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