutdown_callback_client(struct nfs4_client *clp)
433 {
434         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
435
436         if (clnt) {
437                 /*
438                  * Callback threads take a reference on the client, so there
439                  * should be no outstanding callbacks at this point.
440                  */
441                 clp->cl_callback.cb_client = NULL;
442                 rpc_shutdown_client(clnt);
443         }
444 }
445
446 static inline void
447 free_client(struct nfs4_client *clp)
448 {
449         shutdown_callback_client(clp);
450         if (clp->cl_cred.cr_group_info)
451                 put_group_info(clp->cl_cred.cr_group_info);
452         kfree(clp->cl_principal);
453         kfree(clp->cl_name.data);
454         kfree(clp);
455 }
456
457 void
458 put_nfs4_client(struct nfs4_client *clp)
459 {
460         if (atomic_dec_and_test(&clp->cl_count))
461                 free_client(clp);
462 }
463
464 static void
465 expire_client(struct nfs4_client *clp)
466 {
467         struct nfs4_stateowner *sop;
468         struct nfs4_delegation *dp;
469         struct list_head reaplist;
470
471         dprintk("NFSD: expire_client cl_count %d\n",
472                             atomic_read(&clp->cl_count));
473
474         INIT_LIST_HEAD(&reaplist);
475         spin_lock(&recall_lock);
476         while (!list_empty(&clp->cl_delegations)) {
477                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
478                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
479                                 dp->dl_flock);
480                 list_del_init(&dp->dl_perclnt);
481                 list_move(&dp->dl_recall_lru, &reaplist);
482         }
483         spin_unlock(&recall_lock);
484         while (!list_empty(&reaplist)) {
485                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
486                 list_del_init(&dp->dl_recall_lru);
487                 unhash_delegation(dp);
488         }
489         list_del(&clp->cl_idhash);
490         list_del(&clp->cl_strhash);
491         list_del(&clp->cl_lru);
492         while (!list_empty(&clp->cl_openowners)) {
493                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
494                 release_openowner(sop);
495         }
496         put_nfs4_client(clp);
497 }
498
499 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
500 {
501         struct nfs4_client *clp;
502
503         clp = alloc_client(name);
504         if (clp == NULL)
505                 return NULL;
506         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
507         atomic_set(&clp->cl_count, 1);
508         atomic_set(&clp->cl_callback.cb_set, 0);
509         INIT_LIST_HEAD(&clp->cl_idhash);
510         INIT_LIST_HEAD(&clp->cl_strhash);
511         INIT_LIST_HEAD(&clp->cl_openowners);
512         INIT_LIST_HEAD(&clp->cl_delegations);
513         INIT_LIST_HEAD(&clp->cl_lru);
514         return clp;
515 }
516
517 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
518 {
519         memcpy(target->cl_verifier.data, source->data,
520                         sizeof(target->cl_verifier.data));
521 }
522
523 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
524 {
525         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
526         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
527 }
528
529 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
530 {
531         target->cr_uid = source->cr_uid;
532         target->cr_gid = source->cr_gid;
533         target->cr_group_info = source->cr_group_info;
534         get_group_info(target->cr_group_info);
535 }
536
537 static int same_name(const char *n1, const char *n2)
538 {
539         return 0 == memcmp(n1, n2, HEXDIR_LEN);
540 }
541
542 static int
543 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
544 {
545         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
546 }
547
548 static int
549 same_clid(clientid_t *cl1, clientid_t *cl2)
550 {
551         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
552 }
553
554 /* XXX what about NGROUP */
555 static int
556 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
557 {
558         return cr1->cr_uid == cr2->cr_uid;
559 }
560
561 static void gen_clid(struct nfs4_client *clp)
562 {
563         static u32 current_clientid = 1;
564
565         clp->cl_clientid.cl_boot = boot_time;
566         clp->cl_clientid.cl_id = current_clientid++; 
567 }
568
569 static void gen_confirm(struct nfs4_client *clp)
570 {
571         static u32 i;
572         u32 *p;
573
574         p = (u32 *)clp->cl_confirm.data;
575         *p++ = get_seconds();
576         *p++ = i++;
577 }
578
579 static int check_name(struct xdr_netobj name)
580 {
581         if (name.len == 0) 
582                 return 0;
583         if (name.len > NFS4_OPAQUE_LIMIT) {
584                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
585                 return 0;
586         }
587         return 1;
588 }
589
590 static void
591 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
592 {
593         unsigned int idhashval;
594
595         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
596         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
597         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
598         list_add_tail(&clp->cl_lru, &client_lru);
599         clp->cl_time = get_seconds();
600 }
601
602 static void
603 move_to_confirmed(struct nfs4_client *clp)
604 {
605         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
606         unsigned int strhashval;
607
608         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
609         list_del_init(&clp->cl_strhash);
610         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
611         strhashval = clientstr_hashval(clp->cl_recdir);
612         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
613         renew_client(clp);
614 }
615
616 static struct nfs4_client *
617 find_confirmed_client(clientid_t *clid)
618 {
619         struct nfs4_client *clp;
620         unsigned int idhashval = clientid_hashval(clid->cl_id);
621
622         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
623                 if (same_clid(&clp->cl_clientid, clid))
624                         return clp;
625         }
626         return NULL;
627 }
628
629 static struct nfs4_client *
630 find_unconfirmed_client(clientid_t *clid)
631 {
632         struct nfs4_client *clp;
633         unsigned int idhashval = clientid_hashval(clid->cl_id);
634
635         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
636                 if (same_clid(&clp->cl_clientid, clid))
637                         return clp;
638         }
639         return NULL;
640 }
641
642 static struct nfs4_client *
643 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
644 {
645         struct nfs4_client *clp;
646
647         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
648                 if (same_name(clp->cl_recdir, dname))
649                         return clp;
650         }
651         return NULL;
652 }
653
654 static struct nfs4_client *
655 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
656 {
657         struct nfs4_client *clp;
658
659         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
660                 if (same_name(clp->cl_recdir, dname))
661                         return clp;
662         }
663         return NULL;
664 }
665
666 /* a helper function for parse_callback */
667 static int
668 parse_octet(unsigned int *lenp, char **addrp)
669 {
670         unsigned int len = *lenp;
671         char *p = *addrp;
672         int n = -1;
673         char c;
674
675         for (;;) {
676                 if (!len)
677                         break;
678                 len--;
679                 c = *p++;
680                 if (c == '.')
681                         break;
682                 if ((c < '0') || (c > '9')) {
683                         n = -1;
684                         break;
685                 }
686                 if (n < 0)
687                         n = 0;
688                 n = (n * 10) + (c - '0');
689                 if (n > 255) {
690                         n = -1;
691                         break;
692                 }
693         }
694         *lenp = len;
695         *addrp = p;
696         return n;
697 }
698
699 /* parse and set the setclientid ipv4 callback address */
700 static int
701 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
702 {
703         int temp = 0;
704         u32 cbaddr = 0;
705         u16 cbport = 0;
706         u32 addrlen = addr_len;
707         char *addr = addr_val;
708         int i, shift;
709
710         /* ipaddress */
711         shift = 24;
712         for(i = 4; i > 0  ; i--) {
713                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
714                         return 0;
715                 }
716                 cbaddr |= (temp << shift);
717                 if (shift > 0)
718                 shift -= 8;
719         }
720         *cbaddrp = cbaddr;
721
722         /* port */
723         shift = 8;
724         for(i = 2; i > 0  ; i--) {
725                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
726                         return 0;
727                 }
728                 cbport |= (temp << shift);
729                 if (shift > 0)
730                         shift -= 8;
731         }
732         *cbportp = cbport;
733         return 1;
734 }
735
736 static void
737 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
738 {
739         struct nfs4_callback *cb = &clp->cl_callback;
740
741         /* Currently, we only support tcp for the callback channel */
742         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
743                 goto out_err;
744
745         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
746                          &cb->cb_addr, &cb->cb_port)))
747                 goto out_err;
748         cb->cb_prog = se->se_callback_prog;
749         cb->cb_ident = se->se_callback_ident;
750         return;
751 out_err:
752         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
753                 "will not receive delegations\n",
754                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
755
756         return;
757 }
758
759 __be32
760 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
761                   struct nfsd4_setclientid *setclid)
762 {
763         struct sockaddr_in      *sin = svc_addr_in(rqstp);
764         struct xdr_netobj       clname = { 
765                 .len = setclid->se_namelen,
766                 .data = setclid->se_name,
767         };
768         nfs4_verifier           clverifier = setclid->se_verf;
769         unsigned int            strhashval;
770         struct nfs4_client      *conf, *unconf, *new;
771         __be32                  status;
772         char                    *princ;
773         char                    dname[HEXDIR_LEN];
774         
775         if (!check_name(clname))
776                 return nfserr_inval;
777
778         status = nfs4_make_rec_clidname(dname, &clname);
779         if (status)
780                 return status;
781
782         /* 
783          * XXX The Duplicate Request Cache (DRC) has been checked (??)
784          * We get here on a DRC miss.
785          */
786
787         strhashval = clientstr_hashval(dname);
788
789         nfs4_lock_state();
790         conf = find_confirmed_client_by_str(dname, strhashval);
791         if (conf) {
792                 /* RFC 3530 14.2.33 CASE 0: */
793                 status = nfserr_clid_inuse;
794                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
795                                 || conf->cl_addr != sin->sin_addr.s_addr) {
796                         dprintk("NFSD: setclientid: string in use by clientat %pI4\n",
797                                 &conf->cl_addr);
798                         goto out;
799                 }
800         }
801         /*
802          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
803          * has a description of SETCLIENTID request processing consisting
804          * of 5 bullet points, labeled as CASE0 - CASE4 below.
805          */
806         unconf = find_unconfirmed_client_by_str(dname, strhashval);
807         status = nfserr_resource;
808         if (!conf) {
809                 /*
810                  * RFC 3530 14.2.33 CASE 4:
811                  * placed first, because it is the normal case
812                  */
813                 if (unconf)
814                         expire_client(unconf);
815                 new = create_client(clname, dname);
816                 if (new == NULL)
817                         goto out;
818                 gen_clid(new);
819         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
820                 /*
821                  * RFC 3530 14.2.33 CASE 1:
822                  * probable callback update
823                  */
824                 if (unconf) {
825                         /* Note this is removing unconfirmed {*x***},
826                          * which is stronger than RFC recommended {vxc**}.
827                          * This has the advantage that there is at most
828                          * one {*x***} in either list at any time.
829                          */
830                         expire_client(unconf);
831                 }
832                 new = create_client(clname, dname);
833                 if (new == NULL)
834                         goto out;
835                 copy_clid(new, conf);
836         } else if (!unconf) {
837                 /*
838                  * RFC 3530 14.2.33 CASE 2:
839                  * probable client reboot; state will be removed if
840                  * confirmed.
841                  */
842                 new = create_client(clname, dname);
843                 if (new == NULL)
844                         goto out;
845                 gen_clid(new);
846         } else {
847                 /*
848                  * RFC 3530 14.2.33 CASE 3:
849                  * probable client reboot; state will be removed if
850                  * confirmed.
851                  */
852                 expire_client(unconf);
853                 new = create_client(clname, dname);
854                 if (new == NULL)
855                         goto out;
856                 gen_clid(new);
857         }
858         copy_verf(new, &clverifier);
859         new->cl_addr = sin->sin_addr.s_addr;
860         new->cl_flavor = rqstp->rq_flavor;
861         princ = svc_gss_principal(rqstp);
862         if (princ) {
863                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
864                 if (new->cl_principal == NULL) {
865                         free_client(new);
866                         goto out;
867                 }
868         }
869         copy_cred(&new->cl_cred, &rqstp->rq_cred);
870         gen_confirm(new);
871         gen_callback(new, setclid);
872         add_to_unconfirmed(new, strhashval);
873         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
874         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
875         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
876         status = nfs_ok;
877 out:
878         nfs4_unlock_state();
879         return status;
880 }
881
882
883 /*
884  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
885  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
886  * bullets, labeled as CASE1 - CASE4 below.
887  */
888 __be32
889 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
890                          struct nfsd4_compound_state *cstate,
891                          struct nfsd4_setclientid_confirm *setclientid_confirm)
892 {
893         struct sockaddr_in *sin = svc_addr_in(rqstp);
894         struct nfs4_client *conf, *unconf;
895         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
896         clientid_t * clid = &setclientid_confirm->sc_clientid;
897         __be32 status;
898
899         if (STALE_CLIENTID(clid))
900                 return nfserr_stale_clientid;
901         /* 
902          * XXX The Duplicate Request Cache (DRC) has been checked (??)
903          * We get here on a DRC miss.
904          */
905
906         nfs4_lock_state();
907
908         conf = find_confirmed_client(clid);
909         unconf = find_unconfirmed_client(clid);
910
911         status = nfserr_clid_inuse;
912         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
913                 goto out;
914         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
915                 goto out;
916
917         /*
918          * section 14.2.34 of RFC 3530 has a description of
919          * SETCLIENTID_CONFIRM request processing consisting
920          * of 4 bullet points, labeled as CASE1 - CASE4 below.
921          */
922         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
923                 /*
924                  * RFC 3530 14.2.34 CASE 1:
925                  * callback update
926                  */
927                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
928                         status = nfserr_clid_inuse;
929                 else {
930                         /* XXX: We just turn off callbacks until we can handle
931                           * change request correctly. */
932                         atomic_set(&conf->cl_callback.cb_set, 0);
933                         gen_confirm(conf);
934                         nfsd4_remove_clid_dir(unconf);
935                         expire_client(unconf);
936                         status = nfs_ok;
937
938                 }
939         } else if (conf && !unconf) {
940                 /*
941                  * RFC 3530 14.2.34 CASE 2:
942                  * probable retransmitted request; play it safe and
943                  * do nothing.
944                  */
945                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
946                         status = nfserr_clid_inuse;
947                 else
948                         status = nfs_ok;
949         } else if (!conf && unconf
950                         && same_verf(&unconf->cl_confirm, &confirm)) {
951                 /*
952                  * RFC 3530 14.2.34 CASE 3:
953                  * Normal case; new or rebooted client:
954                  */
955                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
956                         status = nfserr_clid_inuse;
957                 } else {
958                         unsigned int hash =
959                                 clientstr_hashval(unconf->cl_recdir);
960                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
961                                                                         hash);
962                         if (conf) {
963                                 nfsd4_remove_clid_dir(conf);
964                                 expire_client(conf);
965                         }
966                         move_to_confirmed(unconf);
967                         conf = unconf;
968                         nfsd4_probe_callback(conf);
969                         status = nfs_ok;
970                 }
971         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
972             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
973                                                                 &confirm)))) {
974                 /*
975                  * RFC 3530 14.2.34 CASE 4:
976                  * Client probably hasn't noticed that we rebooted yet.
977                  */
978                 status = nfserr_stale_clientid;
979         } else {
980                 /* check that we have hit one of the cases...*/
981                 status = nfserr_clid_inuse;
982         }
983 out:
984         nfs4_unlock_state();
985         return status;
986 }
987
988 /* OPEN Share state helper functions */
989 static inline struct nfs4_file *
990 alloc_init_file(struct inode *ino)
991 {
992         struct nfs4_file *fp;
993         unsigned int hashval = file_hashval(ino);
994
995         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
996         if (fp) {
997                 atomic_set(&fp->fi_ref, 1);
998                 INIT_LIST_HEAD(&fp->fi_hash);
999                 INIT_LIST_HEAD(&fp->fi_stateids);
1000                 INIT_LIST_HEAD(&fp->fi_delegations);
1001                 spin_lock(&recall_lock);
1002                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1003                 spin_unlock(&recall_lock);
1004                 fp->fi_inode = igrab(ino);
1005                 fp->fi_id = current_fileid++;
1006                 fp->fi_had_conflict = false;
1007                 return fp;
1008         }
1009         return NULL;
1010 }
1011
1012 static void
1013 nfsd4_free_slab(struct kmem_cache **slab)
1014 {
1015         if (*slab == NULL)
1016                 return;
1017         kmem_cache_destroy(*slab);
1018         *slab = NULL;
1019 }
1020
1021 void
1022 nfsd4_free_slabs(void)
1023 {
1024         nfsd4_free_slab(&stateowner_slab);
1025         nfsd4_free_slab(&file_slab);
1026         nfsd4_free_slab(&stateid_slab);
1027         nfsd4_free_slab(&deleg_slab);
1028 }
1029
1030 static int
1031 nfsd4_init_slabs(void)
1032 {
1033         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1034                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1035         if (stateowner_slab == NULL)
1036                 goto out_nomem;
1037         file_slab = kmem_cache_create("nfsd4_files",
1038                         sizeof(struct nfs4_file), 0, 0, NULL);
1039         if (file_slab == NULL)
1040                 goto out_nomem;
1041         stateid_slab = kmem_cache_create("nfsd4_stateids",
1042                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1043         if (stateid_slab == NULL)
1044                 goto out_nomem;
1045         deleg_slab = kmem_cache_create("nfsd4_delegations",
1046                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1047         if (deleg_slab == NULL)
1048                 goto out_nomem;
1049         return 0;
1050 out_nomem:
1051         nfsd4_free_slabs();
1052         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1053         return -ENOMEM;
1054 }
1055
1056 void
1057 nfs4_free_stateowner(struct kref *kref)
1058 {
1059         struct nfs4_stateowner *sop =
1060                 container_of(kref, struct nfs4_stateowner, so_ref);
1061         kfree(sop->so_owner.data);
1062         kmem_cache_free(stateowner_slab, sop);
1063 }
1064
1065 static inline struct nfs4_stateowner *
1066 alloc_stateowner(struct xdr_netobj *owner)
1067 {
1068         struct nfs4_stateowner *sop;
1069
1070         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1071                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1072                         memcpy(sop->so_owner.data, owner->data, owner->len);
1073                         sop->so_owner.len = owner->len;
1074                         kref_init(&sop->so_ref);
1075                         return sop;
1076                 } 
1077                 kmem_cache_free(stateowner_slab, sop);
1078         }
1079         return NULL;
1080 }
1081
1082 static struct nfs4_stateowner *
1083 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1084         struct nfs4_stateowner *sop;
1085         struct nfs4_replay *rp;
1086         unsigned int idhashval;
1087
1088         if (!(sop = alloc_stateowner(&open->op_owner)))
1089                 return NULL;
1090         idhashval = ownerid_hashval(current_ownerid);
1091         INIT_LIST_HEAD(&sop->so_idhash);
1092         INIT_LIST_HEAD(&sop->so_strhash);
1093         INIT_LIST_HEAD(&sop->so_perclient);
1094         INIT_LIST_HEAD(&sop->so_stateids);
1095         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1096         INIT_LIST_HEAD(&sop->so_close_lru);
1097         sop->so_time = 0;
1098         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1099         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1100         list_add(&sop->so_perclient, &clp->cl_openowners);
1101         sop->so_is_open_owner = 1;
1102         sop->so_id = current_ownerid++;
1103         sop->so_client = clp;
1104         sop->so_seqid = open->op_seqid;
1105         sop->so_confirmed = 0;
1106         rp = &sop->so_replay;
1107         rp->rp_status = nfserr_serverfault;
1108         rp->rp_buflen = 0;
1109         rp->rp_buf = rp->rp_ibuf;
1110         return sop;
1111 }
1112
1113 static inline void
1114 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1115         struct nfs4_stateowner *sop = open->op_stateowner;
1116         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1117
1118         INIT_LIST_HEAD(&stp->st_hash);
1119         INIT_LIST_HEAD(&stp->st_perstateowner);
1120         INIT_LIST_HEAD(&stp->st_lockowners);
1121         INIT_LIST_HEAD(&stp->st_perfile);
1122         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1123         list_add(&stp->st_perstateowner, &sop->so_stateids);
1124         list_add(&stp->st_perfile, &fp->fi_stateids);
1125         stp->st_stateowner = sop;
1126         get_nfs4_file(fp);
1127         stp->st_file = fp;
1128         stp->st_stateid.si_boot = boot_time;
1129         stp->st_stateid.si_stateownerid = sop->so_id;
1130         stp->st_stateid.si_fileid = fp->fi_id;
1131         stp->st_stateid.si_generation = 0;
1132         stp->st_access_bmap = 0;
1133         stp->st_deny_bmap = 0;
1134         __set_bit(open->op_share_access, &stp->st_access_bmap);
1135         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1136         stp->st_openstp = NULL;
1137 }
1138
1139 static void
1140 move_to_close_lru(struct nfs4_stateowner *sop)
1141 {
1142         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1143
1144         list_move_tail(&sop->so_close_lru, &close_lru);
1145         sop->so_time = get_seconds();
1146 }
1147
1148 static int
1149 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1150                                                         clientid_t *clid)
1151 {
1152         return (sop->so_owner.len == owner->len) &&
1153                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1154                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1155 }
1156
1157 static struct nfs4_stateowner *
1158 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1159 {
1160         struct nfs4_stateowner *so = NULL;
1161
1162         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1163                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1164                         return so;
1165         }
1166         return NULL;
1167 }
1168
1169 /* search file_hashtbl[] for file */
1170 static struct nfs4_file *
1171 find_file(struct inode *ino)
1172 {
1173         unsigned int hashval = file_hashval(ino);
1174         struct nfs4_file *fp;
1175
1176         spin_lock(&recall_lock);
1177         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1178                 if (fp->fi_inode == ino) {
1179                         get_nfs4_file(fp);
1180                         spin_unlock(&recall_lock);
1181                         return fp;
1182                 }
1183         }
1184         spin_unlock(&recall_lock);
1185         return NULL;
1186 }
1187
1188 static inline int access_valid(u32 x)
1189 {
1190         if (x < NFS4_SHARE_ACCESS_READ)
1191                 return 0;
1192         if (x > NFS4_SHARE_ACCESS_BOTH)
1193                 return 0;
1194         return 1;
1195 }
1196
1197 static inline int deny_valid(u32 x)
1198 {
1199         /* Note: unlike access bits, deny bits may be zero. */
1200         return x <= NFS4_SHARE_DENY_BOTH;
1201 }
1202
1203 /*
1204  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1205  * st_{access,deny}_bmap field of the stateid, in order to track not
1206  * only what share bits are currently in force, but also what
1207  * combinations of share bits previous opens have used.  This allows us
1208  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1209  * return an error if the client attempt to downgrade to a combination
1210  * of share bits not explicable by closing some of its previous opens.
1211  *
1212  * XXX: This enforcement is actually incomplete, since we don't keep
1213  * track of access/deny bit combinations; so, e.g., we allow:
1214  *
1215  *      OPEN allow read, deny write
1216  *      OPEN allow both, deny none
1217  *      DOWNGRADE allow read, deny none
1218  *
1219  * which we should reject.
1220  */
1221 static void
1222 set_access(unsigned int *access, unsigned long bmap) {
1223         int i;
1224
1225         *access = 0;
1226         for (i = 1; i < 4; i++) {
1227                 if (test_bit(i, &bmap))
1228                         *access |= i;
1229         }
1230 }
1231
1232 static void
1233 set_deny(unsigned int *deny, unsigned long bmap) {
1234         int i;
1235
1236         *deny = 0;
1237         for (i = 0; i < 4; i++) {
1238                 if (test_bit(i, &bmap))
1239                         *deny |= i ;
1240         }
1241 }
1242
1243 static int
1244 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1245         unsigned int access, deny;
1246
1247         set_access(&access, stp->st_access_bmap);
1248         set_deny(&deny, stp->st_deny_bmap);
1249         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1250                 return 0;
1251         return 1;
1252 }
1253
1254 /*
1255  * Called to check deny when READ with all zero stateid or
1256  * WRITE with all zero or all one stateid
1257  */
1258 static __be32
1259 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1260 {
1261         struct inode *ino = current_fh->fh_dentry->d_inode;
1262         struct nfs4_file *fp;
1263         struct nfs4_stateid *stp;
1264         __be32 ret;
1265
1266         dprintk("NFSD: nfs4_share_conflict\n");
1267
1268         fp = find_file(ino);
1269         if (!fp)
1270                 return nfs_ok;
1271         ret = nfserr_locked;
1272         /* Search for conflicting share reservations */
1273         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1274                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1275                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1276                         goto out;
1277         }
1278         ret = nfs_ok;
1279 out:
1280         put_nfs4_file(fp);
1281         return ret;
1282 }
1283
1284 static inline void
1285 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1286 {
1287         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1288                 drop_file_write_access(filp);
1289                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1290         }
1291 }
1292
1293 /*
1294  * Recall a delegation
1295  */
1296 static int
1297 do_recall(void *__dp)
1298 {
1299         struct nfs4_delegation *dp = __dp;
1300
1301         dp->dl_file->fi_had_conflict = true;
1302         nfsd4_cb_recall(dp);
1303         return 0;
1304 }
1305
1306 /*
1307  * Spawn a thread to perform a recall on the delegation represented
1308  * by the lease (file_lock)
1309  *
1310  * Called from break_lease() with lock_kernel() held.
1311  * Note: we assume break_lease will only call this *once* for any given
1312  * lease.
1313  */
1314 static
1315 void nfsd_break_deleg_cb(struct file_lock *fl)
1316 {
1317         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1318         struct task_struct *t;
1319
1320         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1321         if (!dp)
1322                 return;
1323
1324         /* We're assuming the state code never drops its reference
1325          * without first removing the lease.  Since we're in this lease
1326          * callback (and since the lease code is serialized by the kernel
1327          * lock) we know the server hasn't removed the lease yet, we know
1328          * it's safe to take a reference: */
1329         atomic_inc(&dp->dl_count);
1330         atomic_inc(&dp->dl_client->cl_count);
1331
1332         spin_lock(&recall_lock);
1333         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1334         spin_unlock(&recall_lock);
1335
1336         /* only place dl_time is set. protected by lock_kernel*/
1337         dp->dl_time = get_seconds();
1338
1339         /*
1340          * We don't want the locks code to timeout the lease for us;
1341          * we'll remove it ourself if the delegation isn't returned
1342          * in time.
1343          */
1344         fl->fl_break_time = 0;
1345
1346         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1347         if (IS_ERR(t)) {
1348                 struct nfs4_client *clp = dp->dl_client;
1349
1350                 printk(KERN_INFO "NFSD: Callback thread failed for "
1351                         "for client (clientid %08x/%08x)\n",
1352                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1353                 put_nfs4_client(dp->dl_client);
1354                 nfs4_put_delegation(dp);
1355         }
1356 }
1357
1358 /*
1359  * The file_lock is being reapd.
1360  *
1361  * Called by locks_free_lock() with lock_kernel() held.
1362  */
1363 static
1364 void nfsd_release_deleg_cb(struct file_lock *fl)
1365 {
1366         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1367
1368         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1369
1370         if (!(fl->fl_flags & FL_LEASE) || !dp)
1371                 return;
1372         dp->dl_flock = NULL;
1373 }
1374
1375 /*
1376  * Set the delegation file_lock back pointer.
1377  *
1378  * Called from setlease() with lock_kernel() held.
1379  */
1380 static
1381 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1382 {
1383         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1384
1385         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1386         if (!dp)
1387                 return;
1388         dp->dl_flock = new;
1389 }
1390
1391 /*
1392  * Called from setlease() with lock_kernel() held
1393  */
1394 static
1395 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1396 {
1397         struct nfs4_delegation *onlistd =
1398                 (struct nfs4_delegation *)onlist->fl_owner;
1399         struct nfs4_delegation *tryd =
1400                 (struct nfs4_delegation *)try->fl_owner;
1401
1402         if (onlist->fl_lmops != try->fl_lmops)
1403                 return 0;
1404
1405         return onlistd->dl_client == tryd->dl_client;
1406 }
1407
1408
1409 static
1410 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1411 {
1412         if (arg & F_UNLCK)
1413                 return lease_modify(onlist, arg);
1414         else
1415                 return -EAGAIN;
1416 }
1417
1418 static struct lock_manager_operations nfsd_lease_mng_ops = {
1419         .fl_break = nfsd_break_deleg_cb,
1420         .fl_release_private = nfsd_release_deleg_cb,
1421         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1422         .fl_mylease = nfsd_same_client_deleg_cb,
1423         .fl_change = nfsd_change_deleg_cb,
1424 };
1425
1426
1427 __be32
1428 nfsd4_process_open1(struct nfsd4_open *open)
1429 {
1430         clientid_t *clientid = &open->op_clientid;
1431         struct nfs4_client *clp = NULL;
1432         unsigned int strhashval;
1433         struct nfs4_stateowner *sop = NULL;
1434
1435         if (!check_name(open->op_owner))
1436                 return nfserr_inval;
1437
1438         if (STALE_CLIENTID(&open->op_clientid))
1439                 return nfserr_stale_clientid;
1440
1441         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1442         sop = find_openstateowner_str(strhashval, open);
1443         open->op_stateowner = sop;
1444         if (!sop) {
1445                 /* Make sure the client's lease hasn't expired. */
1446                 clp = find_confirmed_client(clientid);
1447                 if (clp == NULL)
1448                         return nfserr_expired;
1449                 goto renew;
1450         }
1451         if (!sop->so_confirmed) {
1452                 /* Replace unconfirmed owners without checking for replay. */
1453                 clp = sop->so_client;
1454                 release_openowner(sop);
1455                 open->op_stateowner = NULL;
1456                 goto renew;
1457         }
1458         if (open->op_seqid == sop->so_seqid - 1) {
1459                 if (sop->so_replay.rp_buflen)
1460                         return nfserr_replay_me;
1461                 /* The original OPEN failed so spectacularly
1462                  * that we don't even have replay data saved!
1463                  * Therefore, we have no choice but to continue
1464                  * processing this OPEN; presumably, we'll
1465                  * fail again for the same reason.
1466                  */
1467                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1468                 goto renew;
1469         }
1470         if (open->op_seqid != sop->so_seqid)
1471                 return nfserr_bad_seqid;
1472 renew:
1473         if (open->op_stateowner == NULL) {
1474                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1475                 if (sop == NULL)
1476                         return nfserr_resource;
1477                 open->op_stateowner = sop;
1478         }
1479         list_del_init(&sop->so_close_lru);
1480         renew_client(sop->so_client);
1481         return nfs_ok;
1482 }
1483
1484 static inline __be32
1485 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1486 {
1487         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1488                 return nfserr_openmode;
1489         else
1490                 return nfs_ok;
1491 }
1492
1493 static struct nfs4_delegation *
1494 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1495 {
1496         struct nfs4_delegation *dp;
1497
1498         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1499                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1500                         return dp;
1501         }
1502         return NULL;
1503 }
1504
1505 static __be32
1506 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1507                 struct nfs4_delegation **dp)
1508 {
1509         int flags;
1510         __be32 status = nfserr_bad_stateid;
1511
1512         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1513         if (*dp == NULL)
1514                 goto out;
1515         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1516                                                 RD_STATE : WR_STATE;
1517         status = nfs4_check_delegmode(*dp, flags);
1518         if (status)
1519                 *dp = NULL;
1520 out:
1521         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1522                 return nfs_ok;
1523         if (status)
1524                 return status;
1525         open->op_stateowner->so_confirmed = 1;
1526         return nfs_ok;
1527 }
1528
1529 static __be32
1530 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1531 {
1532         struct nfs4_stateid *local;
1533         __be32 status = nfserr_share_denied;
1534         struct nfs4_stateowner *sop = open->op_stateowner;
1535
1536         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1537                 /* ignore lock owners */
1538                 if (local->st_stateowner->so_is_open_owner == 0)
1539                         continue;
1540                 /* remember if we have seen this open owner */
1541                 if (local->st_stateowner == sop)
1542                         *stpp = local;
1543                 /* check for conflicting share reservations */
1544                 if (!test_share(local, open))
1545                         goto out;
1546         }
1547         status = 0;
1548 out:
1549         return status;
1550 }
1551
1552 static inline struct nfs4_stateid *
1553 nfs4_alloc_stateid(void)
1554 {
1555         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1556 }
1557
1558 static __be32
1559 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1560                 struct nfs4_delegation *dp,
1561                 struct svc_fh *cur_fh, int flags)
1562 {
1563         struct nfs4_stateid *stp;
1564
1565         stp = nfs4_alloc_stateid();
1566         if (stp == NULL)
1567                 return nfserr_resource;
1568
1569         if (dp) {
1570                 get_file(dp->dl_vfs_file);
1571                 stp->st_vfs_file = dp->dl_vfs_file;
1572         } else {
1573                 __be32 status;
1574                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1575                                 &stp->st_vfs_file);
1576                 if (status) {
1577                         if (status == nfserr_dropit)
1578                                 status = nfserr_jukebox;
1579                         kmem_cache_free(stateid_slab, stp);
1580                         return status;
1581                 }
1582         }
1583         *stpp = stp;
1584         return 0;
1585 }
1586
1587 static inline __be32
1588 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1589                 struct nfsd4_open *open)
1590 {
1591         struct iattr iattr = {
1592                 .ia_valid = ATTR_SIZE,
1593                 .ia_size = 0,
1594         };
1595         if (!open->op_truncate)
1596                 return 0;
1597         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1598                 return nfserr_inval;
1599         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1600 }
1601
1602 static __be32
1603 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1604 {
1605         struct file *filp = stp->st_vfs_file;
1606         struct inode *inode = filp->f_path.dentry->d_inode;
1607         unsigned int share_access, new_writer;
1608         __be32 status;
1609
1610         set_access(&share_access, stp->st_access_bmap);
1611         new_writer = (~share_access) & open->op_share_access
1612                         & NFS4_SHARE_ACCESS_WRITE;
1613
1614         if (new_writer) {
1615                 int err = get_write_access(inode);
1616                 if (err)
1617                         return nfserrno(err);
1618                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1619                 if (err)
1620                         return nfserrno(err);
1621                 file_take_write(filp);
1622         }
1623         status = nfsd4_truncate(rqstp, cur_fh, open);
1624         if (status) {
1625                 if (new_writer)
1626                         put_write_access(inode);
1627                 return status;
1628         }
1629         /* remember the open */
1630         filp->f_mode |= open->op_share_access;
1631         __set_bit(open->op_share_access, &stp->st_access_bmap);
1632         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1633
1634         return nfs_ok;
1635 }
1636
1637
1638 static void
1639 nfs4_set_claim_prev(struct nfsd4_open *open)
1640 {
1641         open->op_stateowner->so_confirmed = 1;
1642         open->op_stateowner->so_client->cl_firststate = 1;
1643 }
1644
1645 /*
1646  * Attempt to hand out a delegation.
1647  */
1648 static void
1649 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1650 {
1651         struct nfs4_delegation *dp;
1652         struct nfs4_stateowner *sop = stp->st_stateowner;
1653         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1654         struct file_lock fl, *flp = &fl;
1655         int status, flag = 0;
1656
1657         flag = NFS4_OPEN_DELEGATE_NONE;
1658         open->op_recall = 0;
1659         switch (open->op_claim_type) {
1660                 case NFS4_OPEN_CLAIM_PREVIOUS:
1661                         if (!atomic_read(&cb->cb_set))
1662                                 open->op_recall = 1;
1663                         flag = open->op_delegate_type;
1664                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1665                                 goto out;
1666                         break;
1667                 case NFS4_OPEN_CLAIM_NULL:
1668                         /* Let's not give out any delegations till everyone's
1669                          * had the chance to reclaim theirs.... */
1670                         if (locks_in_grace())
1671                                 goto out;
1672                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1673                                 goto out;
1674                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1675                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1676                         else
1677                                 flag = NFS4_OPEN_DELEGATE_READ;
1678                         break;
1679                 default:
1680                         goto out;
1681         }
1682
1683         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1684         if (dp == NULL) {
1685                 flag = NFS4_OPEN_DELEGATE_NONE;
1686                 goto out;
1687         }
1688         locks_init_lock(&fl);
1689         fl.fl_lmops = &nfsd_lease_mng_ops;
1690         fl.fl_flags = FL_LEASE;
1691         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1692         fl.fl_end = OFFSET_MAX;
1693         fl.fl_owner =  (fl_owner_t)dp;
1694         fl.fl_file = stp->st_vfs_file;
1695         fl.fl_pid = current->tgid;
1696
1697         /* vfs_setlease checks to see if delegation should be handed out.
1698          * the lock_manager callbacks fl_mylease and fl_change are used
1699          */
1700         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1701                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1702                 unhash_delegation(dp);
1703                 flag = NFS4_OPEN_DELEGATE_NONE;
1704                 goto out;
1705         }
1706
1707         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1708
1709         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1710                      dp->dl_stateid.si_boot,
1711                      dp->dl_stateid.si_stateownerid,
1712                      dp->dl_stateid.si_fileid,
1713                      dp->dl_stateid.si_generation);
1714 out:
1715         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1716                         && flag == NFS4_OPEN_DELEGATE_NONE
1717                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1718                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1719         open->op_delegate_type = flag;
1720 }
1721
1722 /*
1723  * called with nfs4_lock_state() held.
1724  */
1725 __be32
1726 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1727 {
1728         struct nfs4_file *fp = NULL;
1729         struct inode *ino = current_fh->fh_dentry->d_inode;
1730         struct nfs4_stateid *stp = NULL;
1731         struct nfs4_delegation *dp = NULL;
1732         __be32 status;
1733
1734         status = nfserr_inval;
1735         if (!access_valid(open->op_share_access)
1736                         || !deny_valid(open->op_share_deny))
1737                 goto out;
1738         /*
1739          * Lookup file; if found, lookup stateid and check open request,
1740          * and check for delegations in the process of being recalled.
1741          * If not found, create the nfs4_file struct
1742          */
1743         fp = find_file(ino);
1744         if (fp) {
1745                 if ((status = nfs4_check_open(fp, open, &stp)))
1746                         goto out;
1747                 status = nfs4_check_deleg(fp, open, &dp);
1748                 if (status)
1749                         goto out;
1750         } else {
1751                 status = nfserr_bad_stateid;
1752                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1753                         goto out;
1754                 status = nfserr_resource;
1755                 fp = alloc_init_file(ino);
1756                 if (fp == NULL)
1757                         goto out;
1758         }
1759
1760         /*
1761          * OPEN the file, or upgrade an existing OPEN.
1762          * If truncate fails, the OPEN fails.
1763          */
1764         if (stp) {
1765                 /* Stateid was found, this is an OPEN upgrade */
1766                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1767                 if (status)
1768                         goto out;
1769                 update_stateid(&stp->st_stateid);
1770         } else {
1771                 /* Stateid was not found, this is a new OPEN */
1772                 int flags = 0;
1773                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1774                         flags |= NFSD_MAY_READ;
1775                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1776                         flags |= NFSD_MAY_WRITE;
1777                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1778                 if (status)
1779                         goto out;
1780                 init_stateid(stp, fp, open);
1781                 status = nfsd4_truncate(rqstp, current_fh, open);
1782                 if (status) {
1783                         release_open_stateid(stp);
1784                         goto out;
1785                 }
1786         }
1787         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1788
1789         /*
1790         * Attempt to hand out a delegation. No error return, because the
1791         * OPEN succeeds even if we fail.
1792         */
1793         nfs4_open_delegation(current_fh, open, stp);
1794
1795         status = nfs_ok;
1796
1797         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1798                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1799                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1800 out:
1801         if (fp)
1802                 put_nfs4_file(fp);
1803         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1804                 nfs4_set_claim_prev(open);
1805         /*
1806         * To finish the open response, we just need to set the rflags.
1807         */
1808         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1809         if (!open->op_stateowner->so_confirmed)
1810                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1811
1812         return status;
1813 }
1814
1815 __be32
1816 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1817             clientid_t *clid)
1818 {
1819         struct nfs4_client *clp;
1820         __be32 status;
1821
1822         nfs4_lock_state();
1823         dprintk("process_renew(%08x/%08x): starting\n", 
1824                         clid->cl_boot, clid->cl_id);
1825         status = nfserr_stale_clientid;
1826         if (STALE_CLIENTID(clid))
1827                 goto out;
1828         clp = find_confirmed_client(clid);
1829         status = nfserr_expired;
1830         if (clp == NULL) {
1831                 /* We assume the client took too long to RENEW. */
1832                 dprintk("nfsd4_renew: clientid not found!\n");
1833                 goto out;
1834         }
1835         renew_client(clp);
1836         status = nfserr_cb_path_down;
1837         if (!list_empty(&clp->cl_delegations)
1838                         && !atomic_read(&clp->cl_callback.cb_set))
1839                 goto out;
1840         status = nfs_ok;
1841 out:
1842         nfs4_unlock_state();
1843         return status;
1844 }
1845
1846 struct lock_manager nfsd4_manager = {
1847 };
1848
1849 static void
1850 nfsd4_end_grace(void)
1851 {
1852         dprintk("NFSD: end of grace period\n");
1853         nfsd4_recdir_purge_old();
1854         locks_end_grace(&nfsd4_manager);
1855 }
1856
1857 static time_t
1858 nfs4_laundromat(void)
1859 {
1860         struct nfs4_client *clp;
1861         struct nfs4_stateowner *sop;
1862         struct nfs4_delegation *dp;
1863         struct list_head *pos, *next, reaplist;
1864         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1865         time_t t, clientid_val = NFSD_LEASE_TIME;
1866         time_t u, test_val = NFSD_LEASE_TIME;
1867
1868         nfs4_lock_state();
1869
1870         dprintk("NFSD: laundromat service - starting\n");
1871         if (locks_in_grace())
1872                 nfsd4_end_grace();
1873         list_for_each_safe(pos, next, &client_lru) {
1874                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1875                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1876                         t = clp->cl_time - cutoff;
1877                         if (clientid_val > t)
1878                                 clientid_val = t;
1879                         break;
1880                 }
1881                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1882                         clp->cl_clientid.cl_id);
1883                 nfsd4_remove_clid_dir(clp);
1884                 expire_client(clp);
1885         }
1886         INIT_LIST_HEAD(&reaplist);
1887         spin_lock(&recall_lock);
1888         list_for_each_safe(pos, next, &del_recall_lru) {
1889                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1890                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1891                         u = dp->dl_time - cutoff;
1892                         if (test_val > u)
1893                                 test_val = u;
1894                         break;
1895                 }
1896                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1897                                     dp, dp->dl_flock);
1898                 list_move(&dp->dl_recall_lru, &reaplist);
1899         }
1900         spin_unlock(&recall_lock);
1901         list_for_each_safe(pos, next, &reaplist) {
1902                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1903                 list_del_init(&dp->dl_recall_lru);
1904                 unhash_delegation(dp);
1905         }
1906         test_val = NFSD_LEASE_TIME;
1907         list_for_each_safe(pos, next, &close_lru) {
1908                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1909                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1910                         u = sop->so_time - cutoff;
1911                         if (test_val > u)
1912                                 test_val = u;
1913                         break;
1914                 }
1915                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1916                         sop->so_id);
1917                 release_openowner(sop);
1918         }
1919         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1920                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1921         nfs4_unlock_state();
1922         return clientid_val;
1923 }
1924
1925 static struct workqueue_struct *laundry_wq;
1926 static void laundromat_main(struct work_struct *);
1927 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1928
1929 static void
1930 laundromat_main(struct work_struct *not_used)
1931 {
1932         time_t t;
1933
1934         t = nfs4_laundromat();
1935         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1936         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1937 }
1938
1939 static struct nfs4_stateowner *
1940 search_close_lru(u32 st_id, int flags)
1941 {
1942         struct nfs4_stateowner *local = NULL;
1943
1944         if (flags & CLOSE_STATE) {
1945                 list_for_each_entry(local, &close_lru, so_close_lru) {
1946                         if (local->so_id == st_id)
1947                                 return local;
1948                 }
1949         }
1950         return NULL;
1951 }
1952
1953 static inline int
1954 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1955 {
1956         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1957 }
1958
1959 static int
1960 STALE_STATEID(stateid_t *stateid)
1961 {
1962         if (stateid->si_boot == boot_time)
1963                 return 0;
1964         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1965                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1966                 stateid->si_generation);
1967         return 1;
1968 }
1969
1970 static inline int
1971 access_permit_read(unsigned long access_bmap)
1972 {
1973         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1974                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1975                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1976 }
1977
1978 static inline int
1979 access_permit_write(unsigned long access_bmap)
1980 {
1981         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1982                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1983 }
1984
1985 static
1986 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1987 {
1988         __be32 status = nfserr_openmode;
1989
1990         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1991                 goto out;
1992         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1993                 goto out;
1994         status = nfs_ok;
1995 out:
1996         return status;
1997 }
1998
1999 static inline __be32
2000 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2001 {
2002         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2003                 return nfs_ok;
2004         else if (locks_in_grace()) {
2005                 /* Answer in remaining cases depends on existance of
2006                  * conflicting state; so we must wait out the grace period. */
2007                 return nfserr_grace;
2008         } else if (flags & WR_STATE)
2009                 return nfs4_share_conflict(current_fh,
2010                                 NFS4_SHARE_DENY_WRITE);
2011         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2012                 return nfs4_share_conflict(current_fh,
2013                                 NFS4_SHARE_DENY_READ);
2014 }
2015
2016 /*
2017  * Allow READ/WRITE during grace period on recovered state only for files
2018  * that are not able to provide mandatory locking.
2019  */
2020 static inline int
2021 grace_disallows_io(struct inode *inode)
2022 {
2023         return locks_in_grace() && mandatory_lock(inode);
2024 }
2025
2026 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2027 {
2028         /* If the client sends us a stateid from the future, it's buggy: */
2029         if (in->si_generation > ref->si_generation)
2030                 return nfserr_bad_stateid;
2031         /*
2032          * The following, however, can happen.  For example, if the
2033          * client sends an open and some IO at the same time, the open
2034          * may bump si_generation while the IO is still in flight.
2035          * Thanks to hard links and renames, the client never knows what
2036          * file an open will affect.  So it could avoid that situation
2037          * only by serializing all opens and IO from the same open
2038          * owner.  To recover from the old_stateid error, the client
2039          * will just have to retry the IO:
2040          */
2041         if (in->si_generation < ref->si_generation)
2042                 return nfserr_old_stateid;
2043         return nfs_ok;
2044 }
2045
2046 static int is_delegation_stateid(stateid_t *stateid)
2047 {
2048         return stateid->si_fileid == 0;
2049 }
2050
2051 /*
2052 * Checks for stateid operations
2053 */
2054 __be32
2055 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2056 {
2057         struct nfs4_stateid *stp = NULL;
2058         struct nfs4_delegation *dp = NULL;
2059         struct inode *ino = current_fh->fh_dentry->d_inode;
2060         __be32 status;
2061
2062         if (filpp)
2063                 *filpp = NULL;
2064
2065         if (grace_disallows_io(ino))
2066                 return nfserr_grace;
2067
2068         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2069                 return check_special_stateids(current_fh, stateid, flags);
2070
2071         status = nfserr_stale_stateid;
2072         if (STALE_STATEID(stateid)) 
2073                 goto out;
2074
2075         status = nfserr_bad_stateid;
2076         if (is_delegation_stateid(stateid)) {
2077                 dp = find_delegation_stateid(ino, stateid);
2078                 if (!dp)
2079                         goto out;
2080                 status = check_stateid_generation(stateid, &dp->dl_stateid);
2081                 if (status)
2082                         goto out;
2083                 status = nfs4_check_delegmode(dp, flags);
2084                 if (status)
2085                         goto out;
2086                 renew_client(dp->dl_client);
2087                 if (filpp)
2088                         *filpp = dp->dl_vfs_file;
2089         } else { /* open or lock stateid */
2090                 stp = find_stateid(stateid, flags);
2091                 if (!stp)
2092                         goto out;
2093                 if (nfs4_check_fh(current_fh, stp))
2094                         goto out;
2095                 if (!stp->st_stateowner->so_confirmed)
2096                         goto out;
2097                 status = check_stateid_generation(stateid, &stp->st_stateid);
2098                 if (status)
2099                         goto out;
2100                 status = nfs4_check_openmode(stp, flags);
2101                 if (status)
2102                         goto out;
2103                 renew_client(stp->st_stateowner->so_client);
2104                 if (filpp)
2105                         *filpp = stp->st_vfs_file;
2106         }
2107         status = nfs_ok;
2108 out:
2109         return status;
2110 }
2111
2112 static inline int
2113 setlkflg (int type)
2114 {
2115         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2116                 RD_STATE : WR_STATE;
2117 }
2118
2119 /* 
2120  * Checks for sequence id mutating operations. 
2121  */
2122 static __be32
2123 nfs4_preprocess_seqid_op(struct svc_fh *current_fh, u32 seqid, stateid_t *stateid, int flags, struct nfs4_stateowner **sopp, struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2124 {
2125         struct nfs4_stateid *stp;
2126         struct nfs4_stateowner *sop;
2127         __be32 status;
2128
2129         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2130                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2131                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2132                 stateid->si_generation);
2133
2134         *stpp = NULL;
2135         *sopp = NULL;
2136
2137         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2138                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2139                 return nfserr_bad_stateid;
2140         }
2141
2142         if (STALE_STATEID(stateid))
2143                 return nfserr_stale_stateid;
2144         /*
2145         * We return BAD_STATEID if filehandle doesn't match stateid, 
2146         * the confirmed flag is incorrecly set, or the generation 
2147         * number is incorrect.  
2148         */
2149         stp = find_stateid(stateid, flags);
2150         if (stp == NULL) {
2151                 /*
2152                  * Also, we should make sure this isn't just the result of
2153                  * a replayed close:
2154                  */
2155                 sop = search_close_lru(stateid->si_stateownerid, flags);
2156                 if (sop == NULL)
2157                         return nfserr_bad_stateid;
2158                 *sopp = sop;
2159                 goto check_replay;
2160         }
2161
2162         *stpp = stp;
2163         *sopp = sop = stp->st_stateowner;
2164
2165         if (lock) {
2166                 clientid_t *lockclid = &lock->v.new.clientid;
2167                 struct nfs4_client *clp = sop->so_client;
2168                 int lkflg = 0;
2169                 __be32 status;
2170
2171                 lkflg = setlkflg(lock->lk_type);
2172
2173                 if (lock->lk_is_new) {
2174                         if (!sop->so_is_open_owner)
2175                                 return nfserr_bad_stateid;
2176                         if (!same_clid(&clp->cl_clientid, lockclid))
2177                                return nfserr_bad_stateid;
2178                         /* stp is the open stateid */
2179                         status = nfs4_check_openmode(stp, lkflg);
2180                         if (status)
2181                                 return status;
2182                 } else {
2183                         /* stp is the lock stateid */
2184                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2185                         if (status)
2186                                 return status;
2187                }
2188         }
2189
2190         if (nfs4_check_fh(current_fh, stp)) {
2191                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2192                 return nfserr_bad_stateid;
2193         }
2194
2195         /*
2196         *  We now validate the seqid and stateid generation numbers.
2197         *  For the moment, we ignore the possibility of 
2198         *  generation number wraparound.
2199         */
2200         if (seqid != sop->so_seqid)
2201                 goto check_replay;
2202
2203         if (sop->so_confirmed && flags & CONFIRM) {
2204                 dprintk("NFSD: preprocess_seqid_op: expected"
2205                                 " unconfirmed stateowner!\n");
2206                 return nfserr_bad_stateid;
2207         }
2208         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2209                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2210                                 " confirmed yet!\n");
2211                 return nfserr_bad_stateid;
2212         }
2213         status = check_stateid_generation(stateid, &stp->st_stateid);
2214         if (status)
2215                 return status;
2216         renew_client(sop->so_client);
2217         return nfs_ok;
2218
2219 check_replay:
2220         if (seqid == sop->so_seqid - 1) {
2221                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2222                 /* indicate replay to calling function */
2223                 return nfserr_replay_me;
2224         }
2225         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2226                         sop->so_seqid, seqid);
2227         *sopp = NULL;
2228         return nfserr_bad_seqid;
2229 }
2230
2231 __be32
2232 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2233                    struct nfsd4_open_confirm *oc)
2234 {
2235         __be32 status;
2236         struct nfs4_stateowner *sop;
2237         struct nfs4_stateid *stp;
2238
2239         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2240                         (int)cstate->current_fh.fh_dentry->d_name.len,
2241                         cstate->current_fh.fh_dentry->d_name.name);
2242
2243         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2244         if (status)
2245                 return status;
2246
2247         nfs4_lock_state();
2248
2249         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2250                                         oc->oc_seqid, &oc->oc_req_stateid,
2251                                         CONFIRM | OPEN_STATE,
2252                                         &oc->oc_stateowner, &stp, NULL)))
2253                 goto out; 
2254
2255         sop = oc->oc_stateowner;
2256         sop->so_confirmed = 1;
2257         update_stateid(&stp->st_stateid);
2258         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2259         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
2260                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2261                          stp->st_stateid.si_boot,
2262                          stp->st_stateid.si_stateownerid,
2263                          stp->st_stateid.si_fileid,
2264                          stp->st_stateid.si_generation);
2265
2266         nfsd4_create_clid_dir(sop->so_client);
2267 out:
2268         if (oc->oc_stateowner) {
2269                 nfs4_get_stateowner(oc->oc_stateowner);
2270                 cstate->replay_owner = oc->oc_stateowner;
2271         }
2272         nfs4_unlock_state();
2273         return status;
2274 }
2275
2276
2277 /*
2278  * unset all bits in union bitmap (bmap) that
2279  * do not exist in share (from successful OPEN_DOWNGRADE)
2280  */
2281 static void
2282 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2283 {
2284         int i;
2285         for (i = 1; i < 4; i++) {
2286                 if ((i & access) != i)
2287                         __clear_bit(i, bmap);
2288         }
2289 }
2290
2291 static void
2292 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2293 {
2294         int i;
2295         for (i = 0; i < 4; i++) {
2296                 if ((i & deny) != i)
2297                         __clear_bit(i, bmap);
2298         }
2299 }
2300
2301 __be32
2302 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2303                      struct nfsd4_compound_state *cstate,
2304                      struct nfsd4_open_downgrade *od)
2305 {
2306         __be32 status;
2307         struct nfs4_stateid *stp;
2308         unsigned int share_access;
2309
2310         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
2311                         (int)cstate->current_fh.fh_dentry->d_name.len,
2312                         cstate->current_fh.fh_dentry->d_name.name);
2313
2314         if (!access_valid(od->od_share_access)
2315                         || !deny_valid(od->od_share_deny))
2316                 return nfserr_inval;
2317
2318         nfs4_lock_state();
2319         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2320                                         od->od_seqid,
2321                                         &od->od_stateid, 
2322                                         OPEN_STATE,
2323                                         &od->od_stateowner, &stp, NULL)))
2324                 goto out; 
2325
2326         status = nfserr_inval;
2327         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2328                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2329                         stp->st_access_bmap, od->od_share_access);
2330                 goto out;
2331         }
2332         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2333                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2334                         stp->st_deny_bmap, od->od_share_deny);
2335                 goto out;
2336         }
2337         set_access(&share_access, stp->st_access_bmap);
2338         nfs4_file_downgrade(stp->st_vfs_file,
2339                             share_access & ~od->od_share_access);
2340
2341         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2342         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2343
2344         update_stateid(&stp->st_stateid);
2345         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2346         status = nfs_ok;
2347 out:
2348         if (od->od_stateowner) {
2349                 nfs4_get_stateowner(od->od_stateowner);
2350                 cstate->replay_owner = od->od_stateowner;
2351         }
2352         nfs4_unlock_state();
2353         return status;
2354 }
2355
2356 /*
2357  * nfs4_unlock_state() called after encode
2358  */
2359 __be32
2360 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2361             struct nfsd4_close *close)
2362 {
2363         __be32 status;
2364         struct nfs4_stateid *stp;
2365
2366         dprintk("NFSD: nfsd4_close on file %.*s\n", 
2367                         (int)cstate->current_fh.fh_dentry->d_name.len,
2368                         cstate->current_fh.fh_dentry->d_name.name);
2369
2370         nfs4_lock_state();
2371         /* check close_lru for replay */
2372         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2373                                         close->cl_seqid,
2374                                         &close->cl_stateid, 
2375                                         OPEN_STATE | CLOSE_STATE,
2376                                         &close->cl_stateowner, &stp, NULL)))
2377                 goto out; 
2378         status = nfs_ok;
2379         update_stateid(&stp->st_stateid);
2380         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2381
2382         /* release_stateid() calls nfsd_close() if needed */
2383         release_open_stateid(stp);
2384
2385         /* place unused nfs4_stateowners on so_close_lru list to be
2386          * released by the laundromat service after the lease period
2387          * to enable us to handle CLOSE replay
2388          */
2389         if (list_empty(&close->cl_stateowner->so_stateids))
2390                 move_to_close_lru(close->cl_stateowner);
2391 out:
2392         if (close->cl_stateowner) {
2393                 nfs4_get_stateowner(close->cl_stateowner);
2394                 cstate->replay_owner = close->cl_stateowner;
2395         }
2396         nfs4_unlock_state();
2397         return status;
2398 }
2399
2400 __be32
2401 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2402                   struct nfsd4_delegreturn *dr)
2403 {
2404         struct nfs4_delegation *dp;
2405         stateid_t *stateid = &dr->dr_stateid;
2406         struct inode *inode;
2407         __be32 status;
2408
2409         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2410                 return status;
2411         inode = cstate->current_fh.fh_dentry->d_inode;
2412
2413         nfs4_lock_state();
2414         status = nfserr_bad_stateid;
2415         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2416                 goto out;
2417         status = nfserr_stale_stateid;
2418         if (STALE_STATEID(stateid))
2419                 goto out;
2420         status = nfserr_bad_stateid;
2421         if (!is_delegation_stateid(stateid))
2422                 goto out;
2423         dp = find_delegation_stateid(inode, stateid);
2424         if (!dp)
2425                 goto out;
2426         status = check_stateid_generation(stateid, &dp->dl_stateid);
2427         if (status)
2428                 goto out;
2429         renew_client(dp->dl_client);
2430
2431         unhash_delegation(dp);
2432 out:
2433         nfs4_unlock_state();
2434
2435         return status;
2436 }
2437
2438
2439 /* 
2440  * Lock owner state (byte-range locks)
2441  */
2442 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2443 #define LOCK_HASH_BITS              8
2444 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2445 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2446
2447 static inline u64
2448 end_offset(u64 start, u64 len)
2449 {
2450         u64 end;
2451
2452         end = start + len;
2453         return end >= start ? end: NFS4_MAX_UINT64;
2454 }
2455
2456 /* last octet in a range */
2457 static inline u64
2458 last_byte_offset(u64 start, u64 len)
2459 {
2460         u64 end;
2461
2462         BUG_ON(!len);
2463         end = start + len;
2464         return end > start ? end - 1: NFS4_MAX_UINT64;
2465 }
2466
2467 #define lockownerid_hashval(id) \
2468         ((id) & LOCK_HASH_MASK)
2469
2470 static inline unsigned int
2471 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2472                 struct xdr_netobj *ownername)
2473 {
2474         return (file_hashval(inode) + cl_id
2475                         + opaque_hashval(ownername->data, ownername->len))
2476                 & LOCK_HASH_MASK;
2477 }
2478
2479 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2480 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2481 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2482
2483 static struct nfs4_stateid *
2484 find_stateid(stateid_t *stid, int flags)
2485 {
2486         struct nfs4_stateid *local;
2487         u32 st_id = stid->si_stateownerid;
2488         u32 f_id = stid->si_fileid;
2489         unsigned int hashval;
2490
2491         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2492         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
2493                 hashval = stateid_hashval(st_id, f_id);
2494                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2495                         if ((local->st_stateid.si_stateownerid == st_id) &&
2496                             (local->st_stateid.si_fileid == f_id))
2497                                 return local;
2498                 }
2499         } 
2500
2501         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
2502                 hashval = stateid_hashval(st_id, f_id);
2503                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2504                         if ((local->st_stateid.si_stateownerid == st_id) &&
2505                             (local->st_stateid.si_fileid == f_id))
2506                                 return local;
2507                 }
2508         }
2509         return NULL;
2510 }
2511
2512 static struct nfs4_delegation *
2513 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2514 {
2515         struct nfs4_file *fp;
2516         struct nfs4_delegation *dl;
2517
2518         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2519                     stid->si_boot, stid->si_stateownerid,
2520                     stid->si_fileid, stid->si_generation);
2521
2522         fp = find_file(ino);
2523         if (!fp)
2524                 return NULL;
2525         dl = find_delegation_file(fp, stid);
2526         put_nfs4_file(fp);
2527         return dl;
2528 }
2529
2530 /*
2531  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2532  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2533  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2534  * locking, this prevents us from being completely protocol-compliant.  The
2535  * real solution to this problem is to start using unsigned file offsets in
2536  * the VFS, but this is a very deep change!
2537  */
2538 static inline void
2539 nfs4_transform_lock_offset(struct file_lock *lock)
2540 {
2541         if (lock->fl_start < 0)
2542                 lock->fl_start = OFFSET_MAX;
2543         if (lock->fl_end < 0)
2544                 lock->fl_end = OFFSET_MAX;
2545 }
2546
2547 /* Hack!: For now, we're defining this just so we can use a pointer to it
2548  * as a unique cookie to identify our (NFSv4's) posix locks. */
2549 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2550 };
2551
2552 static inline void
2553 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2554 {
2555         struct nfs4_stateowner *sop;
2556         unsigned int hval;
2557
2558         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2559                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2560                 hval = lockownerid_hashval(sop->so_id);
2561                 kref_get(&sop->so_ref);
2562                 deny->ld_sop = sop;
2563                 deny->ld_clientid = sop->so_client->cl_clientid;
2564         } else {
2565                 deny->ld_sop = NULL;
2566                 deny->ld_clientid.cl_boot = 0;
2567                 deny->ld_clientid.cl_id = 0;
2568         }
2569         deny->ld_start = fl->fl_start;
2570         deny->ld_length = NFS4_MAX_UINT64;
2571         if (fl->fl_end != NFS4_MAX_UINT64)
2572                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
2573         deny->ld_type = NFS4_READ_LT;
2574         if (fl->fl_type != F_RDLCK)
2575                 deny->ld_type = NFS4_WRITE_LT;
2576 }
2577
2578 static struct nfs4_stateowner *
2579 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2580                 struct xdr_netobj *owner)
2581 {
2582         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2583         struct nfs4_stateowner *op;
2584
2585         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2586                 if (same_owner_str(op, owner, clid))
2587                         return op;
2588         }
2589         return NULL;
2590 }
2591
2592 /*
2593  * Alloc a lock owner structure.
2594  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
2595  * occured. 
2596  *
2597  * strhashval = lock_ownerstr_hashval 
2598  */
2599
2600 static struct nfs4_stateowner *
2601 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2602         struct nfs4_stateowner *sop;
2603         struct nfs4_replay *rp;
2604         unsigned int idhashval;
2605
2606         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2607                 return NULL;
2608         idhashval = lockownerid_hashval(current_ownerid);
2609         INIT_LIST_HEAD(&sop->so_idhash);
2610         INIT_LIST_HEAD(&sop->so_strhash);
2611         INIT_LIST_HEAD(&sop->so_perclient);
2612         INIT_LIST_HEAD(&sop->so_stateids);
2613         INIT_LIST_HEAD(&sop->so_perstateid);
2614         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2615         sop->so_time = 0;
2616         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2617         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2618         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2619         sop->so_is_open_owner = 0;
2620         sop->so_id = current_ownerid++;
2621         sop->so_client = clp;
2622         /* It is the openowner seqid that will be incremented in encode in the
2623          * case of new lockowners; so increment the lock seqid manually: */
2624         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2625         sop->so_confirmed = 1;
2626         rp = &sop->so_replay;
2627         rp->rp_status = nfserr_serverfault;
2628         rp->rp_buflen = 0;
2629         rp->rp_buf = rp->rp_ibuf;
2630         return sop;
2631 }
2632
2633 static struct nfs4_stateid *
2634 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2635 {
2636         struct nfs4_stateid *stp;
2637         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2638
2639         stp = nfs4_alloc_stateid();
2640         if (stp == NULL)
2641                 goto out;
2642         INIT_LIST_HEAD(&stp->st_hash);
2643         INIT_LIST_HEAD(&stp->st_perfile);
2644         INIT_LIST_HEAD(&stp->st_perstateowner);
2645         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2646         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2647         list_add(&stp->st_perfile, &fp->fi_stateids);
2648         list_add(&stp->st_perstateowner, &sop->so_stateids);
2649         stp->st_stateowner = sop;
2650         get_nfs4_file(fp);
2651         stp->st_file = fp;
2652         stp->st_stateid.si_boot = boot_time;
2653         stp->st_stateid.si_stateownerid = sop->so_id;
2654         stp->st_stateid.si_fileid = fp->fi_id;
2655         stp->st_stateid.si_generation = 0;
2656         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2657         stp->st_access_bmap = open_stp->st_access_bmap;
2658         stp->st_deny_bmap = open_stp->st_deny_bmap;
2659         stp->st_openstp = open_stp;
2660
2661 out:
2662         return stp;
2663 }
2664
2665 static int
2666 check_lock_length(u64 offset, u64 length)
2667 {
2668         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
2669              LOFF_OVERFLOW(offset, length)));
2670 }
2671
2672 /*
2673  *  LOCK operation 
2674  */
2675 __be32
2676 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2677            struct nfsd4_lock *lock)
2678 {
2679         struct nfs4_stateowner *open_sop = NULL;
2680         struct nfs4_stateowner *lock_sop = NULL;
2681         struct nfs4_stateid *lock_stp;
2682         struct file *filp;
2683         struct file_lock file_lock;
2684         struct file_lock conflock;
2685         __be32 status = 0;
2686         unsigned int strhashval;
2687         unsigned int cmd;
2688         int err;
2689
2690         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2691                 (long long) lock->lk_offset,
2692                 (long long) lock->lk_length);
2693
2694         if (check_lock_length(lock->lk_offset, lock->lk_length))
2695                  return nfserr_inval;
2696
2697         if ((status = fh_verify(rqstp, &cstate->current_fh,
2698                                 S_IFREG, NFSD_MAY_LOCK))) {
2699                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2700                 return status;
2701         }
2702
2703         nfs4_lock_state();
2704
2705         if (lock->lk_is_new) {
2706                 /*
2707                  * Client indicates that this is a new lockowner.
2708                  * Use open owner and open stateid to create lock owner and
2709                  * lock stateid.
2710                  */
2711                 struct nfs4_stateid *open_stp = NULL;
2712                 struct nfs4_file *fp;
2713                 
2714                 status = nfserr_stale_clientid;
2715                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2716                         goto out;
2717
2718                 /* validate and update open stateid and open seqid */
2719                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2720                                         lock->lk_new_open_seqid,
2721                                         &lock->lk_new_open_stateid,
2722                                         OPEN_STATE,
2723                                         &lock->lk_replay_owner, &open_stp,
2724                                         lock);
2725                 if (status)
2726                         goto out;
2727                 open_sop = lock->lk_replay_owner;
2728                 /* create lockowner and lock stateid */
2729                 fp = open_stp->st_file;
2730                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
2731                                 open_sop->so_client->cl_clientid.cl_id, 
2732                                 &lock->v.new.owner);
2733                 /* XXX: Do we need to check for duplicate stateowners on
2734                  * the same file, or should they just be allowed (and
2735                  * create new stateids)? */
2736                 status = nfserr_resource;
2737                 lock_sop = alloc_init_lock_stateowner(strhashval,
2738                                 open_sop->so_client, open_stp, lock);
2739                 if (lock_sop == NULL)
2740                         goto out;
2741                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2742                 if (lock_stp == NULL)
2743                         goto out;
2744         } else {
2745                 /* lock (lock owner + lock stateid) already exists */
2746                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2747                                        lock->lk_old_lock_seqid, 
2748                                        &lock->lk_old_lock_stateid, 
2749                                        LOCK_STATE,
2750                                        &lock->lk_replay_owner, &lock_stp, lock);
2751                 if (status)
2752                         goto out;
2753                 lock_sop = lock->lk_replay_owner;
2754         }
2755         /* lock->lk_replay_owner and lock_stp have been created or found */
2756         filp = lock_stp->st_vfs_file;
2757
2758         status = nfserr_grace;
2759         if (locks_in_grace() && !lock->lk_reclaim)
2760                 goto out;
2761         status = nfserr_no_grace;
2762         if (!locks_in_grace() && lock->lk_reclaim)
2763                 goto out;
2764
2765         locks_init_lock(&file_lock);
2766         switch (lock->lk_type) {
2767                 case NFS4_READ_LT:
2768                 case NFS4_READW_LT:
2769                         file_lock.fl_type = F_RDLCK;
2770                         cmd = F_SETLK;
2771                 break;
2772                 case NFS4_WRITE_LT:
2773                 case NFS4_WRITEW_LT:
2774                         file_lock.fl_type = F_WRLCK;
2775                         cmd = F_SETLK;
2776                 break;
2777                 default:
2778                         status = nfserr_inval;
2779                 goto out;
2780         }
2781         file_lock.fl_owner = (fl_owner_t)lock_sop;
2782         file_lock.fl_pid = current->tgid;
2783         file_lock.fl_file = filp;
2784         file_lock.fl_flags = FL_POSIX;
2785         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2786
2787         file_lock.fl_start = lock->lk_offset;
2788         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
2789         nfs4_transform_lock_offset(&file_lock);
2790
2791         /*
2792         * Try to lock the file in the VFS.
2793         * Note: locks.c uses the BKL to protect the inode's lock list.
2794         */
2795
2796         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2797         switch (-err) {
2798         case 0: /* success! */
2799                 update_stateid(&lock_stp->st_stateid);
2800                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
2801                                 sizeof(stateid_t));
2802                 status = 0;
2803                 break;
2804         case (EAGAIN):          /* conflock holds conflicting lock */
2805                 status = nfserr_denied;
2806                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2807                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2808                 break;
2809         case (EDEADLK):
2810                 status = nfserr_deadlock;
2811                 break;
2812         default:        
2813                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2814                 status = nfserr_resource;
2815                 break;
2816         }
2817 out:
2818         if (status && lock->lk_is_new && lock_sop)
2819                 release_lockowner(lock_sop);
2820         if (lock->lk_replay_owner) {
2821                 nfs4_get_stateowner(lock->lk_replay_owner);
2822                 cstate->replay_owner = lock->lk_replay_owner;
2823         }
2824         nfs4_unlock_state();
2825         return status;
2826 }
2827
2828 /*
2829  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
2830  * so we do a temporary open here just to get an open file to pass to
2831  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
2832  * inode operation.)
2833  */
2834 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
2835 {
2836         struct file *file;
2837         int err;
2838
2839         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
2840         if (err)
2841                 return err;
2842         err = vfs_test_lock(file, lock);
2843         nfsd_close(file);
2844         return err;
2845 }
2846
2847 /*
2848  * LOCKT operation
2849  */
2850 __be32
2851 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2852             struct nfsd4_lockt *lockt)
2853 {
2854         struct inode *inode;
2855         struct file_lock file_lock;
2856         int error;
2857         __be32 status;
2858
2859         if (locks_in_grace())
2860                 return nfserr_grace;
2861
2862         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2863                  return nfserr_inval;
2864
2865         lockt->lt_stateowner = NULL;
2866         nfs4_lock_state();
2867
2868         status = nfserr_stale_clientid;
2869         if (STALE_CLIENTID(&lockt->lt_clientid))
2870                 goto out;
2871
2872         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2873                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2874                 if (status == nfserr_symlink)
2875                         status = nfserr_inval;
2876                 goto out;
2877         }
2878
2879         inode = cstate->current_fh.fh_dentry->d_inode;
2880         locks_init_lock(&file_lock);
2881         switch (lockt->lt_type) {
2882                 case NFS4_READ_LT:
2883                 case NFS4_READW_LT:
2884                         file_lock.fl_type = F_RDLCK;
2885                 break;
2886                 case NFS4_WRITE_LT:
2887                 case NFS4_WRITEW_LT:
2888                         file_lock.fl_type = F_WRLCK;
2889                 break;
2890                 default:
2891                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2892                         status = nfserr_inval;
2893                 goto out;
2894         }
2895
2896         lockt->lt_stateowner = find_lockstateowner_str(inode,
2897                         &lockt->lt_clientid, &lockt->lt_owner);
2898         if (lockt->lt_stateowner)
2899                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2900         file_lock.fl_pid = current->tgid;
2901         file_lock.fl_flags = FL_POSIX;
2902
2903         file_lock.fl_start = lockt->lt_offset;
2904         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
2905
2906         nfs4_transform_lock_offset(&file_lock);
2907
2908         status = nfs_ok;
2909         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
2910         if (error) {
2911                 status = nfserrno(error);
2912                 goto out;
2913         }
2914         if (file_lock.fl_type != F_UNLCK) {
2915                 status = nfserr_denied;
2916                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2917         }
2918 out:
2919         nfs4_unlock_state();
2920         return status;
2921 }
2922
2923 __be32
2924 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2925             struct nfsd4_locku *locku)
2926 {
2927         struct nfs4_stateid *stp;
2928         struct file *filp = NULL;
2929         struct file_lock file_lock;
2930         __be32 status;
2931         int err;
2932                                                         
2933         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2934                 (long long) locku->lu_offset,
2935                 (long long) locku->lu_length);
2936
2937         if (check_lock_length(locku->lu_offset, locku->lu_length))
2938                  return nfserr_inval;
2939
2940         nfs4_lock_state();
2941                                                                                 
2942         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2943                                         locku->lu_seqid, 
2944                                         &locku->lu_stateid, 
2945                                         LOCK_STATE,
2946                                         &locku->lu_stateowner, &stp, NULL)))
2947                 goto out;
2948
2949         filp = stp->st_vfs_file;
2950         BUG_ON(!filp);
2951         locks_init_lock(&file_lock);
2952         file_lock.fl_type = F_UNLCK;
2953         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2954         file_lock.fl_pid = current->tgid;
2955         file_lock.fl_file = filp;
2956         file_lock.fl_flags = FL_POSIX; 
2957         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2958         file_lock.fl_start = locku->lu_offset;
2959
2960         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
2961         nfs4_transform_lock_offset(&file_lock);
2962
2963         /*
2964         *  Try to unlock the file in the VFS.
2965         */
2966         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2967         if (err) {
2968                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2969                 goto out_nfserr;
2970         }
2971         /*
2972         * OK, unlock succeeded; the only thing left to do is update the stateid.
2973         */
2974         update_stateid(&stp->st_stateid);
2975         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2976
2977 out:
2978         if (locku->lu_stateowner) {
2979                 nfs4_get_stateowner(locku->lu_stateowner);
2980                 cstate->replay_owner = locku->lu_stateowner;
2981         }
2982         nfs4_unlock_state();
2983         return status;
2984
2985 out_nfserr:
2986         status = nfserrno(err);
2987         goto out;
2988 }
2989
2990 /*
2991  * returns
2992  *      1: locks held by lockowner
2993  *      0: no locks held by lockowner
2994  */
2995 static int
2996 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2997 {
2998         struct file_lock **flpp;
2999         struct inode *inode = filp->f_path.dentry->d_inode;
3000         int status = 0;
3001
3002         lock_kernel();
3003         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3004                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3005                         status = 1;
3006                         goto out;
3007                 }
3008         }
3009 out:
3010         unlock_kernel();
3011         return status;
3012 }
3013
3014 __be32
3015 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3016                         struct nfsd4_compound_state *cstate,
3017                         struct nfsd4_release_lockowner *rlockowner)
3018 {
3019         clientid_t *clid = &rlockowner->rl_clientid;
3020         struct nfs4_stateowner *sop;
3021         struct nfs4_stateid *stp;
3022         struct xdr_netobj *owner = &rlockowner->rl_owner;
3023         struct list_head matches;
3024         int i;
3025         __be32 status;
3026
3027         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3028                 clid->cl_boot, clid->cl_id);
3029
3030         /* XXX check for lease expiration */
3031
3032         status = nfserr_stale_clientid;
3033         if (STALE_CLIENTID(clid))
3034                 return status;
3035
3036         nfs4_lock_state();
3037
3038         status = nfserr_locks_held;
3039         /* XXX: we're doing a linear search through all the lockowners.
3040          * Yipes!  For now we'll just hope clients aren't really using
3041          * release_lockowner much, but eventually we have to fix these
3042          * data structures. */
3043         INIT_LIST_HEAD(&matches);
3044         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3045                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3046                         if (!same_owner_str(sop, owner, clid))
3047                                 continue;
3048                         list_for_each_entry(stp, &sop->so_stateids,
3049                                         st_perstateowner) {
3050                                 if (check_for_locks(stp->st_vfs_file, sop))
3051                                         goto out;
3052                                 /* Note: so_perclient unused for lockowners,
3053                                  * so it's OK to fool with here. */
3054                                 list_add(&sop->so_perclient, &matches);
3055                         }
3056                 }
3057         }
3058         /* Clients probably won't expect us to return with some (but not all)
3059          * of the lockowner state released; so don't release any until all
3060          * have been checked. */
3061         status = nfs_ok;
3062         while (!list_empty(&matches)) {
3063                 sop = list_entry(matches.next, struct nfs4_stateowner,
3064                                                                 so_perclient);
3065                 /* unhash_stateowner deletes so_perclient only
3066                  * for openowners. */
3067                 list_del(&sop->so_perclient);
3068                 release_lockowner(sop);
3069         }
3070 out:
3071         nfs4_unlock_state();
3072         return status;
3073 }
3074
3075 static inline struct nfs4_client_reclaim *
3076 alloc_reclaim(void)
3077 {
3078         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3079 }
3080
3081 int
3082 nfs4_has_reclaimed_state(const char *name)
3083 {
3084         unsigned int strhashval = clientstr_hashval(name);
3085         struct nfs4_client *clp;
3086
3087         clp = find_confirmed_client_by_str(name, strhashval);
3088         return clp ? 1 : 0;
3089 }
3090
3091 /*
3092  * failure => all reset bets are off, nfserr_no_grace...
3093  */
3094 int
3095 nfs4_client_to_reclaim(const char *name)
3096 {
3097         unsigned int strhashval;
3098         struct nfs4_client_reclaim *crp = NULL;
3099
3100         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3101         crp = alloc_reclaim();
3102         if (!crp)
3103                 return 0;
3104         strhashval = clientstr_hashval(name);
3105         INIT_LIST_HEAD(&crp->cr_strhash);
3106         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3107         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3108         reclaim_str_hashtbl_size++;
3109         return 1;
3110 }
3111
3112 static void
3113 nfs4_release_reclaim(void)
3114 {
3115         struct nfs4_client_reclaim *crp = NULL;
3116         int i;
3117
3118         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3119                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3120                         crp = list_entry(reclaim_str_hashtbl[i].next,
3121                                         struct nfs4_client_reclaim, cr_strhash);
3122                         list_del(&crp->cr_strhash);
3123                         kfree(crp);
3124                         reclaim_str_hashtbl_size--;
3125                 }
3126         }
3127         BUG_ON(reclaim_str_hashtbl_size);
3128 }
3129
3130 /*
3131  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3132 static struct nfs4_client_reclaim *
3133 nfs4_find_reclaim_client(clientid_t *clid)
3134 {
3135         unsigned int strhashval;
3136         struct nfs4_client *clp;
3137         struct nfs4_client_reclaim *crp = NULL;
3138
3139
3140         /* find clientid in conf_id_hashtbl */
3141         clp = find_confirmed_client(clid);
3142         if (clp == NULL)
3143                 return NULL;
3144
3145         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3146                             clp->cl_name.len, clp->cl_name.data,
3147                             clp->cl_recdir);
3148
3149         /* find clp->cl_name in reclaim_str_hashtbl */
3150         strhashval = clientstr_hashval(clp->cl_recdir);
3151         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3152                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3153                         return crp;
3154                 }
3155         }
3156         return NULL;
3157 }
3158
3159 /*
3160 * Called from OPEN. Look for clientid in reclaim list.
3161 */
3162 __be32
3163 nfs4_check_open_reclaim(clientid_t *clid)
3164 {
3165         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3166 }
3167
3168 /* initialization to perform at module load time: */
3169
3170 int
3171 nfs4_state_init(void)
3172 {
3173         int i, status;
3174
3175         status = nfsd4_init_slabs();
3176         if (status)
3177                 return status;
3178         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3179                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3180                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3181                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3182                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3183         }
3184         for (i = 0; i < FILE_HASH_SIZE; i++) {
3185                 INIT_LIST_HEAD(&file_hashtbl[i]);
3186         }
3187         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3188                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3189                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3190         }
3191         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3192                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3193                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3194         }
3195         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3196                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3197                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3198         }
3199         memset(&onestateid, ~0, sizeof(stateid_t));
3200         INIT_LIST_HEAD(&close_lru);
3201         INIT_LIST_HEAD(&client_lru);
3202         INIT_LIST_HEAD(&del_recall_lru);
3203         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3204                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3205         reclaim_str_hashtbl_size = 0;
3206         return 0;
3207 }
3208
3209 static void
3210 nfsd4_load_reboot_recovery_data(void)
3211 {
3212         int status;
3213
3214         nfs4_lock_state();
3215         nfsd4_init_recdir(user_recovery_dirname);
3216         status = nfsd4_recdir_load();
3217         nfs4_unlock_state();
3218         if (status)
3219                 printk("NFSD: Failure reading reboot recovery data\n");
3220 }
3221
3222 unsigned long
3223 get_nfs4_grace_period(void)
3224 {
3225         return max(user_lease_time, lease_time) * HZ;
3226 }
3227
3228 /*
3229  * Since the lifetime of a delegation isn't limited to that of an open, a
3230  * client may quite reasonably hang on to a delegation as long as it has
3231  * the inode cached.  This becomes an obvious problem the first time a
3232  * client's inode cache approaches the size of the server's total memory.
3233  *
3234  * For now we avoid this problem by imposing a hard limit on the number
3235  * of delegations, which varies according to the server's memory size.
3236  */
3237 static void
3238 set_max_delegations(void)
3239 {
3240         /*
3241          * Allow at most 4 delegations per megabyte of RAM.  Quick
3242          * estimates suggest that in the worst case (where every delegation
3243          * is for a different inode), a delegation could take about 1.5K,
3244          * giving a worst case usage of about 6% of memory.
3245          */
3246         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3247 }
3248
3249 /* initialization to perform when the nfsd service is started: */
3250
3251 static void
3252 __nfs4_state_start(void)
3253 {
3254         unsigned long grace_time;
3255
3256         boot_time = get_seconds();
3257         grace_time = get_nfs4_grace_period();
3258         lease_time = user_lease_time;
3259         locks_start_grace(&nfsd4_manager);
3260         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3261                grace_time/HZ);
3262         laundry_wq = create_singlethread_workqueue("nfsd4");
3263         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3264         set_max_delegations();
3265 }
3266
3267 void
3268 nfs4_state_start(void)
3269 {
3270         if (nfs4_init)
3271                 return;
3272         nfsd4_load_reboot_recovery_data();
3273         __nfs4_state_start();
3274         nfs4_init = 1;
3275         return;
3276 }
3277
3278 time_t
3279 nfs4_lease_time(void)
3280 {
3281         return lease_time;
3282 }
3283
3284 static void
3285 __nfs4_state_shutdown(void)
3286 {
3287         int i;
3288         struct nfs4_client *clp = NULL;
3289         struct nfs4_delegation *dp = NULL;
3290         struct list_head *pos, *next, reaplist;
3291
3292         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3293                 while (!list_empty(&conf_id_hashtbl[i])) {
3294                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3295                         expire_client(clp);
3296                 }
3297                 while (!list_empty(&unconf_str_hashtbl[i])) {
3298                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3299                         expire_client(clp);
3300                 }
3301         }
3302         INIT_LIST_HEAD(&reaplist);
3303         spin_lock(&recall_lock);
3304         list_for_each_safe(pos, next, &del_recall_lru) {
3305                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3306                 list_move(&dp->dl_recall_lru, &reaplist);
3307         }
3308         spin_unlock(&recall_lock);
3309         list_for_each_safe(pos, next, &reaplist) {
3310                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3311                 list_del_init(&dp->dl_recall_lru);
3312                 unhash_delegation(dp);
3313         }
3314
3315         nfsd4_shutdown_recdir();
3316         nfs4_init = 0;
3317 }
3318
3319 void
3320 nfs4_state_shutdown(void)
3321 {
3322         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3323         destroy_workqueue(laundry_wq);
3324         locks_end_grace(&nfsd4_manager);
3325         nfs4_lock_state();
3326         nfs4_release_reclaim();
3327         __nfs4_state_shutdown();
3328         nfs4_unlock_state();
3329 }
3330
3331 /*
3332  * user_recovery_dirname is protected by the nfsd_mutex since it's only
3333  * accessed when nfsd is starting.
3334  */
3335 static void
3336 nfs4_set_recdir(char *recdir)
3337 {
3338         strcpy(user_recovery_dirname, recdir);
3339 }
3340
3341 /*
3342  * Change the NFSv4 recovery directory to recdir.
3343  */
3344 int
3345 nfs4_reset_recoverydir(char *recdir)
3346 {
3347         int status;
3348         struct path path;
3349
3350         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3351         if (status)
3352                 return status;
3353         status = -ENOTDIR;
3354         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3355                 nfs4_set_recdir(recdir);
3356                 status = 0;
3357         }
3358         path_put(&path);
3359         return status;
3360 }
3361
3362 char *
3363 nfs4_recoverydir(void)
3364 {
3365         return user_recovery_dirname;
3366 }
3367
3368 /*
3369  * Called when leasetime is changed.
3370  *
3371  * The only way the protocol gives us to handle on-the-fly lease changes is to
3372  * simulate a reboot.  Instead of doing that, we just wait till the next time
3373  * we start to register any changes in lease time.  If the administrator
3374  * really wants to change the lease time *now*, they can go ahead and bring
3375  * nfsd down and then back up again after changing the lease time.
3376  *
3377  * user_lease_time is protected by nfsd_mutex since it's only really accessed
3378  * when nfsd is starting
3379  */
3380 void
3381 nfs4_reset_lease(time_t leasetime)
3382 {
3383         user_lease_time = leasetime;
3384 }