Merge branch 'cleanups' into next
[linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/mount.h>
45 #include <linux/workqueue.h>
46 #include <linux/smp_lock.h>
47 #include <linux/kthread.h>
48 #include <linux/nfs4.h>
49 #include <linux/nfsd/state.h>
50 #include <linux/nfsd/xdr4.h>
51 #include <linux/namei.h>
52 #include <linux/swap.h>
53 #include <linux/mutex.h>
54 #include <linux/lockd/bind.h>
55 #include <linux/module.h>
56
57 #define NFSDDBG_FACILITY                NFSDDBG_PROC
58
59 /* Globals */
60 static time_t lease_time = 90;     /* default lease time */
61 static time_t user_lease_time = 90;
62 static time_t boot_time;
63 static int in_grace = 1;
64 static u32 current_ownerid = 1;
65 static u32 current_fileid = 1;
66 static u32 current_delegid = 1;
67 static u32 nfs4_init;
68 static stateid_t zerostateid;             /* bits all 0 */
69 static stateid_t onestateid;              /* bits all 1 */
70
71 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
72 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
73
74 /* forward declarations */
75 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
76 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
77 static void release_stateid_lockowners(struct nfs4_stateid *open_stp);
78 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
79 static void nfs4_set_recdir(char *recdir);
80
81 /* Locking:
82  *
83  * client_mutex:
84  *      protects clientid_hashtbl[], clientstr_hashtbl[],
85  *      unconfstr_hashtbl[], uncofid_hashtbl[].
86  */
87 static DEFINE_MUTEX(client_mutex);
88
89 static struct kmem_cache *stateowner_slab = NULL;
90 static struct kmem_cache *file_slab = NULL;
91 static struct kmem_cache *stateid_slab = NULL;
92 static struct kmem_cache *deleg_slab = NULL;
93
94 void
95 nfs4_lock_state(void)
96 {
97         mutex_lock(&client_mutex);
98 }
99
100 void
101 nfs4_unlock_state(void)
102 {
103         mutex_unlock(&client_mutex);
104 }
105
106 static inline u32
107 opaque_hashval(const void *ptr, int nbytes)
108 {
109         unsigned char *cptr = (unsigned char *) ptr;
110
111         u32 x = 0;
112         while (nbytes--) {
113                 x *= 37;
114                 x += *cptr++;
115         }
116         return x;
117 }
118
119 /* forward declarations */
120 static void release_stateowner(struct nfs4_stateowner *sop);
121 static void release_stateid(struct nfs4_stateid *stp, int flags);
122
123 /*
124  * Delegation state
125  */
126
127 /* recall_lock protects the del_recall_lru */
128 static DEFINE_SPINLOCK(recall_lock);
129 static struct list_head del_recall_lru;
130
131 static void
132 free_nfs4_file(struct kref *kref)
133 {
134         struct nfs4_file *fp = container_of(kref, struct nfs4_file, fi_ref);
135         list_del(&fp->fi_hash);
136         iput(fp->fi_inode);
137         kmem_cache_free(file_slab, fp);
138 }
139
140 static inline void
141 put_nfs4_file(struct nfs4_file *fi)
142 {
143         kref_put(&fi->fi_ref, free_nfs4_file);
144 }
145
146 static inline void
147 get_nfs4_file(struct nfs4_file *fi)
148 {
149         kref_get(&fi->fi_ref);
150 }
151
152 static int num_delegations;
153 unsigned int max_delegations;
154
155 /*
156  * Open owner state (share locks)
157  */
158
159 /* hash tables for nfs4_stateowner */
160 #define OWNER_HASH_BITS              8
161 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
162 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
163
164 #define ownerid_hashval(id) \
165         ((id) & OWNER_HASH_MASK)
166 #define ownerstr_hashval(clientid, ownername) \
167         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
168
169 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
170 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
171
172 /* hash table for nfs4_file */
173 #define FILE_HASH_BITS                   8
174 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
175 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
176 /* hash table for (open)nfs4_stateid */
177 #define STATEID_HASH_BITS              10
178 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
179 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
180
181 #define file_hashval(x) \
182         hash_ptr(x, FILE_HASH_BITS)
183 #define stateid_hashval(owner_id, file_id)  \
184         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
185
186 static struct list_head file_hashtbl[FILE_HASH_SIZE];
187 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
188
189 static struct nfs4_delegation *
190 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
191 {
192         struct nfs4_delegation *dp;
193         struct nfs4_file *fp = stp->st_file;
194         struct nfs4_callback *cb = &stp->st_stateowner->so_client->cl_callback;
195
196         dprintk("NFSD alloc_init_deleg\n");
197         if (fp->fi_had_conflict)
198                 return NULL;
199         if (num_delegations > max_delegations)
200                 return NULL;
201         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
202         if (dp == NULL)
203                 return dp;
204         num_delegations++;
205         INIT_LIST_HEAD(&dp->dl_perfile);
206         INIT_LIST_HEAD(&dp->dl_perclnt);
207         INIT_LIST_HEAD(&dp->dl_recall_lru);
208         dp->dl_client = clp;
209         get_nfs4_file(fp);
210         dp->dl_file = fp;
211         dp->dl_flock = NULL;
212         get_file(stp->st_vfs_file);
213         dp->dl_vfs_file = stp->st_vfs_file;
214         dp->dl_type = type;
215         dp->dl_recall.cbr_dp = NULL;
216         dp->dl_recall.cbr_ident = cb->cb_ident;
217         dp->dl_recall.cbr_trunc = 0;
218         dp->dl_stateid.si_boot = boot_time;
219         dp->dl_stateid.si_stateownerid = current_delegid++;
220         dp->dl_stateid.si_fileid = 0;
221         dp->dl_stateid.si_generation = 0;
222         dp->dl_fhlen = current_fh->fh_handle.fh_size;
223         memcpy(dp->dl_fhval, &current_fh->fh_handle.fh_base,
224                         current_fh->fh_handle.fh_size);
225         dp->dl_time = 0;
226         atomic_set(&dp->dl_count, 1);
227         list_add(&dp->dl_perfile, &fp->fi_delegations);
228         list_add(&dp->dl_perclnt, &clp->cl_delegations);
229         return dp;
230 }
231
232 void
233 nfs4_put_delegation(struct nfs4_delegation *dp)
234 {
235         if (atomic_dec_and_test(&dp->dl_count)) {
236                 dprintk("NFSD: freeing dp %p\n",dp);
237                 put_nfs4_file(dp->dl_file);
238                 kmem_cache_free(deleg_slab, dp);
239                 num_delegations--;
240         }
241 }
242
243 /* Remove the associated file_lock first, then remove the delegation.
244  * lease_modify() is called to remove the FS_LEASE file_lock from
245  * the i_flock list, eventually calling nfsd's lock_manager
246  * fl_release_callback.
247  */
248 static void
249 nfs4_close_delegation(struct nfs4_delegation *dp)
250 {
251         struct file *filp = dp->dl_vfs_file;
252
253         dprintk("NFSD: close_delegation dp %p\n",dp);
254         dp->dl_vfs_file = NULL;
255         /* The following nfsd_close may not actually close the file,
256          * but we want to remove the lease in any case. */
257         if (dp->dl_flock)
258                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
259         nfsd_close(filp);
260 }
261
262 /* Called under the state lock. */
263 static void
264 unhash_delegation(struct nfs4_delegation *dp)
265 {
266         list_del_init(&dp->dl_perfile);
267         list_del_init(&dp->dl_perclnt);
268         spin_lock(&recall_lock);
269         list_del_init(&dp->dl_recall_lru);
270         spin_unlock(&recall_lock);
271         nfs4_close_delegation(dp);
272         nfs4_put_delegation(dp);
273 }
274
275 /* 
276  * SETCLIENTID state 
277  */
278
279 /* Hash tables for nfs4_clientid state */
280 #define CLIENT_HASH_BITS                 4
281 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
282 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
283
284 #define clientid_hashval(id) \
285         ((id) & CLIENT_HASH_MASK)
286 #define clientstr_hashval(name) \
287         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
288 /*
289  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
290  * used in reboot/reset lease grace period processing
291  *
292  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
293  * setclientid_confirmed info. 
294  *
295  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
296  * setclientid info.
297  *
298  * client_lru holds client queue ordered by nfs4_client.cl_time
299  * for lease renewal.
300  *
301  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
302  * for last close replay.
303  */
304 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
305 static int reclaim_str_hashtbl_size = 0;
306 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
307 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
308 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
309 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
310 static struct list_head client_lru;
311 static struct list_head close_lru;
312
313 static inline void
314 renew_client(struct nfs4_client *clp)
315 {
316         /*
317         * Move client to the end to the LRU list.
318         */
319         dprintk("renewing client (clientid %08x/%08x)\n", 
320                         clp->cl_clientid.cl_boot, 
321                         clp->cl_clientid.cl_id);
322         list_move_tail(&clp->cl_lru, &client_lru);
323         clp->cl_time = get_seconds();
324 }
325
326 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
327 static int
328 STALE_CLIENTID(clientid_t *clid)
329 {
330         if (clid->cl_boot == boot_time)
331                 return 0;
332         dprintk("NFSD stale clientid (%08x/%08x)\n", 
333                         clid->cl_boot, clid->cl_id);
334         return 1;
335 }
336
337 /* 
338  * XXX Should we use a slab cache ?
339  * This type of memory management is somewhat inefficient, but we use it
340  * anyway since SETCLIENTID is not a common operation.
341  */
342 static struct nfs4_client *alloc_client(struct xdr_netobj name)
343 {
344         struct nfs4_client *clp;
345
346         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
347         if (clp == NULL)
348                 return NULL;
349         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
350         if (clp->cl_name.data == NULL) {
351                 kfree(clp);
352                 return NULL;
353         }
354         memcpy(clp->cl_name.data, name.data, name.len);
355         clp->cl_name.len = name.len;
356         return clp;
357 }
358
359 static void
360 shutdown_callback_client(struct nfs4_client *clp)
361 {
362         struct rpc_clnt *clnt = clp->cl_callback.cb_client;
363
364         if (clnt) {
365                 /*
366                  * Callback threads take a reference on the client, so there
367                  * should be no outstanding callbacks at this point.
368                  */
369                 clp->cl_callback.cb_client = NULL;
370                 rpc_shutdown_client(clnt);
371         }
372 }
373
374 static inline void
375 free_client(struct nfs4_client *clp)
376 {
377         shutdown_callback_client(clp);
378         if (clp->cl_cred.cr_group_info)
379                 put_group_info(clp->cl_cred.cr_group_info);
380         kfree(clp->cl_name.data);
381         kfree(clp);
382 }
383
384 void
385 put_nfs4_client(struct nfs4_client *clp)
386 {
387         if (atomic_dec_and_test(&clp->cl_count))
388                 free_client(clp);
389 }
390
391 static void
392 expire_client(struct nfs4_client *clp)
393 {
394         struct nfs4_stateowner *sop;
395         struct nfs4_delegation *dp;
396         struct list_head reaplist;
397
398         dprintk("NFSD: expire_client cl_count %d\n",
399                             atomic_read(&clp->cl_count));
400
401         INIT_LIST_HEAD(&reaplist);
402         spin_lock(&recall_lock);
403         while (!list_empty(&clp->cl_delegations)) {
404                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
405                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
406                                 dp->dl_flock);
407                 list_del_init(&dp->dl_perclnt);
408                 list_move(&dp->dl_recall_lru, &reaplist);
409         }
410         spin_unlock(&recall_lock);
411         while (!list_empty(&reaplist)) {
412                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
413                 list_del_init(&dp->dl_recall_lru);
414                 unhash_delegation(dp);
415         }
416         list_del(&clp->cl_idhash);
417         list_del(&clp->cl_strhash);
418         list_del(&clp->cl_lru);
419         while (!list_empty(&clp->cl_openowners)) {
420                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
421                 release_stateowner(sop);
422         }
423         put_nfs4_client(clp);
424 }
425
426 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
427 {
428         struct nfs4_client *clp;
429
430         clp = alloc_client(name);
431         if (clp == NULL)
432                 return NULL;
433         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
434         atomic_set(&clp->cl_count, 1);
435         atomic_set(&clp->cl_callback.cb_set, 0);
436         INIT_LIST_HEAD(&clp->cl_idhash);
437         INIT_LIST_HEAD(&clp->cl_strhash);
438         INIT_LIST_HEAD(&clp->cl_openowners);
439         INIT_LIST_HEAD(&clp->cl_delegations);
440         INIT_LIST_HEAD(&clp->cl_lru);
441         return clp;
442 }
443
444 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
445 {
446         memcpy(target->cl_verifier.data, source->data,
447                         sizeof(target->cl_verifier.data));
448 }
449
450 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
451 {
452         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
453         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
454 }
455
456 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
457 {
458         target->cr_uid = source->cr_uid;
459         target->cr_gid = source->cr_gid;
460         target->cr_group_info = source->cr_group_info;
461         get_group_info(target->cr_group_info);
462 }
463
464 static int 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 gen_clid(struct nfs4_client *clp)
489 {
490         static u32 current_clientid = 1;
491
492         clp->cl_clientid.cl_boot = boot_time;
493         clp->cl_clientid.cl_id = current_clientid++; 
494 }
495
496 static void gen_confirm(struct nfs4_client *clp)
497 {
498         static u32 i;
499         u32 *p;
500
501         p = (u32 *)clp->cl_confirm.data;
502         *p++ = get_seconds();
503         *p++ = i++;
504 }
505
506 static int 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 __be32
687 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
688                   struct nfsd4_setclientid *setclid)
689 {
690         struct sockaddr_in      *sin = svc_addr_in(rqstp);
691         struct xdr_netobj       clname = { 
692                 .len = setclid->se_namelen,
693                 .data = setclid->se_name,
694         };
695         nfs4_verifier           clverifier = setclid->se_verf;
696         unsigned int            strhashval;
697         struct nfs4_client      *conf, *unconf, *new;
698         __be32                  status;
699         char                    dname[HEXDIR_LEN];
700         
701         if (!check_name(clname))
702                 return nfserr_inval;
703
704         status = nfs4_make_rec_clidname(dname, &clname);
705         if (status)
706                 return status;
707
708         /* 
709          * XXX The Duplicate Request Cache (DRC) has been checked (??)
710          * We get here on a DRC miss.
711          */
712
713         strhashval = clientstr_hashval(dname);
714
715         nfs4_lock_state();
716         conf = find_confirmed_client_by_str(dname, strhashval);
717         if (conf) {
718                 /* RFC 3530 14.2.33 CASE 0: */
719                 status = nfserr_clid_inuse;
720                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)
721                                 || conf->cl_addr != sin->sin_addr.s_addr) {
722                         dprintk("NFSD: setclientid: string in use by client"
723                                 "at %u.%u.%u.%u\n", NIPQUAD(conf->cl_addr));
724                         goto out;
725                 }
726         }
727         /*
728          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
729          * has a description of SETCLIENTID request processing consisting
730          * of 5 bullet points, labeled as CASE0 - CASE4 below.
731          */
732         unconf = find_unconfirmed_client_by_str(dname, strhashval);
733         status = nfserr_resource;
734         if (!conf) {
735                 /*
736                  * RFC 3530 14.2.33 CASE 4:
737                  * placed first, because it is the normal case
738                  */
739                 if (unconf)
740                         expire_client(unconf);
741                 new = create_client(clname, dname);
742                 if (new == NULL)
743                         goto out;
744                 gen_clid(new);
745         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
746                 /*
747                  * RFC 3530 14.2.33 CASE 1:
748                  * probable callback update
749                  */
750                 if (unconf) {
751                         /* Note this is removing unconfirmed {*x***},
752                          * which is stronger than RFC recommended {vxc**}.
753                          * This has the advantage that there is at most
754                          * one {*x***} in either list at any time.
755                          */
756                         expire_client(unconf);
757                 }
758                 new = create_client(clname, dname);
759                 if (new == NULL)
760                         goto out;
761                 copy_clid(new, conf);
762         } else if (!unconf) {
763                 /*
764                  * RFC 3530 14.2.33 CASE 2:
765                  * probable client reboot; state will be removed if
766                  * confirmed.
767                  */
768                 new = create_client(clname, dname);
769                 if (new == NULL)
770                         goto out;
771                 gen_clid(new);
772         } else {
773                 /*
774                  * RFC 3530 14.2.33 CASE 3:
775                  * probable client reboot; state will be removed if
776                  * confirmed.
777                  */
778                 expire_client(unconf);
779                 new = create_client(clname, dname);
780                 if (new == NULL)
781                         goto out;
782                 gen_clid(new);
783         }
784         copy_verf(new, &clverifier);
785         new->cl_addr = sin->sin_addr.s_addr;
786         copy_cred(&new->cl_cred, &rqstp->rq_cred);
787         gen_confirm(new);
788         gen_callback(new, setclid);
789         add_to_unconfirmed(new, strhashval);
790         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
791         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
792         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
793         status = nfs_ok;
794 out:
795         nfs4_unlock_state();
796         return status;
797 }
798
799
800 /*
801  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
802  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
803  * bullets, labeled as CASE1 - CASE4 below.
804  */
805 __be32
806 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
807                          struct nfsd4_compound_state *cstate,
808                          struct nfsd4_setclientid_confirm *setclientid_confirm)
809 {
810         struct sockaddr_in *sin = svc_addr_in(rqstp);
811         struct nfs4_client *conf, *unconf;
812         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
813         clientid_t * clid = &setclientid_confirm->sc_clientid;
814         __be32 status;
815
816         if (STALE_CLIENTID(clid))
817                 return nfserr_stale_clientid;
818         /* 
819          * XXX The Duplicate Request Cache (DRC) has been checked (??)
820          * We get here on a DRC miss.
821          */
822
823         nfs4_lock_state();
824
825         conf = find_confirmed_client(clid);
826         unconf = find_unconfirmed_client(clid);
827
828         status = nfserr_clid_inuse;
829         if (conf && conf->cl_addr != sin->sin_addr.s_addr)
830                 goto out;
831         if (unconf && unconf->cl_addr != sin->sin_addr.s_addr)
832                 goto out;
833
834         /*
835          * section 14.2.34 of RFC 3530 has a description of
836          * SETCLIENTID_CONFIRM request processing consisting
837          * of 4 bullet points, labeled as CASE1 - CASE4 below.
838          */
839         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
840                 /*
841                  * RFC 3530 14.2.34 CASE 1:
842                  * callback update
843                  */
844                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
845                         status = nfserr_clid_inuse;
846                 else {
847                         /* XXX: We just turn off callbacks until we can handle
848                           * change request correctly. */
849                         atomic_set(&conf->cl_callback.cb_set, 0);
850                         gen_confirm(conf);
851                         nfsd4_remove_clid_dir(unconf);
852                         expire_client(unconf);
853                         status = nfs_ok;
854
855                 }
856         } else if (conf && !unconf) {
857                 /*
858                  * RFC 3530 14.2.34 CASE 2:
859                  * probable retransmitted request; play it safe and
860                  * do nothing.
861                  */
862                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
863                         status = nfserr_clid_inuse;
864                 else
865                         status = nfs_ok;
866         } else if (!conf && unconf
867                         && same_verf(&unconf->cl_confirm, &confirm)) {
868                 /*
869                  * RFC 3530 14.2.34 CASE 3:
870                  * Normal case; new or rebooted client:
871                  */
872                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
873                         status = nfserr_clid_inuse;
874                 } else {
875                         unsigned int hash =
876                                 clientstr_hashval(unconf->cl_recdir);
877                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
878                                                                         hash);
879                         if (conf) {
880                                 nfsd4_remove_clid_dir(conf);
881                                 expire_client(conf);
882                         }
883                         move_to_confirmed(unconf);
884                         conf = unconf;
885                         nfsd4_probe_callback(conf);
886                         status = nfs_ok;
887                 }
888         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
889             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
890                                                                 &confirm)))) {
891                 /*
892                  * RFC 3530 14.2.34 CASE 4:
893                  * Client probably hasn't noticed that we rebooted yet.
894                  */
895                 status = nfserr_stale_clientid;
896         } else {
897                 /* check that we have hit one of the cases...*/
898                 status = nfserr_clid_inuse;
899         }
900 out:
901         nfs4_unlock_state();
902         return status;
903 }
904
905 /* OPEN Share state helper functions */
906 static inline struct nfs4_file *
907 alloc_init_file(struct inode *ino)
908 {
909         struct nfs4_file *fp;
910         unsigned int hashval = file_hashval(ino);
911
912         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
913         if (fp) {
914                 kref_init(&fp->fi_ref);
915                 INIT_LIST_HEAD(&fp->fi_hash);
916                 INIT_LIST_HEAD(&fp->fi_stateids);
917                 INIT_LIST_HEAD(&fp->fi_delegations);
918                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
919                 fp->fi_inode = igrab(ino);
920                 fp->fi_id = current_fileid++;
921                 fp->fi_had_conflict = false;
922                 return fp;
923         }
924         return NULL;
925 }
926
927 static void
928 nfsd4_free_slab(struct kmem_cache **slab)
929 {
930         if (*slab == NULL)
931                 return;
932         kmem_cache_destroy(*slab);
933         *slab = NULL;
934 }
935
936 void
937 nfsd4_free_slabs(void)
938 {
939         nfsd4_free_slab(&stateowner_slab);
940         nfsd4_free_slab(&file_slab);
941         nfsd4_free_slab(&stateid_slab);
942         nfsd4_free_slab(&deleg_slab);
943 }
944
945 static int
946 nfsd4_init_slabs(void)
947 {
948         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
949                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
950         if (stateowner_slab == NULL)
951                 goto out_nomem;
952         file_slab = kmem_cache_create("nfsd4_files",
953                         sizeof(struct nfs4_file), 0, 0, NULL);
954         if (file_slab == NULL)
955                 goto out_nomem;
956         stateid_slab = kmem_cache_create("nfsd4_stateids",
957                         sizeof(struct nfs4_stateid), 0, 0, NULL);
958         if (stateid_slab == NULL)
959                 goto out_nomem;
960         deleg_slab = kmem_cache_create("nfsd4_delegations",
961                         sizeof(struct nfs4_delegation), 0, 0, NULL);
962         if (deleg_slab == NULL)
963                 goto out_nomem;
964         return 0;
965 out_nomem:
966         nfsd4_free_slabs();
967         dprintk("nfsd4: out of memory while initializing nfsv4\n");
968         return -ENOMEM;
969 }
970
971 void
972 nfs4_free_stateowner(struct kref *kref)
973 {
974         struct nfs4_stateowner *sop =
975                 container_of(kref, struct nfs4_stateowner, so_ref);
976         kfree(sop->so_owner.data);
977         kmem_cache_free(stateowner_slab, sop);
978 }
979
980 static inline struct nfs4_stateowner *
981 alloc_stateowner(struct xdr_netobj *owner)
982 {
983         struct nfs4_stateowner *sop;
984
985         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
986                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
987                         memcpy(sop->so_owner.data, owner->data, owner->len);
988                         sop->so_owner.len = owner->len;
989                         kref_init(&sop->so_ref);
990                         return sop;
991                 } 
992                 kmem_cache_free(stateowner_slab, sop);
993         }
994         return NULL;
995 }
996
997 static struct nfs4_stateowner *
998 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
999         struct nfs4_stateowner *sop;
1000         struct nfs4_replay *rp;
1001         unsigned int idhashval;
1002
1003         if (!(sop = alloc_stateowner(&open->op_owner)))
1004                 return NULL;
1005         idhashval = ownerid_hashval(current_ownerid);
1006         INIT_LIST_HEAD(&sop->so_idhash);
1007         INIT_LIST_HEAD(&sop->so_strhash);
1008         INIT_LIST_HEAD(&sop->so_perclient);
1009         INIT_LIST_HEAD(&sop->so_stateids);
1010         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1011         INIT_LIST_HEAD(&sop->so_close_lru);
1012         sop->so_time = 0;
1013         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1014         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1015         list_add(&sop->so_perclient, &clp->cl_openowners);
1016         sop->so_is_open_owner = 1;
1017         sop->so_id = current_ownerid++;
1018         sop->so_client = clp;
1019         sop->so_seqid = open->op_seqid;
1020         sop->so_confirmed = 0;
1021         rp = &sop->so_replay;
1022         rp->rp_status = nfserr_serverfault;
1023         rp->rp_buflen = 0;
1024         rp->rp_buf = rp->rp_ibuf;
1025         return sop;
1026 }
1027
1028 static void
1029 release_stateid_lockowners(struct nfs4_stateid *open_stp)
1030 {
1031         struct nfs4_stateowner *lock_sop;
1032
1033         while (!list_empty(&open_stp->st_lockowners)) {
1034                 lock_sop = list_entry(open_stp->st_lockowners.next,
1035                                 struct nfs4_stateowner, so_perstateid);
1036                 /* list_del(&open_stp->st_lockowners);  */
1037                 BUG_ON(lock_sop->so_is_open_owner);
1038                 release_stateowner(lock_sop);
1039         }
1040 }
1041
1042 static void
1043 unhash_stateowner(struct nfs4_stateowner *sop)
1044 {
1045         struct nfs4_stateid *stp;
1046
1047         list_del(&sop->so_idhash);
1048         list_del(&sop->so_strhash);
1049         if (sop->so_is_open_owner)
1050                 list_del(&sop->so_perclient);
1051         list_del(&sop->so_perstateid);
1052         while (!list_empty(&sop->so_stateids)) {
1053                 stp = list_entry(sop->so_stateids.next,
1054                         struct nfs4_stateid, st_perstateowner);
1055                 if (sop->so_is_open_owner)
1056                         release_stateid(stp, OPEN_STATE);
1057                 else
1058                         release_stateid(stp, LOCK_STATE);
1059         }
1060 }
1061
1062 static void
1063 release_stateowner(struct nfs4_stateowner *sop)
1064 {
1065         unhash_stateowner(sop);
1066         list_del(&sop->so_close_lru);
1067         nfs4_put_stateowner(sop);
1068 }
1069
1070 static inline void
1071 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1072         struct nfs4_stateowner *sop = open->op_stateowner;
1073         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1074
1075         INIT_LIST_HEAD(&stp->st_hash);
1076         INIT_LIST_HEAD(&stp->st_perstateowner);
1077         INIT_LIST_HEAD(&stp->st_lockowners);
1078         INIT_LIST_HEAD(&stp->st_perfile);
1079         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1080         list_add(&stp->st_perstateowner, &sop->so_stateids);
1081         list_add(&stp->st_perfile, &fp->fi_stateids);
1082         stp->st_stateowner = sop;
1083         get_nfs4_file(fp);
1084         stp->st_file = fp;
1085         stp->st_stateid.si_boot = boot_time;
1086         stp->st_stateid.si_stateownerid = sop->so_id;
1087         stp->st_stateid.si_fileid = fp->fi_id;
1088         stp->st_stateid.si_generation = 0;
1089         stp->st_access_bmap = 0;
1090         stp->st_deny_bmap = 0;
1091         __set_bit(open->op_share_access, &stp->st_access_bmap);
1092         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1093         stp->st_openstp = NULL;
1094 }
1095
1096 static void
1097 release_stateid(struct nfs4_stateid *stp, int flags)
1098 {
1099         struct file *filp = stp->st_vfs_file;
1100
1101         list_del(&stp->st_hash);
1102         list_del(&stp->st_perfile);
1103         list_del(&stp->st_perstateowner);
1104         if (flags & OPEN_STATE) {
1105                 release_stateid_lockowners(stp);
1106                 stp->st_vfs_file = NULL;
1107                 nfsd_close(filp);
1108         } else if (flags & LOCK_STATE)
1109                 locks_remove_posix(filp, (fl_owner_t) stp->st_stateowner);
1110         put_nfs4_file(stp->st_file);
1111         kmem_cache_free(stateid_slab, stp);
1112 }
1113
1114 static void
1115 move_to_close_lru(struct nfs4_stateowner *sop)
1116 {
1117         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1118
1119         list_move_tail(&sop->so_close_lru, &close_lru);
1120         sop->so_time = get_seconds();
1121 }
1122
1123 static int
1124 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1125                                                         clientid_t *clid)
1126 {
1127         return (sop->so_owner.len == owner->len) &&
1128                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1129                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1130 }
1131
1132 static struct nfs4_stateowner *
1133 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1134 {
1135         struct nfs4_stateowner *so = NULL;
1136
1137         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1138                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1139                         return so;
1140         }
1141         return NULL;
1142 }
1143
1144 /* search file_hashtbl[] for file */
1145 static struct nfs4_file *
1146 find_file(struct inode *ino)
1147 {
1148         unsigned int hashval = file_hashval(ino);
1149         struct nfs4_file *fp;
1150
1151         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1152                 if (fp->fi_inode == ino) {
1153                         get_nfs4_file(fp);
1154                         return fp;
1155                 }
1156         }
1157         return NULL;
1158 }
1159
1160 static inline int access_valid(u32 x)
1161 {
1162         if (x < NFS4_SHARE_ACCESS_READ)
1163                 return 0;
1164         if (x > NFS4_SHARE_ACCESS_BOTH)
1165                 return 0;
1166         return 1;
1167 }
1168
1169 static inline int deny_valid(u32 x)
1170 {
1171         /* Note: unlike access bits, deny bits may be zero. */
1172         return x <= NFS4_SHARE_DENY_BOTH;
1173 }
1174
1175 static void
1176 set_access(unsigned int *access, unsigned long bmap) {
1177         int i;
1178
1179         *access = 0;
1180         for (i = 1; i < 4; i++) {
1181                 if (test_bit(i, &bmap))
1182                         *access |= i;
1183         }
1184 }
1185
1186 static void
1187 set_deny(unsigned int *deny, unsigned long bmap) {
1188         int i;
1189
1190         *deny = 0;
1191         for (i = 0; i < 4; i++) {
1192                 if (test_bit(i, &bmap))
1193                         *deny |= i ;
1194         }
1195 }
1196
1197 static int
1198 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1199         unsigned int access, deny;
1200
1201         set_access(&access, stp->st_access_bmap);
1202         set_deny(&deny, stp->st_deny_bmap);
1203         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1204                 return 0;
1205         return 1;
1206 }
1207
1208 /*
1209  * Called to check deny when READ with all zero stateid or
1210  * WRITE with all zero or all one stateid
1211  */
1212 static __be32
1213 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1214 {
1215         struct inode *ino = current_fh->fh_dentry->d_inode;
1216         struct nfs4_file *fp;
1217         struct nfs4_stateid *stp;
1218         __be32 ret;
1219
1220         dprintk("NFSD: nfs4_share_conflict\n");
1221
1222         fp = find_file(ino);
1223         if (!fp)
1224                 return nfs_ok;
1225         ret = nfserr_locked;
1226         /* Search for conflicting share reservations */
1227         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1228                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1229                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1230                         goto out;
1231         }
1232         ret = nfs_ok;
1233 out:
1234         put_nfs4_file(fp);
1235         return ret;
1236 }
1237
1238 static inline void
1239 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1240 {
1241         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1242                 put_write_access(filp->f_path.dentry->d_inode);
1243                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1244         }
1245 }
1246
1247 /*
1248  * Recall a delegation
1249  */
1250 static int
1251 do_recall(void *__dp)
1252 {
1253         struct nfs4_delegation *dp = __dp;
1254
1255         dp->dl_file->fi_had_conflict = true;
1256         nfsd4_cb_recall(dp);
1257         return 0;
1258 }
1259
1260 /*
1261  * Spawn a thread to perform a recall on the delegation represented
1262  * by the lease (file_lock)
1263  *
1264  * Called from break_lease() with lock_kernel() held.
1265  * Note: we assume break_lease will only call this *once* for any given
1266  * lease.
1267  */
1268 static
1269 void nfsd_break_deleg_cb(struct file_lock *fl)
1270 {
1271         struct nfs4_delegation *dp=  (struct nfs4_delegation *)fl->fl_owner;
1272         struct task_struct *t;
1273
1274         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1275         if (!dp)
1276                 return;
1277
1278         /* We're assuming the state code never drops its reference
1279          * without first removing the lease.  Since we're in this lease
1280          * callback (and since the lease code is serialized by the kernel
1281          * lock) we know the server hasn't removed the lease yet, we know
1282          * it's safe to take a reference: */
1283         atomic_inc(&dp->dl_count);
1284         atomic_inc(&dp->dl_client->cl_count);
1285
1286         spin_lock(&recall_lock);
1287         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
1288         spin_unlock(&recall_lock);
1289
1290         /* only place dl_time is set. protected by lock_kernel*/
1291         dp->dl_time = get_seconds();
1292
1293         /*
1294          * We don't want the locks code to timeout the lease for us;
1295          * we'll remove it ourself if the delegation isn't returned
1296          * in time.
1297          */
1298         fl->fl_break_time = 0;
1299
1300         t = kthread_run(do_recall, dp, "%s", "nfs4_cb_recall");
1301         if (IS_ERR(t)) {
1302                 struct nfs4_client *clp = dp->dl_client;
1303
1304                 printk(KERN_INFO "NFSD: Callback thread failed for "
1305                         "for client (clientid %08x/%08x)\n",
1306                         clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1307                 put_nfs4_client(dp->dl_client);
1308                 nfs4_put_delegation(dp);
1309         }
1310 }
1311
1312 /*
1313  * The file_lock is being reapd.
1314  *
1315  * Called by locks_free_lock() with lock_kernel() held.
1316  */
1317 static
1318 void nfsd_release_deleg_cb(struct file_lock *fl)
1319 {
1320         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1321
1322         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
1323
1324         if (!(fl->fl_flags & FL_LEASE) || !dp)
1325                 return;
1326         dp->dl_flock = NULL;
1327 }
1328
1329 /*
1330  * Set the delegation file_lock back pointer.
1331  *
1332  * Called from setlease() with lock_kernel() held.
1333  */
1334 static
1335 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
1336 {
1337         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
1338
1339         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
1340         if (!dp)
1341                 return;
1342         dp->dl_flock = new;
1343 }
1344
1345 /*
1346  * Called from setlease() with lock_kernel() held
1347  */
1348 static
1349 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
1350 {
1351         struct nfs4_delegation *onlistd =
1352                 (struct nfs4_delegation *)onlist->fl_owner;
1353         struct nfs4_delegation *tryd =
1354                 (struct nfs4_delegation *)try->fl_owner;
1355
1356         if (onlist->fl_lmops != try->fl_lmops)
1357                 return 0;
1358
1359         return onlistd->dl_client == tryd->dl_client;
1360 }
1361
1362
1363 static
1364 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
1365 {
1366         if (arg & F_UNLCK)
1367                 return lease_modify(onlist, arg);
1368         else
1369                 return -EAGAIN;
1370 }
1371
1372 static struct lock_manager_operations nfsd_lease_mng_ops = {
1373         .fl_break = nfsd_break_deleg_cb,
1374         .fl_release_private = nfsd_release_deleg_cb,
1375         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
1376         .fl_mylease = nfsd_same_client_deleg_cb,
1377         .fl_change = nfsd_change_deleg_cb,
1378 };
1379
1380
1381 __be32
1382 nfsd4_process_open1(struct nfsd4_open *open)
1383 {
1384         clientid_t *clientid = &open->op_clientid;
1385         struct nfs4_client *clp = NULL;
1386         unsigned int strhashval;
1387         struct nfs4_stateowner *sop = NULL;
1388
1389         if (!check_name(open->op_owner))
1390                 return nfserr_inval;
1391
1392         if (STALE_CLIENTID(&open->op_clientid))
1393                 return nfserr_stale_clientid;
1394
1395         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
1396         sop = find_openstateowner_str(strhashval, open);
1397         open->op_stateowner = sop;
1398         if (!sop) {
1399                 /* Make sure the client's lease hasn't expired. */
1400                 clp = find_confirmed_client(clientid);
1401                 if (clp == NULL)
1402                         return nfserr_expired;
1403                 goto renew;
1404         }
1405         if (!sop->so_confirmed) {
1406                 /* Replace unconfirmed owners without checking for replay. */
1407                 clp = sop->so_client;
1408                 release_stateowner(sop);
1409                 open->op_stateowner = NULL;
1410                 goto renew;
1411         }
1412         if (open->op_seqid == sop->so_seqid - 1) {
1413                 if (sop->so_replay.rp_buflen)
1414                         return nfserr_replay_me;
1415                 /* The original OPEN failed so spectacularly
1416                  * that we don't even have replay data saved!
1417                  * Therefore, we have no choice but to continue
1418                  * processing this OPEN; presumably, we'll
1419                  * fail again for the same reason.
1420                  */
1421                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
1422                 goto renew;
1423         }
1424         if (open->op_seqid != sop->so_seqid)
1425                 return nfserr_bad_seqid;
1426 renew:
1427         if (open->op_stateowner == NULL) {
1428                 sop = alloc_init_open_stateowner(strhashval, clp, open);
1429                 if (sop == NULL)
1430                         return nfserr_resource;
1431                 open->op_stateowner = sop;
1432         }
1433         list_del_init(&sop->so_close_lru);
1434         renew_client(sop->so_client);
1435         return nfs_ok;
1436 }
1437
1438 static inline __be32
1439 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
1440 {
1441         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
1442                 return nfserr_openmode;
1443         else
1444                 return nfs_ok;
1445 }
1446
1447 static struct nfs4_delegation *
1448 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
1449 {
1450         struct nfs4_delegation *dp;
1451
1452         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
1453                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
1454                         return dp;
1455         }
1456         return NULL;
1457 }
1458
1459 static __be32
1460 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
1461                 struct nfs4_delegation **dp)
1462 {
1463         int flags;
1464         __be32 status = nfserr_bad_stateid;
1465
1466         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
1467         if (*dp == NULL)
1468                 goto out;
1469         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
1470                                                 RD_STATE : WR_STATE;
1471         status = nfs4_check_delegmode(*dp, flags);
1472         if (status)
1473                 *dp = NULL;
1474 out:
1475         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
1476                 return nfs_ok;
1477         if (status)
1478                 return status;
1479         open->op_stateowner->so_confirmed = 1;
1480         return nfs_ok;
1481 }
1482
1483 static __be32
1484 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
1485 {
1486         struct nfs4_stateid *local;
1487         __be32 status = nfserr_share_denied;
1488         struct nfs4_stateowner *sop = open->op_stateowner;
1489
1490         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
1491                 /* ignore lock owners */
1492                 if (local->st_stateowner->so_is_open_owner == 0)
1493                         continue;
1494                 /* remember if we have seen this open owner */
1495                 if (local->st_stateowner == sop)
1496                         *stpp = local;
1497                 /* check for conflicting share reservations */
1498                 if (!test_share(local, open))
1499                         goto out;
1500         }
1501         status = 0;
1502 out:
1503         return status;
1504 }
1505
1506 static inline struct nfs4_stateid *
1507 nfs4_alloc_stateid(void)
1508 {
1509         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
1510 }
1511
1512 static __be32
1513 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
1514                 struct nfs4_delegation *dp,
1515                 struct svc_fh *cur_fh, int flags)
1516 {
1517         struct nfs4_stateid *stp;
1518
1519         stp = nfs4_alloc_stateid();
1520         if (stp == NULL)
1521                 return nfserr_resource;
1522
1523         if (dp) {
1524                 get_file(dp->dl_vfs_file);
1525                 stp->st_vfs_file = dp->dl_vfs_file;
1526         } else {
1527                 __be32 status;
1528                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
1529                                 &stp->st_vfs_file);
1530                 if (status) {
1531                         if (status == nfserr_dropit)
1532                                 status = nfserr_jukebox;
1533                         kmem_cache_free(stateid_slab, stp);
1534                         return status;
1535                 }
1536         }
1537         *stpp = stp;
1538         return 0;
1539 }
1540
1541 static inline __be32
1542 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
1543                 struct nfsd4_open *open)
1544 {
1545         struct iattr iattr = {
1546                 .ia_valid = ATTR_SIZE,
1547                 .ia_size = 0,
1548         };
1549         if (!open->op_truncate)
1550                 return 0;
1551         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
1552                 return nfserr_inval;
1553         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
1554 }
1555
1556 static __be32
1557 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
1558 {
1559         struct file *filp = stp->st_vfs_file;
1560         struct inode *inode = filp->f_path.dentry->d_inode;
1561         unsigned int share_access, new_writer;
1562         __be32 status;
1563
1564         set_access(&share_access, stp->st_access_bmap);
1565         new_writer = (~share_access) & open->op_share_access
1566                         & NFS4_SHARE_ACCESS_WRITE;
1567
1568         if (new_writer) {
1569                 int err = get_write_access(inode);
1570                 if (err)
1571                         return nfserrno(err);
1572         }
1573         status = nfsd4_truncate(rqstp, cur_fh, open);
1574         if (status) {
1575                 if (new_writer)
1576                         put_write_access(inode);
1577                 return status;
1578         }
1579         /* remember the open */
1580         filp->f_mode |= open->op_share_access;
1581         set_bit(open->op_share_access, &stp->st_access_bmap);
1582         set_bit(open->op_share_deny, &stp->st_deny_bmap);
1583
1584         return nfs_ok;
1585 }
1586
1587
1588 static void
1589 nfs4_set_claim_prev(struct nfsd4_open *open)
1590 {
1591         open->op_stateowner->so_confirmed = 1;
1592         open->op_stateowner->so_client->cl_firststate = 1;
1593 }
1594
1595 /*
1596  * Attempt to hand out a delegation.
1597  */
1598 static void
1599 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
1600 {
1601         struct nfs4_delegation *dp;
1602         struct nfs4_stateowner *sop = stp->st_stateowner;
1603         struct nfs4_callback *cb = &sop->so_client->cl_callback;
1604         struct file_lock fl, *flp = &fl;
1605         int status, flag = 0;
1606
1607         flag = NFS4_OPEN_DELEGATE_NONE;
1608         open->op_recall = 0;
1609         switch (open->op_claim_type) {
1610                 case NFS4_OPEN_CLAIM_PREVIOUS:
1611                         if (!atomic_read(&cb->cb_set))
1612                                 open->op_recall = 1;
1613                         flag = open->op_delegate_type;
1614                         if (flag == NFS4_OPEN_DELEGATE_NONE)
1615                                 goto out;
1616                         break;
1617                 case NFS4_OPEN_CLAIM_NULL:
1618                         /* Let's not give out any delegations till everyone's
1619                          * had the chance to reclaim theirs.... */
1620                         if (nfs4_in_grace())
1621                                 goto out;
1622                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
1623                                 goto out;
1624                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1625                                 flag = NFS4_OPEN_DELEGATE_WRITE;
1626                         else
1627                                 flag = NFS4_OPEN_DELEGATE_READ;
1628                         break;
1629                 default:
1630                         goto out;
1631         }
1632
1633         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
1634         if (dp == NULL) {
1635                 flag = NFS4_OPEN_DELEGATE_NONE;
1636                 goto out;
1637         }
1638         locks_init_lock(&fl);
1639         fl.fl_lmops = &nfsd_lease_mng_ops;
1640         fl.fl_flags = FL_LEASE;
1641         fl.fl_end = OFFSET_MAX;
1642         fl.fl_owner =  (fl_owner_t)dp;
1643         fl.fl_file = stp->st_vfs_file;
1644         fl.fl_pid = current->tgid;
1645
1646         /* vfs_setlease checks to see if delegation should be handed out.
1647          * the lock_manager callbacks fl_mylease and fl_change are used
1648          */
1649         if ((status = vfs_setlease(stp->st_vfs_file,
1650                 flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK, &flp))) {
1651                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
1652                 unhash_delegation(dp);
1653                 flag = NFS4_OPEN_DELEGATE_NONE;
1654                 goto out;
1655         }
1656
1657         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
1658
1659         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
1660                      dp->dl_stateid.si_boot,
1661                      dp->dl_stateid.si_stateownerid,
1662                      dp->dl_stateid.si_fileid,
1663                      dp->dl_stateid.si_generation);
1664 out:
1665         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
1666                         && flag == NFS4_OPEN_DELEGATE_NONE
1667                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
1668                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
1669         open->op_delegate_type = flag;
1670 }
1671
1672 /*
1673  * called with nfs4_lock_state() held.
1674  */
1675 __be32
1676 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
1677 {
1678         struct nfs4_file *fp = NULL;
1679         struct inode *ino = current_fh->fh_dentry->d_inode;
1680         struct nfs4_stateid *stp = NULL;
1681         struct nfs4_delegation *dp = NULL;
1682         __be32 status;
1683
1684         status = nfserr_inval;
1685         if (!access_valid(open->op_share_access)
1686                         || !deny_valid(open->op_share_deny))
1687                 goto out;
1688         /*
1689          * Lookup file; if found, lookup stateid and check open request,
1690          * and check for delegations in the process of being recalled.
1691          * If not found, create the nfs4_file struct
1692          */
1693         fp = find_file(ino);
1694         if (fp) {
1695                 if ((status = nfs4_check_open(fp, open, &stp)))
1696                         goto out;
1697                 status = nfs4_check_deleg(fp, open, &dp);
1698                 if (status)
1699                         goto out;
1700         } else {
1701                 status = nfserr_bad_stateid;
1702                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
1703                         goto out;
1704                 status = nfserr_resource;
1705                 fp = alloc_init_file(ino);
1706                 if (fp == NULL)
1707                         goto out;
1708         }
1709
1710         /*
1711          * OPEN the file, or upgrade an existing OPEN.
1712          * If truncate fails, the OPEN fails.
1713          */
1714         if (stp) {
1715                 /* Stateid was found, this is an OPEN upgrade */
1716                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
1717                 if (status)
1718                         goto out;
1719                 update_stateid(&stp->st_stateid);
1720         } else {
1721                 /* Stateid was not found, this is a new OPEN */
1722                 int flags = 0;
1723                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
1724                         flags |= MAY_READ;
1725                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
1726                         flags |= MAY_WRITE;
1727                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
1728                 if (status)
1729                         goto out;
1730                 init_stateid(stp, fp, open);
1731                 status = nfsd4_truncate(rqstp, current_fh, open);
1732                 if (status) {
1733                         release_stateid(stp, OPEN_STATE);
1734                         goto out;
1735                 }
1736         }
1737         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
1738
1739         /*
1740         * Attempt to hand out a delegation. No error return, because the
1741         * OPEN succeeds even if we fail.
1742         */
1743         nfs4_open_delegation(current_fh, open, stp);
1744
1745         status = nfs_ok;
1746
1747         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
1748                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
1749                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
1750 out:
1751         if (fp)
1752                 put_nfs4_file(fp);
1753         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
1754                 nfs4_set_claim_prev(open);
1755         /*
1756         * To finish the open response, we just need to set the rflags.
1757         */
1758         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
1759         if (!open->op_stateowner->so_confirmed)
1760                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
1761
1762         return status;
1763 }
1764
1765 static struct workqueue_struct *laundry_wq;
1766 static void laundromat_main(struct work_struct *);
1767 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
1768
1769 __be32
1770 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1771             clientid_t *clid)
1772 {
1773         struct nfs4_client *clp;
1774         __be32 status;
1775
1776         nfs4_lock_state();
1777         dprintk("process_renew(%08x/%08x): starting\n", 
1778                         clid->cl_boot, clid->cl_id);
1779         status = nfserr_stale_clientid;
1780         if (STALE_CLIENTID(clid))
1781                 goto out;
1782         clp = find_confirmed_client(clid);
1783         status = nfserr_expired;
1784         if (clp == NULL) {
1785                 /* We assume the client took too long to RENEW. */
1786                 dprintk("nfsd4_renew: clientid not found!\n");
1787                 goto out;
1788         }
1789         renew_client(clp);
1790         status = nfserr_cb_path_down;
1791         if (!list_empty(&clp->cl_delegations)
1792                         && !atomic_read(&clp->cl_callback.cb_set))
1793                 goto out;
1794         status = nfs_ok;
1795 out:
1796         nfs4_unlock_state();
1797         return status;
1798 }
1799
1800 static void
1801 end_grace(void)
1802 {
1803         dprintk("NFSD: end of grace period\n");
1804         nfsd4_recdir_purge_old();
1805         in_grace = 0;
1806 }
1807
1808 static time_t
1809 nfs4_laundromat(void)
1810 {
1811         struct nfs4_client *clp;
1812         struct nfs4_stateowner *sop;
1813         struct nfs4_delegation *dp;
1814         struct list_head *pos, *next, reaplist;
1815         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
1816         time_t t, clientid_val = NFSD_LEASE_TIME;
1817         time_t u, test_val = NFSD_LEASE_TIME;
1818
1819         nfs4_lock_state();
1820
1821         dprintk("NFSD: laundromat service - starting\n");
1822         if (in_grace)
1823                 end_grace();
1824         list_for_each_safe(pos, next, &client_lru) {
1825                 clp = list_entry(pos, struct nfs4_client, cl_lru);
1826                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
1827                         t = clp->cl_time - cutoff;
1828                         if (clientid_val > t)
1829                                 clientid_val = t;
1830                         break;
1831                 }
1832                 dprintk("NFSD: purging unused client (clientid %08x)\n",
1833                         clp->cl_clientid.cl_id);
1834                 nfsd4_remove_clid_dir(clp);
1835                 expire_client(clp);
1836         }
1837         INIT_LIST_HEAD(&reaplist);
1838         spin_lock(&recall_lock);
1839         list_for_each_safe(pos, next, &del_recall_lru) {
1840                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1841                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
1842                         u = dp->dl_time - cutoff;
1843                         if (test_val > u)
1844                                 test_val = u;
1845                         break;
1846                 }
1847                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
1848                                     dp, dp->dl_flock);
1849                 list_move(&dp->dl_recall_lru, &reaplist);
1850         }
1851         spin_unlock(&recall_lock);
1852         list_for_each_safe(pos, next, &reaplist) {
1853                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
1854                 list_del_init(&dp->dl_recall_lru);
1855                 unhash_delegation(dp);
1856         }
1857         test_val = NFSD_LEASE_TIME;
1858         list_for_each_safe(pos, next, &close_lru) {
1859                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
1860                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
1861                         u = sop->so_time - cutoff;
1862                         if (test_val > u)
1863                                 test_val = u;
1864                         break;
1865                 }
1866                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
1867                         sop->so_id);
1868                 release_stateowner(sop);
1869         }
1870         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
1871                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
1872         nfs4_unlock_state();
1873         return clientid_val;
1874 }
1875
1876 void
1877 laundromat_main(struct work_struct *not_used)
1878 {
1879         time_t t;
1880
1881         t = nfs4_laundromat();
1882         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
1883         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
1884 }
1885
1886 static struct nfs4_stateowner *
1887 search_close_lru(u32 st_id, int flags)
1888 {
1889         struct nfs4_stateowner *local = NULL;
1890
1891         if (flags & CLOSE_STATE) {
1892                 list_for_each_entry(local, &close_lru, so_close_lru) {
1893                         if (local->so_id == st_id)
1894                                 return local;
1895                 }
1896         }
1897         return NULL;
1898 }
1899
1900 static inline int
1901 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
1902 {
1903         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
1904 }
1905
1906 static int
1907 STALE_STATEID(stateid_t *stateid)
1908 {
1909         if (stateid->si_boot == boot_time)
1910                 return 0;
1911         dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
1912                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
1913                 stateid->si_generation);
1914         return 1;
1915 }
1916
1917 static inline int
1918 access_permit_read(unsigned long access_bmap)
1919 {
1920         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
1921                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
1922                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
1923 }
1924
1925 static inline int
1926 access_permit_write(unsigned long access_bmap)
1927 {
1928         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
1929                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
1930 }
1931
1932 static
1933 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
1934 {
1935         __be32 status = nfserr_openmode;
1936
1937         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
1938                 goto out;
1939         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
1940                 goto out;
1941         status = nfs_ok;
1942 out:
1943         return status;
1944 }
1945
1946 static inline __be32
1947 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
1948 {
1949         /* Trying to call delegreturn with a special stateid? Yuch: */
1950         if (!(flags & (RD_STATE | WR_STATE)))
1951                 return nfserr_bad_stateid;
1952         else if (ONE_STATEID(stateid) && (flags & RD_STATE))
1953                 return nfs_ok;
1954         else if (nfs4_in_grace()) {
1955                 /* Answer in remaining cases depends on existance of
1956                  * conflicting state; so we must wait out the grace period. */
1957                 return nfserr_grace;
1958         } else if (flags & WR_STATE)
1959                 return nfs4_share_conflict(current_fh,
1960                                 NFS4_SHARE_DENY_WRITE);
1961         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
1962                 return nfs4_share_conflict(current_fh,
1963                                 NFS4_SHARE_DENY_READ);
1964 }
1965
1966 /*
1967  * Allow READ/WRITE during grace period on recovered state only for files
1968  * that are not able to provide mandatory locking.
1969  */
1970 static inline int
1971 io_during_grace_disallowed(struct inode *inode, int flags)
1972 {
1973         return nfs4_in_grace() && (flags & (RD_STATE | WR_STATE))
1974                 && mandatory_lock(inode);
1975 }
1976
1977 /*
1978 * Checks for stateid operations
1979 */
1980 __be32
1981 nfs4_preprocess_stateid_op(struct svc_fh *current_fh, stateid_t *stateid, int flags, struct file **filpp)
1982 {
1983         struct nfs4_stateid *stp = NULL;
1984         struct nfs4_delegation *dp = NULL;
1985         stateid_t *stidp;
1986         struct inode *ino = current_fh->fh_dentry->d_inode;
1987         __be32 status;
1988
1989         dprintk("NFSD: preprocess_stateid_op: stateid = (%08x/%08x/%08x/%08x)\n",
1990                 stateid->si_boot, stateid->si_stateownerid, 
1991                 stateid->si_fileid, stateid->si_generation); 
1992         if (filpp)
1993                 *filpp = NULL;
1994
1995         if (io_during_grace_disallowed(ino, flags))
1996                 return nfserr_grace;
1997
1998         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
1999                 return check_special_stateids(current_fh, stateid, flags);
2000
2001         /* STALE STATEID */
2002         status = nfserr_stale_stateid;
2003         if (STALE_STATEID(stateid)) 
2004                 goto out;
2005
2006         /* BAD STATEID */
2007         status = nfserr_bad_stateid;
2008         if (!stateid->si_fileid) { /* delegation stateid */
2009                 if(!(dp = find_delegation_stateid(ino, stateid))) {
2010                         dprintk("NFSD: delegation stateid not found\n");
2011                         goto out;
2012                 }
2013                 stidp = &dp->dl_stateid;
2014         } else { /* open or lock stateid */
2015                 if (!(stp = find_stateid(stateid, flags))) {
2016                         dprintk("NFSD: open or lock stateid not found\n");
2017                         goto out;
2018                 }
2019                 if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp))
2020                         goto out;
2021                 if (!stp->st_stateowner->so_confirmed)
2022                         goto out;
2023                 stidp = &stp->st_stateid;
2024         }
2025         if (stateid->si_generation > stidp->si_generation)
2026                 goto out;
2027
2028         /* OLD STATEID */
2029         status = nfserr_old_stateid;
2030         if (stateid->si_generation < stidp->si_generation)
2031                 goto out;
2032         if (stp) {
2033                 if ((status = nfs4_check_openmode(stp,flags)))
2034                         goto out;
2035                 renew_client(stp->st_stateowner->so_client);
2036                 if (filpp)
2037                         *filpp = stp->st_vfs_file;
2038         } else if (dp) {
2039                 if ((status = nfs4_check_delegmode(dp, flags)))
2040                         goto out;
2041                 renew_client(dp->dl_client);
2042                 if (flags & DELEG_RET)
2043                         unhash_delegation(dp);
2044                 if (filpp)
2045                         *filpp = dp->dl_vfs_file;
2046         }
2047         status = nfs_ok;
2048 out:
2049         return status;
2050 }
2051
2052 static inline int
2053 setlkflg (int type)
2054 {
2055         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2056                 RD_STATE : WR_STATE;
2057 }
2058
2059 /* 
2060  * Checks for sequence id mutating operations. 
2061  */
2062 static __be32
2063 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)
2064 {
2065         struct nfs4_stateid *stp;
2066         struct nfs4_stateowner *sop;
2067
2068         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2069                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2070                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2071                 stateid->si_generation);
2072
2073         *stpp = NULL;
2074         *sopp = NULL;
2075
2076         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2077                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2078                 return nfserr_bad_stateid;
2079         }
2080
2081         if (STALE_STATEID(stateid))
2082                 return nfserr_stale_stateid;
2083         /*
2084         * We return BAD_STATEID if filehandle doesn't match stateid, 
2085         * the confirmed flag is incorrecly set, or the generation 
2086         * number is incorrect.  
2087         */
2088         stp = find_stateid(stateid, flags);
2089         if (stp == NULL) {
2090                 /*
2091                  * Also, we should make sure this isn't just the result of
2092                  * a replayed close:
2093                  */
2094                 sop = search_close_lru(stateid->si_stateownerid, flags);
2095                 if (sop == NULL)
2096                         return nfserr_bad_stateid;
2097                 *sopp = sop;
2098                 goto check_replay;
2099         }
2100
2101         *stpp = stp;
2102         *sopp = sop = stp->st_stateowner;
2103
2104         if (lock) {
2105                 clientid_t *lockclid = &lock->v.new.clientid;
2106                 struct nfs4_client *clp = sop->so_client;
2107                 int lkflg = 0;
2108                 __be32 status;
2109
2110                 lkflg = setlkflg(lock->lk_type);
2111
2112                 if (lock->lk_is_new) {
2113                         if (!sop->so_is_open_owner)
2114                                 return nfserr_bad_stateid;
2115                         if (!same_clid(&clp->cl_clientid, lockclid))
2116                                return nfserr_bad_stateid;
2117                         /* stp is the open stateid */
2118                         status = nfs4_check_openmode(stp, lkflg);
2119                         if (status)
2120                                 return status;
2121                 } else {
2122                         /* stp is the lock stateid */
2123                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2124                         if (status)
2125                                 return status;
2126                }
2127         }
2128
2129         if ((flags & CHECK_FH) && nfs4_check_fh(current_fh, stp)) {
2130                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2131                 return nfserr_bad_stateid;
2132         }
2133
2134         /*
2135         *  We now validate the seqid and stateid generation numbers.
2136         *  For the moment, we ignore the possibility of 
2137         *  generation number wraparound.
2138         */
2139         if (seqid != sop->so_seqid)
2140                 goto check_replay;
2141
2142         if (sop->so_confirmed && flags & CONFIRM) {
2143                 dprintk("NFSD: preprocess_seqid_op: expected"
2144                                 " unconfirmed stateowner!\n");
2145                 return nfserr_bad_stateid;
2146         }
2147         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2148                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2149                                 " confirmed yet!\n");
2150                 return nfserr_bad_stateid;
2151         }
2152         if (stateid->si_generation > stp->st_stateid.si_generation) {
2153                 dprintk("NFSD: preprocess_seqid_op: future stateid?!\n");
2154                 return nfserr_bad_stateid;
2155         }
2156
2157         if (stateid->si_generation < stp->st_stateid.si_generation) {
2158                 dprintk("NFSD: preprocess_seqid_op: old stateid!\n");
2159                 return nfserr_old_stateid;
2160         }
2161         renew_client(sop->so_client);
2162         return nfs_ok;
2163
2164 check_replay:
2165         if (seqid == sop->so_seqid - 1) {
2166                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2167                 /* indicate replay to calling function */
2168                 return nfserr_replay_me;
2169         }
2170         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2171                         sop->so_seqid, seqid);
2172         *sopp = NULL;
2173         return nfserr_bad_seqid;
2174 }
2175
2176 __be32
2177 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2178                    struct nfsd4_open_confirm *oc)
2179 {
2180         __be32 status;
2181         struct nfs4_stateowner *sop;
2182         struct nfs4_stateid *stp;
2183
2184         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2185                         (int)cstate->current_fh.fh_dentry->d_name.len,
2186                         cstate->current_fh.fh_dentry->d_name.name);
2187
2188         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2189         if (status)
2190                 return status;
2191
2192         nfs4_lock_state();
2193
2194         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2195                                         oc->oc_seqid, &oc->oc_req_stateid,
2196                                         CHECK_FH | CONFIRM | OPEN_STATE,
2197                                         &oc->oc_stateowner, &stp, NULL)))
2198                 goto out; 
2199
2200         sop = oc->oc_stateowner;
2201         sop->so_confirmed = 1;
2202         update_stateid(&stp->st_stateid);
2203         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2204         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
2205                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
2206                          stp->st_stateid.si_boot,
2207                          stp->st_stateid.si_stateownerid,
2208                          stp->st_stateid.si_fileid,
2209                          stp->st_stateid.si_generation);
2210
2211         nfsd4_create_clid_dir(sop->so_client);
2212 out:
2213         if (oc->oc_stateowner) {
2214                 nfs4_get_stateowner(oc->oc_stateowner);
2215                 cstate->replay_owner = oc->oc_stateowner;
2216         }
2217         nfs4_unlock_state();
2218         return status;
2219 }
2220
2221
2222 /*
2223  * unset all bits in union bitmap (bmap) that
2224  * do not exist in share (from successful OPEN_DOWNGRADE)
2225  */
2226 static void
2227 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
2228 {
2229         int i;
2230         for (i = 1; i < 4; i++) {
2231                 if ((i & access) != i)
2232                         __clear_bit(i, bmap);
2233         }
2234 }
2235
2236 static void
2237 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
2238 {
2239         int i;
2240         for (i = 0; i < 4; i++) {
2241                 if ((i & deny) != i)
2242                         __clear_bit(i, bmap);
2243         }
2244 }
2245
2246 __be32
2247 nfsd4_open_downgrade(struct svc_rqst *rqstp,
2248                      struct nfsd4_compound_state *cstate,
2249                      struct nfsd4_open_downgrade *od)
2250 {
2251         __be32 status;
2252         struct nfs4_stateid *stp;
2253         unsigned int share_access;
2254
2255         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
2256                         (int)cstate->current_fh.fh_dentry->d_name.len,
2257                         cstate->current_fh.fh_dentry->d_name.name);
2258
2259         if (!access_valid(od->od_share_access)
2260                         || !deny_valid(od->od_share_deny))
2261                 return nfserr_inval;
2262
2263         nfs4_lock_state();
2264         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2265                                         od->od_seqid,
2266                                         &od->od_stateid, 
2267                                         CHECK_FH | OPEN_STATE, 
2268                                         &od->od_stateowner, &stp, NULL)))
2269                 goto out; 
2270
2271         status = nfserr_inval;
2272         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
2273                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
2274                         stp->st_access_bmap, od->od_share_access);
2275                 goto out;
2276         }
2277         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
2278                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
2279                         stp->st_deny_bmap, od->od_share_deny);
2280                 goto out;
2281         }
2282         set_access(&share_access, stp->st_access_bmap);
2283         nfs4_file_downgrade(stp->st_vfs_file,
2284                             share_access & ~od->od_share_access);
2285
2286         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
2287         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
2288
2289         update_stateid(&stp->st_stateid);
2290         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
2291         status = nfs_ok;
2292 out:
2293         if (od->od_stateowner) {
2294                 nfs4_get_stateowner(od->od_stateowner);
2295                 cstate->replay_owner = od->od_stateowner;
2296         }
2297         nfs4_unlock_state();
2298         return status;
2299 }
2300
2301 /*
2302  * nfs4_unlock_state() called after encode
2303  */
2304 __be32
2305 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2306             struct nfsd4_close *close)
2307 {
2308         __be32 status;
2309         struct nfs4_stateid *stp;
2310
2311         dprintk("NFSD: nfsd4_close on file %.*s\n", 
2312                         (int)cstate->current_fh.fh_dentry->d_name.len,
2313                         cstate->current_fh.fh_dentry->d_name.name);
2314
2315         nfs4_lock_state();
2316         /* check close_lru for replay */
2317         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2318                                         close->cl_seqid,
2319                                         &close->cl_stateid, 
2320                                         CHECK_FH | OPEN_STATE | CLOSE_STATE,
2321                                         &close->cl_stateowner, &stp, NULL)))
2322                 goto out; 
2323         status = nfs_ok;
2324         update_stateid(&stp->st_stateid);
2325         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
2326
2327         /* release_stateid() calls nfsd_close() if needed */
2328         release_stateid(stp, OPEN_STATE);
2329
2330         /* place unused nfs4_stateowners on so_close_lru list to be
2331          * released by the laundromat service after the lease period
2332          * to enable us to handle CLOSE replay
2333          */
2334         if (list_empty(&close->cl_stateowner->so_stateids))
2335                 move_to_close_lru(close->cl_stateowner);
2336 out:
2337         if (close->cl_stateowner) {
2338                 nfs4_get_stateowner(close->cl_stateowner);
2339                 cstate->replay_owner = close->cl_stateowner;
2340         }
2341         nfs4_unlock_state();
2342         return status;
2343 }
2344
2345 __be32
2346 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2347                   struct nfsd4_delegreturn *dr)
2348 {
2349         __be32 status;
2350
2351         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
2352                 goto out;
2353
2354         nfs4_lock_state();
2355         status = nfs4_preprocess_stateid_op(&cstate->current_fh,
2356                                             &dr->dr_stateid, DELEG_RET, NULL);
2357         nfs4_unlock_state();
2358 out:
2359         return status;
2360 }
2361
2362
2363 /* 
2364  * Lock owner state (byte-range locks)
2365  */
2366 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
2367 #define LOCK_HASH_BITS              8
2368 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
2369 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
2370
2371 #define lockownerid_hashval(id) \
2372         ((id) & LOCK_HASH_MASK)
2373
2374 static inline unsigned int
2375 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
2376                 struct xdr_netobj *ownername)
2377 {
2378         return (file_hashval(inode) + cl_id
2379                         + opaque_hashval(ownername->data, ownername->len))
2380                 & LOCK_HASH_MASK;
2381 }
2382
2383 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
2384 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
2385 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
2386
2387 static struct nfs4_stateid *
2388 find_stateid(stateid_t *stid, int flags)
2389 {
2390         struct nfs4_stateid *local = NULL;
2391         u32 st_id = stid->si_stateownerid;
2392         u32 f_id = stid->si_fileid;
2393         unsigned int hashval;
2394
2395         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
2396         if ((flags & LOCK_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2397                 hashval = stateid_hashval(st_id, f_id);
2398                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
2399                         if ((local->st_stateid.si_stateownerid == st_id) &&
2400                             (local->st_stateid.si_fileid == f_id))
2401                                 return local;
2402                 }
2403         } 
2404         if ((flags & OPEN_STATE) || (flags & RD_STATE) || (flags & WR_STATE)) {
2405                 hashval = stateid_hashval(st_id, f_id);
2406                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
2407                         if ((local->st_stateid.si_stateownerid == st_id) &&
2408                             (local->st_stateid.si_fileid == f_id))
2409                                 return local;
2410                 }
2411         }
2412         return NULL;
2413 }
2414
2415 static struct nfs4_delegation *
2416 find_delegation_stateid(struct inode *ino, stateid_t *stid)
2417 {
2418         struct nfs4_file *fp;
2419         struct nfs4_delegation *dl;
2420
2421         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
2422                     stid->si_boot, stid->si_stateownerid,
2423                     stid->si_fileid, stid->si_generation);
2424
2425         fp = find_file(ino);
2426         if (!fp)
2427                 return NULL;
2428         dl = find_delegation_file(fp, stid);
2429         put_nfs4_file(fp);
2430         return dl;
2431 }
2432
2433 /*
2434  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
2435  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
2436  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
2437  * locking, this prevents us from being completely protocol-compliant.  The
2438  * real solution to this problem is to start using unsigned file offsets in
2439  * the VFS, but this is a very deep change!
2440  */
2441 static inline void
2442 nfs4_transform_lock_offset(struct file_lock *lock)
2443 {
2444         if (lock->fl_start < 0)
2445                 lock->fl_start = OFFSET_MAX;
2446         if (lock->fl_end < 0)
2447                 lock->fl_end = OFFSET_MAX;
2448 }
2449
2450 /* Hack!: For now, we're defining this just so we can use a pointer to it
2451  * as a unique cookie to identify our (NFSv4's) posix locks. */
2452 static struct lock_manager_operations nfsd_posix_mng_ops  = {
2453 };
2454
2455 static inline void
2456 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
2457 {
2458         struct nfs4_stateowner *sop;
2459         unsigned int hval;
2460
2461         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
2462                 sop = (struct nfs4_stateowner *) fl->fl_owner;
2463                 hval = lockownerid_hashval(sop->so_id);
2464                 kref_get(&sop->so_ref);
2465                 deny->ld_sop = sop;
2466                 deny->ld_clientid = sop->so_client->cl_clientid;
2467         } else {
2468                 deny->ld_sop = NULL;
2469                 deny->ld_clientid.cl_boot = 0;
2470                 deny->ld_clientid.cl_id = 0;
2471         }
2472         deny->ld_start = fl->fl_start;
2473         deny->ld_length = ~(u64)0;
2474         if (fl->fl_end != ~(u64)0)
2475                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
2476         deny->ld_type = NFS4_READ_LT;
2477         if (fl->fl_type != F_RDLCK)
2478                 deny->ld_type = NFS4_WRITE_LT;
2479 }
2480
2481 static struct nfs4_stateowner *
2482 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
2483                 struct xdr_netobj *owner)
2484 {
2485         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
2486         struct nfs4_stateowner *op;
2487
2488         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
2489                 if (same_owner_str(op, owner, clid))
2490                         return op;
2491         }
2492         return NULL;
2493 }
2494
2495 /*
2496  * Alloc a lock owner structure.
2497  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
2498  * occured. 
2499  *
2500  * strhashval = lock_ownerstr_hashval 
2501  */
2502
2503 static struct nfs4_stateowner *
2504 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
2505         struct nfs4_stateowner *sop;
2506         struct nfs4_replay *rp;
2507         unsigned int idhashval;
2508
2509         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
2510                 return NULL;
2511         idhashval = lockownerid_hashval(current_ownerid);
2512         INIT_LIST_HEAD(&sop->so_idhash);
2513         INIT_LIST_HEAD(&sop->so_strhash);
2514         INIT_LIST_HEAD(&sop->so_perclient);
2515         INIT_LIST_HEAD(&sop->so_stateids);
2516         INIT_LIST_HEAD(&sop->so_perstateid);
2517         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
2518         sop->so_time = 0;
2519         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
2520         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
2521         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
2522         sop->so_is_open_owner = 0;
2523         sop->so_id = current_ownerid++;
2524         sop->so_client = clp;
2525         /* It is the openowner seqid that will be incremented in encode in the
2526          * case of new lockowners; so increment the lock seqid manually: */
2527         sop->so_seqid = lock->lk_new_lock_seqid + 1;
2528         sop->so_confirmed = 1;
2529         rp = &sop->so_replay;
2530         rp->rp_status = nfserr_serverfault;
2531         rp->rp_buflen = 0;
2532         rp->rp_buf = rp->rp_ibuf;
2533         return sop;
2534 }
2535
2536 static struct nfs4_stateid *
2537 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
2538 {
2539         struct nfs4_stateid *stp;
2540         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2541
2542         stp = nfs4_alloc_stateid();
2543         if (stp == NULL)
2544                 goto out;
2545         INIT_LIST_HEAD(&stp->st_hash);
2546         INIT_LIST_HEAD(&stp->st_perfile);
2547         INIT_LIST_HEAD(&stp->st_perstateowner);
2548         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
2549         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
2550         list_add(&stp->st_perfile, &fp->fi_stateids);
2551         list_add(&stp->st_perstateowner, &sop->so_stateids);
2552         stp->st_stateowner = sop;
2553         get_nfs4_file(fp);
2554         stp->st_file = fp;
2555         stp->st_stateid.si_boot = boot_time;
2556         stp->st_stateid.si_stateownerid = sop->so_id;
2557         stp->st_stateid.si_fileid = fp->fi_id;
2558         stp->st_stateid.si_generation = 0;
2559         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
2560         stp->st_access_bmap = open_stp->st_access_bmap;
2561         stp->st_deny_bmap = open_stp->st_deny_bmap;
2562         stp->st_openstp = open_stp;
2563
2564 out:
2565         return stp;
2566 }
2567
2568 static int
2569 check_lock_length(u64 offset, u64 length)
2570 {
2571         return ((length == 0)  || ((length != ~(u64)0) &&
2572              LOFF_OVERFLOW(offset, length)));
2573 }
2574
2575 /*
2576  *  LOCK operation 
2577  */
2578 __be32
2579 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2580            struct nfsd4_lock *lock)
2581 {
2582         struct nfs4_stateowner *open_sop = NULL;
2583         struct nfs4_stateowner *lock_sop = NULL;
2584         struct nfs4_stateid *lock_stp;
2585         struct file *filp;
2586         struct file_lock file_lock;
2587         struct file_lock conflock;
2588         __be32 status = 0;
2589         unsigned int strhashval;
2590         unsigned int cmd;
2591         int err;
2592
2593         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
2594                 (long long) lock->lk_offset,
2595                 (long long) lock->lk_length);
2596
2597         if (check_lock_length(lock->lk_offset, lock->lk_length))
2598                  return nfserr_inval;
2599
2600         if ((status = fh_verify(rqstp, &cstate->current_fh,
2601                                 S_IFREG, MAY_LOCK))) {
2602                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
2603                 return status;
2604         }
2605
2606         nfs4_lock_state();
2607
2608         if (lock->lk_is_new) {
2609                 /*
2610                  * Client indicates that this is a new lockowner.
2611                  * Use open owner and open stateid to create lock owner and
2612                  * lock stateid.
2613                  */
2614                 struct nfs4_stateid *open_stp = NULL;
2615                 struct nfs4_file *fp;
2616                 
2617                 status = nfserr_stale_clientid;
2618                 if (STALE_CLIENTID(&lock->lk_new_clientid))
2619                         goto out;
2620
2621                 /* validate and update open stateid and open seqid */
2622                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2623                                         lock->lk_new_open_seqid,
2624                                         &lock->lk_new_open_stateid,
2625                                         CHECK_FH | OPEN_STATE,
2626                                         &lock->lk_replay_owner, &open_stp,
2627                                         lock);
2628                 if (status)
2629                         goto out;
2630                 open_sop = lock->lk_replay_owner;
2631                 /* create lockowner and lock stateid */
2632                 fp = open_stp->st_file;
2633                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
2634                                 open_sop->so_client->cl_clientid.cl_id, 
2635                                 &lock->v.new.owner);
2636                 /* XXX: Do we need to check for duplicate stateowners on
2637                  * the same file, or should they just be allowed (and
2638                  * create new stateids)? */
2639                 status = nfserr_resource;
2640                 lock_sop = alloc_init_lock_stateowner(strhashval,
2641                                 open_sop->so_client, open_stp, lock);
2642                 if (lock_sop == NULL)
2643                         goto out;
2644                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
2645                 if (lock_stp == NULL)
2646                         goto out;
2647         } else {
2648                 /* lock (lock owner + lock stateid) already exists */
2649                 status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2650                                        lock->lk_old_lock_seqid, 
2651                                        &lock->lk_old_lock_stateid, 
2652                                        CHECK_FH | LOCK_STATE, 
2653                                        &lock->lk_replay_owner, &lock_stp, lock);
2654                 if (status)
2655                         goto out;
2656                 lock_sop = lock->lk_replay_owner;
2657         }
2658         /* lock->lk_replay_owner and lock_stp have been created or found */
2659         filp = lock_stp->st_vfs_file;
2660
2661         status = nfserr_grace;
2662         if (nfs4_in_grace() && !lock->lk_reclaim)
2663                 goto out;
2664         status = nfserr_no_grace;
2665         if (!nfs4_in_grace() && lock->lk_reclaim)
2666                 goto out;
2667
2668         locks_init_lock(&file_lock);
2669         switch (lock->lk_type) {
2670                 case NFS4_READ_LT:
2671                 case NFS4_READW_LT:
2672                         file_lock.fl_type = F_RDLCK;
2673                         cmd = F_SETLK;
2674                 break;
2675                 case NFS4_WRITE_LT:
2676                 case NFS4_WRITEW_LT:
2677                         file_lock.fl_type = F_WRLCK;
2678                         cmd = F_SETLK;
2679                 break;
2680                 default:
2681                         status = nfserr_inval;
2682                 goto out;
2683         }
2684         file_lock.fl_owner = (fl_owner_t)lock_sop;
2685         file_lock.fl_pid = current->tgid;
2686         file_lock.fl_file = filp;
2687         file_lock.fl_flags = FL_POSIX;
2688         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2689
2690         file_lock.fl_start = lock->lk_offset;
2691         if ((lock->lk_length == ~(u64)0) || 
2692                         LOFF_OVERFLOW(lock->lk_offset, lock->lk_length))
2693                 file_lock.fl_end = ~(u64)0;
2694         else
2695                 file_lock.fl_end = lock->lk_offset + lock->lk_length - 1;
2696         nfs4_transform_lock_offset(&file_lock);
2697
2698         /*
2699         * Try to lock the file in the VFS.
2700         * Note: locks.c uses the BKL to protect the inode's lock list.
2701         */
2702
2703         /* XXX?: Just to divert the locks_release_private at the start of
2704          * locks_copy_lock: */
2705         locks_init_lock(&conflock);
2706         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
2707         switch (-err) {
2708         case 0: /* success! */
2709                 update_stateid(&lock_stp->st_stateid);
2710                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
2711                                 sizeof(stateid_t));
2712                 status = 0;
2713                 break;
2714         case (EAGAIN):          /* conflock holds conflicting lock */
2715                 status = nfserr_denied;
2716                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
2717                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
2718                 break;
2719         case (EDEADLK):
2720                 status = nfserr_deadlock;
2721                 break;
2722         default:        
2723                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
2724                 status = nfserr_resource;
2725                 break;
2726         }
2727 out:
2728         if (status && lock->lk_is_new && lock_sop)
2729                 release_stateowner(lock_sop);
2730         if (lock->lk_replay_owner) {
2731                 nfs4_get_stateowner(lock->lk_replay_owner);
2732                 cstate->replay_owner = lock->lk_replay_owner;
2733         }
2734         nfs4_unlock_state();
2735         return status;
2736 }
2737
2738 /*
2739  * LOCKT operation
2740  */
2741 __be32
2742 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2743             struct nfsd4_lockt *lockt)
2744 {
2745         struct inode *inode;
2746         struct file file;
2747         struct file_lock file_lock;
2748         int error;
2749         __be32 status;
2750
2751         if (nfs4_in_grace())
2752                 return nfserr_grace;
2753
2754         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
2755                  return nfserr_inval;
2756
2757         lockt->lt_stateowner = NULL;
2758         nfs4_lock_state();
2759
2760         status = nfserr_stale_clientid;
2761         if (STALE_CLIENTID(&lockt->lt_clientid))
2762                 goto out;
2763
2764         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
2765                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
2766                 if (status == nfserr_symlink)
2767                         status = nfserr_inval;
2768                 goto out;
2769         }
2770
2771         inode = cstate->current_fh.fh_dentry->d_inode;
2772         locks_init_lock(&file_lock);
2773         switch (lockt->lt_type) {
2774                 case NFS4_READ_LT:
2775                 case NFS4_READW_LT:
2776                         file_lock.fl_type = F_RDLCK;
2777                 break;
2778                 case NFS4_WRITE_LT:
2779                 case NFS4_WRITEW_LT:
2780                         file_lock.fl_type = F_WRLCK;
2781                 break;
2782                 default:
2783                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
2784                         status = nfserr_inval;
2785                 goto out;
2786         }
2787
2788         lockt->lt_stateowner = find_lockstateowner_str(inode,
2789                         &lockt->lt_clientid, &lockt->lt_owner);
2790         if (lockt->lt_stateowner)
2791                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
2792         file_lock.fl_pid = current->tgid;
2793         file_lock.fl_flags = FL_POSIX;
2794         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2795
2796         file_lock.fl_start = lockt->lt_offset;
2797         if ((lockt->lt_length == ~(u64)0) || LOFF_OVERFLOW(lockt->lt_offset, lockt->lt_length))
2798                 file_lock.fl_end = ~(u64)0;
2799         else
2800                 file_lock.fl_end = lockt->lt_offset + lockt->lt_length - 1;
2801
2802         nfs4_transform_lock_offset(&file_lock);
2803
2804         /* vfs_test_lock uses the struct file _only_ to resolve the inode.
2805          * since LOCKT doesn't require an OPEN, and therefore a struct
2806          * file may not exist, pass vfs_test_lock a struct file with
2807          * only the dentry:inode set.
2808          */
2809         memset(&file, 0, sizeof (struct file));
2810         file.f_path.dentry = cstate->current_fh.fh_dentry;
2811
2812         status = nfs_ok;
2813         error = vfs_test_lock(&file, &file_lock);
2814         if (error) {
2815                 status = nfserrno(error);
2816                 goto out;
2817         }
2818         if (file_lock.fl_type != F_UNLCK) {
2819                 status = nfserr_denied;
2820                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
2821         }
2822 out:
2823         nfs4_unlock_state();
2824         return status;
2825 }
2826
2827 __be32
2828 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2829             struct nfsd4_locku *locku)
2830 {
2831         struct nfs4_stateid *stp;
2832         struct file *filp = NULL;
2833         struct file_lock file_lock;
2834         __be32 status;
2835         int err;
2836                                                         
2837         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
2838                 (long long) locku->lu_offset,
2839                 (long long) locku->lu_length);
2840
2841         if (check_lock_length(locku->lu_offset, locku->lu_length))
2842                  return nfserr_inval;
2843
2844         nfs4_lock_state();
2845                                                                                 
2846         if ((status = nfs4_preprocess_seqid_op(&cstate->current_fh,
2847                                         locku->lu_seqid, 
2848                                         &locku->lu_stateid, 
2849                                         CHECK_FH | LOCK_STATE, 
2850                                         &locku->lu_stateowner, &stp, NULL)))
2851                 goto out;
2852
2853         filp = stp->st_vfs_file;
2854         BUG_ON(!filp);
2855         locks_init_lock(&file_lock);
2856         file_lock.fl_type = F_UNLCK;
2857         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
2858         file_lock.fl_pid = current->tgid;
2859         file_lock.fl_file = filp;
2860         file_lock.fl_flags = FL_POSIX; 
2861         file_lock.fl_lmops = &nfsd_posix_mng_ops;
2862         file_lock.fl_start = locku->lu_offset;
2863
2864         if ((locku->lu_length == ~(u64)0) || LOFF_OVERFLOW(locku->lu_offset, locku->lu_length))
2865                 file_lock.fl_end = ~(u64)0;
2866         else
2867                 file_lock.fl_end = locku->lu_offset + locku->lu_length - 1;
2868         nfs4_transform_lock_offset(&file_lock);
2869
2870         /*
2871         *  Try to unlock the file in the VFS.
2872         */
2873         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
2874         if (err) {
2875                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
2876                 goto out_nfserr;
2877         }
2878         /*
2879         * OK, unlock succeeded; the only thing left to do is update the stateid.
2880         */
2881         update_stateid(&stp->st_stateid);
2882         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
2883
2884 out:
2885         if (locku->lu_stateowner) {
2886                 nfs4_get_stateowner(locku->lu_stateowner);
2887                 cstate->replay_owner = locku->lu_stateowner;
2888         }
2889         nfs4_unlock_state();
2890         return status;
2891
2892 out_nfserr:
2893         status = nfserrno(err);
2894         goto out;
2895 }
2896
2897 /*
2898  * returns
2899  *      1: locks held by lockowner
2900  *      0: no locks held by lockowner
2901  */
2902 static int
2903 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
2904 {
2905         struct file_lock **flpp;
2906         struct inode *inode = filp->f_path.dentry->d_inode;
2907         int status = 0;
2908
2909         lock_kernel();
2910         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
2911                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
2912                         status = 1;
2913                         goto out;
2914                 }
2915         }
2916 out:
2917         unlock_kernel();
2918         return status;
2919 }
2920
2921 __be32
2922 nfsd4_release_lockowner(struct svc_rqst *rqstp,
2923                         struct nfsd4_compound_state *cstate,
2924                         struct nfsd4_release_lockowner *rlockowner)
2925 {
2926         clientid_t *clid = &rlockowner->rl_clientid;
2927         struct nfs4_stateowner *sop;
2928         struct nfs4_stateid *stp;
2929         struct xdr_netobj *owner = &rlockowner->rl_owner;
2930         struct list_head matches;
2931         int i;
2932         __be32 status;
2933
2934         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
2935                 clid->cl_boot, clid->cl_id);
2936
2937         /* XXX check for lease expiration */
2938
2939         status = nfserr_stale_clientid;
2940         if (STALE_CLIENTID(clid))
2941                 return status;
2942
2943         nfs4_lock_state();
2944
2945         status = nfserr_locks_held;
2946         /* XXX: we're doing a linear search through all the lockowners.
2947          * Yipes!  For now we'll just hope clients aren't really using
2948          * release_lockowner much, but eventually we have to fix these
2949          * data structures. */
2950         INIT_LIST_HEAD(&matches);
2951         for (i = 0; i < LOCK_HASH_SIZE; i++) {
2952                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
2953                         if (!same_owner_str(sop, owner, clid))
2954                                 continue;
2955                         list_for_each_entry(stp, &sop->so_stateids,
2956                                         st_perstateowner) {
2957                                 if (check_for_locks(stp->st_vfs_file, sop))
2958                                         goto out;
2959                                 /* Note: so_perclient unused for lockowners,
2960                                  * so it's OK to fool with here. */
2961                                 list_add(&sop->so_perclient, &matches);
2962                         }
2963                 }
2964         }
2965         /* Clients probably won't expect us to return with some (but not all)
2966          * of the lockowner state released; so don't release any until all
2967          * have been checked. */
2968         status = nfs_ok;
2969         while (!list_empty(&matches)) {
2970                 sop = list_entry(matches.next, struct nfs4_stateowner,
2971                                                                 so_perclient);
2972                 /* unhash_stateowner deletes so_perclient only
2973                  * for openowners. */
2974                 list_del(&sop->so_perclient);
2975                 release_stateowner(sop);
2976         }
2977 out:
2978         nfs4_unlock_state();
2979         return status;
2980 }
2981
2982 static inline struct nfs4_client_reclaim *
2983 alloc_reclaim(void)
2984 {
2985         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
2986 }
2987
2988 int
2989 nfs4_has_reclaimed_state(const char *name)
2990 {
2991         unsigned int strhashval = clientstr_hashval(name);
2992         struct nfs4_client *clp;
2993
2994         clp = find_confirmed_client_by_str(name, strhashval);
2995         return clp ? 1 : 0;
2996 }
2997
2998 /*
2999  * failure => all reset bets are off, nfserr_no_grace...
3000  */
3001 int
3002 nfs4_client_to_reclaim(const char *name)
3003 {
3004         unsigned int strhashval;
3005         struct nfs4_client_reclaim *crp = NULL;
3006
3007         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3008         crp = alloc_reclaim();
3009         if (!crp)
3010                 return 0;
3011         strhashval = clientstr_hashval(name);
3012         INIT_LIST_HEAD(&crp->cr_strhash);
3013         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3014         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3015         reclaim_str_hashtbl_size++;
3016         return 1;
3017 }
3018
3019 static void
3020 nfs4_release_reclaim(void)
3021 {
3022         struct nfs4_client_reclaim *crp = NULL;
3023         int i;
3024
3025         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3026                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3027                         crp = list_entry(reclaim_str_hashtbl[i].next,
3028                                         struct nfs4_client_reclaim, cr_strhash);
3029                         list_del(&crp->cr_strhash);
3030                         kfree(crp);
3031                         reclaim_str_hashtbl_size--;
3032                 }
3033         }
3034         BUG_ON(reclaim_str_hashtbl_size);
3035 }
3036
3037 /*
3038  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3039 static struct nfs4_client_reclaim *
3040 nfs4_find_reclaim_client(clientid_t *clid)
3041 {
3042         unsigned int strhashval;
3043         struct nfs4_client *clp;
3044         struct nfs4_client_reclaim *crp = NULL;
3045
3046
3047         /* find clientid in conf_id_hashtbl */
3048         clp = find_confirmed_client(clid);
3049         if (clp == NULL)
3050                 return NULL;
3051
3052         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3053                             clp->cl_name.len, clp->cl_name.data,
3054                             clp->cl_recdir);
3055
3056         /* find clp->cl_name in reclaim_str_hashtbl */
3057         strhashval = clientstr_hashval(clp->cl_recdir);
3058         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3059                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3060                         return crp;
3061                 }
3062         }
3063         return NULL;
3064 }
3065
3066 /*
3067 * Called from OPEN. Look for clientid in reclaim list.
3068 */
3069 __be32
3070 nfs4_check_open_reclaim(clientid_t *clid)
3071 {
3072         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3073 }
3074
3075 /* initialization to perform at module load time: */
3076
3077 int
3078 nfs4_state_init(void)
3079 {
3080         int i, status;
3081
3082         status = nfsd4_init_slabs();
3083         if (status)
3084                 return status;
3085         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3086                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3087                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3088                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3089                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3090         }
3091         for (i = 0; i < FILE_HASH_SIZE; i++) {
3092                 INIT_LIST_HEAD(&file_hashtbl[i]);
3093         }
3094         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3095                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3096                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3097         }
3098         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3099                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3100                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3101         }
3102         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3103                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3104                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3105         }
3106         memset(&onestateid, ~0, sizeof(stateid_t));
3107         INIT_LIST_HEAD(&close_lru);
3108         INIT_LIST_HEAD(&client_lru);
3109         INIT_LIST_HEAD(&del_recall_lru);
3110         for (i = 0; i < CLIENT_HASH_SIZE; i++)
3111                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3112         reclaim_str_hashtbl_size = 0;
3113         return 0;
3114 }
3115
3116 static void
3117 nfsd4_load_reboot_recovery_data(void)
3118 {
3119         int status;
3120
3121         nfs4_lock_state();
3122         nfsd4_init_recdir(user_recovery_dirname);
3123         status = nfsd4_recdir_load();
3124         nfs4_unlock_state();
3125         if (status)
3126                 printk("NFSD: Failure reading reboot recovery data\n");
3127 }
3128
3129 unsigned long
3130 get_nfs4_grace_period(void)
3131 {
3132         return max(user_lease_time, lease_time) * HZ;
3133 }
3134
3135 /*
3136  * Since the lifetime of a delegation isn't limited to that of an open, a
3137  * client may quite reasonably hang on to a delegation as long as it has
3138  * the inode cached.  This becomes an obvious problem the first time a
3139  * client's inode cache approaches the size of the server's total memory.
3140  *
3141  * For now we avoid this problem by imposing a hard limit on the number
3142  * of delegations, which varies according to the server's memory size.
3143  */
3144 static void
3145 set_max_delegations(void)
3146 {
3147         /*
3148          * Allow at most 4 delegations per megabyte of RAM.  Quick
3149          * estimates suggest that in the worst case (where every delegation
3150          * is for a different inode), a delegation could take about 1.5K,
3151          * giving a worst case usage of about 6% of memory.
3152          */
3153         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3154 }
3155
3156 /* initialization to perform when the nfsd service is started: */
3157
3158 static void
3159 __nfs4_state_start(void)
3160 {
3161         unsigned long grace_time;
3162
3163         boot_time = get_seconds();
3164         grace_time = get_nfs_grace_period();
3165         lease_time = user_lease_time;
3166         in_grace = 1;
3167         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3168                grace_time/HZ);
3169         laundry_wq = create_singlethread_workqueue("nfsd4");
3170         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
3171         set_max_delegations();
3172 }
3173
3174 void
3175 nfs4_state_start(void)
3176 {
3177         if (nfs4_init)
3178                 return;
3179         nfsd4_load_reboot_recovery_data();
3180         __nfs4_state_start();
3181         nfs4_init = 1;
3182         return;
3183 }
3184
3185 int
3186 nfs4_in_grace(void)
3187 {
3188         return in_grace;
3189 }
3190
3191 time_t
3192 nfs4_lease_time(void)
3193 {
3194         return lease_time;
3195 }
3196
3197 static void
3198 __nfs4_state_shutdown(void)
3199 {
3200         int i;
3201         struct nfs4_client *clp = NULL;
3202         struct nfs4_delegation *dp = NULL;
3203         struct list_head *pos, *next, reaplist;
3204
3205         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3206                 while (!list_empty(&conf_id_hashtbl[i])) {
3207                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
3208                         expire_client(clp);
3209                 }
3210                 while (!list_empty(&unconf_str_hashtbl[i])) {
3211                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
3212                         expire_client(clp);
3213                 }
3214         }
3215         INIT_LIST_HEAD(&reaplist);
3216         spin_lock(&recall_lock);
3217         list_for_each_safe(pos, next, &del_recall_lru) {
3218                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3219                 list_move(&dp->dl_recall_lru, &reaplist);
3220         }
3221         spin_unlock(&recall_lock);
3222         list_for_each_safe(pos, next, &reaplist) {
3223                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3224                 list_del_init(&dp->dl_recall_lru);
3225                 unhash_delegation(dp);
3226         }
3227
3228         nfsd4_shutdown_recdir();
3229         nfs4_init = 0;
3230 }
3231
3232 void
3233 nfs4_state_shutdown(void)
3234 {
3235         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
3236         destroy_workqueue(laundry_wq);
3237         nfs4_lock_state();
3238         nfs4_release_reclaim();
3239         __nfs4_state_shutdown();
3240         nfs4_unlock_state();
3241 }
3242
3243 static void
3244 nfs4_set_recdir(char *recdir)
3245 {
3246         nfs4_lock_state();
3247         strcpy(user_recovery_dirname, recdir);
3248         nfs4_unlock_state();
3249 }
3250
3251 /*
3252  * Change the NFSv4 recovery directory to recdir.
3253  */
3254 int
3255 nfs4_reset_recoverydir(char *recdir)
3256 {
3257         int status;
3258         struct nameidata nd;
3259
3260         status = path_lookup(recdir, LOOKUP_FOLLOW, &nd);
3261         if (status)
3262                 return status;
3263         status = -ENOTDIR;
3264         if (S_ISDIR(nd.path.dentry->d_inode->i_mode)) {
3265                 nfs4_set_recdir(recdir);
3266                 status = 0;
3267         }
3268         path_put(&nd.path);
3269         return status;
3270 }
3271
3272 /*
3273  * Called when leasetime is changed.
3274  *
3275  * The only way the protocol gives us to handle on-the-fly lease changes is to
3276  * simulate a reboot.  Instead of doing that, we just wait till the next time
3277  * we start to register any changes in lease time.  If the administrator
3278  * really wants to change the lease time *now*, they can go ahead and bring
3279  * nfsd down and then back up again after changing the lease time.
3280  */
3281 void
3282 nfs4_reset_lease(time_t leasetime)
3283 {
3284         lock_kernel();
3285         user_lease_time = leasetime;
3286         unlock_kernel();
3287 }