nfsd: kill unused macro definition
[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
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_type = type;
234         dp->dl_stateid.si_boot = boot_time;
235         dp->dl_stateid.si_stateownerid = current_delegid++;
236         dp->dl_stateid.si_fileid = 0;
237         dp->dl_stateid.si_generation = 0;
238         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
239         dp->dl_time = 0;
240         atomic_set(&dp->dl_count, 1);
241         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
242         return dp;
243 }
244
245 void
246 nfs4_put_delegation(struct nfs4_delegation *dp)
247 {
248         if (atomic_dec_and_test(&dp->dl_count)) {
249                 dprintk("NFSD: freeing dp %p\n",dp);
250                 put_nfs4_file(dp->dl_file);
251                 kmem_cache_free(deleg_slab, dp);
252                 num_delegations--;
253         }
254 }
255
256 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
257 {
258         if (atomic_dec_and_test(&fp->fi_delegees)) {
259                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
260                 fp->fi_lease = NULL;
261                 fp->fi_deleg_file = NULL;
262         }
263 }
264
265 /* Called under the state lock. */
266 static void
267 unhash_delegation(struct nfs4_delegation *dp)
268 {
269         list_del_init(&dp->dl_perclnt);
270         spin_lock(&recall_lock);
271         list_del_init(&dp->dl_perfile);
272         list_del_init(&dp->dl_recall_lru);
273         spin_unlock(&recall_lock);
274         nfs4_put_deleg_lease(dp->dl_file);
275         nfs4_put_delegation(dp);
276 }
277
278 /* 
279  * SETCLIENTID state 
280  */
281
282 /* client_lock protects the client lru list and session hash table */
283 static DEFINE_SPINLOCK(client_lock);
284
285 /* Hash tables for nfs4_clientid state */
286 #define CLIENT_HASH_BITS                 4
287 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
288 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
289
290 #define clientid_hashval(id) \
291         ((id) & CLIENT_HASH_MASK)
292 #define clientstr_hashval(name) \
293         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
294 /*
295  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
296  * used in reboot/reset lease grace period processing
297  *
298  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
299  * setclientid_confirmed info. 
300  *
301  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
302  * setclientid info.
303  *
304  * client_lru holds client queue ordered by nfs4_client.cl_time
305  * for lease renewal.
306  *
307  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
308  * for last close replay.
309  */
310 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
311 static int reclaim_str_hashtbl_size = 0;
312 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
313 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
314 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
315 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
316 static struct list_head client_lru;
317 static struct list_head close_lru;
318
319 static void unhash_generic_stateid(struct nfs4_stateid *stp)
320 {
321         list_del(&stp->st_hash);
322         list_del(&stp->st_perfile);
323         list_del(&stp->st_perstateowner);
324 }
325
326 static void free_generic_stateid(struct nfs4_stateid *stp)
327 {
328         put_nfs4_file(stp->st_file);
329         kmem_cache_free(stateid_slab, stp);
330 }
331
332 static void release_lock_stateid(struct nfs4_stateid *stp)
333 {
334         struct file *file;
335
336         unhash_generic_stateid(stp);
337         file = find_any_file(stp->st_file);
338         if (file)
339                 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
340         free_generic_stateid(stp);
341 }
342
343 static void unhash_lockowner(struct nfs4_stateowner *sop)
344 {
345         struct nfs4_stateid *stp;
346
347         list_del(&sop->so_idhash);
348         list_del(&sop->so_strhash);
349         list_del(&sop->so_perstateid);
350         while (!list_empty(&sop->so_stateids)) {
351                 stp = list_first_entry(&sop->so_stateids,
352                                 struct nfs4_stateid, st_perstateowner);
353                 release_lock_stateid(stp);
354         }
355 }
356
357 static void release_lockowner(struct nfs4_stateowner *sop)
358 {
359         unhash_lockowner(sop);
360         nfs4_put_stateowner(sop);
361 }
362
363 static void
364 release_stateid_lockowners(struct nfs4_stateid *open_stp)
365 {
366         struct nfs4_stateowner *lock_sop;
367
368         while (!list_empty(&open_stp->st_lockowners)) {
369                 lock_sop = list_entry(open_stp->st_lockowners.next,
370                                 struct nfs4_stateowner, so_perstateid);
371                 /* list_del(&open_stp->st_lockowners);  */
372                 BUG_ON(lock_sop->so_is_open_owner);
373                 release_lockowner(lock_sop);
374         }
375 }
376
377 /*
378  * We store the NONE, READ, WRITE, and BOTH bits separately in the
379  * st_{access,deny}_bmap field of the stateid, in order to track not
380  * only what share bits are currently in force, but also what
381  * combinations of share bits previous opens have used.  This allows us
382  * to enforce the recommendation of rfc 3530 14.2.19 that the server
383  * return an error if the client attempt to downgrade to a combination
384  * of share bits not explicable by closing some of its previous opens.
385  *
386  * XXX: This enforcement is actually incomplete, since we don't keep
387  * track of access/deny bit combinations; so, e.g., we allow:
388  *
389  *      OPEN allow read, deny write
390  *      OPEN allow both, deny none
391  *      DOWNGRADE allow read, deny none
392  *
393  * which we should reject.
394  */
395 static void
396 set_access(unsigned int *access, unsigned long bmap) {
397         int i;
398
399         *access = 0;
400         for (i = 1; i < 4; i++) {
401                 if (test_bit(i, &bmap))
402                         *access |= i;
403         }
404 }
405
406 static void
407 set_deny(unsigned int *deny, unsigned long bmap) {
408         int i;
409
410         *deny = 0;
411         for (i = 0; i < 4; i++) {
412                 if (test_bit(i, &bmap))
413                         *deny |= i ;
414         }
415 }
416
417 static int
418 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
419         unsigned int access, deny;
420
421         set_access(&access, stp->st_access_bmap);
422         set_deny(&deny, stp->st_deny_bmap);
423         if ((access & open->op_share_deny) || (deny & open->op_share_access))
424                 return 0;
425         return 1;
426 }
427
428 static int nfs4_access_to_omode(u32 access)
429 {
430         switch (access & NFS4_SHARE_ACCESS_BOTH) {
431         case NFS4_SHARE_ACCESS_READ:
432                 return O_RDONLY;
433         case NFS4_SHARE_ACCESS_WRITE:
434                 return O_WRONLY;
435         case NFS4_SHARE_ACCESS_BOTH:
436                 return O_RDWR;
437         }
438         BUG();
439 }
440
441 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
442 {
443         unsigned int access;
444
445         set_access(&access, stp->st_access_bmap);
446         return nfs4_access_to_omode(access);
447 }
448
449 static void release_open_stateid(struct nfs4_stateid *stp)
450 {
451         int oflag = nfs4_access_bmap_to_omode(stp);
452
453         unhash_generic_stateid(stp);
454         release_stateid_lockowners(stp);
455         nfs4_file_put_access(stp->st_file, oflag);
456         free_generic_stateid(stp);
457 }
458
459 static void unhash_openowner(struct nfs4_stateowner *sop)
460 {
461         struct nfs4_stateid *stp;
462
463         list_del(&sop->so_idhash);
464         list_del(&sop->so_strhash);
465         list_del(&sop->so_perclient);
466         list_del(&sop->so_perstateid); /* XXX: necessary? */
467         while (!list_empty(&sop->so_stateids)) {
468                 stp = list_first_entry(&sop->so_stateids,
469                                 struct nfs4_stateid, st_perstateowner);
470                 release_open_stateid(stp);
471         }
472 }
473
474 static void release_openowner(struct nfs4_stateowner *sop)
475 {
476         unhash_openowner(sop);
477         list_del(&sop->so_close_lru);
478         nfs4_put_stateowner(sop);
479 }
480
481 #define SESSION_HASH_SIZE       512
482 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
483
484 static inline int
485 hash_sessionid(struct nfs4_sessionid *sessionid)
486 {
487         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
488
489         return sid->sequence % SESSION_HASH_SIZE;
490 }
491
492 static inline void
493 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
494 {
495         u32 *ptr = (u32 *)(&sessionid->data[0]);
496         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
497 }
498
499 static void
500 gen_sessionid(struct nfsd4_session *ses)
501 {
502         struct nfs4_client *clp = ses->se_client;
503         struct nfsd4_sessionid *sid;
504
505         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
506         sid->clientid = clp->cl_clientid;
507         sid->sequence = current_sessionid++;
508         sid->reserved = 0;
509 }
510
511 /*
512  * The protocol defines ca_maxresponssize_cached to include the size of
513  * the rpc header, but all we need to cache is the data starting after
514  * the end of the initial SEQUENCE operation--the rest we regenerate
515  * each time.  Therefore we can advertise a ca_maxresponssize_cached
516  * value that is the number of bytes in our cache plus a few additional
517  * bytes.  In order to stay on the safe side, and not promise more than
518  * we can cache, those additional bytes must be the minimum possible: 24
519  * bytes of rpc header (xid through accept state, with AUTH_NULL
520  * verifier), 12 for the compound header (with zero-length tag), and 44
521  * for the SEQUENCE op response:
522  */
523 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
524
525 static void
526 free_session_slots(struct nfsd4_session *ses)
527 {
528         int i;
529
530         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
531                 kfree(ses->se_slots[i]);
532 }
533
534 /*
535  * We don't actually need to cache the rpc and session headers, so we
536  * can allocate a little less for each slot:
537  */
538 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
539 {
540         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
541 }
542
543 static int nfsd4_sanitize_slot_size(u32 size)
544 {
545         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
546         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
547
548         return size;
549 }
550
551 /*
552  * XXX: If we run out of reserved DRC memory we could (up to a point)
553  * re-negotiate active sessions and reduce their slot usage to make
554  * rooom for new connections. For now we just fail the create session.
555  */
556 static int nfsd4_get_drc_mem(int slotsize, u32 num)
557 {
558         int avail;
559
560         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
561
562         spin_lock(&nfsd_drc_lock);
563         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
564                         nfsd_drc_max_mem - nfsd_drc_mem_used);
565         num = min_t(int, num, avail / slotsize);
566         nfsd_drc_mem_used += num * slotsize;
567         spin_unlock(&nfsd_drc_lock);
568
569         return num;
570 }
571
572 static void nfsd4_put_drc_mem(int slotsize, int num)
573 {
574         spin_lock(&nfsd_drc_lock);
575         nfsd_drc_mem_used -= slotsize * num;
576         spin_unlock(&nfsd_drc_lock);
577 }
578
579 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
580 {
581         struct nfsd4_session *new;
582         int mem, i;
583
584         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
585                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
586         mem = numslots * sizeof(struct nfsd4_slot *);
587
588         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
589         if (!new)
590                 return NULL;
591         /* allocate each struct nfsd4_slot and data cache in one piece */
592         for (i = 0; i < numslots; i++) {
593                 mem = sizeof(struct nfsd4_slot) + slotsize;
594                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
595                 if (!new->se_slots[i])
596                         goto out_free;
597         }
598         return new;
599 out_free:
600         while (i--)
601                 kfree(new->se_slots[i]);
602         kfree(new);
603         return NULL;
604 }
605
606 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
607 {
608         u32 maxrpc = nfsd_serv->sv_max_mesg;
609
610         new->maxreqs = numslots;
611         new->maxresp_cached = slotsize + NFSD_MIN_HDR_SEQ_SZ;
612         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
613         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
614         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
615 }
616
617 static void free_conn(struct nfsd4_conn *c)
618 {
619         svc_xprt_put(c->cn_xprt);
620         kfree(c);
621 }
622
623 static void nfsd4_conn_lost(struct svc_xpt_user *u)
624 {
625         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
626         struct nfs4_client *clp = c->cn_session->se_client;
627
628         spin_lock(&clp->cl_lock);
629         if (!list_empty(&c->cn_persession)) {
630                 list_del(&c->cn_persession);
631                 free_conn(c);
632         }
633         spin_unlock(&clp->cl_lock);
634         nfsd4_probe_callback(clp);
635 }
636
637 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
638 {
639         struct nfsd4_conn *conn;
640
641         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
642         if (!conn)
643                 return NULL;
644         svc_xprt_get(rqstp->rq_xprt);
645         conn->cn_xprt = rqstp->rq_xprt;
646         conn->cn_flags = flags;
647         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
648         return conn;
649 }
650
651 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
652 {
653         conn->cn_session = ses;
654         list_add(&conn->cn_persession, &ses->se_conns);
655 }
656
657 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
658 {
659         struct nfs4_client *clp = ses->se_client;
660
661         spin_lock(&clp->cl_lock);
662         __nfsd4_hash_conn(conn, ses);
663         spin_unlock(&clp->cl_lock);
664 }
665
666 static int nfsd4_register_conn(struct nfsd4_conn *conn)
667 {
668         conn->cn_xpt_user.callback = nfsd4_conn_lost;
669         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
670 }
671
672 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
673 {
674         struct nfsd4_conn *conn;
675         int ret;
676
677         conn = alloc_conn(rqstp, dir);
678         if (!conn)
679                 return nfserr_jukebox;
680         nfsd4_hash_conn(conn, ses);
681         ret = nfsd4_register_conn(conn);
682         if (ret)
683                 /* oops; xprt is already down: */
684                 nfsd4_conn_lost(&conn->cn_xpt_user);
685         return nfs_ok;
686 }
687
688 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
689 {
690         u32 dir = NFS4_CDFC4_FORE;
691
692         if (ses->se_flags & SESSION4_BACK_CHAN)
693                 dir |= NFS4_CDFC4_BACK;
694
695         return nfsd4_new_conn(rqstp, ses, dir);
696 }
697
698 /* must be called under client_lock */
699 static void nfsd4_del_conns(struct nfsd4_session *s)
700 {
701         struct nfs4_client *clp = s->se_client;
702         struct nfsd4_conn *c;
703
704         spin_lock(&clp->cl_lock);
705         while (!list_empty(&s->se_conns)) {
706                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
707                 list_del_init(&c->cn_persession);
708                 spin_unlock(&clp->cl_lock);
709
710                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
711                 free_conn(c);
712
713                 spin_lock(&clp->cl_lock);
714         }
715         spin_unlock(&clp->cl_lock);
716 }
717
718 void free_session(struct kref *kref)
719 {
720         struct nfsd4_session *ses;
721         int mem;
722
723         ses = container_of(kref, struct nfsd4_session, se_ref);
724         nfsd4_del_conns(ses);
725         spin_lock(&nfsd_drc_lock);
726         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
727         nfsd_drc_mem_used -= mem;
728         spin_unlock(&nfsd_drc_lock);
729         free_session_slots(ses);
730         kfree(ses);
731 }
732
733 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
734 {
735         struct nfsd4_session *new;
736         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
737         int numslots, slotsize;
738         int status;
739         int idx;
740
741         /*
742          * Note decreasing slot size below client's request may
743          * make it difficult for client to function correctly, whereas
744          * decreasing the number of slots will (just?) affect
745          * performance.  When short on memory we therefore prefer to
746          * decrease number of slots instead of their size.
747          */
748         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
749         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
750         if (numslots < 1)
751                 return NULL;
752
753         new = alloc_session(slotsize, numslots);
754         if (!new) {
755                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
756                 return NULL;
757         }
758         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
759
760         new->se_client = clp;
761         gen_sessionid(new);
762
763         INIT_LIST_HEAD(&new->se_conns);
764
765         new->se_cb_seq_nr = 1;
766         new->se_flags = cses->flags;
767         new->se_cb_prog = cses->callback_prog;
768         kref_init(&new->se_ref);
769         idx = hash_sessionid(&new->se_sessionid);
770         spin_lock(&client_lock);
771         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
772         spin_lock(&clp->cl_lock);
773         list_add(&new->se_perclnt, &clp->cl_sessions);
774         spin_unlock(&clp->cl_lock);
775         spin_unlock(&client_lock);
776
777         status = nfsd4_new_conn_from_crses(rqstp, new);
778         /* whoops: benny points out, status is ignored! (err, or bogus) */
779         if (status) {
780                 free_session(&new->se_ref);
781                 return NULL;
782         }
783         if (cses->flags & SESSION4_BACK_CHAN) {
784                 struct sockaddr *sa = svc_addr(rqstp);
785                 /*
786                  * This is a little silly; with sessions there's no real
787                  * use for the callback address.  Use the peer address
788                  * as a reasonable default for now, but consider fixing
789                  * the rpc client not to require an address in the
790                  * future:
791                  */
792                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
793                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
794         }
795         nfsd4_probe_callback(clp);
796         return new;
797 }
798
799 /* caller must hold client_lock */
800 static struct nfsd4_session *
801 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
802 {
803         struct nfsd4_session *elem;
804         int idx;
805
806         dump_sessionid(__func__, sessionid);
807         idx = hash_sessionid(sessionid);
808         /* Search in the appropriate list */
809         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
810                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
811                             NFS4_MAX_SESSIONID_LEN)) {
812                         return elem;
813                 }
814         }
815
816         dprintk("%s: session not found\n", __func__);
817         return NULL;
818 }
819
820 /* caller must hold client_lock */
821 static void
822 unhash_session(struct nfsd4_session *ses)
823 {
824         list_del(&ses->se_hash);
825         spin_lock(&ses->se_client->cl_lock);
826         list_del(&ses->se_perclnt);
827         spin_unlock(&ses->se_client->cl_lock);
828 }
829
830 /* must be called under the client_lock */
831 static inline void
832 renew_client_locked(struct nfs4_client *clp)
833 {
834         if (is_client_expired(clp)) {
835                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
836                         __func__,
837                         clp->cl_clientid.cl_boot,
838                         clp->cl_clientid.cl_id);
839                 return;
840         }
841
842         /*
843         * Move client to the end to the LRU list.
844         */
845         dprintk("renewing client (clientid %08x/%08x)\n", 
846                         clp->cl_clientid.cl_boot, 
847                         clp->cl_clientid.cl_id);
848         list_move_tail(&clp->cl_lru, &client_lru);
849         clp->cl_time = get_seconds();
850 }
851
852 static inline void
853 renew_client(struct nfs4_client *clp)
854 {
855         spin_lock(&client_lock);
856         renew_client_locked(clp);
857         spin_unlock(&client_lock);
858 }
859
860 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
861 static int
862 STALE_CLIENTID(clientid_t *clid)
863 {
864         if (clid->cl_boot == boot_time)
865                 return 0;
866         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
867                 clid->cl_boot, clid->cl_id, boot_time);
868         return 1;
869 }
870
871 /* 
872  * XXX Should we use a slab cache ?
873  * This type of memory management is somewhat inefficient, but we use it
874  * anyway since SETCLIENTID is not a common operation.
875  */
876 static struct nfs4_client *alloc_client(struct xdr_netobj name)
877 {
878         struct nfs4_client *clp;
879
880         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
881         if (clp == NULL)
882                 return NULL;
883         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
884         if (clp->cl_name.data == NULL) {
885                 kfree(clp);
886                 return NULL;
887         }
888         memcpy(clp->cl_name.data, name.data, name.len);
889         clp->cl_name.len = name.len;
890         return clp;
891 }
892
893 static inline void
894 free_client(struct nfs4_client *clp)
895 {
896         while (!list_empty(&clp->cl_sessions)) {
897                 struct nfsd4_session *ses;
898                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
899                                 se_perclnt);
900                 list_del(&ses->se_perclnt);
901                 nfsd4_put_session(ses);
902         }
903         if (clp->cl_cred.cr_group_info)
904                 put_group_info(clp->cl_cred.cr_group_info);
905         kfree(clp->cl_principal);
906         kfree(clp->cl_name.data);
907         kfree(clp);
908 }
909
910 void
911 release_session_client(struct nfsd4_session *session)
912 {
913         struct nfs4_client *clp = session->se_client;
914
915         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
916                 return;
917         if (is_client_expired(clp)) {
918                 free_client(clp);
919                 session->se_client = NULL;
920         } else
921                 renew_client_locked(clp);
922         spin_unlock(&client_lock);
923 }
924
925 /* must be called under the client_lock */
926 static inline void
927 unhash_client_locked(struct nfs4_client *clp)
928 {
929         struct nfsd4_session *ses;
930
931         mark_client_expired(clp);
932         list_del(&clp->cl_lru);
933         spin_lock(&clp->cl_lock);
934         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
935                 list_del_init(&ses->se_hash);
936         spin_unlock(&clp->cl_lock);
937 }
938
939 static void
940 expire_client(struct nfs4_client *clp)
941 {
942         struct nfs4_stateowner *sop;
943         struct nfs4_delegation *dp;
944         struct list_head reaplist;
945
946         INIT_LIST_HEAD(&reaplist);
947         spin_lock(&recall_lock);
948         while (!list_empty(&clp->cl_delegations)) {
949                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
950                 list_del_init(&dp->dl_perclnt);
951                 list_move(&dp->dl_recall_lru, &reaplist);
952         }
953         spin_unlock(&recall_lock);
954         while (!list_empty(&reaplist)) {
955                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
956                 list_del_init(&dp->dl_recall_lru);
957                 unhash_delegation(dp);
958         }
959         while (!list_empty(&clp->cl_openowners)) {
960                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
961                 release_openowner(sop);
962         }
963         nfsd4_shutdown_callback(clp);
964         if (clp->cl_cb_conn.cb_xprt)
965                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
966         list_del(&clp->cl_idhash);
967         list_del(&clp->cl_strhash);
968         spin_lock(&client_lock);
969         unhash_client_locked(clp);
970         if (atomic_read(&clp->cl_refcount) == 0)
971                 free_client(clp);
972         spin_unlock(&client_lock);
973 }
974
975 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
976 {
977         memcpy(target->cl_verifier.data, source->data,
978                         sizeof(target->cl_verifier.data));
979 }
980
981 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
982 {
983         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
984         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
985 }
986
987 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
988 {
989         target->cr_uid = source->cr_uid;
990         target->cr_gid = source->cr_gid;
991         target->cr_group_info = source->cr_group_info;
992         get_group_info(target->cr_group_info);
993 }
994
995 static int same_name(const char *n1, const char *n2)
996 {
997         return 0 == memcmp(n1, n2, HEXDIR_LEN);
998 }
999
1000 static int
1001 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1002 {
1003         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1004 }
1005
1006 static int
1007 same_clid(clientid_t *cl1, clientid_t *cl2)
1008 {
1009         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1010 }
1011
1012 /* XXX what about NGROUP */
1013 static int
1014 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1015 {
1016         return cr1->cr_uid == cr2->cr_uid;
1017 }
1018
1019 static void gen_clid(struct nfs4_client *clp)
1020 {
1021         static u32 current_clientid = 1;
1022
1023         clp->cl_clientid.cl_boot = boot_time;
1024         clp->cl_clientid.cl_id = current_clientid++; 
1025 }
1026
1027 static void gen_confirm(struct nfs4_client *clp)
1028 {
1029         static u32 i;
1030         u32 *p;
1031
1032         p = (u32 *)clp->cl_confirm.data;
1033         *p++ = get_seconds();
1034         *p++ = i++;
1035 }
1036
1037 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1038                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1039 {
1040         struct nfs4_client *clp;
1041         struct sockaddr *sa = svc_addr(rqstp);
1042         char *princ;
1043
1044         clp = alloc_client(name);
1045         if (clp == NULL)
1046                 return NULL;
1047
1048         INIT_LIST_HEAD(&clp->cl_sessions);
1049
1050         princ = svc_gss_principal(rqstp);
1051         if (princ) {
1052                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1053                 if (clp->cl_principal == NULL) {
1054                         free_client(clp);
1055                         return NULL;
1056                 }
1057         }
1058
1059         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1060         atomic_set(&clp->cl_refcount, 0);
1061         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1062         INIT_LIST_HEAD(&clp->cl_idhash);
1063         INIT_LIST_HEAD(&clp->cl_strhash);
1064         INIT_LIST_HEAD(&clp->cl_openowners);
1065         INIT_LIST_HEAD(&clp->cl_delegations);
1066         INIT_LIST_HEAD(&clp->cl_lru);
1067         INIT_LIST_HEAD(&clp->cl_callbacks);
1068         spin_lock_init(&clp->cl_lock);
1069         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1070         clp->cl_time = get_seconds();
1071         clear_bit(0, &clp->cl_cb_slot_busy);
1072         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1073         copy_verf(clp, verf);
1074         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1075         clp->cl_flavor = rqstp->rq_flavor;
1076         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1077         gen_confirm(clp);
1078         clp->cl_cb_session = NULL;
1079         return clp;
1080 }
1081
1082 static int check_name(struct xdr_netobj name)
1083 {
1084         if (name.len == 0) 
1085                 return 0;
1086         if (name.len > NFS4_OPAQUE_LIMIT) {
1087                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1088                 return 0;
1089         }
1090         return 1;
1091 }
1092
1093 static void
1094 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1095 {
1096         unsigned int idhashval;
1097
1098         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1099         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1100         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1101         renew_client(clp);
1102 }
1103
1104 static void
1105 move_to_confirmed(struct nfs4_client *clp)
1106 {
1107         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1108         unsigned int strhashval;
1109
1110         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1111         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1112         strhashval = clientstr_hashval(clp->cl_recdir);
1113         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1114         renew_client(clp);
1115 }
1116
1117 static struct nfs4_client *
1118 find_confirmed_client(clientid_t *clid)
1119 {
1120         struct nfs4_client *clp;
1121         unsigned int idhashval = clientid_hashval(clid->cl_id);
1122
1123         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1124                 if (same_clid(&clp->cl_clientid, clid))
1125                         return clp;
1126         }
1127         return NULL;
1128 }
1129
1130 static struct nfs4_client *
1131 find_unconfirmed_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, &unconf_id_hashtbl[idhashval], cl_idhash) {
1137                 if (same_clid(&clp->cl_clientid, clid))
1138                         return clp;
1139         }
1140         return NULL;
1141 }
1142
1143 static bool clp_used_exchangeid(struct nfs4_client *clp)
1144 {
1145         return clp->cl_exchange_flags != 0;
1146
1147
1148 static struct nfs4_client *
1149 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1150 {
1151         struct nfs4_client *clp;
1152
1153         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1154                 if (same_name(clp->cl_recdir, dname))
1155                         return clp;
1156         }
1157         return NULL;
1158 }
1159
1160 static struct nfs4_client *
1161 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1162 {
1163         struct nfs4_client *clp;
1164
1165         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1166                 if (same_name(clp->cl_recdir, dname))
1167                         return clp;
1168         }
1169         return NULL;
1170 }
1171
1172 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1173 {
1174         switch (family) {
1175         case AF_INET:
1176                 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1177                 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1178                 return;
1179         case AF_INET6:
1180                 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1181                 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1182                 return;
1183         }
1184 }
1185
1186 static void
1187 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1188 {
1189         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1190         struct sockaddr *sa = svc_addr(rqstp);
1191         u32 scopeid = rpc_get_scope_id(sa);
1192         unsigned short expected_family;
1193
1194         /* Currently, we only support tcp and tcp6 for the callback channel */
1195         if (se->se_callback_netid_len == 3 &&
1196             !memcmp(se->se_callback_netid_val, "tcp", 3))
1197                 expected_family = AF_INET;
1198         else if (se->se_callback_netid_len == 4 &&
1199                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1200                 expected_family = AF_INET6;
1201         else
1202                 goto out_err;
1203
1204         conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1205                                             se->se_callback_addr_len,
1206                                             (struct sockaddr *)&conn->cb_addr,
1207                                             sizeof(conn->cb_addr));
1208
1209         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1210                 goto out_err;
1211
1212         if (conn->cb_addr.ss_family == AF_INET6)
1213                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1214
1215         conn->cb_prog = se->se_callback_prog;
1216         conn->cb_ident = se->se_callback_ident;
1217         rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1218         return;
1219 out_err:
1220         conn->cb_addr.ss_family = AF_UNSPEC;
1221         conn->cb_addrlen = 0;
1222         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1223                 "will not receive delegations\n",
1224                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1225
1226         return;
1227 }
1228
1229 /*
1230  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1231  */
1232 void
1233 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1234 {
1235         struct nfsd4_slot *slot = resp->cstate.slot;
1236         unsigned int base;
1237
1238         dprintk("--> %s slot %p\n", __func__, slot);
1239
1240         slot->sl_opcnt = resp->opcnt;
1241         slot->sl_status = resp->cstate.status;
1242
1243         if (nfsd4_not_cached(resp)) {
1244                 slot->sl_datalen = 0;
1245                 return;
1246         }
1247         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1248         base = (char *)resp->cstate.datap -
1249                                         (char *)resp->xbuf->head[0].iov_base;
1250         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1251                                     slot->sl_datalen))
1252                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1253         return;
1254 }
1255
1256 /*
1257  * Encode the replay sequence operation from the slot values.
1258  * If cachethis is FALSE encode the uncached rep error on the next
1259  * operation which sets resp->p and increments resp->opcnt for
1260  * nfs4svc_encode_compoundres.
1261  *
1262  */
1263 static __be32
1264 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1265                           struct nfsd4_compoundres *resp)
1266 {
1267         struct nfsd4_op *op;
1268         struct nfsd4_slot *slot = resp->cstate.slot;
1269
1270         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1271                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1272
1273         /* Encode the replayed sequence operation */
1274         op = &args->ops[resp->opcnt - 1];
1275         nfsd4_encode_operation(resp, op);
1276
1277         /* Return nfserr_retry_uncached_rep in next operation. */
1278         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1279                 op = &args->ops[resp->opcnt++];
1280                 op->status = nfserr_retry_uncached_rep;
1281                 nfsd4_encode_operation(resp, op);
1282         }
1283         return op->status;
1284 }
1285
1286 /*
1287  * The sequence operation is not cached because we can use the slot and
1288  * session values.
1289  */
1290 __be32
1291 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1292                          struct nfsd4_sequence *seq)
1293 {
1294         struct nfsd4_slot *slot = resp->cstate.slot;
1295         __be32 status;
1296
1297         dprintk("--> %s slot %p\n", __func__, slot);
1298
1299         /* Either returns 0 or nfserr_retry_uncached */
1300         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1301         if (status == nfserr_retry_uncached_rep)
1302                 return status;
1303
1304         /* The sequence operation has been encoded, cstate->datap set. */
1305         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1306
1307         resp->opcnt = slot->sl_opcnt;
1308         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1309         status = slot->sl_status;
1310
1311         return status;
1312 }
1313
1314 /*
1315  * Set the exchange_id flags returned by the server.
1316  */
1317 static void
1318 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1319 {
1320         /* pNFS is not supported */
1321         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1322
1323         /* Referrals are supported, Migration is not. */
1324         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1325
1326         /* set the wire flags to return to client. */
1327         clid->flags = new->cl_exchange_flags;
1328 }
1329
1330 __be32
1331 nfsd4_exchange_id(struct svc_rqst *rqstp,
1332                   struct nfsd4_compound_state *cstate,
1333                   struct nfsd4_exchange_id *exid)
1334 {
1335         struct nfs4_client *unconf, *conf, *new;
1336         int status;
1337         unsigned int            strhashval;
1338         char                    dname[HEXDIR_LEN];
1339         char                    addr_str[INET6_ADDRSTRLEN];
1340         nfs4_verifier           verf = exid->verifier;
1341         struct sockaddr         *sa = svc_addr(rqstp);
1342
1343         rpc_ntop(sa, addr_str, sizeof(addr_str));
1344         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1345                 "ip_addr=%s flags %x, spa_how %d\n",
1346                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1347                 addr_str, exid->flags, exid->spa_how);
1348
1349         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1350                 return nfserr_inval;
1351
1352         /* Currently only support SP4_NONE */
1353         switch (exid->spa_how) {
1354         case SP4_NONE:
1355                 break;
1356         case SP4_SSV:
1357                 return nfserr_serverfault;
1358         default:
1359                 BUG();                          /* checked by xdr code */
1360         case SP4_MACH_CRED:
1361                 return nfserr_serverfault;      /* no excuse :-/ */
1362         }
1363
1364         status = nfs4_make_rec_clidname(dname, &exid->clname);
1365
1366         if (status)
1367                 goto error;
1368
1369         strhashval = clientstr_hashval(dname);
1370
1371         nfs4_lock_state();
1372         status = nfs_ok;
1373
1374         conf = find_confirmed_client_by_str(dname, strhashval);
1375         if (conf) {
1376                 if (!clp_used_exchangeid(conf)) {
1377                         status = nfserr_clid_inuse; /* XXX: ? */
1378                         goto out;
1379                 }
1380                 if (!same_verf(&verf, &conf->cl_verifier)) {
1381                         /* 18.35.4 case 8 */
1382                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1383                                 status = nfserr_not_same;
1384                                 goto out;
1385                         }
1386                         /* Client reboot: destroy old state */
1387                         expire_client(conf);
1388                         goto out_new;
1389                 }
1390                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1391                         /* 18.35.4 case 9 */
1392                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1393                                 status = nfserr_perm;
1394                                 goto out;
1395                         }
1396                         expire_client(conf);
1397                         goto out_new;
1398                 }
1399                 /*
1400                  * Set bit when the owner id and verifier map to an already
1401                  * confirmed client id (18.35.3).
1402                  */
1403                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1404
1405                 /*
1406                  * Falling into 18.35.4 case 2, possible router replay.
1407                  * Leave confirmed record intact and return same result.
1408                  */
1409                 copy_verf(conf, &verf);
1410                 new = conf;
1411                 goto out_copy;
1412         }
1413
1414         /* 18.35.4 case 7 */
1415         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1416                 status = nfserr_noent;
1417                 goto out;
1418         }
1419
1420         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1421         if (unconf) {
1422                 /*
1423                  * Possible retry or client restart.  Per 18.35.4 case 4,
1424                  * a new unconfirmed record should be generated regardless
1425                  * of whether any properties have changed.
1426                  */
1427                 expire_client(unconf);
1428         }
1429
1430 out_new:
1431         /* Normal case */
1432         new = create_client(exid->clname, dname, rqstp, &verf);
1433         if (new == NULL) {
1434                 status = nfserr_jukebox;
1435                 goto out;
1436         }
1437
1438         gen_clid(new);
1439         add_to_unconfirmed(new, strhashval);
1440 out_copy:
1441         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1442         exid->clientid.cl_id = new->cl_clientid.cl_id;
1443
1444         exid->seqid = 1;
1445         nfsd4_set_ex_flags(new, exid);
1446
1447         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1448                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1449         status = nfs_ok;
1450
1451 out:
1452         nfs4_unlock_state();
1453 error:
1454         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1455         return status;
1456 }
1457
1458 static int
1459 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1460 {
1461         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1462                 slot_seqid);
1463
1464         /* The slot is in use, and no response has been sent. */
1465         if (slot_inuse) {
1466                 if (seqid == slot_seqid)
1467                         return nfserr_jukebox;
1468                 else
1469                         return nfserr_seq_misordered;
1470         }
1471         /* Normal */
1472         if (likely(seqid == slot_seqid + 1))
1473                 return nfs_ok;
1474         /* Replay */
1475         if (seqid == slot_seqid)
1476                 return nfserr_replay_cache;
1477         /* Wraparound */
1478         if (seqid == 1 && (slot_seqid + 1) == 0)
1479                 return nfs_ok;
1480         /* Misordered replay or misordered new request */
1481         return nfserr_seq_misordered;
1482 }
1483
1484 /*
1485  * Cache the create session result into the create session single DRC
1486  * slot cache by saving the xdr structure. sl_seqid has been set.
1487  * Do this for solo or embedded create session operations.
1488  */
1489 static void
1490 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1491                            struct nfsd4_clid_slot *slot, int nfserr)
1492 {
1493         slot->sl_status = nfserr;
1494         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1495 }
1496
1497 static __be32
1498 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1499                             struct nfsd4_clid_slot *slot)
1500 {
1501         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1502         return slot->sl_status;
1503 }
1504
1505 __be32
1506 nfsd4_create_session(struct svc_rqst *rqstp,
1507                      struct nfsd4_compound_state *cstate,
1508                      struct nfsd4_create_session *cr_ses)
1509 {
1510         struct sockaddr *sa = svc_addr(rqstp);
1511         struct nfs4_client *conf, *unconf;
1512         struct nfsd4_session *new;
1513         struct nfsd4_clid_slot *cs_slot = NULL;
1514         bool confirm_me = false;
1515         int status = 0;
1516
1517         nfs4_lock_state();
1518         unconf = find_unconfirmed_client(&cr_ses->clientid);
1519         conf = find_confirmed_client(&cr_ses->clientid);
1520
1521         if (conf) {
1522                 cs_slot = &conf->cl_cs_slot;
1523                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1524                 if (status == nfserr_replay_cache) {
1525                         dprintk("Got a create_session replay! seqid= %d\n",
1526                                 cs_slot->sl_seqid);
1527                         /* Return the cached reply status */
1528                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1529                         goto out;
1530                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1531                         status = nfserr_seq_misordered;
1532                         dprintk("Sequence misordered!\n");
1533                         dprintk("Expected seqid= %d but got seqid= %d\n",
1534                                 cs_slot->sl_seqid, cr_ses->seqid);
1535                         goto out;
1536                 }
1537         } else if (unconf) {
1538                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1539                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1540                         status = nfserr_clid_inuse;
1541                         goto out;
1542                 }
1543
1544                 cs_slot = &unconf->cl_cs_slot;
1545                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1546                 if (status) {
1547                         /* an unconfirmed replay returns misordered */
1548                         status = nfserr_seq_misordered;
1549                         goto out;
1550                 }
1551
1552                 confirm_me = true;
1553                 conf = unconf;
1554         } else {
1555                 status = nfserr_stale_clientid;
1556                 goto out;
1557         }
1558
1559         /*
1560          * XXX: we should probably set this at creation time, and check
1561          * for consistent minorversion use throughout:
1562          */
1563         conf->cl_minorversion = 1;
1564         /*
1565          * We do not support RDMA or persistent sessions
1566          */
1567         cr_ses->flags &= ~SESSION4_PERSIST;
1568         cr_ses->flags &= ~SESSION4_RDMA;
1569
1570         status = nfserr_jukebox;
1571         new = alloc_init_session(rqstp, conf, cr_ses);
1572         if (!new)
1573                 goto out;
1574         status = nfs_ok;
1575         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1576                NFS4_MAX_SESSIONID_LEN);
1577         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1578                 sizeof(struct nfsd4_channel_attrs));
1579         cs_slot->sl_seqid++;
1580         cr_ses->seqid = cs_slot->sl_seqid;
1581
1582         /* cache solo and embedded create sessions under the state lock */
1583         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1584         if (confirm_me)
1585                 move_to_confirmed(conf);
1586 out:
1587         nfs4_unlock_state();
1588         dprintk("%s returns %d\n", __func__, ntohl(status));
1589         return status;
1590 }
1591
1592 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1593 {
1594         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1595         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1596
1597         return argp->opcnt == resp->opcnt;
1598 }
1599
1600 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1601 {
1602         switch (*dir) {
1603         case NFS4_CDFC4_FORE:
1604         case NFS4_CDFC4_BACK:
1605                 return nfs_ok;
1606         case NFS4_CDFC4_FORE_OR_BOTH:
1607         case NFS4_CDFC4_BACK_OR_BOTH:
1608                 *dir = NFS4_CDFC4_BOTH;
1609                 return nfs_ok;
1610         };
1611         return nfserr_inval;
1612 }
1613
1614 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1615                      struct nfsd4_compound_state *cstate,
1616                      struct nfsd4_bind_conn_to_session *bcts)
1617 {
1618         __be32 status;
1619
1620         if (!nfsd4_last_compound_op(rqstp))
1621                 return nfserr_not_only_op;
1622         spin_lock(&client_lock);
1623         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1624         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1625          * client_lock iself: */
1626         if (cstate->session) {
1627                 nfsd4_get_session(cstate->session);
1628                 atomic_inc(&cstate->session->se_client->cl_refcount);
1629         }
1630         spin_unlock(&client_lock);
1631         if (!cstate->session)
1632                 return nfserr_badsession;
1633
1634         status = nfsd4_map_bcts_dir(&bcts->dir);
1635         nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1636         return nfs_ok;
1637 }
1638
1639 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1640 {
1641         if (!session)
1642                 return 0;
1643         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1644 }
1645
1646 __be32
1647 nfsd4_destroy_session(struct svc_rqst *r,
1648                       struct nfsd4_compound_state *cstate,
1649                       struct nfsd4_destroy_session *sessionid)
1650 {
1651         struct nfsd4_session *ses;
1652         u32 status = nfserr_badsession;
1653
1654         /* Notes:
1655          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1656          * - Should we return nfserr_back_chan_busy if waiting for
1657          *   callbacks on to-be-destroyed session?
1658          * - Do we need to clear any callback info from previous session?
1659          */
1660
1661         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1662                 if (!nfsd4_last_compound_op(r))
1663                         return nfserr_not_only_op;
1664         }
1665         dump_sessionid(__func__, &sessionid->sessionid);
1666         spin_lock(&client_lock);
1667         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1668         if (!ses) {
1669                 spin_unlock(&client_lock);
1670                 goto out;
1671         }
1672
1673         unhash_session(ses);
1674         spin_unlock(&client_lock);
1675
1676         nfs4_lock_state();
1677         nfsd4_probe_callback_sync(ses->se_client);
1678         nfs4_unlock_state();
1679
1680         nfsd4_del_conns(ses);
1681
1682         nfsd4_put_session(ses);
1683         status = nfs_ok;
1684 out:
1685         dprintk("%s returns %d\n", __func__, ntohl(status));
1686         return status;
1687 }
1688
1689 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1690 {
1691         struct nfsd4_conn *c;
1692
1693         list_for_each_entry(c, &s->se_conns, cn_persession) {
1694                 if (c->cn_xprt == xpt) {
1695                         return c;
1696                 }
1697         }
1698         return NULL;
1699 }
1700
1701 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1702 {
1703         struct nfs4_client *clp = ses->se_client;
1704         struct nfsd4_conn *c;
1705         int ret;
1706
1707         spin_lock(&clp->cl_lock);
1708         c = __nfsd4_find_conn(new->cn_xprt, ses);
1709         if (c) {
1710                 spin_unlock(&clp->cl_lock);
1711                 free_conn(new);
1712                 return;
1713         }
1714         __nfsd4_hash_conn(new, ses);
1715         spin_unlock(&clp->cl_lock);
1716         ret = nfsd4_register_conn(new);
1717         if (ret)
1718                 /* oops; xprt is already down: */
1719                 nfsd4_conn_lost(&new->cn_xpt_user);
1720         return;
1721 }
1722
1723 __be32
1724 nfsd4_sequence(struct svc_rqst *rqstp,
1725                struct nfsd4_compound_state *cstate,
1726                struct nfsd4_sequence *seq)
1727 {
1728         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1729         struct nfsd4_session *session;
1730         struct nfsd4_slot *slot;
1731         struct nfsd4_conn *conn;
1732         int status;
1733
1734         if (resp->opcnt != 1)
1735                 return nfserr_sequence_pos;
1736
1737         /*
1738          * Will be either used or freed by nfsd4_sequence_check_conn
1739          * below.
1740          */
1741         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1742         if (!conn)
1743                 return nfserr_jukebox;
1744
1745         spin_lock(&client_lock);
1746         status = nfserr_badsession;
1747         session = find_in_sessionid_hashtbl(&seq->sessionid);
1748         if (!session)
1749                 goto out;
1750
1751         status = nfserr_badslot;
1752         if (seq->slotid >= session->se_fchannel.maxreqs)
1753                 goto out;
1754
1755         slot = session->se_slots[seq->slotid];
1756         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1757
1758         /* We do not negotiate the number of slots yet, so set the
1759          * maxslots to the session maxreqs which is used to encode
1760          * sr_highest_slotid and the sr_target_slot id to maxslots */
1761         seq->maxslots = session->se_fchannel.maxreqs;
1762
1763         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1764         if (status == nfserr_replay_cache) {
1765                 cstate->slot = slot;
1766                 cstate->session = session;
1767                 /* Return the cached reply status and set cstate->status
1768                  * for nfsd4_proc_compound processing */
1769                 status = nfsd4_replay_cache_entry(resp, seq);
1770                 cstate->status = nfserr_replay_cache;
1771                 goto out;
1772         }
1773         if (status)
1774                 goto out;
1775
1776         nfsd4_sequence_check_conn(conn, session);
1777         conn = NULL;
1778
1779         /* Success! bump slot seqid */
1780         slot->sl_inuse = true;
1781         slot->sl_seqid = seq->seqid;
1782         slot->sl_cachethis = seq->cachethis;
1783
1784         cstate->slot = slot;
1785         cstate->session = session;
1786
1787 out:
1788         /* Hold a session reference until done processing the compound. */
1789         if (cstate->session) {
1790                 struct nfs4_client *clp = session->se_client;
1791
1792                 nfsd4_get_session(cstate->session);
1793                 atomic_inc(&clp->cl_refcount);
1794                 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1795                         seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1796         }
1797         kfree(conn);
1798         spin_unlock(&client_lock);
1799         dprintk("%s: return %d\n", __func__, ntohl(status));
1800         return status;
1801 }
1802
1803 __be32
1804 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1805 {
1806         if (rc->rca_one_fs) {
1807                 if (!cstate->current_fh.fh_dentry)
1808                         return nfserr_nofilehandle;
1809                 /*
1810                  * We don't take advantage of the rca_one_fs case.
1811                  * That's OK, it's optional, we can safely ignore it.
1812                  */
1813                  return nfs_ok;
1814         }
1815         nfs4_lock_state();
1816         if (is_client_expired(cstate->session->se_client)) {
1817                 nfs4_unlock_state();
1818                 /*
1819                  * The following error isn't really legal.
1820                  * But we only get here if the client just explicitly
1821                  * destroyed the client.  Surely it no longer cares what
1822                  * error it gets back on an operation for the dead
1823                  * client.
1824                  */
1825                 return nfserr_stale_clientid;
1826         }
1827         nfsd4_create_clid_dir(cstate->session->se_client);
1828         nfs4_unlock_state();
1829         return nfs_ok;
1830 }
1831
1832 __be32
1833 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1834                   struct nfsd4_setclientid *setclid)
1835 {
1836         struct xdr_netobj       clname = { 
1837                 .len = setclid->se_namelen,
1838                 .data = setclid->se_name,
1839         };
1840         nfs4_verifier           clverifier = setclid->se_verf;
1841         unsigned int            strhashval;
1842         struct nfs4_client      *conf, *unconf, *new;
1843         __be32                  status;
1844         char                    dname[HEXDIR_LEN];
1845         
1846         if (!check_name(clname))
1847                 return nfserr_inval;
1848
1849         status = nfs4_make_rec_clidname(dname, &clname);
1850         if (status)
1851                 return status;
1852
1853         /* 
1854          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1855          * We get here on a DRC miss.
1856          */
1857
1858         strhashval = clientstr_hashval(dname);
1859
1860         nfs4_lock_state();
1861         conf = find_confirmed_client_by_str(dname, strhashval);
1862         if (conf) {
1863                 /* RFC 3530 14.2.33 CASE 0: */
1864                 status = nfserr_clid_inuse;
1865                 if (clp_used_exchangeid(conf))
1866                         goto out;
1867                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1868                         char addr_str[INET6_ADDRSTRLEN];
1869                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1870                                  sizeof(addr_str));
1871                         dprintk("NFSD: setclientid: string in use by client "
1872                                 "at %s\n", addr_str);
1873                         goto out;
1874                 }
1875         }
1876         /*
1877          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1878          * has a description of SETCLIENTID request processing consisting
1879          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1880          */
1881         unconf = find_unconfirmed_client_by_str(dname, strhashval);
1882         status = nfserr_resource;
1883         if (!conf) {
1884                 /*
1885                  * RFC 3530 14.2.33 CASE 4:
1886                  * placed first, because it is the normal case
1887                  */
1888                 if (unconf)
1889                         expire_client(unconf);
1890                 new = create_client(clname, dname, rqstp, &clverifier);
1891                 if (new == NULL)
1892                         goto out;
1893                 gen_clid(new);
1894         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1895                 /*
1896                  * RFC 3530 14.2.33 CASE 1:
1897                  * probable callback update
1898                  */
1899                 if (unconf) {
1900                         /* Note this is removing unconfirmed {*x***},
1901                          * which is stronger than RFC recommended {vxc**}.
1902                          * This has the advantage that there is at most
1903                          * one {*x***} in either list at any time.
1904                          */
1905                         expire_client(unconf);
1906                 }
1907                 new = create_client(clname, dname, rqstp, &clverifier);
1908                 if (new == NULL)
1909                         goto out;
1910                 copy_clid(new, conf);
1911         } else if (!unconf) {
1912                 /*
1913                  * RFC 3530 14.2.33 CASE 2:
1914                  * probable client reboot; state will be removed if
1915                  * confirmed.
1916                  */
1917                 new = create_client(clname, dname, rqstp, &clverifier);
1918                 if (new == NULL)
1919                         goto out;
1920                 gen_clid(new);
1921         } else {
1922                 /*
1923                  * RFC 3530 14.2.33 CASE 3:
1924                  * probable client reboot; state will be removed if
1925                  * confirmed.
1926                  */
1927                 expire_client(unconf);
1928                 new = create_client(clname, dname, rqstp, &clverifier);
1929                 if (new == NULL)
1930                         goto out;
1931                 gen_clid(new);
1932         }
1933         /*
1934          * XXX: we should probably set this at creation time, and check
1935          * for consistent minorversion use throughout:
1936          */
1937         new->cl_minorversion = 0;
1938         gen_callback(new, setclid, rqstp);
1939         add_to_unconfirmed(new, strhashval);
1940         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1941         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1942         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1943         status = nfs_ok;
1944 out:
1945         nfs4_unlock_state();
1946         return status;
1947 }
1948
1949
1950 /*
1951  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1952  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1953  * bullets, labeled as CASE1 - CASE4 below.
1954  */
1955 __be32
1956 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1957                          struct nfsd4_compound_state *cstate,
1958                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1959 {
1960         struct sockaddr *sa = svc_addr(rqstp);
1961         struct nfs4_client *conf, *unconf;
1962         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1963         clientid_t * clid = &setclientid_confirm->sc_clientid;
1964         __be32 status;
1965
1966         if (STALE_CLIENTID(clid))
1967                 return nfserr_stale_clientid;
1968         /* 
1969          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1970          * We get here on a DRC miss.
1971          */
1972
1973         nfs4_lock_state();
1974
1975         conf = find_confirmed_client(clid);
1976         unconf = find_unconfirmed_client(clid);
1977
1978         status = nfserr_clid_inuse;
1979         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1980                 goto out;
1981         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1982                 goto out;
1983
1984         /*
1985          * section 14.2.34 of RFC 3530 has a description of
1986          * SETCLIENTID_CONFIRM request processing consisting
1987          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1988          */
1989         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1990                 /*
1991                  * RFC 3530 14.2.34 CASE 1:
1992                  * callback update
1993                  */
1994                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1995                         status = nfserr_clid_inuse;
1996                 else {
1997                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
1998                         nfsd4_probe_callback(conf);
1999                         expire_client(unconf);
2000                         status = nfs_ok;
2001
2002                 }
2003         } else if (conf && !unconf) {
2004                 /*
2005                  * RFC 3530 14.2.34 CASE 2:
2006                  * probable retransmitted request; play it safe and
2007                  * do nothing.
2008                  */
2009                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2010                         status = nfserr_clid_inuse;
2011                 else
2012                         status = nfs_ok;
2013         } else if (!conf && unconf
2014                         && same_verf(&unconf->cl_confirm, &confirm)) {
2015                 /*
2016                  * RFC 3530 14.2.34 CASE 3:
2017                  * Normal case; new or rebooted client:
2018                  */
2019                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2020                         status = nfserr_clid_inuse;
2021                 } else {
2022                         unsigned int hash =
2023                                 clientstr_hashval(unconf->cl_recdir);
2024                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2025                                                             hash);
2026                         if (conf) {
2027                                 nfsd4_remove_clid_dir(conf);
2028                                 expire_client(conf);
2029                         }
2030                         move_to_confirmed(unconf);
2031                         conf = unconf;
2032                         nfsd4_probe_callback(conf);
2033                         status = nfs_ok;
2034                 }
2035         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2036             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2037                                                                 &confirm)))) {
2038                 /*
2039                  * RFC 3530 14.2.34 CASE 4:
2040                  * Client probably hasn't noticed that we rebooted yet.
2041                  */
2042                 status = nfserr_stale_clientid;
2043         } else {
2044                 /* check that we have hit one of the cases...*/
2045                 status = nfserr_clid_inuse;
2046         }
2047 out:
2048         nfs4_unlock_state();
2049         return status;
2050 }
2051
2052 /* OPEN Share state helper functions */
2053 static inline struct nfs4_file *
2054 alloc_init_file(struct inode *ino)
2055 {
2056         struct nfs4_file *fp;
2057         unsigned int hashval = file_hashval(ino);
2058
2059         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2060         if (fp) {
2061                 atomic_set(&fp->fi_ref, 1);
2062                 INIT_LIST_HEAD(&fp->fi_hash);
2063                 INIT_LIST_HEAD(&fp->fi_stateids);
2064                 INIT_LIST_HEAD(&fp->fi_delegations);
2065                 fp->fi_inode = igrab(ino);
2066                 fp->fi_id = current_fileid++;
2067                 fp->fi_had_conflict = false;
2068                 fp->fi_lease = NULL;
2069                 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2070                 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2071                 spin_lock(&recall_lock);
2072                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2073                 spin_unlock(&recall_lock);
2074                 return fp;
2075         }
2076         return NULL;
2077 }
2078
2079 static void
2080 nfsd4_free_slab(struct kmem_cache **slab)
2081 {
2082         if (*slab == NULL)
2083                 return;
2084         kmem_cache_destroy(*slab);
2085         *slab = NULL;
2086 }
2087
2088 void
2089 nfsd4_free_slabs(void)
2090 {
2091         nfsd4_free_slab(&stateowner_slab);
2092         nfsd4_free_slab(&file_slab);
2093         nfsd4_free_slab(&stateid_slab);
2094         nfsd4_free_slab(&deleg_slab);
2095 }
2096
2097 static int
2098 nfsd4_init_slabs(void)
2099 {
2100         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2101                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
2102         if (stateowner_slab == NULL)
2103                 goto out_nomem;
2104         file_slab = kmem_cache_create("nfsd4_files",
2105                         sizeof(struct nfs4_file), 0, 0, NULL);
2106         if (file_slab == NULL)
2107                 goto out_nomem;
2108         stateid_slab = kmem_cache_create("nfsd4_stateids",
2109                         sizeof(struct nfs4_stateid), 0, 0, NULL);
2110         if (stateid_slab == NULL)
2111                 goto out_nomem;
2112         deleg_slab = kmem_cache_create("nfsd4_delegations",
2113                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2114         if (deleg_slab == NULL)
2115                 goto out_nomem;
2116         return 0;
2117 out_nomem:
2118         nfsd4_free_slabs();
2119         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2120         return -ENOMEM;
2121 }
2122
2123 void
2124 nfs4_free_stateowner(struct kref *kref)
2125 {
2126         struct nfs4_stateowner *sop =
2127                 container_of(kref, struct nfs4_stateowner, so_ref);
2128         kfree(sop->so_owner.data);
2129         kmem_cache_free(stateowner_slab, sop);
2130 }
2131
2132 static inline struct nfs4_stateowner *
2133 alloc_stateowner(struct xdr_netobj *owner)
2134 {
2135         struct nfs4_stateowner *sop;
2136
2137         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2138                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2139                         memcpy(sop->so_owner.data, owner->data, owner->len);
2140                         sop->so_owner.len = owner->len;
2141                         kref_init(&sop->so_ref);
2142                         return sop;
2143                 } 
2144                 kmem_cache_free(stateowner_slab, sop);
2145         }
2146         return NULL;
2147 }
2148
2149 static struct nfs4_stateowner *
2150 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2151         struct nfs4_stateowner *sop;
2152         struct nfs4_replay *rp;
2153         unsigned int idhashval;
2154
2155         if (!(sop = alloc_stateowner(&open->op_owner)))
2156                 return NULL;
2157         idhashval = ownerid_hashval(current_ownerid);
2158         INIT_LIST_HEAD(&sop->so_idhash);
2159         INIT_LIST_HEAD(&sop->so_strhash);
2160         INIT_LIST_HEAD(&sop->so_perclient);
2161         INIT_LIST_HEAD(&sop->so_stateids);
2162         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
2163         INIT_LIST_HEAD(&sop->so_close_lru);
2164         sop->so_time = 0;
2165         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2166         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2167         list_add(&sop->so_perclient, &clp->cl_openowners);
2168         sop->so_is_open_owner = 1;
2169         sop->so_id = current_ownerid++;
2170         sop->so_client = clp;
2171         sop->so_seqid = open->op_seqid;
2172         sop->so_confirmed = 0;
2173         rp = &sop->so_replay;
2174         rp->rp_status = nfserr_serverfault;
2175         rp->rp_buflen = 0;
2176         rp->rp_buf = rp->rp_ibuf;
2177         return sop;
2178 }
2179
2180 static inline void
2181 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2182         struct nfs4_stateowner *sop = open->op_stateowner;
2183         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2184
2185         INIT_LIST_HEAD(&stp->st_hash);
2186         INIT_LIST_HEAD(&stp->st_perstateowner);
2187         INIT_LIST_HEAD(&stp->st_lockowners);
2188         INIT_LIST_HEAD(&stp->st_perfile);
2189         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2190         list_add(&stp->st_perstateowner, &sop->so_stateids);
2191         list_add(&stp->st_perfile, &fp->fi_stateids);
2192         stp->st_stateowner = sop;
2193         get_nfs4_file(fp);
2194         stp->st_file = fp;
2195         stp->st_stateid.si_boot = boot_time;
2196         stp->st_stateid.si_stateownerid = sop->so_id;
2197         stp->st_stateid.si_fileid = fp->fi_id;
2198         stp->st_stateid.si_generation = 0;
2199         stp->st_access_bmap = 0;
2200         stp->st_deny_bmap = 0;
2201         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2202                   &stp->st_access_bmap);
2203         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2204         stp->st_openstp = NULL;
2205 }
2206
2207 static void
2208 move_to_close_lru(struct nfs4_stateowner *sop)
2209 {
2210         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2211
2212         list_move_tail(&sop->so_close_lru, &close_lru);
2213         sop->so_time = get_seconds();
2214 }
2215
2216 static int
2217 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2218                                                         clientid_t *clid)
2219 {
2220         return (sop->so_owner.len == owner->len) &&
2221                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2222                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2223 }
2224
2225 static struct nfs4_stateowner *
2226 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2227 {
2228         struct nfs4_stateowner *so = NULL;
2229
2230         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2231                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2232                         return so;
2233         }
2234         return NULL;
2235 }
2236
2237 /* search file_hashtbl[] for file */
2238 static struct nfs4_file *
2239 find_file(struct inode *ino)
2240 {
2241         unsigned int hashval = file_hashval(ino);
2242         struct nfs4_file *fp;
2243
2244         spin_lock(&recall_lock);
2245         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2246                 if (fp->fi_inode == ino) {
2247                         get_nfs4_file(fp);
2248                         spin_unlock(&recall_lock);
2249                         return fp;
2250                 }
2251         }
2252         spin_unlock(&recall_lock);
2253         return NULL;
2254 }
2255
2256 static inline int access_valid(u32 x, u32 minorversion)
2257 {
2258         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2259                 return 0;
2260         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2261                 return 0;
2262         x &= ~NFS4_SHARE_ACCESS_MASK;
2263         if (minorversion && x) {
2264                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2265                         return 0;
2266                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2267                         return 0;
2268                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2269         }
2270         if (x)
2271                 return 0;
2272         return 1;
2273 }
2274
2275 static inline int deny_valid(u32 x)
2276 {
2277         /* Note: unlike access bits, deny bits may be zero. */
2278         return x <= NFS4_SHARE_DENY_BOTH;
2279 }
2280
2281 /*
2282  * Called to check deny when READ with all zero stateid or
2283  * WRITE with all zero or all one stateid
2284  */
2285 static __be32
2286 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2287 {
2288         struct inode *ino = current_fh->fh_dentry->d_inode;
2289         struct nfs4_file *fp;
2290         struct nfs4_stateid *stp;
2291         __be32 ret;
2292
2293         dprintk("NFSD: nfs4_share_conflict\n");
2294
2295         fp = find_file(ino);
2296         if (!fp)
2297                 return nfs_ok;
2298         ret = nfserr_locked;
2299         /* Search for conflicting share reservations */
2300         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2301                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2302                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2303                         goto out;
2304         }
2305         ret = nfs_ok;
2306 out:
2307         put_nfs4_file(fp);
2308         return ret;
2309 }
2310
2311 static inline void
2312 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2313 {
2314         if (share_access & NFS4_SHARE_ACCESS_WRITE)
2315                 nfs4_file_put_access(fp, O_WRONLY);
2316         if (share_access & NFS4_SHARE_ACCESS_READ)
2317                 nfs4_file_put_access(fp, O_RDONLY);
2318 }
2319
2320 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2321 {
2322         /* We're assuming the state code never drops its reference
2323          * without first removing the lease.  Since we're in this lease
2324          * callback (and since the lease code is serialized by the kernel
2325          * lock) we know the server hasn't removed the lease yet, we know
2326          * it's safe to take a reference: */
2327         atomic_inc(&dp->dl_count);
2328
2329         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2330
2331         /* only place dl_time is set. protected by lock_flocks*/
2332         dp->dl_time = get_seconds();
2333
2334         nfsd4_cb_recall(dp);
2335 }
2336
2337 /* Called from break_lease() with lock_flocks() held. */
2338 static void nfsd_break_deleg_cb(struct file_lock *fl)
2339 {
2340         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2341         struct nfs4_delegation *dp;
2342
2343         BUG_ON(!fp);
2344         /* We assume break_lease is only called once per lease: */
2345         BUG_ON(fp->fi_had_conflict);
2346         /*
2347          * We don't want the locks code to timeout the lease for us;
2348          * we'll remove it ourself if a delegation isn't returned
2349          * in time:
2350          */
2351         fl->fl_break_time = 0;
2352
2353         spin_lock(&recall_lock);
2354         fp->fi_had_conflict = true;
2355         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2356                 nfsd_break_one_deleg(dp);
2357         spin_unlock(&recall_lock);
2358 }
2359
2360 static
2361 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2362 {
2363         if (arg & F_UNLCK)
2364                 return lease_modify(onlist, arg);
2365         else
2366                 return -EAGAIN;
2367 }
2368
2369 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2370         .fl_break = nfsd_break_deleg_cb,
2371         .fl_change = nfsd_change_deleg_cb,
2372 };
2373
2374
2375 __be32
2376 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2377                     struct nfsd4_open *open)
2378 {
2379         clientid_t *clientid = &open->op_clientid;
2380         struct nfs4_client *clp = NULL;
2381         unsigned int strhashval;
2382         struct nfs4_stateowner *sop = NULL;
2383
2384         if (!check_name(open->op_owner))
2385                 return nfserr_inval;
2386
2387         if (STALE_CLIENTID(&open->op_clientid))
2388                 return nfserr_stale_clientid;
2389
2390         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2391         sop = find_openstateowner_str(strhashval, open);
2392         open->op_stateowner = sop;
2393         if (!sop) {
2394                 /* Make sure the client's lease hasn't expired. */
2395                 clp = find_confirmed_client(clientid);
2396                 if (clp == NULL)
2397                         return nfserr_expired;
2398                 goto renew;
2399         }
2400         /* When sessions are used, skip open sequenceid processing */
2401         if (nfsd4_has_session(cstate))
2402                 goto renew;
2403         if (!sop->so_confirmed) {
2404                 /* Replace unconfirmed owners without checking for replay. */
2405                 clp = sop->so_client;
2406                 release_openowner(sop);
2407                 open->op_stateowner = NULL;
2408                 goto renew;
2409         }
2410         if (open->op_seqid == sop->so_seqid - 1) {
2411                 if (sop->so_replay.rp_buflen)
2412                         return nfserr_replay_me;
2413                 /* The original OPEN failed so spectacularly
2414                  * that we don't even have replay data saved!
2415                  * Therefore, we have no choice but to continue
2416                  * processing this OPEN; presumably, we'll
2417                  * fail again for the same reason.
2418                  */
2419                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2420                 goto renew;
2421         }
2422         if (open->op_seqid != sop->so_seqid)
2423                 return nfserr_bad_seqid;
2424 renew:
2425         if (open->op_stateowner == NULL) {
2426                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2427                 if (sop == NULL)
2428                         return nfserr_resource;
2429                 open->op_stateowner = sop;
2430         }
2431         list_del_init(&sop->so_close_lru);
2432         renew_client(sop->so_client);
2433         return nfs_ok;
2434 }
2435
2436 static inline __be32
2437 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2438 {
2439         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2440                 return nfserr_openmode;
2441         else
2442                 return nfs_ok;
2443 }
2444
2445 static struct nfs4_delegation *
2446 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2447 {
2448         struct nfs4_delegation *dp = NULL;
2449
2450         spin_lock(&recall_lock);
2451         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2452                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2453                         break;
2454         }
2455         spin_unlock(&recall_lock);
2456         return dp;
2457 }
2458
2459 int share_access_to_flags(u32 share_access)
2460 {
2461         share_access &= ~NFS4_SHARE_WANT_MASK;
2462
2463         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2464 }
2465
2466 static __be32
2467 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2468                 struct nfs4_delegation **dp)
2469 {
2470         int flags;
2471         __be32 status = nfserr_bad_stateid;
2472
2473         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2474         if (*dp == NULL)
2475                 goto out;
2476         flags = share_access_to_flags(open->op_share_access);
2477         status = nfs4_check_delegmode(*dp, flags);
2478         if (status)
2479                 *dp = NULL;
2480 out:
2481         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2482                 return nfs_ok;
2483         if (status)
2484                 return status;
2485         open->op_stateowner->so_confirmed = 1;
2486         return nfs_ok;
2487 }
2488
2489 static __be32
2490 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2491 {
2492         struct nfs4_stateid *local;
2493         __be32 status = nfserr_share_denied;
2494         struct nfs4_stateowner *sop = open->op_stateowner;
2495
2496         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2497                 /* ignore lock owners */
2498                 if (local->st_stateowner->so_is_open_owner == 0)
2499                         continue;
2500                 /* remember if we have seen this open owner */
2501                 if (local->st_stateowner == sop)
2502                         *stpp = local;
2503                 /* check for conflicting share reservations */
2504                 if (!test_share(local, open))
2505                         goto out;
2506         }
2507         status = 0;
2508 out:
2509         return status;
2510 }
2511
2512 static inline struct nfs4_stateid *
2513 nfs4_alloc_stateid(void)
2514 {
2515         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2516 }
2517
2518 static inline int nfs4_access_to_access(u32 nfs4_access)
2519 {
2520         int flags = 0;
2521
2522         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2523                 flags |= NFSD_MAY_READ;
2524         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2525                 flags |= NFSD_MAY_WRITE;
2526         return flags;
2527 }
2528
2529 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2530 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2531 {
2532         __be32 status;
2533         int oflag = nfs4_access_to_omode(nfs4_access);
2534         int access = nfs4_access_to_access(nfs4_access);
2535
2536         if (!fp->fi_fds[oflag]) {
2537                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2538                         &fp->fi_fds[oflag]);
2539                 if (status)
2540                         return status;
2541         }
2542         nfs4_file_get_access(fp, oflag);
2543
2544         return nfs_ok;
2545 }
2546
2547 static __be32
2548 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2549                 struct nfs4_file *fp, struct svc_fh *cur_fh,
2550                 struct nfsd4_open *open)
2551 {
2552         struct nfs4_stateid *stp;
2553         __be32 status;
2554
2555         stp = nfs4_alloc_stateid();
2556         if (stp == NULL)
2557                 return nfserr_resource;
2558
2559         status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2560         if (status) {
2561                 kmem_cache_free(stateid_slab, stp);
2562                 return status;
2563         }
2564         *stpp = stp;
2565         return 0;
2566 }
2567
2568 static inline __be32
2569 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2570                 struct nfsd4_open *open)
2571 {
2572         struct iattr iattr = {
2573                 .ia_valid = ATTR_SIZE,
2574                 .ia_size = 0,
2575         };
2576         if (!open->op_truncate)
2577                 return 0;
2578         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2579                 return nfserr_inval;
2580         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2581 }
2582
2583 static __be32
2584 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2585 {
2586         u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2587         bool new_access;
2588         __be32 status;
2589
2590         new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2591         if (new_access) {
2592                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, op_share_access);
2593                 if (status)
2594                         return status;
2595         }
2596         status = nfsd4_truncate(rqstp, cur_fh, open);
2597         if (status) {
2598                 if (new_access) {
2599                         int oflag = nfs4_access_to_omode(new_access);
2600                         nfs4_file_put_access(fp, oflag);
2601                 }
2602                 return status;
2603         }
2604         /* remember the open */
2605         __set_bit(op_share_access, &stp->st_access_bmap);
2606         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2607
2608         return nfs_ok;
2609 }
2610
2611
2612 static void
2613 nfs4_set_claim_prev(struct nfsd4_open *open)
2614 {
2615         open->op_stateowner->so_confirmed = 1;
2616         open->op_stateowner->so_client->cl_firststate = 1;
2617 }
2618
2619 /* Should we give out recallable state?: */
2620 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2621 {
2622         if (clp->cl_cb_state == NFSD4_CB_UP)
2623                 return true;
2624         /*
2625          * In the sessions case, since we don't have to establish a
2626          * separate connection for callbacks, we assume it's OK
2627          * until we hear otherwise:
2628          */
2629         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2630 }
2631
2632 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2633 {
2634         struct file_lock *fl;
2635
2636         fl = locks_alloc_lock();
2637         if (!fl)
2638                 return NULL;
2639         locks_init_lock(fl);
2640         fl->fl_lmops = &nfsd_lease_mng_ops;
2641         fl->fl_flags = FL_LEASE;
2642         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2643         fl->fl_end = OFFSET_MAX;
2644         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2645         fl->fl_pid = current->tgid;
2646         return fl;
2647 }
2648
2649 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2650 {
2651         struct nfs4_file *fp = dp->dl_file;
2652         struct file_lock *fl;
2653         int status;
2654
2655         fl = nfs4_alloc_init_lease(dp, flag);
2656         if (!fl)
2657                 return -ENOMEM;
2658         fl->fl_file = find_readable_file(fp);
2659         list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2660         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2661         if (status) {
2662                 list_del_init(&dp->dl_perclnt);
2663                 locks_free_lock(fl);
2664                 return -ENOMEM;
2665         }
2666         fp->fi_lease = fl;
2667         fp->fi_deleg_file = fl->fl_file;
2668         get_file(fp->fi_deleg_file);
2669         atomic_set(&fp->fi_delegees, 1);
2670         list_add(&dp->dl_perfile, &fp->fi_delegations);
2671         return 0;
2672 }
2673
2674 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2675 {
2676         struct nfs4_file *fp = dp->dl_file;
2677
2678         if (!fp->fi_lease)
2679                 return nfs4_setlease(dp, flag);
2680         spin_lock(&recall_lock);
2681         if (fp->fi_had_conflict) {
2682                 spin_unlock(&recall_lock);
2683                 return -EAGAIN;
2684         }
2685         atomic_inc(&fp->fi_delegees);
2686         list_add(&dp->dl_perfile, &fp->fi_delegations);
2687         spin_unlock(&recall_lock);
2688         list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2689         return 0;
2690 }
2691
2692 /*
2693  * Attempt to hand out a delegation.
2694  */
2695 static void
2696 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2697 {
2698         struct nfs4_delegation *dp;
2699         struct nfs4_stateowner *sop = stp->st_stateowner;
2700         int cb_up;
2701         int status, flag = 0;
2702
2703         cb_up = nfsd4_cb_channel_good(sop->so_client);
2704         flag = NFS4_OPEN_DELEGATE_NONE;
2705         open->op_recall = 0;
2706         switch (open->op_claim_type) {
2707                 case NFS4_OPEN_CLAIM_PREVIOUS:
2708                         if (!cb_up)
2709                                 open->op_recall = 1;
2710                         flag = open->op_delegate_type;
2711                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2712                                 goto out;
2713                         break;
2714                 case NFS4_OPEN_CLAIM_NULL:
2715                         /* Let's not give out any delegations till everyone's
2716                          * had the chance to reclaim theirs.... */
2717                         if (locks_in_grace())
2718                                 goto out;
2719                         if (!cb_up || !sop->so_confirmed)
2720                                 goto out;
2721                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2722                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2723                         else
2724                                 flag = NFS4_OPEN_DELEGATE_READ;
2725                         break;
2726                 default:
2727                         goto out;
2728         }
2729
2730         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2731         if (dp == NULL)
2732                 goto out_no_deleg;
2733         status = nfs4_set_delegation(dp, flag);
2734         if (status)
2735                 goto out_free;
2736
2737         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2738
2739         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2740                 STATEID_VAL(&dp->dl_stateid));
2741 out:
2742         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2743                         && flag == NFS4_OPEN_DELEGATE_NONE
2744                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2745                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2746         open->op_delegate_type = flag;
2747         return;
2748 out_free:
2749         nfs4_put_delegation(dp);
2750 out_no_deleg:
2751         flag = NFS4_OPEN_DELEGATE_NONE;
2752         goto out;
2753 }
2754
2755 /*
2756  * called with nfs4_lock_state() held.
2757  */
2758 __be32
2759 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2760 {
2761         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2762         struct nfs4_file *fp = NULL;
2763         struct inode *ino = current_fh->fh_dentry->d_inode;
2764         struct nfs4_stateid *stp = NULL;
2765         struct nfs4_delegation *dp = NULL;
2766         __be32 status;
2767
2768         status = nfserr_inval;
2769         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2770                         || !deny_valid(open->op_share_deny))
2771                 goto out;
2772         /*
2773          * Lookup file; if found, lookup stateid and check open request,
2774          * and check for delegations in the process of being recalled.
2775          * If not found, create the nfs4_file struct
2776          */
2777         fp = find_file(ino);
2778         if (fp) {
2779                 if ((status = nfs4_check_open(fp, open, &stp)))
2780                         goto out;
2781                 status = nfs4_check_deleg(fp, open, &dp);
2782                 if (status)
2783                         goto out;
2784         } else {
2785                 status = nfserr_bad_stateid;
2786                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2787                         goto out;
2788                 status = nfserr_resource;
2789                 fp = alloc_init_file(ino);
2790                 if (fp == NULL)
2791                         goto out;
2792         }
2793
2794         /*
2795          * OPEN the file, or upgrade an existing OPEN.
2796          * If truncate fails, the OPEN fails.
2797          */
2798         if (stp) {
2799                 /* Stateid was found, this is an OPEN upgrade */
2800                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2801                 if (status)
2802                         goto out;
2803                 update_stateid(&stp->st_stateid);
2804         } else {
2805                 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2806                 if (status)
2807                         goto out;
2808                 init_stateid(stp, fp, open);
2809                 status = nfsd4_truncate(rqstp, current_fh, open);
2810                 if (status) {
2811                         release_open_stateid(stp);
2812                         goto out;
2813                 }
2814                 if (nfsd4_has_session(&resp->cstate))
2815                         update_stateid(&stp->st_stateid);
2816         }
2817         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2818
2819         if (nfsd4_has_session(&resp->cstate))
2820                 open->op_stateowner->so_confirmed = 1;
2821
2822         /*
2823         * Attempt to hand out a delegation. No error return, because the
2824         * OPEN succeeds even if we fail.
2825         */
2826         nfs4_open_delegation(current_fh, open, stp);
2827
2828         status = nfs_ok;
2829
2830         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2831                 STATEID_VAL(&stp->st_stateid));
2832 out:
2833         if (fp)
2834                 put_nfs4_file(fp);
2835         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2836                 nfs4_set_claim_prev(open);
2837         /*
2838         * To finish the open response, we just need to set the rflags.
2839         */
2840         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2841         if (!open->op_stateowner->so_confirmed &&
2842             !nfsd4_has_session(&resp->cstate))
2843                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2844
2845         return status;
2846 }
2847
2848 __be32
2849 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2850             clientid_t *clid)
2851 {
2852         struct nfs4_client *clp;
2853         __be32 status;
2854
2855         nfs4_lock_state();
2856         dprintk("process_renew(%08x/%08x): starting\n", 
2857                         clid->cl_boot, clid->cl_id);
2858         status = nfserr_stale_clientid;
2859         if (STALE_CLIENTID(clid))
2860                 goto out;
2861         clp = find_confirmed_client(clid);
2862         status = nfserr_expired;
2863         if (clp == NULL) {
2864                 /* We assume the client took too long to RENEW. */
2865                 dprintk("nfsd4_renew: clientid not found!\n");
2866                 goto out;
2867         }
2868         renew_client(clp);
2869         status = nfserr_cb_path_down;
2870         if (!list_empty(&clp->cl_delegations)
2871                         && clp->cl_cb_state != NFSD4_CB_UP)
2872                 goto out;
2873         status = nfs_ok;
2874 out:
2875         nfs4_unlock_state();
2876         return status;
2877 }
2878
2879 struct lock_manager nfsd4_manager = {
2880 };
2881
2882 static void
2883 nfsd4_end_grace(void)
2884 {
2885         dprintk("NFSD: end of grace period\n");
2886         nfsd4_recdir_purge_old();
2887         locks_end_grace(&nfsd4_manager);
2888         /*
2889          * Now that every NFSv4 client has had the chance to recover and
2890          * to see the (possibly new, possibly shorter) lease time, we
2891          * can safely set the next grace time to the current lease time:
2892          */
2893         nfsd4_grace = nfsd4_lease;
2894 }
2895
2896 static time_t
2897 nfs4_laundromat(void)
2898 {
2899         struct nfs4_client *clp;
2900         struct nfs4_stateowner *sop;
2901         struct nfs4_delegation *dp;
2902         struct list_head *pos, *next, reaplist;
2903         time_t cutoff = get_seconds() - nfsd4_lease;
2904         time_t t, clientid_val = nfsd4_lease;
2905         time_t u, test_val = nfsd4_lease;
2906
2907         nfs4_lock_state();
2908
2909         dprintk("NFSD: laundromat service - starting\n");
2910         if (locks_in_grace())
2911                 nfsd4_end_grace();
2912         INIT_LIST_HEAD(&reaplist);
2913         spin_lock(&client_lock);
2914         list_for_each_safe(pos, next, &client_lru) {
2915                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2916                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2917                         t = clp->cl_time - cutoff;
2918                         if (clientid_val > t)
2919                                 clientid_val = t;
2920                         break;
2921                 }
2922                 if (atomic_read(&clp->cl_refcount)) {
2923                         dprintk("NFSD: client in use (clientid %08x)\n",
2924                                 clp->cl_clientid.cl_id);
2925                         continue;
2926                 }
2927                 unhash_client_locked(clp);
2928                 list_add(&clp->cl_lru, &reaplist);
2929         }
2930         spin_unlock(&client_lock);
2931         list_for_each_safe(pos, next, &reaplist) {
2932                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2933                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2934                         clp->cl_clientid.cl_id);
2935                 nfsd4_remove_clid_dir(clp);
2936                 expire_client(clp);
2937         }
2938         spin_lock(&recall_lock);
2939         list_for_each_safe(pos, next, &del_recall_lru) {
2940                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2941                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2942                         u = dp->dl_time - cutoff;
2943                         if (test_val > u)
2944                                 test_val = u;
2945                         break;
2946                 }
2947                 list_move(&dp->dl_recall_lru, &reaplist);
2948         }
2949         spin_unlock(&recall_lock);
2950         list_for_each_safe(pos, next, &reaplist) {
2951                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2952                 list_del_init(&dp->dl_recall_lru);
2953                 unhash_delegation(dp);
2954         }
2955         test_val = nfsd4_lease;
2956         list_for_each_safe(pos, next, &close_lru) {
2957                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2958                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2959                         u = sop->so_time - cutoff;
2960                         if (test_val > u)
2961                                 test_val = u;
2962                         break;
2963                 }
2964                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2965                         sop->so_id);
2966                 release_openowner(sop);
2967         }
2968         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2969                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2970         nfs4_unlock_state();
2971         return clientid_val;
2972 }
2973
2974 static struct workqueue_struct *laundry_wq;
2975 static void laundromat_main(struct work_struct *);
2976 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2977
2978 static void
2979 laundromat_main(struct work_struct *not_used)
2980 {
2981         time_t t;
2982
2983         t = nfs4_laundromat();
2984         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2985         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2986 }
2987
2988 static struct nfs4_stateowner *
2989 search_close_lru(u32 st_id, int flags)
2990 {
2991         struct nfs4_stateowner *local = NULL;
2992
2993         if (flags & CLOSE_STATE) {
2994                 list_for_each_entry(local, &close_lru, so_close_lru) {
2995                         if (local->so_id == st_id)
2996                                 return local;
2997                 }
2998         }
2999         return NULL;
3000 }
3001
3002 static inline int
3003 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
3004 {
3005         return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
3006 }
3007
3008 static int
3009 STALE_STATEID(stateid_t *stateid)
3010 {
3011         if (stateid->si_boot == boot_time)
3012                 return 0;
3013         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3014                 STATEID_VAL(stateid));
3015         return 1;
3016 }
3017
3018 static inline int
3019 access_permit_read(unsigned long access_bmap)
3020 {
3021         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3022                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3023                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3024 }
3025
3026 static inline int
3027 access_permit_write(unsigned long access_bmap)
3028 {
3029         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3030                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3031 }
3032
3033 static
3034 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
3035 {
3036         __be32 status = nfserr_openmode;
3037
3038         /* For lock stateid's, we test the parent open, not the lock: */
3039         if (stp->st_openstp)
3040                 stp = stp->st_openstp;
3041         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3042                 goto out;
3043         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3044                 goto out;
3045         status = nfs_ok;
3046 out:
3047         return status;
3048 }
3049
3050 static inline __be32
3051 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3052 {
3053         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3054                 return nfs_ok;
3055         else if (locks_in_grace()) {
3056                 /* Answer in remaining cases depends on existance of
3057                  * conflicting state; so we must wait out the grace period. */
3058                 return nfserr_grace;
3059         } else if (flags & WR_STATE)
3060                 return nfs4_share_conflict(current_fh,
3061                                 NFS4_SHARE_DENY_WRITE);
3062         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3063                 return nfs4_share_conflict(current_fh,
3064                                 NFS4_SHARE_DENY_READ);
3065 }
3066
3067 /*
3068  * Allow READ/WRITE during grace period on recovered state only for files
3069  * that are not able to provide mandatory locking.
3070  */
3071 static inline int
3072 grace_disallows_io(struct inode *inode)
3073 {
3074         return locks_in_grace() && mandatory_lock(inode);
3075 }
3076
3077 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
3078 {
3079         /*
3080          * When sessions are used the stateid generation number is ignored
3081          * when it is zero.
3082          */
3083         if ((flags & HAS_SESSION) && in->si_generation == 0)
3084                 goto out;
3085
3086         /* If the client sends us a stateid from the future, it's buggy: */
3087         if (in->si_generation > ref->si_generation)
3088                 return nfserr_bad_stateid;
3089         /*
3090          * The following, however, can happen.  For example, if the
3091          * client sends an open and some IO at the same time, the open
3092          * may bump si_generation while the IO is still in flight.
3093          * Thanks to hard links and renames, the client never knows what
3094          * file an open will affect.  So it could avoid that situation
3095          * only by serializing all opens and IO from the same open
3096          * owner.  To recover from the old_stateid error, the client
3097          * will just have to retry the IO:
3098          */
3099         if (in->si_generation < ref->si_generation)
3100                 return nfserr_old_stateid;
3101 out:
3102         return nfs_ok;
3103 }
3104
3105 static int is_delegation_stateid(stateid_t *stateid)
3106 {
3107         return stateid->si_fileid == 0;
3108 }
3109
3110 /*
3111 * Checks for stateid operations
3112 */
3113 __be32
3114 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3115                            stateid_t *stateid, int flags, struct file **filpp)
3116 {
3117         struct nfs4_stateid *stp = NULL;
3118         struct nfs4_delegation *dp = NULL;
3119         struct svc_fh *current_fh = &cstate->current_fh;
3120         struct inode *ino = current_fh->fh_dentry->d_inode;
3121         __be32 status;
3122
3123         if (filpp)
3124                 *filpp = NULL;
3125
3126         if (grace_disallows_io(ino))
3127                 return nfserr_grace;
3128
3129         if (nfsd4_has_session(cstate))
3130                 flags |= HAS_SESSION;
3131
3132         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3133                 return check_special_stateids(current_fh, stateid, flags);
3134
3135         status = nfserr_stale_stateid;
3136         if (STALE_STATEID(stateid)) 
3137                 goto out;
3138
3139         /*
3140          * We assume that any stateid that has the current boot time,
3141          * but that we can't find, is expired:
3142          */
3143         status = nfserr_expired;
3144         if (is_delegation_stateid(stateid)) {
3145                 dp = find_delegation_stateid(ino, stateid);
3146                 if (!dp)
3147                         goto out;
3148                 status = check_stateid_generation(stateid, &dp->dl_stateid,
3149                                                   flags);
3150                 if (status)
3151                         goto out;
3152                 status = nfs4_check_delegmode(dp, flags);
3153                 if (status)
3154                         goto out;
3155                 renew_client(dp->dl_client);
3156                 if (filpp) {
3157                         *filpp = dp->dl_file->fi_deleg_file;
3158                         BUG_ON(!*filpp);
3159                 }
3160         } else { /* open or lock stateid */
3161                 stp = find_stateid(stateid, flags);
3162                 if (!stp)
3163                         goto out;
3164                 status = nfserr_bad_stateid;
3165                 if (nfs4_check_fh(current_fh, stp))
3166                         goto out;
3167                 if (!stp->st_stateowner->so_confirmed)
3168                         goto out;
3169                 status = check_stateid_generation(stateid, &stp->st_stateid,
3170                                                   flags);
3171                 if (status)
3172                         goto out;
3173                 status = nfs4_check_openmode(stp, flags);
3174                 if (status)
3175                         goto out;
3176                 renew_client(stp->st_stateowner->so_client);
3177                 if (filpp) {
3178                         if (flags & RD_STATE)
3179                                 *filpp = find_readable_file(stp->st_file);
3180                         else
3181                                 *filpp = find_writeable_file(stp->st_file);
3182                 }
3183         }
3184         status = nfs_ok;
3185 out:
3186         return status;
3187 }
3188
3189 static inline int
3190 setlkflg (int type)
3191 {
3192         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3193                 RD_STATE : WR_STATE;
3194 }
3195
3196 /* 
3197  * Checks for sequence id mutating operations. 
3198  */
3199 static __be32
3200 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3201                          stateid_t *stateid, int flags,
3202                          struct nfs4_stateowner **sopp,
3203                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3204 {
3205         struct nfs4_stateid *stp;
3206         struct nfs4_stateowner *sop;
3207         struct svc_fh *current_fh = &cstate->current_fh;
3208         __be32 status;
3209
3210         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3211                 seqid, STATEID_VAL(stateid));
3212
3213         *stpp = NULL;
3214         *sopp = NULL;
3215
3216         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3217                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3218                 return nfserr_bad_stateid;
3219         }
3220
3221         if (STALE_STATEID(stateid))
3222                 return nfserr_stale_stateid;
3223
3224         if (nfsd4_has_session(cstate))
3225                 flags |= HAS_SESSION;
3226
3227         /*
3228         * We return BAD_STATEID if filehandle doesn't match stateid, 
3229         * the confirmed flag is incorrecly set, or the generation 
3230         * number is incorrect.  
3231         */
3232         stp = find_stateid(stateid, flags);
3233         if (stp == NULL) {
3234                 /*
3235                  * Also, we should make sure this isn't just the result of
3236                  * a replayed close:
3237                  */
3238                 sop = search_close_lru(stateid->si_stateownerid, flags);
3239                 /* It's not stale; let's assume it's expired: */
3240                 if (sop == NULL)
3241                         return nfserr_expired;
3242                 *sopp = sop;
3243                 goto check_replay;
3244         }
3245
3246         *stpp = stp;
3247         *sopp = sop = stp->st_stateowner;
3248
3249         if (lock) {
3250                 clientid_t *lockclid = &lock->v.new.clientid;
3251                 struct nfs4_client *clp = sop->so_client;
3252                 int lkflg = 0;
3253                 __be32 status;
3254
3255                 lkflg = setlkflg(lock->lk_type);
3256
3257                 if (lock->lk_is_new) {
3258                         if (!sop->so_is_open_owner)
3259                                 return nfserr_bad_stateid;
3260                         if (!(flags & HAS_SESSION) &&
3261                             !same_clid(&clp->cl_clientid, lockclid))
3262                                 return nfserr_bad_stateid;
3263                         /* stp is the open stateid */
3264                         status = nfs4_check_openmode(stp, lkflg);
3265                         if (status)
3266                                 return status;
3267                 } else {
3268                         /* stp is the lock stateid */
3269                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3270                         if (status)
3271                                 return status;
3272                }
3273         }
3274
3275         if (nfs4_check_fh(current_fh, stp)) {
3276                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3277                 return nfserr_bad_stateid;
3278         }
3279
3280         /*
3281         *  We now validate the seqid and stateid generation numbers.
3282         *  For the moment, we ignore the possibility of 
3283         *  generation number wraparound.
3284         */
3285         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3286                 goto check_replay;
3287
3288         if (sop->so_confirmed && flags & CONFIRM) {
3289                 dprintk("NFSD: preprocess_seqid_op: expected"
3290                                 " unconfirmed stateowner!\n");
3291                 return nfserr_bad_stateid;
3292         }
3293         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3294                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3295                                 " confirmed yet!\n");
3296                 return nfserr_bad_stateid;
3297         }
3298         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3299         if (status)
3300                 return status;
3301         renew_client(sop->so_client);
3302         return nfs_ok;
3303
3304 check_replay:
3305         if (seqid == sop->so_seqid - 1) {
3306                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3307                 /* indicate replay to calling function */
3308                 return nfserr_replay_me;
3309         }
3310         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3311                         sop->so_seqid, seqid);
3312         *sopp = NULL;
3313         return nfserr_bad_seqid;
3314 }
3315
3316 __be32
3317 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3318                    struct nfsd4_open_confirm *oc)
3319 {
3320         __be32 status;
3321         struct nfs4_stateowner *sop;
3322         struct nfs4_stateid *stp;
3323
3324         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3325                         (int)cstate->current_fh.fh_dentry->d_name.len,
3326                         cstate->current_fh.fh_dentry->d_name.name);
3327
3328         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3329         if (status)
3330                 return status;
3331
3332         nfs4_lock_state();
3333
3334         if ((status = nfs4_preprocess_seqid_op(cstate,
3335                                         oc->oc_seqid, &oc->oc_req_stateid,
3336                                         CONFIRM | OPEN_STATE,
3337                                         &oc->oc_stateowner, &stp, NULL)))
3338                 goto out; 
3339
3340         sop = oc->oc_stateowner;
3341         sop->so_confirmed = 1;
3342         update_stateid(&stp->st_stateid);
3343         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3344         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3345                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3346
3347         nfsd4_create_clid_dir(sop->so_client);
3348 out:
3349         if (oc->oc_stateowner) {
3350                 nfs4_get_stateowner(oc->oc_stateowner);
3351                 cstate->replay_owner = oc->oc_stateowner;
3352         }
3353         nfs4_unlock_state();
3354         return status;
3355 }
3356
3357
3358 /*
3359  * unset all bits in union bitmap (bmap) that
3360  * do not exist in share (from successful OPEN_DOWNGRADE)
3361  */
3362 static void
3363 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3364 {
3365         int i;
3366         for (i = 1; i < 4; i++) {
3367                 if ((i & access) != i)
3368                         __clear_bit(i, bmap);
3369         }
3370 }
3371
3372 static void
3373 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3374 {
3375         int i;
3376         for (i = 0; i < 4; i++) {
3377                 if ((i & deny) != i)
3378                         __clear_bit(i, bmap);
3379         }
3380 }
3381
3382 __be32
3383 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3384                      struct nfsd4_compound_state *cstate,
3385                      struct nfsd4_open_downgrade *od)
3386 {
3387         __be32 status;
3388         struct nfs4_stateid *stp;
3389         unsigned int share_access;
3390
3391         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3392                         (int)cstate->current_fh.fh_dentry->d_name.len,
3393                         cstate->current_fh.fh_dentry->d_name.name);
3394
3395         if (!access_valid(od->od_share_access, cstate->minorversion)
3396                         || !deny_valid(od->od_share_deny))
3397                 return nfserr_inval;
3398
3399         nfs4_lock_state();
3400         if ((status = nfs4_preprocess_seqid_op(cstate,
3401                                         od->od_seqid,
3402                                         &od->od_stateid, 
3403                                         OPEN_STATE,
3404                                         &od->od_stateowner, &stp, NULL)))
3405                 goto out; 
3406
3407         status = nfserr_inval;
3408         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3409                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3410                         stp->st_access_bmap, od->od_share_access);
3411                 goto out;
3412         }
3413         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3414                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3415                         stp->st_deny_bmap, od->od_share_deny);
3416                 goto out;
3417         }
3418         set_access(&share_access, stp->st_access_bmap);
3419         nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3420
3421         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3422         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3423
3424         update_stateid(&stp->st_stateid);
3425         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3426         status = nfs_ok;
3427 out:
3428         if (od->od_stateowner) {
3429                 nfs4_get_stateowner(od->od_stateowner);
3430                 cstate->replay_owner = od->od_stateowner;
3431         }
3432         nfs4_unlock_state();
3433         return status;
3434 }
3435
3436 /*
3437  * nfs4_unlock_state() called after encode
3438  */
3439 __be32
3440 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3441             struct nfsd4_close *close)
3442 {
3443         __be32 status;
3444         struct nfs4_stateid *stp;
3445
3446         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3447                         (int)cstate->current_fh.fh_dentry->d_name.len,
3448                         cstate->current_fh.fh_dentry->d_name.name);
3449
3450         nfs4_lock_state();
3451         /* check close_lru for replay */
3452         if ((status = nfs4_preprocess_seqid_op(cstate,
3453                                         close->cl_seqid,
3454                                         &close->cl_stateid, 
3455                                         OPEN_STATE | CLOSE_STATE,
3456                                         &close->cl_stateowner, &stp, NULL)))
3457                 goto out; 
3458         status = nfs_ok;
3459         update_stateid(&stp->st_stateid);
3460         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3461
3462         /* release_stateid() calls nfsd_close() if needed */
3463         release_open_stateid(stp);
3464
3465         /* place unused nfs4_stateowners on so_close_lru list to be
3466          * released by the laundromat service after the lease period
3467          * to enable us to handle CLOSE replay
3468          */
3469         if (list_empty(&close->cl_stateowner->so_stateids))
3470                 move_to_close_lru(close->cl_stateowner);
3471 out:
3472         if (close->cl_stateowner) {
3473                 nfs4_get_stateowner(close->cl_stateowner);
3474                 cstate->replay_owner = close->cl_stateowner;
3475         }
3476         nfs4_unlock_state();
3477         return status;
3478 }
3479
3480 __be32
3481 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3482                   struct nfsd4_delegreturn *dr)
3483 {
3484         struct nfs4_delegation *dp;
3485         stateid_t *stateid = &dr->dr_stateid;
3486         struct inode *inode;
3487         __be32 status;
3488         int flags = 0;
3489
3490         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3491                 return status;
3492         inode = cstate->current_fh.fh_dentry->d_inode;
3493
3494         if (nfsd4_has_session(cstate))
3495                 flags |= HAS_SESSION;
3496         nfs4_lock_state();
3497         status = nfserr_bad_stateid;
3498         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3499                 goto out;
3500         status = nfserr_stale_stateid;
3501         if (STALE_STATEID(stateid))
3502                 goto out;
3503         status = nfserr_bad_stateid;
3504         if (!is_delegation_stateid(stateid))
3505                 goto out;
3506         status = nfserr_expired;
3507         dp = find_delegation_stateid(inode, stateid);
3508         if (!dp)
3509                 goto out;
3510         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3511         if (status)
3512                 goto out;
3513         renew_client(dp->dl_client);
3514
3515         unhash_delegation(dp);
3516 out:
3517         nfs4_unlock_state();
3518
3519         return status;
3520 }
3521
3522
3523 /* 
3524  * Lock owner state (byte-range locks)
3525  */
3526 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3527 #define LOCK_HASH_BITS              8
3528 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3529 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3530
3531 static inline u64
3532 end_offset(u64 start, u64 len)
3533 {
3534         u64 end;
3535
3536         end = start + len;
3537         return end >= start ? end: NFS4_MAX_UINT64;
3538 }
3539
3540 /* last octet in a range */
3541 static inline u64
3542 last_byte_offset(u64 start, u64 len)
3543 {
3544         u64 end;
3545
3546         BUG_ON(!len);
3547         end = start + len;
3548         return end > start ? end - 1: NFS4_MAX_UINT64;
3549 }
3550
3551 #define lockownerid_hashval(id) \
3552         ((id) & LOCK_HASH_MASK)
3553
3554 static inline unsigned int
3555 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3556                 struct xdr_netobj *ownername)
3557 {
3558         return (file_hashval(inode) + cl_id
3559                         + opaque_hashval(ownername->data, ownername->len))
3560                 & LOCK_HASH_MASK;
3561 }
3562
3563 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3564 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3565 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3566
3567 static struct nfs4_stateid *
3568 find_stateid(stateid_t *stid, int flags)
3569 {
3570         struct nfs4_stateid *local;
3571         u32 st_id = stid->si_stateownerid;
3572         u32 f_id = stid->si_fileid;
3573         unsigned int hashval;
3574
3575         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3576         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3577                 hashval = stateid_hashval(st_id, f_id);
3578                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3579                         if ((local->st_stateid.si_stateownerid == st_id) &&
3580                             (local->st_stateid.si_fileid == f_id))
3581                                 return local;
3582                 }
3583         } 
3584
3585         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3586                 hashval = stateid_hashval(st_id, f_id);
3587                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3588                         if ((local->st_stateid.si_stateownerid == st_id) &&
3589                             (local->st_stateid.si_fileid == f_id))
3590                                 return local;
3591                 }
3592         }
3593         return NULL;
3594 }
3595
3596 static struct nfs4_delegation *
3597 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3598 {
3599         struct nfs4_file *fp;
3600         struct nfs4_delegation *dl;
3601
3602         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3603                 STATEID_VAL(stid));
3604
3605         fp = find_file(ino);
3606         if (!fp)
3607                 return NULL;
3608         dl = find_delegation_file(fp, stid);
3609         put_nfs4_file(fp);
3610         return dl;
3611 }
3612
3613 /*
3614  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3615  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3616  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3617  * locking, this prevents us from being completely protocol-compliant.  The
3618  * real solution to this problem is to start using unsigned file offsets in
3619  * the VFS, but this is a very deep change!
3620  */
3621 static inline void
3622 nfs4_transform_lock_offset(struct file_lock *lock)
3623 {
3624         if (lock->fl_start < 0)
3625                 lock->fl_start = OFFSET_MAX;
3626         if (lock->fl_end < 0)
3627                 lock->fl_end = OFFSET_MAX;
3628 }
3629
3630 /* Hack!: For now, we're defining this just so we can use a pointer to it
3631  * as a unique cookie to identify our (NFSv4's) posix locks. */
3632 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3633 };
3634
3635 static inline void
3636 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3637 {
3638         struct nfs4_stateowner *sop;
3639
3640         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3641                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3642                 kref_get(&sop->so_ref);
3643                 deny->ld_sop = sop;
3644                 deny->ld_clientid = sop->so_client->cl_clientid;
3645         } else {
3646                 deny->ld_sop = NULL;
3647                 deny->ld_clientid.cl_boot = 0;
3648                 deny->ld_clientid.cl_id = 0;
3649         }
3650         deny->ld_start = fl->fl_start;
3651         deny->ld_length = NFS4_MAX_UINT64;
3652         if (fl->fl_end != NFS4_MAX_UINT64)
3653                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3654         deny->ld_type = NFS4_READ_LT;
3655         if (fl->fl_type != F_RDLCK)
3656                 deny->ld_type = NFS4_WRITE_LT;
3657 }
3658
3659 static struct nfs4_stateowner *
3660 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3661                 struct xdr_netobj *owner)
3662 {
3663         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3664         struct nfs4_stateowner *op;
3665
3666         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3667                 if (same_owner_str(op, owner, clid))
3668                         return op;
3669         }
3670         return NULL;
3671 }
3672
3673 /*
3674  * Alloc a lock owner structure.
3675  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3676  * occured. 
3677  *
3678  * strhashval = lock_ownerstr_hashval 
3679  */
3680
3681 static struct nfs4_stateowner *
3682 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3683         struct nfs4_stateowner *sop;
3684         struct nfs4_replay *rp;
3685         unsigned int idhashval;
3686
3687         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3688                 return NULL;
3689         idhashval = lockownerid_hashval(current_ownerid);
3690         INIT_LIST_HEAD(&sop->so_idhash);
3691         INIT_LIST_HEAD(&sop->so_strhash);
3692         INIT_LIST_HEAD(&sop->so_perclient);
3693         INIT_LIST_HEAD(&sop->so_stateids);
3694         INIT_LIST_HEAD(&sop->so_perstateid);
3695         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3696         sop->so_time = 0;
3697         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3698         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3699         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3700         sop->so_is_open_owner = 0;
3701         sop->so_id = current_ownerid++;
3702         sop->so_client = clp;
3703         /* It is the openowner seqid that will be incremented in encode in the
3704          * case of new lockowners; so increment the lock seqid manually: */
3705         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3706         sop->so_confirmed = 1;
3707         rp = &sop->so_replay;
3708         rp->rp_status = nfserr_serverfault;
3709         rp->rp_buflen = 0;
3710         rp->rp_buf = rp->rp_ibuf;
3711         return sop;
3712 }
3713
3714 static struct nfs4_stateid *
3715 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3716 {
3717         struct nfs4_stateid *stp;
3718         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3719
3720         stp = nfs4_alloc_stateid();
3721         if (stp == NULL)
3722                 goto out;
3723         INIT_LIST_HEAD(&stp->st_hash);
3724         INIT_LIST_HEAD(&stp->st_perfile);
3725         INIT_LIST_HEAD(&stp->st_perstateowner);
3726         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3727         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3728         list_add(&stp->st_perfile, &fp->fi_stateids);
3729         list_add(&stp->st_perstateowner, &sop->so_stateids);
3730         stp->st_stateowner = sop;
3731         get_nfs4_file(fp);
3732         stp->st_file = fp;
3733         stp->st_stateid.si_boot = boot_time;
3734         stp->st_stateid.si_stateownerid = sop->so_id;
3735         stp->st_stateid.si_fileid = fp->fi_id;
3736         stp->st_stateid.si_generation = 0;
3737         stp->st_deny_bmap = open_stp->st_deny_bmap;
3738         stp->st_openstp = open_stp;
3739
3740 out:
3741         return stp;
3742 }
3743
3744 static int
3745 check_lock_length(u64 offset, u64 length)
3746 {
3747         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3748              LOFF_OVERFLOW(offset, length)));
3749 }
3750
3751 /*
3752  *  LOCK operation 
3753  */
3754 __be32
3755 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3756            struct nfsd4_lock *lock)
3757 {
3758         struct nfs4_stateowner *open_sop = NULL;
3759         struct nfs4_stateowner *lock_sop = NULL;
3760         struct nfs4_stateid *lock_stp;
3761         struct nfs4_file *fp;
3762         struct file *filp = NULL;
3763         struct file_lock file_lock;
3764         struct file_lock conflock;
3765         __be32 status = 0;
3766         unsigned int strhashval;
3767         unsigned int cmd;
3768         int err;
3769
3770         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3771                 (long long) lock->lk_offset,
3772                 (long long) lock->lk_length);
3773
3774         if (check_lock_length(lock->lk_offset, lock->lk_length))
3775                  return nfserr_inval;
3776
3777         if ((status = fh_verify(rqstp, &cstate->current_fh,
3778                                 S_IFREG, NFSD_MAY_LOCK))) {
3779                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3780                 return status;
3781         }
3782
3783         nfs4_lock_state();
3784
3785         if (lock->lk_is_new) {
3786                 /*
3787                  * Client indicates that this is a new lockowner.
3788                  * Use open owner and open stateid to create lock owner and
3789                  * lock stateid.
3790                  */
3791                 struct nfs4_stateid *open_stp = NULL;
3792                 
3793                 status = nfserr_stale_clientid;
3794                 if (!nfsd4_has_session(cstate) &&
3795                     STALE_CLIENTID(&lock->lk_new_clientid))
3796                         goto out;
3797
3798                 /* validate and update open stateid and open seqid */
3799                 status = nfs4_preprocess_seqid_op(cstate,
3800                                         lock->lk_new_open_seqid,
3801                                         &lock->lk_new_open_stateid,
3802                                         OPEN_STATE,
3803                                         &lock->lk_replay_owner, &open_stp,
3804                                         lock);
3805                 if (status)
3806                         goto out;
3807                 open_sop = lock->lk_replay_owner;
3808                 /* create lockowner and lock stateid */
3809                 fp = open_stp->st_file;
3810                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3811                                 open_sop->so_client->cl_clientid.cl_id, 
3812                                 &lock->v.new.owner);
3813                 /* XXX: Do we need to check for duplicate stateowners on
3814                  * the same file, or should they just be allowed (and
3815                  * create new stateids)? */
3816                 status = nfserr_resource;
3817                 lock_sop = alloc_init_lock_stateowner(strhashval,
3818                                 open_sop->so_client, open_stp, lock);
3819                 if (lock_sop == NULL)
3820                         goto out;
3821                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3822                 if (lock_stp == NULL)
3823                         goto out;
3824         } else {
3825                 /* lock (lock owner + lock stateid) already exists */
3826                 status = nfs4_preprocess_seqid_op(cstate,
3827                                        lock->lk_old_lock_seqid, 
3828                                        &lock->lk_old_lock_stateid, 
3829                                        LOCK_STATE,
3830                                        &lock->lk_replay_owner, &lock_stp, lock);
3831                 if (status)
3832                         goto out;
3833                 lock_sop = lock->lk_replay_owner;
3834                 fp = lock_stp->st_file;
3835         }
3836         /* lock->lk_replay_owner and lock_stp have been created or found */
3837
3838         status = nfserr_grace;
3839         if (locks_in_grace() && !lock->lk_reclaim)
3840                 goto out;
3841         status = nfserr_no_grace;
3842         if (!locks_in_grace() && lock->lk_reclaim)
3843                 goto out;
3844
3845         locks_init_lock(&file_lock);
3846         switch (lock->lk_type) {
3847                 case NFS4_READ_LT:
3848                 case NFS4_READW_LT:
3849                         if (find_readable_file(lock_stp->st_file)) {
3850                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_READ);
3851                                 filp = find_readable_file(lock_stp->st_file);
3852                         }
3853                         file_lock.fl_type = F_RDLCK;
3854                         cmd = F_SETLK;
3855                 break;
3856                 case NFS4_WRITE_LT:
3857                 case NFS4_WRITEW_LT:
3858                         if (find_writeable_file(lock_stp->st_file)) {
3859                                 nfs4_get_vfs_file(rqstp, fp, &cstate->current_fh, NFS4_SHARE_ACCESS_WRITE);
3860                                 filp = find_writeable_file(lock_stp->st_file);
3861                         }
3862                         file_lock.fl_type = F_WRLCK;
3863                         cmd = F_SETLK;
3864                 break;
3865                 default:
3866                         status = nfserr_inval;
3867                 goto out;
3868         }
3869         if (!filp) {
3870                 status = nfserr_openmode;
3871                 goto out;
3872         }
3873         file_lock.fl_owner = (fl_owner_t)lock_sop;
3874         file_lock.fl_pid = current->tgid;
3875         file_lock.fl_file = filp;
3876         file_lock.fl_flags = FL_POSIX;
3877         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3878
3879         file_lock.fl_start = lock->lk_offset;
3880         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3881         nfs4_transform_lock_offset(&file_lock);
3882
3883         /*
3884         * Try to lock the file in the VFS.
3885         * Note: locks.c uses the BKL to protect the inode's lock list.
3886         */
3887
3888         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3889         switch (-err) {
3890         case 0: /* success! */
3891                 update_stateid(&lock_stp->st_stateid);
3892                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3893                                 sizeof(stateid_t));
3894                 status = 0;
3895                 break;
3896         case (EAGAIN):          /* conflock holds conflicting lock */
3897                 status = nfserr_denied;
3898                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3899                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3900                 break;
3901         case (EDEADLK):
3902                 status = nfserr_deadlock;
3903                 break;
3904         default:        
3905                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3906                 status = nfserr_resource;
3907                 break;
3908         }
3909 out:
3910         if (status && lock->lk_is_new && lock_sop)
3911                 release_lockowner(lock_sop);
3912         if (lock->lk_replay_owner) {
3913                 nfs4_get_stateowner(lock->lk_replay_owner);
3914                 cstate->replay_owner = lock->lk_replay_owner;
3915         }
3916         nfs4_unlock_state();
3917         return status;
3918 }
3919
3920 /*
3921  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3922  * so we do a temporary open here just to get an open file to pass to
3923  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3924  * inode operation.)
3925  */
3926 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3927 {
3928         struct file *file;
3929         int err;
3930
3931         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3932         if (err)
3933                 return err;
3934         err = vfs_test_lock(file, lock);
3935         nfsd_close(file);
3936         return err;
3937 }
3938
3939 /*
3940  * LOCKT operation
3941  */
3942 __be32
3943 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3944             struct nfsd4_lockt *lockt)
3945 {
3946         struct inode *inode;
3947         struct file_lock file_lock;
3948         int error;
3949         __be32 status;
3950
3951         if (locks_in_grace())
3952                 return nfserr_grace;
3953
3954         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3955                  return nfserr_inval;
3956
3957         lockt->lt_stateowner = NULL;
3958         nfs4_lock_state();
3959
3960         status = nfserr_stale_clientid;
3961         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3962                 goto out;
3963
3964         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3965                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3966                 if (status == nfserr_symlink)
3967                         status = nfserr_inval;
3968                 goto out;
3969         }
3970
3971         inode = cstate->current_fh.fh_dentry->d_inode;
3972         locks_init_lock(&file_lock);
3973         switch (lockt->lt_type) {
3974                 case NFS4_READ_LT:
3975                 case NFS4_READW_LT:
3976                         file_lock.fl_type = F_RDLCK;
3977                 break;
3978                 case NFS4_WRITE_LT:
3979                 case NFS4_WRITEW_LT:
3980                         file_lock.fl_type = F_WRLCK;
3981                 break;
3982                 default:
3983                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3984                         status = nfserr_inval;
3985                 goto out;
3986         }
3987
3988         lockt->lt_stateowner = find_lockstateowner_str(inode,
3989                         &lockt->lt_clientid, &lockt->lt_owner);
3990         if (lockt->lt_stateowner)
3991                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3992         file_lock.fl_pid = current->tgid;
3993         file_lock.fl_flags = FL_POSIX;
3994
3995         file_lock.fl_start = lockt->lt_offset;
3996         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3997
3998         nfs4_transform_lock_offset(&file_lock);
3999
4000         status = nfs_ok;
4001         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
4002         if (error) {
4003                 status = nfserrno(error);
4004                 goto out;
4005         }
4006         if (file_lock.fl_type != F_UNLCK) {
4007                 status = nfserr_denied;
4008                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
4009         }
4010 out:
4011         nfs4_unlock_state();
4012         return status;
4013 }
4014
4015 __be32
4016 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4017             struct nfsd4_locku *locku)
4018 {
4019         struct nfs4_stateid *stp;
4020         struct file *filp = NULL;
4021         struct file_lock file_lock;
4022         __be32 status;
4023         int err;
4024                                                         
4025         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4026                 (long long) locku->lu_offset,
4027                 (long long) locku->lu_length);
4028
4029         if (check_lock_length(locku->lu_offset, locku->lu_length))
4030                  return nfserr_inval;
4031
4032         nfs4_lock_state();
4033                                                                                 
4034         if ((status = nfs4_preprocess_seqid_op(cstate,
4035                                         locku->lu_seqid, 
4036                                         &locku->lu_stateid, 
4037                                         LOCK_STATE,
4038                                         &locku->lu_stateowner, &stp, NULL)))
4039                 goto out;
4040
4041         filp = find_any_file(stp->st_file);
4042         if (!filp) {
4043                 status = nfserr_lock_range;
4044                 goto out;
4045         }
4046         BUG_ON(!filp);
4047         locks_init_lock(&file_lock);
4048         file_lock.fl_type = F_UNLCK;
4049         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
4050         file_lock.fl_pid = current->tgid;
4051         file_lock.fl_file = filp;
4052         file_lock.fl_flags = FL_POSIX; 
4053         file_lock.fl_lmops = &nfsd_posix_mng_ops;
4054         file_lock.fl_start = locku->lu_offset;
4055
4056         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
4057         nfs4_transform_lock_offset(&file_lock);
4058
4059         /*
4060         *  Try to unlock the file in the VFS.
4061         */
4062         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
4063         if (err) {
4064                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4065                 goto out_nfserr;
4066         }
4067         /*
4068         * OK, unlock succeeded; the only thing left to do is update the stateid.
4069         */
4070         update_stateid(&stp->st_stateid);
4071         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
4072
4073 out:
4074         if (locku->lu_stateowner) {
4075                 nfs4_get_stateowner(locku->lu_stateowner);
4076                 cstate->replay_owner = locku->lu_stateowner;
4077         }
4078         nfs4_unlock_state();
4079         return status;
4080
4081 out_nfserr:
4082         status = nfserrno(err);
4083         goto out;
4084 }
4085
4086 /*
4087  * returns
4088  *      1: locks held by lockowner
4089  *      0: no locks held by lockowner
4090  */
4091 static int
4092 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
4093 {
4094         struct file_lock **flpp;
4095         struct inode *inode = filp->fi_inode;
4096         int status = 0;
4097
4098         lock_flocks();
4099         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4100                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4101                         status = 1;
4102                         goto out;
4103                 }
4104         }
4105 out:
4106         unlock_flocks();
4107         return status;
4108 }
4109
4110 __be32
4111 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4112                         struct nfsd4_compound_state *cstate,
4113                         struct nfsd4_release_lockowner *rlockowner)
4114 {
4115         clientid_t *clid = &rlockowner->rl_clientid;
4116         struct nfs4_stateowner *sop;
4117         struct nfs4_stateid *stp;
4118         struct xdr_netobj *owner = &rlockowner->rl_owner;
4119         struct list_head matches;
4120         int i;
4121         __be32 status;
4122
4123         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4124                 clid->cl_boot, clid->cl_id);
4125
4126         /* XXX check for lease expiration */
4127
4128         status = nfserr_stale_clientid;
4129         if (STALE_CLIENTID(clid))
4130                 return status;
4131
4132         nfs4_lock_state();
4133
4134         status = nfserr_locks_held;
4135         /* XXX: we're doing a linear search through all the lockowners.
4136          * Yipes!  For now we'll just hope clients aren't really using
4137          * release_lockowner much, but eventually we have to fix these
4138          * data structures. */
4139         INIT_LIST_HEAD(&matches);
4140         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4141                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
4142                         if (!same_owner_str(sop, owner, clid))
4143                                 continue;
4144                         list_for_each_entry(stp, &sop->so_stateids,
4145                                         st_perstateowner) {
4146                                 if (check_for_locks(stp->st_file, sop))
4147                                         goto out;
4148                                 /* Note: so_perclient unused for lockowners,
4149                                  * so it's OK to fool with here. */
4150                                 list_add(&sop->so_perclient, &matches);
4151                         }
4152                 }
4153         }
4154         /* Clients probably won't expect us to return with some (but not all)
4155          * of the lockowner state released; so don't release any until all
4156          * have been checked. */
4157         status = nfs_ok;
4158         while (!list_empty(&matches)) {
4159                 sop = list_entry(matches.next, struct nfs4_stateowner,
4160                                                                 so_perclient);
4161                 /* unhash_stateowner deletes so_perclient only
4162                  * for openowners. */
4163                 list_del(&sop->so_perclient);
4164                 release_lockowner(sop);
4165         }
4166 out:
4167         nfs4_unlock_state();
4168         return status;
4169 }
4170
4171 static inline struct nfs4_client_reclaim *
4172 alloc_reclaim(void)
4173 {
4174         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4175 }
4176
4177 int
4178 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
4179 {
4180         unsigned int strhashval = clientstr_hashval(name);
4181         struct nfs4_client *clp;
4182
4183         clp = find_confirmed_client_by_str(name, strhashval);
4184         return clp ? 1 : 0;
4185 }
4186
4187 /*
4188  * failure => all reset bets are off, nfserr_no_grace...
4189  */
4190 int
4191 nfs4_client_to_reclaim(const char *name)
4192 {
4193         unsigned int strhashval;
4194         struct nfs4_client_reclaim *crp = NULL;
4195
4196         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4197         crp = alloc_reclaim();
4198         if (!crp)
4199                 return 0;
4200         strhashval = clientstr_hashval(name);
4201         INIT_LIST_HEAD(&crp->cr_strhash);
4202         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4203         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4204         reclaim_str_hashtbl_size++;
4205         return 1;
4206 }
4207
4208 static void
4209 nfs4_release_reclaim(void)
4210 {
4211         struct nfs4_client_reclaim *crp = NULL;
4212         int i;
4213
4214         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4215                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4216                         crp = list_entry(reclaim_str_hashtbl[i].next,
4217                                         struct nfs4_client_reclaim, cr_strhash);
4218                         list_del(&crp->cr_strhash);
4219                         kfree(crp);
4220                         reclaim_str_hashtbl_size--;
4221                 }
4222         }
4223         BUG_ON(reclaim_str_hashtbl_size);
4224 }
4225
4226 /*
4227  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4228 static struct nfs4_client_reclaim *
4229 nfs4_find_reclaim_client(clientid_t *clid)
4230 {
4231         unsigned int strhashval;
4232         struct nfs4_client *clp;
4233         struct nfs4_client_reclaim *crp = NULL;
4234
4235
4236         /* find clientid in conf_id_hashtbl */
4237         clp = find_confirmed_client(clid);
4238         if (clp == NULL)
4239                 return NULL;
4240
4241         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4242                             clp->cl_name.len, clp->cl_name.data,
4243                             clp->cl_recdir);
4244
4245         /* find clp->cl_name in reclaim_str_hashtbl */
4246         strhashval = clientstr_hashval(clp->cl_recdir);
4247         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4248                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4249                         return crp;
4250                 }
4251         }
4252         return NULL;
4253 }
4254
4255 /*
4256 * Called from OPEN. Look for clientid in reclaim list.
4257 */
4258 __be32
4259 nfs4_check_open_reclaim(clientid_t *clid)
4260 {
4261         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4262 }
4263
4264 /* initialization to perform at module load time: */
4265
4266 int
4267 nfs4_state_init(void)
4268 {
4269         int i, status;
4270
4271         status = nfsd4_init_slabs();
4272         if (status)
4273                 return status;
4274         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4275                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4276                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4277                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4278                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4279                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4280         }
4281         for (i = 0; i < SESSION_HASH_SIZE; i++)
4282                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4283         for (i = 0; i < FILE_HASH_SIZE; i++) {
4284                 INIT_LIST_HEAD(&file_hashtbl[i]);
4285         }
4286         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4287                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4288                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4289         }
4290         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4291                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4292                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4293         }
4294         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4295                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4296                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4297         }
4298         memset(&onestateid, ~0, sizeof(stateid_t));
4299         INIT_LIST_HEAD(&close_lru);
4300         INIT_LIST_HEAD(&client_lru);
4301         INIT_LIST_HEAD(&del_recall_lru);
4302         reclaim_str_hashtbl_size = 0;
4303         return 0;
4304 }
4305
4306 static void
4307 nfsd4_load_reboot_recovery_data(void)
4308 {
4309         int status;
4310
4311         nfs4_lock_state();
4312         nfsd4_init_recdir(user_recovery_dirname);
4313         status = nfsd4_recdir_load();
4314         nfs4_unlock_state();
4315         if (status)
4316                 printk("NFSD: Failure reading reboot recovery data\n");
4317 }
4318
4319 /*
4320  * Since the lifetime of a delegation isn't limited to that of an open, a
4321  * client may quite reasonably hang on to a delegation as long as it has
4322  * the inode cached.  This becomes an obvious problem the first time a
4323  * client's inode cache approaches the size of the server's total memory.
4324  *
4325  * For now we avoid this problem by imposing a hard limit on the number
4326  * of delegations, which varies according to the server's memory size.
4327  */
4328 static void
4329 set_max_delegations(void)
4330 {
4331         /*
4332          * Allow at most 4 delegations per megabyte of RAM.  Quick
4333          * estimates suggest that in the worst case (where every delegation
4334          * is for a different inode), a delegation could take about 1.5K,
4335          * giving a worst case usage of about 6% of memory.
4336          */
4337         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4338 }
4339
4340 /* initialization to perform when the nfsd service is started: */
4341
4342 static int
4343 __nfs4_state_start(void)
4344 {
4345         int ret;
4346
4347         boot_time = get_seconds();
4348         locks_start_grace(&nfsd4_manager);
4349         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4350                nfsd4_grace);
4351         ret = set_callback_cred();
4352         if (ret)
4353                 return -ENOMEM;
4354         laundry_wq = create_singlethread_workqueue("nfsd4");
4355         if (laundry_wq == NULL)
4356                 return -ENOMEM;
4357         ret = nfsd4_create_callback_queue();
4358         if (ret)
4359                 goto out_free_laundry;
4360         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4361         set_max_delegations();
4362         return 0;
4363 out_free_laundry:
4364         destroy_workqueue(laundry_wq);
4365         return ret;
4366 }
4367
4368 int
4369 nfs4_state_start(void)
4370 {
4371         nfsd4_load_reboot_recovery_data();
4372         return __nfs4_state_start();
4373 }
4374
4375 static void
4376 __nfs4_state_shutdown(void)
4377 {
4378         int i;
4379         struct nfs4_client *clp = NULL;
4380         struct nfs4_delegation *dp = NULL;
4381         struct list_head *pos, *next, reaplist;
4382
4383         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4384                 while (!list_empty(&conf_id_hashtbl[i])) {
4385                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4386                         expire_client(clp);
4387                 }
4388                 while (!list_empty(&unconf_str_hashtbl[i])) {
4389                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4390                         expire_client(clp);
4391                 }
4392         }
4393         INIT_LIST_HEAD(&reaplist);
4394         spin_lock(&recall_lock);
4395         list_for_each_safe(pos, next, &del_recall_lru) {
4396                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4397                 list_move(&dp->dl_recall_lru, &reaplist);
4398         }
4399         spin_unlock(&recall_lock);
4400         list_for_each_safe(pos, next, &reaplist) {
4401                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4402                 list_del_init(&dp->dl_recall_lru);
4403                 unhash_delegation(dp);
4404         }
4405
4406         nfsd4_shutdown_recdir();
4407 }
4408
4409 void
4410 nfs4_state_shutdown(void)
4411 {
4412         cancel_delayed_work_sync(&laundromat_work);
4413         destroy_workqueue(laundry_wq);
4414         locks_end_grace(&nfsd4_manager);
4415         nfs4_lock_state();
4416         nfs4_release_reclaim();
4417         __nfs4_state_shutdown();
4418         nfs4_unlock_state();
4419         nfsd4_destroy_callback_queue();
4420 }
4421
4422 /*
4423  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4424  * accessed when nfsd is starting.
4425  */
4426 static void
4427 nfs4_set_recdir(char *recdir)
4428 {
4429         strcpy(user_recovery_dirname, recdir);
4430 }
4431
4432 /*
4433  * Change the NFSv4 recovery directory to recdir.
4434  */
4435 int
4436 nfs4_reset_recoverydir(char *recdir)
4437 {
4438         int status;
4439         struct path path;
4440
4441         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4442         if (status)
4443                 return status;
4444         status = -ENOTDIR;
4445         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4446                 nfs4_set_recdir(recdir);
4447                 status = 0;
4448         }
4449         path_put(&path);
4450         return status;
4451 }
4452
4453 char *
4454 nfs4_recoverydir(void)
4455 {
4456         return user_recovery_dirname;
4457 }