13e0e074dbb8154aa4cc49a45405478efe929cb7
[linux-2.6.git] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58
59 #define NFSDDBG_FACILITY                NFSDDBG_PROC
60
61 /* Globals */
62 static time_t lease_time = 90;     /* default lease time */
63 static time_t user_lease_time = 90;
64 static time_t boot_time;
65 static u32 current_ownerid = 1;
66 static u32 current_fileid = 1;
67 static u32 current_delegid = 1;
68 static u32 nfs4_init;
69 static stateid_t zerostateid;             /* bits all 0 */
70 static stateid_t onestateid;              /* bits all 1 */
71
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 void release_stateid_lockowners(struct nfs4_stateid *open_stp);
79 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
80 static void nfs4_set_recdir(char *recdir);
81
82 /* Locking:
83  *
84  * client_mutex:
85  *      protects clientid_hashtbl[], clientstr_hashtbl[],
86  *      unconfstr_hashtbl[], uncofid_hashtbl[].
87  */
88 static DEFINE_MUTEX(client_mutex);
89
90 static struct kmem_cache *stateowner_slab = NULL;
91 static struct kmem_cache *file_slab = NULL;
92 static struct kmem_cache *stateid_slab = NULL;
93 static struct kmem_cache *deleg_slab = NULL;
94
95 void
96 nfs4_lock_state(void)
97 {
98         mutex_lock(&client_mutex);
99 }
100
101 void
102 nfs4_unlock_state(void)
103 {
104         mutex_unlock(&client_mutex);
105 }
106
107 static inline u32
108 opaque_hashval(const void *ptr, int nbytes)
109 {
110         unsigned char *cptr = (unsigned char *) ptr;
111
112         u32 x = 0;
113         while (nbytes--) {
114                 x *= 37;
115                 x += *cptr++;
116         }
117         return x;
118 }
119
120 /* forward declarations */
121 static void release_stateowner(struct nfs4_stateowner *sop);
122 static void release_stateid(struct nfs4_stateid *stp, int flags);
123
124 /*
125  * Delegation state
126  */
127
128 /* recall_lock protects the del_recall_lru */
129 static DEFINE_SPINLOCK(recall_lock);
130 static struct list_head del_recall_lru;
131
132 static void
133 free_nfs4_file(struct kref *kref)
134 {
135         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
136         list_del(&fp->fi_hash);
137         iput(fp->fi_inode);
138         kmem_cache_free(file_slab, fp);
139 }
140
141 static inline void
142 put_nfs4_file(struct nfs4_file *fi)
143 {
144         kref_put(&fi->fi_ref, free_nfs4_file);
145 }
146
147 static inline void
148 get_nfs4_file(struct nfs4_file *fi)
149 {
150         kref_get(&fi->fi_ref);
151 }
152
153 static int num_delegations;
154 unsigned int max_delegations;
155
156 /*
157  * Open owner state (share locks)
158  */
159
160 /* hash tables for nfs4_stateowner */
161 #define OWNER_HASH_BITS              8
162 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
163 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
164
165 #define ownerid_hashval(id) \
166         ((id) & OWNER_HASH_MASK)
167 #define ownerstr_hashval(clientid, ownername) \
168         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
169
170 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
171 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
172
173 /* hash table for nfs4_file */
174 #define FILE_HASH_BITS                   8
175 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
176 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
177 /* hash table for (open)nfs4_stateid */
178 #define STATEID_HASH_BITS              10
179 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
180 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
181
182 #define file_hashval(x) \
183         hash_ptr(x, FILE_HASH_BITS)
184 #define stateid_hashval(owner_id, file_id)  \
185         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
186
187 static struct list_head file_hashtbl[FILE_HASH_SIZE];
188 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
189
190 static struct nfs4_delegation *
191 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
192 {
193         struct nfs4_delegation *dp;
194         struct nfs4_file *fp = stp->st_file;
195         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
196
197         dprintk("NFSD alloc_init_deleg\n");
198         if (fp->fi_had_conflict)
199                 return NULL;
200         if (num_delegations > max_delegations)
201                 return NULL;
202         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
203         if (dp == NULL)
204                 return dp;
205         num_delegations++;
206         INIT_LIST_HEAD(&dp->dl_perfile);
207         INIT_LIST_HEAD(&dp->dl_perclnt);
208         INIT_LIST_HEAD(&dp->dl_recall_lru);
209         dp->dl_client = clp;
210         get_nfs4_file(fp);
211         dp->dl_file = fp;
212         dp->dl_flock = NULL;
213         get_file(stp->st_vfs_file);
214         dp->dl_vfs_file = stp->st_vfs_file;
215         dp->dl_type = type;
216         dp->dl_recall.cbr_dp = NULL;
217         dp->dl_recall.cbr_ident = cb->cb_ident;
218         dp->dl_recall.cbr_trunc = 0;
219         dp->dl_stateid.si_boot = boot_time;
220         dp->dl_stateid.si_stateownerid = current_delegid++;
221         dp->dl_stateid.si_fileid = 0;
222         dp->dl_stateid.si_generation = 0;
223         dp->dl_fhlen = current_fh->fh_handle.fh_size;
224         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
225                         current_fh->fh_handle.fh_size);
226         dp->dl_time = 0;
227         atomic_set(&dp->dl_count, 1);
228         list_add(&dp->dl_perfile, &fp->fi_delegations);
229         list_add(&dp->dl_perclnt, &clp->cl_delegations);
230         return dp;
231 }
232
233 void
234 nfs4_put_delegation(struct nfs4_delegation *dp)
235 {
236         if (atomic_dec_and_test(&dp->dl_count)) {
237                 dprintk("NFSD: freeing dp %p\n",dp);
238                 put_nfs4_file(dp->dl_file);
239                 kmem_cache_free(deleg_slab, dp);
240                 num_delegations--;
241         }
242 }
243
244 /* Remove the associated file_lock first, then remove the delegation.
245  * lease_modify() is called to remove the FS_LEASE file_lock from
246  * the i_flock list, eventually calling nfsd's lock_manager
247  * fl_release_callback.
248  */
249 static void
250 nfs4_close_delegation(struct nfs4_delegation *dp)
251 {
252         struct file *filp = dp->dl_vfs_file;
253
254         dprintk("NFSD: close_delegation dp %p\n",dp);
255         dp->dl_vfs_file = NULL;
256         /* The following nfsd_close may not actually close the file,
257          * but we want to remove the lease in any case. */
258         if (dp->dl_flock)
259                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
260         nfsd_close(filp);
261 }
262
263 /* Called under the state lock. */
264 static void
265 unhash_delegation(struct nfs4_delegation *dp)
266 {
267         list_del_init(&dp->dl_perfile);
268         list_del_init(&dp->dl_perclnt);
269         spin_lock(&recall_lock);
270         list_del_init(&dp->dl_recall_lru);
271         spin_unlock(&recall_lock);
272         nfs4_close_delegation(dp);
273         nfs4_put_delegation(dp);
274 }
275
276 /* 
277  * SETCLIENTID state 
278  */
279
280 /* Hash tables for nfs4_clientid state */
281 #define CLIENT_HASH_BITS                 4
282 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
283 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
284
285 #define clientid_hashval(id) \
286         ((id) & CLIENT_HASH_MASK)
287 #define clientstr_hashval(name) \
288         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
289 /*
290  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
291  * used in reboot/reset lease grace period processing
292  *
293  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
294  * setclientid_confirmed info. 
295  *
296  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
297  * setclientid info.
298  *
299  * client_lru holds client queue ordered by nfs4_client.cl_time
300  * for lease renewal.
301  *
302  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
303  * for last close replay.
304  */
305 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
306 static int reclaim_str_hashtbl_size = 0;
307 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
308 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
309 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
310 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
311 static struct list_head client_lru;
312 static struct list_head close_lru;
313
314 static inline void
315 renew_client(struct nfs4_client *clp)
316 {
317         /*
318         * Move client to the end to the LRU list.
319         */
320         dprintk("renewing client (clientid %08x/%08x)\n", 
321                         clp->cl_clientid.cl_boot, 
322                         clp->cl_clientid.cl_id);
323         list_move_tail(&clp->cl_lru, &client_lru);
324         clp->cl_time = get_seconds();
325 }
326
327 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
328 static int
329 STALE_CLIENTID(clientid_t *clid)
330 {
331         if (clid->cl_boot == boot_time)
332                 return 0;
333         dprintk("NFSD stale clientid (%08x/%08x)\n", 
334                         clid->cl_boot, clid->cl_id);
335         return 1;
336 }
337
338 /* 
339  * XXX Should we use a slab cache ?
340  * This type of memory management is somewhat inefficient, but we use it
341  * anyway since SETCLIENTID is not a common operation.
342  */
343 static struct nfs4_client *alloc_client(struct xdr_netobj name)
344 {
345         struct nfs4_client *clp;
346
347         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
348         if (clp == NULL)
349                 return NULL;
350         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
351         if (clp->cl_name.data == NULL) {
352                 kfree(clp);
353                 return NULL;
354         }
355         memcpy(clp->cl_name.data, name.data, name.len);
356         clp->cl_name.len = name.len;
357         return clp;
358 }
359
360 static void
361 shutdown_callback_client(struct nfs4_client *clp)
362 {
363         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
364
365         if (clnt) {
366                 /*
367                  * Callback threads take a reference on the client, so there
368                  * should be no outstanding callbacks at this point.
369                  */
370                 clp->cl_callback.cb_client = NULL;
371                 rpc_shutdown_client(clnt);
372         }
373 }
374
375 static inline void
376 free_client(struct nfs4_client *clp)
377 {
378         shutdown_callback_client(clp);
379         if (clp->cl_cred.cr_group_info)
380                 put_group_info(clp->cl_cred.cr_group_info);
381         kfree(clp->cl_principal);
382         kfree(clp->cl_name.data);
383         kfree(clp);
384 }
385
386 void
387 put_nfs4_client(struct nfs4_client *clp)
388 {
389         if (atomic_dec_and_test(&clp->cl_count))
390                 free_client(clp);
391 }
392
393 static void
394 expire_client(struct nfs4_client *clp)
395 {
396         struct nfs4_stateowner *sop;
397         struct nfs4_delegation *dp;
398         struct list_head reaplist;
399
400         dprintk("NFSD: expire_client cl_count %d\n",
401                             atomic_read(&clp->cl_count));
402
403         INIT_LIST_HEAD(&reaplist);
404         spin_lock(&recall_lock);
405         while (!list_empty(&clp->cl_delegations)) {
406                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
407                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
408                                 dp->dl_flock);
409                 list_del_init(&dp->dl_perclnt);
410                 list_move(&dp->dl_recall_lru, &reaplist);
411         }
412         spin_unlock(&recall_lock);
413         while (!list_empty(&reaplist)) {
414                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
415                 list_del_init(&dp->dl_recall_lru);
416                 unhash_delegation(dp);
417         }
418         list_del(&clp->cl_idhash);
419         list_del(&clp->cl_strhash);
420         list_del(&clp->cl_lru);
421         while (!list_empty(&clp->cl_openowners)) {
422                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
423                 release_stateowner(sop);
424         }
425         put_nfs4_client(clp);
426 }
427
428 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
429 {
430         struct nfs4_client *clp;
431
432         clp = alloc_client(name);
433         if (clp == NULL)
434                 return NULL;
435         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
436         atomic_set(&clp->cl_count, 1);
437         atomic_set(&clp->cl_callback.cb_set, 0);
438         INIT_LIST_HEAD(&clp->cl_idhash);
439         INIT_LIST_HEAD(&clp->cl_strhash);
440         INIT_LIST_HEAD(&clp->cl_openowners);
441         INIT_LIST_HEAD(&clp->cl_delegations);
442         INIT_LIST_HEAD(&clp->cl_lru);
443         return clp;
444 }
445
446 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
447 {
448         memcpy(target->cl_verifier.data, source->data,
449                         sizeof(target->cl_verifier.data));
450 }
451
452 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
453 {
454         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
455         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
456 }
457
458 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
459 {
460         target->cr_uid = source->cr_uid;
461         target->cr_gid = source->cr_gid;
462         target->cr_group_info = source->cr_group_info;
463         get_group_info(target->cr_group_info);
464 }
465
466 static int same_name(const char *n1, const char *n2)
467 {
468         return 0 == memcmp(n1, n2, HEXDIR_LEN);
469 }
470
471 static int
472 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
473 {
474         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
475 }
476
477 static int
478 same_clid(clientid_t *cl1, clientid_t *cl2)
479 {
480         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
481 }
482
483 /* XXX what about NGROUP */
484 static int
485 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
486 {
487         return cr1->cr_uid == cr2->cr_uid;
488 }
489
490 static void gen_clid(struct nfs4_client *clp)
491 {
492         static u32 current_clientid = 1;
493
494         clp->cl_clientid.cl_boot = boot_time;
495         clp->cl_clientid.cl_id = current_clientid++; 
496 }
497
498 static void gen_confirm(struct nfs4_client *clp)
499 {
500         static u32 i;
501         u32 *p;
502
503         p = (u32 *)clp->cl_confirm.data;
504         *p++ = get_seconds();
505         *p++ = i++;
506 }
507
508 static int check_name(struct xdr_netobj name)
509 {
510         if (name.len == 0) 
511                 return 0;
512         if (name.len > NFS4_OPAQUE_LIMIT) {
513                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
514                 return 0;
515         }
516         return 1;
517 }
518
519 static void
520 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
521 {
522         unsigned int idhashval;
523
524         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
525         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
526         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
527         list_add_tail(&clp->cl_lru, &client_lru);
528         clp->cl_time = get_seconds();
529 }
530
531 static void
532 move_to_confirmed(struct nfs4_client *clp)
533 {
534         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
535         unsigned int strhashval;
536
537         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
538         list_del_init(&clp->cl_strhash);
539         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
540         strhashval = clientstr_hashval(clp->cl_recdir);
541         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
542         renew_client(clp);
543 }
544
545 static struct nfs4_client *
546 find_confirmed_client(clientid_t *clid)
547 {
548         struct nfs4_client *clp;
549         unsigned int idhashval = clientid_hashval(clid->cl_id);
550
551         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
552                 if (same_clid(&clp->cl_clientid, clid))
553                         return clp;
554         }
555         return NULL;
556 }
557
558 static struct nfs4_client *
559 find_unconfirmed_client(clientid_t *clid)
560 {
561         struct nfs4_client *clp;
562         unsigned int idhashval = clientid_hashval(clid->cl_id);
563
564         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
565                 if (same_clid(&clp->cl_clientid, clid))
566                         return clp;
567         }
568         return NULL;
569 }
570
571 static struct nfs4_client *
572 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
573 {
574         struct nfs4_client *clp;
575
576         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
577                 if (same_name(clp->cl_recdir, dname))
578                         return clp;
579         }
580         return NULL;
581 }
582
583 static struct nfs4_client *
584 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
585 {
586         struct nfs4_client *clp;
587
588         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
589                 if (same_name(clp->cl_recdir, dname))
590                         return clp;
591         }
592         return NULL;
593 }
594
595 /* a helper function for parse_callback */
596 static int
597 parse_octet(unsigned int *lenp, char **addrp)
598 {
599         unsigned int len = *lenp;
600         char *p = *addrp;
601         int n = -1;
602         char c;
603
604         for (;;) {
605                 if (!len)
606                         break;
607                 len--;
608                 c = *p++;
609                 if (c == '.')
610                         break;
611                 if ((c < '0') || (c > '9')) {
612                         n = -1;
613                         break;
614                 }
615                 if (n < 0)
616                         n = 0;
617                 n = (n * 10) + (c - '0');
618                 if (n > 255) {
619                         n = -1;
620                         break;
621                 }
622         }
623         *lenp = len;
624         *addrp = p;
625         return n;
626 }
627
628 /* parse and set the setclientid ipv4 callback address */
629 static int
630 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
631 {
632         int temp = 0;
633         u32 cbaddr = 0;
634         u16 cbport = 0;
635         u32 addrlen = addr_len;
636         char *addr = addr_val;
637         int i, shift;
638
639         /* ipaddress */
640         shift = 24;
641         for(i = 4; i > 0  ; i--) {
642                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
643                         return 0;
644                 }
645                 cbaddr |= (temp << shift);
646                 if (shift > 0)
647                 shift -= 8;
648         }
649         *cbaddrp = cbaddr;
650
651         /* port */
652         shift = 8;
653         for(i = 2; i > 0  ; i--) {
654                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
655                         return 0;
656                 }
657                 cbport |= (temp << shift);
658                 if (shift > 0)
659                         shift -= 8;
660         }
661         *cbportp = cbport;
662         return 1;
663 }
664
665 static void
666 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
667 {
668         struct nfs4_callback *cb = &clp->cl_callback;
669
670         /* Currently, we only support tcp for the callback channel */
671         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
672                 goto out_err;
673
674         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
675                          &cb->cb_addr, &cb->cb_port)))
676                 goto out_err;
677         cb->cb_prog = se->se_callback_prog;
678         cb->cb_ident = se->se_callback_ident;
679         return;
680 out_err:
681         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
682                 "will not receive delegations\n",
683                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
684
685         return;
686 }
687
688 __be32
689 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
690                   struct nfsd4_setclientid *setclid)
691 {
692         struct sockaddr_in      *sin = svc_addr_in(rqstp);
693         struct xdr_netobj       clname = { 
694                 .len = setclid->se_namelen,
695                 .data = setclid->se_name,
696         };
697         nfs4_verifier           clverifier = setclid->se_verf;
698         unsigned int            strhashval;
699         struct nfs4_client      *conf, *unconf, *new;
700         __be32                  status;
701         char                    *princ;
702         char                    dname[HEXDIR_LEN];
703         
704         if (!check_name(clname))
705                 return nfserr_inval;
706
707         status = nfs4_make_rec_clidname(dname, &clname);
708         if (status)
709                 return status;
710
711         /* 
712          * XXX The Duplicate Request Cache (DRC) has been checked (??)
713          * We get here on a DRC miss.
714          */
715
716         strhashval = clientstr_hashval(dname);
717
718         nfs4_lock_state();
719         conf = find_confirmed_client_by_str(dname, strhashval);
720         if (conf) {
721                 /* RFC 3530 14.2.33 CASE 0: */
722                 status = nfserr_clid_inuse;
723                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
724                                 || conf->cl_addr != sin->sin_addr.s_addr) {
725                         dprintk("NFSD: setclientid: string in use by clientat %pI4\n",
726                                 &conf->cl_addr);
727                         goto out;
728                 }
729         }
730         /*
731          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
732          * has a description of SETCLIENTID request processing consisting
733          * of 5 bullet points, labeled as CASE0 - CASE4 below.
734          */
735         unconf = find_unconfirmed_client_by_str(dname, strhashval);
736         status = nfserr_resource;
737         if (!conf) {
738                 /*
739                  * RFC 3530 14.2.33 CASE 4:
740                  * placed first, because it is the normal case
741                  */
742                 if (unconf)
743                         expire_client(unconf);
744                 new = create_client(clname, dname);
745                 if (new == NULL)
746                         goto out;
747                 gen_clid(new);
748         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
749                 /*
750                  * RFC 3530 14.2.33 CASE 1:
751                  * probable callback update
752                  */
753                 if (unconf) {
754                         /* Note this is removing unconfirmed {*x***},
755                          * which is stronger than RFC recommended {vxc**}.
756                          * This has the advantage that there is at most
757                          * one {*x***} in either list at any time.
758                          */
759                         expire_client(unconf);
760                 }
761                 new = create_client(clname, dname);
762                 if (new == NULL)
763                         goto out;
764                 copy_clid(new, conf);
765         } else if (!unconf) {
766                 /*
767                  * RFC 3530 14.2.33 CASE 2:
768                  * probable client reboot; state will be removed if
769                  * confirmed.
770                  */
771                 new = create_client(clname, dname);
772                 if (new == NULL)
773                         goto out;
774                 gen_clid(new);
775         } else {
776                 /*
777                  * RFC 3530 14.2.33 CASE 3:
778                  * probable client reboot; state will be removed if
779                  * confirmed.
780                  */
781                 expire_client(unconf);
782                 new = create_client(clname, dname);
783                 if (new == NULL)
784                         goto out;
785                 gen_clid(new);
786         }
787         copy_verf(new, &clverifier);
788         new->cl_addr = sin->sin_addr.s_addr;
789         new->cl_flavor = rqstp->rq_flavor;
790         princ = svc_gss_principal(rqstp);
791         if (princ) {
792                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
793                 if (new->cl_principal == NULL) {
794                         free_client(new);
795                         goto out;
796                 }
797         }
798         copy_cred(&new->cl_cred, &rqstp->rq_cred);
799         gen_confirm(new);
800         gen_callback(new, setclid);
801         add_to_unconfirmed(new, strhashval);
802         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
803         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
804         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
805         status = nfs_ok;
806 out:
807         nfs4_unlock_state();
808         return status;
809 }
810
811
812 /*
813  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
814  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
815  * bullets, labeled as CASE1 - CASE4 below.
816  */
817 __be32
818 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
819                          struct nfsd4_compound_state *cstate,
820                          struct nfsd4_setclientid_confirm *setclientid_confirm)
821 {
822         struct sockaddr_in *sin = svc_addr_in(rqstp);
823         struct nfs4_client *conf, *unconf;
824         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
825         clientid_t * clid = &setclientid_confirm->sc_clientid;
826         __be32 status;
827
828         if (STALE_CLIENTID(clid))
829                 return nfserr_stale_clientid;
830         /* 
831          * XXX The Duplicate Request Cache (DRC) has been checked (??)
832          * We get here on a DRC miss.
833          */
834
835         nfs4_lock_state();
836
837         conf = find_confirmed_client(clid);
838         unconf = find_unconfirmed_client(clid);
839
840         status = nfserr_clid_inuse;
841         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
842                 goto out;
843         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
844                 goto out;
845
846         /*
847          * section 14.2.34 of RFC 3530 has a description of
848          * SETCLIENTID_CONFIRM request processing consisting
849          * of 4 bullet points, labeled as CASE1 - CASE4 below.
850          */
851         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
852                 /*
853                  * RFC 3530 14.2.34 CASE 1:
854                  * callback update
855                  */
856                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
857                         status = nfserr_clid_inuse;
858                 else {
859                         /* XXX: We just turn off callbacks until we can handle
860                           * change request correctly. */
861                         atomic_set(&conf->cl_callback.cb_set, 0);
862                         gen_confirm(conf);
863                         nfsd4_remove_clid_dir(unconf);
864                         expire_client(unconf);
865                         status = nfs_ok;
866
867                 }
868         } else if (conf && !unconf) {
869                 /*
870                  * RFC 3530 14.2.34 CASE 2:
871                  * probable retransmitted request; play it safe and
872                  * do nothing.
873                  */
874                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
875                         status = nfserr_clid_inuse;
876                 else
877                         status = nfs_ok;
878         } else if (!conf && unconf
879                         && same_verf(&unconf->cl_confirm, &confirm)) {
880                 /*
881                  * RFC 3530 14.2.34 CASE 3:
882                  * Normal case; new or rebooted client:
883                  */
884                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
885                         status = nfserr_clid_inuse;
886                 } else {
887                         unsigned int hash =
888                                 clientstr_hashval(unconf->cl_recdir);
889                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
890                                                                         hash);
891                         if (conf) {
892                                 nfsd4_remove_clid_dir(conf);
893                                 expire_client(conf);
894                         }
895                         move_to_confirmed(unconf);
896                         conf = unconf;
897                         nfsd4_probe_callback(conf);
898                         status = nfs_ok;
899                 }
900         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
901             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
902                                                                 &confirm)))) {
903                 /*
904                  * RFC 3530 14.2.34 CASE 4:
905                  * Client probably hasn't noticed that we rebooted yet.
906                  */
907                 status = nfserr_stale_clientid;
908         } else {
909                 /* check that we have hit one of the cases...*/
910                 status = nfserr_clid_inuse;
911         }
912 out:
913         nfs4_unlock_state();
914         return status;
915 }
916
917 /* OPEN Share state helper functions */
918 static inline struct nfs4_file *
919 alloc_init_file(struct inode *ino)
920 {
921         struct nfs4_file *fp;
922         unsigned int hashval = file_hashval(ino);
923
924         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
925         if (fp) {
926                 kref_init(&fp->fi_ref);
927                 INIT_LIST_HEAD(&fp->fi_hash);
928                 INIT_LIST_HEAD(&fp->fi_stateids);
929                 INIT_LIST_HEAD(&fp->fi_delegations);
930                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
931                 fp->fi_inode = igrab(ino);
932                 fp->fi_id = current_fileid++;
933                 fp->fi_had_conflict = false;
934                 return fp;
935         }
936         return NULL;
937 }
938
939 static void
940 nfsd4_free_slab(struct kmem_cache **slab)
941 {
942         if (*slab == NULL)
943                 return;
944         kmem_cache_destroy(*slab);
945         *slab = NULL;
946 }
947
948 void
949 nfsd4_free_slabs(void)
950 {
951         nfsd4_free_slab(&stateowner_slab);
952         nfsd4_free_slab(&file_slab);
953         nfsd4_free_slab(&stateid_slab);
954         nfsd4_free_slab(&deleg_slab);
955 }
956
957 static int
958 nfsd4_init_slabs(void)
959 {
960         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
961                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
962         if (stateowner_slab == NULL)
963                 goto out_nomem;
964         file_slab = kmem_cache_create("nfsd4_files",
965                         sizeof(struct nfs4_file), 0, 0, NULL);
966         if (file_slab == NULL)
967                 goto out_nomem;
968         stateid_slab = kmem_cache_create("nfsd4_stateids",
969                         sizeof(struct nfs4_stateid), 0, 0, NULL);
970         if (stateid_slab == NULL)
971                 goto out_nomem;
972         deleg_slab = kmem_cache_create("nfsd4_delegations",
973                         sizeof(struct nfs4_delegation), 0, 0, NULL);
974         if (deleg_slab == NULL)
975                 goto out_nomem;
976         return 0;
977 out_nomem:
978         nfsd4_free_slabs();
979         dprintk("nfsd4: out of memory while initializing nfsv4\n");
980         return -ENOMEM;
981 }
982
983 void
984 nfs4_free_stateowner(struct kref *kref)
985 {
986         struct nfs4_stateowner *sop =
987                 container_of(kref, struct nfs4_stateowner, so_ref);
988         kfree(sop->so_owner.data);
989         kmem_cache_free(stateowner_slab, sop);
990 }
991
992 static inline struct nfs4_stateowner *
993 alloc_stateowner(struct xdr_netobj *owner)
994 {
995         struct nfs4_stateowner *sop;
996
997         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
998                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
999                         memcpy(sop->so_owner.data, owner->data, owner->len);
1000                         sop->so_owner.len = owner->len;
1001                         kref_init(&sop->so_ref);
1002                         return sop;
1003                 } 
1004                 kmem_cache_free(stateowner_slab, sop);
1005         }
1006         return NULL;
1007 }
1008
1009 static struct nfs4_stateowner *
1010 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1011         struct nfs4_stateowner *sop;
1012         struct nfs4_replay *rp;
1013         unsigned int idhashval;
1014
1015         if (!(sop = alloc_stateowner(&open->op_owner)))
1016                 return NULL;
1017         idhashval = ownerid_hashval(current_ownerid);
1018         INIT_LIST_HEAD(&sop->so_idhash);
1019         INIT_LIST_HEAD(&sop->so_strhash);
1020         INIT_LIST_HEAD(&sop->so_perclient);
1021         INIT_LIST_HEAD(&sop->so_stateids);
1022         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1023         INIT_LIST_HEAD(&sop->so_close_lru);
1024         sop->so_time = 0;
1025         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1026         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1027         list_add(&sop->so_perclient, &clp->cl_openowners);
1028         sop->so_is_open_owner = 1;
1029         sop->so_id = current_ownerid++;
1030         sop->so_client = clp;
1031         sop->so_seqid = open->op_seqid;
1032         sop->so_confirmed = 0;
1033         rp = &sop->so_replay;
1034         rp->rp_status = nfserr_serverfault;
1035         rp->rp_buflen = 0;
1036         rp->rp_buf = rp->rp_ibuf;
1037         return sop;
1038 }
1039
1040 static void
1041 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1042 {
1043         struct nfs4_stateowner *lock_sop;
1044
1045         while (!list_empty(&open_stp->st_lockowners)) {
1046                 lock_sop = list_entry(open_stp->st_lockowners.next,
1047                                 struct nfs4_stateowner, so_perstateid);
1048                 /* list_del(&open_stp->st_lockowners);  */
1049                 BUG_ON(lock_sop->so_is_open_owner);
1050                 release_stateowner(lock_sop);
1051         }
1052 }
1053
1054 static void
1055 unhash_stateowner(struct nfs4_stateowner *sop)
1056 {
1057         struct nfs4_stateid *stp;
1058
1059         list_del(&sop->so_idhash);
1060         list_del(&sop->so_strhash);
1061         if (sop->so_is_open_owner)
1062                 list_del(&sop->so_perclient);
1063         list_del(&sop->so_perstateid);
1064         while (!list_empty(&sop->so_stateids)) {
1065                 stp = list_entry(sop->so_stateids.next,
1066                         struct nfs4_stateid, st_perstateowner);
1067                 if (sop->so_is_open_owner)
1068                         release_stateid(stp, OPEN_STATE);
1069                 else
1070                         release_stateid(stp, LOCK_STATE);
1071         }
1072 }
1073
1074 static void
1075 release_stateowner(struct nfs4_stateowner *sop)
1076 {
1077         unhash_stateowner(sop);
1078         list_del(&sop->so_close_lru);
1079         nfs4_put_stateowner(sop);
1080 }
1081
1082 static inline void
1083 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1084         struct nfs4_stateowner *sop = open->op_stateowner;
1085         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1086
1087         INIT_LIST_HEAD(&stp->st_hash);
1088         INIT_LIST_HEAD(&stp->st_perstateowner);
1089         INIT_LIST_HEAD(&stp->st_lockowners);
1090         INIT_LIST_HEAD(&stp->st_perfile);
1091         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1092         list_add(&stp->st_perstateowner, &sop->so_stateids);
1093         list_add(&stp->st_perfile, &fp->fi_stateids);
1094         stp->st_stateowner = sop;
1095         get_nfs4_file(fp);
1096         stp->st_file = fp;
1097         stp->st_stateid.si_boot = boot_time;
1098         stp->st_stateid.si_stateownerid = sop->so_id;
1099         stp->st_stateid.si_fileid = fp->fi_id;
1100         stp->st_stateid.si_generation = 0;
1101         stp->st_access_bmap = 0;
1102         stp->st_deny_bmap = 0;
1103         __set_bit(open->op_share_access, &stp->st_access_bmap);
1104         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1105         stp->st_openstp = NULL;
1106 }
1107
1108 static void
1109 release_stateid(struct nfs4_stateid *stp, int flags)
1110 {
1111         struct file *filp = stp->st_vfs_file;
1112
1113         list_del(&stp->st_hash);
1114         list_del(&stp->st_perfile);
1115         list_del(&stp->st_perstateowner);
1116         if (flags & OPEN_STATE) {
1117                 release_stateid_lockowners(stp);
1118                 stp->st_vfs_file = NULL;
1119                 nfsd_close(filp);
1120         } else if (flags & LOCK_STATE)
1121                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1122         put_nfs4_file(stp->st_file);
1123         kmem_cache_free(stateid_slab, stp);
1124 }
1125
1126 static void
1127 move_to_close_lru(struct nfs4_stateowner *sop)
1128 {
1129         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1130
1131         list_move_tail(&sop->so_close_lru, &close_lru);
1132         sop->so_time = get_seconds();
1133 }
1134
1135 static int
1136 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1137                                                         clientid_t *clid)
1138 {
1139         return (sop->so_owner.len == owner->len) &&
1140                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1141                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1142 }
1143
1144 static struct nfs4_stateowner *
1145 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1146 {
1147         struct nfs4_stateowner *so = NULL;
1148
1149         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1150                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1151                         return so;
1152         }
1153         return NULL;
1154 }
1155
1156 /* search file_hashtbl[] for file */
1157 static struct nfs4_file *
1158 find_file(struct inode *ino)
1159 {
1160         unsigned int hashval = file_hashval(ino);
1161         struct nfs4_file *fp;
1162
1163         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1164                 if (fp->fi_inode == ino) {
1165                         get_nfs4_file(fp);
1166                         return fp;
1167                 }
1168         }
1169         return NULL;
1170 }
1171
1172 static inline int access_valid(u32 x)
1173 {
1174         if (x < NFS4_SHARE_ACCESS_READ)
1175                 return 0;
1176         if (x > NFS4_SHARE_ACCESS_BOTH)
1177                 return 0;
1178         return 1;
1179 }
1180
1181 static inline int deny_valid(u32 x)
1182 {
1183         /* Note: unlike access bits, deny bits may be zero. */
1184         return x <= NFS4_SHARE_DENY_BOTH;
1185 }
1186
1187 /*
1188  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1189  * st_{access,deny}_bmap field of the stateid, in order to track not
1190  * only what share bits are currently in force, but also what
1191  * combinations of share bits previous opens have used.  This allows us
1192  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1193  * return an error if the client attempt to downgrade to a combination
1194  * of share bits not explicable by closing some of its previous opens.
1195  *
1196  * XXX: This enforcement is actually incomplete, since we don't keep
1197  * track of access/deny bit combinations; so, e.g., we allow:
1198  *
1199  *      OPEN allow read, deny write
1200  *      OPEN allow both, deny none
1201  *      DOWNGRADE allow read, deny none
1202  *
1203  * which we should reject.
1204  */
1205 static void
1206 set_access(unsigned int *access, unsigned long bmap) {
1207         int i;
1208
1209         *access = 0;
1210         for (i = 1; i < 4; i++) {
1211                 if (test_bit(i, &bmap))
1212                         *access |= i;
1213         }
1214 }
1215
1216 static void
1217 set_deny(unsigned int *deny, unsigned long bmap) {
1218         int i;
1219
1220         *deny = 0;
1221         for (i = 0; i < 4; i++) {
1222                 if (test_bit(i, &bmap))
1223                         *deny |= i ;
1224         }
1225 }
1226
1227 static int
1228 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1229         unsigned int access, deny;
1230
1231         set_access(&access, stp->st_access_bmap);
1232         set_deny(&deny, stp->st_deny_bmap);
1233         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1234                 return 0;
1235         return 1;
1236 }
1237
1238 /*
1239  * Called to check deny when READ with all zero stateid or
1240  * WRITE with all zero or all one stateid
1241  */
1242 static __be32
1243 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1244 {
1245         struct inode *ino = current_fh->fh_dentry->d_inode;
1246         struct nfs4_file *fp;
1247         struct nfs4_stateid *stp;
1248         __be32 ret;
1249
1250         dprintk("NFSD: nfs4_share_conflict\n");
1251
1252         fp = find_file(ino);
1253         if (!fp)
1254                 return nfs_ok;
1255         ret = nfserr_locked;
1256         /* Search for conflicting share reservations */
1257         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1258                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1259                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1260                         goto out;
1261         }
1262         ret = nfs_ok;
1263 out:
1264         put_nfs4_file(fp);
1265         return ret;
1266 }
1267
1268 static inline void
1269 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1270 {
1271         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1272                 drop_file_write_access(filp);
1273                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1274         }
1275 }
1276
1277 /*
1278  * Recall a delegation
1279  */
1280 static int
1281 do_recall(void *__dp)
1282 {
1283         struct nfs4_delegation *dp = __dp;
1284
1285         dp->dl_file->fi_had_conflict = true;
1286         nfsd4_cb_recall(dp);
1287         return 0;
1288 }
1289
1290 /*
1291  * Spawn a thread to perform a recall on the delegation represented
1292  * by the lease (file_lock)
1293  *
1294  * Called from break_lease() with lock_kernel() held.
1295  * Note: we assume break_lease will only call this *once* for any given
1296  * lease.
1297  */
1298 static
1299 void nfsd_break_deleg_cb(struct file_lock *fl)
1300 {
1301         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1302         struct task_struct *t;
1303
1304         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1305         if (!dp)
1306                 return;
1307
1308         /* We're assuming the state code never drops its reference
1309          * without first removing the lease.  Since we're in this lease
1310          * callback (and since the lease code is serialized by the kernel
1311          * lock) we know the server hasn't removed the lease yet, we know
1312          * it's safe to take a reference: */
1313         atomic_inc(&dp->dl_count);
1314         atomic_inc(&dp->dl_client->cl_count);
1315
1316         spin_lock(&recall_lock);
1317         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1318         spin_unlock(&recall_lock);
1319
1320         /* only place dl_time is set. protected by lock_kernel*/
1321         dp->dl_time = get_seconds();
1322
1323         /*
1324          * We don't want the locks code to timeout the lease for us;
1325          * we'll remove it ourself if the delegation isn't returned
1326          * in time.
1327          */
1328         fl->fl_break_time = 0;
1329
1330         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1331         if (IS_ERR(t)) {
1332                 struct nfs4_client *clp = dp->dl_client;
1333
1334                 printk(KERN_INFO "NFSD: Callback thread failed for "
1335                         "for client (clientid %08x/%08x)\n",
1336                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1337                 put_nfs4_client(dp->dl_client);
1338                 nfs4_put_delegation(dp);
1339         }
1340 }
1341
1342 /*
1343  * The file_lock is being reapd.
1344  *
1345  * Called by locks_free_lock() with lock_kernel() held.
1346  */
1347 static
1348 void nfsd_release_deleg_cb(struct file_lock *fl)
1349 {
1350         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1351
1352         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1353
1354         if (!(fl->fl_flags & FL_LEASE) || !dp)
1355                 return;
1356         dp->dl_flock = NULL;
1357 }
1358
1359 /*
1360  * Set the delegation file_lock back pointer.
1361  *
1362  * Called from setlease() with lock_kernel() held.
1363  */
1364 static
1365 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1366 {
1367         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1368
1369         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1370         if (!dp)
1371                 return;
1372         dp->dl_flock = new;
1373 }
1374
1375 /*
1376  * Called from setlease() with lock_kernel() held
1377  */
1378 static
1379 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1380 {
1381         struct nfs4_delegation *onlistd =
1382                 (struct nfs4_delegation *)onlist->fl_owner;
1383         struct nfs4_delegation *tryd =
1384                 (struct nfs4_delegation *)try->fl_owner;
1385
1386         if (onlist->fl_lmops != try->fl_lmops)
1387                 return 0;
1388
1389         return onlistd->dl_client == tryd->dl_client;
1390 }
1391
1392
1393 static
1394 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1395 {
1396         if (arg & F_UNLCK)
1397                 return lease_modify(onlist, arg);
1398         else
1399                 return -EAGAIN;
1400 }
1401
1402 static struct lock_manager_operations nfsd_lease_mng_ops = {
1403         .fl_break = nfsd_break_deleg_cb,
1404         .fl_release_private = nfsd_release_deleg_cb,
1405         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1406         .fl_mylease = nfsd_same_client_deleg_cb,
1407         .fl_change = nfsd_change_deleg_cb,
1408 };
1409
1410
1411 __be32
1412 nfsd4_process_open1(struct nfsd4_open *open)
1413 {
1414         clientid_t *clientid = &open->op_clientid;
1415         struct nfs4_client *clp = NULL;
1416         unsigned int strhashval;
1417         struct nfs4_stateowner *sop = NULL;
1418
1419         if (!check_name(open->op_owner))
1420                 return nfserr_inval;
1421
1422         if (STALE_CLIENTID(&open->op_clientid))
1423                 return nfserr_stale_clientid;
1424
1425         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1426         sop = find_openstateowner_str(strhashval, open);
1427         open->op_stateowner = sop;
1428         if (!sop) {
1429                 /* Make sure the client's lease hasn't expired. */
1430                 clp = find_confirmed_client(clientid);
1431                 if (clp == NULL)
1432                         return nfserr_expired;
1433                 goto renew;
1434         }
1435         if (!sop->so_confirmed) {
1436                 /* Replace unconfirmed owners without checking for replay. */
1437                 clp = sop->so_client;
1438                 release_stateowner(sop);
1439                 open->op_stateowner = NULL;
1440                 goto renew;
1441         }
1442         if (open->op_seqid == sop->so_seqid - 1) {
1443                 if (sop->so_replay.rp_buflen)
1444                         return nfserr_replay_me;
1445                 /* The original OPEN failed so spectacularly
1446                  * that we don't even have replay data saved!
1447                  * Therefore, we have no choice but to continue
1448                  * processing this OPEN; presumably, we'll
1449                  * fail again for the same reason.
1450                  */
1451                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1452                 goto renew;
1453         }
1454         if (open->op_seqid != sop->so_seqid)
1455                 return nfserr_bad_seqid;
1456 renew:
1457         if (open->op_stateowner == NULL) {
1458                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1459                 if (sop == NULL)
1460                         return nfserr_resource;
1461                 open->op_stateowner = sop;
1462         }
1463         list_del_init(&sop->so_close_lru);
1464         renew_client(sop->so_client);
1465         return nfs_ok;
1466 }
1467
1468 static inline __be32
1469 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1470 {
1471         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1472                 return nfserr_openmode;
1473         else
1474                 return nfs_ok;
1475 }
1476
1477 static struct nfs4_delegation *
1478 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1479 {
1480         struct nfs4_delegation *dp;
1481
1482         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1483                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1484                         return dp;
1485         }
1486         return NULL;
1487 }
1488
1489 static __be32
1490 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1491                 struct nfs4_delegation **dp)
1492 {
1493         int flags;
1494         __be32 status = nfserr_bad_stateid;
1495
1496         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1497         if (*dp == NULL)
1498                 goto out;
1499         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1500                                                 RD_STATE : WR_STATE;
1501         status = nfs4_check_delegmode(*dp, flags);
1502         if (status)
1503                 *dp = NULL;
1504 out:
1505         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1506                 return nfs_ok;
1507         if (status)
1508                 return status;
1509         open->op_stateowner->so_confirmed = 1;
1510         return nfs_ok;
1511 }
1512
1513 static __be32
1514 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1515 {
1516         struct nfs4_stateid *local;
1517         __be32 status = nfserr_share_denied;
1518         struct nfs4_stateowner *sop = open->op_stateowner;
1519
1520         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1521                 /* ignore lock owners */
1522                 if (local->st_stateowner->so_is_open_owner == 0)
1523                         continue;
1524                 /* remember if we have seen this open owner */
1525                 if (local->st_stateowner == sop)
1526                         *stpp = local;
1527                 /* check for conflicting share reservations */
1528                 if (!test_share(local, open))
1529                         goto out;
1530         }
1531         status = 0;
1532 out:
1533         return status;
1534 }
1535
1536 static inline struct nfs4_stateid *
1537 nfs4_alloc_stateid(void)
1538 {
1539         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1540 }
1541
1542 static __be32
1543 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1544                 struct nfs4_delegation *dp,
1545                 struct svc_fh *cur_fh, int flags)
1546 {
1547         struct nfs4_stateid *stp;
1548
1549         stp = nfs4_alloc_stateid();
1550         if (stp == NULL)
1551                 return nfserr_resource;
1552
1553         if (dp) {
1554                 get_file(dp->dl_vfs_file);
1555                 stp->st_vfs_file = dp->dl_vfs_file;
1556         } else {
1557                 __be32 status;
1558                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1559                                 &stp->st_vfs_file);
1560                 if (status) {
1561                         if (status == nfserr_dropit)
1562                                 status = nfserr_jukebox;
1563                         kmem_cache_free(stateid_slab, stp);
1564                         return status;
1565                 }
1566         }
1567         *stpp = stp;
1568         return 0;
1569 }
1570
1571 static inline __be32
1572 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1573                 struct nfsd4_open *open)
1574 {
1575         struct iattr iattr = {
1576                 .ia_valid = ATTR_SIZE,
1577                 .ia_size = 0,
1578         };
1579         if (!open->op_truncate)
1580                 return 0;
1581         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1582                 return nfserr_inval;
1583         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1584 }
1585
1586 static __be32
1587 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1588 {
1589         struct file *filp = stp->st_vfs_file;
1590         struct inode *inode = filp->f_path.dentry->d_inode;
1591         unsigned int share_access, new_writer;
1592         __be32 status;
1593
1594         set_access(&share_access, stp->st_access_bmap);
1595         new_writer = (~share_access) & open->op_share_access
1596                         & NFS4_SHARE_ACCESS_WRITE;
1597
1598         if (new_writer) {
1599                 int err = get_write_access(inode);
1600                 if (err)
1601                         return nfserrno(err);
1602                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
1603                 if (err)
1604                         return nfserrno(err);
1605                 file_take_write(filp);
1606         }
1607         status = nfsd4_truncate(rqstp, cur_fh, open);
1608         if (status) {
1609                 if (new_writer)
1610                         put_write_access(inode);
1611                 return status;
1612         }
1613         /* remember the open */
1614         filp->f_mode |= open->op_share_access;
1615         __set_bit(open->op_share_access, &stp->st_access_bmap);
1616         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1617
1618         return nfs_ok;
1619 }
1620
1621
1622 static void
1623 nfs4_set_claim_prev(struct nfsd4_open *open)
1624 {
1625         open->op_stateowner->so_confirmed = 1;
1626         open->op_stateowner->so_client->cl_firststate = 1;
1627 }
1628
1629 /*
1630  * Attempt to hand out a delegation.
1631  */
1632 static void
1633 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1634 {
1635         struct nfs4_delegation *dp;
1636         struct nfs4_stateowner *sop = stp->st_stateowner;
1637         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1638         struct file_lock fl, *flp = &fl;
1639         int status, flag = 0;
1640
1641         flag = NFS4_OPEN_DELEGATE_NONE;
1642         open->op_recall = 0;
1643         switch (open->op_claim_type) {
1644                 case NFS4_OPEN_CLAIM_PREVIOUS:
1645                         if (!atomic_read(&cb->cb_set))
1646                                 open->op_recall = 1;
1647                         flag = open->op_delegate_type;
1648                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1649                                 goto out;
1650                         break;
1651                 case NFS4_OPEN_CLAIM_NULL:
1652                         /* Let's not give out any delegations till everyone's
1653                          * had the chance to reclaim theirs.... */
1654                         if (locks_in_grace())
1655                                 goto out;
1656                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1657                                 goto out;
1658                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1659                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1660                         else
1661                                 flag = NFS4_OPEN_DELEGATE_READ;
1662                         break;
1663                 default:
1664                         goto out;
1665         }
1666
1667         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1668         if (dp == NULL) {
1669                 flag = NFS4_OPEN_DELEGATE_NONE;
1670                 goto out;
1671         }
1672         locks_init_lock(&fl);
1673         fl.fl_lmops = &nfsd_lease_mng_ops;
1674         fl.fl_flags = FL_LEASE;
1675         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
1676         fl.fl_end = OFFSET_MAX;
1677         fl.fl_owner =  (fl_owner_t)dp;
1678         fl.fl_file = stp->st_vfs_file;
1679         fl.fl_pid = current->tgid;
1680
1681         /* vfs_setlease checks to see if delegation should be handed out.
1682          * the lock_manager callbacks fl_mylease and fl_change are used
1683          */
1684         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
1685                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1686                 unhash_delegation(dp);
1687                 flag = NFS4_OPEN_DELEGATE_NONE;
1688                 goto out;
1689         }
1690
1691         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1692
1693         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1694                      dp->dl_stateid.si_boot,
1695                      dp->dl_stateid.si_stateownerid,
1696                      dp->dl_stateid.si_fileid,
1697                      dp->dl_stateid.si_generation);
1698 out:
1699         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1700                         && flag == NFS4_OPEN_DELEGATE_NONE
1701                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1702                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1703         open->op_delegate_type = flag;
1704 }
1705
1706 /*
1707  * called with nfs4_lock_state() held.
1708  */
1709 __be32
1710 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1711 {
1712         struct nfs4_file *fp = NULL;
1713         struct inode *ino = current_fh->fh_dentry->d_inode;
1714         struct nfs4_stateid *stp = NULL;
1715         struct nfs4_delegation *dp = NULL;
1716         __be32 status;
1717
1718         status = nfserr_inval;
1719         if (!access_valid(open->op_share_access)
1720                         || !deny_valid(open->op_share_deny))
1721                 goto out;
1722         /*
1723          * Lookup file; if found, lookup stateid and check open request,
1724          * and check for delegations in the process of being recalled.
1725          * If not found, create the nfs4_file struct
1726          */
1727         fp = find_file(ino);
1728         if (fp) {
1729                 if ((status = nfs4_check_open(fp, open, &stp)))
1730                         goto out;
1731                 status = nfs4_check_deleg(fp, open, &dp);
1732                 if (status)
1733                         goto out;
1734         } else {
1735                 status = nfserr_bad_stateid;
1736                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1737                         goto out;
1738                 status = nfserr_resource;
1739                 fp = alloc_init_file(ino);
1740                 if (fp == NULL)
1741                         goto out;
1742         }
1743
1744         /*
1745          * OPEN the file, or upgrade an existing OPEN.
1746          * If truncate fails, the OPEN fails.
1747          */
1748         if (stp) {
1749                 /* Stateid was found, this is an OPEN upgrade */
1750                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1751                 if (status)
1752                         goto out;
1753                 update_stateid(&stp->st_stateid);
1754         } else {
1755                 /* Stateid was not found, this is a new OPEN */
1756                 int flags = 0;
1757                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1758                         flags |= NFSD_MAY_READ;
1759                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1760                         flags |= NFSD_MAY_WRITE;
1761                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1762                 if (status)
1763                         goto out;
1764                 init_stateid(stp, fp, open);
1765                 status = nfsd4_truncate(rqstp, current_fh, open);
1766                 if (status) {
1767                         release_stateid(stp, OPEN_STATE);
1768                         goto out;
1769                 }
1770         }
1771         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1772
1773         /*
1774         * Attempt to hand out a delegation. No error return, because the
1775         * OPEN succeeds even if we fail.
1776         */
1777         nfs4_open_delegation(current_fh, open, stp);
1778
1779         status = nfs_ok;
1780
1781         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1782                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1783                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1784 out:
1785         if (fp)
1786                 put_nfs4_file(fp);
1787         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1788                 nfs4_set_claim_prev(open);
1789         /*
1790         * To finish the open response, we just need to set the rflags.
1791         */
1792         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1793         if (!open->op_stateowner->so_confirmed)
1794                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1795
1796         return status;
1797 }
1798
1799 __be32
1800 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1801             clientid_t *clid)
1802 {
1803         struct nfs4_client *clp;
1804         __be32 status;
1805
1806         nfs4_lock_state();
1807         dprintk("process_renew(%08x/%08x): starting\n", 
1808                         clid->cl_boot, clid->cl_id);
1809         status = nfserr_stale_clientid;
1810         if (STALE_CLIENTID(clid))
1811                 goto out;
1812         clp = find_confirmed_client(clid);
1813         status = nfserr_expired;
1814         if (clp == NULL) {
1815                 /* We assume the client took too long to RENEW. */
1816                 dprintk("nfsd4_renew: clientid not found!\n");
1817                 goto out;
1818         }
1819         renew_client(clp);
1820         status = nfserr_cb_path_down;
1821         if (!list_empty(&clp->cl_delegations)
1822                         && !atomic_read(&clp->cl_callback.cb_set))
1823                 goto out;
1824         status = nfs_ok;
1825 out:
1826         nfs4_unlock_state();
1827         return status;
1828 }
1829
1830 struct lock_manager nfsd4_manager = {
1831 };
1832
1833 static void
1834 nfsd4_end_grace(void)
1835 {
1836         dprintk("NFSD: end of grace period\n");
1837         nfsd4_recdir_purge_old();
1838         locks_end_grace(&nfsd4_manager);
1839 }
1840
1841 static time_t
1842 nfs4_laundromat(void)
1843 {
1844         struct nfs4_client *clp;
1845         struct nfs4_stateowner *sop;
1846         struct nfs4_delegation *dp;
1847         struct list_head *pos, *next, reaplist;
1848         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1849         time_t t, clientid_val = NFSD_LEASE_TIME;
1850         time_t u, test_val = NFSD_LEASE_TIME;
1851
1852         nfs4_lock_state();
1853
1854         dprintk("NFSD: laundromat service - starting\n");
1855         if (locks_in_grace())
1856                 nfsd4_end_grace();
1857         list_for_each_safe(pos, next, &client_lru) {
1858                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1859                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1860                         t = clp->cl_time - cutoff;
1861                         if (clientid_val > t)
1862                                 clientid_val = t;
1863                         break;
1864                 }
1865                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1866                         clp->cl_clientid.cl_id);
1867                 nfsd4_remove_clid_dir(clp);
1868                 expire_client(clp);
1869         }
1870         INIT_LIST_HEAD(&reaplist);
1871         spin_lock(&recall_lock);
1872         list_for_each_safe(pos, next, &del_recall_lru) {
1873                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1874                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1875                         u = dp->dl_time - cutoff;
1876                         if (test_val > u)
1877                                 test_val = u;
1878                         break;
1879                 }
1880                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1881                                     dp, dp->dl_flock);
1882                 list_move(&dp->dl_recall_lru, &reaplist);
1883         }
1884         spin_unlock(&recall_lock);
1885         list_for_each_safe(pos, next, &reaplist) {
1886                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1887                 list_del_init(&dp->dl_recall_lru);
1888                 unhash_delegation(dp);
1889         }
1890         test_val = NFSD_LEASE_TIME;
1891         list_for_each_safe(pos, next, &close_lru) {
1892                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1893                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1894                         u = sop->so_time - cutoff;
1895                         if (test_val > u)
1896                                 test_val = u;
1897                         break;
1898                 }
1899                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1900                         sop->so_id);
1901                 release_stateowner(sop);
1902         }
1903         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1904                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1905         nfs4_unlock_state();
1906         return clientid_val;
1907 }
1908
1909 static struct workqueue_struct *laundry_wq;
1910 static void laundromat_main(struct work_struct *);
1911 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1912
1913 static void
1914 laundromat_main(struct work_struct *not_used)
1915 {
1916         time_t t;
1917
1918         t = nfs4_laundromat();
1919         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1920         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1921 }
1922
1923 static struct nfs4_stateowner *
1924 search_close_lru(u32 st_id, int flags)
1925 {
1926         struct nfs4_stateowner *local = NULL;
1927
1928         if (flags & CLOSE_STATE) {
1929                 list_for_each_entry(local, &close_lru, so_close_lru) {
1930                         if (local->so_id == st_id)
1931                                 return local;
1932                 }
1933         }
1934         return NULL;
1935 }
1936
1937 static inline int
1938 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1939 {
1940         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1941 }
1942
1943 static int
1944 STALE_STATEID(stateid_t *stateid)
1945 {
1946         if (stateid->si_boot == boot_time)
1947                 return 0;
1948         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1949                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1950                 stateid->si_generation);
1951         return 1;
1952 }
1953
1954 static inline int
1955 access_permit_read(unsigned long access_bmap)
1956 {
1957         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1958                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1959                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1960 }
1961
1962 static inline int
1963 access_permit_write(unsigned long access_bmap)
1964 {
1965         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1966                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1967 }
1968
1969 static
1970 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1971 {
1972         __be32 status = nfserr_openmode;
1973
1974         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1975                 goto out;
1976         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1977                 goto out;
1978         status = nfs_ok;
1979 out:
1980         return status;
1981 }
1982
1983 static inline __be32
1984 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
1985 {
1986         /* Trying to call delegreturn with a special stateid? Yuch: */
1987         if (!(flags & (RD_STATE | WR_STATE)))
1988                 return nfserr_bad_stateid;
1989         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
1990                 return nfs_ok;
1991         else if (locks_in_grace()) {
1992                 /* Answer in remaining cases depends on existance of
1993                  * conflicting state; so we must wait out the grace period. */
1994                 return nfserr_grace;
1995         } else if (flags & WR_STATE)
1996                 return nfs4_share_conflict(current_fh,
1997                                 NFS4_SHARE_DENY_WRITE);
1998         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
1999                 return nfs4_share_conflict(current_fh,
2000                                 NFS4_SHARE_DENY_READ);
2001 }
2002
2003 /*
2004  * Allow READ/WRITE during grace period on recovered state only for files
2005  * that are not able to provide mandatory locking.
2006  */
2007 static inline int
2008 io_during_grace_disallowed(struct inode *inode, int flags)
2009 {
2010         return locks_in_grace() && (flags & (RD_STATE | WR_STATE))
2011                 && mandatory_lock(inode);
2012 }
2013
2014 static int check_stateid_generation(stateid_t *in, stateid_t *ref)
2015 {
2016         /* If the client sends us a stateid from the future, it's buggy: */
2017         if (in->si_generation > ref->si_generation)
2018                 return nfserr_bad_stateid;
2019         /*
2020          * The following, however, can happen.  For example, if the
2021          * client sends an open and some IO at the same time, the open
2022          * may bump si_generation while the IO is still in flight.
2023          * Thanks to hard links and renames, the client never knows what
2024          * file an open will affect.  So it could avoid that situation
2025          * only by serializing all opens and IO from the same open
2026          * owner.  To recover from the old_stateid error, the client
2027          * will just have to retry the IO:
2028          */
2029         if (in->si_generation < ref->si_generation)
2030                 return nfserr_old_stateid;
2031         return nfs_ok;
2032 }
2033
2034 /*
2035 * Checks for stateid operations
2036 */
2037 __be32
2038 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2039 {
2040         struct nfs4_stateid *stp = NULL;
2041         struct nfs4_delegation *dp = NULL;
2042         stateid_t *stidp;
2043         struct inode *ino = current_fh->fh_dentry->d_inode;
2044         __be32 status;
2045
2046         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2047                 stateid->si_boot, stateid->si_stateownerid, 
2048                 stateid->si_fileid, stateid->si_generation); 
2049         if (filpp)
2050                 *filpp = NULL;
2051
2052         if (io_during_grace_disallowed(ino, flags))
2053                 return nfserr_grace;
2054
2055         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2056                 return check_special_stateids(current_fh, stateid, flags);
2057
2058         /* STALE STATEID */
2059         status = nfserr_stale_stateid;
2060         if (STALE_STATEID(stateid)) 
2061                 goto out;
2062
2063         /* BAD STATEID */
2064         status = nfserr_bad_stateid;
2065         if (!stateid->si_fileid) { /* delegation stateid */
2066                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2067                         dprintk("NFSD: delegation stateid not found\n");
2068                         goto out;
2069                 }
2070                 stidp = &dp->dl_stateid;
2071         } else { /* open or lock stateid */
2072                 if (!(stp = find_stateid(stateid, flags))) {
2073                         dprintk("NFSD: open or lock stateid not found\n");
2074                         goto out;
2075                 }
2076                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2077                         goto out;
2078                 if (!stp->st_stateowner->so_confirmed)
2079                         goto out;
2080                 stidp = &stp->st_stateid;
2081         }
2082         status = check_stateid_generation(stateid, stidp);
2083         if (status)
2084                 goto out;
2085         if (stp) {
2086                 if ((status = nfs4_check_openmode(stp,flags)))
2087                         goto out;
2088                 renew_client(stp->st_stateowner->so_client);
2089                 if (filpp)
2090                         *filpp = stp->st_vfs_file;
2091         } else {
2092                 if ((status = nfs4_check_delegmode(dp, flags)))
2093                         goto out;
2094                 renew_client(dp->dl_client);
2095                 if (flags & DELEG_RET)
2096                         unhash_delegation(dp);
2097                 if (filpp)
2098                         *filpp = dp->dl_vfs_file;
2099         }
2100         status = nfs_ok;
2101 out:
2102         return status;
2103 }
2104
2105 static inline int
2106 setlkflg (int type)
2107 {
2108         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2109                 RD_STATE : WR_STATE;
2110 }
2111
2112 /* 
2113  * Checks for sequence id mutating operations. 
2114  */
2115 static __be32
2116 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)
2117 {
2118         struct nfs4_stateid *stp;
2119         struct nfs4_stateowner *sop;
2120         __be32 status;
2121
2122         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2123                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2124                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2125                 stateid->si_generation);
2126
2127         *stpp = NULL;
2128         *sopp = NULL;
2129
2130         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2131                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2132                 return nfserr_bad_stateid;
2133         }
2134
2135         if (STALE_STATEID(stateid))
2136                 return nfserr_stale_stateid;
2137         /*
2138         * We return BAD_STATEID if filehandle doesn't match stateid, 
2139         * the confirmed flag is incorrecly set, or the generation 
2140         * number is incorrect.  
2141         */
2142         stp = find_stateid(stateid, flags);
2143         if (stp == NULL) {
2144                 /*
2145                  * Also, we should make sure this isn't just the result of
2146                  * a replayed close:
2147                  */
2148                 sop = search_close_lru(stateid->si_stateownerid, flags);
2149                 if (sop == NULL)
2150                         return nfserr_bad_stateid;
2151                 *sopp = sop;
2152                 goto check_replay;
2153         }
2154
2155         *stpp = stp;
2156         *sopp = sop = stp->st_stateowner;
2157
2158         if (lock) {
2159                 clientid_t *lockclid = &lock->v.new.clientid;
2160                 struct nfs4_client *clp = sop->so_client;
2161                 int lkflg = 0;
2162                 __be32 status;
2163
2164                 lkflg = setlkflg(lock->lk_type);
2165
2166                 if (lock->lk_is_new) {
2167                         if (!sop->so_is_open_owner)
2168                                 return nfserr_bad_stateid;
2169                         if (!same_clid(&clp->cl_clientid, lockclid))
2170                                return nfserr_bad_stateid;
2171                         /* stp is the open stateid */
2172                         status = nfs4_check_openmode(stp, lkflg);
2173                         if (status)
2174                                 return status;
2175                 } else {
2176                         /* stp is the lock stateid */
2177                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2178                         if (status)
2179                                 return status;
2180                }
2181         }
2182
2183         if (nfs4_check_fh(current_fh, stp)) {
2184                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2185                 return nfserr_bad_stateid;
2186         }
2187
2188         /*
2189         *  We now validate the seqid and stateid generation numbers.
2190         *  For the moment, we ignore the possibility of 
2191         *  generation number wraparound.
2192         */
2193         if (seqid != sop->so_seqid)
2194                 goto check_replay;
2195
2196         if (sop->so_confirmed && flags & CONFIRM) {
2197                 dprintk("NFSD: preprocess_seqid_op: expected"
2198                                 " unconfirmed stateowner!\n");
2199                 return nfserr_bad_stateid;
2200         }
2201         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2202                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2203                                 " confirmed yet!\n");
2204                 return nfserr_bad_stateid;
2205         }
2206         status = check_stateid_generation(stateid, &stp->st_stateid);
2207         if (status)
2208                 return status;
2209         renew_client(sop->so_client);
2210         return nfs_ok;
2211
2212 check_replay:
2213         if (seqid == sop->so_seqid - 1) {
2214                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2215                 /* indicate replay to calling function */
2216                 return nfserr_replay_me;
2217         }
2218         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2219                         sop->so_seqid, seqid);
2220         *sopp = NULL;
2221         return nfserr_bad_seqid;
2222 }
2223
2224 __be32
2225 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2226                    struct nfsd4_open_confirm *oc)
2227 {
2228         __be32 status;
2229         struct nfs4_stateowner *sop;
2230         struct nfs4_stateid *stp;
2231
2232         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2233                         (int)cstate->current_fh.fh_dentry->d_name.len,
2234                         cstate->current_fh.fh_dentry->d_name.name);
2235
2236         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2237         if (status)
2238                 return status;
2239
2240         nfs4_lock_state();
2241
2242         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2243                                         oc->oc_seqid, &oc->oc_req_stateid,
2244                                         CONFIRM | OPEN_STATE,
2245                                         &oc->oc_stateowner, &stp, NULL)))
2246                 goto out; 
2247
2248         sop = oc->oc_stateowner;
2249         sop->so_confirmed = 1;
2250         update_stateid(&stp->st_stateid);
2251         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2252         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
2253                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2254                          stp->st_stateid.si_boot,
2255                          stp->st_stateid.si_stateownerid,
2256                          stp->st_stateid.si_fileid,
2257                          stp->st_stateid.si_generation);
2258
2259         nfsd4_create_clid_dir(sop->so_client);
2260 out:
2261         if (oc->oc_stateowner) {
2262                 nfs4_get_stateowner(oc->oc_stateowner);
2263                 cstate->replay_owner = oc->oc_stateowner;
2264         }
2265         nfs4_unlock_state();
2266         return status;
2267 }
2268
2269
2270 /*
2271  * unset all bits in union bitmap (bmap) that
2272  * do not exist in share (from successful OPEN_DOWNGRADE)
2273  */
2274 static void
2275 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2276 {
2277         int i;
2278         for (i = 1; i < 4; i++) {
2279                 if ((i & access) != i)
2280                         __clear_bit(i, bmap);
2281         }
2282 }
2283
2284 static void
2285 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2286 {
2287         int i;
2288         for (i = 0; i < 4; i++) {
2289                 if ((i & deny) != i)
2290                         __clear_bit(i, bmap);
2291         }
2292 }
2293
2294 __be32
2295 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2296                      struct nfsd4_compound_state *cstate,
2297                      struct nfsd4_open_downgrade *od)
2298 {
2299         __be32 status;
2300         struct nfs4_stateid *stp;
2301         unsigned int share_access;
2302
2303         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
2304                         (int)cstate->current_fh.fh_dentry->d_name.len,
2305                         cstate->current_fh.fh_dentry->d_name.name);
2306
2307         if (!access_valid(od->od_share_access)
2308                         || !deny_valid(od->od_share_deny))
2309                 return nfserr_inval;
2310
2311         nfs4_lock_state();
2312         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2313                                         od->od_seqid,
2314                                         &od->od_stateid, 
2315                                         OPEN_STATE,
2316                                         &od->od_stateowner, &stp, NULL)))
2317                 goto out; 
2318
2319         status = nfserr_inval;
2320         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2321                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2322                         stp->st_access_bmap, od->od_share_access);
2323                 goto out;
2324         }
2325         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2326                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2327                         stp->st_deny_bmap, od->od_share_deny);
2328                 goto out;
2329         }
2330         set_access(&share_access, stp->st_access_bmap);
2331         nfs4_file_downgrade(stp->st_vfs_file,
2332                             share_access & ~od->od_share_access);
2333
2334         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2335         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2336
2337         update_stateid(&stp->st_stateid);
2338         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2339         status = nfs_ok;
2340 out:
2341         if (od->od_stateowner) {
2342                 nfs4_get_stateowner(od->od_stateowner);
2343                 cstate->replay_owner = od->od_stateowner;
2344         }
2345         nfs4_unlock_state();
2346         return status;
2347 }
2348
2349 /*
2350  * nfs4_unlock_state() called after encode
2351  */
2352 __be32
2353 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2354             struct nfsd4_close *close)
2355 {
2356         __be32 status;
2357         struct nfs4_stateid *stp;
2358
2359         dprintk("NFSD: nfsd4_close on file %.*s\n", 
2360                         (int)cstate->current_fh.fh_dentry->d_name.len,
2361                         cstate->current_fh.fh_dentry->d_name.name);
2362
2363         nfs4_lock_state();
2364         /* check close_lru for replay */
2365         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2366                                         close->cl_seqid,
2367                                         &close->cl_stateid, 
2368                                         OPEN_STATE | CLOSE_STATE,
2369                                         &close->cl_stateowner, &stp, NULL)))
2370                 goto out; 
2371         status = nfs_ok;
2372         update_stateid(&stp->st_stateid);
2373         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2374
2375         /* release_stateid() calls nfsd_close() if needed */
2376         release_stateid(stp, OPEN_STATE);
2377
2378         /* place unused nfs4_stateowners on so_close_lru list to be
2379          * released by the laundromat service after the lease period
2380          * to enable us to handle CLOSE replay
2381          */
2382         if (list_empty(&close->cl_stateowner->so_stateids))
2383                 move_to_close_lru(close->cl_stateowner);
2384 out:
2385         if (close->cl_stateowner) {
2386                 nfs4_get_stateowner(close->cl_stateowner);
2387                 cstate->replay_owner = close->cl_stateowner;
2388         }
2389         nfs4_unlock_state();
2390         return status;
2391 }
2392
2393 __be32
2394 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2395                   struct nfsd4_delegreturn *dr)
2396 {
2397         __be32 status;
2398
2399         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2400                 goto out;
2401
2402         nfs4_lock_state();
2403         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2404                                             &dr->dr_stateid, DELEG_RET, NULL);
2405         nfs4_unlock_state();
2406 out:
2407         return status;
2408 }
2409
2410
2411 /* 
2412  * Lock owner state (byte-range locks)
2413  */
2414 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2415 #define LOCK_HASH_BITS              8
2416 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2417 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2418
2419 #define lockownerid_hashval(id) \
2420         ((id) & LOCK_HASH_MASK)
2421
2422 static inline unsigned int
2423 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2424                 struct xdr_netobj *ownername)
2425 {
2426         return (file_hashval(inode) + cl_id
2427                         + opaque_hashval(ownername->data, ownername->len))
2428                 & LOCK_HASH_MASK;
2429 }
2430
2431 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2432 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2433 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2434
2435 static struct nfs4_stateid *
2436 find_stateid(stateid_t *stid, int flags)
2437 {
2438         struct nfs4_stateid *local = NULL;
2439         u32 st_id = stid->si_stateownerid;
2440         u32 f_id = stid->si_fileid;
2441         unsigned int hashval;
2442
2443         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2444         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2445                 hashval = stateid_hashval(st_id, f_id);
2446                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2447                         if ((local->st_stateid.si_stateownerid == st_id) &&
2448                             (local->st_stateid.si_fileid == f_id))
2449                                 return local;
2450                 }
2451         } 
2452         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2453                 hashval = stateid_hashval(st_id, f_id);
2454                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2455                         if ((local->st_stateid.si_stateownerid == st_id) &&
2456                             (local->st_stateid.si_fileid == f_id))
2457                                 return local;
2458                 }
2459         }
2460         return NULL;
2461 }
2462
2463 static struct nfs4_delegation *
2464 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2465 {
2466         struct nfs4_file *fp;
2467         struct nfs4_delegation *dl;
2468
2469         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2470                     stid->si_boot, stid->si_stateownerid,
2471                     stid->si_fileid, stid->si_generation);
2472
2473         fp = find_file(ino);
2474         if (!fp)
2475                 return NULL;
2476         dl = find_delegation_file(fp, stid);
2477         put_nfs4_file(fp);
2478         return dl;
2479 }
2480
2481 /*
2482  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2483  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2484  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2485  * locking, this prevents us from being completely protocol-compliant.  The
2486  * real solution to this problem is to start using unsigned file offsets in
2487  * the VFS, but this is a very deep change!
2488  */
2489 static inline void
2490 nfs4_transform_lock_offset(struct file_lock *lock)
2491 {
2492         if (lock->fl_start < 0)
2493                 lock->fl_start = OFFSET_MAX;
2494         if (lock->fl_end < 0)
2495                 lock->fl_end = OFFSET_MAX;
2496 }
2497
2498 /* Hack!: For now, we're defining this just so we can use a pointer to it
2499  * as a unique cookie to identify our (NFSv4's) posix locks. */
2500 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2501 };
2502
2503 static inline void
2504 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2505 {
2506         struct nfs4_stateowner *sop;
2507         unsigned int hval;
2508
2509         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2510                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2511                 hval = lockownerid_hashval(sop->so_id);
2512                 kref_get(&sop->so_ref);
2513                 deny->ld_sop = sop;
2514                 deny->ld_clientid = sop->so_client->cl_clientid;
2515         } else {
2516                 deny->ld_sop = NULL;
2517                 deny->ld_clientid.cl_boot = 0;
2518                 deny->ld_clientid.cl_id = 0;
2519         }
2520         deny->ld_start = fl->fl_start;
2521         deny->ld_length = ~(u64)0;
2522         if (fl->fl_end != ~(u64)0)
2523                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
2524         deny->ld_type = NFS4_READ_LT;
2525         if (fl->fl_type != F_RDLCK)
2526                 deny->ld_type = NFS4_WRITE_LT;
2527 }
2528
2529 static struct nfs4_stateowner *
2530 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2531                 struct xdr_netobj *owner)
2532 {
2533         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2534         struct nfs4_stateowner *op;
2535
2536         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2537                 if (same_owner_str(op, owner, clid))
2538                         return op;
2539         }
2540         return NULL;
2541 }
2542
2543 /*
2544  * Alloc a lock owner structure.
2545  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
2546  * occured. 
2547  *
2548  * strhashval = lock_ownerstr_hashval 
2549  */
2550
2551 static struct nfs4_stateowner *
2552 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2553         struct nfs4_stateowner *sop;
2554         struct nfs4_replay *rp;
2555         unsigned int idhashval;
2556
2557         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2558                 return NULL;
2559         idhashval = lockownerid_hashval(current_ownerid);
2560         INIT_LIST_HEAD(&sop->so_idhash);
2561         INIT_LIST_HEAD(&sop->so_strhash);
2562         INIT_LIST_HEAD(&sop->so_perclient);
2563         INIT_LIST_HEAD(&sop->so_stateids);
2564         INIT_LIST_HEAD(&sop->so_perstateid);
2565         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2566         sop->so_time = 0;
2567         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2568         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2569         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2570         sop->so_is_open_owner = 0;
2571         sop->so_id = current_ownerid++;
2572         sop->so_client = clp;
2573         /* It is the openowner seqid that will be incremented in encode in the
2574          * case of new lockowners; so increment the lock seqid manually: */
2575         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2576         sop->so_confirmed = 1;
2577         rp = &sop->so_replay;
2578         rp->rp_status = nfserr_serverfault;
2579         rp->rp_buflen = 0;
2580         rp->rp_buf = rp->rp_ibuf;
2581         return sop;
2582 }
2583
2584 static struct nfs4_stateid *
2585 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2586 {
2587         struct nfs4_stateid *stp;
2588         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2589
2590         stp = nfs4_alloc_stateid();
2591         if (stp == NULL)
2592                 goto out;
2593         INIT_LIST_HEAD(&stp->st_hash);
2594         INIT_LIST_HEAD(&stp->st_perfile);
2595         INIT_LIST_HEAD(&stp->st_perstateowner);
2596         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2597         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2598         list_add(&stp->st_perfile, &fp->fi_stateids);
2599         list_add(&stp->st_perstateowner, &sop->so_stateids);
2600         stp->st_stateowner = sop;
2601         get_nfs4_file(fp);
2602         stp->st_file = fp;
2603         stp->st_stateid.si_boot = boot_time;
2604         stp->st_stateid.si_stateownerid = sop->so_id;
2605         stp->st_stateid.si_fileid = fp->fi_id;
2606         stp->st_stateid.si_generation = 0;
2607         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2608         stp->st_access_bmap = open_stp->st_access_bmap;
2609         stp->st_deny_bmap = open_stp->st_deny_bmap;
2610         stp->st_openstp = open_stp;
2611
2612 out:
2613         return stp;
2614 }
2615
2616 static int
2617 check_lock_length(u64 offset, u64 length)
2618 {
2619         return ((length == 0)  || ((length != ~(u64)0) &&
2620              LOFF_OVERFLOW(offset, length)));
2621 }
2622
2623 /*
2624  *  LOCK operation 
2625  */
2626 __be32
2627 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2628            struct nfsd4_lock *lock)
2629 {
2630         struct nfs4_stateowner *open_sop = NULL;
2631         struct nfs4_stateowner *lock_sop = NULL;
2632         struct nfs4_stateid *lock_stp;
2633         struct file *filp;
2634         struct file_lock file_lock;
2635         struct file_lock conflock;
2636         __be32 status = 0;
2637         unsigned int strhashval;
2638         unsigned int cmd;
2639         int err;
2640
2641         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2642                 (long long) lock->lk_offset,
2643                 (long long) lock->lk_length);
2644
2645         if (check_lock_length(lock->lk_offset, lock->lk_length))
2646                  return nfserr_inval;
2647
2648         if ((status = fh_verify(rqstp, &cstate->current_fh,
2649                                 S_IFREG, NFSD_MAY_LOCK))) {
2650                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2651                 return status;
2652         }
2653
2654         nfs4_lock_state();
2655
2656         if (lock->lk_is_new) {
2657                 /*
2658                  * Client indicates that this is a new lockowner.
2659                  * Use open owner and open stateid to create lock owner and
2660                  * lock stateid.
2661                  */
2662                 struct nfs4_stateid *open_stp = NULL;
2663                 struct nfs4_file *fp;
2664                 
2665                 status = nfserr_stale_clientid;
2666                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2667                         goto out;
2668
2669                 /* validate and update open stateid and open seqid */
2670                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2671                                         lock->lk_new_open_seqid,
2672                                         &lock->lk_new_open_stateid,
2673                                         OPEN_STATE,
2674                                         &lock->lk_replay_owner, &open_stp,
2675                                         lock);
2676                 if (status)
2677                         goto out;
2678                 open_sop = lock->lk_replay_owner;
2679                 /* create lockowner and lock stateid */
2680                 fp = open_stp->st_file;
2681                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
2682                                 open_sop->so_client->cl_clientid.cl_id, 
2683                                 &lock->v.new.owner);
2684                 /* XXX: Do we need to check for duplicate stateowners on
2685                  * the same file, or should they just be allowed (and
2686                  * create new stateids)? */
2687                 status = nfserr_resource;
2688                 lock_sop = alloc_init_lock_stateowner(strhashval,
2689                                 open_sop->so_client, open_stp, lock);
2690                 if (lock_sop == NULL)
2691                         goto out;
2692                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2693                 if (lock_stp == NULL)
2694                         goto out;
2695         } else {
2696                 /* lock (lock owner + lock stateid) already exists */
2697                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2698                                        lock->lk_old_lock_seqid, 
2699                                        &lock->lk_old_lock_stateid, 
2700                                        LOCK_STATE,
2701                                        &lock->lk_replay_owner, &lock_stp, lock);
2702                 if (status)
2703                         goto out;
2704                 lock_sop = lock->lk_replay_owner;
2705         }
2706         /* lock->lk_replay_owner and lock_stp have been created or found */
2707         filp = lock_stp->st_vfs_file;
2708
2709         status = nfserr_grace;
2710         if (locks_in_grace() && !lock->lk_reclaim)
2711                 goto out;
2712         status = nfserr_no_grace;
2713         if (!locks_in_grace() && lock->lk_reclaim)
2714                 goto out;
2715
2716         locks_init_lock(&file_lock);
2717         switch (lock->lk_type) {
2718                 case NFS4_READ_LT:
2719                 case NFS4_READW_LT:
2720                         file_lock.fl_type = F_RDLCK;
2721                         cmd = F_SETLK;
2722                 break;
2723                 case NFS4_WRITE_LT:
2724                 case NFS4_WRITEW_LT:
2725                         file_lock.fl_type = F_WRLCK;
2726                         cmd = F_SETLK;
2727                 break;
2728                 default:
2729                         status = nfserr_inval;
2730                 goto out;
2731         }
2732         file_lock.fl_owner = (fl_owner_t)lock_sop;
2733         file_lock.fl_pid = current->tgid;
2734         file_lock.fl_file = filp;
2735         file_lock.fl_flags = FL_POSIX;
2736         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2737
2738         file_lock.fl_start = lock->lk_offset;
2739         if ((lock->lk_length == ~(u64)0) || 
2740                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2741                 file_lock.fl_end = ~(u64)0;
2742         else
2743                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2744         nfs4_transform_lock_offset(&file_lock);
2745
2746         /*
2747         * Try to lock the file in the VFS.
2748         * Note: locks.c uses the BKL to protect the inode's lock list.
2749         */
2750
2751         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2752         switch (-err) {
2753         case 0: /* success! */
2754                 update_stateid(&lock_stp->st_stateid);
2755                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
2756                                 sizeof(stateid_t));
2757                 status = 0;
2758                 break;
2759         case (EAGAIN):          /* conflock holds conflicting lock */
2760                 status = nfserr_denied;
2761                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2762                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2763                 break;
2764         case (EDEADLK):
2765                 status = nfserr_deadlock;
2766                 break;
2767         default:        
2768                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2769                 status = nfserr_resource;
2770                 break;
2771         }
2772 out:
2773         if (status && lock->lk_is_new && lock_sop)
2774                 release_stateowner(lock_sop);
2775         if (lock->lk_replay_owner) {
2776                 nfs4_get_stateowner(lock->lk_replay_owner);
2777                 cstate->replay_owner = lock->lk_replay_owner;
2778         }
2779         nfs4_unlock_state();
2780         return status;
2781 }
2782
2783 /*
2784  * LOCKT operation
2785  */
2786 __be32
2787 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2788             struct nfsd4_lockt *lockt)
2789 {
2790         struct inode *inode;
2791         struct file file;
2792         struct file_lock file_lock;
2793         int error;
2794         __be32 status;
2795
2796         if (locks_in_grace())
2797                 return nfserr_grace;
2798
2799         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2800                  return nfserr_inval;
2801
2802         lockt->lt_stateowner = NULL;
2803         nfs4_lock_state();
2804
2805         status = nfserr_stale_clientid;
2806         if (STALE_CLIENTID(&lockt->lt_clientid))
2807                 goto out;
2808
2809         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2810                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2811                 if (status == nfserr_symlink)
2812                         status = nfserr_inval;
2813                 goto out;
2814         }
2815
2816         inode = cstate->current_fh.fh_dentry->d_inode;
2817         locks_init_lock(&file_lock);
2818         switch (lockt->lt_type) {
2819                 case NFS4_READ_LT:
2820                 case NFS4_READW_LT:
2821                         file_lock.fl_type = F_RDLCK;
2822                 break;
2823                 case NFS4_WRITE_LT:
2824                 case NFS4_WRITEW_LT:
2825                         file_lock.fl_type = F_WRLCK;
2826                 break;
2827                 default:
2828                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2829                         status = nfserr_inval;
2830                 goto out;
2831         }
2832
2833         lockt->lt_stateowner = find_lockstateowner_str(inode,
2834                         &lockt->lt_clientid, &lockt->lt_owner);
2835         if (lockt->lt_stateowner)
2836                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2837         file_lock.fl_pid = current->tgid;
2838         file_lock.fl_flags = FL_POSIX;
2839         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2840
2841         file_lock.fl_start = lockt->lt_offset;
2842         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2843                 file_lock.fl_end = ~(u64)0;
2844         else
2845                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2846
2847         nfs4_transform_lock_offset(&file_lock);
2848
2849         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2850          * since LOCKT doesn't require an OPEN, and therefore a struct
2851          * file may not exist, pass vfs_test_lock a struct file with
2852          * only the dentry:inode set.
2853          */
2854         memset(&file, 0, sizeof (struct file));
2855         file.f_path.dentry = cstate->current_fh.fh_dentry;
2856
2857         status = nfs_ok;
2858         error = vfs_test_lock(&file, &file_lock);
2859         if (error) {
2860                 status = nfserrno(error);
2861                 goto out;
2862         }
2863         if (file_lock.fl_type != F_UNLCK) {
2864                 status = nfserr_denied;
2865                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2866         }
2867 out:
2868         nfs4_unlock_state();
2869         return status;
2870 }
2871
2872 __be32
2873 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2874             struct nfsd4_locku *locku)
2875 {
2876         struct nfs4_stateid *stp;
2877         struct file *filp = NULL;
2878         struct file_lock file_lock;
2879         __be32 status;
2880         int err;
2881                                                         
2882         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2883                 (long long) locku->lu_offset,
2884                 (long long) locku->lu_length);
2885
2886         if (check_lock_length(locku->lu_offset, locku->lu_length))
2887                  return nfserr_inval;
2888
2889         nfs4_lock_state();
2890                                                                                 
2891         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2892                                         locku->lu_seqid, 
2893                                         &locku->lu_stateid, 
2894                                         LOCK_STATE,
2895                                         &locku->lu_stateowner, &stp, NULL)))
2896                 goto out;
2897
2898         filp = stp->st_vfs_file;
2899         BUG_ON(!filp);
2900         locks_init_lock(&file_lock);
2901         file_lock.fl_type = F_UNLCK;
2902         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2903         file_lock.fl_pid = current->tgid;
2904         file_lock.fl_file = filp;
2905         file_lock.fl_flags = FL_POSIX; 
2906         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2907         file_lock.fl_start = locku->lu_offset;
2908
2909         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2910                 file_lock.fl_end = ~(u64)0;
2911         else
2912                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2913         nfs4_transform_lock_offset(&file_lock);
2914
2915         /*
2916         *  Try to unlock the file in the VFS.
2917         */
2918         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2919         if (err) {
2920                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2921                 goto out_nfserr;
2922         }
2923         /*
2924         * OK, unlock succeeded; the only thing left to do is update the stateid.
2925         */
2926         update_stateid(&stp->st_stateid);
2927         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2928
2929 out:
2930         if (locku->lu_stateowner) {
2931                 nfs4_get_stateowner(locku->lu_stateowner);
2932                 cstate->replay_owner = locku->lu_stateowner;
2933         }
2934         nfs4_unlock_state();
2935         return status;
2936
2937 out_nfserr:
2938         status = nfserrno(err);
2939         goto out;
2940 }
2941
2942 /*
2943  * returns
2944  *      1: locks held by lockowner
2945  *      0: no locks held by lockowner
2946  */
2947 static int
2948 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2949 {
2950         struct file_lock **flpp;
2951         struct inode *inode = filp->f_path.dentry->d_inode;
2952         int status = 0;
2953
2954         lock_kernel();
2955         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2956                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2957                         status = 1;
2958                         goto out;
2959                 }
2960         }
2961 out:
2962         unlock_kernel();
2963         return status;
2964 }
2965
2966 __be32
2967 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2968                         struct nfsd4_compound_state *cstate,
2969                         struct nfsd4_release_lockowner *rlockowner)
2970 {
2971         clientid_t *clid = &rlockowner->rl_clientid;
2972         struct nfs4_stateowner *sop;
2973         struct nfs4_stateid *stp;
2974         struct xdr_netobj *owner = &rlockowner->rl_owner;
2975         struct list_head matches;
2976         int i;
2977         __be32 status;
2978
2979         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2980                 clid->cl_boot, clid->cl_id);
2981
2982         /* XXX check for lease expiration */
2983
2984         status = nfserr_stale_clientid;
2985         if (STALE_CLIENTID(clid))
2986                 return status;
2987
2988         nfs4_lock_state();
2989
2990         status = nfserr_locks_held;
2991         /* XXX: we're doing a linear search through all the lockowners.
2992          * Yipes!  For now we'll just hope clients aren't really using
2993          * release_lockowner much, but eventually we have to fix these
2994          * data structures. */
2995         INIT_LIST_HEAD(&matches);
2996         for (i = 0; i < LOCK_HASH_SIZE; i++) {
2997                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
2998                         if (!same_owner_str(sop, owner, clid))
2999                                 continue;
3000                         list_for_each_entry(stp, &sop->so_stateids,
3001                                         st_perstateowner) {
3002                                 if (check_for_locks(stp->st_vfs_file, sop))
3003                                         goto out;
3004                                 /* Note: so_perclient unused for lockowners,
3005                                  * so it's OK to fool with here. */
3006                                 list_add(&sop->so_perclient, &matches);
3007                         }
3008                 }
3009         }
3010         /* Clients probably won't expect us to return with some (but not all)
3011          * of the lockowner state released; so don't release any until all
3012          * have been checked. */
3013         status = nfs_ok;
3014         while (!list_empty(&matches)) {
3015                 sop = list_entry(matches.next, struct nfs4_stateowner,
3016                                                                 so_perclient);
3017                 /* unhash_stateowner deletes so_perclient only
3018                  * for openowners. */
3019                 list_del(&sop->so_perclient);
3020                 release_stateowner(sop);
3021         }
3022 out:
3023         nfs4_unlock_state();
3024         return status;
3025 }
3026
3027 static inline struct nfs4_client_reclaim *
3028 alloc_reclaim(void)
3029 {
3030         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3031 }
3032
3033 int
3034 nfs4_has_reclaimed_state(const char *name)
3035 {
3036         unsigned int strhashval = clientstr_hashval(name);
3037         struct nfs4_client *clp;
3038
3039         clp = find_confirmed_client_by_str(name, strhashval);
3040         return clp ? 1 : 0;
3041 }
3042
3043 /*
3044  * failure => all reset bets are off, nfserr_no_grace...
3045  */
3046 int
3047 nfs4_client_to_reclaim(const char *name)
3048 {
3049         unsigned int strhashval;
3050         struct nfs4_client_reclaim *crp = NULL;
3051
3052         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3053         crp = alloc_reclaim();
3054         if (!crp)
3055                 return 0;
3056         strhashval = clientstr_hashval(name);
3057         INIT_LIST_HEAD(&crp->cr_strhash);
3058         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3059         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3060         reclaim_str_hashtbl_size++;
3061         return 1;
3062 }
3063
3064 static void
3065 nfs4_release_reclaim(void)
3066 {
3067         struct nfs4_client_reclaim *crp = NULL;
3068         int i;
3069
3070         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3071                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3072                         crp = list_entry(reclaim_str_hashtbl[i].next,
3073                                         struct nfs4_client_reclaim, cr_strhash);
3074                         list_del(&crp->cr_strhash);
3075                         kfree(crp);
3076                         reclaim_str_hashtbl_size--;
3077                 }
3078         }
3079         BUG_ON(reclaim_str_hashtbl_size);
3080 }
3081
3082 /*
3083  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3084 static struct nfs4_client_reclaim *
3085 nfs4_find_reclaim_client(clientid_t *clid)
3086 {
3087         unsigned int strhashval;
3088         struct nfs4_client *clp;
3089         struct nfs4_client_reclaim *crp = NULL;
3090
3091
3092         /* find clientid in conf_id_hashtbl */
3093         clp = find_confirmed_client(clid);
3094         if (clp == NULL)
3095                 return NULL;
3096
3097         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3098                             clp->cl_name.len, clp->cl_name.data,
3099                             clp->cl_recdir);
3100
3101         /* find clp->cl_name in reclaim_str_hashtbl */
3102         strhashval = clientstr_hashval(clp->cl_recdir);
3103         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3104                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3105                         return crp;
3106                 }
3107         }
3108         return NULL;
3109 }
3110
3111 /*
3112 * Called from OPEN. Look for clientid in reclaim list.
3113 */
3114 __be32
3115 nfs4_check_open_reclaim(clientid_t *clid)
3116 {
3117         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3118 }
3119
3120 /* initialization to perform at module load time: */
3121
3122 int
3123 nfs4_state_init(void)
3124 {
3125         int i, status;
3126
3127         status = nfsd4_init_slabs();
3128         if (status)
3129                 return status;
3130         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3131                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3132                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3133                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3134                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3135         }
3136         for (i = 0; i < FILE_HASH_SIZE; i++) {
3137                 INIT_LIST_HEAD(&file_hashtbl[i]);
3138         }
3139         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3140                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3141                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3142         }
3143         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3144                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3145                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3146         }
3147         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3148                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3149                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3150         }
3151         memset(&onestateid, ~0, sizeof(stateid_t));
3152         INIT_LIST_HEAD(&close_lru);
3153         INIT_LIST_HEAD(&client_lru);
3154         INIT_LIST_HEAD(&del_recall_lru);
3155         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3156                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3157         reclaim_str_hashtbl_size = 0;
3158         return 0;
3159 }
3160
3161 static void
3162 nfsd4_load_reboot_recovery_data(void)
3163 {
3164         int status;
3165
3166         nfs4_lock_state();
3167         nfsd4_init_recdir(user_recovery_dirname);
3168         status = nfsd4_recdir_load();
3169         nfs4_unlock_state();
3170         if (status)
3171                 printk("NFSD: Failure reading reboot recovery data\n");
3172 }
3173
3174 unsigned long
3175 get_nfs4_grace_period(void)
3176 {
3177         return max(user_lease_time, lease_time) * HZ;
3178 }
3179
3180 /*
3181  * Since the lifetime of a delegation isn't limited to that of an open, a
3182  * client may quite reasonably hang on to a delegation as long as it has
3183  * the inode cached.  This becomes an obvious problem the first time a
3184  * client's inode cache approaches the size of the server's total memory.
3185  *
3186  * For now we avoid this problem by imposing a hard limit on the number
3187  * of delegations, which varies according to the server's memory size.
3188  */
3189 static void
3190 set_max_delegations(void)
3191 {
3192         /*
3193          * Allow at most 4 delegations per megabyte of RAM.  Quick
3194          * estimates suggest that in the worst case (where every delegation
3195          * is for a different inode), a delegation could take about 1.5K,
3196          * giving a worst case usage of about 6% of memory.
3197          */
3198         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3199 }
3200
3201 /* initialization to perform when the nfsd service is started: */
3202
3203 static void
3204 __nfs4_state_start(void)
3205 {
3206         unsigned long grace_time;
3207
3208         boot_time = get_seconds();
3209         grace_time = get_nfs4_grace_period();
3210         lease_time = user_lease_time;
3211         locks_start_grace(&nfsd4_manager);
3212         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3213                grace_time/HZ);
3214         laundry_wq = create_singlethread_workqueue("nfsd4");
3215         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3216         set_max_delegations();
3217 }
3218
3219 void
3220 nfs4_state_start(void)
3221 {
3222         if (nfs4_init)
3223                 return;
3224         nfsd4_load_reboot_recovery_data();
3225         __nfs4_state_start();
3226         nfs4_init = 1;
3227         return;
3228 }
3229
3230 time_t
3231 nfs4_lease_time(void)
3232 {
3233         return lease_time;
3234 }
3235
3236 static void
3237 __nfs4_state_shutdown(void)
3238 {
3239         int i;
3240         struct nfs4_client *clp = NULL;
3241         struct nfs4_delegation *dp = NULL;
3242         struct list_head *pos, *next, reaplist;
3243
3244         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3245                 while (!list_empty(&conf_id_hashtbl[i])) {
3246                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3247                         expire_client(clp);
3248                 }
3249                 while (!list_empty(&unconf_str_hashtbl[i])) {
3250                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3251                         expire_client(clp);
3252                 }
3253         }
3254         INIT_LIST_HEAD(&reaplist);
3255         spin_lock(&recall_lock);
3256         list_for_each_safe(pos, next, &del_recall_lru) {
3257                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3258                 list_move(&dp->dl_recall_lru, &reaplist);
3259         }
3260         spin_unlock(&recall_lock);
3261         list_for_each_safe(pos, next, &reaplist) {
3262                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3263                 list_del_init(&dp->dl_recall_lru);
3264                 unhash_delegation(dp);
3265         }
3266
3267         nfsd4_shutdown_recdir();
3268         nfs4_init = 0;
3269 }
3270
3271 void
3272 nfs4_state_shutdown(void)
3273 {
3274         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3275         destroy_workqueue(laundry_wq);
3276         locks_end_grace(&nfsd4_manager);
3277         nfs4_lock_state();
3278         nfs4_release_reclaim();
3279         __nfs4_state_shutdown();
3280         nfs4_unlock_state();
3281 }
3282
3283 /*
3284  * user_recovery_dirname is protected by the nfsd_mutex since it's only
3285  * accessed when nfsd is starting.
3286  */
3287 static void
3288 nfs4_set_recdir(char *recdir)
3289 {
3290         strcpy(user_recovery_dirname, recdir);
3291 }
3292
3293 /*
3294  * Change the NFSv4 recovery directory to recdir.
3295  */
3296 int
3297 nfs4_reset_recoverydir(char *recdir)
3298 {
3299         int status;
3300         struct path path;
3301
3302         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
3303         if (status)
3304                 return status;
3305         status = -ENOTDIR;
3306         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
3307                 nfs4_set_recdir(recdir);
3308                 status = 0;
3309         }
3310         path_put(&path);
3311         return status;
3312 }
3313
3314 char *
3315 nfs4_recoverydir(void)
3316 {
3317         return user_recovery_dirname;
3318 }
3319
3320 /*
3321  * Called when leasetime is changed.
3322  *
3323  * The only way the protocol gives us to handle on-the-fly lease changes is to
3324  * simulate a reboot.  Instead of doing that, we just wait till the next time
3325  * we start to register any changes in lease time.  If the administrator
3326  * really wants to change the lease time *now*, they can go ahead and bring
3327  * nfsd down and then back up again after changing the lease time.
3328  *
3329  * user_lease_time is protected by nfsd_mutex since it's only really accessed
3330  * when nfsd is starting
3331  */
3332 void
3333 nfs4_reset_lease(time_t leasetime)
3334 {
3335         user_lease_time = leasetime;
3336 }