Pull vector-domain into release branch
[linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/mount.h>
45 #include <linux/workqueue.h>
46 #include <linux/smp_lock.h>
47 #include <linux/kthread.h>
48 #include <linux/nfs4.h>
49 #include <linux/nfsd/state.h>
50 #include <linux/nfsd/xdr4.h>
51 #include <linux/namei.h>
52 #include <linux/swap.h>
53 #include <linux/mutex.h>
54 #include <linux/lockd/bind.h>
55 #include <linux/module.h>
56
57 #define NFSDDBG_FACILITY                NFSDDBG_PROC
58
59 /* Globals */
60 static time_t lease_time = 90;     /* default lease time */
61 static time_t user_lease_time = 90;
62 static time_t boot_time;
63 static int in_grace = 1;
64 static u32 current_clientid = 1;
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 inline struct nfs4_client *
344 alloc_client(struct xdr_netobj name)
345 {
346         struct nfs4_client *clp;
347
348         if ((clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL))!= NULL) {
349                 if ((clp->cl_name.data = kmalloc(name.len, GFP_KERNEL)) != NULL) {
350                         memcpy(clp->cl_name.data, name.data, name.len);
351                         clp->cl_name.len = name.len;
352                 }
353                 else {
354                         kfree(clp);
355                         clp = NULL;
356                 }
357         }
358         return clp;
359 }
360
361 static inline void
362 free_client(struct nfs4_client *clp)
363 {
364         if (clp->cl_cred.cr_group_info)
365                 put_group_info(clp->cl_cred.cr_group_info);
366         kfree(clp->cl_name.data);
367         kfree(clp);
368 }
369
370 void
371 put_nfs4_client(struct nfs4_client *clp)
372 {
373         if (atomic_dec_and_test(&clp->cl_count))
374                 free_client(clp);
375 }
376
377 static void
378 shutdown_callback_client(struct nfs4_client *clp)
379 {
380         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
381
382         /* shutdown rpc client, ending any outstanding recall rpcs */
383         if (clnt) {
384                 clp->cl_callback.cb_client = NULL;
385                 rpc_shutdown_client(clnt);
386         }
387 }
388
389 static void
390 expire_client(struct nfs4_client *clp)
391 {
392         struct nfs4_stateowner *sop;
393         struct nfs4_delegation *dp;
394         struct list_head reaplist;
395
396         dprintk("NFSD: expire_client cl_count %d\n",
397                             atomic_read(&clp->cl_count));
398
399         shutdown_callback_client(clp);
400
401         INIT_LIST_HEAD(&reaplist);
402         spin_lock(&recall_lock);
403         while (!list_empty(&clp->cl_delegations)) {
404                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
405                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
406                                 dp->dl_flock);
407                 list_del_init(&dp->dl_perclnt);
408                 list_move(&dp->dl_recall_lru, &reaplist);
409         }
410         spin_unlock(&recall_lock);
411         while (!list_empty(&reaplist)) {
412                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
413                 list_del_init(&dp->dl_recall_lru);
414                 unhash_delegation(dp);
415         }
416         list_del(&clp->cl_idhash);
417         list_del(&clp->cl_strhash);
418         list_del(&clp->cl_lru);
419         while (!list_empty(&clp->cl_openowners)) {
420                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
421                 release_stateowner(sop);
422         }
423         put_nfs4_client(clp);
424 }
425
426 static struct nfs4_client *
427 create_client(struct xdr_netobj name, char *recdir) {
428         struct nfs4_client *clp;
429
430         if (!(clp = alloc_client(name)))
431                 goto out;
432         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
433         atomic_set(&clp->cl_count, 1);
434         atomic_set(&clp->cl_callback.cb_set, 0);
435         INIT_LIST_HEAD(&clp->cl_idhash);
436         INIT_LIST_HEAD(&clp->cl_strhash);
437         INIT_LIST_HEAD(&clp->cl_openowners);
438         INIT_LIST_HEAD(&clp->cl_delegations);
439         INIT_LIST_HEAD(&clp->cl_lru);
440 out:
441         return clp;
442 }
443
444 static void
445 copy_verf(struct nfs4_client *target, nfs4_verifier *source) {
446         memcpy(target->cl_verifier.data, source->data, sizeof(target->cl_verifier.data));
447 }
448
449 static void
450 copy_clid(struct nfs4_client *target, struct nfs4_client *source) {
451         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
452         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
453 }
454
455 static void
456 copy_cred(struct svc_cred *target, struct svc_cred *source) {
457
458         target->cr_uid = source->cr_uid;
459         target->cr_gid = source->cr_gid;
460         target->cr_group_info = source->cr_group_info;
461         get_group_info(target->cr_group_info);
462 }
463
464 static inline int
465 same_name(const char *n1, const char *n2) {
466         return 0 == memcmp(n1, n2, HEXDIR_LEN);
467 }
468
469 static int
470 cmp_verf(nfs4_verifier *v1, nfs4_verifier *v2) {
471         return(!memcmp(v1->data,v2->data,sizeof(v1->data)));
472 }
473
474 static int
475 cmp_clid(clientid_t * cl1, clientid_t * cl2) {
476         return((cl1->cl_boot == cl2->cl_boot) &&
477                 (cl1->cl_id == cl2->cl_id));
478 }
479
480 /* XXX what about NGROUP */
481 static int
482 cmp_creds(struct svc_cred *cr1, struct svc_cred *cr2){
483         return(cr1->cr_uid == cr2->cr_uid);
484
485 }
486
487 static void
488 gen_clid(struct nfs4_client *clp) {
489         clp->cl_clientid.cl_boot = boot_time;
490         clp->cl_clientid.cl_id = current_clientid++; 
491 }
492
493 static void
494 gen_confirm(struct nfs4_client *clp) {
495         struct timespec         tv;
496         u32 *                   p;
497
498         tv = CURRENT_TIME;
499         p = (u32 *)clp->cl_confirm.data;
500         *p++ = tv.tv_sec;
501         *p++ = tv.tv_nsec;
502 }
503
504 static int
505 check_name(struct xdr_netobj name) {
506
507         if (name.len == 0) 
508                 return 0;
509         if (name.len > NFS4_OPAQUE_LIMIT) {
510                 printk("NFSD: check_name: name too long(%d)!\n", name.len);
511                 return 0;
512         }
513         return 1;
514 }
515
516 static void
517 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
518 {
519         unsigned int idhashval;
520
521         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
522         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
523         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
524         list_add_tail(&clp->cl_lru, &client_lru);
525         clp->cl_time = get_seconds();
526 }
527
528 static void
529 move_to_confirmed(struct nfs4_client *clp)
530 {
531         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
532         unsigned int strhashval;
533
534         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
535         list_del_init(&clp->cl_strhash);
536         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
537         strhashval = clientstr_hashval(clp->cl_recdir);
538         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
539         renew_client(clp);
540 }
541
542 static struct nfs4_client *
543 find_confirmed_client(clientid_t *clid)
544 {
545         struct nfs4_client *clp;
546         unsigned int idhashval = clientid_hashval(clid->cl_id);
547
548         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
549                 if (cmp_clid(&clp->cl_clientid, clid))
550                         return clp;
551         }
552         return NULL;
553 }
554
555 static struct nfs4_client *
556 find_unconfirmed_client(clientid_t *clid)
557 {
558         struct nfs4_client *clp;
559         unsigned int idhashval = clientid_hashval(clid->cl_id);
560
561         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
562                 if (cmp_clid(&clp->cl_clientid, clid))
563                         return clp;
564         }
565         return NULL;
566 }
567
568 static struct nfs4_client *
569 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
570 {
571         struct nfs4_client *clp;
572
573         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
574                 if (same_name(clp->cl_recdir, dname))
575                         return clp;
576         }
577         return NULL;
578 }
579
580 static struct nfs4_client *
581 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
582 {
583         struct nfs4_client *clp;
584
585         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
586                 if (same_name(clp->cl_recdir, dname))
587                         return clp;
588         }
589         return NULL;
590 }
591
592 /* a helper function for parse_callback */
593 static int
594 parse_octet(unsigned int *lenp, char **addrp)
595 {
596         unsigned int len = *lenp;
597         char *p = *addrp;
598         int n = -1;
599         char c;
600
601         for (;;) {
602                 if (!len)
603                         break;
604                 len--;
605                 c = *p++;
606                 if (c == '.')
607                         break;
608                 if ((c < '0') || (c > '9')) {
609                         n = -1;
610                         break;
611                 }
612                 if (n < 0)
613                         n = 0;
614                 n = (n * 10) + (c - '0');
615                 if (n > 255) {
616                         n = -1;
617                         break;
618                 }
619         }
620         *lenp = len;
621         *addrp = p;
622         return n;
623 }
624
625 /* parse and set the setclientid ipv4 callback address */
626 static int
627 parse_ipv4(unsigned int addr_len, char *addr_val, unsigned int *cbaddrp, unsigned short *cbportp)
628 {
629         int temp = 0;
630         u32 cbaddr = 0;
631         u16 cbport = 0;
632         u32 addrlen = addr_len;
633         char *addr = addr_val;
634         int i, shift;
635
636         /* ipaddress */
637         shift = 24;
638         for(i = 4; i > 0  ; i--) {
639                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
640                         return 0;
641                 }
642                 cbaddr |= (temp << shift);
643                 if (shift > 0)
644                 shift -= 8;
645         }
646         *cbaddrp = cbaddr;
647
648         /* port */
649         shift = 8;
650         for(i = 2; i > 0  ; i--) {
651                 if ((temp = parse_octet(&addrlen, &addr)) < 0) {
652                         return 0;
653                 }
654                 cbport |= (temp << shift);
655                 if (shift > 0)
656                         shift -= 8;
657         }
658         *cbportp = cbport;
659         return 1;
660 }
661
662 static void
663 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se)
664 {
665         struct nfs4_callback *cb = &clp->cl_callback;
666
667         /* Currently, we only support tcp for the callback channel */
668         if ((se->se_callback_netid_len != 3) || memcmp((char *)se->se_callback_netid_val, "tcp", 3))
669                 goto out_err;
670
671         if ( !(parse_ipv4(se->se_callback_addr_len, se->se_callback_addr_val,
672                          &cb->cb_addr, &cb->cb_port)))
673                 goto out_err;
674         cb->cb_prog = se->se_callback_prog;
675         cb->cb_ident = se->se_callback_ident;
676         return;
677 out_err:
678         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
679                 "will not receive delegations\n",
680                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
681
682         return;
683 }
684
685 /*
686  * RFC 3010 has a complex implmentation description of processing a 
687  * SETCLIENTID request consisting of 5 bullets, labeled as 
688  * CASE0 - CASE4 below.
689  *
690  * NOTES:
691  *      callback information will be processed in a future patch
692  *
693  *      an unconfirmed record is added when:
694  *      NORMAL (part of CASE 4): there is no confirmed nor unconfirmed record.
695  *      CASE 1: confirmed record found with matching name, principal,
696  *              verifier, and clientid.
697  *      CASE 2: confirmed record found with matching name, principal,
698  *              and there is no unconfirmed record with matching
699  *              name and principal
700  *
701  *      an unconfirmed record is replaced when:
702  *      CASE 3: confirmed record found with matching name, principal,
703  *              and an unconfirmed record is found with matching 
704  *              name, principal, and with clientid and
705  *              confirm that does not match the confirmed record.
706  *      CASE 4: there is no confirmed record with matching name and 
707  *              principal. there is an unconfirmed record with 
708  *              matching name, principal.
709  *
710  *      an unconfirmed record is deleted when:
711  *      CASE 1: an unconfirmed record that matches input name, verifier,
712  *              and confirmed clientid.
713  *      CASE 4: any unconfirmed records with matching name and principal
714  *              that exist after an unconfirmed record has been replaced
715  *              as described above.
716  *
717  */
718 __be32
719 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
720                   struct nfsd4_setclientid *setclid)
721 {
722         struct sockaddr_in      *sin = svc_addr_in(rqstp);
723         struct xdr_netobj       clname = { 
724                 .len = setclid->se_namelen,
725                 .data = setclid->se_name,
726         };
727         nfs4_verifier           clverifier = setclid->se_verf;
728         unsigned int            strhashval;
729         struct nfs4_client      *conf, *unconf, *new;
730         __be32                  status;
731         char                    dname[HEXDIR_LEN];
732         
733         if (!check_name(clname))
734                 return nfserr_inval;
735
736         status = nfs4_make_rec_clidname(dname, &clname);
737         if (status)
738                 return status;
739
740         /* 
741          * XXX The Duplicate Request Cache (DRC) has been checked (??)
742          * We get here on a DRC miss.
743          */
744
745         strhashval = clientstr_hashval(dname);
746
747         nfs4_lock_state();
748         conf = find_confirmed_client_by_str(dname, strhashval);
749         if (conf) {
750                 /* 
751                  * CASE 0:
752                  * clname match, confirmed, different principal
753                  * or different ip_address
754                  */
755                 status = nfserr_clid_inuse;
756                 if (!cmp_creds(&conf->cl_cred, &rqstp->rq_cred)
757                                 || conf->cl_addr != sin->sin_addr.s_addr) {
758                         dprintk("NFSD: setclientid: string in use by client"
759                                 "at %u.%u.%u.%u\n", NIPQUAD(conf->cl_addr));
760                         goto out;
761                 }
762         }
763         unconf = find_unconfirmed_client_by_str(dname, strhashval);
764         status = nfserr_resource;
765         if (!conf) {
766                 /* 
767                  * CASE 4:
768                  * placed first, because it is the normal case.
769                  */
770                 if (unconf)
771                         expire_client(unconf);
772                 new = create_client(clname, dname);
773                 if (new == NULL)
774                         goto out;
775                 copy_verf(new, &clverifier);
776                 new->cl_addr = sin->sin_addr.s_addr;
777                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
778                 gen_clid(new);
779                 gen_confirm(new);
780                 gen_callback(new, setclid);
781                 add_to_unconfirmed(new, strhashval);
782         } else if (cmp_verf(&conf->cl_verifier, &clverifier)) {
783                 /*
784                  * CASE 1:
785                  * cl_name match, confirmed, principal match
786                  * verifier match: probable callback update
787                  *
788                  * remove any unconfirmed nfs4_client with 
789                  * matching cl_name, cl_verifier, and cl_clientid
790                  *
791                  * create and insert an unconfirmed nfs4_client with same 
792                  * cl_name, cl_verifier, and cl_clientid as existing 
793                  * nfs4_client,  but with the new callback info and a 
794                  * new cl_confirm
795                  */
796                 if (unconf) {
797                         /* Note this is removing unconfirmed {*x***},
798                          * which is stronger than RFC recommended {vxc**}.
799                          * This has the advantage that there is at most
800                          * one {*x***} in either list at any time.
801                          */
802                         expire_client(unconf);
803                 }
804                 new = create_client(clname, dname);
805                 if (new == NULL)
806                         goto out;
807                 copy_verf(new,&conf->cl_verifier);
808                 new->cl_addr = sin->sin_addr.s_addr;
809                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
810                 copy_clid(new, conf);
811                 gen_confirm(new);
812                 gen_callback(new, setclid);
813                 add_to_unconfirmed(new,strhashval);
814         } else if (!unconf) {
815                 /*
816                  * CASE 2:
817                  * clname match, confirmed, principal match
818                  * verfier does not match
819                  * no unconfirmed. create a new unconfirmed nfs4_client
820                  * using input clverifier, clname, and callback info
821                  * and generate a new cl_clientid and cl_confirm.
822                  */
823                 new = create_client(clname, dname);
824                 if (new == NULL)
825                         goto out;
826                 copy_verf(new,&clverifier);
827                 new->cl_addr = sin->sin_addr.s_addr;
828                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
829                 gen_clid(new);
830                 gen_confirm(new);
831                 gen_callback(new, setclid);
832                 add_to_unconfirmed(new, strhashval);
833         } else if (!cmp_verf(&conf->cl_confirm, &unconf->cl_confirm)) {
834                 /*      
835                  * CASE3:
836                  * confirmed found (name, principal match)
837                  * confirmed verifier does not match input clverifier
838                  *
839                  * unconfirmed found (name match)
840                  * confirmed->cl_confirm != unconfirmed->cl_confirm
841                  *
842                  * remove unconfirmed.
843                  *
844                  * create an unconfirmed nfs4_client 
845                  * with same cl_name as existing confirmed nfs4_client, 
846                  * but with new callback info, new cl_clientid,
847                  * new cl_verifier and a new cl_confirm
848                  */
849                 expire_client(unconf);
850                 new = create_client(clname, dname);
851                 if (new == NULL)
852                         goto out;
853                 copy_verf(new,&clverifier);
854                 new->cl_addr = sin->sin_addr.s_addr;
855                 copy_cred(&new->cl_cred,&rqstp->rq_cred);
856                 gen_clid(new);
857                 gen_confirm(new);
858                 gen_callback(new, setclid);
859                 add_to_unconfirmed(new, strhashval);
860         } else {
861                 /* No cases hit !!! */
862                 status = nfserr_inval;
863                 goto out;
864
865         }
866         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
867         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
868         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
869         status = nfs_ok;
870 out:
871         nfs4_unlock_state();
872         return status;
873 }
874
875
876 /*
877  * RFC 3010 has a complex implmentation description of processing a 
878  * SETCLIENTID_CONFIRM request consisting of 4 bullets describing
879  * processing on a DRC miss, labeled as CASE1 - CASE4 below.
880  *
881  * NOTE: callback information will be processed here in a future patch
882  */
883 __be32
884 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
885                          struct nfsd4_compound_state *cstate,
886                          struct nfsd4_setclientid_confirm *setclientid_confirm)
887 {
888         struct sockaddr_in *sin = svc_addr_in(rqstp);
889         struct nfs4_client *conf, *unconf;
890         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
891         clientid_t * clid = &setclientid_confirm->sc_clientid;
892         __be32 status;
893
894         if (STALE_CLIENTID(clid))
895                 return nfserr_stale_clientid;
896         /* 
897          * XXX The Duplicate Request Cache (DRC) has been checked (??)
898          * We get here on a DRC miss.
899          */
900
901         nfs4_lock_state();
902
903         conf = find_confirmed_client(clid);
904         unconf = find_unconfirmed_client(clid);
905
906         status = nfserr_clid_inuse;
907         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
908                 goto out;
909         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
910                 goto out;
911
912         if ((conf && unconf) && 
913             (cmp_verf(&unconf->cl_confirm, &confirm)) &&
914             (cmp_verf(&conf->cl_verifier, &unconf->cl_verifier)) &&
915             (same_name(conf->cl_recdir,unconf->cl_recdir))  &&
916             (!cmp_verf(&conf->cl_confirm, &unconf->cl_confirm))) {
917                 /* CASE 1:
918                 * unconf record that matches input clientid and input confirm.
919                 * conf record that matches input clientid.
920                 * conf and unconf records match names, verifiers
921                 */
922                 if (!cmp_creds(&conf->cl_cred, &unconf->cl_cred)) 
923                         status = nfserr_clid_inuse;
924                 else {
925                         /* XXX: We just turn off callbacks until we can handle
926                           * change request correctly. */
927                         atomic_set(&conf->cl_callback.cb_set, 0);
928                         gen_confirm(conf);
929                         nfsd4_remove_clid_dir(unconf);
930                         expire_client(unconf);
931                         status = nfs_ok;
932
933                 }
934         } else if ((conf && !unconf) ||
935             ((conf && unconf) && 
936              (!cmp_verf(&conf->cl_verifier, &unconf->cl_verifier) ||
937               !same_name(conf->cl_recdir, unconf->cl_recdir)))) {
938                 /* CASE 2:
939                  * conf record that matches input clientid.
940                  * if unconf record matches input clientid, then
941                  * unconf->cl_name or unconf->cl_verifier don't match the
942                  * conf record.
943                  */
944                 if (!cmp_creds(&conf->cl_cred,&rqstp->rq_cred))
945                         status = nfserr_clid_inuse;
946                 else
947                         status = nfs_ok;
948         } else if (!conf && unconf
949                         && cmp_verf(&unconf->cl_confirm, &confirm)) {
950                 /* CASE 3:
951                  * conf record not found.
952                  * unconf record found.
953                  * unconf->cl_confirm matches input confirm
954                  */
955                 if (!cmp_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
956                         status = nfserr_clid_inuse;
957                 } else {
958                         unsigned int hash =
959                                 clientstr_hashval(unconf->cl_recdir);
960                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
961                                                                         hash);
962                         if (conf) {
963                                 nfsd4_remove_clid_dir(conf);
964                                 expire_client(conf);
965                         }
966                         move_to_confirmed(unconf);
967                         conf = unconf;
968                         status = nfs_ok;
969                 }
970         } else if ((!conf || (conf && !cmp_verf(&conf->cl_confirm, &confirm)))
971             && (!unconf || (unconf && !cmp_verf(&unconf->cl_confirm,
972                                                                 &confirm)))) {
973                 /* CASE 4:
974                  * conf record not found, or if conf, conf->cl_confirm does not
975                  * match input confirm.
976                  * unconf record not found, or if unconf, unconf->cl_confirm
977                  * does not match input confirm.
978                  */
979                 status = nfserr_stale_clientid;
980         } else {
981                 /* check that we have hit one of the cases...*/
982                 status = nfserr_clid_inuse;
983         }
984 out:
985         if (!status)
986                 nfsd4_probe_callback(conf);
987         nfs4_unlock_state();
988         return status;
989 }
990
991 /* OPEN Share state helper functions */
992 static inline struct nfs4_file *
993 alloc_init_file(struct inode *ino)
994 {
995         struct nfs4_file *fp;
996         unsigned int hashval = file_hashval(ino);
997
998         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
999         if (fp) {
1000                 kref_init(&fp->fi_ref);
1001                 INIT_LIST_HEAD(&fp->fi_hash);
1002                 INIT_LIST_HEAD(&fp->fi_stateids);
1003                 INIT_LIST_HEAD(&fp->fi_delegations);
1004                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1005                 fp->fi_inode = igrab(ino);
1006                 fp->fi_id = current_fileid++;
1007                 fp->fi_had_conflict = false;
1008                 return fp;
1009         }
1010         return NULL;
1011 }
1012
1013 static void
1014 nfsd4_free_slab(struct kmem_cache **slab)
1015 {
1016         if (*slab == NULL)
1017                 return;
1018         kmem_cache_destroy(*slab);
1019         *slab = NULL;
1020 }
1021
1022 static void
1023 nfsd4_free_slabs(void)
1024 {
1025         nfsd4_free_slab(&stateowner_slab);
1026         nfsd4_free_slab(&file_slab);
1027         nfsd4_free_slab(&stateid_slab);
1028         nfsd4_free_slab(&deleg_slab);
1029 }
1030
1031 static int
1032 nfsd4_init_slabs(void)
1033 {
1034         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1035                         sizeof(struct nfs4_stateowner), 0, 0, NULL, NULL);
1036         if (stateowner_slab == NULL)
1037                 goto out_nomem;
1038         file_slab = kmem_cache_create("nfsd4_files",
1039                         sizeof(struct nfs4_file), 0, 0, NULL, NULL);
1040         if (file_slab == NULL)
1041                 goto out_nomem;
1042         stateid_slab = kmem_cache_create("nfsd4_stateids",
1043                         sizeof(struct nfs4_stateid), 0, 0, NULL, NULL);
1044         if (stateid_slab == NULL)
1045                 goto out_nomem;
1046         deleg_slab = kmem_cache_create("nfsd4_delegations",
1047                         sizeof(struct nfs4_delegation), 0, 0, NULL, NULL);
1048         if (deleg_slab == NULL)
1049                 goto out_nomem;
1050         return 0;
1051 out_nomem:
1052         nfsd4_free_slabs();
1053         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1054         return -ENOMEM;
1055 }
1056
1057 void
1058 nfs4_free_stateowner(struct kref *kref)
1059 {
1060         struct nfs4_stateowner *sop =
1061                 container_of(kref, struct nfs4_stateowner, so_ref);
1062         kfree(sop->so_owner.data);
1063         kmem_cache_free(stateowner_slab, sop);
1064 }
1065
1066 static inline struct nfs4_stateowner *
1067 alloc_stateowner(struct xdr_netobj *owner)
1068 {
1069         struct nfs4_stateowner *sop;
1070
1071         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1072                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1073                         memcpy(sop->so_owner.data, owner->data, owner->len);
1074                         sop->so_owner.len = owner->len;
1075                         kref_init(&sop->so_ref);
1076                         return sop;
1077                 } 
1078                 kmem_cache_free(stateowner_slab, sop);
1079         }
1080         return NULL;
1081 }
1082
1083 static struct nfs4_stateowner *
1084 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1085         struct nfs4_stateowner *sop;
1086         struct nfs4_replay *rp;
1087         unsigned int idhashval;
1088
1089         if (!(sop = alloc_stateowner(&open->op_owner)))
1090                 return NULL;
1091         idhashval = ownerid_hashval(current_ownerid);
1092         INIT_LIST_HEAD(&sop->so_idhash);
1093         INIT_LIST_HEAD(&sop->so_strhash);
1094         INIT_LIST_HEAD(&sop->so_perclient);
1095         INIT_LIST_HEAD(&sop->so_stateids);
1096         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1097         INIT_LIST_HEAD(&sop->so_close_lru);
1098         sop->so_time = 0;
1099         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1100         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1101         list_add(&sop->so_perclient, &clp->cl_openowners);
1102         sop->so_is_open_owner = 1;
1103         sop->so_id = current_ownerid++;
1104         sop->so_client = clp;
1105         sop->so_seqid = open->op_seqid;
1106         sop->so_confirmed = 0;
1107         rp = &sop->so_replay;
1108         rp->rp_status = nfserr_serverfault;
1109         rp->rp_buflen = 0;
1110         rp->rp_buf = rp->rp_ibuf;
1111         return sop;
1112 }
1113
1114 static void
1115 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1116 {
1117         struct nfs4_stateowner *lock_sop;
1118
1119         while (!list_empty(&open_stp->st_lockowners)) {
1120                 lock_sop = list_entry(open_stp->st_lockowners.next,
1121                                 struct nfs4_stateowner, so_perstateid);
1122                 /* list_del(&open_stp->st_lockowners);  */
1123                 BUG_ON(lock_sop->so_is_open_owner);
1124                 release_stateowner(lock_sop);
1125         }
1126 }
1127
1128 static void
1129 unhash_stateowner(struct nfs4_stateowner *sop)
1130 {
1131         struct nfs4_stateid *stp;
1132
1133         list_del(&sop->so_idhash);
1134         list_del(&sop->so_strhash);
1135         if (sop->so_is_open_owner)
1136                 list_del(&sop->so_perclient);
1137         list_del(&sop->so_perstateid);
1138         while (!list_empty(&sop->so_stateids)) {
1139                 stp = list_entry(sop->so_stateids.next,
1140                         struct nfs4_stateid, st_perstateowner);
1141                 if (sop->so_is_open_owner)
1142                         release_stateid(stp, OPEN_STATE);
1143                 else
1144                         release_stateid(stp, LOCK_STATE);
1145         }
1146 }
1147
1148 static void
1149 release_stateowner(struct nfs4_stateowner *sop)
1150 {
1151         unhash_stateowner(sop);
1152         list_del(&sop->so_close_lru);
1153         nfs4_put_stateowner(sop);
1154 }
1155
1156 static inline void
1157 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1158         struct nfs4_stateowner *sop = open->op_stateowner;
1159         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1160
1161         INIT_LIST_HEAD(&stp->st_hash);
1162         INIT_LIST_HEAD(&stp->st_perstateowner);
1163         INIT_LIST_HEAD(&stp->st_lockowners);
1164         INIT_LIST_HEAD(&stp->st_perfile);
1165         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1166         list_add(&stp->st_perstateowner, &sop->so_stateids);
1167         list_add(&stp->st_perfile, &fp->fi_stateids);
1168         stp->st_stateowner = sop;
1169         get_nfs4_file(fp);
1170         stp->st_file = fp;
1171         stp->st_stateid.si_boot = boot_time;
1172         stp->st_stateid.si_stateownerid = sop->so_id;
1173         stp->st_stateid.si_fileid = fp->fi_id;
1174         stp->st_stateid.si_generation = 0;
1175         stp->st_access_bmap = 0;
1176         stp->st_deny_bmap = 0;
1177         __set_bit(open->op_share_access, &stp->st_access_bmap);
1178         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1179         stp->st_openstp = NULL;
1180 }
1181
1182 static void
1183 release_stateid(struct nfs4_stateid *stp, int flags)
1184 {
1185         struct file *filp = stp->st_vfs_file;
1186
1187         list_del(&stp->st_hash);
1188         list_del(&stp->st_perfile);
1189         list_del(&stp->st_perstateowner);
1190         if (flags & OPEN_STATE) {
1191                 release_stateid_lockowners(stp);
1192                 stp->st_vfs_file = NULL;
1193                 nfsd_close(filp);
1194         } else if (flags & LOCK_STATE)
1195                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1196         put_nfs4_file(stp->st_file);
1197         kmem_cache_free(stateid_slab, stp);
1198 }
1199
1200 static void
1201 move_to_close_lru(struct nfs4_stateowner *sop)
1202 {
1203         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1204
1205         list_move_tail(&sop->so_close_lru, &close_lru);
1206         sop->so_time = get_seconds();
1207 }
1208
1209 static int
1210 cmp_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner, clientid_t *clid) {
1211         return ((sop->so_owner.len == owner->len) && 
1212          !memcmp(sop->so_owner.data, owner->data, owner->len) && 
1213           (sop->so_client->cl_clientid.cl_id == clid->cl_id));
1214 }
1215
1216 static struct nfs4_stateowner *
1217 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1218 {
1219         struct nfs4_stateowner *so = NULL;
1220
1221         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1222                 if (cmp_owner_str(so, &open->op_owner, &open->op_clientid))
1223                         return so;
1224         }
1225         return NULL;
1226 }
1227
1228 /* search file_hashtbl[] for file */
1229 static struct nfs4_file *
1230 find_file(struct inode *ino)
1231 {
1232         unsigned int hashval = file_hashval(ino);
1233         struct nfs4_file *fp;
1234
1235         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1236                 if (fp->fi_inode == ino) {
1237                         get_nfs4_file(fp);
1238                         return fp;
1239                 }
1240         }
1241         return NULL;
1242 }
1243
1244 static int access_valid(u32 x)
1245 {
1246         return (x > 0 && x < 4);
1247 }
1248
1249 static int deny_valid(u32 x)
1250 {
1251         return (x >= 0 && x < 5);
1252 }
1253
1254 static void
1255 set_access(unsigned int *access, unsigned long bmap) {
1256         int i;
1257
1258         *access = 0;
1259         for (i = 1; i < 4; i++) {
1260                 if (test_bit(i, &bmap))
1261                         *access |= i;
1262         }
1263 }
1264
1265 static void
1266 set_deny(unsigned int *deny, unsigned long bmap) {
1267         int i;
1268
1269         *deny = 0;
1270         for (i = 0; i < 4; i++) {
1271                 if (test_bit(i, &bmap))
1272                         *deny |= i ;
1273         }
1274 }
1275
1276 static int
1277 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1278         unsigned int access, deny;
1279
1280         set_access(&access, stp->st_access_bmap);
1281         set_deny(&deny, stp->st_deny_bmap);
1282         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1283                 return 0;
1284         return 1;
1285 }
1286
1287 /*
1288  * Called to check deny when READ with all zero stateid or
1289  * WRITE with all zero or all one stateid
1290  */
1291 static __be32
1292 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1293 {
1294         struct inode *ino = current_fh->fh_dentry->d_inode;
1295         struct nfs4_file *fp;
1296         struct nfs4_stateid *stp;
1297         __be32 ret;
1298
1299         dprintk("NFSD: nfs4_share_conflict\n");
1300
1301         fp = find_file(ino);
1302         if (!fp)
1303                 return nfs_ok;
1304         ret = nfserr_locked;
1305         /* Search for conflicting share reservations */
1306         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1307                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1308                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1309                         goto out;
1310         }
1311         ret = nfs_ok;
1312 out:
1313         put_nfs4_file(fp);
1314         return ret;
1315 }
1316
1317 static inline void
1318 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1319 {
1320         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1321                 put_write_access(filp->f_path.dentry->d_inode);
1322                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1323         }
1324 }
1325
1326 /*
1327  * Recall a delegation
1328  */
1329 static int
1330 do_recall(void *__dp)
1331 {
1332         struct nfs4_delegation *dp = __dp;
1333
1334         dp->dl_file->fi_had_conflict = true;
1335         nfsd4_cb_recall(dp);
1336         return 0;
1337 }
1338
1339 /*
1340  * Spawn a thread to perform a recall on the delegation represented
1341  * by the lease (file_lock)
1342  *
1343  * Called from break_lease() with lock_kernel() held.
1344  * Note: we assume break_lease will only call this *once* for any given
1345  * lease.
1346  */
1347 static
1348 void nfsd_break_deleg_cb(struct file_lock *fl)
1349 {
1350         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1351         struct task_struct *t;
1352
1353         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1354         if (!dp)
1355                 return;
1356
1357         /* We're assuming the state code never drops its reference
1358          * without first removing the lease.  Since we're in this lease
1359          * callback (and since the lease code is serialized by the kernel
1360          * lock) we know the server hasn't removed the lease yet, we know
1361          * it's safe to take a reference: */
1362         atomic_inc(&dp->dl_count);
1363
1364         spin_lock(&recall_lock);
1365         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1366         spin_unlock(&recall_lock);
1367
1368         /* only place dl_time is set. protected by lock_kernel*/
1369         dp->dl_time = get_seconds();
1370
1371         /* XXX need to merge NFSD_LEASE_TIME with fs/locks.c:lease_break_time */
1372         fl->fl_break_time = jiffies + NFSD_LEASE_TIME * HZ;
1373
1374         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1375         if (IS_ERR(t)) {
1376                 struct nfs4_client *clp = dp->dl_client;
1377
1378                 printk(KERN_INFO "NFSD: Callback thread failed for "
1379                         "for client (clientid %08x/%08x)\n",
1380                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1381                 nfs4_put_delegation(dp);
1382         }
1383 }
1384
1385 /*
1386  * The file_lock is being reapd.
1387  *
1388  * Called by locks_free_lock() with lock_kernel() held.
1389  */
1390 static
1391 void nfsd_release_deleg_cb(struct file_lock *fl)
1392 {
1393         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1394
1395         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1396
1397         if (!(fl->fl_flags & FL_LEASE) || !dp)
1398                 return;
1399         dp->dl_flock = NULL;
1400 }
1401
1402 /*
1403  * Set the delegation file_lock back pointer.
1404  *
1405  * Called from setlease() with lock_kernel() held.
1406  */
1407 static
1408 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1409 {
1410         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1411
1412         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1413         if (!dp)
1414                 return;
1415         dp->dl_flock = new;
1416 }
1417
1418 /*
1419  * Called from setlease() with lock_kernel() held
1420  */
1421 static
1422 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1423 {
1424         struct nfs4_delegation *onlistd =
1425                 (struct nfs4_delegation *)onlist->fl_owner;
1426         struct nfs4_delegation *tryd =
1427                 (struct nfs4_delegation *)try->fl_owner;
1428
1429         if (onlist->fl_lmops != try->fl_lmops)
1430                 return 0;
1431
1432         return onlistd->dl_client == tryd->dl_client;
1433 }
1434
1435
1436 static
1437 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1438 {
1439         if (arg & F_UNLCK)
1440                 return lease_modify(onlist, arg);
1441         else
1442                 return -EAGAIN;
1443 }
1444
1445 static struct lock_manager_operations nfsd_lease_mng_ops = {
1446         .fl_break = nfsd_break_deleg_cb,
1447         .fl_release_private = nfsd_release_deleg_cb,
1448         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1449         .fl_mylease = nfsd_same_client_deleg_cb,
1450         .fl_change = nfsd_change_deleg_cb,
1451 };
1452
1453
1454 __be32
1455 nfsd4_process_open1(struct nfsd4_open *open)
1456 {
1457         clientid_t *clientid = &open->op_clientid;
1458         struct nfs4_client *clp = NULL;
1459         unsigned int strhashval;
1460         struct nfs4_stateowner *sop = NULL;
1461
1462         if (!check_name(open->op_owner))
1463                 return nfserr_inval;
1464
1465         if (STALE_CLIENTID(&open->op_clientid))
1466                 return nfserr_stale_clientid;
1467
1468         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1469         sop = find_openstateowner_str(strhashval, open);
1470         open->op_stateowner = sop;
1471         if (!sop) {
1472                 /* Make sure the client's lease hasn't expired. */
1473                 clp = find_confirmed_client(clientid);
1474                 if (clp == NULL)
1475                         return nfserr_expired;
1476                 goto renew;
1477         }
1478         if (!sop->so_confirmed) {
1479                 /* Replace unconfirmed owners without checking for replay. */
1480                 clp = sop->so_client;
1481                 release_stateowner(sop);
1482                 open->op_stateowner = NULL;
1483                 goto renew;
1484         }
1485         if (open->op_seqid == sop->so_seqid - 1) {
1486                 if (sop->so_replay.rp_buflen)
1487                         return nfserr_replay_me;
1488                 /* The original OPEN failed so spectacularly
1489                  * that we don't even have replay data saved!
1490                  * Therefore, we have no choice but to continue
1491                  * processing this OPEN; presumably, we'll
1492                  * fail again for the same reason.
1493                  */
1494                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1495                 goto renew;
1496         }
1497         if (open->op_seqid != sop->so_seqid)
1498                 return nfserr_bad_seqid;
1499 renew:
1500         if (open->op_stateowner == NULL) {
1501                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1502                 if (sop == NULL)
1503                         return nfserr_resource;
1504                 open->op_stateowner = sop;
1505         }
1506         list_del_init(&sop->so_close_lru);
1507         renew_client(sop->so_client);
1508         return nfs_ok;
1509 }
1510
1511 static inline __be32
1512 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1513 {
1514         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1515                 return nfserr_openmode;
1516         else
1517                 return nfs_ok;
1518 }
1519
1520 static struct nfs4_delegation *
1521 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1522 {
1523         struct nfs4_delegation *dp;
1524
1525         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1526                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1527                         return dp;
1528         }
1529         return NULL;
1530 }
1531
1532 static __be32
1533 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1534                 struct nfs4_delegation **dp)
1535 {
1536         int flags;
1537         __be32 status = nfserr_bad_stateid;
1538
1539         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1540         if (*dp == NULL)
1541                 goto out;
1542         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1543                                                 RD_STATE : WR_STATE;
1544         status = nfs4_check_delegmode(*dp, flags);
1545         if (status)
1546                 *dp = NULL;
1547 out:
1548         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1549                 return nfs_ok;
1550         if (status)
1551                 return status;
1552         open->op_stateowner->so_confirmed = 1;
1553         return nfs_ok;
1554 }
1555
1556 static __be32
1557 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1558 {
1559         struct nfs4_stateid *local;
1560         __be32 status = nfserr_share_denied;
1561         struct nfs4_stateowner *sop = open->op_stateowner;
1562
1563         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1564                 /* ignore lock owners */
1565                 if (local->st_stateowner->so_is_open_owner == 0)
1566                         continue;
1567                 /* remember if we have seen this open owner */
1568                 if (local->st_stateowner == sop)
1569                         *stpp = local;
1570                 /* check for conflicting share reservations */
1571                 if (!test_share(local, open))
1572                         goto out;
1573         }
1574         status = 0;
1575 out:
1576         return status;
1577 }
1578
1579 static inline struct nfs4_stateid *
1580 nfs4_alloc_stateid(void)
1581 {
1582         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1583 }
1584
1585 static __be32
1586 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1587                 struct nfs4_delegation *dp,
1588                 struct svc_fh *cur_fh, int flags)
1589 {
1590         struct nfs4_stateid *stp;
1591
1592         stp = nfs4_alloc_stateid();
1593         if (stp == NULL)
1594                 return nfserr_resource;
1595
1596         if (dp) {
1597                 get_file(dp->dl_vfs_file);
1598                 stp->st_vfs_file = dp->dl_vfs_file;
1599         } else {
1600                 __be32 status;
1601                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1602                                 &stp->st_vfs_file);
1603                 if (status) {
1604                         if (status == nfserr_dropit)
1605                                 status = nfserr_jukebox;
1606                         kmem_cache_free(stateid_slab, stp);
1607                         return status;
1608                 }
1609         }
1610         *stpp = stp;
1611         return 0;
1612 }
1613
1614 static inline __be32
1615 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1616                 struct nfsd4_open *open)
1617 {
1618         struct iattr iattr = {
1619                 .ia_valid = ATTR_SIZE,
1620                 .ia_size = 0,
1621         };
1622         if (!open->op_truncate)
1623                 return 0;
1624         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1625                 return nfserr_inval;
1626         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1627 }
1628
1629 static __be32
1630 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1631 {
1632         struct file *filp = stp->st_vfs_file;
1633         struct inode *inode = filp->f_path.dentry->d_inode;
1634         unsigned int share_access, new_writer;
1635         __be32 status;
1636
1637         set_access(&share_access, stp->st_access_bmap);
1638         new_writer = (~share_access) & open->op_share_access
1639                         & NFS4_SHARE_ACCESS_WRITE;
1640
1641         if (new_writer) {
1642                 int err = get_write_access(inode);
1643                 if (err)
1644                         return nfserrno(err);
1645         }
1646         status = nfsd4_truncate(rqstp, cur_fh, open);
1647         if (status) {
1648                 if (new_writer)
1649                         put_write_access(inode);
1650                 return status;
1651         }
1652         /* remember the open */
1653         filp->f_mode |= open->op_share_access;
1654         set_bit(open->op_share_access, &stp->st_access_bmap);
1655         set_bit(open->op_share_deny, &stp->st_deny_bmap);
1656
1657         return nfs_ok;
1658 }
1659
1660
1661 static void
1662 nfs4_set_claim_prev(struct nfsd4_open *open)
1663 {
1664         open->op_stateowner->so_confirmed = 1;
1665         open->op_stateowner->so_client->cl_firststate = 1;
1666 }
1667
1668 /*
1669  * Attempt to hand out a delegation.
1670  */
1671 static void
1672 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1673 {
1674         struct nfs4_delegation *dp;
1675         struct nfs4_stateowner *sop = stp->st_stateowner;
1676         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1677         struct file_lock fl, *flp = &fl;
1678         int status, flag = 0;
1679
1680         flag = NFS4_OPEN_DELEGATE_NONE;
1681         open->op_recall = 0;
1682         switch (open->op_claim_type) {
1683                 case NFS4_OPEN_CLAIM_PREVIOUS:
1684                         if (!atomic_read(&cb->cb_set))
1685                                 open->op_recall = 1;
1686                         flag = open->op_delegate_type;
1687                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1688                                 goto out;
1689                         break;
1690                 case NFS4_OPEN_CLAIM_NULL:
1691                         /* Let's not give out any delegations till everyone's
1692                          * had the chance to reclaim theirs.... */
1693                         if (nfs4_in_grace())
1694                                 goto out;
1695                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1696                                 goto out;
1697                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1698                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1699                         else
1700                                 flag = NFS4_OPEN_DELEGATE_READ;
1701                         break;
1702                 default:
1703                         goto out;
1704         }
1705
1706         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1707         if (dp == NULL) {
1708                 flag = NFS4_OPEN_DELEGATE_NONE;
1709                 goto out;
1710         }
1711         locks_init_lock(&fl);
1712         fl.fl_lmops = &nfsd_lease_mng_ops;
1713         fl.fl_flags = FL_LEASE;
1714         fl.fl_end = OFFSET_MAX;
1715         fl.fl_owner =  (fl_owner_t)dp;
1716         fl.fl_file = stp->st_vfs_file;
1717         fl.fl_pid = current->tgid;
1718
1719         /* vfs_setlease checks to see if delegation should be handed out.
1720          * the lock_manager callbacks fl_mylease and fl_change are used
1721          */
1722         if ((status = vfs_setlease(stp->st_vfs_file,
1723                 flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK, &flp))) {
1724                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1725                 unhash_delegation(dp);
1726                 flag = NFS4_OPEN_DELEGATE_NONE;
1727                 goto out;
1728         }
1729
1730         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1731
1732         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1733                      dp->dl_stateid.si_boot,
1734                      dp->dl_stateid.si_stateownerid,
1735                      dp->dl_stateid.si_fileid,
1736                      dp->dl_stateid.si_generation);
1737 out:
1738         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1739                         && flag == NFS4_OPEN_DELEGATE_NONE
1740                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1741                 printk("NFSD: WARNING: refusing delegation reclaim\n");
1742         open->op_delegate_type = flag;
1743 }
1744
1745 /*
1746  * called with nfs4_lock_state() held.
1747  */
1748 __be32
1749 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1750 {
1751         struct nfs4_file *fp = NULL;
1752         struct inode *ino = current_fh->fh_dentry->d_inode;
1753         struct nfs4_stateid *stp = NULL;
1754         struct nfs4_delegation *dp = NULL;
1755         __be32 status;
1756
1757         status = nfserr_inval;
1758         if (!access_valid(open->op_share_access)
1759                         || !deny_valid(open->op_share_deny))
1760                 goto out;
1761         /*
1762          * Lookup file; if found, lookup stateid and check open request,
1763          * and check for delegations in the process of being recalled.
1764          * If not found, create the nfs4_file struct
1765          */
1766         fp = find_file(ino);
1767         if (fp) {
1768                 if ((status = nfs4_check_open(fp, open, &stp)))
1769                         goto out;
1770                 status = nfs4_check_deleg(fp, open, &dp);
1771                 if (status)
1772                         goto out;
1773         } else {
1774                 status = nfserr_bad_stateid;
1775                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1776                         goto out;
1777                 status = nfserr_resource;
1778                 fp = alloc_init_file(ino);
1779                 if (fp == NULL)
1780                         goto out;
1781         }
1782
1783         /*
1784          * OPEN the file, or upgrade an existing OPEN.
1785          * If truncate fails, the OPEN fails.
1786          */
1787         if (stp) {
1788                 /* Stateid was found, this is an OPEN upgrade */
1789                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1790                 if (status)
1791                         goto out;
1792                 update_stateid(&stp->st_stateid);
1793         } else {
1794                 /* Stateid was not found, this is a new OPEN */
1795                 int flags = 0;
1796                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1797                         flags |= MAY_READ;
1798                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1799                         flags |= MAY_WRITE;
1800                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1801                 if (status)
1802                         goto out;
1803                 init_stateid(stp, fp, open);
1804                 status = nfsd4_truncate(rqstp, current_fh, open);
1805                 if (status) {
1806                         release_stateid(stp, OPEN_STATE);
1807                         goto out;
1808                 }
1809         }
1810         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1811
1812         /*
1813         * Attempt to hand out a delegation. No error return, because the
1814         * OPEN succeeds even if we fail.
1815         */
1816         nfs4_open_delegation(current_fh, open, stp);
1817
1818         status = nfs_ok;
1819
1820         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1821                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1822                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1823 out:
1824         if (fp)
1825                 put_nfs4_file(fp);
1826         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1827                 nfs4_set_claim_prev(open);
1828         /*
1829         * To finish the open response, we just need to set the rflags.
1830         */
1831         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1832         if (!open->op_stateowner->so_confirmed)
1833                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1834
1835         return status;
1836 }
1837
1838 static struct workqueue_struct *laundry_wq;
1839 static void laundromat_main(struct work_struct *);
1840 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1841
1842 __be32
1843 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1844             clientid_t *clid)
1845 {
1846         struct nfs4_client *clp;
1847         __be32 status;
1848
1849         nfs4_lock_state();
1850         dprintk("process_renew(%08x/%08x): starting\n", 
1851                         clid->cl_boot, clid->cl_id);
1852         status = nfserr_stale_clientid;
1853         if (STALE_CLIENTID(clid))
1854                 goto out;
1855         clp = find_confirmed_client(clid);
1856         status = nfserr_expired;
1857         if (clp == NULL) {
1858                 /* We assume the client took too long to RENEW. */
1859                 dprintk("nfsd4_renew: clientid not found!\n");
1860                 goto out;
1861         }
1862         renew_client(clp);
1863         status = nfserr_cb_path_down;
1864         if (!list_empty(&clp->cl_delegations)
1865                         && !atomic_read(&clp->cl_callback.cb_set))
1866                 goto out;
1867         status = nfs_ok;
1868 out:
1869         nfs4_unlock_state();
1870         return status;
1871 }
1872
1873 static void
1874 end_grace(void)
1875 {
1876         dprintk("NFSD: end of grace period\n");
1877         nfsd4_recdir_purge_old();
1878         in_grace = 0;
1879 }
1880
1881 static time_t
1882 nfs4_laundromat(void)
1883 {
1884         struct nfs4_client *clp;
1885         struct nfs4_stateowner *sop;
1886         struct nfs4_delegation *dp;
1887         struct list_head *pos, *next, reaplist;
1888         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1889         time_t t, clientid_val = NFSD_LEASE_TIME;
1890         time_t u, test_val = NFSD_LEASE_TIME;
1891
1892         nfs4_lock_state();
1893
1894         dprintk("NFSD: laundromat service - starting\n");
1895         if (in_grace)
1896                 end_grace();
1897         list_for_each_safe(pos, next, &client_lru) {
1898                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1899                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1900                         t = clp->cl_time - cutoff;
1901                         if (clientid_val > t)
1902                                 clientid_val = t;
1903                         break;
1904                 }
1905                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1906                         clp->cl_clientid.cl_id);
1907                 nfsd4_remove_clid_dir(clp);
1908                 expire_client(clp);
1909         }
1910         INIT_LIST_HEAD(&reaplist);
1911         spin_lock(&recall_lock);
1912         list_for_each_safe(pos, next, &del_recall_lru) {
1913                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1914                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1915                         u = dp->dl_time - cutoff;
1916                         if (test_val > u)
1917                                 test_val = u;
1918                         break;
1919                 }
1920                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1921                                     dp, dp->dl_flock);
1922                 list_move(&dp->dl_recall_lru, &reaplist);
1923         }
1924         spin_unlock(&recall_lock);
1925         list_for_each_safe(pos, next, &reaplist) {
1926                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1927                 list_del_init(&dp->dl_recall_lru);
1928                 unhash_delegation(dp);
1929         }
1930         test_val = NFSD_LEASE_TIME;
1931         list_for_each_safe(pos, next, &close_lru) {
1932                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1933                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1934                         u = sop->so_time - cutoff;
1935                         if (test_val > u)
1936                                 test_val = u;
1937                         break;
1938                 }
1939                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1940                         sop->so_id);
1941                 release_stateowner(sop);
1942         }
1943         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1944                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1945         nfs4_unlock_state();
1946         return clientid_val;
1947 }
1948
1949 void
1950 laundromat_main(struct work_struct *not_used)
1951 {
1952         time_t t;
1953
1954         t = nfs4_laundromat();
1955         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1956         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1957 }
1958
1959 static struct nfs4_stateowner *
1960 search_close_lru(u32 st_id, int flags)
1961 {
1962         struct nfs4_stateowner *local = NULL;
1963
1964         if (flags & CLOSE_STATE) {
1965                 list_for_each_entry(local, &close_lru, so_close_lru) {
1966                         if (local->so_id == st_id)
1967                                 return local;
1968                 }
1969         }
1970         return NULL;
1971 }
1972
1973 static inline int
1974 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1975 {
1976         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1977 }
1978
1979 static int
1980 STALE_STATEID(stateid_t *stateid)
1981 {
1982         if (stateid->si_boot == boot_time)
1983                 return 0;
1984         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1985                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1986                 stateid->si_generation);
1987         return 1;
1988 }
1989
1990 static inline int
1991 access_permit_read(unsigned long access_bmap)
1992 {
1993         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1994                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1995                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1996 }
1997
1998 static inline int
1999 access_permit_write(unsigned long access_bmap)
2000 {
2001         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2002                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2003 }
2004
2005 static
2006 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2007 {
2008         __be32 status = nfserr_openmode;
2009
2010         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2011                 goto out;
2012         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2013                 goto out;
2014         status = nfs_ok;
2015 out:
2016         return status;
2017 }
2018
2019 static inline __be32
2020 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2021 {
2022         /* Trying to call delegreturn with a special stateid? Yuch: */
2023         if (!(flags & (RD_STATE | WR_STATE)))
2024                 return nfserr_bad_stateid;
2025         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
2026                 return nfs_ok;
2027         else if (nfs4_in_grace()) {
2028                 /* Answer in remaining cases depends on existance of
2029                  * conflicting state; so we must wait out the grace period. */
2030                 return nfserr_grace;
2031         } else if (flags & WR_STATE)
2032                 return nfs4_share_conflict(current_fh,
2033                                 NFS4_SHARE_DENY_WRITE);
2034         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2035                 return nfs4_share_conflict(current_fh,
2036                                 NFS4_SHARE_DENY_READ);
2037 }
2038
2039 /*
2040  * Allow READ/WRITE during grace period on recovered state only for files
2041  * that are not able to provide mandatory locking.
2042  */
2043 static inline int
2044 io_during_grace_disallowed(struct inode *inode, int flags)
2045 {
2046         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
2047                 && MANDATORY_LOCK(inode);
2048 }
2049
2050 /*
2051 * Checks for stateid operations
2052 */
2053 __be32
2054 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
2055 {
2056         struct nfs4_stateid *stp = NULL;
2057         struct nfs4_delegation *dp = NULL;
2058         stateid_t *stidp;
2059         struct inode *ino = current_fh->fh_dentry->d_inode;
2060         __be32 status;
2061
2062         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
2063                 stateid->si_boot, stateid->si_stateownerid, 
2064                 stateid->si_fileid, stateid->si_generation); 
2065         if (filpp)
2066                 *filpp = NULL;
2067
2068         if (io_during_grace_disallowed(ino, flags))
2069                 return nfserr_grace;
2070
2071         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2072                 return check_special_stateids(current_fh, stateid, flags);
2073
2074         /* STALE STATEID */
2075         status = nfserr_stale_stateid;
2076         if (STALE_STATEID(stateid)) 
2077                 goto out;
2078
2079         /* BAD STATEID */
2080         status = nfserr_bad_stateid;
2081         if (!stateid->si_fileid) { /* delegation stateid */
2082                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2083                         dprintk("NFSD: delegation stateid not found\n");
2084                         goto out;
2085                 }
2086                 stidp = &dp->dl_stateid;
2087         } else { /* open or lock stateid */
2088                 if (!(stp = find_stateid(stateid, flags))) {
2089                         dprintk("NFSD: open or lock stateid not found\n");
2090                         goto out;
2091                 }
2092                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2093                         goto out;
2094                 if (!stp->st_stateowner->so_confirmed)
2095                         goto out;
2096                 stidp = &stp->st_stateid;
2097         }
2098         if (stateid->si_generation > stidp->si_generation)
2099                 goto out;
2100
2101         /* OLD STATEID */
2102         status = nfserr_old_stateid;
2103         if (stateid->si_generation < stidp->si_generation)
2104                 goto out;
2105         if (stp) {
2106                 if ((status = nfs4_check_openmode(stp,flags)))
2107                         goto out;
2108                 renew_client(stp->st_stateowner->so_client);
2109                 if (filpp)
2110                         *filpp = stp->st_vfs_file;
2111         } else if (dp) {
2112                 if ((status = nfs4_check_delegmode(dp, flags)))
2113                         goto out;
2114                 renew_client(dp->dl_client);
2115                 if (flags & DELEG_RET)
2116                         unhash_delegation(dp);
2117                 if (filpp)
2118                         *filpp = dp->dl_vfs_file;
2119         }
2120         status = nfs_ok;
2121 out:
2122         return status;
2123 }
2124
2125 static inline int
2126 setlkflg (int type)
2127 {
2128         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2129                 RD_STATE : WR_STATE;
2130 }
2131
2132 /* 
2133  * Checks for sequence id mutating operations. 
2134  */
2135 static __be32
2136 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)
2137 {
2138         struct nfs4_stateid *stp;
2139         struct nfs4_stateowner *sop;
2140
2141         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2142                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2143                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2144                 stateid->si_generation);
2145
2146         *stpp = NULL;
2147         *sopp = NULL;
2148
2149         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2150                 printk("NFSD: preprocess_seqid_op: magic stateid!\n");
2151                 return nfserr_bad_stateid;
2152         }
2153
2154         if (STALE_STATEID(stateid))
2155                 return nfserr_stale_stateid;
2156         /*
2157         * We return BAD_STATEID if filehandle doesn't match stateid, 
2158         * the confirmed flag is incorrecly set, or the generation 
2159         * number is incorrect.  
2160         */
2161         stp = find_stateid(stateid, flags);
2162         if (stp == NULL) {
2163                 /*
2164                  * Also, we should make sure this isn't just the result of
2165                  * a replayed close:
2166                  */
2167                 sop = search_close_lru(stateid->si_stateownerid, flags);
2168                 if (sop == NULL)
2169                         return nfserr_bad_stateid;
2170                 *sopp = sop;
2171                 goto check_replay;
2172         }
2173
2174         if (lock) {
2175                 struct nfs4_stateowner *sop = stp->st_stateowner;
2176                 clientid_t *lockclid = &lock->v.new.clientid;
2177                 struct nfs4_client *clp = sop->so_client;
2178                 int lkflg = 0;
2179                 __be32 status;
2180
2181                 lkflg = setlkflg(lock->lk_type);
2182
2183                 if (lock->lk_is_new) {
2184                        if (!sop->so_is_open_owner)
2185                                return nfserr_bad_stateid;
2186                        if (!cmp_clid(&clp->cl_clientid, lockclid))
2187                                return nfserr_bad_stateid;
2188                        /* stp is the open stateid */
2189                        status = nfs4_check_openmode(stp, lkflg);
2190                        if (status)
2191                                return status;
2192                } else {
2193                        /* stp is the lock stateid */
2194                        status = nfs4_check_openmode(stp->st_openstp, lkflg);
2195                        if (status)
2196                                return status;
2197                }
2198
2199         }
2200
2201         if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp)) {
2202                 printk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2203                 return nfserr_bad_stateid;
2204         }
2205
2206         *stpp = stp;
2207         *sopp = sop = stp->st_stateowner;
2208
2209         /*
2210         *  We now validate the seqid and stateid generation numbers.
2211         *  For the moment, we ignore the possibility of 
2212         *  generation number wraparound.
2213         */
2214         if (seqid != sop->so_seqid)
2215                 goto check_replay;
2216
2217         if (sop->so_confirmed && flags & CONFIRM) {
2218                 printk("NFSD: preprocess_seqid_op: expected"
2219                                 " unconfirmed stateowner!\n");
2220                 return nfserr_bad_stateid;
2221         }
2222         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2223                 printk("NFSD: preprocess_seqid_op: stateowner not"
2224                                 " confirmed yet!\n");
2225                 return nfserr_bad_stateid;
2226         }
2227         if (stateid->si_generation > stp->st_stateid.si_generation) {
2228                 printk("NFSD: preprocess_seqid_op: future stateid?!\n");
2229                 return nfserr_bad_stateid;
2230         }
2231
2232         if (stateid->si_generation < stp->st_stateid.si_generation) {
2233                 printk("NFSD: preprocess_seqid_op: old stateid!\n");
2234                 return nfserr_old_stateid;
2235         }
2236         renew_client(sop->so_client);
2237         return nfs_ok;
2238
2239 check_replay:
2240         if (seqid == sop->so_seqid - 1) {
2241                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2242                 /* indicate replay to calling function */
2243                 return nfserr_replay_me;
2244         }
2245         printk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2246                         sop->so_seqid, seqid);
2247         *sopp = NULL;
2248         return nfserr_bad_seqid;
2249 }
2250
2251 __be32
2252 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2253                    struct nfsd4_open_confirm *oc)
2254 {
2255         __be32 status;
2256         struct nfs4_stateowner *sop;
2257         struct nfs4_stateid *stp;
2258
2259         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2260                         (int)cstate->current_fh.fh_dentry->d_name.len,
2261                         cstate->current_fh.fh_dentry->d_name.name);
2262
2263         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2264         if (status)
2265                 return status;
2266
2267         nfs4_lock_state();
2268
2269         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2270                                         oc->oc_seqid, &oc->oc_req_stateid,
2271                                         CHECK_FH | CONFIRM | OPEN_STATE,
2272                                         &oc->oc_stateowner, &stp, NULL)))
2273                 goto out; 
2274
2275         sop = oc->oc_stateowner;
2276         sop->so_confirmed = 1;
2277         update_stateid(&stp->st_stateid);
2278         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2279         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
2280                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2281                          stp->st_stateid.si_boot,
2282                          stp->st_stateid.si_stateownerid,
2283                          stp->st_stateid.si_fileid,
2284                          stp->st_stateid.si_generation);
2285
2286         nfsd4_create_clid_dir(sop->so_client);
2287 out:
2288         if (oc->oc_stateowner) {
2289                 nfs4_get_stateowner(oc->oc_stateowner);
2290                 cstate->replay_owner = oc->oc_stateowner;
2291         }
2292         nfs4_unlock_state();
2293         return status;
2294 }
2295
2296
2297 /*
2298  * unset all bits in union bitmap (bmap) that
2299  * do not exist in share (from successful OPEN_DOWNGRADE)
2300  */
2301 static void
2302 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2303 {
2304         int i;
2305         for (i = 1; i < 4; i++) {
2306                 if ((i & access) != i)
2307                         __clear_bit(i, bmap);
2308         }
2309 }
2310
2311 static void
2312 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2313 {
2314         int i;
2315         for (i = 0; i < 4; i++) {
2316                 if ((i & deny) != i)
2317                         __clear_bit(i, bmap);
2318         }
2319 }
2320
2321 __be32
2322 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2323                      struct nfsd4_compound_state *cstate,
2324                      struct nfsd4_open_downgrade *od)
2325 {
2326         __be32 status;
2327         struct nfs4_stateid *stp;
2328         unsigned int share_access;
2329
2330         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
2331                         (int)cstate->current_fh.fh_dentry->d_name.len,
2332                         cstate->current_fh.fh_dentry->d_name.name);
2333
2334         if (!access_valid(od->od_share_access)
2335                         || !deny_valid(od->od_share_deny))
2336                 return nfserr_inval;
2337
2338         nfs4_lock_state();
2339         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2340                                         od->od_seqid,
2341                                         &od->od_stateid, 
2342                                         CHECK_FH | OPEN_STATE, 
2343                                         &od->od_stateowner, &stp, NULL)))
2344                 goto out; 
2345
2346         status = nfserr_inval;
2347         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2348                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2349                         stp->st_access_bmap, od->od_share_access);
2350                 goto out;
2351         }
2352         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2353                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2354                         stp->st_deny_bmap, od->od_share_deny);
2355                 goto out;
2356         }
2357         set_access(&share_access, stp->st_access_bmap);
2358         nfs4_file_downgrade(stp->st_vfs_file,
2359                             share_access & ~od->od_share_access);
2360
2361         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2362         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2363
2364         update_stateid(&stp->st_stateid);
2365         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2366         status = nfs_ok;
2367 out:
2368         if (od->od_stateowner) {
2369                 nfs4_get_stateowner(od->od_stateowner);
2370                 cstate->replay_owner = od->od_stateowner;
2371         }
2372         nfs4_unlock_state();
2373         return status;
2374 }
2375
2376 /*
2377  * nfs4_unlock_state() called after encode
2378  */
2379 __be32
2380 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2381             struct nfsd4_close *close)
2382 {
2383         __be32 status;
2384         struct nfs4_stateid *stp;
2385
2386         dprintk("NFSD: nfsd4_close on file %.*s\n", 
2387                         (int)cstate->current_fh.fh_dentry->d_name.len,
2388                         cstate->current_fh.fh_dentry->d_name.name);
2389
2390         nfs4_lock_state();
2391         /* check close_lru for replay */
2392         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2393                                         close->cl_seqid,
2394                                         &close->cl_stateid, 
2395                                         CHECK_FH | OPEN_STATE | CLOSE_STATE,
2396                                         &close->cl_stateowner, &stp, NULL)))
2397                 goto out; 
2398         status = nfs_ok;
2399         update_stateid(&stp->st_stateid);
2400         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2401
2402         /* release_stateid() calls nfsd_close() if needed */
2403         release_stateid(stp, OPEN_STATE);
2404
2405         /* place unused nfs4_stateowners on so_close_lru list to be
2406          * released by the laundromat service after the lease period
2407          * to enable us to handle CLOSE replay
2408          */
2409         if (list_empty(&close->cl_stateowner->so_stateids))
2410                 move_to_close_lru(close->cl_stateowner);
2411 out:
2412         if (close->cl_stateowner) {
2413                 nfs4_get_stateowner(close->cl_stateowner);
2414                 cstate->replay_owner = close->cl_stateowner;
2415         }
2416         nfs4_unlock_state();
2417         return status;
2418 }
2419
2420 __be32
2421 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2422                   struct nfsd4_delegreturn *dr)
2423 {
2424         __be32 status;
2425
2426         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2427                 goto out;
2428
2429         nfs4_lock_state();
2430         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2431                                             &dr->dr_stateid, DELEG_RET, NULL);
2432         nfs4_unlock_state();
2433 out:
2434         return status;
2435 }
2436
2437
2438 /* 
2439  * Lock owner state (byte-range locks)
2440  */
2441 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2442 #define LOCK_HASH_BITS              8
2443 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2444 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2445
2446 #define lockownerid_hashval(id) \
2447         ((id) & LOCK_HASH_MASK)
2448
2449 static inline unsigned int
2450 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2451                 struct xdr_netobj *ownername)
2452 {
2453         return (file_hashval(inode) + cl_id
2454                         + opaque_hashval(ownername->data, ownername->len))
2455                 & LOCK_HASH_MASK;
2456 }
2457
2458 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2459 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2460 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2461
2462 static struct nfs4_stateid *
2463 find_stateid(stateid_t *stid, int flags)
2464 {
2465         struct nfs4_stateid *local = NULL;
2466         u32 st_id = stid->si_stateownerid;
2467         u32 f_id = stid->si_fileid;
2468         unsigned int hashval;
2469
2470         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2471         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2472                 hashval = stateid_hashval(st_id, f_id);
2473                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2474                         if ((local->st_stateid.si_stateownerid == st_id) &&
2475                             (local->st_stateid.si_fileid == f_id))
2476                                 return local;
2477                 }
2478         } 
2479         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2480                 hashval = stateid_hashval(st_id, f_id);
2481                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2482                         if ((local->st_stateid.si_stateownerid == st_id) &&
2483                             (local->st_stateid.si_fileid == f_id))
2484                                 return local;
2485                 }
2486         }
2487         return NULL;
2488 }
2489
2490 static struct nfs4_delegation *
2491 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2492 {
2493         struct nfs4_file *fp;
2494         struct nfs4_delegation *dl;
2495
2496         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2497                     stid->si_boot, stid->si_stateownerid,
2498                     stid->si_fileid, stid->si_generation);
2499
2500         fp = find_file(ino);
2501         if (!fp)
2502                 return NULL;
2503         dl = find_delegation_file(fp, stid);
2504         put_nfs4_file(fp);
2505         return dl;
2506 }
2507
2508 /*
2509  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2510  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2511  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2512  * locking, this prevents us from being completely protocol-compliant.  The
2513  * real solution to this problem is to start using unsigned file offsets in
2514  * the VFS, but this is a very deep change!
2515  */
2516 static inline void
2517 nfs4_transform_lock_offset(struct file_lock *lock)
2518 {
2519         if (lock->fl_start < 0)
2520                 lock->fl_start = OFFSET_MAX;
2521         if (lock->fl_end < 0)
2522                 lock->fl_end = OFFSET_MAX;
2523 }
2524
2525 /* Hack!: For now, we're defining this just so we can use a pointer to it
2526  * as a unique cookie to identify our (NFSv4's) posix locks. */
2527 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2528 };
2529
2530 static inline void
2531 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2532 {
2533         struct nfs4_stateowner *sop;
2534         unsigned int hval;
2535
2536         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2537                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2538                 hval = lockownerid_hashval(sop->so_id);
2539                 kref_get(&sop->so_ref);
2540                 deny->ld_sop = sop;
2541                 deny->ld_clientid = sop->so_client->cl_clientid;
2542         } else {
2543                 deny->ld_sop = NULL;
2544                 deny->ld_clientid.cl_boot = 0;
2545                 deny->ld_clientid.cl_id = 0;
2546         }
2547         deny->ld_start = fl->fl_start;
2548         deny->ld_length = ~(u64)0;
2549         if (fl->fl_end != ~(u64)0)
2550                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
2551         deny->ld_type = NFS4_READ_LT;
2552         if (fl->fl_type != F_RDLCK)
2553                 deny->ld_type = NFS4_WRITE_LT;
2554 }
2555
2556 static struct nfs4_stateowner *
2557 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2558                 struct xdr_netobj *owner)
2559 {
2560         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2561         struct nfs4_stateowner *op;
2562
2563         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2564                 if (cmp_owner_str(op, owner, clid))
2565                         return op;
2566         }
2567         return NULL;
2568 }
2569
2570 /*
2571  * Alloc a lock owner structure.
2572  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
2573  * occured. 
2574  *
2575  * strhashval = lock_ownerstr_hashval 
2576  */
2577
2578 static struct nfs4_stateowner *
2579 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2580         struct nfs4_stateowner *sop;
2581         struct nfs4_replay *rp;
2582         unsigned int idhashval;
2583
2584         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2585                 return NULL;
2586         idhashval = lockownerid_hashval(current_ownerid);
2587         INIT_LIST_HEAD(&sop->so_idhash);
2588         INIT_LIST_HEAD(&sop->so_strhash);
2589         INIT_LIST_HEAD(&sop->so_perclient);
2590         INIT_LIST_HEAD(&sop->so_stateids);
2591         INIT_LIST_HEAD(&sop->so_perstateid);
2592         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2593         sop->so_time = 0;
2594         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2595         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2596         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2597         sop->so_is_open_owner = 0;
2598         sop->so_id = current_ownerid++;
2599         sop->so_client = clp;
2600         /* It is the openowner seqid that will be incremented in encode in the
2601          * case of new lockowners; so increment the lock seqid manually: */
2602         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2603         sop->so_confirmed = 1;
2604         rp = &sop->so_replay;
2605         rp->rp_status = nfserr_serverfault;
2606         rp->rp_buflen = 0;
2607         rp->rp_buf = rp->rp_ibuf;
2608         return sop;
2609 }
2610
2611 static struct nfs4_stateid *
2612 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2613 {
2614         struct nfs4_stateid *stp;
2615         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2616
2617         stp = nfs4_alloc_stateid();
2618         if (stp == NULL)
2619                 goto out;
2620         INIT_LIST_HEAD(&stp->st_hash);
2621         INIT_LIST_HEAD(&stp->st_perfile);
2622         INIT_LIST_HEAD(&stp->st_perstateowner);
2623         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2624         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2625         list_add(&stp->st_perfile, &fp->fi_stateids);
2626         list_add(&stp->st_perstateowner, &sop->so_stateids);
2627         stp->st_stateowner = sop;
2628         get_nfs4_file(fp);
2629         stp->st_file = fp;
2630         stp->st_stateid.si_boot = boot_time;
2631         stp->st_stateid.si_stateownerid = sop->so_id;
2632         stp->st_stateid.si_fileid = fp->fi_id;
2633         stp->st_stateid.si_generation = 0;
2634         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2635         stp->st_access_bmap = open_stp->st_access_bmap;
2636         stp->st_deny_bmap = open_stp->st_deny_bmap;
2637         stp->st_openstp = open_stp;
2638
2639 out:
2640         return stp;
2641 }
2642
2643 static int
2644 check_lock_length(u64 offset, u64 length)
2645 {
2646         return ((length == 0)  || ((length != ~(u64)0) &&
2647              LOFF_OVERFLOW(offset, length)));
2648 }
2649
2650 /*
2651  *  LOCK operation 
2652  */
2653 __be32
2654 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2655            struct nfsd4_lock *lock)
2656 {
2657         struct nfs4_stateowner *open_sop = NULL;
2658         struct nfs4_stateowner *lock_sop = NULL;
2659         struct nfs4_stateid *lock_stp;
2660         struct file *filp;
2661         struct file_lock file_lock;
2662         struct file_lock conflock;
2663         __be32 status = 0;
2664         unsigned int strhashval;
2665         unsigned int cmd;
2666         int err;
2667
2668         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2669                 (long long) lock->lk_offset,
2670                 (long long) lock->lk_length);
2671
2672         if (check_lock_length(lock->lk_offset, lock->lk_length))
2673                  return nfserr_inval;
2674
2675         if ((status = fh_verify(rqstp, &cstate->current_fh,
2676                                 S_IFREG, MAY_LOCK))) {
2677                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2678                 return status;
2679         }
2680
2681         nfs4_lock_state();
2682
2683         if (lock->lk_is_new) {
2684                 /*
2685                  * Client indicates that this is a new lockowner.
2686                  * Use open owner and open stateid to create lock owner and
2687                  * lock stateid.
2688                  */
2689                 struct nfs4_stateid *open_stp = NULL;
2690                 struct nfs4_file *fp;
2691                 
2692                 status = nfserr_stale_clientid;
2693                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2694                         goto out;
2695
2696                 /* validate and update open stateid and open seqid */
2697                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2698                                         lock->lk_new_open_seqid,
2699                                         &lock->lk_new_open_stateid,
2700                                         CHECK_FH | OPEN_STATE,
2701                                         &lock->lk_replay_owner, &open_stp,
2702                                         lock);
2703                 if (status)
2704                         goto out;
2705                 open_sop = lock->lk_replay_owner;
2706                 /* create lockowner and lock stateid */
2707                 fp = open_stp->st_file;
2708                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
2709                                 open_sop->so_client->cl_clientid.cl_id, 
2710                                 &lock->v.new.owner);
2711                 /* XXX: Do we need to check for duplicate stateowners on
2712                  * the same file, or should they just be allowed (and
2713                  * create new stateids)? */
2714                 status = nfserr_resource;
2715                 lock_sop = alloc_init_lock_stateowner(strhashval,
2716                                 open_sop->so_client, open_stp, lock);
2717                 if (lock_sop == NULL)
2718                         goto out;
2719                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2720                 if (lock_stp == NULL)
2721                         goto out;
2722         } else {
2723                 /* lock (lock owner + lock stateid) already exists */
2724                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2725                                        lock->lk_old_lock_seqid, 
2726                                        &lock->lk_old_lock_stateid, 
2727                                        CHECK_FH | LOCK_STATE, 
2728                                        &lock->lk_replay_owner, &lock_stp, lock);
2729                 if (status)
2730                         goto out;
2731                 lock_sop = lock->lk_replay_owner;
2732         }
2733         /* lock->lk_replay_owner and lock_stp have been created or found */
2734         filp = lock_stp->st_vfs_file;
2735
2736         status = nfserr_grace;
2737         if (nfs4_in_grace() && !lock->lk_reclaim)
2738                 goto out;
2739         status = nfserr_no_grace;
2740         if (!nfs4_in_grace() && lock->lk_reclaim)
2741                 goto out;
2742
2743         locks_init_lock(&file_lock);
2744         switch (lock->lk_type) {
2745                 case NFS4_READ_LT:
2746                 case NFS4_READW_LT:
2747                         file_lock.fl_type = F_RDLCK;
2748                         cmd = F_SETLK;
2749                 break;
2750                 case NFS4_WRITE_LT:
2751                 case NFS4_WRITEW_LT:
2752                         file_lock.fl_type = F_WRLCK;
2753                         cmd = F_SETLK;
2754                 break;
2755                 default:
2756                         status = nfserr_inval;
2757                 goto out;
2758         }
2759         file_lock.fl_owner = (fl_owner_t)lock_sop;
2760         file_lock.fl_pid = current->tgid;
2761         file_lock.fl_file = filp;
2762         file_lock.fl_flags = FL_POSIX;
2763         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2764
2765         file_lock.fl_start = lock->lk_offset;
2766         if ((lock->lk_length == ~(u64)0) || 
2767                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2768                 file_lock.fl_end = ~(u64)0;
2769         else
2770                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2771         nfs4_transform_lock_offset(&file_lock);
2772
2773         /*
2774         * Try to lock the file in the VFS.
2775         * Note: locks.c uses the BKL to protect the inode's lock list.
2776         */
2777
2778         /* XXX?: Just to divert the locks_release_private at the start of
2779          * locks_copy_lock: */
2780         locks_init_lock(&conflock);
2781         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2782         switch (-err) {
2783         case 0: /* success! */
2784                 update_stateid(&lock_stp->st_stateid);
2785                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
2786                                 sizeof(stateid_t));
2787                 status = 0;
2788                 break;
2789         case (EAGAIN):          /* conflock holds conflicting lock */
2790                 status = nfserr_denied;
2791                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2792                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2793                 break;
2794         case (EDEADLK):
2795                 status = nfserr_deadlock;
2796                 break;
2797         default:        
2798                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2799                 status = nfserr_resource;
2800                 break;
2801         }
2802 out:
2803         if (status && lock->lk_is_new && lock_sop)
2804                 release_stateowner(lock_sop);
2805         if (lock->lk_replay_owner) {
2806                 nfs4_get_stateowner(lock->lk_replay_owner);
2807                 cstate->replay_owner = lock->lk_replay_owner;
2808         }
2809         nfs4_unlock_state();
2810         return status;
2811 }
2812
2813 /*
2814  * LOCKT operation
2815  */
2816 __be32
2817 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2818             struct nfsd4_lockt *lockt)
2819 {
2820         struct inode *inode;
2821         struct file file;
2822         struct file_lock file_lock;
2823         int error;
2824         __be32 status;
2825
2826         if (nfs4_in_grace())
2827                 return nfserr_grace;
2828
2829         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2830                  return nfserr_inval;
2831
2832         lockt->lt_stateowner = NULL;
2833         nfs4_lock_state();
2834
2835         status = nfserr_stale_clientid;
2836         if (STALE_CLIENTID(&lockt->lt_clientid))
2837                 goto out;
2838
2839         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2840                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2841                 if (status == nfserr_symlink)
2842                         status = nfserr_inval;
2843                 goto out;
2844         }
2845
2846         inode = cstate->current_fh.fh_dentry->d_inode;
2847         locks_init_lock(&file_lock);
2848         switch (lockt->lt_type) {
2849                 case NFS4_READ_LT:
2850                 case NFS4_READW_LT:
2851                         file_lock.fl_type = F_RDLCK;
2852                 break;
2853                 case NFS4_WRITE_LT:
2854                 case NFS4_WRITEW_LT:
2855                         file_lock.fl_type = F_WRLCK;
2856                 break;
2857                 default:
2858                         printk("NFSD: nfs4_lockt: bad lock type!\n");
2859                         status = nfserr_inval;
2860                 goto out;
2861         }
2862
2863         lockt->lt_stateowner = find_lockstateowner_str(inode,
2864                         &lockt->lt_clientid, &lockt->lt_owner);
2865         if (lockt->lt_stateowner)
2866                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2867         file_lock.fl_pid = current->tgid;
2868         file_lock.fl_flags = FL_POSIX;
2869         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2870
2871         file_lock.fl_start = lockt->lt_offset;
2872         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2873                 file_lock.fl_end = ~(u64)0;
2874         else
2875                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2876
2877         nfs4_transform_lock_offset(&file_lock);
2878
2879         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2880          * since LOCKT doesn't require an OPEN, and therefore a struct
2881          * file may not exist, pass vfs_test_lock a struct file with
2882          * only the dentry:inode set.
2883          */
2884         memset(&file, 0, sizeof (struct file));
2885         file.f_path.dentry = cstate->current_fh.fh_dentry;
2886
2887         status = nfs_ok;
2888         error = vfs_test_lock(&file, &file_lock);
2889         if (error) {
2890                 status = nfserrno(error);
2891                 goto out;
2892         }
2893         if (file_lock.fl_type != F_UNLCK) {
2894                 status = nfserr_denied;
2895                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2896         }
2897 out:
2898         nfs4_unlock_state();
2899         return status;
2900 }
2901
2902 __be32
2903 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2904             struct nfsd4_locku *locku)
2905 {
2906         struct nfs4_stateid *stp;
2907         struct file *filp = NULL;
2908         struct file_lock file_lock;
2909         __be32 status;
2910         int err;
2911                                                         
2912         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2913                 (long long) locku->lu_offset,
2914                 (long long) locku->lu_length);
2915
2916         if (check_lock_length(locku->lu_offset, locku->lu_length))
2917                  return nfserr_inval;
2918
2919         nfs4_lock_state();
2920                                                                                 
2921         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2922                                         locku->lu_seqid, 
2923                                         &locku->lu_stateid, 
2924                                         CHECK_FH | LOCK_STATE, 
2925                                         &locku->lu_stateowner, &stp, NULL)))
2926                 goto out;
2927
2928         filp = stp->st_vfs_file;
2929         BUG_ON(!filp);
2930         locks_init_lock(&file_lock);
2931         file_lock.fl_type = F_UNLCK;
2932         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2933         file_lock.fl_pid = current->tgid;
2934         file_lock.fl_file = filp;
2935         file_lock.fl_flags = FL_POSIX; 
2936         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2937         file_lock.fl_start = locku->lu_offset;
2938
2939         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2940                 file_lock.fl_end = ~(u64)0;
2941         else
2942                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2943         nfs4_transform_lock_offset(&file_lock);
2944
2945         /*
2946         *  Try to unlock the file in the VFS.
2947         */
2948         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2949         if (err) {
2950                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2951                 goto out_nfserr;
2952         }
2953         /*
2954         * OK, unlock succeeded; the only thing left to do is update the stateid.
2955         */
2956         update_stateid(&stp->st_stateid);
2957         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2958
2959 out:
2960         if (locku->lu_stateowner) {
2961                 nfs4_get_stateowner(locku->lu_stateowner);
2962                 cstate->replay_owner = locku->lu_stateowner;
2963         }
2964         nfs4_unlock_state();
2965         return status;
2966
2967 out_nfserr:
2968         status = nfserrno(err);
2969         goto out;
2970 }
2971
2972 /*
2973  * returns
2974  *      1: locks held by lockowner
2975  *      0: no locks held by lockowner
2976  */
2977 static int
2978 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2979 {
2980         struct file_lock **flpp;
2981         struct inode *inode = filp->f_path.dentry->d_inode;
2982         int status = 0;
2983
2984         lock_kernel();
2985         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2986                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2987                         status = 1;
2988                         goto out;
2989                 }
2990         }
2991 out:
2992         unlock_kernel();
2993         return status;
2994 }
2995
2996 __be32
2997 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2998                         struct nfsd4_compound_state *cstate,
2999                         struct nfsd4_release_lockowner *rlockowner)
3000 {
3001         clientid_t *clid = &rlockowner->rl_clientid;
3002         struct nfs4_stateowner *sop;
3003         struct nfs4_stateid *stp;
3004         struct xdr_netobj *owner = &rlockowner->rl_owner;
3005         struct list_head matches;
3006         int i;
3007         __be32 status;
3008
3009         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3010                 clid->cl_boot, clid->cl_id);
3011
3012         /* XXX check for lease expiration */
3013
3014         status = nfserr_stale_clientid;
3015         if (STALE_CLIENTID(clid))
3016                 return status;
3017
3018         nfs4_lock_state();
3019
3020         status = nfserr_locks_held;
3021         /* XXX: we're doing a linear search through all the lockowners.
3022          * Yipes!  For now we'll just hope clients aren't really using
3023          * release_lockowner much, but eventually we have to fix these
3024          * data structures. */
3025         INIT_LIST_HEAD(&matches);
3026         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3027                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3028                         if (!cmp_owner_str(sop, owner, clid))
3029                                 continue;
3030                         list_for_each_entry(stp, &sop->so_stateids,
3031                                         st_perstateowner) {
3032                                 if (check_for_locks(stp->st_vfs_file, sop))
3033                                         goto out;
3034                                 /* Note: so_perclient unused for lockowners,
3035                                  * so it's OK to fool with here. */
3036                                 list_add(&sop->so_perclient, &matches);
3037                         }
3038                 }
3039         }
3040         /* Clients probably won't expect us to return with some (but not all)
3041          * of the lockowner state released; so don't release any until all
3042          * have been checked. */
3043         status = nfs_ok;
3044         while (!list_empty(&matches)) {
3045                 sop = list_entry(matches.next, struct nfs4_stateowner,
3046                                                                 so_perclient);
3047                 /* unhash_stateowner deletes so_perclient only
3048                  * for openowners. */
3049                 list_del(&sop->so_perclient);
3050                 release_stateowner(sop);
3051         }
3052 out:
3053         nfs4_unlock_state();
3054         return status;
3055 }
3056
3057 static inline struct nfs4_client_reclaim *
3058 alloc_reclaim(void)
3059 {
3060         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3061 }
3062
3063 int
3064 nfs4_has_reclaimed_state(const char *name)
3065 {
3066         unsigned int strhashval = clientstr_hashval(name);
3067         struct nfs4_client *clp;
3068
3069         clp = find_confirmed_client_by_str(name, strhashval);
3070         return clp ? 1 : 0;
3071 }
3072
3073 /*
3074  * failure => all reset bets are off, nfserr_no_grace...
3075  */
3076 int
3077 nfs4_client_to_reclaim(const char *name)
3078 {
3079         unsigned int strhashval;
3080         struct nfs4_client_reclaim *crp = NULL;
3081
3082         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3083         crp = alloc_reclaim();
3084         if (!crp)
3085                 return 0;
3086         strhashval = clientstr_hashval(name);
3087         INIT_LIST_HEAD(&crp->cr_strhash);
3088         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3089         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3090         reclaim_str_hashtbl_size++;
3091         return 1;
3092 }
3093
3094 static void
3095 nfs4_release_reclaim(void)
3096 {
3097         struct nfs4_client_reclaim *crp = NULL;
3098         int i;
3099
3100         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3101                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3102                         crp = list_entry(reclaim_str_hashtbl[i].next,
3103                                         struct nfs4_client_reclaim, cr_strhash);
3104                         list_del(&crp->cr_strhash);
3105                         kfree(crp);
3106                         reclaim_str_hashtbl_size--;
3107                 }
3108         }
3109         BUG_ON(reclaim_str_hashtbl_size);
3110 }
3111
3112 /*
3113  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3114 static struct nfs4_client_reclaim *
3115 nfs4_find_reclaim_client(clientid_t *clid)
3116 {
3117         unsigned int strhashval;
3118         struct nfs4_client *clp;
3119         struct nfs4_client_reclaim *crp = NULL;
3120
3121
3122         /* find clientid in conf_id_hashtbl */
3123         clp = find_confirmed_client(clid);
3124         if (clp == NULL)
3125                 return NULL;
3126
3127         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3128                             clp->cl_name.len, clp->cl_name.data,
3129                             clp->cl_recdir);
3130
3131         /* find clp->cl_name in reclaim_str_hashtbl */
3132         strhashval = clientstr_hashval(clp->cl_recdir);
3133         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3134                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3135                         return crp;
3136                 }
3137         }
3138         return NULL;
3139 }
3140
3141 /*
3142 * Called from OPEN. Look for clientid in reclaim list.
3143 */
3144 __be32
3145 nfs4_check_open_reclaim(clientid_t *clid)
3146 {
3147         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3148 }
3149
3150 /* initialization to perform at module load time: */
3151
3152 void
3153 nfs4_state_init(void)
3154 {
3155         int i;
3156
3157         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3158                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3159                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3160                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3161                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3162         }
3163         for (i = 0; i < FILE_HASH_SIZE; i++) {
3164                 INIT_LIST_HEAD(&file_hashtbl[i]);
3165         }
3166         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3167                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3168                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3169         }
3170         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3171                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3172                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3173         }
3174         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3175                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3176                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3177         }
3178         memset(&onestateid, ~0, sizeof(stateid_t));
3179         INIT_LIST_HEAD(&close_lru);
3180         INIT_LIST_HEAD(&client_lru);
3181         INIT_LIST_HEAD(&del_recall_lru);
3182         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3183                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3184         reclaim_str_hashtbl_size = 0;
3185 }
3186
3187 static void
3188 nfsd4_load_reboot_recovery_data(void)
3189 {
3190         int status;
3191
3192         nfs4_lock_state();
3193         nfsd4_init_recdir(user_recovery_dirname);
3194         status = nfsd4_recdir_load();
3195         nfs4_unlock_state();
3196         if (status)
3197                 printk("NFSD: Failure reading reboot recovery data\n");
3198 }
3199
3200 unsigned long
3201 get_nfs4_grace_period(void)
3202 {
3203         return max(user_lease_time, lease_time) * HZ;
3204 }
3205
3206 /*
3207  * Since the lifetime of a delegation isn't limited to that of an open, a
3208  * client may quite reasonably hang on to a delegation as long as it has
3209  * the inode cached.  This becomes an obvious problem the first time a
3210  * client's inode cache approaches the size of the server's total memory.
3211  *
3212  * For now we avoid this problem by imposing a hard limit on the number
3213  * of delegations, which varies according to the server's memory size.
3214  */
3215 static void
3216 set_max_delegations(void)
3217 {
3218         /*
3219          * Allow at most 4 delegations per megabyte of RAM.  Quick
3220          * estimates suggest that in the worst case (where every delegation
3221          * is for a different inode), a delegation could take about 1.5K,
3222          * giving a worst case usage of about 6% of memory.
3223          */
3224         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3225 }
3226
3227 /* initialization to perform when the nfsd service is started: */
3228
3229 static void
3230 __nfs4_state_start(void)
3231 {
3232         unsigned long grace_time;
3233
3234         boot_time = get_seconds();
3235         grace_time = get_nfs_grace_period();
3236         lease_time = user_lease_time;
3237         in_grace = 1;
3238         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3239                grace_time/HZ);
3240         laundry_wq = create_singlethread_workqueue("nfsd4");
3241         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3242         set_max_delegations();
3243 }
3244
3245 int
3246 nfs4_state_start(void)
3247 {
3248         int status;
3249
3250         if (nfs4_init)
3251                 return 0;
3252         status = nfsd4_init_slabs();
3253         if (status)
3254                 return status;
3255         nfsd4_load_reboot_recovery_data();
3256         __nfs4_state_start();
3257         nfs4_init = 1;
3258         return 0;
3259 }
3260
3261 int
3262 nfs4_in_grace(void)
3263 {
3264         return in_grace;
3265 }
3266
3267 time_t
3268 nfs4_lease_time(void)
3269 {
3270         return lease_time;
3271 }
3272
3273 static void
3274 __nfs4_state_shutdown(void)
3275 {
3276         int i;
3277         struct nfs4_client *clp = NULL;
3278         struct nfs4_delegation *dp = NULL;
3279         struct list_head *pos, *next, reaplist;
3280
3281         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3282                 while (!list_empty(&conf_id_hashtbl[i])) {
3283                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3284                         expire_client(clp);
3285                 }
3286                 while (!list_empty(&unconf_str_hashtbl[i])) {
3287                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3288                         expire_client(clp);
3289                 }
3290         }
3291         INIT_LIST_HEAD(&reaplist);
3292         spin_lock(&recall_lock);
3293         list_for_each_safe(pos, next, &del_recall_lru) {
3294                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3295                 list_move(&dp->dl_recall_lru, &reaplist);
3296         }
3297         spin_unlock(&recall_lock);
3298         list_for_each_safe(pos, next, &reaplist) {
3299                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3300                 list_del_init(&dp->dl_recall_lru);
3301                 unhash_delegation(dp);
3302         }
3303
3304         nfsd4_shutdown_recdir();
3305         nfs4_init = 0;
3306 }
3307
3308 void
3309 nfs4_state_shutdown(void)
3310 {
3311         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3312         destroy_workqueue(laundry_wq);
3313         nfs4_lock_state();
3314         nfs4_release_reclaim();
3315         __nfs4_state_shutdown();
3316         nfsd4_free_slabs();
3317         nfs4_unlock_state();
3318 }
3319
3320 static void
3321 nfs4_set_recdir(char *recdir)
3322 {
3323         nfs4_lock_state();
3324         strcpy(user_recovery_dirname, recdir);
3325         nfs4_unlock_state();
3326 }
3327
3328 /*
3329  * Change the NFSv4 recovery directory to recdir.
3330  */
3331 int
3332 nfs4_reset_recoverydir(char *recdir)
3333 {
3334         int status;
3335         struct nameidata nd;
3336
3337         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3338         if (status)
3339                 return status;
3340         status = -ENOTDIR;
3341         if (S_ISDIR(nd.dentry->d_inode->i_mode)) {
3342                 nfs4_set_recdir(recdir);
3343                 status = 0;
3344         }
3345         path_release(&nd);
3346         return status;
3347 }
3348
3349 /*
3350  * Called when leasetime is changed.
3351  *
3352  * The only way the protocol gives us to handle on-the-fly lease changes is to
3353  * simulate a reboot.  Instead of doing that, we just wait till the next time
3354  * we start to register any changes in lease time.  If the administrator
3355  * really wants to change the lease time *now*, they can go ahead and bring
3356  * nfsd down and then back up again after changing the lease time.
3357  */
3358 void
3359 nfs4_reset_lease(time_t leasetime)
3360 {
3361         lock_kernel();
3362         user_lease_time = leasetime;
3363         unlock_kernel();
3364 }