65978a9aa8772ad279c1847c50eb695b94baaa17
[linux-2.6.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
42 #include "xdr4.h"
43 #include "vfs.h"
44
45 #define NFSDDBG_FACILITY                NFSDDBG_PROC
46
47 /* Globals */
48 time_t nfsd4_lease = 90;     /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
166 {
167         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168         atomic_inc(&fp->fi_access[oflag]);
169 }
170
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
172 {
173         if (oflag == O_RDWR) {
174                 __nfs4_file_get_access(fp, O_RDONLY);
175                 __nfs4_file_get_access(fp, O_WRONLY);
176         } else
177                 __nfs4_file_get_access(fp, oflag);
178 }
179
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
181 {
182         if (fp->fi_fds[oflag]) {
183                 fput(fp->fi_fds[oflag]);
184                 fp->fi_fds[oflag] = NULL;
185         }
186 }
187
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
189 {
190         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191                 nfs4_file_put_fd(fp, O_RDWR);
192                 nfs4_file_put_fd(fp, oflag);
193         }
194 }
195
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
197 {
198         if (oflag == O_RDWR) {
199                 __nfs4_file_put_access(fp, O_RDONLY);
200                 __nfs4_file_put_access(fp, O_WRONLY);
201         } else
202                 __nfs4_file_put_access(fp, oflag);
203 }
204
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
207 {
208         struct nfs4_delegation *dp;
209         struct nfs4_file *fp = stp->st_file;
210
211         dprintk("NFSD alloc_init_deleg\n");
212         /*
213          * Major work on the lease subsystem (for example, to support
214          * calbacks on stat) will be required before we can support
215          * write delegations properly.
216          */
217         if (type != NFS4_OPEN_DELEGATE_READ)
218                 return NULL;
219         if (fp->fi_had_conflict)
220                 return NULL;
221         if (num_delegations > max_delegations)
222                 return NULL;
223         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
224         if (dp == NULL)
225                 return dp;
226         num_delegations++;
227         INIT_LIST_HEAD(&dp->dl_perfile);
228         INIT_LIST_HEAD(&dp->dl_perclnt);
229         INIT_LIST_HEAD(&dp->dl_recall_lru);
230         dp->dl_client = clp;
231         get_nfs4_file(fp);
232         dp->dl_file = fp;
233         dp->dl_vfs_file = find_readable_file(fp);
234         get_file(dp->dl_vfs_file);
235         dp->dl_flock = NULL;
236         dp->dl_type = type;
237         dp->dl_stateid.si_boot = boot_time;
238         dp->dl_stateid.si_stateownerid = current_delegid++;
239         dp->dl_stateid.si_fileid = 0;
240         dp->dl_stateid.si_generation = 0;
241         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
242         dp->dl_time = 0;
243         atomic_set(&dp->dl_count, 1);
244         list_add(&dp->dl_perfile, &fp->fi_delegations);
245         list_add(&dp->dl_perclnt, &clp->cl_delegations);
246         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
247         return dp;
248 }
249
250 void
251 nfs4_put_delegation(struct nfs4_delegation *dp)
252 {
253         if (atomic_dec_and_test(&dp->dl_count)) {
254                 dprintk("NFSD: freeing dp %p\n",dp);
255                 put_nfs4_file(dp->dl_file);
256                 fput(dp->dl_vfs_file);
257                 kmem_cache_free(deleg_slab, dp);
258                 num_delegations--;
259         }
260 }
261
262 /* Remove the associated file_lock first, then remove the delegation.
263  * lease_modify() is called to remove the FS_LEASE file_lock from
264  * the i_flock list, eventually calling nfsd's lock_manager
265  * fl_release_callback.
266  */
267 static void
268 nfs4_close_delegation(struct nfs4_delegation *dp)
269 {
270         dprintk("NFSD: close_delegation dp %p\n",dp);
271         /* XXX: do we even need this check?: */
272         if (dp->dl_flock)
273                 vfs_setlease(dp->dl_vfs_file, F_UNLCK, &dp->dl_flock);
274 }
275
276 /* Called under the state lock. */
277 static void
278 unhash_delegation(struct nfs4_delegation *dp)
279 {
280         list_del_init(&dp->dl_perfile);
281         list_del_init(&dp->dl_perclnt);
282         spin_lock(&recall_lock);
283         list_del_init(&dp->dl_recall_lru);
284         spin_unlock(&recall_lock);
285         nfs4_close_delegation(dp);
286         nfs4_put_delegation(dp);
287 }
288
289 /* 
290  * SETCLIENTID state 
291  */
292
293 /* client_lock protects the client lru list and session hash table */
294 static DEFINE_SPINLOCK(client_lock);
295
296 /* Hash tables for nfs4_clientid state */
297 #define CLIENT_HASH_BITS                 4
298 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
299 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
300
301 #define clientid_hashval(id) \
302         ((id) & CLIENT_HASH_MASK)
303 #define clientstr_hashval(name) \
304         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
305 /*
306  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
307  * used in reboot/reset lease grace period processing
308  *
309  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
310  * setclientid_confirmed info. 
311  *
312  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
313  * setclientid info.
314  *
315  * client_lru holds client queue ordered by nfs4_client.cl_time
316  * for lease renewal.
317  *
318  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
319  * for last close replay.
320  */
321 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
322 static int reclaim_str_hashtbl_size = 0;
323 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
324 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
325 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head client_lru;
328 static struct list_head close_lru;
329
330 static void unhash_generic_stateid(struct nfs4_stateid *stp)
331 {
332         list_del(&stp->st_hash);
333         list_del(&stp->st_perfile);
334         list_del(&stp->st_perstateowner);
335 }
336
337 static void free_generic_stateid(struct nfs4_stateid *stp)
338 {
339         put_nfs4_file(stp->st_file);
340         kmem_cache_free(stateid_slab, stp);
341 }
342
343 static void release_lock_stateid(struct nfs4_stateid *stp)
344 {
345         struct file *file;
346
347         unhash_generic_stateid(stp);
348         file = find_any_file(stp->st_file);
349         if (file)
350                 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
351         free_generic_stateid(stp);
352 }
353
354 static void unhash_lockowner(struct nfs4_stateowner *sop)
355 {
356         struct nfs4_stateid *stp;
357
358         list_del(&sop->so_idhash);
359         list_del(&sop->so_strhash);
360         list_del(&sop->so_perstateid);
361         while (!list_empty(&sop->so_stateids)) {
362                 stp = list_first_entry(&sop->so_stateids,
363                                 struct nfs4_stateid, st_perstateowner);
364                 release_lock_stateid(stp);
365         }
366 }
367
368 static void release_lockowner(struct nfs4_stateowner *sop)
369 {
370         unhash_lockowner(sop);
371         nfs4_put_stateowner(sop);
372 }
373
374 static void
375 release_stateid_lockowners(struct nfs4_stateid *open_stp)
376 {
377         struct nfs4_stateowner *lock_sop;
378
379         while (!list_empty(&open_stp->st_lockowners)) {
380                 lock_sop = list_entry(open_stp->st_lockowners.next,
381                                 struct nfs4_stateowner, so_perstateid);
382                 /* list_del(&open_stp->st_lockowners);  */
383                 BUG_ON(lock_sop->so_is_open_owner);
384                 release_lockowner(lock_sop);
385         }
386 }
387
388 /*
389  * We store the NONE, READ, WRITE, and BOTH bits separately in the
390  * st_{access,deny}_bmap field of the stateid, in order to track not
391  * only what share bits are currently in force, but also what
392  * combinations of share bits previous opens have used.  This allows us
393  * to enforce the recommendation of rfc 3530 14.2.19 that the server
394  * return an error if the client attempt to downgrade to a combination
395  * of share bits not explicable by closing some of its previous opens.
396  *
397  * XXX: This enforcement is actually incomplete, since we don't keep
398  * track of access/deny bit combinations; so, e.g., we allow:
399  *
400  *      OPEN allow read, deny write
401  *      OPEN allow both, deny none
402  *      DOWNGRADE allow read, deny none
403  *
404  * which we should reject.
405  */
406 static void
407 set_access(unsigned int *access, unsigned long bmap) {
408         int i;
409
410         *access = 0;
411         for (i = 1; i < 4; i++) {
412                 if (test_bit(i, &bmap))
413                         *access |= i;
414         }
415 }
416
417 static void
418 set_deny(unsigned int *deny, unsigned long bmap) {
419         int i;
420
421         *deny = 0;
422         for (i = 0; i < 4; i++) {
423                 if (test_bit(i, &bmap))
424                         *deny |= i ;
425         }
426 }
427
428 static int
429 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
430         unsigned int access, deny;
431
432         set_access(&access, stp->st_access_bmap);
433         set_deny(&deny, stp->st_deny_bmap);
434         if ((access & open->op_share_deny) || (deny & open->op_share_access))
435                 return 0;
436         return 1;
437 }
438
439 static int nfs4_access_to_omode(u32 access)
440 {
441         switch (access & NFS4_SHARE_ACCESS_BOTH) {
442         case NFS4_SHARE_ACCESS_READ:
443                 return O_RDONLY;
444         case NFS4_SHARE_ACCESS_WRITE:
445                 return O_WRONLY;
446         case NFS4_SHARE_ACCESS_BOTH:
447                 return O_RDWR;
448         }
449         BUG();
450 }
451
452 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
453 {
454         unsigned int access;
455
456         set_access(&access, stp->st_access_bmap);
457         return nfs4_access_to_omode(access);
458 }
459
460 static void release_open_stateid(struct nfs4_stateid *stp)
461 {
462         int oflag = nfs4_access_bmap_to_omode(stp);
463
464         unhash_generic_stateid(stp);
465         release_stateid_lockowners(stp);
466         nfs4_file_put_access(stp->st_file, oflag);
467         free_generic_stateid(stp);
468 }
469
470 static void unhash_openowner(struct nfs4_stateowner *sop)
471 {
472         struct nfs4_stateid *stp;
473
474         list_del(&sop->so_idhash);
475         list_del(&sop->so_strhash);
476         list_del(&sop->so_perclient);
477         list_del(&sop->so_perstateid); /* XXX: necessary? */
478         while (!list_empty(&sop->so_stateids)) {
479                 stp = list_first_entry(&sop->so_stateids,
480                                 struct nfs4_stateid, st_perstateowner);
481                 release_open_stateid(stp);
482         }
483 }
484
485 static void release_openowner(struct nfs4_stateowner *sop)
486 {
487         unhash_openowner(sop);
488         list_del(&sop->so_close_lru);
489         nfs4_put_stateowner(sop);
490 }
491
492 #define SESSION_HASH_SIZE       512
493 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
494
495 static inline int
496 hash_sessionid(struct nfs4_sessionid *sessionid)
497 {
498         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
499
500         return sid->sequence % SESSION_HASH_SIZE;
501 }
502
503 static inline void
504 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
505 {
506         u32 *ptr = (u32 *)(&sessionid->data[0]);
507         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
508 }
509
510 static void
511 gen_sessionid(struct nfsd4_session *ses)
512 {
513         struct nfs4_client *clp = ses->se_client;
514         struct nfsd4_sessionid *sid;
515
516         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
517         sid->clientid = clp->cl_clientid;
518         sid->sequence = current_sessionid++;
519         sid->reserved = 0;
520 }
521
522 /*
523  * The protocol defines ca_maxresponssize_cached to include the size of
524  * the rpc header, but all we need to cache is the data starting after
525  * the end of the initial SEQUENCE operation--the rest we regenerate
526  * each time.  Therefore we can advertise a ca_maxresponssize_cached
527  * value that is the number of bytes in our cache plus a few additional
528  * bytes.  In order to stay on the safe side, and not promise more than
529  * we can cache, those additional bytes must be the minimum possible: 24
530  * bytes of rpc header (xid through accept state, with AUTH_NULL
531  * verifier), 12 for the compound header (with zero-length tag), and 44
532  * for the SEQUENCE op response:
533  */
534 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
535
536 static void
537 free_session_slots(struct nfsd4_session *ses)
538 {
539         int i;
540
541         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
542                 kfree(ses->se_slots[i]);
543 }
544
545 /*
546  * We don't actually need to cache the rpc and session headers, so we
547  * can allocate a little less for each slot:
548  */
549 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
550 {
551         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
552 }
553
554 static int nfsd4_sanitize_slot_size(u32 size)
555 {
556         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
557         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
558
559         return size;
560 }
561
562 /*
563  * XXX: If we run out of reserved DRC memory we could (up to a point)
564  * re-negotiate active sessions and reduce their slot usage to make
565  * rooom for new connections. For now we just fail the create session.
566  */
567 static int nfsd4_get_drc_mem(int slotsize, u32 num)
568 {
569         int avail;
570
571         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
572
573         spin_lock(&nfsd_drc_lock);
574         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
575                         nfsd_drc_max_mem - nfsd_drc_mem_used);
576         num = min_t(int, num, avail / slotsize);
577         nfsd_drc_mem_used += num * slotsize;
578         spin_unlock(&nfsd_drc_lock);
579
580         return num;
581 }
582
583 static void nfsd4_put_drc_mem(int slotsize, int num)
584 {
585         spin_lock(&nfsd_drc_lock);
586         nfsd_drc_mem_used -= slotsize * num;
587         spin_unlock(&nfsd_drc_lock);
588 }
589
590 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
591 {
592         struct nfsd4_session *new;
593         int mem, i;
594
595         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
596                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
597         mem = numslots * sizeof(struct nfsd4_slot *);
598
599         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
600         if (!new)
601                 return NULL;
602         /* allocate each struct nfsd4_slot and data cache in one piece */
603         for (i = 0; i < numslots; i++) {
604                 mem = sizeof(struct nfsd4_slot) + slotsize;
605                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
606                 if (!new->se_slots[i])
607                         goto out_free;
608         }
609         return new;
610 out_free:
611         while (i--)
612                 kfree(new->se_slots[i]);
613         kfree(new);
614         return NULL;
615 }
616
617 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
618 {
619         u32 maxrpc = nfsd_serv->sv_max_mesg;
620
621         new->maxreqs = numslots;
622         new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
623         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
624         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
625         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
626 }
627
628 static void free_conn(struct nfsd4_conn *c)
629 {
630         svc_xprt_put(c->cn_xprt);
631         kfree(c);
632 }
633
634 static void nfsd4_conn_lost(struct svc_xpt_user *u)
635 {
636         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
637         struct nfs4_client *clp = c->cn_session->se_client;
638
639         spin_lock(&clp->cl_lock);
640         if (!list_empty(&c->cn_persession)) {
641                 list_del(&c->cn_persession);
642                 free_conn(c);
643         }
644         spin_unlock(&clp->cl_lock);
645         nfsd4_probe_callback(clp);
646 }
647
648 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
649 {
650         struct nfsd4_conn *conn;
651
652         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
653         if (!conn)
654                 return NULL;
655         svc_xprt_get(rqstp->rq_xprt);
656         conn->cn_xprt = rqstp->rq_xprt;
657         conn->cn_flags = flags;
658         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
659         return conn;
660 }
661
662 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
663 {
664         conn->cn_session = ses;
665         list_add(&conn->cn_persession, &ses->se_conns);
666 }
667
668 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
669 {
670         struct nfs4_client *clp = ses->se_client;
671
672         spin_lock(&clp->cl_lock);
673         __nfsd4_hash_conn(conn, ses);
674         spin_unlock(&clp->cl_lock);
675 }
676
677 static int nfsd4_register_conn(struct nfsd4_conn *conn)
678 {
679         conn->cn_xpt_user.callback = nfsd4_conn_lost;
680         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
681 }
682
683 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
684 {
685         struct nfsd4_conn *conn;
686         int ret;
687
688         conn = alloc_conn(rqstp, dir);
689         if (!conn)
690                 return nfserr_jukebox;
691         nfsd4_hash_conn(conn, ses);
692         ret = nfsd4_register_conn(conn);
693         if (ret)
694                 /* oops; xprt is already down: */
695                 nfsd4_conn_lost(&conn->cn_xpt_user);
696         return nfs_ok;
697 }
698
699 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
700 {
701         u32 dir = NFS4_CDFC4_FORE;
702
703         if (ses->se_flags & SESSION4_BACK_CHAN)
704                 dir |= NFS4_CDFC4_BACK;
705
706         return nfsd4_new_conn(rqstp, ses, dir);
707 }
708
709 /* must be called under client_lock */
710 static void nfsd4_del_conns(struct nfsd4_session *s)
711 {
712         struct nfs4_client *clp = s->se_client;
713         struct nfsd4_conn *c;
714
715         spin_lock(&clp->cl_lock);
716         while (!list_empty(&s->se_conns)) {
717                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
718                 list_del_init(&c->cn_persession);
719                 spin_unlock(&clp->cl_lock);
720
721                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
722                 free_conn(c);
723
724                 spin_lock(&clp->cl_lock);
725         }
726         spin_unlock(&clp->cl_lock);
727 }
728
729 void free_session(struct kref *kref)
730 {
731         struct nfsd4_session *ses;
732         int mem;
733
734         ses = container_of(kref, struct nfsd4_session, se_ref);
735         nfsd4_del_conns(ses);
736         spin_lock(&nfsd_drc_lock);
737         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
738         nfsd_drc_mem_used -= mem;
739         spin_unlock(&nfsd_drc_lock);
740         free_session_slots(ses);
741         kfree(ses);
742 }
743
744 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
745 {
746         struct nfsd4_session *new;
747         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
748         int numslots, slotsize;
749         int status;
750         int idx;
751
752         /*
753          * Note decreasing slot size below client's request may
754          * make it difficult for client to function correctly, whereas
755          * decreasing the number of slots will (just?) affect
756          * performance.  When short on memory we therefore prefer to
757          * decrease number of slots instead of their size.
758          */
759         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
760         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
761         if (numslots < 1)
762                 return NULL;
763
764         new = alloc_session(slotsize, numslots);
765         if (!new) {
766                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
767                 return NULL;
768         }
769         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
770
771         new->se_client = clp;
772         gen_sessionid(new);
773
774         INIT_LIST_HEAD(&new->se_conns);
775
776         new->se_cb_seq_nr = 1;
777         new->se_flags = cses->flags;
778         new->se_cb_prog = cses->callback_prog;
779         kref_init(&new->se_ref);
780         idx = hash_sessionid(&new->se_sessionid);
781         spin_lock(&client_lock);
782         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
783         spin_lock(&clp->cl_lock);
784         list_add(&new->se_perclnt, &clp->cl_sessions);
785         spin_unlock(&clp->cl_lock);
786         spin_unlock(&client_lock);
787
788         status = nfsd4_new_conn_from_crses(rqstp, new);
789         /* whoops: benny points out, status is ignored! (err, or bogus) */
790         if (status) {
791                 free_session(&new->se_ref);
792                 return NULL;
793         }
794         if (cses->flags & SESSION4_BACK_CHAN) {
795                 struct sockaddr *sa = svc_addr(rqstp);
796                 /*
797                  * This is a little silly; with sessions there's no real
798                  * use for the callback address.  Use the peer address
799                  * as a reasonable default for now, but consider fixing
800                  * the rpc client not to require an address in the
801                  * future:
802                  */
803                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
804                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
805         }
806         nfsd4_probe_callback(clp);
807         return new;
808 }
809
810 /* caller must hold client_lock */
811 static struct nfsd4_session *
812 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
813 {
814         struct nfsd4_session *elem;
815         int idx;
816
817         dump_sessionid(__func__, sessionid);
818         idx = hash_sessionid(sessionid);
819         /* Search in the appropriate list */
820         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
821                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
822                             NFS4_MAX_SESSIONID_LEN)) {
823                         return elem;
824                 }
825         }
826
827         dprintk("%s: session not found\n", __func__);
828         return NULL;
829 }
830
831 /* caller must hold client_lock */
832 static void
833 unhash_session(struct nfsd4_session *ses)
834 {
835         list_del(&ses->se_hash);
836         spin_lock(&ses->se_client->cl_lock);
837         list_del(&ses->se_perclnt);
838         spin_unlock(&ses->se_client->cl_lock);
839 }
840
841 /* must be called under the client_lock */
842 static inline void
843 renew_client_locked(struct nfs4_client *clp)
844 {
845         if (is_client_expired(clp)) {
846                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
847                         __func__,
848                         clp->cl_clientid.cl_boot,
849                         clp->cl_clientid.cl_id);
850                 return;
851         }
852
853         /*
854         * Move client to the end to the LRU list.
855         */
856         dprintk("renewing client (clientid %08x/%08x)\n", 
857                         clp->cl_clientid.cl_boot, 
858                         clp->cl_clientid.cl_id);
859         list_move_tail(&clp->cl_lru, &client_lru);
860         clp->cl_time = get_seconds();
861 }
862
863 static inline void
864 renew_client(struct nfs4_client *clp)
865 {
866         spin_lock(&client_lock);
867         renew_client_locked(clp);
868         spin_unlock(&client_lock);
869 }
870
871 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
872 static int
873 STALE_CLIENTID(clientid_t *clid)
874 {
875         if (clid->cl_boot == boot_time)
876                 return 0;
877         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
878                 clid->cl_boot, clid->cl_id, boot_time);
879         return 1;
880 }
881
882 /* 
883  * XXX Should we use a slab cache ?
884  * This type of memory management is somewhat inefficient, but we use it
885  * anyway since SETCLIENTID is not a common operation.
886  */
887 static struct nfs4_client *alloc_client(struct xdr_netobj name)
888 {
889         struct nfs4_client *clp;
890
891         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
892         if (clp == NULL)
893                 return NULL;
894         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
895         if (clp->cl_name.data == NULL) {
896                 kfree(clp);
897                 return NULL;
898         }
899         memcpy(clp->cl_name.data, name.data, name.len);
900         clp->cl_name.len = name.len;
901         return clp;
902 }
903
904 static inline void
905 free_client(struct nfs4_client *clp)
906 {
907         while (!list_empty(&clp->cl_sessions)) {
908                 struct nfsd4_session *ses;
909                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
910                                 se_perclnt);
911                 list_del(&ses->se_perclnt);
912                 nfsd4_put_session(ses);
913         }
914         if (clp->cl_cred.cr_group_info)
915                 put_group_info(clp->cl_cred.cr_group_info);
916         kfree(clp->cl_principal);
917         kfree(clp->cl_name.data);
918         kfree(clp);
919 }
920
921 void
922 release_session_client(struct nfsd4_session *session)
923 {
924         struct nfs4_client *clp = session->se_client;
925
926         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
927                 return;
928         if (is_client_expired(clp)) {
929                 free_client(clp);
930                 session->se_client = NULL;
931         } else
932                 renew_client_locked(clp);
933         spin_unlock(&client_lock);
934 }
935
936 /* must be called under the client_lock */
937 static inline void
938 unhash_client_locked(struct nfs4_client *clp)
939 {
940         struct nfsd4_session *ses;
941
942         mark_client_expired(clp);
943         list_del(&clp->cl_lru);
944         spin_lock(&clp->cl_lock);
945         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
946                 list_del_init(&ses->se_hash);
947         spin_unlock(&clp->cl_lock);
948 }
949
950 static void
951 expire_client(struct nfs4_client *clp)
952 {
953         struct nfs4_stateowner *sop;
954         struct nfs4_delegation *dp;
955         struct list_head reaplist;
956
957         INIT_LIST_HEAD(&reaplist);
958         spin_lock(&recall_lock);
959         while (!list_empty(&clp->cl_delegations)) {
960                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
961                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
962                                 dp->dl_flock);
963                 list_del_init(&dp->dl_perclnt);
964                 list_move(&dp->dl_recall_lru, &reaplist);
965         }
966         spin_unlock(&recall_lock);
967         while (!list_empty(&reaplist)) {
968                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
969                 list_del_init(&dp->dl_recall_lru);
970                 unhash_delegation(dp);
971         }
972         while (!list_empty(&clp->cl_openowners)) {
973                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
974                 release_openowner(sop);
975         }
976         nfsd4_shutdown_callback(clp);
977         if (clp->cl_cb_conn.cb_xprt)
978                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
979         list_del(&clp->cl_idhash);
980         list_del(&clp->cl_strhash);
981         spin_lock(&client_lock);
982         unhash_client_locked(clp);
983         if (atomic_read(&clp->cl_refcount) == 0)
984                 free_client(clp);
985         spin_unlock(&client_lock);
986 }
987
988 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
989 {
990         memcpy(target->cl_verifier.data, source->data,
991                         sizeof(target->cl_verifier.data));
992 }
993
994 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
995 {
996         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
997         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
998 }
999
1000 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1001 {
1002         target->cr_uid = source->cr_uid;
1003         target->cr_gid = source->cr_gid;
1004         target->cr_group_info = source->cr_group_info;
1005         get_group_info(target->cr_group_info);
1006 }
1007
1008 static int same_name(const char *n1, const char *n2)
1009 {
1010         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1011 }
1012
1013 static int
1014 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1015 {
1016         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1017 }
1018
1019 static int
1020 same_clid(clientid_t *cl1, clientid_t *cl2)
1021 {
1022         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1023 }
1024
1025 /* XXX what about NGROUP */
1026 static int
1027 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1028 {
1029         return cr1->cr_uid == cr2->cr_uid;
1030 }
1031
1032 static void gen_clid(struct nfs4_client *clp)
1033 {
1034         static u32 current_clientid = 1;
1035
1036         clp->cl_clientid.cl_boot = boot_time;
1037         clp->cl_clientid.cl_id = current_clientid++; 
1038 }
1039
1040 static void gen_confirm(struct nfs4_client *clp)
1041 {
1042         static u32 i;
1043         u32 *p;
1044
1045         p = (u32 *)clp->cl_confirm.data;
1046         *p++ = get_seconds();
1047         *p++ = i++;
1048 }
1049
1050 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1051                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1052 {
1053         struct nfs4_client *clp;
1054         struct sockaddr *sa = svc_addr(rqstp);
1055         char *princ;
1056
1057         clp = alloc_client(name);
1058         if (clp == NULL)
1059                 return NULL;
1060
1061         INIT_LIST_HEAD(&clp->cl_sessions);
1062
1063         princ = svc_gss_principal(rqstp);
1064         if (princ) {
1065                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1066                 if (clp->cl_principal == NULL) {
1067                         free_client(clp);
1068                         return NULL;
1069                 }
1070         }
1071
1072         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1073         atomic_set(&clp->cl_refcount, 0);
1074         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1075         INIT_LIST_HEAD(&clp->cl_idhash);
1076         INIT_LIST_HEAD(&clp->cl_strhash);
1077         INIT_LIST_HEAD(&clp->cl_openowners);
1078         INIT_LIST_HEAD(&clp->cl_delegations);
1079         INIT_LIST_HEAD(&clp->cl_lru);
1080         INIT_LIST_HEAD(&clp->cl_callbacks);
1081         spin_lock_init(&clp->cl_lock);
1082         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1083         clp->cl_time = get_seconds();
1084         clear_bit(0, &clp->cl_cb_slot_busy);
1085         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1086         copy_verf(clp, verf);
1087         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1088         clp->cl_flavor = rqstp->rq_flavor;
1089         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1090         gen_confirm(clp);
1091         clp->cl_cb_session = NULL;
1092         return clp;
1093 }
1094
1095 static int check_name(struct xdr_netobj name)
1096 {
1097         if (name.len == 0) 
1098                 return 0;
1099         if (name.len > NFS4_OPAQUE_LIMIT) {
1100                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1101                 return 0;
1102         }
1103         return 1;
1104 }
1105
1106 static void
1107 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1108 {
1109         unsigned int idhashval;
1110
1111         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1112         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1113         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1114         renew_client(clp);
1115 }
1116
1117 static void
1118 move_to_confirmed(struct nfs4_client *clp)
1119 {
1120         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1121         unsigned int strhashval;
1122
1123         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1124         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1125         strhashval = clientstr_hashval(clp->cl_recdir);
1126         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1127         renew_client(clp);
1128 }
1129
1130 static struct nfs4_client *
1131 find_confirmed_client(clientid_t *clid)
1132 {
1133         struct nfs4_client *clp;
1134         unsigned int idhashval = clientid_hashval(clid->cl_id);
1135
1136         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1137                 if (same_clid(&clp->cl_clientid, clid))
1138                         return clp;
1139         }
1140         return NULL;
1141 }
1142
1143 static struct nfs4_client *
1144 find_unconfirmed_client(clientid_t *clid)
1145 {
1146         struct nfs4_client *clp;
1147         unsigned int idhashval = clientid_hashval(clid->cl_id);
1148
1149         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1150                 if (same_clid(&clp->cl_clientid, clid))
1151                         return clp;
1152         }
1153         return NULL;
1154 }
1155
1156 static bool clp_used_exchangeid(struct nfs4_client *clp)
1157 {
1158         return clp->cl_exchange_flags != 0;
1159
1160
1161 static struct nfs4_client *
1162 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1163 {
1164         struct nfs4_client *clp;
1165
1166         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1167                 if (same_name(clp->cl_recdir, dname))
1168                         return clp;
1169         }
1170         return NULL;
1171 }
1172
1173 static struct nfs4_client *
1174 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1175 {
1176         struct nfs4_client *clp;
1177
1178         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1179                 if (same_name(clp->cl_recdir, dname))
1180                         return clp;
1181         }
1182         return NULL;
1183 }
1184
1185 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1186 {
1187         switch (family) {
1188         case AF_INET:
1189                 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1190                 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1191                 return;
1192         case AF_INET6:
1193                 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1194                 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1195                 return;
1196         }
1197 }
1198
1199 static void
1200 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1201 {
1202         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1203         struct sockaddr *sa = svc_addr(rqstp);
1204         u32 scopeid = rpc_get_scope_id(sa);
1205         unsigned short expected_family;
1206
1207         /* Currently, we only support tcp and tcp6 for the callback channel */
1208         if (se->se_callback_netid_len == 3 &&
1209             !memcmp(se->se_callback_netid_val, "tcp", 3))
1210                 expected_family = AF_INET;
1211         else if (se->se_callback_netid_len == 4 &&
1212                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1213                 expected_family = AF_INET6;
1214         else
1215                 goto out_err;
1216
1217         conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1218                                             se->se_callback_addr_len,
1219                                             (struct sockaddr *)&conn->cb_addr,
1220                                             sizeof(conn->cb_addr));
1221
1222         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1223                 goto out_err;
1224
1225         if (conn->cb_addr.ss_family == AF_INET6)
1226                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1227
1228         conn->cb_prog = se->se_callback_prog;
1229         conn->cb_ident = se->se_callback_ident;
1230         rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1231         return;
1232 out_err:
1233         conn->cb_addr.ss_family = AF_UNSPEC;
1234         conn->cb_addrlen = 0;
1235         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1236                 "will not receive delegations\n",
1237                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1238
1239         return;
1240 }
1241
1242 /*
1243  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1244  */
1245 void
1246 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1247 {
1248         struct nfsd4_slot *slot = resp->cstate.slot;
1249         unsigned int base;
1250
1251         dprintk("--> %s slot %p\n", __func__, slot);
1252
1253         slot->sl_opcnt = resp->opcnt;
1254         slot->sl_status = resp->cstate.status;
1255
1256         if (nfsd4_not_cached(resp)) {
1257                 slot->sl_datalen = 0;
1258                 return;
1259         }
1260         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1261         base = (char *)resp->cstate.datap -
1262                                         (char *)resp->xbuf->head[0].iov_base;
1263         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1264                                     slot->sl_datalen))
1265                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1266         return;
1267 }
1268
1269 /*
1270  * Encode the replay sequence operation from the slot values.
1271  * If cachethis is FALSE encode the uncached rep error on the next
1272  * operation which sets resp->p and increments resp->opcnt for
1273  * nfs4svc_encode_compoundres.
1274  *
1275  */
1276 static __be32
1277 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1278                           struct nfsd4_compoundres *resp)
1279 {
1280         struct nfsd4_op *op;
1281         struct nfsd4_slot *slot = resp->cstate.slot;
1282
1283         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1284                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1285
1286         /* Encode the replayed sequence operation */
1287         op = &args->ops[resp->opcnt - 1];
1288         nfsd4_encode_operation(resp, op);
1289
1290         /* Return nfserr_retry_uncached_rep in next operation. */
1291         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1292                 op = &args->ops[resp->opcnt++];
1293                 op->status = nfserr_retry_uncached_rep;
1294                 nfsd4_encode_operation(resp, op);
1295         }
1296         return op->status;
1297 }
1298
1299 /*
1300  * The sequence operation is not cached because we can use the slot and
1301  * session values.
1302  */
1303 __be32
1304 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1305                          struct nfsd4_sequence *seq)
1306 {
1307         struct nfsd4_slot *slot = resp->cstate.slot;
1308         __be32 status;
1309
1310         dprintk("--> %s slot %p\n", __func__, slot);
1311
1312         /* Either returns 0 or nfserr_retry_uncached */
1313         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1314         if (status == nfserr_retry_uncached_rep)
1315                 return status;
1316
1317         /* The sequence operation has been encoded, cstate->datap set. */
1318         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1319
1320         resp->opcnt = slot->sl_opcnt;
1321         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1322         status = slot->sl_status;
1323
1324         return status;
1325 }
1326
1327 /*
1328  * Set the exchange_id flags returned by the server.
1329  */
1330 static void
1331 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1332 {
1333         /* pNFS is not supported */
1334         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1335
1336         /* Referrals are supported, Migration is not. */
1337         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1338
1339         /* set the wire flags to return to client. */
1340         clid->flags = new->cl_exchange_flags;
1341 }
1342
1343 __be32
1344 nfsd4_exchange_id(struct svc_rqst *rqstp,
1345                   struct nfsd4_compound_state *cstate,
1346                   struct nfsd4_exchange_id *exid)
1347 {
1348         struct nfs4_client *unconf, *conf, *new;
1349         int status;
1350         unsigned int            strhashval;
1351         char                    dname[HEXDIR_LEN];
1352         char                    addr_str[INET6_ADDRSTRLEN];
1353         nfs4_verifier           verf = exid->verifier;
1354         struct sockaddr         *sa = svc_addr(rqstp);
1355
1356         rpc_ntop(sa, addr_str, sizeof(addr_str));
1357         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1358                 "ip_addr=%s flags %x, spa_how %d\n",
1359                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1360                 addr_str, exid->flags, exid->spa_how);
1361
1362         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1363                 return nfserr_inval;
1364
1365         /* Currently only support SP4_NONE */
1366         switch (exid->spa_how) {
1367         case SP4_NONE:
1368                 break;
1369         case SP4_SSV:
1370                 return nfserr_serverfault;
1371         default:
1372                 BUG();                          /* checked by xdr code */
1373         case SP4_MACH_CRED:
1374                 return nfserr_serverfault;      /* no excuse :-/ */
1375         }
1376
1377         status = nfs4_make_rec_clidname(dname, &exid->clname);
1378
1379         if (status)
1380                 goto error;
1381
1382         strhashval = clientstr_hashval(dname);
1383
1384         nfs4_lock_state();
1385         status = nfs_ok;
1386
1387         conf = find_confirmed_client_by_str(dname, strhashval);
1388         if (conf) {
1389                 if (!clp_used_exchangeid(conf)) {
1390                         status = nfserr_clid_inuse; /* XXX: ? */
1391                         goto out;
1392                 }
1393                 if (!same_verf(&verf, &conf->cl_verifier)) {
1394                         /* 18.35.4 case 8 */
1395                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1396                                 status = nfserr_not_same;
1397                                 goto out;
1398                         }
1399                         /* Client reboot: destroy old state */
1400                         expire_client(conf);
1401                         goto out_new;
1402                 }
1403                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1404                         /* 18.35.4 case 9 */
1405                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1406                                 status = nfserr_perm;
1407                                 goto out;
1408                         }
1409                         expire_client(conf);
1410                         goto out_new;
1411                 }
1412                 /*
1413                  * Set bit when the owner id and verifier map to an already
1414                  * confirmed client id (18.35.3).
1415                  */
1416                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1417
1418                 /*
1419                  * Falling into 18.35.4 case 2, possible router replay.
1420                  * Leave confirmed record intact and return same result.
1421                  */
1422                 copy_verf(conf, &verf);
1423                 new = conf;
1424                 goto out_copy;
1425         }
1426
1427         /* 18.35.4 case 7 */
1428         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1429                 status = nfserr_noent;
1430                 goto out;
1431         }
1432
1433         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1434         if (unconf) {
1435                 /*
1436                  * Possible retry or client restart.  Per 18.35.4 case 4,
1437                  * a new unconfirmed record should be generated regardless
1438                  * of whether any properties have changed.
1439                  */
1440                 expire_client(unconf);
1441         }
1442
1443 out_new:
1444         /* Normal case */
1445         new = create_client(exid->clname, dname, rqstp, &verf);
1446         if (new == NULL) {
1447                 status = nfserr_jukebox;
1448                 goto out;
1449         }
1450
1451         gen_clid(new);
1452         add_to_unconfirmed(new, strhashval);
1453 out_copy:
1454         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1455         exid->clientid.cl_id = new->cl_clientid.cl_id;
1456
1457         exid->seqid = 1;
1458         nfsd4_set_ex_flags(new, exid);
1459
1460         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1461                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1462         status = nfs_ok;
1463
1464 out:
1465         nfs4_unlock_state();
1466 error:
1467         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1468         return status;
1469 }
1470
1471 static int
1472 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1473 {
1474         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1475                 slot_seqid);
1476
1477         /* The slot is in use, and no response has been sent. */
1478         if (slot_inuse) {
1479                 if (seqid == slot_seqid)
1480                         return nfserr_jukebox;
1481                 else
1482                         return nfserr_seq_misordered;
1483         }
1484         /* Normal */
1485         if (likely(seqid == slot_seqid + 1))
1486                 return nfs_ok;
1487         /* Replay */
1488         if (seqid == slot_seqid)
1489                 return nfserr_replay_cache;
1490         /* Wraparound */
1491         if (seqid == 1 && (slot_seqid + 1) == 0)
1492                 return nfs_ok;
1493         /* Misordered replay or misordered new request */
1494         return nfserr_seq_misordered;
1495 }
1496
1497 /*
1498  * Cache the create session result into the create session single DRC
1499  * slot cache by saving the xdr structure. sl_seqid has been set.
1500  * Do this for solo or embedded create session operations.
1501  */
1502 static void
1503 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1504                            struct nfsd4_clid_slot *slot, int nfserr)
1505 {
1506         slot->sl_status = nfserr;
1507         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1508 }
1509
1510 static __be32
1511 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1512                             struct nfsd4_clid_slot *slot)
1513 {
1514         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1515         return slot->sl_status;
1516 }
1517
1518 __be32
1519 nfsd4_create_session(struct svc_rqst *rqstp,
1520                      struct nfsd4_compound_state *cstate,
1521                      struct nfsd4_create_session *cr_ses)
1522 {
1523         struct sockaddr *sa = svc_addr(rqstp);
1524         struct nfs4_client *conf, *unconf;
1525         struct nfsd4_session *new;
1526         struct nfsd4_clid_slot *cs_slot = NULL;
1527         bool confirm_me = false;
1528         int status = 0;
1529
1530         nfs4_lock_state();
1531         unconf = find_unconfirmed_client(&cr_ses->clientid);
1532         conf = find_confirmed_client(&cr_ses->clientid);
1533
1534         if (conf) {
1535                 cs_slot = &conf->cl_cs_slot;
1536                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1537                 if (status == nfserr_replay_cache) {
1538                         dprintk("Got a create_session replay! seqid= %d\n",
1539                                 cs_slot->sl_seqid);
1540                         /* Return the cached reply status */
1541                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1542                         goto out;
1543                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1544                         status = nfserr_seq_misordered;
1545                         dprintk("Sequence misordered!\n");
1546                         dprintk("Expected seqid= %d but got seqid= %d\n",
1547                                 cs_slot->sl_seqid, cr_ses->seqid);
1548                         goto out;
1549                 }
1550         } else if (unconf) {
1551                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1552                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1553                         status = nfserr_clid_inuse;
1554                         goto out;
1555                 }
1556
1557                 cs_slot = &unconf->cl_cs_slot;
1558                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1559                 if (status) {
1560                         /* an unconfirmed replay returns misordered */
1561                         status = nfserr_seq_misordered;
1562                         goto out;
1563                 }
1564
1565                 confirm_me = true;
1566                 conf = unconf;
1567         } else {
1568                 status = nfserr_stale_clientid;
1569                 goto out;
1570         }
1571
1572         /*
1573          * XXX: we should probably set this at creation time, and check
1574          * for consistent minorversion use throughout:
1575          */
1576         conf->cl_minorversion = 1;
1577         /*
1578          * We do not support RDMA or persistent sessions
1579          */
1580         cr_ses->flags &= ~SESSION4_PERSIST;
1581         cr_ses->flags &= ~SESSION4_RDMA;
1582
1583         status = nfserr_jukebox;
1584         new = alloc_init_session(rqstp, conf, cr_ses);
1585         if (!new)
1586                 goto out;
1587         status = nfs_ok;
1588         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1589                NFS4_MAX_SESSIONID_LEN);
1590         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1591                 sizeof(struct nfsd4_channel_attrs));
1592         cs_slot->sl_seqid++;
1593         cr_ses->seqid = cs_slot->sl_seqid;
1594
1595         /* cache solo and embedded create sessions under the state lock */
1596         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1597         if (confirm_me)
1598                 move_to_confirmed(conf);
1599 out:
1600         nfs4_unlock_state();
1601         dprintk("%s returns %d\n", __func__, ntohl(status));
1602         return status;
1603 }
1604
1605 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1606 {
1607         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1608         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1609
1610         return argp->opcnt == resp->opcnt;
1611 }
1612
1613 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1614 {
1615         switch (*dir) {
1616         case NFS4_CDFC4_FORE:
1617         case NFS4_CDFC4_BACK:
1618                 return nfs_ok;
1619         case NFS4_CDFC4_FORE_OR_BOTH:
1620         case NFS4_CDFC4_BACK_OR_BOTH:
1621                 *dir = NFS4_CDFC4_BOTH;
1622                 return nfs_ok;
1623         };
1624         return nfserr_inval;
1625 }
1626
1627 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1628                      struct nfsd4_compound_state *cstate,
1629                      struct nfsd4_bind_conn_to_session *bcts)
1630 {
1631         __be32 status;
1632
1633         if (!nfsd4_last_compound_op(rqstp))
1634                 return nfserr_not_only_op;
1635         spin_lock(&client_lock);
1636         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1637         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1638          * client_lock iself: */
1639         if (cstate->session) {
1640                 nfsd4_get_session(cstate->session);
1641                 atomic_inc(&cstate->session->se_client->cl_refcount);
1642         }
1643         spin_unlock(&client_lock);
1644         if (!cstate->session)
1645                 return nfserr_badsession;
1646
1647         status = nfsd4_map_bcts_dir(&bcts->dir);
1648         nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1649         return nfs_ok;
1650 }
1651
1652 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1653 {
1654         if (!session)
1655                 return 0;
1656         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1657 }
1658
1659 __be32
1660 nfsd4_destroy_session(struct svc_rqst *r,
1661                       struct nfsd4_compound_state *cstate,
1662                       struct nfsd4_destroy_session *sessionid)
1663 {
1664         struct nfsd4_session *ses;
1665         u32 status = nfserr_badsession;
1666
1667         /* Notes:
1668          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1669          * - Should we return nfserr_back_chan_busy if waiting for
1670          *   callbacks on to-be-destroyed session?
1671          * - Do we need to clear any callback info from previous session?
1672          */
1673
1674         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1675                 if (!nfsd4_last_compound_op(r))
1676                         return nfserr_not_only_op;
1677         }
1678         dump_sessionid(__func__, &sessionid->sessionid);
1679         spin_lock(&client_lock);
1680         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1681         if (!ses) {
1682                 spin_unlock(&client_lock);
1683                 goto out;
1684         }
1685
1686         unhash_session(ses);
1687         spin_unlock(&client_lock);
1688
1689         nfs4_lock_state();
1690         nfsd4_probe_callback_sync(ses->se_client);
1691         nfs4_unlock_state();
1692
1693         nfsd4_del_conns(ses);
1694
1695         nfsd4_put_session(ses);
1696         status = nfs_ok;
1697 out:
1698         dprintk("%s returns %d\n", __func__, ntohl(status));
1699         return status;
1700 }
1701
1702 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1703 {
1704         struct nfsd4_conn *c;
1705
1706         list_for_each_entry(c, &s->se_conns, cn_persession) {
1707                 if (c->cn_xprt == xpt) {
1708                         return c;
1709                 }
1710         }
1711         return NULL;
1712 }
1713
1714 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1715 {
1716         struct nfs4_client *clp = ses->se_client;
1717         struct nfsd4_conn *c;
1718         int ret;
1719
1720         spin_lock(&clp->cl_lock);
1721         c = __nfsd4_find_conn(new->cn_xprt, ses);
1722         if (c) {
1723                 spin_unlock(&clp->cl_lock);
1724                 free_conn(new);
1725                 return;
1726         }
1727         __nfsd4_hash_conn(new, ses);
1728         spin_unlock(&clp->cl_lock);
1729         ret = nfsd4_register_conn(new);
1730         if (ret)
1731                 /* oops; xprt is already down: */
1732                 nfsd4_conn_lost(&new->cn_xpt_user);
1733         return;
1734 }
1735
1736 __be32
1737 nfsd4_sequence(struct svc_rqst *rqstp,
1738                struct nfsd4_compound_state *cstate,
1739                struct nfsd4_sequence *seq)
1740 {
1741         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1742         struct nfsd4_session *session;
1743         struct nfsd4_slot *slot;
1744         struct nfsd4_conn *conn;
1745         int status;
1746
1747         if (resp->opcnt != 1)
1748                 return nfserr_sequence_pos;
1749
1750         /*
1751          * Will be either used or freed by nfsd4_sequence_check_conn
1752          * below.
1753          */
1754         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1755         if (!conn)
1756                 return nfserr_jukebox;
1757
1758         spin_lock(&client_lock);
1759         status = nfserr_badsession;
1760         session = find_in_sessionid_hashtbl(&seq->sessionid);
1761         if (!session)
1762                 goto out;
1763
1764         status = nfserr_badslot;
1765         if (seq->slotid >= session->se_fchannel.maxreqs)
1766                 goto out;
1767
1768         slot = session->se_slots[seq->slotid];
1769         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1770
1771         /* We do not negotiate the number of slots yet, so set the
1772          * maxslots to the session maxreqs which is used to encode
1773          * sr_highest_slotid and the sr_target_slot id to maxslots */
1774         seq->maxslots = session->se_fchannel.maxreqs;
1775
1776         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1777         if (status == nfserr_replay_cache) {
1778                 cstate->slot = slot;
1779                 cstate->session = session;
1780                 /* Return the cached reply status and set cstate->status
1781                  * for nfsd4_proc_compound processing */
1782                 status = nfsd4_replay_cache_entry(resp, seq);
1783                 cstate->status = nfserr_replay_cache;
1784                 goto out;
1785         }
1786         if (status)
1787                 goto out;
1788
1789         nfsd4_sequence_check_conn(conn, session);
1790         conn = NULL;
1791
1792         /* Success! bump slot seqid */
1793         slot->sl_inuse = true;
1794         slot->sl_seqid = seq->seqid;
1795         slot->sl_cachethis = seq->cachethis;
1796
1797         cstate->slot = slot;
1798         cstate->session = session;
1799
1800 out:
1801         /* Hold a session reference until done processing the compound. */
1802         if (cstate->session) {
1803                 struct nfs4_client *clp = session->se_client;
1804
1805                 nfsd4_get_session(cstate->session);
1806                 atomic_inc(&clp->cl_refcount);
1807                 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1808                         seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1809         }
1810         kfree(conn);
1811         spin_unlock(&client_lock);
1812         dprintk("%s: return %d\n", __func__, ntohl(status));
1813         return status;
1814 }
1815
1816 __be32
1817 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1818 {
1819         if (rc->rca_one_fs) {
1820                 if (!cstate->current_fh.fh_dentry)
1821                         return nfserr_nofilehandle;
1822                 /*
1823                  * We don't take advantage of the rca_one_fs case.
1824                  * That's OK, it's optional, we can safely ignore it.
1825                  */
1826                  return nfs_ok;
1827         }
1828         nfs4_lock_state();
1829         if (is_client_expired(cstate->session->se_client)) {
1830                 nfs4_unlock_state();
1831                 /*
1832                  * The following error isn't really legal.
1833                  * But we only get here if the client just explicitly
1834                  * destroyed the client.  Surely it no longer cares what
1835                  * error it gets back on an operation for the dead
1836                  * client.
1837                  */
1838                 return nfserr_stale_clientid;
1839         }
1840         nfsd4_create_clid_dir(cstate->session->se_client);
1841         nfs4_unlock_state();
1842         return nfs_ok;
1843 }
1844
1845 __be32
1846 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1847                   struct nfsd4_setclientid *setclid)
1848 {
1849         struct xdr_netobj       clname = { 
1850                 .len = setclid->se_namelen,
1851                 .data = setclid->se_name,
1852         };
1853         nfs4_verifier           clverifier = setclid->se_verf;
1854         unsigned int            strhashval;
1855         struct nfs4_client      *conf, *unconf, *new;
1856         __be32                  status;
1857         char                    dname[HEXDIR_LEN];
1858         
1859         if (!check_name(clname))
1860                 return nfserr_inval;
1861
1862         status = nfs4_make_rec_clidname(dname, &clname);
1863         if (status)
1864                 return status;
1865
1866         /* 
1867          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1868          * We get here on a DRC miss.
1869          */
1870
1871         strhashval = clientstr_hashval(dname);
1872
1873         nfs4_lock_state();
1874         conf = find_confirmed_client_by_str(dname, strhashval);
1875         if (conf) {
1876                 /* RFC 3530 14.2.33 CASE 0: */
1877                 status = nfserr_clid_inuse;
1878                 if (clp_used_exchangeid(conf))
1879                         goto out;
1880                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1881                         char addr_str[INET6_ADDRSTRLEN];
1882                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1883                                  sizeof(addr_str));
1884                         dprintk("NFSD: setclientid: string in use by client "
1885                                 "at %s\n", addr_str);
1886                         goto out;
1887                 }
1888         }
1889         /*
1890          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1891          * has a description of SETCLIENTID request processing consisting
1892          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1893          */
1894         unconf = find_unconfirmed_client_by_str(dname, strhashval);
1895         status = nfserr_resource;
1896         if (!conf) {
1897                 /*
1898                  * RFC 3530 14.2.33 CASE 4:
1899                  * placed first, because it is the normal case
1900                  */
1901                 if (unconf)
1902                         expire_client(unconf);
1903                 new = create_client(clname, dname, rqstp, &clverifier);
1904                 if (new == NULL)
1905                         goto out;
1906                 gen_clid(new);
1907         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1908                 /*
1909                  * RFC 3530 14.2.33 CASE 1:
1910                  * probable callback update
1911                  */
1912                 if (unconf) {
1913                         /* Note this is removing unconfirmed {*x***},
1914                          * which is stronger than RFC recommended {vxc**}.
1915                          * This has the advantage that there is at most
1916                          * one {*x***} in either list at any time.
1917                          */
1918                         expire_client(unconf);
1919                 }
1920                 new = create_client(clname, dname, rqstp, &clverifier);
1921                 if (new == NULL)
1922                         goto out;
1923                 copy_clid(new, conf);
1924         } else if (!unconf) {
1925                 /*
1926                  * RFC 3530 14.2.33 CASE 2:
1927                  * probable client reboot; state will be removed if
1928                  * confirmed.
1929                  */
1930                 new = create_client(clname, dname, rqstp, &clverifier);
1931                 if (new == NULL)
1932                         goto out;
1933                 gen_clid(new);
1934         } else {
1935                 /*
1936                  * RFC 3530 14.2.33 CASE 3:
1937                  * probable client reboot; state will be removed if
1938                  * confirmed.
1939                  */
1940                 expire_client(unconf);
1941                 new = create_client(clname, dname, rqstp, &clverifier);
1942                 if (new == NULL)
1943                         goto out;
1944                 gen_clid(new);
1945         }
1946         /*
1947          * XXX: we should probably set this at creation time, and check
1948          * for consistent minorversion use throughout:
1949          */
1950         new->cl_minorversion = 0;
1951         gen_callback(new, setclid, rqstp);
1952         add_to_unconfirmed(new, strhashval);
1953         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1954         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1955         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1956         status = nfs_ok;
1957 out:
1958         nfs4_unlock_state();
1959         return status;
1960 }
1961
1962
1963 /*
1964  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1965  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1966  * bullets, labeled as CASE1 - CASE4 below.
1967  */
1968 __be32
1969 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1970                          struct nfsd4_compound_state *cstate,
1971                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1972 {
1973         struct sockaddr *sa = svc_addr(rqstp);
1974         struct nfs4_client *conf, *unconf;
1975         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1976         clientid_t * clid = &setclientid_confirm->sc_clientid;
1977         __be32 status;
1978
1979         if (STALE_CLIENTID(clid))
1980                 return nfserr_stale_clientid;
1981         /* 
1982          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1983          * We get here on a DRC miss.
1984          */
1985
1986         nfs4_lock_state();
1987
1988         conf = find_confirmed_client(clid);
1989         unconf = find_unconfirmed_client(clid);
1990
1991         status = nfserr_clid_inuse;
1992         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1993                 goto out;
1994         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1995                 goto out;
1996
1997         /*
1998          * section 14.2.34 of RFC 3530 has a description of
1999          * SETCLIENTID_CONFIRM request processing consisting
2000          * of 4 bullet points, labeled as CASE1 - CASE4 below.
2001          */
2002         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2003                 /*
2004                  * RFC 3530 14.2.34 CASE 1:
2005                  * callback update
2006                  */
2007                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2008                         status = nfserr_clid_inuse;
2009                 else {
2010                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2011                         nfsd4_probe_callback(conf);
2012                         expire_client(unconf);
2013                         status = nfs_ok;
2014
2015                 }
2016         } else if (conf && !unconf) {
2017                 /*
2018                  * RFC 3530 14.2.34 CASE 2:
2019                  * probable retransmitted request; play it safe and
2020                  * do nothing.
2021                  */
2022                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2023                         status = nfserr_clid_inuse;
2024                 else
2025                         status = nfs_ok;
2026         } else if (!conf && unconf
2027                         && same_verf(&unconf->cl_confirm, &confirm)) {
2028                 /*
2029                  * RFC 3530 14.2.34 CASE 3:
2030                  * Normal case; new or rebooted client:
2031                  */
2032                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2033                         status = nfserr_clid_inuse;
2034                 } else {
2035                         unsigned int hash =
2036                                 clientstr_hashval(unconf->cl_recdir);
2037                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2038                                                             hash);
2039                         if (conf) {
2040                                 nfsd4_remove_clid_dir(conf);
2041                                 expire_client(conf);
2042                         }
2043                         move_to_confirmed(unconf);
2044                         conf = unconf;
2045                         nfsd4_probe_callback(conf);
2046                         status = nfs_ok;
2047                 }
2048         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2049             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2050                                                                 &confirm)))) {
2051                 /*
2052                  * RFC 3530 14.2.34 CASE 4:
2053                  * Client probably hasn't noticed that we rebooted yet.
2054                  */
2055                 status = nfserr_stale_clientid;
2056         } else {
2057                 /* check that we have hit one of the cases...*/
2058                 status = nfserr_clid_inuse;
2059         }
2060 out:
2061         nfs4_unlock_state();
2062         return status;
2063 }
2064
2065 /* OPEN Share state helper functions */
2066 static inline struct nfs4_file *
2067 alloc_init_file(struct inode *ino)
2068 {
2069         struct nfs4_file *fp;
2070         unsigned int hashval = file_hashval(ino);
2071
2072         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2073         if (fp) {
2074                 atomic_set(&fp->fi_ref, 1);
2075                 INIT_LIST_HEAD(&fp->fi_hash);
2076                 INIT_LIST_HEAD(&fp->fi_stateids);
2077                 INIT_LIST_HEAD(&fp->fi_delegations);
2078                 fp->fi_inode = igrab(ino);
2079                 fp->fi_id = current_fileid++;
2080                 fp->fi_had_conflict = false;
2081                 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2082                 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2083                 spin_lock(&recall_lock);
2084                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2085                 spin_unlock(&recall_lock);
2086                 return fp;
2087         }
2088         return NULL;
2089 }
2090
2091 static void
2092 nfsd4_free_slab(struct kmem_cache **slab)
2093 {
2094         if (*slab == NULL)
2095                 return;
2096         kmem_cache_destroy(*slab);
2097         *slab = NULL;
2098 }
2099
2100 void
2101 nfsd4_free_slabs(void)
2102 {
2103         nfsd4_free_slab(&stateowner_slab);
2104         nfsd4_free_slab(&file_slab);
2105         nfsd4_free_slab(&stateid_slab);
2106         nfsd4_free_slab(&deleg_slab);
2107 }
2108
2109 static int
2110 nfsd4_init_slabs(void)
2111 {
2112         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2113                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
2114         if (stateowner_slab == NULL)
2115                 goto out_nomem;
2116         file_slab = kmem_cache_create("nfsd4_files",
2117                         sizeof(struct nfs4_file), 0, 0, NULL);
2118         if (file_slab == NULL)
2119                 goto out_nomem;
2120         stateid_slab = kmem_cache_create("nfsd4_stateids",
2121                         sizeof(struct nfs4_stateid), 0, 0, NULL);
2122         if (stateid_slab == NULL)
2123                 goto out_nomem;
2124         deleg_slab = kmem_cache_create("nfsd4_delegations",
2125                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2126         if (deleg_slab == NULL)
2127                 goto out_nomem;
2128         return 0;
2129 out_nomem:
2130         nfsd4_free_slabs();
2131         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2132         return -ENOMEM;
2133 }
2134
2135 void
2136 nfs4_free_stateowner(struct kref *kref)
2137 {
2138         struct nfs4_stateowner *sop =
2139                 container_of(kref, struct nfs4_stateowner, so_ref);
2140         kfree(sop->so_owner.data);
2141         kmem_cache_free(stateowner_slab, sop);
2142 }
2143
2144 static inline struct nfs4_stateowner *
2145 alloc_stateowner(struct xdr_netobj *owner)
2146 {
2147         struct nfs4_stateowner *sop;
2148
2149         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2150                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2151                         memcpy(sop->so_owner.data, owner->data, owner->len);
2152                         sop->so_owner.len = owner->len;
2153                         kref_init(&sop->so_ref);
2154                         return sop;
2155                 } 
2156                 kmem_cache_free(stateowner_slab, sop);
2157         }
2158         return NULL;
2159 }
2160
2161 static struct nfs4_stateowner *
2162 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2163         struct nfs4_stateowner *sop;
2164         struct nfs4_replay *rp;
2165         unsigned int idhashval;
2166
2167         if (!(sop = alloc_stateowner(&open->op_owner)))
2168                 return NULL;
2169         idhashval = ownerid_hashval(current_ownerid);
2170         INIT_LIST_HEAD(&sop->so_idhash);
2171         INIT_LIST_HEAD(&sop->so_strhash);
2172         INIT_LIST_HEAD(&sop->so_perclient);
2173         INIT_LIST_HEAD(&sop->so_stateids);
2174         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
2175         INIT_LIST_HEAD(&sop->so_close_lru);
2176         sop->so_time = 0;
2177         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2178         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2179         list_add(&sop->so_perclient, &clp->cl_openowners);
2180         sop->so_is_open_owner = 1;
2181         sop->so_id = current_ownerid++;
2182         sop->so_client = clp;
2183         sop->so_seqid = open->op_seqid;
2184         sop->so_confirmed = 0;
2185         rp = &sop->so_replay;
2186         rp->rp_status = nfserr_serverfault;
2187         rp->rp_buflen = 0;
2188         rp->rp_buf = rp->rp_ibuf;
2189         return sop;
2190 }
2191
2192 static inline void
2193 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2194         struct nfs4_stateowner *sop = open->op_stateowner;
2195         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2196
2197         INIT_LIST_HEAD(&stp->st_hash);
2198         INIT_LIST_HEAD(&stp->st_perstateowner);
2199         INIT_LIST_HEAD(&stp->st_lockowners);
2200         INIT_LIST_HEAD(&stp->st_perfile);
2201         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2202         list_add(&stp->st_perstateowner, &sop->so_stateids);
2203         list_add(&stp->st_perfile, &fp->fi_stateids);
2204         stp->st_stateowner = sop;
2205         get_nfs4_file(fp);
2206         stp->st_file = fp;
2207         stp->st_stateid.si_boot = boot_time;
2208         stp->st_stateid.si_stateownerid = sop->so_id;
2209         stp->st_stateid.si_fileid = fp->fi_id;
2210         stp->st_stateid.si_generation = 0;
2211         stp->st_access_bmap = 0;
2212         stp->st_deny_bmap = 0;
2213         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2214                   &stp->st_access_bmap);
2215         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2216         stp->st_openstp = NULL;
2217 }
2218
2219 static void
2220 move_to_close_lru(struct nfs4_stateowner *sop)
2221 {
2222         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2223
2224         list_move_tail(&sop->so_close_lru, &close_lru);
2225         sop->so_time = get_seconds();
2226 }
2227
2228 static int
2229 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2230                                                         clientid_t *clid)
2231 {
2232         return (sop->so_owner.len == owner->len) &&
2233                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2234                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2235 }
2236
2237 static struct nfs4_stateowner *
2238 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2239 {
2240         struct nfs4_stateowner *so = NULL;
2241
2242         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2243                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2244                         return so;
2245         }
2246         return NULL;
2247 }
2248
2249 /* search file_hashtbl[] for file */
2250 static struct nfs4_file *
2251 find_file(struct inode *ino)
2252 {
2253         unsigned int hashval = file_hashval(ino);
2254         struct nfs4_file *fp;
2255
2256         spin_lock(&recall_lock);
2257         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2258                 if (fp->fi_inode == ino) {
2259                         get_nfs4_file(fp);
2260                         spin_unlock(&recall_lock);
2261                         return fp;
2262                 }
2263         }
2264         spin_unlock(&recall_lock);
2265         return NULL;
2266 }
2267
2268 static inline int access_valid(u32 x, u32 minorversion)
2269 {
2270         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2271                 return 0;
2272         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2273                 return 0;
2274         x &= ~NFS4_SHARE_ACCESS_MASK;
2275         if (minorversion && x) {
2276                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2277                         return 0;
2278                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2279                         return 0;
2280                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2281         }
2282         if (x)
2283                 return 0;
2284         return 1;
2285 }
2286
2287 static inline int deny_valid(u32 x)
2288 {
2289         /* Note: unlike access bits, deny bits may be zero. */
2290         return x <= NFS4_SHARE_DENY_BOTH;
2291 }
2292
2293 /*
2294  * Called to check deny when READ with all zero stateid or
2295  * WRITE with all zero or all one stateid
2296  */
2297 static __be32
2298 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2299 {
2300         struct inode *ino = current_fh->fh_dentry->d_inode;
2301         struct nfs4_file *fp;
2302         struct nfs4_stateid *stp;
2303         __be32 ret;
2304
2305         dprintk("NFSD: nfs4_share_conflict\n");
2306
2307         fp = find_file(ino);
2308         if (!fp)
2309                 return nfs_ok;
2310         ret = nfserr_locked;
2311         /* Search for conflicting share reservations */
2312         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2313                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2314                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2315                         goto out;
2316         }
2317         ret = nfs_ok;
2318 out:
2319         put_nfs4_file(fp);
2320         return ret;
2321 }
2322
2323 static inline void
2324 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2325 {
2326         if (share_access & NFS4_SHARE_ACCESS_WRITE)
2327                 nfs4_file_put_access(fp, O_WRONLY);
2328         if (share_access & NFS4_SHARE_ACCESS_READ)
2329                 nfs4_file_put_access(fp, O_RDONLY);
2330 }
2331
2332 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2333 {
2334         /* We're assuming the state code never drops its reference
2335          * without first removing the lease.  Since we're in this lease
2336          * callback (and since the lease code is serialized by the kernel
2337          * lock) we know the server hasn't removed the lease yet, we know
2338          * it's safe to take a reference: */
2339         atomic_inc(&dp->dl_count);
2340
2341         spin_lock(&recall_lock);
2342         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2343         spin_unlock(&recall_lock);
2344
2345         /* only place dl_time is set. protected by lock_flocks*/
2346         dp->dl_time = get_seconds();
2347
2348         nfsd4_cb_recall(dp);
2349 }
2350
2351 /*
2352  * Called from break_lease() with lock_flocks() held.
2353  * Note: we assume break_lease will only call this *once* for any given
2354  * lease.
2355  */
2356 static void nfsd_break_deleg_cb(struct file_lock *fl)
2357 {
2358         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2359
2360         BUG_ON(!dp);
2361         /*
2362          * We don't want the locks code to timeout the lease for us;
2363          * we'll remove it ourself if the delegation isn't returned
2364          * in time:
2365          */
2366         fl->fl_break_time = 0;
2367
2368         nfsd_break_one_deleg(dp);
2369         dp->dl_file->fi_had_conflict = true;
2370 }
2371
2372 static
2373 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2374 {
2375         if (arg & F_UNLCK)
2376                 return lease_modify(onlist, arg);
2377         else
2378                 return -EAGAIN;
2379 }
2380
2381 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2382         .fl_break = nfsd_break_deleg_cb,
2383         .fl_change = nfsd_change_deleg_cb,
2384 };
2385
2386
2387 __be32
2388 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2389                     struct nfsd4_open *open)
2390 {
2391         clientid_t *clientid = &open->op_clientid;
2392         struct nfs4_client *clp = NULL;
2393         unsigned int strhashval;
2394         struct nfs4_stateowner *sop = NULL;
2395
2396         if (!check_name(open->op_owner))
2397                 return nfserr_inval;
2398
2399         if (STALE_CLIENTID(&open->op_clientid))
2400                 return nfserr_stale_clientid;
2401
2402         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2403         sop = find_openstateowner_str(strhashval, open);
2404         open->op_stateowner = sop;
2405         if (!sop) {
2406                 /* Make sure the client's lease hasn't expired. */
2407                 clp = find_confirmed_client(clientid);
2408                 if (clp == NULL)
2409                         return nfserr_expired;
2410                 goto renew;
2411         }
2412         /* When sessions are used, skip open sequenceid processing */
2413         if (nfsd4_has_session(cstate))
2414                 goto renew;
2415         if (!sop->so_confirmed) {
2416                 /* Replace unconfirmed owners without checking for replay. */
2417                 clp = sop->so_client;
2418                 release_openowner(sop);
2419                 open->op_stateowner = NULL;
2420                 goto renew;
2421         }
2422         if (open->op_seqid == sop->so_seqid - 1) {
2423                 if (sop->so_replay.rp_buflen)
2424                         return nfserr_replay_me;
2425                 /* The original OPEN failed so spectacularly
2426                  * that we don't even have replay data saved!
2427                  * Therefore, we have no choice but to continue
2428                  * processing this OPEN; presumably, we'll
2429                  * fail again for the same reason.
2430                  */
2431                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2432                 goto renew;
2433         }
2434         if (open->op_seqid != sop->so_seqid)
2435                 return nfserr_bad_seqid;
2436 renew:
2437         if (open->op_stateowner == NULL) {
2438                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2439                 if (sop == NULL)
2440                         return nfserr_resource;
2441                 open->op_stateowner = sop;
2442         }
2443         list_del_init(&sop->so_close_lru);
2444         renew_client(sop->so_client);
2445         return nfs_ok;
2446 }
2447
2448 static inline __be32
2449 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2450 {
2451         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2452                 return nfserr_openmode;
2453         else
2454                 return nfs_ok;
2455 }
2456
2457 static struct nfs4_delegation *
2458 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2459 {
2460         struct nfs4_delegation *dp;
2461
2462         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2463                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2464                         return dp;
2465         }
2466         return NULL;
2467 }
2468
2469 int share_access_to_flags(u32 share_access)
2470 {
2471         share_access &= ~NFS4_SHARE_WANT_MASK;
2472
2473         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2474 }
2475
2476 static __be32
2477 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2478                 struct nfs4_delegation **dp)
2479 {
2480         int flags;
2481         __be32 status = nfserr_bad_stateid;
2482
2483         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2484         if (*dp == NULL)
2485                 goto out;
2486         flags = share_access_to_flags(open->op_share_access);
2487         status = nfs4_check_delegmode(*dp, flags);
2488         if (status)
2489                 *dp = NULL;
2490 out:
2491         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2492                 return nfs_ok;
2493         if (status)
2494                 return status;
2495         open->op_stateowner->so_confirmed = 1;
2496         return nfs_ok;
2497 }
2498
2499 static __be32
2500 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2501 {
2502         struct nfs4_stateid *local;
2503         __be32 status = nfserr_share_denied;
2504         struct nfs4_stateowner *sop = open->op_stateowner;
2505
2506         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2507                 /* ignore lock owners */
2508                 if (local->st_stateowner->so_is_open_owner == 0)
2509                         continue;
2510                 /* remember if we have seen this open owner */
2511                 if (local->st_stateowner == sop)
2512                         *stpp = local;
2513                 /* check for conflicting share reservations */
2514                 if (!test_share(local, open))
2515                         goto out;
2516         }
2517         status = 0;
2518 out:
2519         return status;
2520 }
2521
2522 static inline struct nfs4_stateid *
2523 nfs4_alloc_stateid(void)
2524 {
2525         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2526 }
2527
2528 static inline int nfs4_access_to_access(u32 nfs4_access)
2529 {
2530         int flags = 0;
2531
2532         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2533                 flags |= NFSD_MAY_READ;
2534         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2535                 flags |= NFSD_MAY_WRITE;
2536         return flags;
2537 }
2538
2539 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2540 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2541 {
2542         __be32 status;
2543         int oflag = nfs4_access_to_omode(nfs4_access);
2544         int access = nfs4_access_to_access(nfs4_access);
2545
2546         if (!fp->fi_fds[oflag]) {
2547                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2548                         &fp->fi_fds[oflag]);
2549                 if (status)
2550                         return status;
2551         }
2552         nfs4_file_get_access(fp, oflag);
2553
2554         return nfs_ok;
2555 }
2556
2557 static __be32
2558 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2559                 struct nfs4_file *fp, struct svc_fh *cur_fh,
2560                 struct nfsd4_open *open)
2561 {
2562         struct nfs4_stateid *stp;
2563         __be32 status;
2564
2565         stp = nfs4_alloc_stateid();
2566         if (stp == NULL)
2567                 return nfserr_resource;
2568
2569         status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2570         if (status) {
2571                 kmem_cache_free(stateid_slab, stp);
2572                 return status;
2573         }
2574         *stpp = stp;
2575         return 0;
2576 }
2577
2578 static inline __be32
2579 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2580                 struct nfsd4_open *open)
2581 {
2582         struct iattr iattr = {
2583                 .ia_valid = ATTR_SIZE,
2584                 .ia_size = 0,
2585         };
2586         if (!open->op_truncate)
2587                 return 0;
2588         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2589                 return nfserr_inval;
2590         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2591 }
2592
2593 static __be32
2594 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2595 {
2596         u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2597         bool new_access;
2598         __be32 status;
2599
2600         new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2601         if (new_access) {
2602                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2603                 if (status)
2604                         return status;
2605         }
2606         status = nfsd4_truncate(rqstp, cur_fh, open);
2607         if (status) {
2608                 if (new_access) {
2609                         int oflag = nfs4_access_to_omode(new_access);
2610                         nfs4_file_put_access(fp, oflag);
2611                 }
2612                 return status;
2613         }
2614         /* remember the open */
2615         __set_bit(op_share_access, &stp->st_access_bmap);
2616         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2617
2618         return nfs_ok;
2619 }
2620
2621
2622 static void
2623 nfs4_set_claim_prev(struct nfsd4_open *open)
2624 {
2625         open->op_stateowner->so_confirmed = 1;
2626         open->op_stateowner->so_client->cl_firststate = 1;
2627 }
2628
2629 /* Should we give out recallable state?: */
2630 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2631 {
2632         if (clp->cl_cb_state == NFSD4_CB_UP)
2633                 return true;
2634         /*
2635          * In the sessions case, since we don't have to establish a
2636          * separate connection for callbacks, we assume it's OK
2637          * until we hear otherwise:
2638          */
2639         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2640 }
2641
2642 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2643 {
2644         struct file_lock *fl;
2645
2646         fl = locks_alloc_lock();
2647         if (!fl)
2648                 return NULL;
2649         locks_init_lock(fl);
2650         fl->fl_lmops = &nfsd_lease_mng_ops;
2651         fl->fl_flags = FL_LEASE;
2652         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2653         fl->fl_end = OFFSET_MAX;
2654         fl->fl_owner = (fl_owner_t)dp;
2655         fl->fl_file = dp->dl_vfs_file;
2656         BUG_ON(!fl->fl_file);
2657         fl->fl_pid = current->tgid;
2658         dp->dl_flock = fl;
2659         return fl;
2660 }
2661
2662 /*
2663  * Attempt to hand out a delegation.
2664  */
2665 static void
2666 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2667 {
2668         struct nfs4_delegation *dp;
2669         struct nfs4_stateowner *sop = stp->st_stateowner;
2670         int cb_up;
2671         struct file_lock *fl;
2672         int status, flag = 0;
2673
2674         cb_up = nfsd4_cb_channel_good(sop->so_client);
2675         flag = NFS4_OPEN_DELEGATE_NONE;
2676         open->op_recall = 0;
2677         switch (open->op_claim_type) {
2678                 case NFS4_OPEN_CLAIM_PREVIOUS:
2679                         if (!cb_up)
2680                                 open->op_recall = 1;
2681                         flag = open->op_delegate_type;
2682                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2683                                 goto out;
2684                         break;
2685                 case NFS4_OPEN_CLAIM_NULL:
2686                         /* Let's not give out any delegations till everyone's
2687                          * had the chance to reclaim theirs.... */
2688                         if (locks_in_grace())
2689                                 goto out;
2690                         if (!cb_up || !sop->so_confirmed)
2691                                 goto out;
2692                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2693                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2694                         else
2695                                 flag = NFS4_OPEN_DELEGATE_READ;
2696                         break;
2697                 default:
2698                         goto out;
2699         }
2700
2701         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2702         if (dp == NULL) {
2703                 flag = NFS4_OPEN_DELEGATE_NONE;
2704                 goto out;
2705         }
2706         status = -ENOMEM;
2707         fl = nfs4_alloc_init_lease(dp, flag);
2708         if (!fl)
2709                 goto out;
2710         /* vfs_setlease checks to see if delegation should be handed out.
2711          * the lock_manager callback fl_change is used
2712          */
2713         if ((status = vfs_setlease(fl->fl_file, fl->fl_type, &fl))) {
2714                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2715                 dp->dl_flock = NULL;
2716                 locks_free_lock(fl);
2717                 unhash_delegation(dp);
2718                 flag = NFS4_OPEN_DELEGATE_NONE;
2719                 goto out;
2720         }
2721
2722         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2723
2724         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2725                 STATEID_VAL(&dp->dl_stateid));
2726 out:
2727         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2728                         && flag == NFS4_OPEN_DELEGATE_NONE
2729                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2730                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2731         open->op_delegate_type = flag;
2732 }
2733
2734 /*
2735  * called with nfs4_lock_state() held.
2736  */
2737 __be32
2738 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2739 {
2740         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2741         struct nfs4_file *fp = NULL;
2742         struct inode *ino = current_fh->fh_dentry->d_inode;
2743         struct nfs4_stateid *stp = NULL;
2744         struct nfs4_delegation *dp = NULL;
2745         __be32 status;
2746
2747         status = nfserr_inval;
2748         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2749                         || !deny_valid(open->op_share_deny))
2750                 goto out;
2751         /*
2752          * Lookup file; if found, lookup stateid and check open request,
2753          * and check for delegations in the process of being recalled.
2754          * If not found, create the nfs4_file struct
2755          */
2756         fp = find_file(ino);
2757         if (fp) {
2758                 if ((status = nfs4_check_open(fp, open, &stp)))
2759                         goto out;
2760                 status = nfs4_check_deleg(fp, open, &dp);
2761                 if (status)
2762                         goto out;
2763         } else {
2764                 status = nfserr_bad_stateid;
2765                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2766                         goto out;
2767                 status = nfserr_resource;
2768                 fp = alloc_init_file(ino);
2769                 if (fp == NULL)
2770                         goto out;
2771         }
2772
2773         /*
2774          * OPEN the file, or upgrade an existing OPEN.
2775          * If truncate fails, the OPEN fails.
2776          */
2777         if (stp) {
2778                 /* Stateid was found, this is an OPEN upgrade */
2779                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2780                 if (status)
2781                         goto out;
2782                 update_stateid(&stp->st_stateid);
2783         } else {
2784                 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2785                 if (status)
2786                         goto out;
2787                 init_stateid(stp, fp, open);
2788                 status = nfsd4_truncate(rqstp, current_fh, open);
2789                 if (status) {
2790                         release_open_stateid(stp);
2791                         goto out;
2792                 }
2793                 if (nfsd4_has_session(&resp->cstate))
2794                         update_stateid(&stp->st_stateid);
2795         }
2796         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2797
2798         if (nfsd4_has_session(&resp->cstate))
2799                 open->op_stateowner->so_confirmed = 1;
2800
2801         /*
2802         * Attempt to hand out a delegation. No error return, because the
2803         * OPEN succeeds even if we fail.
2804         */
2805         nfs4_open_delegation(current_fh, open, stp);
2806
2807         status = nfs_ok;
2808
2809         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2810                 STATEID_VAL(&stp->st_stateid));
2811 out:
2812         if (fp)
2813                 put_nfs4_file(fp);
2814         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2815                 nfs4_set_claim_prev(open);
2816         /*
2817         * To finish the open response, we just need to set the rflags.
2818         */
2819         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2820         if (!open->op_stateowner->so_confirmed &&
2821             !nfsd4_has_session(&resp->cstate))
2822                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2823
2824         return status;
2825 }
2826
2827 __be32
2828 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2829             clientid_t *clid)
2830 {
2831         struct nfs4_client *clp;
2832         __be32 status;
2833
2834         nfs4_lock_state();
2835         dprintk("process_renew(%08x/%08x): starting\n", 
2836                         clid->cl_boot, clid->cl_id);
2837         status = nfserr_stale_clientid;
2838         if (STALE_CLIENTID(clid))
2839                 goto out;
2840         clp = find_confirmed_client(clid);
2841         status = nfserr_expired;
2842         if (clp == NULL) {
2843                 /* We assume the client took too long to RENEW. */
2844                 dprintk("nfsd4_renew: clientid not found!\n");
2845                 goto out;
2846         }
2847         renew_client(clp);
2848         status = nfserr_cb_path_down;
2849         if (!list_empty(&clp->cl_delegations)
2850                         && clp->cl_cb_state != NFSD4_CB_UP)
2851                 goto out;
2852         status = nfs_ok;
2853 out:
2854         nfs4_unlock_state();
2855         return status;
2856 }
2857
2858 struct lock_manager nfsd4_manager = {
2859 };
2860
2861 static void
2862 nfsd4_end_grace(void)
2863 {
2864         dprintk("NFSD: end of grace period\n");
2865         nfsd4_recdir_purge_old();
2866         locks_end_grace(&nfsd4_manager);
2867         /*
2868          * Now that every NFSv4 client has had the chance to recover and
2869          * to see the (possibly new, possibly shorter) lease time, we
2870          * can safely set the next grace time to the current lease time:
2871          */
2872         nfsd4_grace = nfsd4_lease;
2873 }
2874
2875 static time_t
2876 nfs4_laundromat(void)
2877 {
2878         struct nfs4_client *clp;
2879         struct nfs4_stateowner *sop;
2880         struct nfs4_delegation *dp;
2881         struct list_head *pos, *next, reaplist;
2882         time_t cutoff = get_seconds() - nfsd4_lease;
2883         time_t t, clientid_val = nfsd4_lease;
2884         time_t u, test_val = nfsd4_lease;
2885
2886         nfs4_lock_state();
2887
2888         dprintk("NFSD: laundromat service - starting\n");
2889         if (locks_in_grace())
2890                 nfsd4_end_grace();
2891         INIT_LIST_HEAD(&reaplist);
2892         spin_lock(&client_lock);
2893         list_for_each_safe(pos, next, &client_lru) {
2894                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2895                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2896                         t = clp->cl_time - cutoff;
2897                         if (clientid_val > t)
2898                                 clientid_val = t;
2899                         break;
2900                 }
2901                 if (atomic_read(&clp->cl_refcount)) {
2902                         dprintk("NFSD: client in use (clientid %08x)\n",
2903                                 clp->cl_clientid.cl_id);
2904                         continue;
2905                 }
2906                 unhash_client_locked(clp);
2907                 list_add(&clp->cl_lru, &reaplist);
2908         }
2909         spin_unlock(&client_lock);
2910         list_for_each_safe(pos, next, &reaplist) {
2911                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2912                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2913                         clp->cl_clientid.cl_id);
2914                 nfsd4_remove_clid_dir(clp);
2915                 expire_client(clp);
2916         }
2917         spin_lock(&recall_lock);
2918         list_for_each_safe(pos, next, &del_recall_lru) {
2919                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2920                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2921                         u = dp->dl_time - cutoff;
2922                         if (test_val > u)
2923                                 test_val = u;
2924                         break;
2925                 }
2926                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2927                                     dp, dp->dl_flock);
2928                 list_move(&dp->dl_recall_lru, &reaplist);
2929         }
2930         spin_unlock(&recall_lock);
2931         list_for_each_safe(pos, next, &reaplist) {
2932                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2933                 list_del_init(&dp->dl_recall_lru);
2934                 unhash_delegation(dp);
2935         }
2936         test_val = nfsd4_lease;
2937         list_for_each_safe(pos, next, &close_lru) {
2938                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2939                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2940                         u = sop->so_time - cutoff;
2941                         if (test_val > u)
2942                                 test_val = u;
2943                         break;
2944                 }
2945                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2946                         sop->so_id);
2947                 release_openowner(sop);
2948         }
2949         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2950                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2951         nfs4_unlock_state();
2952         return clientid_val;
2953 }
2954
2955 static struct workqueue_struct *laundry_wq;
2956 static void laundromat_main(struct work_struct *);
2957 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2958
2959 static void
2960 laundromat_main(struct work_struct *not_used)
2961 {
2962         time_t t;
2963
2964         t = nfs4_laundromat();
2965         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2966         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2967 }
2968
2969 static struct nfs4_stateowner *
2970 search_close_lru(u32 st_id, int flags)
2971 {
2972         struct nfs4_stateowner *local = NULL;
2973
2974         if (flags & CLOSE_STATE) {
2975                 list_for_each_entry(local, &close_lru, so_close_lru) {
2976                         if (local->so_id == st_id)
2977                                 return local;
2978                 }
2979         }
2980         return NULL;
2981 }
2982
2983 static inline int
2984 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2985 {
2986         return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2987 }
2988
2989 static int
2990 STALE_STATEID(stateid_t *stateid)
2991 {
2992         if (stateid->si_boot == boot_time)
2993                 return 0;
2994         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2995                 STATEID_VAL(stateid));
2996         return 1;
2997 }
2998
2999 static inline int
3000 access_permit_read(unsigned long access_bmap)
3001 {
3002         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3003                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3004                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3005 }
3006
3007 static inline int
3008 access_permit_write(unsigned long access_bmap)
3009 {
3010         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3011                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3012 }
3013
3014 static
3015 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3016 {
3017         __be32 status = nfserr_openmode;
3018
3019         /* For lock stateid's, we test the parent open, not the lock: */
3020         if (stp->st_openstp)
3021                 stp = stp->st_openstp;
3022         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3023                 goto out;
3024         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3025                 goto out;
3026         status = nfs_ok;
3027 out:
3028         return status;
3029 }
3030
3031 static inline __be32
3032 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3033 {
3034         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3035                 return nfs_ok;
3036         else if (locks_in_grace()) {
3037                 /* Answer in remaining cases depends on existance of
3038                  * conflicting state; so we must wait out the grace period. */
3039                 return nfserr_grace;
3040         } else if (flags & WR_STATE)
3041                 return nfs4_share_conflict(current_fh,
3042                                 NFS4_SHARE_DENY_WRITE);
3043         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3044                 return nfs4_share_conflict(current_fh,
3045                                 NFS4_SHARE_DENY_READ);
3046 }
3047
3048 /*
3049  * Allow READ/WRITE during grace period on recovered state only for files
3050  * that are not able to provide mandatory locking.
3051  */
3052 static inline int
3053 grace_disallows_io(struct inode *inode)
3054 {
3055         return locks_in_grace() && mandatory_lock(inode);
3056 }
3057
3058 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3059 {
3060         /*
3061          * When sessions are used the stateid generation number is ignored
3062          * when it is zero.
3063          */
3064         if ((flags & HAS_SESSION) && in->si_generation == 0)
3065                 goto out;
3066
3067         /* If the client sends us a stateid from the future, it's buggy: */
3068         if (in->si_generation > ref->si_generation)
3069                 return nfserr_bad_stateid;
3070         /*
3071          * The following, however, can happen.  For example, if the
3072          * client sends an open and some IO at the same time, the open
3073          * may bump si_generation while the IO is still in flight.
3074          * Thanks to hard links and renames, the client never knows what
3075          * file an open will affect.  So it could avoid that situation
3076          * only by serializing all opens and IO from the same open
3077          * owner.  To recover from the old_stateid error, the client
3078          * will just have to retry the IO:
3079          */
3080         if (in->si_generation < ref->si_generation)
3081                 return nfserr_old_stateid;
3082 out:
3083         return nfs_ok;
3084 }
3085
3086 static int is_delegation_stateid(stateid_t *stateid)
3087 {
3088         return stateid->si_fileid == 0;
3089 }
3090
3091 /*
3092 * Checks for stateid operations
3093 */
3094 __be32
3095 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3096                            stateid_t *stateid, int flags, struct file **filpp)
3097 {
3098         struct nfs4_stateid *stp = NULL;
3099         struct nfs4_delegation *dp = NULL;
3100         struct svc_fh *current_fh = &cstate->current_fh;
3101         struct inode *ino = current_fh->fh_dentry->d_inode;
3102         __be32 status;
3103
3104         if (filpp)
3105                 *filpp = NULL;
3106
3107         if (grace_disallows_io(ino))
3108                 return nfserr_grace;
3109
3110         if (nfsd4_has_session(cstate))
3111                 flags |= HAS_SESSION;
3112
3113         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3114                 return check_special_stateids(current_fh, stateid, flags);
3115
3116         status = nfserr_stale_stateid;
3117         if (STALE_STATEID(stateid)) 
3118                 goto out;
3119
3120         /*
3121          * We assume that any stateid that has the current boot time,
3122          * but that we can't find, is expired:
3123          */
3124         status = nfserr_expired;
3125         if (is_delegation_stateid(stateid)) {
3126                 dp = find_delegation_stateid(ino, stateid);
3127                 if (!dp)
3128                         goto out;
3129                 status = check_stateid_generation(stateid, &dp->dl_stateid,
3130                                                   flags);
3131                 if (status)
3132                         goto out;
3133                 status = nfs4_check_delegmode(dp, flags);
3134                 if (status)
3135                         goto out;
3136                 renew_client(dp->dl_client);
3137                 if (filpp) {
3138                         *filpp = find_readable_file(dp->dl_file);
3139                         BUG_ON(!*filpp);
3140                 }
3141         } else { /* open or lock stateid */
3142                 stp = find_stateid(stateid, flags);
3143                 if (!stp)
3144                         goto out;
3145                 status = nfserr_bad_stateid;
3146                 if (nfs4_check_fh(current_fh, stp))
3147                         goto out;
3148                 if (!stp->st_stateowner->so_confirmed)
3149                         goto out;
3150                 status = check_stateid_generation(stateid, &stp->st_stateid,
3151                                                   flags);
3152                 if (status)
3153                         goto out;
3154                 status = nfs4_check_openmode(stp, flags);
3155                 if (status)
3156                         goto out;
3157                 renew_client(stp->st_stateowner->so_client);
3158                 if (filpp) {
3159                         if (flags & RD_STATE)
3160                                 *filpp = find_readable_file(stp->st_file);
3161                         else
3162                                 *filpp = find_writeable_file(stp->st_file);
3163                 }
3164         }
3165         status = nfs_ok;
3166 out:
3167         return status;
3168 }
3169
3170 static inline int
3171 setlkflg (int type)
3172 {
3173         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3174                 RD_STATE : WR_STATE;
3175 }
3176
3177 /* 
3178  * Checks for sequence id mutating operations. 
3179  */
3180 static __be32
3181 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3182                          stateid_t *stateid, int flags,
3183                          struct nfs4_stateowner **sopp,
3184                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3185 {
3186         struct nfs4_stateid *stp;
3187         struct nfs4_stateowner *sop;
3188         struct svc_fh *current_fh = &cstate->current_fh;
3189         __be32 status;
3190
3191         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3192                 seqid, STATEID_VAL(stateid));
3193
3194         *stpp = NULL;
3195         *sopp = NULL;
3196
3197         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3198                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3199                 return nfserr_bad_stateid;
3200         }
3201
3202         if (STALE_STATEID(stateid))
3203                 return nfserr_stale_stateid;
3204
3205         if (nfsd4_has_session(cstate))
3206                 flags |= HAS_SESSION;
3207
3208         /*
3209         * We return BAD_STATEID if filehandle doesn't match stateid, 
3210         * the confirmed flag is incorrecly set, or the generation 
3211         * number is incorrect.  
3212         */
3213         stp = find_stateid(stateid, flags);
3214         if (stp == NULL) {
3215                 /*
3216                  * Also, we should make sure this isn't just the result of
3217                  * a replayed close:
3218                  */
3219                 sop = search_close_lru(stateid->si_stateownerid, flags);
3220                 /* It's not stale; let's assume it's expired: */
3221                 if (sop == NULL)
3222                         return nfserr_expired;
3223                 *sopp = sop;
3224                 goto check_replay;
3225         }
3226
3227         *stpp = stp;
3228         *sopp = sop = stp->st_stateowner;
3229
3230         if (lock) {
3231                 clientid_t *lockclid = &lock->v.new.clientid;
3232                 struct nfs4_client *clp = sop->so_client;
3233                 int lkflg = 0;
3234                 __be32 status;
3235
3236                 lkflg = setlkflg(lock->lk_type);
3237
3238                 if (lock->lk_is_new) {
3239                         if (!sop->so_is_open_owner)
3240                                 return nfserr_bad_stateid;
3241                         if (!(flags & HAS_SESSION) &&
3242                             !same_clid(&clp->cl_clientid, lockclid))
3243                                 return nfserr_bad_stateid;
3244                         /* stp is the open stateid */
3245                         status = nfs4_check_openmode(stp, lkflg);
3246                         if (status)
3247                                 return status;
3248                 } else {
3249                         /* stp is the lock stateid */
3250                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3251                         if (status)
3252                                 return status;
3253                }
3254         }
3255
3256         if (nfs4_check_fh(current_fh, stp)) {
3257                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3258                 return nfserr_bad_stateid;
3259         }
3260
3261         /*
3262         *  We now validate the seqid and stateid generation numbers.
3263         *  For the moment, we ignore the possibility of 
3264         *  generation number wraparound.
3265         */
3266         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3267                 goto check_replay;
3268
3269         if (sop->so_confirmed && flags & CONFIRM) {
3270                 dprintk("NFSD: preprocess_seqid_op: expected"
3271                                 " unconfirmed stateowner!\n");
3272                 return nfserr_bad_stateid;
3273         }
3274         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3275                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3276                                 " confirmed yet!\n");
3277                 return nfserr_bad_stateid;
3278         }
3279         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3280         if (status)
3281                 return status;
3282         renew_client(sop->so_client);
3283         return nfs_ok;
3284
3285 check_replay:
3286         if (seqid == sop->so_seqid - 1) {
3287                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3288                 /* indicate replay to calling function */
3289                 return nfserr_replay_me;
3290         }
3291         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3292                         sop->so_seqid, seqid);
3293         *sopp = NULL;
3294         return nfserr_bad_seqid;
3295 }
3296
3297 __be32
3298 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3299                    struct nfsd4_open_confirm *oc)
3300 {
3301         __be32 status;
3302         struct nfs4_stateowner *sop;
3303         struct nfs4_stateid *stp;
3304
3305         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3306                         (int)cstate->current_fh.fh_dentry->d_name.len,
3307                         cstate->current_fh.fh_dentry->d_name.name);
3308
3309         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3310         if (status)
3311                 return status;
3312
3313         nfs4_lock_state();
3314
3315         if ((status = nfs4_preprocess_seqid_op(cstate,
3316                                         oc->oc_seqid, &oc->oc_req_stateid,
3317                                         CONFIRM | OPEN_STATE,
3318                                         &oc->oc_stateowner, &stp, NULL)))
3319                 goto out; 
3320
3321         sop = oc->oc_stateowner;
3322         sop->so_confirmed = 1;
3323         update_stateid(&stp->st_stateid);
3324         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3325         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3326                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3327
3328         nfsd4_create_clid_dir(sop->so_client);
3329 out:
3330         if (oc->oc_stateowner) {
3331                 nfs4_get_stateowner(oc->oc_stateowner);
3332                 cstate->replay_owner = oc->oc_stateowner;
3333         }
3334         nfs4_unlock_state();
3335         return status;
3336 }
3337
3338
3339 /*
3340  * unset all bits in union bitmap (bmap) that
3341  * do not exist in share (from successful OPEN_DOWNGRADE)
3342  */
3343 static void
3344 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3345 {
3346         int i;
3347         for (i = 1; i < 4; i++) {
3348                 if ((i & access) != i)
3349                         __clear_bit(i, bmap);
3350         }
3351 }
3352
3353 static void
3354 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3355 {
3356         int i;
3357         for (i = 0; i < 4; i++) {
3358                 if ((i & deny) != i)
3359                         __clear_bit(i, bmap);
3360         }
3361 }
3362
3363 __be32
3364 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3365                      struct nfsd4_compound_state *cstate,
3366                      struct nfsd4_open_downgrade *od)
3367 {
3368         __be32 status;
3369         struct nfs4_stateid *stp;
3370         unsigned int share_access;
3371
3372         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3373                         (int)cstate->current_fh.fh_dentry->d_name.len,
3374                         cstate->current_fh.fh_dentry->d_name.name);
3375
3376         if (!access_valid(od->od_share_access, cstate->minorversion)
3377                         || !deny_valid(od->od_share_deny))
3378                 return nfserr_inval;
3379
3380         nfs4_lock_state();
3381         if ((status = nfs4_preprocess_seqid_op(cstate,
3382                                         od->od_seqid,
3383                                         &od->od_stateid, 
3384                                         OPEN_STATE,
3385                                         &od->od_stateowner, &stp, NULL)))
3386                 goto out; 
3387
3388         status = nfserr_inval;
3389         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3390                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3391                         stp->st_access_bmap, od->od_share_access);
3392                 goto out;
3393         }
3394         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3395                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3396                         stp->st_deny_bmap, od->od_share_deny);
3397                 goto out;
3398         }
3399         set_access(&share_access, stp->st_access_bmap);
3400         nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3401
3402         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3403         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3404
3405         update_stateid(&stp->st_stateid);
3406         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3407         status = nfs_ok;
3408 out:
3409         if (od->od_stateowner) {
3410                 nfs4_get_stateowner(od->od_stateowner);
3411                 cstate->replay_owner = od->od_stateowner;
3412         }
3413         nfs4_unlock_state();
3414         return status;
3415 }
3416
3417 /*
3418  * nfs4_unlock_state() called after encode
3419  */
3420 __be32
3421 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3422             struct nfsd4_close *close)
3423 {
3424         __be32 status;
3425         struct nfs4_stateid *stp;
3426
3427         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3428                         (int)cstate->current_fh.fh_dentry->d_name.len,
3429                         cstate->current_fh.fh_dentry->d_name.name);
3430
3431         nfs4_lock_state();
3432         /* check close_lru for replay */
3433         if ((status = nfs4_preprocess_seqid_op(cstate,
3434                                         close->cl_seqid,
3435                                         &close->cl_stateid, 
3436                                         OPEN_STATE | CLOSE_STATE,
3437                                         &close->cl_stateowner, &stp, NULL)))
3438                 goto out; 
3439         status = nfs_ok;
3440         update_stateid(&stp->st_stateid);
3441         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3442
3443         /* release_stateid() calls nfsd_close() if needed */
3444         release_open_stateid(stp);
3445
3446         /* place unused nfs4_stateowners on so_close_lru list to be
3447          * released by the laundromat service after the lease period
3448          * to enable us to handle CLOSE replay
3449          */
3450         if (list_empty(&close->cl_stateowner->so_stateids))
3451                 move_to_close_lru(close->cl_stateowner);
3452 out:
3453         if (close->cl_stateowner) {
3454                 nfs4_get_stateowner(close->cl_stateowner);
3455                 cstate->replay_owner = close->cl_stateowner;
3456         }
3457         nfs4_unlock_state();
3458         return status;
3459 }
3460
3461 __be32
3462 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3463                   struct nfsd4_delegreturn *dr)
3464 {
3465         struct nfs4_delegation *dp;
3466         stateid_t *stateid = &dr->dr_stateid;
3467         struct inode *inode;
3468         __be32 status;
3469         int flags = 0;
3470
3471         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3472                 return status;
3473         inode = cstate->current_fh.fh_dentry->d_inode;
3474
3475         if (nfsd4_has_session(cstate))
3476                 flags |= HAS_SESSION;
3477         nfs4_lock_state();
3478         status = nfserr_bad_stateid;
3479         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3480                 goto out;
3481         status = nfserr_stale_stateid;
3482         if (STALE_STATEID(stateid))
3483                 goto out;
3484         status = nfserr_bad_stateid;
3485         if (!is_delegation_stateid(stateid))
3486                 goto out;
3487         status = nfserr_expired;
3488         dp = find_delegation_stateid(inode, stateid);
3489         if (!dp)
3490                 goto out;
3491         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3492         if (status)
3493                 goto out;
3494         renew_client(dp->dl_client);
3495
3496         unhash_delegation(dp);
3497 out:
3498         nfs4_unlock_state();
3499
3500         return status;
3501 }
3502
3503
3504 /* 
3505  * Lock owner state (byte-range locks)
3506  */
3507 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3508 #define LOCK_HASH_BITS              8
3509 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3510 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3511
3512 static inline u64
3513 end_offset(u64 start, u64 len)
3514 {
3515         u64 end;
3516
3517         end = start + len;
3518         return end >= start ? end: NFS4_MAX_UINT64;
3519 }
3520
3521 /* last octet in a range */
3522 static inline u64
3523 last_byte_offset(u64 start, u64 len)
3524 {
3525         u64 end;
3526
3527         BUG_ON(!len);
3528         end = start + len;
3529         return end > start ? end - 1: NFS4_MAX_UINT64;
3530 }
3531
3532 #define lockownerid_hashval(id) \
3533         ((id) & LOCK_HASH_MASK)
3534
3535 static inline unsigned int
3536 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3537                 struct xdr_netobj *ownername)
3538 {
3539         return (file_hashval(inode) + cl_id
3540                         + opaque_hashval(ownername->data, ownername->len))
3541                 & LOCK_HASH_MASK;
3542 }
3543
3544 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3545 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3546 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3547
3548 static struct nfs4_stateid *
3549 find_stateid(stateid_t *stid, int flags)
3550 {
3551         struct nfs4_stateid *local;
3552         u32 st_id = stid->si_stateownerid;
3553         u32 f_id = stid->si_fileid;
3554         unsigned int hashval;
3555
3556         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3557         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3558                 hashval = stateid_hashval(st_id, f_id);
3559                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3560                         if ((local->st_stateid.si_stateownerid == st_id) &&
3561                             (local->st_stateid.si_fileid == f_id))
3562                                 return local;
3563                 }
3564         } 
3565
3566         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3567                 hashval = stateid_hashval(st_id, f_id);
3568                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3569                         if ((local->st_stateid.si_stateownerid == st_id) &&
3570                             (local->st_stateid.si_fileid == f_id))
3571                                 return local;
3572                 }
3573         }
3574         return NULL;
3575 }
3576
3577 static struct nfs4_delegation *
3578 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3579 {
3580         struct nfs4_file *fp;
3581         struct nfs4_delegation *dl;
3582
3583         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3584                 STATEID_VAL(stid));
3585
3586         fp = find_file(ino);
3587         if (!fp)
3588                 return NULL;
3589         dl = find_delegation_file(fp, stid);
3590         put_nfs4_file(fp);
3591         return dl;
3592 }
3593
3594 /*
3595  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3596  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3597  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3598  * locking, this prevents us from being completely protocol-compliant.  The
3599  * real solution to this problem is to start using unsigned file offsets in
3600  * the VFS, but this is a very deep change!
3601  */
3602 static inline void
3603 nfs4_transform_lock_offset(struct file_lock *lock)
3604 {
3605         if (lock->fl_start < 0)
3606                 lock->fl_start = OFFSET_MAX;
3607         if (lock->fl_end < 0)
3608                 lock->fl_end = OFFSET_MAX;
3609 }
3610
3611 /* Hack!: For now, we're defining this just so we can use a pointer to it
3612  * as a unique cookie to identify our (NFSv4's) posix locks. */
3613 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3614 };
3615
3616 static inline void
3617 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3618 {
3619         struct nfs4_stateowner *sop;
3620
3621         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3622                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3623                 kref_get(&sop->so_ref);
3624                 deny->ld_sop = sop;
3625                 deny->ld_clientid = sop->so_client->cl_clientid;
3626         } else {
3627                 deny->ld_sop = NULL;
3628                 deny->ld_clientid.cl_boot = 0;
3629                 deny->ld_clientid.cl_id = 0;
3630         }
3631         deny->ld_start = fl->fl_start;
3632         deny->ld_length = NFS4_MAX_UINT64;
3633         if (fl->fl_end != NFS4_MAX_UINT64)
3634                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3635         deny->ld_type = NFS4_READ_LT;
3636         if (fl->fl_type != F_RDLCK)
3637                 deny->ld_type = NFS4_WRITE_LT;
3638 }
3639
3640 static struct nfs4_stateowner *
3641 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3642                 struct xdr_netobj *owner)
3643 {
3644         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3645         struct nfs4_stateowner *op;
3646
3647         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3648                 if (same_owner_str(op, owner, clid))
3649                         return op;
3650         }
3651         return NULL;
3652 }
3653
3654 /*
3655  * Alloc a lock owner structure.
3656  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3657  * occured. 
3658  *
3659  * strhashval = lock_ownerstr_hashval 
3660  */
3661
3662 static struct nfs4_stateowner *
3663 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3664         struct nfs4_stateowner *sop;
3665         struct nfs4_replay *rp;
3666         unsigned int idhashval;
3667
3668         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3669                 return NULL;
3670         idhashval = lockownerid_hashval(current_ownerid);
3671         INIT_LIST_HEAD(&sop->so_idhash);
3672         INIT_LIST_HEAD(&sop->so_strhash);
3673         INIT_LIST_HEAD(&sop->so_perclient);
3674         INIT_LIST_HEAD(&sop->so_stateids);
3675         INIT_LIST_HEAD(&sop->so_perstateid);
3676         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3677         sop->so_time = 0;
3678         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3679         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3680         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3681         sop->so_is_open_owner = 0;
3682         sop->so_id = current_ownerid++;
3683         sop->so_client = clp;
3684         /* It is the openowner seqid that will be incremented in encode in the
3685          * case of new lockowners; so increment the lock seqid manually: */
3686         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3687         sop->so_confirmed = 1;
3688         rp = &sop->so_replay;
3689         rp->rp_status = nfserr_serverfault;
3690         rp->rp_buflen = 0;
3691         rp->rp_buf = rp->rp_ibuf;
3692         return sop;
3693 }
3694
3695 static struct nfs4_stateid *
3696 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3697 {
3698         struct nfs4_stateid *stp;
3699         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3700
3701         stp = nfs4_alloc_stateid();
3702         if (stp == NULL)
3703                 goto out;
3704         INIT_LIST_HEAD(&stp->st_hash);
3705         INIT_LIST_HEAD(&stp->st_perfile);
3706         INIT_LIST_HEAD(&stp->st_perstateowner);
3707         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3708         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3709         list_add(&stp->st_perfile, &fp->fi_stateids);
3710         list_add(&stp->st_perstateowner, &sop->so_stateids);
3711         stp->st_stateowner = sop;
3712         get_nfs4_file(fp);
3713         stp->st_file = fp;
3714         stp->st_stateid.si_boot = boot_time;
3715         stp->st_stateid.si_stateownerid = sop->so_id;
3716         stp->st_stateid.si_fileid = fp->fi_id;
3717         stp->st_stateid.si_generation = 0;
3718         stp->st_deny_bmap = open_stp->st_deny_bmap;
3719         stp->st_openstp = open_stp;
3720
3721 out:
3722         return stp;
3723 }
3724
3725 static int
3726 check_lock_length(u64 offset, u64 length)
3727 {
3728         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3729              LOFF_OVERFLOW(offset, length)));
3730 }
3731
3732 /*
3733  *  LOCK operation 
3734  */
3735 __be32
3736 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3737            struct nfsd4_lock *lock)
3738 {
3739         struct nfs4_stateowner *open_sop = NULL;
3740         struct nfs4_stateowner *lock_sop = NULL;
3741         struct nfs4_stateid *lock_stp;
3742         struct nfs4_file *fp;
3743         struct file *filp = NULL;
3744         struct file_lock file_lock;
3745         struct file_lock conflock;
3746         __be32 status = 0;
3747         unsigned int strhashval;
3748         unsigned int cmd;
3749         int err;
3750
3751         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3752                 (long long) lock->lk_offset,
3753                 (long long) lock->lk_length);
3754
3755         if (check_lock_length(lock->lk_offset, lock->lk_length))
3756                  return nfserr_inval;
3757
3758         if ((status = fh_verify(rqstp, &cstate->current_fh,
3759                                 S_IFREG, NFSD_MAY_LOCK))) {
3760                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3761                 return status;
3762         }
3763
3764         nfs4_lock_state();
3765
3766         if (lock->lk_is_new) {
3767                 /*
3768                  * Client indicates that this is a new lockowner.
3769                  * Use open owner and open stateid to create lock owner and
3770                  * lock stateid.
3771                  */
3772                 struct nfs4_stateid *open_stp = NULL;
3773                 
3774                 status = nfserr_stale_clientid;
3775                 if (!nfsd4_has_session(cstate) &&
3776                     STALE_CLIENTID(&lock->lk_new_clientid))
3777                         goto out;
3778
3779                 /* validate and update open stateid and open seqid */
3780                 status = nfs4_preprocess_seqid_op(cstate,
3781                                         lock->lk_new_open_seqid,
3782                                         &lock->lk_new_open_stateid,
3783                                         OPEN_STATE,
3784                                         &lock->lk_replay_owner, &open_stp,
3785                                         lock);
3786                 if (status)
3787                         goto out;
3788                 open_sop = lock->lk_replay_owner;
3789                 /* create lockowner and lock stateid */
3790                 fp = open_stp->st_file;
3791                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3792                                 open_sop->so_client->cl_clientid.cl_id, 
3793                                 &lock->v.new.owner);
3794                 /* XXX: Do we need to check for duplicate stateowners on
3795                  * the same file, or should they just be allowed (and
3796                  * create new stateids)? */
3797                 status = nfserr_resource;
3798                 lock_sop = alloc_init_lock_stateowner(strhashval,
3799                                 open_sop->so_client, open_stp, lock);
3800                 if (lock_sop == NULL)
3801                         goto out;
3802                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3803                 if (lock_stp == NULL)
3804                         goto out;
3805         } else {
3806                 /* lock (lock owner + lock stateid) already exists */
3807                 status = nfs4_preprocess_seqid_op(cstate,
3808                                        lock->lk_old_lock_seqid, 
3809                                        &lock->lk_old_lock_stateid, 
3810                                        LOCK_STATE,
3811                                        &lock->lk_replay_owner, &lock_stp, lock);
3812                 if (status)
3813                         goto out;
3814                 lock_sop = lock->lk_replay_owner;
3815                 fp = lock_stp->st_file;
3816         }
3817         /* lock->lk_replay_owner and lock_stp have been created or found */
3818
3819         status = nfserr_grace;
3820         if (locks_in_grace() && !lock->lk_reclaim)
3821                 goto out;
3822         status = nfserr_no_grace;
3823         if (!locks_in_grace() && lock->lk_reclaim)
3824                 goto out;
3825
3826         locks_init_lock(&file_lock);
3827         switch (lock->lk_type) {
3828                 case NFS4_READ_LT:
3829                 case NFS4_READW_LT:
3830                         if (find_readable_file(lock_stp->st_file)) {
3831                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3832                                 filp = find_readable_file(lock_stp->st_file);
3833                         }
3834                         file_lock.fl_type = F_RDLCK;
3835                         cmd = F_SETLK;
3836                 break;
3837                 case NFS4_WRITE_LT:
3838                 case NFS4_WRITEW_LT:
3839                         if (find_writeable_file(lock_stp->st_file)) {
3840                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3841                                 filp = find_writeable_file(lock_stp->st_file);
3842                         }
3843                         file_lock.fl_type = F_WRLCK;
3844                         cmd = F_SETLK;
3845                 break;
3846                 default:
3847                         status = nfserr_inval;
3848                 goto out;
3849         }
3850         if (!filp) {
3851                 status = nfserr_openmode;
3852                 goto out;
3853         }
3854         file_lock.fl_owner = (fl_owner_t)lock_sop;
3855         file_lock.fl_pid = current->tgid;
3856         file_lock.fl_file = filp;
3857         file_lock.fl_flags = FL_POSIX;
3858         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3859
3860         file_lock.fl_start = lock->lk_offset;
3861         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3862         nfs4_transform_lock_offset(&file_lock);
3863
3864         /*
3865         * Try to lock the file in the VFS.
3866         * Note: locks.c uses the BKL to protect the inode's lock list.
3867         */
3868
3869         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3870         switch (-err) {
3871         case 0: /* success! */
3872                 update_stateid(&lock_stp->st_stateid);
3873                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3874                                 sizeof(stateid_t));
3875                 status = 0;
3876                 break;
3877         case (EAGAIN):          /* conflock holds conflicting lock */
3878                 status = nfserr_denied;
3879                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3880                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3881                 break;
3882         case (EDEADLK):
3883                 status = nfserr_deadlock;
3884                 break;
3885         default:        
3886                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3887                 status = nfserr_resource;
3888                 break;
3889         }
3890 out:
3891         if (status && lock->lk_is_new && lock_sop)
3892                 release_lockowner(lock_sop);
3893         if (lock->lk_replay_owner) {
3894                 nfs4_get_stateowner(lock->lk_replay_owner);
3895                 cstate->replay_owner = lock->lk_replay_owner;
3896         }
3897         nfs4_unlock_state();
3898         return status;
3899 }
3900
3901 /*
3902  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3903  * so we do a temporary open here just to get an open file to pass to
3904  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3905  * inode operation.)
3906  */
3907 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3908 {
3909         struct file *file;
3910         int err;
3911
3912         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3913         if (err)
3914                 return err;
3915         err = vfs_test_lock(file, lock);
3916         nfsd_close(file);
3917         return err;
3918 }
3919
3920 /*
3921  * LOCKT operation
3922  */
3923 __be32
3924 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3925             struct nfsd4_lockt *lockt)
3926 {
3927         struct inode *inode;
3928         struct file_lock file_lock;
3929         int error;
3930         __be32 status;
3931
3932         if (locks_in_grace())
3933                 return nfserr_grace;
3934
3935         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3936                  return nfserr_inval;
3937
3938         lockt->lt_stateowner = NULL;
3939         nfs4_lock_state();
3940
3941         status = nfserr_stale_clientid;
3942         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3943                 goto out;
3944
3945         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3946                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3947                 if (status == nfserr_symlink)
3948                         status = nfserr_inval;
3949                 goto out;
3950         }
3951
3952         inode = cstate->current_fh.fh_dentry->d_inode;
3953         locks_init_lock(&file_lock);
3954         switch (lockt->lt_type) {
3955                 case NFS4_READ_LT:
3956                 case NFS4_READW_LT:
3957                         file_lock.fl_type = F_RDLCK;
3958                 break;
3959                 case NFS4_WRITE_LT:
3960                 case NFS4_WRITEW_LT:
3961                         file_lock.fl_type = F_WRLCK;
3962                 break;
3963                 default:
3964                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3965                         status = nfserr_inval;
3966                 goto out;
3967         }
3968
3969         lockt->lt_stateowner = find_lockstateowner_str(inode,
3970                         &lockt->lt_clientid, &lockt->lt_owner);
3971         if (lockt->lt_stateowner)
3972                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3973         file_lock.fl_pid = current->tgid;
3974         file_lock.fl_flags = FL_POSIX;
3975
3976         file_lock.fl_start = lockt->lt_offset;
3977         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3978
3979         nfs4_transform_lock_offset(&file_lock);
3980
3981         status = nfs_ok;
3982         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3983         if (error) {
3984                 status = nfserrno(error);
3985                 goto out;
3986         }
3987         if (file_lock.fl_type != F_UNLCK) {
3988                 status = nfserr_denied;
3989                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3990         }
3991 out:
3992         nfs4_unlock_state();
3993         return status;
3994 }
3995
3996 __be32
3997 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3998             struct nfsd4_locku *locku)
3999 {
4000         struct nfs4_stateid *stp;
4001         struct file *filp = NULL;
4002         struct file_lock file_lock;
4003         __be32 status;
4004         int err;
4005                                                         
4006         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4007                 (long long) locku->lu_offset,
4008                 (long long) locku->lu_length);
4009
4010         if (check_lock_length(locku->lu_offset, locku->lu_length))
4011                  return nfserr_inval;
4012
4013         nfs4_lock_state();
4014                                                                                 
4015         if ((status = nfs4_preprocess_seqid_op(cstate,
4016                                         locku->lu_seqid, 
4017                                         &locku->lu_stateid, 
4018                                         LOCK_STATE,
4019                                         &locku->lu_stateowner, &stp, NULL)))
4020                 goto out;
4021
4022         filp = find_any_file(stp->st_file);
4023         if (!filp) {
4024                 status = nfserr_lock_range;
4025                 goto out;
4026         }
4027         BUG_ON(!filp);
4028         locks_init_lock(&file_lock);
4029         file_lock.fl_type = F_UNLCK;
4030         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4031         file_lock.fl_pid = current->tgid;
4032         file_lock.fl_file = filp;
4033         file_lock.fl_flags = FL_POSIX; 
4034         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4035         file_lock.fl_start = locku->lu_offset;
4036
4037         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4038         nfs4_transform_lock_offset(&file_lock);
4039
4040         /*
4041         *  Try to unlock the file in the VFS.
4042         */
4043         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4044         if (err) {
4045                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4046                 goto out_nfserr;
4047         }
4048         /*
4049         * OK, unlock succeeded; the only thing left to do is update the stateid.
4050         */
4051         update_stateid(&stp->st_stateid);
4052         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4053
4054 out:
4055         if (locku->lu_stateowner) {
4056                 nfs4_get_stateowner(locku->lu_stateowner);
4057                 cstate->replay_owner = locku->lu_stateowner;
4058         }
4059         nfs4_unlock_state();
4060         return status;
4061
4062 out_nfserr:
4063         status = nfserrno(err);
4064         goto out;
4065 }
4066
4067 /*
4068  * returns
4069  *      1: locks held by lockowner
4070  *      0: no locks held by lockowner
4071  */
4072 static int
4073 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4074 {
4075         struct file_lock **flpp;
4076         struct inode *inode = filp->fi_inode;
4077         int status = 0;
4078
4079         lock_flocks();
4080         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4081                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4082                         status = 1;
4083                         goto out;
4084                 }
4085         }
4086 out:
4087         unlock_flocks();
4088         return status;
4089 }
4090
4091 __be32
4092 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4093                         struct nfsd4_compound_state *cstate,
4094                         struct nfsd4_release_lockowner *rlockowner)
4095 {
4096         clientid_t *clid = &rlockowner->rl_clientid;
4097         struct nfs4_stateowner *sop;
4098         struct nfs4_stateid *stp;
4099         struct xdr_netobj *owner = &rlockowner->rl_owner;
4100         struct list_head matches;
4101         int i;
4102         __be32 status;
4103
4104         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4105                 clid->cl_boot, clid->cl_id);
4106
4107         /* XXX check for lease expiration */
4108
4109         status = nfserr_stale_clientid;
4110         if (STALE_CLIENTID(clid))
4111                 return status;
4112
4113         nfs4_lock_state();
4114
4115         status = nfserr_locks_held;
4116         /* XXX: we're doing a linear search through all the lockowners.
4117          * Yipes!  For now we'll just hope clients aren't really using
4118          * release_lockowner much, but eventually we have to fix these
4119          * data structures. */
4120         INIT_LIST_HEAD(&matches);
4121         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4122                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4123                         if (!same_owner_str(sop, owner, clid))
4124                                 continue;
4125                         list_for_each_entry(stp, &sop->so_stateids,
4126                                         st_perstateowner) {
4127                                 if (check_for_locks(stp->st_file, sop))
4128                                         goto out;
4129                                 /* Note: so_perclient unused for lockowners,
4130                                  * so it's OK to fool with here. */
4131                                 list_add(&sop->so_perclient, &matches);
4132                         }
4133                 }
4134         }
4135         /* Clients probably won't expect us to return with some (but not all)
4136          * of the lockowner state released; so don't release any until all
4137          * have been checked. */
4138         status = nfs_ok;
4139         while (!list_empty(&matches)) {
4140                 sop = list_entry(matches.next, struct nfs4_stateowner,
4141                                                                 so_perclient);
4142                 /* unhash_stateowner deletes so_perclient only
4143                  * for openowners. */
4144                 list_del(&sop->so_perclient);
4145                 release_lockowner(sop);
4146         }
4147 out:
4148         nfs4_unlock_state();
4149         return status;
4150 }
4151
4152 static inline struct nfs4_client_reclaim *
4153 alloc_reclaim(void)
4154 {
4155         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4156 }
4157
4158 int
4159 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4160 {
4161         unsigned int strhashval = clientstr_hashval(name);
4162         struct nfs4_client *clp;
4163
4164         clp = find_confirmed_client_by_str(name, strhashval);
4165         return clp ? 1 : 0;
4166 }
4167
4168 /*
4169  * failure => all reset bets are off, nfserr_no_grace...
4170  */
4171 int
4172 nfs4_client_to_reclaim(const char *name)
4173 {
4174         unsigned int strhashval;
4175         struct nfs4_client_reclaim *crp = NULL;
4176
4177         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4178         crp = alloc_reclaim();
4179         if (!crp)
4180                 return 0;
4181         strhashval = clientstr_hashval(name);
4182         INIT_LIST_HEAD(&crp->cr_strhash);
4183         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4184         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4185         reclaim_str_hashtbl_size++;
4186         return 1;
4187 }
4188
4189 static void
4190 nfs4_release_reclaim(void)
4191 {
4192         struct nfs4_client_reclaim *crp = NULL;
4193         int i;
4194
4195         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4196                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4197                         crp = list_entry(reclaim_str_hashtbl[i].next,
4198                                         struct nfs4_client_reclaim, cr_strhash);
4199                         list_del(&crp->cr_strhash);
4200                         kfree(crp);
4201                         reclaim_str_hashtbl_size--;
4202                 }
4203         }
4204         BUG_ON(reclaim_str_hashtbl_size);
4205 }
4206
4207 /*
4208  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4209 static struct nfs4_client_reclaim *
4210 nfs4_find_reclaim_client(clientid_t *clid)
4211 {
4212         unsigned int strhashval;
4213         struct nfs4_client *clp;
4214         struct nfs4_client_reclaim *crp = NULL;
4215
4216
4217         /* find clientid in conf_id_hashtbl */
4218         clp = find_confirmed_client(clid);
4219         if (clp == NULL)
4220                 return NULL;
4221
4222         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4223                             clp->cl_name.len, clp->cl_name.data,
4224                             clp->cl_recdir);
4225
4226         /* find clp->cl_name in reclaim_str_hashtbl */
4227         strhashval = clientstr_hashval(clp->cl_recdir);
4228         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4229                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4230                         return crp;
4231                 }
4232         }
4233         return NULL;
4234 }
4235
4236 /*
4237 * Called from OPEN. Look for clientid in reclaim list.
4238 */
4239 __be32
4240 nfs4_check_open_reclaim(clientid_t *clid)
4241 {
4242         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4243 }
4244
4245 /* initialization to perform at module load time: */
4246
4247 int
4248 nfs4_state_init(void)
4249 {
4250         int i, status;
4251
4252         status = nfsd4_init_slabs();
4253         if (status)
4254                 return status;
4255         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4256                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4257                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4258                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4259                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4260                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4261         }
4262         for (i = 0; i < SESSION_HASH_SIZE; i++)
4263                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4264         for (i = 0; i < FILE_HASH_SIZE; i++) {
4265                 INIT_LIST_HEAD(&file_hashtbl[i]);
4266         }
4267         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4268                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4269                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4270         }
4271         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4272                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4273                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4274         }
4275         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4276                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4277                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4278         }
4279         memset(&onestateid, ~0, sizeof(stateid_t));
4280         INIT_LIST_HEAD(&close_lru);
4281         INIT_LIST_HEAD(&client_lru);
4282         INIT_LIST_HEAD(&del_recall_lru);
4283         reclaim_str_hashtbl_size = 0;
4284         return 0;
4285 }
4286
4287 static void
4288 nfsd4_load_reboot_recovery_data(void)
4289 {
4290         int status;
4291
4292         nfs4_lock_state();
4293         nfsd4_init_recdir(user_recovery_dirname);
4294         status = nfsd4_recdir_load();
4295         nfs4_unlock_state();
4296         if (status)
4297                 printk("NFSD: Failure reading reboot recovery data\n");
4298 }
4299
4300 /*
4301  * Since the lifetime of a delegation isn't limited to that of an open, a
4302  * client may quite reasonably hang on to a delegation as long as it has
4303  * the inode cached.  This becomes an obvious problem the first time a
4304  * client's inode cache approaches the size of the server's total memory.
4305  *
4306  * For now we avoid this problem by imposing a hard limit on the number
4307  * of delegations, which varies according to the server's memory size.
4308  */
4309 static void
4310 set_max_delegations(void)
4311 {
4312         /*
4313          * Allow at most 4 delegations per megabyte of RAM.  Quick
4314          * estimates suggest that in the worst case (where every delegation
4315          * is for a different inode), a delegation could take about 1.5K,
4316          * giving a worst case usage of about 6% of memory.
4317          */
4318         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4319 }
4320
4321 /* initialization to perform when the nfsd service is started: */
4322
4323 static int
4324 __nfs4_state_start(void)
4325 {
4326         int ret;
4327
4328         boot_time = get_seconds();
4329         locks_start_grace(&nfsd4_manager);
4330         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4331                nfsd4_grace);
4332         ret = set_callback_cred();
4333         if (ret)
4334                 return -ENOMEM;
4335         laundry_wq = create_singlethread_workqueue("nfsd4");
4336         if (laundry_wq == NULL)
4337                 return -ENOMEM;
4338         ret = nfsd4_create_callback_queue();
4339         if (ret)
4340                 goto out_free_laundry;
4341         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4342         set_max_delegations();
4343         return 0;
4344 out_free_laundry:
4345         destroy_workqueue(laundry_wq);
4346         return ret;
4347 }
4348
4349 int
4350 nfs4_state_start(void)
4351 {
4352         nfsd4_load_reboot_recovery_data();
4353         return __nfs4_state_start();
4354 }
4355
4356 static void
4357 __nfs4_state_shutdown(void)
4358 {
4359         int i;
4360         struct nfs4_client *clp = NULL;
4361         struct nfs4_delegation *dp = NULL;
4362         struct list_head *pos, *next, reaplist;
4363
4364         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4365                 while (!list_empty(&conf_id_hashtbl[i])) {
4366                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4367                         expire_client(clp);
4368                 }
4369                 while (!list_empty(&unconf_str_hashtbl[i])) {
4370                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4371                         expire_client(clp);
4372                 }
4373         }
4374         INIT_LIST_HEAD(&reaplist);
4375         spin_lock(&recall_lock);
4376         list_for_each_safe(pos, next, &del_recall_lru) {
4377                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4378                 list_move(&dp->dl_recall_lru, &reaplist);
4379         }
4380         spin_unlock(&recall_lock);
4381         list_for_each_safe(pos, next, &reaplist) {
4382                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4383                 list_del_init(&dp->dl_recall_lru);
4384                 unhash_delegation(dp);
4385         }
4386
4387         nfsd4_shutdown_recdir();
4388 }
4389
4390 void
4391 nfs4_state_shutdown(void)
4392 {
4393         cancel_delayed_work_sync(&laundromat_work);
4394         destroy_workqueue(laundry_wq);
4395         locks_end_grace(&nfsd4_manager);
4396         nfs4_lock_state();
4397         nfs4_release_reclaim();
4398         __nfs4_state_shutdown();
4399         nfs4_unlock_state();
4400         nfsd4_destroy_callback_queue();
4401 }
4402
4403 /*
4404  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4405  * accessed when nfsd is starting.
4406  */
4407 static void
4408 nfs4_set_recdir(char *recdir)
4409 {
4410         strcpy(user_recovery_dirname, recdir);
4411 }
4412
4413 /*
4414  * Change the NFSv4 recovery directory to recdir.
4415  */
4416 int
4417 nfs4_reset_recoverydir(char *recdir)
4418 {
4419         int status;
4420         struct path path;
4421
4422         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4423         if (status)
4424                 return status;
4425         status = -ENOTDIR;
4426         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4427                 nfs4_set_recdir(recdir);
4428                 status = 0;
4429         }
4430         path_put(&path);
4431         return status;
4432 }
4433
4434 char *
4435 nfs4_recoverydir(void)
4436 {
4437         return user_recovery_dirname;
4438 }