46e9ac5268720cde89232139e5604703b4fc5ad0
[safe/jmp/linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  linux/fs/nfsd/nfs4state.c
3 *
4 *  Copyright (c) 2001 The Regents of the University of Michigan.
5 *  All rights reserved.
6 *
7 *  Kendrick Smith <kmsmith@umich.edu>
8 *  Andy Adamson <kandros@umich.edu>
9 *
10 *  Redistribution and use in source and binary forms, with or without
11 *  modification, are permitted provided that the following conditions
12 *  are met:
13 *
14 *  1. Redistributions of source code must retain the above copyright
15 *     notice, this list of conditions and the following disclaimer.
16 *  2. Redistributions in binary form must reproduce the above copyright
17 *     notice, this list of conditions and the following disclaimer in the
18 *     documentation and/or other materials provided with the distribution.
19 *  3. Neither the name of the University nor the names of its
20 *     contributors may be used to endorse or promote products derived
21 *     from this software without specific prior written permission.
22 *
23 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
24 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
25 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
26 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
31 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
32 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 *
35 */
36
37 #include <linux/param.h>
38 #include <linux/major.h>
39 #include <linux/slab.h>
40
41 #include <linux/sunrpc/svc.h>
42 #include <linux/nfsd/nfsd.h>
43 #include <linux/nfsd/cache.h>
44 #include <linux/file.h>
45 #include <linux/mount.h>
46 #include <linux/workqueue.h>
47 #include <linux/smp_lock.h>
48 #include <linux/kthread.h>
49 #include <linux/nfs4.h>
50 #include <linux/nfsd/state.h>
51 #include <linux/nfsd/xdr4.h>
52 #include <linux/namei.h>
53 #include <linux/swap.h>
54 #include <linux/mutex.h>
55 #include <linux/lockd/bind.h>
56 #include <linux/module.h>
57 #include <linux/sunrpc/svcauth_gss.h>
58 #include <linux/sunrpc/clnt.h>
59
60 #define NFSDDBG_FACILITY                NFSDDBG_PROC
61
62 /* Globals */
63 static time_t lease_time = 90;     /* default lease time */
64 static time_t user_lease_time = 90;
65 static time_t boot_time;
66 static u32 current_ownerid = 1;
67 static u32 current_fileid = 1;
68 static u32 current_delegid = 1;
69 static u32 nfs4_init;
70 static stateid_t zerostateid;             /* bits all 0 */
71 static stateid_t onestateid;              /* bits all 1 */
72 static u64 current_sessionid = 1;
73
74 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
75 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
76
77 /* forward declarations */
78 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
79 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
80 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
81 static void nfs4_set_recdir(char *recdir);
82
83 /* Locking: */
84
85 /* Currently used for almost all code touching nfsv4 state: */
86 static DEFINE_MUTEX(client_mutex);
87
88 /*
89  * Currently used for the del_recall_lru and file hash table.  In an
90  * effort to decrease the scope of the client_mutex, this spinlock may
91  * eventually cover more:
92  */
93 static DEFINE_SPINLOCK(recall_lock);
94
95 static struct kmem_cache *stateowner_slab = NULL;
96 static struct kmem_cache *file_slab = NULL;
97 static struct kmem_cache *stateid_slab = NULL;
98 static struct kmem_cache *deleg_slab = NULL;
99
100 void
101 nfs4_lock_state(void)
102 {
103         mutex_lock(&client_mutex);
104 }
105
106 void
107 nfs4_unlock_state(void)
108 {
109         mutex_unlock(&client_mutex);
110 }
111
112 static inline u32
113 opaque_hashval(const void *ptr, int nbytes)
114 {
115         unsigned char *cptr = (unsigned char *) ptr;
116
117         u32 x = 0;
118         while (nbytes--) {
119                 x *= 37;
120                 x += *cptr++;
121         }
122         return x;
123 }
124
125 static struct list_head del_recall_lru;
126
127 static inline void
128 put_nfs4_file(struct nfs4_file *fi)
129 {
130         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
131                 list_del(&fi->fi_hash);
132                 spin_unlock(&recall_lock);
133                 iput(fi->fi_inode);
134                 kmem_cache_free(file_slab, fi);
135         }
136 }
137
138 static inline void
139 get_nfs4_file(struct nfs4_file *fi)
140 {
141         atomic_inc(&fi->fi_ref);
142 }
143
144 static int num_delegations;
145 unsigned int max_delegations;
146
147 /*
148  * Open owner state (share locks)
149  */
150
151 /* hash tables for nfs4_stateowner */
152 #define OWNER_HASH_BITS              8
153 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
154 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
155
156 #define ownerid_hashval(id) \
157         ((id) & OWNER_HASH_MASK)
158 #define ownerstr_hashval(clientid, ownername) \
159         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
160
161 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
162 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
163
164 /* hash table for nfs4_file */
165 #define FILE_HASH_BITS                   8
166 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
167 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
168 /* hash table for (open)nfs4_stateid */
169 #define STATEID_HASH_BITS              10
170 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
171 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
172
173 #define file_hashval(x) \
174         hash_ptr(x, FILE_HASH_BITS)
175 #define stateid_hashval(owner_id, file_id)  \
176         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
177
178 static struct list_head file_hashtbl[FILE_HASH_SIZE];
179 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
180
181 static struct nfs4_delegation *
182 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
183 {
184         struct nfs4_delegation *dp;
185         struct nfs4_file *fp = stp->st_file;
186         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
187
188         dprintk("NFSD alloc_init_deleg\n");
189         if (fp->fi_had_conflict)
190                 return NULL;
191         if (num_delegations > max_delegations)
192                 return NULL;
193         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
194         if (dp == NULL)
195                 return dp;
196         num_delegations++;
197         INIT_LIST_HEAD(&dp->dl_perfile);
198         INIT_LIST_HEAD(&dp->dl_perclnt);
199         INIT_LIST_HEAD(&dp->dl_recall_lru);
200         dp->dl_client = clp;
201         get_nfs4_file(fp);
202         dp->dl_file = fp;
203         dp->dl_flock = NULL;
204         get_file(stp->st_vfs_file);
205         dp->dl_vfs_file = stp->st_vfs_file;
206         dp->dl_type = type;
207         dp->dl_ident = cb->cb_ident;
208         dp->dl_stateid.si_boot = get_seconds();
209         dp->dl_stateid.si_stateownerid = current_delegid++;
210         dp->dl_stateid.si_fileid = 0;
211         dp->dl_stateid.si_generation = 0;
212         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
213         dp->dl_time = 0;
214         atomic_set(&dp->dl_count, 1);
215         list_add(&dp->dl_perfile, &fp->fi_delegations);
216         list_add(&dp->dl_perclnt, &clp->cl_delegations);
217         return dp;
218 }
219
220 void
221 nfs4_put_delegation(struct nfs4_delegation *dp)
222 {
223         if (atomic_dec_and_test(&dp->dl_count)) {
224                 dprintk("NFSD: freeing dp %p\n",dp);
225                 put_nfs4_file(dp->dl_file);
226                 kmem_cache_free(deleg_slab, dp);
227                 num_delegations--;
228         }
229 }
230
231 /* Remove the associated file_lock first, then remove the delegation.
232  * lease_modify() is called to remove the FS_LEASE file_lock from
233  * the i_flock list, eventually calling nfsd's lock_manager
234  * fl_release_callback.
235  */
236 static void
237 nfs4_close_delegation(struct nfs4_delegation *dp)
238 {
239         struct file *filp = dp->dl_vfs_file;
240
241         dprintk("NFSD: close_delegation dp %p\n",dp);
242         dp->dl_vfs_file = NULL;
243         /* The following nfsd_close may not actually close the file,
244          * but we want to remove the lease in any case. */
245         if (dp->dl_flock)
246                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
247         nfsd_close(filp);
248 }
249
250 /* Called under the state lock. */
251 static void
252 unhash_delegation(struct nfs4_delegation *dp)
253 {
254         list_del_init(&dp->dl_perfile);
255         list_del_init(&dp->dl_perclnt);
256         spin_lock(&recall_lock);
257         list_del_init(&dp->dl_recall_lru);
258         spin_unlock(&recall_lock);
259         nfs4_close_delegation(dp);
260         nfs4_put_delegation(dp);
261 }
262
263 /* 
264  * SETCLIENTID state 
265  */
266
267 /* Hash tables for nfs4_clientid state */
268 #define CLIENT_HASH_BITS                 4
269 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
270 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
271
272 #define clientid_hashval(id) \
273         ((id) & CLIENT_HASH_MASK)
274 #define clientstr_hashval(name) \
275         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
276 /*
277  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
278  * used in reboot/reset lease grace period processing
279  *
280  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
281  * setclientid_confirmed info. 
282  *
283  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
284  * setclientid info.
285  *
286  * client_lru holds client queue ordered by nfs4_client.cl_time
287  * for lease renewal.
288  *
289  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
290  * for last close replay.
291  */
292 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
293 static int reclaim_str_hashtbl_size = 0;
294 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
295 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
296 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
297 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
298 static struct list_head client_lru;
299 static struct list_head close_lru;
300
301 static void unhash_generic_stateid(struct nfs4_stateid *stp)
302 {
303         list_del(&stp->st_hash);
304         list_del(&stp->st_perfile);
305         list_del(&stp->st_perstateowner);
306 }
307
308 static void free_generic_stateid(struct nfs4_stateid *stp)
309 {
310         put_nfs4_file(stp->st_file);
311         kmem_cache_free(stateid_slab, stp);
312 }
313
314 static void release_lock_stateid(struct nfs4_stateid *stp)
315 {
316         unhash_generic_stateid(stp);
317         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
318         free_generic_stateid(stp);
319 }
320
321 static void unhash_lockowner(struct nfs4_stateowner *sop)
322 {
323         struct nfs4_stateid *stp;
324
325         list_del(&sop->so_idhash);
326         list_del(&sop->so_strhash);
327         list_del(&sop->so_perstateid);
328         while (!list_empty(&sop->so_stateids)) {
329                 stp = list_first_entry(&sop->so_stateids,
330                                 struct nfs4_stateid, st_perstateowner);
331                 release_lock_stateid(stp);
332         }
333 }
334
335 static void release_lockowner(struct nfs4_stateowner *sop)
336 {
337         unhash_lockowner(sop);
338         nfs4_put_stateowner(sop);
339 }
340
341 static void
342 release_stateid_lockowners(struct nfs4_stateid *open_stp)
343 {
344         struct nfs4_stateowner *lock_sop;
345
346         while (!list_empty(&open_stp->st_lockowners)) {
347                 lock_sop = list_entry(open_stp->st_lockowners.next,
348                                 struct nfs4_stateowner, so_perstateid);
349                 /* list_del(&open_stp->st_lockowners);  */
350                 BUG_ON(lock_sop->so_is_open_owner);
351                 release_lockowner(lock_sop);
352         }
353 }
354
355 static void release_open_stateid(struct nfs4_stateid *stp)
356 {
357         unhash_generic_stateid(stp);
358         release_stateid_lockowners(stp);
359         nfsd_close(stp->st_vfs_file);
360         free_generic_stateid(stp);
361 }
362
363 static void unhash_openowner(struct nfs4_stateowner *sop)
364 {
365         struct nfs4_stateid *stp;
366
367         list_del(&sop->so_idhash);
368         list_del(&sop->so_strhash);
369         list_del(&sop->so_perclient);
370         list_del(&sop->so_perstateid); /* XXX: necessary? */
371         while (!list_empty(&sop->so_stateids)) {
372                 stp = list_first_entry(&sop->so_stateids,
373                                 struct nfs4_stateid, st_perstateowner);
374                 release_open_stateid(stp);
375         }
376 }
377
378 static void release_openowner(struct nfs4_stateowner *sop)
379 {
380         unhash_openowner(sop);
381         list_del(&sop->so_close_lru);
382         nfs4_put_stateowner(sop);
383 }
384
385 static DEFINE_SPINLOCK(sessionid_lock);
386 #define SESSION_HASH_SIZE       512
387 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
388
389 static inline int
390 hash_sessionid(struct nfs4_sessionid *sessionid)
391 {
392         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
393
394         return sid->sequence % SESSION_HASH_SIZE;
395 }
396
397 static inline void
398 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
399 {
400         u32 *ptr = (u32 *)(&sessionid->data[0]);
401         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
402 }
403
404 static void
405 gen_sessionid(struct nfsd4_session *ses)
406 {
407         struct nfs4_client *clp = ses->se_client;
408         struct nfsd4_sessionid *sid;
409
410         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
411         sid->clientid = clp->cl_clientid;
412         sid->sequence = current_sessionid++;
413         sid->reserved = 0;
414 }
415
416 /*
417  * The protocol defines ca_maxresponssize_cached to include the size of
418  * the rpc header, but all we need to cache is the data starting after
419  * the end of the initial SEQUENCE operation--the rest we regenerate
420  * each time.  Therefore we can advertise a ca_maxresponssize_cached
421  * value that is the number of bytes in our cache plus a few additional
422  * bytes.  In order to stay on the safe side, and not promise more than
423  * we can cache, those additional bytes must be the minimum possible: 24
424  * bytes of rpc header (xid through accept state, with AUTH_NULL
425  * verifier), 12 for the compound header (with zero-length tag), and 44
426  * for the SEQUENCE op response:
427  */
428 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
429
430 /*
431  * Give the client the number of ca_maxresponsesize_cached slots it
432  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
433  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
434  * than NFSD_MAX_SLOTS_PER_SESSION.
435  *
436  * If we run out of reserved DRC memory we should (up to a point)
437  * re-negotiate active sessions and reduce their slot usage to make
438  * rooom for new connections. For now we just fail the create session.
439  */
440 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
441 {
442         int mem, size = fchan->maxresp_cached;
443
444         if (fchan->maxreqs < 1)
445                 return nfserr_inval;
446
447         if (size < NFSD_MIN_HDR_SEQ_SZ)
448                 size = NFSD_MIN_HDR_SEQ_SZ;
449         size -= NFSD_MIN_HDR_SEQ_SZ;
450         if (size > NFSD_SLOT_CACHE_SIZE)
451                 size = NFSD_SLOT_CACHE_SIZE;
452
453         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
454         mem = fchan->maxreqs * size;
455         if (mem > NFSD_MAX_MEM_PER_SESSION) {
456                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
457                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
458                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
459                 mem = fchan->maxreqs * size;
460         }
461
462         spin_lock(&nfsd_drc_lock);
463         /* bound the total session drc memory ussage */
464         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
465                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
466                 mem = fchan->maxreqs * size;
467         }
468         nfsd_drc_mem_used += mem;
469         spin_unlock(&nfsd_drc_lock);
470
471         if (fchan->maxreqs == 0)
472                 return nfserr_serverfault;
473
474         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
475         return 0;
476 }
477
478 /*
479  * fchan holds the client values on input, and the server values on output
480  */
481 static int init_forechannel_attrs(struct svc_rqst *rqstp,
482                                   struct nfsd4_channel_attrs *session_fchan,
483                                   struct nfsd4_channel_attrs *fchan)
484 {
485         int status = 0;
486         __u32   maxcount = svc_max_payload(rqstp);
487
488         /* headerpadsz set to zero in encode routine */
489
490         /* Use the client's max request and max response size if possible */
491         if (fchan->maxreq_sz > maxcount)
492                 fchan->maxreq_sz = maxcount;
493         session_fchan->maxreq_sz = fchan->maxreq_sz;
494
495         if (fchan->maxresp_sz > maxcount)
496                 fchan->maxresp_sz = maxcount;
497         session_fchan->maxresp_sz = fchan->maxresp_sz;
498
499         /* Use the client's maxops if possible */
500         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
501                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
502         session_fchan->maxops = fchan->maxops;
503
504         /* FIXME: Error means no more DRC pages so the server should
505          * recover pages from existing sessions. For now fail session
506          * creation.
507          */
508         status = set_forechannel_drc_size(fchan);
509
510         session_fchan->maxresp_cached = fchan->maxresp_cached;
511         session_fchan->maxreqs = fchan->maxreqs;
512
513         dprintk("%s status %d\n", __func__, status);
514         return status;
515 }
516
517 static void
518 free_session_slots(struct nfsd4_session *ses)
519 {
520         int i;
521
522         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
523                 kfree(ses->se_slots[i]);
524 }
525
526 static int
527 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
528                    struct nfsd4_create_session *cses)
529 {
530         struct nfsd4_session *new, tmp;
531         struct nfsd4_slot *sp;
532         int idx, slotsize, cachesize, i;
533         int status;
534
535         memset(&tmp, 0, sizeof(tmp));
536
537         /* FIXME: For now, we just accept the client back channel attributes. */
538         tmp.se_bchannel = cses->back_channel;
539         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
540                                         &cses->fore_channel);
541         if (status)
542                 goto out;
543
544         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
545                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
546
547         status = nfserr_serverfault;
548         /* allocate struct nfsd4_session and slot table pointers in one piece */
549         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
550         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
551         if (!new)
552                 goto out;
553
554         memcpy(new, &tmp, sizeof(*new));
555
556         /* allocate each struct nfsd4_slot and data cache in one piece */
557         cachesize = new->se_fchannel.maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
558         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
559                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
560                 if (!sp)
561                         goto out_free;
562                 new->se_slots[i] = sp;
563         }
564
565         new->se_client = clp;
566         gen_sessionid(new);
567         idx = hash_sessionid(&new->se_sessionid);
568         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
569                NFS4_MAX_SESSIONID_LEN);
570
571         new->se_flags = cses->flags;
572         kref_init(&new->se_ref);
573         spin_lock(&sessionid_lock);
574         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
575         list_add(&new->se_perclnt, &clp->cl_sessions);
576         spin_unlock(&sessionid_lock);
577
578         status = nfs_ok;
579 out:
580         return status;
581 out_free:
582         free_session_slots(new);
583         kfree(new);
584         goto out;
585 }
586
587 /* caller must hold sessionid_lock */
588 static struct nfsd4_session *
589 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
590 {
591         struct nfsd4_session *elem;
592         int idx;
593
594         dump_sessionid(__func__, sessionid);
595         idx = hash_sessionid(sessionid);
596         dprintk("%s: idx is %d\n", __func__, idx);
597         /* Search in the appropriate list */
598         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
599                 dump_sessionid("list traversal", &elem->se_sessionid);
600                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
601                             NFS4_MAX_SESSIONID_LEN)) {
602                         return elem;
603                 }
604         }
605
606         dprintk("%s: session not found\n", __func__);
607         return NULL;
608 }
609
610 /* caller must hold sessionid_lock */
611 static void
612 unhash_session(struct nfsd4_session *ses)
613 {
614         list_del(&ses->se_hash);
615         list_del(&ses->se_perclnt);
616 }
617
618 static void
619 release_session(struct nfsd4_session *ses)
620 {
621         spin_lock(&sessionid_lock);
622         unhash_session(ses);
623         spin_unlock(&sessionid_lock);
624         nfsd4_put_session(ses);
625 }
626
627 void
628 free_session(struct kref *kref)
629 {
630         struct nfsd4_session *ses;
631
632         ses = container_of(kref, struct nfsd4_session, se_ref);
633         spin_lock(&nfsd_drc_lock);
634         nfsd_drc_mem_used -= ses->se_fchannel.maxreqs * NFSD_SLOT_CACHE_SIZE;
635         spin_unlock(&nfsd_drc_lock);
636         free_session_slots(ses);
637         kfree(ses);
638 }
639
640 static inline void
641 renew_client(struct nfs4_client *clp)
642 {
643         /*
644         * Move client to the end to the LRU list.
645         */
646         dprintk("renewing client (clientid %08x/%08x)\n", 
647                         clp->cl_clientid.cl_boot, 
648                         clp->cl_clientid.cl_id);
649         list_move_tail(&clp->cl_lru, &client_lru);
650         clp->cl_time = get_seconds();
651 }
652
653 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
654 static int
655 STALE_CLIENTID(clientid_t *clid)
656 {
657         if (clid->cl_boot == boot_time)
658                 return 0;
659         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
660                 clid->cl_boot, clid->cl_id, boot_time);
661         return 1;
662 }
663
664 /* 
665  * XXX Should we use a slab cache ?
666  * This type of memory management is somewhat inefficient, but we use it
667  * anyway since SETCLIENTID is not a common operation.
668  */
669 static struct nfs4_client *alloc_client(struct xdr_netobj name)
670 {
671         struct nfs4_client *clp;
672
673         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
674         if (clp == NULL)
675                 return NULL;
676         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
677         if (clp->cl_name.data == NULL) {
678                 kfree(clp);
679                 return NULL;
680         }
681         memcpy(clp->cl_name.data, name.data, name.len);
682         clp->cl_name.len = name.len;
683         return clp;
684 }
685
686 static void
687 shutdown_callback_client(struct nfs4_client *clp)
688 {
689         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
690
691         if (clnt) {
692                 /*
693                  * Callback threads take a reference on the client, so there
694                  * should be no outstanding callbacks at this point.
695                  */
696                 clp->cl_cb_conn.cb_client = NULL;
697                 rpc_shutdown_client(clnt);
698         }
699         if (clp->cl_cb_conn.cb_cred) {
700                 put_rpccred(clp->cl_cb_conn.cb_cred);
701                 clp->cl_cb_conn.cb_cred = NULL;
702         }
703 }
704
705 static inline void
706 free_client(struct nfs4_client *clp)
707 {
708         shutdown_callback_client(clp);
709         if (clp->cl_cred.cr_group_info)
710                 put_group_info(clp->cl_cred.cr_group_info);
711         kfree(clp->cl_principal);
712         kfree(clp->cl_name.data);
713         kfree(clp);
714 }
715
716 void
717 put_nfs4_client(struct nfs4_client *clp)
718 {
719         if (atomic_dec_and_test(&clp->cl_count))
720                 free_client(clp);
721 }
722
723 static void
724 expire_client(struct nfs4_client *clp)
725 {
726         struct nfs4_stateowner *sop;
727         struct nfs4_delegation *dp;
728         struct list_head reaplist;
729
730         dprintk("NFSD: expire_client cl_count %d\n",
731                             atomic_read(&clp->cl_count));
732
733         INIT_LIST_HEAD(&reaplist);
734         spin_lock(&recall_lock);
735         while (!list_empty(&clp->cl_delegations)) {
736                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
737                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
738                                 dp->dl_flock);
739                 list_del_init(&dp->dl_perclnt);
740                 list_move(&dp->dl_recall_lru, &reaplist);
741         }
742         spin_unlock(&recall_lock);
743         while (!list_empty(&reaplist)) {
744                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
745                 list_del_init(&dp->dl_recall_lru);
746                 unhash_delegation(dp);
747         }
748         list_del(&clp->cl_idhash);
749         list_del(&clp->cl_strhash);
750         list_del(&clp->cl_lru);
751         while (!list_empty(&clp->cl_openowners)) {
752                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
753                 release_openowner(sop);
754         }
755         while (!list_empty(&clp->cl_sessions)) {
756                 struct nfsd4_session  *ses;
757                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
758                                  se_perclnt);
759                 release_session(ses);
760         }
761         put_nfs4_client(clp);
762 }
763
764 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir)
765 {
766         struct nfs4_client *clp;
767
768         clp = alloc_client(name);
769         if (clp == NULL)
770                 return NULL;
771         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
772         atomic_set(&clp->cl_count, 1);
773         atomic_set(&clp->cl_cb_conn.cb_set, 0);
774         INIT_LIST_HEAD(&clp->cl_idhash);
775         INIT_LIST_HEAD(&clp->cl_strhash);
776         INIT_LIST_HEAD(&clp->cl_openowners);
777         INIT_LIST_HEAD(&clp->cl_delegations);
778         INIT_LIST_HEAD(&clp->cl_sessions);
779         INIT_LIST_HEAD(&clp->cl_lru);
780         return clp;
781 }
782
783 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
784 {
785         memcpy(target->cl_verifier.data, source->data,
786                         sizeof(target->cl_verifier.data));
787 }
788
789 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
790 {
791         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
792         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
793 }
794
795 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
796 {
797         target->cr_uid = source->cr_uid;
798         target->cr_gid = source->cr_gid;
799         target->cr_group_info = source->cr_group_info;
800         get_group_info(target->cr_group_info);
801 }
802
803 static int same_name(const char *n1, const char *n2)
804 {
805         return 0 == memcmp(n1, n2, HEXDIR_LEN);
806 }
807
808 static int
809 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
810 {
811         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
812 }
813
814 static int
815 same_clid(clientid_t *cl1, clientid_t *cl2)
816 {
817         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
818 }
819
820 /* XXX what about NGROUP */
821 static int
822 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
823 {
824         return cr1->cr_uid == cr2->cr_uid;
825 }
826
827 static void gen_clid(struct nfs4_client *clp)
828 {
829         static u32 current_clientid = 1;
830
831         clp->cl_clientid.cl_boot = boot_time;
832         clp->cl_clientid.cl_id = current_clientid++; 
833 }
834
835 static void gen_confirm(struct nfs4_client *clp)
836 {
837         static u32 i;
838         u32 *p;
839
840         p = (u32 *)clp->cl_confirm.data;
841         *p++ = get_seconds();
842         *p++ = i++;
843 }
844
845 static int check_name(struct xdr_netobj name)
846 {
847         if (name.len == 0) 
848                 return 0;
849         if (name.len > NFS4_OPAQUE_LIMIT) {
850                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
851                 return 0;
852         }
853         return 1;
854 }
855
856 static void
857 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
858 {
859         unsigned int idhashval;
860
861         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
862         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
863         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
864         list_add_tail(&clp->cl_lru, &client_lru);
865         clp->cl_time = get_seconds();
866 }
867
868 static void
869 move_to_confirmed(struct nfs4_client *clp)
870 {
871         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
872         unsigned int strhashval;
873
874         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
875         list_del_init(&clp->cl_strhash);
876         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
877         strhashval = clientstr_hashval(clp->cl_recdir);
878         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
879         renew_client(clp);
880 }
881
882 static struct nfs4_client *
883 find_confirmed_client(clientid_t *clid)
884 {
885         struct nfs4_client *clp;
886         unsigned int idhashval = clientid_hashval(clid->cl_id);
887
888         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
889                 if (same_clid(&clp->cl_clientid, clid))
890                         return clp;
891         }
892         return NULL;
893 }
894
895 static struct nfs4_client *
896 find_unconfirmed_client(clientid_t *clid)
897 {
898         struct nfs4_client *clp;
899         unsigned int idhashval = clientid_hashval(clid->cl_id);
900
901         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
902                 if (same_clid(&clp->cl_clientid, clid))
903                         return clp;
904         }
905         return NULL;
906 }
907
908 /*
909  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
910  * parameter. Matching is based on the fact the at least one of the
911  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
912  *
913  * FIXME: we need to unify the clientid namespaces for nfsv4.x
914  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
915  * and SET_CLIENTID{,_CONFIRM}
916  */
917 static inline int
918 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
919 {
920         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
921         return use_exchange_id == has_exchange_flags;
922 }
923
924 static struct nfs4_client *
925 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
926                              bool use_exchange_id)
927 {
928         struct nfs4_client *clp;
929
930         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
931                 if (same_name(clp->cl_recdir, dname) &&
932                     match_clientid_establishment(clp, use_exchange_id))
933                         return clp;
934         }
935         return NULL;
936 }
937
938 static struct nfs4_client *
939 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
940                                bool use_exchange_id)
941 {
942         struct nfs4_client *clp;
943
944         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
945                 if (same_name(clp->cl_recdir, dname) &&
946                     match_clientid_establishment(clp, use_exchange_id))
947                         return clp;
948         }
949         return NULL;
950 }
951
952 static void
953 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
954 {
955         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
956         unsigned short expected_family;
957
958         /* Currently, we only support tcp and tcp6 for the callback channel */
959         if (se->se_callback_netid_len == 3 &&
960             !memcmp(se->se_callback_netid_val, "tcp", 3))
961                 expected_family = AF_INET;
962         else if (se->se_callback_netid_len == 4 &&
963                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
964                 expected_family = AF_INET6;
965         else
966                 goto out_err;
967
968         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
969                                             se->se_callback_addr_len,
970                                             (struct sockaddr *) &cb->cb_addr,
971                                             sizeof(cb->cb_addr));
972
973         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
974                 goto out_err;
975
976         if (cb->cb_addr.ss_family == AF_INET6)
977                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
978
979         cb->cb_minorversion = 0;
980         cb->cb_prog = se->se_callback_prog;
981         cb->cb_ident = se->se_callback_ident;
982         return;
983 out_err:
984         cb->cb_addr.ss_family = AF_UNSPEC;
985         cb->cb_addrlen = 0;
986         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
987                 "will not receive delegations\n",
988                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
989
990         return;
991 }
992
993 /*
994  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
995  */
996 void
997 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
998 {
999         struct nfsd4_slot *slot = resp->cstate.slot;
1000         unsigned int base;
1001
1002         dprintk("--> %s slot %p\n", __func__, slot);
1003
1004         slot->sl_opcnt = resp->opcnt;
1005         slot->sl_status = resp->cstate.status;
1006
1007         if (nfsd4_not_cached(resp)) {
1008                 slot->sl_datalen = 0;
1009                 return;
1010         }
1011         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1012         base = (char *)resp->cstate.datap -
1013                                         (char *)resp->xbuf->head[0].iov_base;
1014         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1015                                     slot->sl_datalen))
1016                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1017         return;
1018 }
1019
1020 /*
1021  * Encode the replay sequence operation from the slot values.
1022  * If cachethis is FALSE encode the uncached rep error on the next
1023  * operation which sets resp->p and increments resp->opcnt for
1024  * nfs4svc_encode_compoundres.
1025  *
1026  */
1027 static __be32
1028 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1029                           struct nfsd4_compoundres *resp)
1030 {
1031         struct nfsd4_op *op;
1032         struct nfsd4_slot *slot = resp->cstate.slot;
1033
1034         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1035                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1036
1037         /* Encode the replayed sequence operation */
1038         op = &args->ops[resp->opcnt - 1];
1039         nfsd4_encode_operation(resp, op);
1040
1041         /* Return nfserr_retry_uncached_rep in next operation. */
1042         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1043                 op = &args->ops[resp->opcnt++];
1044                 op->status = nfserr_retry_uncached_rep;
1045                 nfsd4_encode_operation(resp, op);
1046         }
1047         return op->status;
1048 }
1049
1050 /*
1051  * The sequence operation is not cached because we can use the slot and
1052  * session values.
1053  */
1054 __be32
1055 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1056                          struct nfsd4_sequence *seq)
1057 {
1058         struct nfsd4_slot *slot = resp->cstate.slot;
1059         __be32 status;
1060
1061         dprintk("--> %s slot %p\n", __func__, slot);
1062
1063         /* Either returns 0 or nfserr_retry_uncached */
1064         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1065         if (status == nfserr_retry_uncached_rep)
1066                 return status;
1067
1068         /* The sequence operation has been encoded, cstate->datap set. */
1069         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1070
1071         resp->opcnt = slot->sl_opcnt;
1072         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1073         status = slot->sl_status;
1074
1075         return status;
1076 }
1077
1078 /*
1079  * Set the exchange_id flags returned by the server.
1080  */
1081 static void
1082 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1083 {
1084         /* pNFS is not supported */
1085         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1086
1087         /* Referrals are supported, Migration is not. */
1088         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1089
1090         /* set the wire flags to return to client. */
1091         clid->flags = new->cl_exchange_flags;
1092 }
1093
1094 __be32
1095 nfsd4_exchange_id(struct svc_rqst *rqstp,
1096                   struct nfsd4_compound_state *cstate,
1097                   struct nfsd4_exchange_id *exid)
1098 {
1099         struct nfs4_client *unconf, *conf, *new;
1100         int status;
1101         unsigned int            strhashval;
1102         char                    dname[HEXDIR_LEN];
1103         char                    addr_str[INET6_ADDRSTRLEN];
1104         nfs4_verifier           verf = exid->verifier;
1105         struct sockaddr         *sa = svc_addr(rqstp);
1106
1107         rpc_ntop(sa, addr_str, sizeof(addr_str));
1108         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1109                 "ip_addr=%s flags %x, spa_how %d\n",
1110                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1111                 addr_str, exid->flags, exid->spa_how);
1112
1113         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1114                 return nfserr_inval;
1115
1116         /* Currently only support SP4_NONE */
1117         switch (exid->spa_how) {
1118         case SP4_NONE:
1119                 break;
1120         case SP4_SSV:
1121                 return nfserr_encr_alg_unsupp;
1122         default:
1123                 BUG();                          /* checked by xdr code */
1124         case SP4_MACH_CRED:
1125                 return nfserr_serverfault;      /* no excuse :-/ */
1126         }
1127
1128         status = nfs4_make_rec_clidname(dname, &exid->clname);
1129
1130         if (status)
1131                 goto error;
1132
1133         strhashval = clientstr_hashval(dname);
1134
1135         nfs4_lock_state();
1136         status = nfs_ok;
1137
1138         conf = find_confirmed_client_by_str(dname, strhashval, true);
1139         if (conf) {
1140                 if (!same_verf(&verf, &conf->cl_verifier)) {
1141                         /* 18.35.4 case 8 */
1142                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1143                                 status = nfserr_not_same;
1144                                 goto out;
1145                         }
1146                         /* Client reboot: destroy old state */
1147                         expire_client(conf);
1148                         goto out_new;
1149                 }
1150                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1151                         /* 18.35.4 case 9 */
1152                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1153                                 status = nfserr_perm;
1154                                 goto out;
1155                         }
1156                         expire_client(conf);
1157                         goto out_new;
1158                 }
1159                 /*
1160                  * Set bit when the owner id and verifier map to an already
1161                  * confirmed client id (18.35.3).
1162                  */
1163                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1164
1165                 /*
1166                  * Falling into 18.35.4 case 2, possible router replay.
1167                  * Leave confirmed record intact and return same result.
1168                  */
1169                 copy_verf(conf, &verf);
1170                 new = conf;
1171                 goto out_copy;
1172         }
1173
1174         /* 18.35.4 case 7 */
1175         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1176                 status = nfserr_noent;
1177                 goto out;
1178         }
1179
1180         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1181         if (unconf) {
1182                 /*
1183                  * Possible retry or client restart.  Per 18.35.4 case 4,
1184                  * a new unconfirmed record should be generated regardless
1185                  * of whether any properties have changed.
1186                  */
1187                 expire_client(unconf);
1188         }
1189
1190 out_new:
1191         /* Normal case */
1192         new = create_client(exid->clname, dname);
1193         if (new == NULL) {
1194                 status = nfserr_serverfault;
1195                 goto out;
1196         }
1197
1198         copy_verf(new, &verf);
1199         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1200         rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1201         gen_clid(new);
1202         gen_confirm(new);
1203         add_to_unconfirmed(new, strhashval);
1204 out_copy:
1205         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1206         exid->clientid.cl_id = new->cl_clientid.cl_id;
1207
1208         exid->seqid = 1;
1209         nfsd4_set_ex_flags(new, exid);
1210
1211         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1212                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1213         status = nfs_ok;
1214
1215 out:
1216         nfs4_unlock_state();
1217 error:
1218         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1219         return status;
1220 }
1221
1222 static int
1223 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1224 {
1225         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1226                 slot_seqid);
1227
1228         /* The slot is in use, and no response has been sent. */
1229         if (slot_inuse) {
1230                 if (seqid == slot_seqid)
1231                         return nfserr_jukebox;
1232                 else
1233                         return nfserr_seq_misordered;
1234         }
1235         /* Normal */
1236         if (likely(seqid == slot_seqid + 1))
1237                 return nfs_ok;
1238         /* Replay */
1239         if (seqid == slot_seqid)
1240                 return nfserr_replay_cache;
1241         /* Wraparound */
1242         if (seqid == 1 && (slot_seqid + 1) == 0)
1243                 return nfs_ok;
1244         /* Misordered replay or misordered new request */
1245         return nfserr_seq_misordered;
1246 }
1247
1248 /*
1249  * Cache the create session result into the create session single DRC
1250  * slot cache by saving the xdr structure. sl_seqid has been set.
1251  * Do this for solo or embedded create session operations.
1252  */
1253 static void
1254 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1255                            struct nfsd4_clid_slot *slot, int nfserr)
1256 {
1257         slot->sl_status = nfserr;
1258         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1259 }
1260
1261 static __be32
1262 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1263                             struct nfsd4_clid_slot *slot)
1264 {
1265         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1266         return slot->sl_status;
1267 }
1268
1269 __be32
1270 nfsd4_create_session(struct svc_rqst *rqstp,
1271                      struct nfsd4_compound_state *cstate,
1272                      struct nfsd4_create_session *cr_ses)
1273 {
1274         struct sockaddr *sa = svc_addr(rqstp);
1275         struct nfs4_client *conf, *unconf;
1276         struct nfsd4_clid_slot *cs_slot = NULL;
1277         int status = 0;
1278
1279         nfs4_lock_state();
1280         unconf = find_unconfirmed_client(&cr_ses->clientid);
1281         conf = find_confirmed_client(&cr_ses->clientid);
1282
1283         if (conf) {
1284                 cs_slot = &conf->cl_cs_slot;
1285                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1286                 if (status == nfserr_replay_cache) {
1287                         dprintk("Got a create_session replay! seqid= %d\n",
1288                                 cs_slot->sl_seqid);
1289                         /* Return the cached reply status */
1290                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1291                         goto out;
1292                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1293                         status = nfserr_seq_misordered;
1294                         dprintk("Sequence misordered!\n");
1295                         dprintk("Expected seqid= %d but got seqid= %d\n",
1296                                 cs_slot->sl_seqid, cr_ses->seqid);
1297                         goto out;
1298                 }
1299                 cs_slot->sl_seqid++;
1300         } else if (unconf) {
1301                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1302                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1303                         status = nfserr_clid_inuse;
1304                         goto out;
1305                 }
1306
1307                 cs_slot = &unconf->cl_cs_slot;
1308                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1309                 if (status) {
1310                         /* an unconfirmed replay returns misordered */
1311                         status = nfserr_seq_misordered;
1312                         goto out_cache;
1313                 }
1314
1315                 cs_slot->sl_seqid++; /* from 0 to 1 */
1316                 move_to_confirmed(unconf);
1317
1318                 /*
1319                  * We do not support RDMA or persistent sessions
1320                  */
1321                 cr_ses->flags &= ~SESSION4_PERSIST;
1322                 cr_ses->flags &= ~SESSION4_RDMA;
1323
1324                 conf = unconf;
1325         } else {
1326                 status = nfserr_stale_clientid;
1327                 goto out;
1328         }
1329
1330         status = alloc_init_session(rqstp, conf, cr_ses);
1331         if (status)
1332                 goto out;
1333
1334         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1335                NFS4_MAX_SESSIONID_LEN);
1336         cr_ses->seqid = cs_slot->sl_seqid;
1337
1338 out_cache:
1339         /* cache solo and embedded create sessions under the state lock */
1340         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1341 out:
1342         nfs4_unlock_state();
1343         dprintk("%s returns %d\n", __func__, ntohl(status));
1344         return status;
1345 }
1346
1347 __be32
1348 nfsd4_destroy_session(struct svc_rqst *r,
1349                       struct nfsd4_compound_state *cstate,
1350                       struct nfsd4_destroy_session *sessionid)
1351 {
1352         struct nfsd4_session *ses;
1353         u32 status = nfserr_badsession;
1354
1355         /* Notes:
1356          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1357          * - Should we return nfserr_back_chan_busy if waiting for
1358          *   callbacks on to-be-destroyed session?
1359          * - Do we need to clear any callback info from previous session?
1360          */
1361
1362         dump_sessionid(__func__, &sessionid->sessionid);
1363         spin_lock(&sessionid_lock);
1364         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1365         if (!ses) {
1366                 spin_unlock(&sessionid_lock);
1367                 goto out;
1368         }
1369
1370         unhash_session(ses);
1371         spin_unlock(&sessionid_lock);
1372
1373         /* wait for callbacks */
1374         shutdown_callback_client(ses->se_client);
1375         nfsd4_put_session(ses);
1376         status = nfs_ok;
1377 out:
1378         dprintk("%s returns %d\n", __func__, ntohl(status));
1379         return status;
1380 }
1381
1382 __be32
1383 nfsd4_sequence(struct svc_rqst *rqstp,
1384                struct nfsd4_compound_state *cstate,
1385                struct nfsd4_sequence *seq)
1386 {
1387         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1388         struct nfsd4_session *session;
1389         struct nfsd4_slot *slot;
1390         int status;
1391
1392         if (resp->opcnt != 1)
1393                 return nfserr_sequence_pos;
1394
1395         spin_lock(&sessionid_lock);
1396         status = nfserr_badsession;
1397         session = find_in_sessionid_hashtbl(&seq->sessionid);
1398         if (!session)
1399                 goto out;
1400
1401         status = nfserr_badslot;
1402         if (seq->slotid >= session->se_fchannel.maxreqs)
1403                 goto out;
1404
1405         slot = session->se_slots[seq->slotid];
1406         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1407
1408         /* We do not negotiate the number of slots yet, so set the
1409          * maxslots to the session maxreqs which is used to encode
1410          * sr_highest_slotid and the sr_target_slot id to maxslots */
1411         seq->maxslots = session->se_fchannel.maxreqs;
1412
1413         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1414         if (status == nfserr_replay_cache) {
1415                 cstate->slot = slot;
1416                 cstate->session = session;
1417                 /* Return the cached reply status and set cstate->status
1418                  * for nfsd4_proc_compound processing */
1419                 status = nfsd4_replay_cache_entry(resp, seq);
1420                 cstate->status = nfserr_replay_cache;
1421                 goto out;
1422         }
1423         if (status)
1424                 goto out;
1425
1426         /* Success! bump slot seqid */
1427         slot->sl_inuse = true;
1428         slot->sl_seqid = seq->seqid;
1429         slot->sl_cachethis = seq->cachethis;
1430
1431         cstate->slot = slot;
1432         cstate->session = session;
1433
1434         /* Hold a session reference until done processing the compound:
1435          * nfsd4_put_session called only if the cstate slot is set.
1436          */
1437         nfsd4_get_session(session);
1438 out:
1439         spin_unlock(&sessionid_lock);
1440         /* Renew the clientid on success and on replay */
1441         if (cstate->session) {
1442                 nfs4_lock_state();
1443                 renew_client(session->se_client);
1444                 nfs4_unlock_state();
1445         }
1446         dprintk("%s: return %d\n", __func__, ntohl(status));
1447         return status;
1448 }
1449
1450 __be32
1451 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1452                   struct nfsd4_setclientid *setclid)
1453 {
1454         struct sockaddr         *sa = svc_addr(rqstp);
1455         struct xdr_netobj       clname = { 
1456                 .len = setclid->se_namelen,
1457                 .data = setclid->se_name,
1458         };
1459         nfs4_verifier           clverifier = setclid->se_verf;
1460         unsigned int            strhashval;
1461         struct nfs4_client      *conf, *unconf, *new;
1462         __be32                  status;
1463         char                    *princ;
1464         char                    dname[HEXDIR_LEN];
1465         
1466         if (!check_name(clname))
1467                 return nfserr_inval;
1468
1469         status = nfs4_make_rec_clidname(dname, &clname);
1470         if (status)
1471                 return status;
1472
1473         /* 
1474          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1475          * We get here on a DRC miss.
1476          */
1477
1478         strhashval = clientstr_hashval(dname);
1479
1480         nfs4_lock_state();
1481         conf = find_confirmed_client_by_str(dname, strhashval, false);
1482         if (conf) {
1483                 /* RFC 3530 14.2.33 CASE 0: */
1484                 status = nfserr_clid_inuse;
1485                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1486                         char addr_str[INET6_ADDRSTRLEN];
1487                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1488                                  sizeof(addr_str));
1489                         dprintk("NFSD: setclientid: string in use by client "
1490                                 "at %s\n", addr_str);
1491                         goto out;
1492                 }
1493         }
1494         /*
1495          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1496          * has a description of SETCLIENTID request processing consisting
1497          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1498          */
1499         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1500         status = nfserr_resource;
1501         if (!conf) {
1502                 /*
1503                  * RFC 3530 14.2.33 CASE 4:
1504                  * placed first, because it is the normal case
1505                  */
1506                 if (unconf)
1507                         expire_client(unconf);
1508                 new = create_client(clname, dname);
1509                 if (new == NULL)
1510                         goto out;
1511                 gen_clid(new);
1512         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1513                 /*
1514                  * RFC 3530 14.2.33 CASE 1:
1515                  * probable callback update
1516                  */
1517                 if (unconf) {
1518                         /* Note this is removing unconfirmed {*x***},
1519                          * which is stronger than RFC recommended {vxc**}.
1520                          * This has the advantage that there is at most
1521                          * one {*x***} in either list at any time.
1522                          */
1523                         expire_client(unconf);
1524                 }
1525                 new = create_client(clname, dname);
1526                 if (new == NULL)
1527                         goto out;
1528                 copy_clid(new, conf);
1529         } else if (!unconf) {
1530                 /*
1531                  * RFC 3530 14.2.33 CASE 2:
1532                  * probable client reboot; state will be removed if
1533                  * confirmed.
1534                  */
1535                 new = create_client(clname, dname);
1536                 if (new == NULL)
1537                         goto out;
1538                 gen_clid(new);
1539         } else {
1540                 /*
1541                  * RFC 3530 14.2.33 CASE 3:
1542                  * probable client reboot; state will be removed if
1543                  * confirmed.
1544                  */
1545                 expire_client(unconf);
1546                 new = create_client(clname, dname);
1547                 if (new == NULL)
1548                         goto out;
1549                 gen_clid(new);
1550         }
1551         copy_verf(new, &clverifier);
1552         rpc_copy_addr((struct sockaddr *) &new->cl_addr, sa);
1553         new->cl_flavor = rqstp->rq_flavor;
1554         princ = svc_gss_principal(rqstp);
1555         if (princ) {
1556                 new->cl_principal = kstrdup(princ, GFP_KERNEL);
1557                 if (new->cl_principal == NULL) {
1558                         free_client(new);
1559                         goto out;
1560                 }
1561         }
1562         copy_cred(&new->cl_cred, &rqstp->rq_cred);
1563         gen_confirm(new);
1564         gen_callback(new, setclid, rpc_get_scope_id(sa));
1565         add_to_unconfirmed(new, strhashval);
1566         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1567         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1568         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1569         status = nfs_ok;
1570 out:
1571         nfs4_unlock_state();
1572         return status;
1573 }
1574
1575
1576 /*
1577  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1578  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1579  * bullets, labeled as CASE1 - CASE4 below.
1580  */
1581 __be32
1582 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1583                          struct nfsd4_compound_state *cstate,
1584                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1585 {
1586         struct sockaddr *sa = svc_addr(rqstp);
1587         struct nfs4_client *conf, *unconf;
1588         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1589         clientid_t * clid = &setclientid_confirm->sc_clientid;
1590         __be32 status;
1591
1592         if (STALE_CLIENTID(clid))
1593                 return nfserr_stale_clientid;
1594         /* 
1595          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1596          * We get here on a DRC miss.
1597          */
1598
1599         nfs4_lock_state();
1600
1601         conf = find_confirmed_client(clid);
1602         unconf = find_unconfirmed_client(clid);
1603
1604         status = nfserr_clid_inuse;
1605         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1606                 goto out;
1607         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1608                 goto out;
1609
1610         /*
1611          * section 14.2.34 of RFC 3530 has a description of
1612          * SETCLIENTID_CONFIRM request processing consisting
1613          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1614          */
1615         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1616                 /*
1617                  * RFC 3530 14.2.34 CASE 1:
1618                  * callback update
1619                  */
1620                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1621                         status = nfserr_clid_inuse;
1622                 else {
1623                         /* XXX: We just turn off callbacks until we can handle
1624                           * change request correctly. */
1625                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1626                         expire_client(unconf);
1627                         status = nfs_ok;
1628
1629                 }
1630         } else if (conf && !unconf) {
1631                 /*
1632                  * RFC 3530 14.2.34 CASE 2:
1633                  * probable retransmitted request; play it safe and
1634                  * do nothing.
1635                  */
1636                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1637                         status = nfserr_clid_inuse;
1638                 else
1639                         status = nfs_ok;
1640         } else if (!conf && unconf
1641                         && same_verf(&unconf->cl_confirm, &confirm)) {
1642                 /*
1643                  * RFC 3530 14.2.34 CASE 3:
1644                  * Normal case; new or rebooted client:
1645                  */
1646                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1647                         status = nfserr_clid_inuse;
1648                 } else {
1649                         unsigned int hash =
1650                                 clientstr_hashval(unconf->cl_recdir);
1651                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1652                                                             hash, false);
1653                         if (conf) {
1654                                 nfsd4_remove_clid_dir(conf);
1655                                 expire_client(conf);
1656                         }
1657                         move_to_confirmed(unconf);
1658                         conf = unconf;
1659                         nfsd4_probe_callback(conf);
1660                         status = nfs_ok;
1661                 }
1662         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1663             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1664                                                                 &confirm)))) {
1665                 /*
1666                  * RFC 3530 14.2.34 CASE 4:
1667                  * Client probably hasn't noticed that we rebooted yet.
1668                  */
1669                 status = nfserr_stale_clientid;
1670         } else {
1671                 /* check that we have hit one of the cases...*/
1672                 status = nfserr_clid_inuse;
1673         }
1674 out:
1675         nfs4_unlock_state();
1676         return status;
1677 }
1678
1679 /* OPEN Share state helper functions */
1680 static inline struct nfs4_file *
1681 alloc_init_file(struct inode *ino)
1682 {
1683         struct nfs4_file *fp;
1684         unsigned int hashval = file_hashval(ino);
1685
1686         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1687         if (fp) {
1688                 atomic_set(&fp->fi_ref, 1);
1689                 INIT_LIST_HEAD(&fp->fi_hash);
1690                 INIT_LIST_HEAD(&fp->fi_stateids);
1691                 INIT_LIST_HEAD(&fp->fi_delegations);
1692                 spin_lock(&recall_lock);
1693                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1694                 spin_unlock(&recall_lock);
1695                 fp->fi_inode = igrab(ino);
1696                 fp->fi_id = current_fileid++;
1697                 fp->fi_had_conflict = false;
1698                 return fp;
1699         }
1700         return NULL;
1701 }
1702
1703 static void
1704 nfsd4_free_slab(struct kmem_cache **slab)
1705 {
1706         if (*slab == NULL)
1707                 return;
1708         kmem_cache_destroy(*slab);
1709         *slab = NULL;
1710 }
1711
1712 void
1713 nfsd4_free_slabs(void)
1714 {
1715         nfsd4_free_slab(&stateowner_slab);
1716         nfsd4_free_slab(&file_slab);
1717         nfsd4_free_slab(&stateid_slab);
1718         nfsd4_free_slab(&deleg_slab);
1719 }
1720
1721 static int
1722 nfsd4_init_slabs(void)
1723 {
1724         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1725                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1726         if (stateowner_slab == NULL)
1727                 goto out_nomem;
1728         file_slab = kmem_cache_create("nfsd4_files",
1729                         sizeof(struct nfs4_file), 0, 0, NULL);
1730         if (file_slab == NULL)
1731                 goto out_nomem;
1732         stateid_slab = kmem_cache_create("nfsd4_stateids",
1733                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1734         if (stateid_slab == NULL)
1735                 goto out_nomem;
1736         deleg_slab = kmem_cache_create("nfsd4_delegations",
1737                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1738         if (deleg_slab == NULL)
1739                 goto out_nomem;
1740         return 0;
1741 out_nomem:
1742         nfsd4_free_slabs();
1743         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1744         return -ENOMEM;
1745 }
1746
1747 void
1748 nfs4_free_stateowner(struct kref *kref)
1749 {
1750         struct nfs4_stateowner *sop =
1751                 container_of(kref, struct nfs4_stateowner, so_ref);
1752         kfree(sop->so_owner.data);
1753         kmem_cache_free(stateowner_slab, sop);
1754 }
1755
1756 static inline struct nfs4_stateowner *
1757 alloc_stateowner(struct xdr_netobj *owner)
1758 {
1759         struct nfs4_stateowner *sop;
1760
1761         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1762                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1763                         memcpy(sop->so_owner.data, owner->data, owner->len);
1764                         sop->so_owner.len = owner->len;
1765                         kref_init(&sop->so_ref);
1766                         return sop;
1767                 } 
1768                 kmem_cache_free(stateowner_slab, sop);
1769         }
1770         return NULL;
1771 }
1772
1773 static struct nfs4_stateowner *
1774 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1775         struct nfs4_stateowner *sop;
1776         struct nfs4_replay *rp;
1777         unsigned int idhashval;
1778
1779         if (!(sop = alloc_stateowner(&open->op_owner)))
1780                 return NULL;
1781         idhashval = ownerid_hashval(current_ownerid);
1782         INIT_LIST_HEAD(&sop->so_idhash);
1783         INIT_LIST_HEAD(&sop->so_strhash);
1784         INIT_LIST_HEAD(&sop->so_perclient);
1785         INIT_LIST_HEAD(&sop->so_stateids);
1786         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1787         INIT_LIST_HEAD(&sop->so_close_lru);
1788         sop->so_time = 0;
1789         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1790         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1791         list_add(&sop->so_perclient, &clp->cl_openowners);
1792         sop->so_is_open_owner = 1;
1793         sop->so_id = current_ownerid++;
1794         sop->so_client = clp;
1795         sop->so_seqid = open->op_seqid;
1796         sop->so_confirmed = 0;
1797         rp = &sop->so_replay;
1798         rp->rp_status = nfserr_serverfault;
1799         rp->rp_buflen = 0;
1800         rp->rp_buf = rp->rp_ibuf;
1801         return sop;
1802 }
1803
1804 static inline void
1805 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1806         struct nfs4_stateowner *sop = open->op_stateowner;
1807         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1808
1809         INIT_LIST_HEAD(&stp->st_hash);
1810         INIT_LIST_HEAD(&stp->st_perstateowner);
1811         INIT_LIST_HEAD(&stp->st_lockowners);
1812         INIT_LIST_HEAD(&stp->st_perfile);
1813         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1814         list_add(&stp->st_perstateowner, &sop->so_stateids);
1815         list_add(&stp->st_perfile, &fp->fi_stateids);
1816         stp->st_stateowner = sop;
1817         get_nfs4_file(fp);
1818         stp->st_file = fp;
1819         stp->st_stateid.si_boot = get_seconds();
1820         stp->st_stateid.si_stateownerid = sop->so_id;
1821         stp->st_stateid.si_fileid = fp->fi_id;
1822         stp->st_stateid.si_generation = 0;
1823         stp->st_access_bmap = 0;
1824         stp->st_deny_bmap = 0;
1825         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1826                   &stp->st_access_bmap);
1827         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1828         stp->st_openstp = NULL;
1829 }
1830
1831 static void
1832 move_to_close_lru(struct nfs4_stateowner *sop)
1833 {
1834         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1835
1836         list_move_tail(&sop->so_close_lru, &close_lru);
1837         sop->so_time = get_seconds();
1838 }
1839
1840 static int
1841 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1842                                                         clientid_t *clid)
1843 {
1844         return (sop->so_owner.len == owner->len) &&
1845                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1846                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1847 }
1848
1849 static struct nfs4_stateowner *
1850 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1851 {
1852         struct nfs4_stateowner *so = NULL;
1853
1854         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1855                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1856                         return so;
1857         }
1858         return NULL;
1859 }
1860
1861 /* search file_hashtbl[] for file */
1862 static struct nfs4_file *
1863 find_file(struct inode *ino)
1864 {
1865         unsigned int hashval = file_hashval(ino);
1866         struct nfs4_file *fp;
1867
1868         spin_lock(&recall_lock);
1869         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1870                 if (fp->fi_inode == ino) {
1871                         get_nfs4_file(fp);
1872                         spin_unlock(&recall_lock);
1873                         return fp;
1874                 }
1875         }
1876         spin_unlock(&recall_lock);
1877         return NULL;
1878 }
1879
1880 static inline int access_valid(u32 x, u32 minorversion)
1881 {
1882         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1883                 return 0;
1884         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1885                 return 0;
1886         x &= ~NFS4_SHARE_ACCESS_MASK;
1887         if (minorversion && x) {
1888                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1889                         return 0;
1890                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1891                         return 0;
1892                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1893         }
1894         if (x)
1895                 return 0;
1896         return 1;
1897 }
1898
1899 static inline int deny_valid(u32 x)
1900 {
1901         /* Note: unlike access bits, deny bits may be zero. */
1902         return x <= NFS4_SHARE_DENY_BOTH;
1903 }
1904
1905 /*
1906  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1907  * st_{access,deny}_bmap field of the stateid, in order to track not
1908  * only what share bits are currently in force, but also what
1909  * combinations of share bits previous opens have used.  This allows us
1910  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1911  * return an error if the client attempt to downgrade to a combination
1912  * of share bits not explicable by closing some of its previous opens.
1913  *
1914  * XXX: This enforcement is actually incomplete, since we don't keep
1915  * track of access/deny bit combinations; so, e.g., we allow:
1916  *
1917  *      OPEN allow read, deny write
1918  *      OPEN allow both, deny none
1919  *      DOWNGRADE allow read, deny none
1920  *
1921  * which we should reject.
1922  */
1923 static void
1924 set_access(unsigned int *access, unsigned long bmap) {
1925         int i;
1926
1927         *access = 0;
1928         for (i = 1; i < 4; i++) {
1929                 if (test_bit(i, &bmap))
1930                         *access |= i;
1931         }
1932 }
1933
1934 static void
1935 set_deny(unsigned int *deny, unsigned long bmap) {
1936         int i;
1937
1938         *deny = 0;
1939         for (i = 0; i < 4; i++) {
1940                 if (test_bit(i, &bmap))
1941                         *deny |= i ;
1942         }
1943 }
1944
1945 static int
1946 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1947         unsigned int access, deny;
1948
1949         set_access(&access, stp->st_access_bmap);
1950         set_deny(&deny, stp->st_deny_bmap);
1951         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1952                 return 0;
1953         return 1;
1954 }
1955
1956 /*
1957  * Called to check deny when READ with all zero stateid or
1958  * WRITE with all zero or all one stateid
1959  */
1960 static __be32
1961 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1962 {
1963         struct inode *ino = current_fh->fh_dentry->d_inode;
1964         struct nfs4_file *fp;
1965         struct nfs4_stateid *stp;
1966         __be32 ret;
1967
1968         dprintk("NFSD: nfs4_share_conflict\n");
1969
1970         fp = find_file(ino);
1971         if (!fp)
1972                 return nfs_ok;
1973         ret = nfserr_locked;
1974         /* Search for conflicting share reservations */
1975         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1976                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1977                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1978                         goto out;
1979         }
1980         ret = nfs_ok;
1981 out:
1982         put_nfs4_file(fp);
1983         return ret;
1984 }
1985
1986 static inline void
1987 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1988 {
1989         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1990                 drop_file_write_access(filp);
1991                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1992         }
1993 }
1994
1995 /*
1996  * Spawn a thread to perform a recall on the delegation represented
1997  * by the lease (file_lock)
1998  *
1999  * Called from break_lease() with lock_kernel() held.
2000  * Note: we assume break_lease will only call this *once* for any given
2001  * lease.
2002  */
2003 static
2004 void nfsd_break_deleg_cb(struct file_lock *fl)
2005 {
2006         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2007
2008         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2009         if (!dp)
2010                 return;
2011
2012         /* We're assuming the state code never drops its reference
2013          * without first removing the lease.  Since we're in this lease
2014          * callback (and since the lease code is serialized by the kernel
2015          * lock) we know the server hasn't removed the lease yet, we know
2016          * it's safe to take a reference: */
2017         atomic_inc(&dp->dl_count);
2018         atomic_inc(&dp->dl_client->cl_count);
2019
2020         spin_lock(&recall_lock);
2021         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2022         spin_unlock(&recall_lock);
2023
2024         /* only place dl_time is set. protected by lock_kernel*/
2025         dp->dl_time = get_seconds();
2026
2027         /*
2028          * We don't want the locks code to timeout the lease for us;
2029          * we'll remove it ourself if the delegation isn't returned
2030          * in time.
2031          */
2032         fl->fl_break_time = 0;
2033
2034         dp->dl_file->fi_had_conflict = true;
2035         nfsd4_cb_recall(dp);
2036 }
2037
2038 /*
2039  * The file_lock is being reapd.
2040  *
2041  * Called by locks_free_lock() with lock_kernel() held.
2042  */
2043 static
2044 void nfsd_release_deleg_cb(struct file_lock *fl)
2045 {
2046         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2047
2048         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2049
2050         if (!(fl->fl_flags & FL_LEASE) || !dp)
2051                 return;
2052         dp->dl_flock = NULL;
2053 }
2054
2055 /*
2056  * Set the delegation file_lock back pointer.
2057  *
2058  * Called from setlease() with lock_kernel() held.
2059  */
2060 static
2061 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2062 {
2063         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2064
2065         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2066         if (!dp)
2067                 return;
2068         dp->dl_flock = new;
2069 }
2070
2071 /*
2072  * Called from setlease() with lock_kernel() held
2073  */
2074 static
2075 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2076 {
2077         struct nfs4_delegation *onlistd =
2078                 (struct nfs4_delegation *)onlist->fl_owner;
2079         struct nfs4_delegation *tryd =
2080                 (struct nfs4_delegation *)try->fl_owner;
2081
2082         if (onlist->fl_lmops != try->fl_lmops)
2083                 return 0;
2084
2085         return onlistd->dl_client == tryd->dl_client;
2086 }
2087
2088
2089 static
2090 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2091 {
2092         if (arg & F_UNLCK)
2093                 return lease_modify(onlist, arg);
2094         else
2095                 return -EAGAIN;
2096 }
2097
2098 static struct lock_manager_operations nfsd_lease_mng_ops = {
2099         .fl_break = nfsd_break_deleg_cb,
2100         .fl_release_private = nfsd_release_deleg_cb,
2101         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2102         .fl_mylease = nfsd_same_client_deleg_cb,
2103         .fl_change = nfsd_change_deleg_cb,
2104 };
2105
2106
2107 __be32
2108 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2109                     struct nfsd4_open *open)
2110 {
2111         clientid_t *clientid = &open->op_clientid;
2112         struct nfs4_client *clp = NULL;
2113         unsigned int strhashval;
2114         struct nfs4_stateowner *sop = NULL;
2115
2116         if (!check_name(open->op_owner))
2117                 return nfserr_inval;
2118
2119         if (STALE_CLIENTID(&open->op_clientid))
2120                 return nfserr_stale_clientid;
2121
2122         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2123         sop = find_openstateowner_str(strhashval, open);
2124         open->op_stateowner = sop;
2125         if (!sop) {
2126                 /* Make sure the client's lease hasn't expired. */
2127                 clp = find_confirmed_client(clientid);
2128                 if (clp == NULL)
2129                         return nfserr_expired;
2130                 goto renew;
2131         }
2132         /* When sessions are used, skip open sequenceid processing */
2133         if (nfsd4_has_session(cstate))
2134                 goto renew;
2135         if (!sop->so_confirmed) {
2136                 /* Replace unconfirmed owners without checking for replay. */
2137                 clp = sop->so_client;
2138                 release_openowner(sop);
2139                 open->op_stateowner = NULL;
2140                 goto renew;
2141         }
2142         if (open->op_seqid == sop->so_seqid - 1) {
2143                 if (sop->so_replay.rp_buflen)
2144                         return nfserr_replay_me;
2145                 /* The original OPEN failed so spectacularly
2146                  * that we don't even have replay data saved!
2147                  * Therefore, we have no choice but to continue
2148                  * processing this OPEN; presumably, we'll
2149                  * fail again for the same reason.
2150                  */
2151                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2152                 goto renew;
2153         }
2154         if (open->op_seqid != sop->so_seqid)
2155                 return nfserr_bad_seqid;
2156 renew:
2157         if (open->op_stateowner == NULL) {
2158                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2159                 if (sop == NULL)
2160                         return nfserr_resource;
2161                 open->op_stateowner = sop;
2162         }
2163         list_del_init(&sop->so_close_lru);
2164         renew_client(sop->so_client);
2165         return nfs_ok;
2166 }
2167
2168 static inline __be32
2169 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2170 {
2171         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2172                 return nfserr_openmode;
2173         else
2174                 return nfs_ok;
2175 }
2176
2177 static struct nfs4_delegation *
2178 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2179 {
2180         struct nfs4_delegation *dp;
2181
2182         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2183                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2184                         return dp;
2185         }
2186         return NULL;
2187 }
2188
2189 static __be32
2190 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2191                 struct nfs4_delegation **dp)
2192 {
2193         int flags;
2194         __be32 status = nfserr_bad_stateid;
2195
2196         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2197         if (*dp == NULL)
2198                 goto out;
2199         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2200                                                 RD_STATE : WR_STATE;
2201         status = nfs4_check_delegmode(*dp, flags);
2202         if (status)
2203                 *dp = NULL;
2204 out:
2205         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2206                 return nfs_ok;
2207         if (status)
2208                 return status;
2209         open->op_stateowner->so_confirmed = 1;
2210         return nfs_ok;
2211 }
2212
2213 static __be32
2214 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2215 {
2216         struct nfs4_stateid *local;
2217         __be32 status = nfserr_share_denied;
2218         struct nfs4_stateowner *sop = open->op_stateowner;
2219
2220         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2221                 /* ignore lock owners */
2222                 if (local->st_stateowner->so_is_open_owner == 0)
2223                         continue;
2224                 /* remember if we have seen this open owner */
2225                 if (local->st_stateowner == sop)
2226                         *stpp = local;
2227                 /* check for conflicting share reservations */
2228                 if (!test_share(local, open))
2229                         goto out;
2230         }
2231         status = 0;
2232 out:
2233         return status;
2234 }
2235
2236 static inline struct nfs4_stateid *
2237 nfs4_alloc_stateid(void)
2238 {
2239         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2240 }
2241
2242 static __be32
2243 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2244                 struct nfs4_delegation *dp,
2245                 struct svc_fh *cur_fh, int flags)
2246 {
2247         struct nfs4_stateid *stp;
2248
2249         stp = nfs4_alloc_stateid();
2250         if (stp == NULL)
2251                 return nfserr_resource;
2252
2253         if (dp) {
2254                 get_file(dp->dl_vfs_file);
2255                 stp->st_vfs_file = dp->dl_vfs_file;
2256         } else {
2257                 __be32 status;
2258                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2259                                 &stp->st_vfs_file);
2260                 if (status) {
2261                         if (status == nfserr_dropit)
2262                                 status = nfserr_jukebox;
2263                         kmem_cache_free(stateid_slab, stp);
2264                         return status;
2265                 }
2266         }
2267         *stpp = stp;
2268         return 0;
2269 }
2270
2271 static inline __be32
2272 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2273                 struct nfsd4_open *open)
2274 {
2275         struct iattr iattr = {
2276                 .ia_valid = ATTR_SIZE,
2277                 .ia_size = 0,
2278         };
2279         if (!open->op_truncate)
2280                 return 0;
2281         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2282                 return nfserr_inval;
2283         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2284 }
2285
2286 static __be32
2287 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2288 {
2289         struct file *filp = stp->st_vfs_file;
2290         struct inode *inode = filp->f_path.dentry->d_inode;
2291         unsigned int share_access, new_writer;
2292         __be32 status;
2293
2294         set_access(&share_access, stp->st_access_bmap);
2295         new_writer = (~share_access) & open->op_share_access
2296                         & NFS4_SHARE_ACCESS_WRITE;
2297
2298         if (new_writer) {
2299                 int err = get_write_access(inode);
2300                 if (err)
2301                         return nfserrno(err);
2302                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2303                 if (err)
2304                         return nfserrno(err);
2305                 file_take_write(filp);
2306         }
2307         status = nfsd4_truncate(rqstp, cur_fh, open);
2308         if (status) {
2309                 if (new_writer)
2310                         put_write_access(inode);
2311                 return status;
2312         }
2313         /* remember the open */
2314         filp->f_mode |= open->op_share_access;
2315         __set_bit(open->op_share_access, &stp->st_access_bmap);
2316         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2317
2318         return nfs_ok;
2319 }
2320
2321
2322 static void
2323 nfs4_set_claim_prev(struct nfsd4_open *open)
2324 {
2325         open->op_stateowner->so_confirmed = 1;
2326         open->op_stateowner->so_client->cl_firststate = 1;
2327 }
2328
2329 /*
2330  * Attempt to hand out a delegation.
2331  */
2332 static void
2333 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2334 {
2335         struct nfs4_delegation *dp;
2336         struct nfs4_stateowner *sop = stp->st_stateowner;
2337         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2338         struct file_lock fl, *flp = &fl;
2339         int status, flag = 0;
2340
2341         flag = NFS4_OPEN_DELEGATE_NONE;
2342         open->op_recall = 0;
2343         switch (open->op_claim_type) {
2344                 case NFS4_OPEN_CLAIM_PREVIOUS:
2345                         if (!atomic_read(&cb->cb_set))
2346                                 open->op_recall = 1;
2347                         flag = open->op_delegate_type;
2348                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2349                                 goto out;
2350                         break;
2351                 case NFS4_OPEN_CLAIM_NULL:
2352                         /* Let's not give out any delegations till everyone's
2353                          * had the chance to reclaim theirs.... */
2354                         if (locks_in_grace())
2355                                 goto out;
2356                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2357                                 goto out;
2358                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2359                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2360                         else
2361                                 flag = NFS4_OPEN_DELEGATE_READ;
2362                         break;
2363                 default:
2364                         goto out;
2365         }
2366
2367         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2368         if (dp == NULL) {
2369                 flag = NFS4_OPEN_DELEGATE_NONE;
2370                 goto out;
2371         }
2372         locks_init_lock(&fl);
2373         fl.fl_lmops = &nfsd_lease_mng_ops;
2374         fl.fl_flags = FL_LEASE;
2375         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2376         fl.fl_end = OFFSET_MAX;
2377         fl.fl_owner =  (fl_owner_t)dp;
2378         fl.fl_file = stp->st_vfs_file;
2379         fl.fl_pid = current->tgid;
2380
2381         /* vfs_setlease checks to see if delegation should be handed out.
2382          * the lock_manager callbacks fl_mylease and fl_change are used
2383          */
2384         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2385                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2386                 unhash_delegation(dp);
2387                 flag = NFS4_OPEN_DELEGATE_NONE;
2388                 goto out;
2389         }
2390
2391         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2392
2393         dprintk("NFSD: delegation stateid=(%08x/%08x/%08x/%08x)\n\n",
2394                      dp->dl_stateid.si_boot,
2395                      dp->dl_stateid.si_stateownerid,
2396                      dp->dl_stateid.si_fileid,
2397                      dp->dl_stateid.si_generation);
2398 out:
2399         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2400                         && flag == NFS4_OPEN_DELEGATE_NONE
2401                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2402                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2403         open->op_delegate_type = flag;
2404 }
2405
2406 /*
2407  * called with nfs4_lock_state() held.
2408  */
2409 __be32
2410 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2411 {
2412         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2413         struct nfs4_file *fp = NULL;
2414         struct inode *ino = current_fh->fh_dentry->d_inode;
2415         struct nfs4_stateid *stp = NULL;
2416         struct nfs4_delegation *dp = NULL;
2417         __be32 status;
2418
2419         status = nfserr_inval;
2420         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2421                         || !deny_valid(open->op_share_deny))
2422                 goto out;
2423         /*
2424          * Lookup file; if found, lookup stateid and check open request,
2425          * and check for delegations in the process of being recalled.
2426          * If not found, create the nfs4_file struct
2427          */
2428         fp = find_file(ino);
2429         if (fp) {
2430                 if ((status = nfs4_check_open(fp, open, &stp)))
2431                         goto out;
2432                 status = nfs4_check_deleg(fp, open, &dp);
2433                 if (status)
2434                         goto out;
2435         } else {
2436                 status = nfserr_bad_stateid;
2437                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2438                         goto out;
2439                 status = nfserr_resource;
2440                 fp = alloc_init_file(ino);
2441                 if (fp == NULL)
2442                         goto out;
2443         }
2444
2445         /*
2446          * OPEN the file, or upgrade an existing OPEN.
2447          * If truncate fails, the OPEN fails.
2448          */
2449         if (stp) {
2450                 /* Stateid was found, this is an OPEN upgrade */
2451                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2452                 if (status)
2453                         goto out;
2454                 update_stateid(&stp->st_stateid);
2455         } else {
2456                 /* Stateid was not found, this is a new OPEN */
2457                 int flags = 0;
2458                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2459                         flags |= NFSD_MAY_READ;
2460                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2461                         flags |= NFSD_MAY_WRITE;
2462                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2463                 if (status)
2464                         goto out;
2465                 init_stateid(stp, fp, open);
2466                 status = nfsd4_truncate(rqstp, current_fh, open);
2467                 if (status) {
2468                         release_open_stateid(stp);
2469                         goto out;
2470                 }
2471                 if (nfsd4_has_session(&resp->cstate))
2472                         update_stateid(&stp->st_stateid);
2473         }
2474         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2475
2476         if (nfsd4_has_session(&resp->cstate))
2477                 open->op_stateowner->so_confirmed = 1;
2478
2479         /*
2480         * Attempt to hand out a delegation. No error return, because the
2481         * OPEN succeeds even if we fail.
2482         */
2483         nfs4_open_delegation(current_fh, open, stp);
2484
2485         status = nfs_ok;
2486
2487         dprintk("nfs4_process_open2: stateid=(%08x/%08x/%08x/%08x)\n",
2488                     stp->st_stateid.si_boot, stp->st_stateid.si_stateownerid,
2489                     stp->st_stateid.si_fileid, stp->st_stateid.si_generation);
2490 out:
2491         if (fp)
2492                 put_nfs4_file(fp);
2493         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2494                 nfs4_set_claim_prev(open);
2495         /*
2496         * To finish the open response, we just need to set the rflags.
2497         */
2498         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2499         if (!open->op_stateowner->so_confirmed &&
2500             !nfsd4_has_session(&resp->cstate))
2501                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2502
2503         return status;
2504 }
2505
2506 __be32
2507 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2508             clientid_t *clid)
2509 {
2510         struct nfs4_client *clp;
2511         __be32 status;
2512
2513         nfs4_lock_state();
2514         dprintk("process_renew(%08x/%08x): starting\n", 
2515                         clid->cl_boot, clid->cl_id);
2516         status = nfserr_stale_clientid;
2517         if (STALE_CLIENTID(clid))
2518                 goto out;
2519         clp = find_confirmed_client(clid);
2520         status = nfserr_expired;
2521         if (clp == NULL) {
2522                 /* We assume the client took too long to RENEW. */
2523                 dprintk("nfsd4_renew: clientid not found!\n");
2524                 goto out;
2525         }
2526         renew_client(clp);
2527         status = nfserr_cb_path_down;
2528         if (!list_empty(&clp->cl_delegations)
2529                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2530                 goto out;
2531         status = nfs_ok;
2532 out:
2533         nfs4_unlock_state();
2534         return status;
2535 }
2536
2537 struct lock_manager nfsd4_manager = {
2538 };
2539
2540 static void
2541 nfsd4_end_grace(void)
2542 {
2543         dprintk("NFSD: end of grace period\n");
2544         nfsd4_recdir_purge_old();
2545         locks_end_grace(&nfsd4_manager);
2546 }
2547
2548 static time_t
2549 nfs4_laundromat(void)
2550 {
2551         struct nfs4_client *clp;
2552         struct nfs4_stateowner *sop;
2553         struct nfs4_delegation *dp;
2554         struct list_head *pos, *next, reaplist;
2555         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2556         time_t t, clientid_val = NFSD_LEASE_TIME;
2557         time_t u, test_val = NFSD_LEASE_TIME;
2558
2559         nfs4_lock_state();
2560
2561         dprintk("NFSD: laundromat service - starting\n");
2562         if (locks_in_grace())
2563                 nfsd4_end_grace();
2564         list_for_each_safe(pos, next, &client_lru) {
2565                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2566                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2567                         t = clp->cl_time - cutoff;
2568                         if (clientid_val > t)
2569                                 clientid_val = t;
2570                         break;
2571                 }
2572                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2573                         clp->cl_clientid.cl_id);
2574                 nfsd4_remove_clid_dir(clp);
2575                 expire_client(clp);
2576         }
2577         INIT_LIST_HEAD(&reaplist);
2578         spin_lock(&recall_lock);
2579         list_for_each_safe(pos, next, &del_recall_lru) {
2580                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2581                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2582                         u = dp->dl_time - cutoff;
2583                         if (test_val > u)
2584                                 test_val = u;
2585                         break;
2586                 }
2587                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2588                                     dp, dp->dl_flock);
2589                 list_move(&dp->dl_recall_lru, &reaplist);
2590         }
2591         spin_unlock(&recall_lock);
2592         list_for_each_safe(pos, next, &reaplist) {
2593                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2594                 list_del_init(&dp->dl_recall_lru);
2595                 unhash_delegation(dp);
2596         }
2597         test_val = NFSD_LEASE_TIME;
2598         list_for_each_safe(pos, next, &close_lru) {
2599                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2600                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2601                         u = sop->so_time - cutoff;
2602                         if (test_val > u)
2603                                 test_val = u;
2604                         break;
2605                 }
2606                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2607                         sop->so_id);
2608                 release_openowner(sop);
2609         }
2610         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2611                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2612         nfs4_unlock_state();
2613         return clientid_val;
2614 }
2615
2616 static struct workqueue_struct *laundry_wq;
2617 static void laundromat_main(struct work_struct *);
2618 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2619
2620 static void
2621 laundromat_main(struct work_struct *not_used)
2622 {
2623         time_t t;
2624
2625         t = nfs4_laundromat();
2626         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2627         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2628 }
2629
2630 static struct nfs4_stateowner *
2631 search_close_lru(u32 st_id, int flags)
2632 {
2633         struct nfs4_stateowner *local = NULL;
2634
2635         if (flags & CLOSE_STATE) {
2636                 list_for_each_entry(local, &close_lru, so_close_lru) {
2637                         if (local->so_id == st_id)
2638                                 return local;
2639                 }
2640         }
2641         return NULL;
2642 }
2643
2644 static inline int
2645 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2646 {
2647         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2648 }
2649
2650 static int
2651 STALE_STATEID(stateid_t *stateid)
2652 {
2653         if (time_after((unsigned long)boot_time,
2654                         (unsigned long)stateid->si_boot)) {
2655                 dprintk("NFSD: stale stateid (%08x/%08x/%08x/%08x)!\n",
2656                         stateid->si_boot, stateid->si_stateownerid,
2657                         stateid->si_fileid, stateid->si_generation);
2658                 return 1;
2659         }
2660         return 0;
2661 }
2662
2663 static int
2664 EXPIRED_STATEID(stateid_t *stateid)
2665 {
2666         if (time_before((unsigned long)boot_time,
2667                         ((unsigned long)stateid->si_boot)) &&
2668             time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2669                 dprintk("NFSD: expired stateid (%08x/%08x/%08x/%08x)!\n",
2670                         stateid->si_boot, stateid->si_stateownerid,
2671                         stateid->si_fileid, stateid->si_generation);
2672                 return 1;
2673         }
2674         return 0;
2675 }
2676
2677 static __be32
2678 stateid_error_map(stateid_t *stateid)
2679 {
2680         if (STALE_STATEID(stateid))
2681                 return nfserr_stale_stateid;
2682         if (EXPIRED_STATEID(stateid))
2683                 return nfserr_expired;
2684
2685         dprintk("NFSD: bad stateid (%08x/%08x/%08x/%08x)!\n",
2686                 stateid->si_boot, stateid->si_stateownerid,
2687                 stateid->si_fileid, stateid->si_generation);
2688         return nfserr_bad_stateid;
2689 }
2690
2691 static inline int
2692 access_permit_read(unsigned long access_bmap)
2693 {
2694         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2695                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2696                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2697 }
2698
2699 static inline int
2700 access_permit_write(unsigned long access_bmap)
2701 {
2702         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2703                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2704 }
2705
2706 static
2707 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2708 {
2709         __be32 status = nfserr_openmode;
2710
2711         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2712                 goto out;
2713         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2714                 goto out;
2715         status = nfs_ok;
2716 out:
2717         return status;
2718 }
2719
2720 static inline __be32
2721 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2722 {
2723         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2724                 return nfs_ok;
2725         else if (locks_in_grace()) {
2726                 /* Answer in remaining cases depends on existance of
2727                  * conflicting state; so we must wait out the grace period. */
2728                 return nfserr_grace;
2729         } else if (flags & WR_STATE)
2730                 return nfs4_share_conflict(current_fh,
2731                                 NFS4_SHARE_DENY_WRITE);
2732         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2733                 return nfs4_share_conflict(current_fh,
2734                                 NFS4_SHARE_DENY_READ);
2735 }
2736
2737 /*
2738  * Allow READ/WRITE during grace period on recovered state only for files
2739  * that are not able to provide mandatory locking.
2740  */
2741 static inline int
2742 grace_disallows_io(struct inode *inode)
2743 {
2744         return locks_in_grace() && mandatory_lock(inode);
2745 }
2746
2747 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2748 {
2749         /*
2750          * When sessions are used the stateid generation number is ignored
2751          * when it is zero.
2752          */
2753         if ((flags & HAS_SESSION) && in->si_generation == 0)
2754                 goto out;
2755
2756         /* If the client sends us a stateid from the future, it's buggy: */
2757         if (in->si_generation > ref->si_generation)
2758                 return nfserr_bad_stateid;
2759         /*
2760          * The following, however, can happen.  For example, if the
2761          * client sends an open and some IO at the same time, the open
2762          * may bump si_generation while the IO is still in flight.
2763          * Thanks to hard links and renames, the client never knows what
2764          * file an open will affect.  So it could avoid that situation
2765          * only by serializing all opens and IO from the same open
2766          * owner.  To recover from the old_stateid error, the client
2767          * will just have to retry the IO:
2768          */
2769         if (in->si_generation < ref->si_generation)
2770                 return nfserr_old_stateid;
2771 out:
2772         return nfs_ok;
2773 }
2774
2775 static int is_delegation_stateid(stateid_t *stateid)
2776 {
2777         return stateid->si_fileid == 0;
2778 }
2779
2780 /*
2781 * Checks for stateid operations
2782 */
2783 __be32
2784 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2785                            stateid_t *stateid, int flags, struct file **filpp)
2786 {
2787         struct nfs4_stateid *stp = NULL;
2788         struct nfs4_delegation *dp = NULL;
2789         struct svc_fh *current_fh = &cstate->current_fh;
2790         struct inode *ino = current_fh->fh_dentry->d_inode;
2791         __be32 status;
2792
2793         if (filpp)
2794                 *filpp = NULL;
2795
2796         if (grace_disallows_io(ino))
2797                 return nfserr_grace;
2798
2799         if (nfsd4_has_session(cstate))
2800                 flags |= HAS_SESSION;
2801
2802         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2803                 return check_special_stateids(current_fh, stateid, flags);
2804
2805         status = nfserr_stale_stateid;
2806         if (STALE_STATEID(stateid)) 
2807                 goto out;
2808
2809         status = nfserr_bad_stateid;
2810         if (is_delegation_stateid(stateid)) {
2811                 dp = find_delegation_stateid(ino, stateid);
2812                 if (!dp) {
2813                         status = stateid_error_map(stateid);
2814                         goto out;
2815                 }
2816                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2817                                                   flags);
2818                 if (status)
2819                         goto out;
2820                 status = nfs4_check_delegmode(dp, flags);
2821                 if (status)
2822                         goto out;
2823                 renew_client(dp->dl_client);
2824                 if (filpp)
2825                         *filpp = dp->dl_vfs_file;
2826         } else { /* open or lock stateid */
2827                 stp = find_stateid(stateid, flags);
2828                 if (!stp) {
2829                         status = stateid_error_map(stateid);
2830                         goto out;
2831                 }
2832                 if (nfs4_check_fh(current_fh, stp))
2833                         goto out;
2834                 if (!stp->st_stateowner->so_confirmed)
2835                         goto out;
2836                 status = check_stateid_generation(stateid, &stp->st_stateid,
2837                                                   flags);
2838                 if (status)
2839                         goto out;
2840                 status = nfs4_check_openmode(stp, flags);
2841                 if (status)
2842                         goto out;
2843                 renew_client(stp->st_stateowner->so_client);
2844                 if (filpp)
2845                         *filpp = stp->st_vfs_file;
2846         }
2847         status = nfs_ok;
2848 out:
2849         return status;
2850 }
2851
2852 static inline int
2853 setlkflg (int type)
2854 {
2855         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2856                 RD_STATE : WR_STATE;
2857 }
2858
2859 /* 
2860  * Checks for sequence id mutating operations. 
2861  */
2862 static __be32
2863 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2864                          stateid_t *stateid, int flags,
2865                          struct nfs4_stateowner **sopp,
2866                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2867 {
2868         struct nfs4_stateid *stp;
2869         struct nfs4_stateowner *sop;
2870         struct svc_fh *current_fh = &cstate->current_fh;
2871         __be32 status;
2872
2873         dprintk("NFSD: preprocess_seqid_op: seqid=%d " 
2874                         "stateid = (%08x/%08x/%08x/%08x)\n", seqid,
2875                 stateid->si_boot, stateid->si_stateownerid, stateid->si_fileid,
2876                 stateid->si_generation);
2877
2878         *stpp = NULL;
2879         *sopp = NULL;
2880
2881         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2882                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2883                 return nfserr_bad_stateid;
2884         }
2885
2886         if (STALE_STATEID(stateid))
2887                 return nfserr_stale_stateid;
2888
2889         if (nfsd4_has_session(cstate))
2890                 flags |= HAS_SESSION;
2891
2892         /*
2893         * We return BAD_STATEID if filehandle doesn't match stateid, 
2894         * the confirmed flag is incorrecly set, or the generation 
2895         * number is incorrect.  
2896         */
2897         stp = find_stateid(stateid, flags);
2898         if (stp == NULL) {
2899                 /*
2900                  * Also, we should make sure this isn't just the result of
2901                  * a replayed close:
2902                  */
2903                 sop = search_close_lru(stateid->si_stateownerid, flags);
2904                 if (sop == NULL)
2905                         return stateid_error_map(stateid);
2906                 *sopp = sop;
2907                 goto check_replay;
2908         }
2909
2910         *stpp = stp;
2911         *sopp = sop = stp->st_stateowner;
2912
2913         if (lock) {
2914                 clientid_t *lockclid = &lock->v.new.clientid;
2915                 struct nfs4_client *clp = sop->so_client;
2916                 int lkflg = 0;
2917                 __be32 status;
2918
2919                 lkflg = setlkflg(lock->lk_type);
2920
2921                 if (lock->lk_is_new) {
2922                         if (!sop->so_is_open_owner)
2923                                 return nfserr_bad_stateid;
2924                         if (!(flags & HAS_SESSION) &&
2925                             !same_clid(&clp->cl_clientid, lockclid))
2926                                 return nfserr_bad_stateid;
2927                         /* stp is the open stateid */
2928                         status = nfs4_check_openmode(stp, lkflg);
2929                         if (status)
2930                                 return status;
2931                 } else {
2932                         /* stp is the lock stateid */
2933                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2934                         if (status)
2935                                 return status;
2936                }
2937         }
2938
2939         if (nfs4_check_fh(current_fh, stp)) {
2940                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2941                 return nfserr_bad_stateid;
2942         }
2943
2944         /*
2945         *  We now validate the seqid and stateid generation numbers.
2946         *  For the moment, we ignore the possibility of 
2947         *  generation number wraparound.
2948         */
2949         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2950                 goto check_replay;
2951
2952         if (sop->so_confirmed && flags & CONFIRM) {
2953                 dprintk("NFSD: preprocess_seqid_op: expected"
2954                                 " unconfirmed stateowner!\n");
2955                 return nfserr_bad_stateid;
2956         }
2957         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2958                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2959                                 " confirmed yet!\n");
2960                 return nfserr_bad_stateid;
2961         }
2962         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2963         if (status)
2964                 return status;
2965         renew_client(sop->so_client);
2966         return nfs_ok;
2967
2968 check_replay:
2969         if (seqid == sop->so_seqid - 1) {
2970                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2971                 /* indicate replay to calling function */
2972                 return nfserr_replay_me;
2973         }
2974         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2975                         sop->so_seqid, seqid);
2976         *sopp = NULL;
2977         return nfserr_bad_seqid;
2978 }
2979
2980 __be32
2981 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2982                    struct nfsd4_open_confirm *oc)
2983 {
2984         __be32 status;
2985         struct nfs4_stateowner *sop;
2986         struct nfs4_stateid *stp;
2987
2988         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2989                         (int)cstate->current_fh.fh_dentry->d_name.len,
2990                         cstate->current_fh.fh_dentry->d_name.name);
2991
2992         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2993         if (status)
2994                 return status;
2995
2996         nfs4_lock_state();
2997
2998         if ((status = nfs4_preprocess_seqid_op(cstate,
2999                                         oc->oc_seqid, &oc->oc_req_stateid,
3000                                         CONFIRM | OPEN_STATE,
3001                                         &oc->oc_stateowner, &stp, NULL)))
3002                 goto out; 
3003
3004         sop = oc->oc_stateowner;
3005         sop->so_confirmed = 1;
3006         update_stateid(&stp->st_stateid);
3007         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3008         dprintk("NFSD: nfsd4_open_confirm: success, seqid=%d " 
3009                 "stateid=(%08x/%08x/%08x/%08x)\n", oc->oc_seqid,
3010                          stp->st_stateid.si_boot,
3011                          stp->st_stateid.si_stateownerid,
3012                          stp->st_stateid.si_fileid,
3013                          stp->st_stateid.si_generation);
3014
3015         nfsd4_create_clid_dir(sop->so_client);
3016 out:
3017         if (oc->oc_stateowner) {
3018                 nfs4_get_stateowner(oc->oc_stateowner);
3019                 cstate->replay_owner = oc->oc_stateowner;
3020         }
3021         nfs4_unlock_state();
3022         return status;
3023 }
3024
3025
3026 /*
3027  * unset all bits in union bitmap (bmap) that
3028  * do not exist in share (from successful OPEN_DOWNGRADE)
3029  */
3030 static void
3031 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3032 {
3033         int i;
3034         for (i = 1; i < 4; i++) {
3035                 if ((i & access) != i)
3036                         __clear_bit(i, bmap);
3037         }
3038 }
3039
3040 static void
3041 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3042 {
3043         int i;
3044         for (i = 0; i < 4; i++) {
3045                 if ((i & deny) != i)
3046                         __clear_bit(i, bmap);
3047         }
3048 }
3049
3050 __be32
3051 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3052                      struct nfsd4_compound_state *cstate,
3053                      struct nfsd4_open_downgrade *od)
3054 {
3055         __be32 status;
3056         struct nfs4_stateid *stp;
3057         unsigned int share_access;
3058
3059         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3060                         (int)cstate->current_fh.fh_dentry->d_name.len,
3061                         cstate->current_fh.fh_dentry->d_name.name);
3062
3063         if (!access_valid(od->od_share_access, cstate->minorversion)
3064                         || !deny_valid(od->od_share_deny))
3065                 return nfserr_inval;
3066
3067         nfs4_lock_state();
3068         if ((status = nfs4_preprocess_seqid_op(cstate,
3069                                         od->od_seqid,
3070                                         &od->od_stateid, 
3071                                         OPEN_STATE,
3072                                         &od->od_stateowner, &stp, NULL)))
3073                 goto out; 
3074
3075         status = nfserr_inval;
3076         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3077                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3078                         stp->st_access_bmap, od->od_share_access);
3079                 goto out;
3080         }
3081         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3082                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3083                         stp->st_deny_bmap, od->od_share_deny);
3084                 goto out;
3085         }
3086         set_access(&share_access, stp->st_access_bmap);
3087         nfs4_file_downgrade(stp->st_vfs_file,
3088                             share_access & ~od->od_share_access);
3089
3090         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3091         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3092
3093         update_stateid(&stp->st_stateid);
3094         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3095         status = nfs_ok;
3096 out:
3097         if (od->od_stateowner) {
3098                 nfs4_get_stateowner(od->od_stateowner);
3099                 cstate->replay_owner = od->od_stateowner;
3100         }
3101         nfs4_unlock_state();
3102         return status;
3103 }
3104
3105 /*
3106  * nfs4_unlock_state() called after encode
3107  */
3108 __be32
3109 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3110             struct nfsd4_close *close)
3111 {
3112         __be32 status;
3113         struct nfs4_stateid *stp;
3114
3115         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3116                         (int)cstate->current_fh.fh_dentry->d_name.len,
3117                         cstate->current_fh.fh_dentry->d_name.name);
3118
3119         nfs4_lock_state();
3120         /* check close_lru for replay */
3121         if ((status = nfs4_preprocess_seqid_op(cstate,
3122                                         close->cl_seqid,
3123                                         &close->cl_stateid, 
3124                                         OPEN_STATE | CLOSE_STATE,
3125                                         &close->cl_stateowner, &stp, NULL)))
3126                 goto out; 
3127         status = nfs_ok;
3128         update_stateid(&stp->st_stateid);
3129         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3130
3131         /* release_stateid() calls nfsd_close() if needed */
3132         release_open_stateid(stp);
3133
3134         /* place unused nfs4_stateowners on so_close_lru list to be
3135          * released by the laundromat service after the lease period
3136          * to enable us to handle CLOSE replay
3137          */
3138         if (list_empty(&close->cl_stateowner->so_stateids))
3139                 move_to_close_lru(close->cl_stateowner);
3140 out:
3141         if (close->cl_stateowner) {
3142                 nfs4_get_stateowner(close->cl_stateowner);
3143                 cstate->replay_owner = close->cl_stateowner;
3144         }
3145         nfs4_unlock_state();
3146         return status;
3147 }
3148
3149 __be32
3150 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3151                   struct nfsd4_delegreturn *dr)
3152 {
3153         struct nfs4_delegation *dp;
3154         stateid_t *stateid = &dr->dr_stateid;
3155         struct inode *inode;
3156         __be32 status;
3157         int flags = 0;
3158
3159         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3160                 return status;
3161         inode = cstate->current_fh.fh_dentry->d_inode;
3162
3163         if (nfsd4_has_session(cstate))
3164                 flags |= HAS_SESSION;
3165         nfs4_lock_state();
3166         status = nfserr_bad_stateid;
3167         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3168                 goto out;
3169         status = nfserr_stale_stateid;
3170         if (STALE_STATEID(stateid))
3171                 goto out;
3172         status = nfserr_bad_stateid;
3173         if (!is_delegation_stateid(stateid))
3174                 goto out;
3175         dp = find_delegation_stateid(inode, stateid);
3176         if (!dp) {
3177                 status = stateid_error_map(stateid);
3178                 goto out;
3179         }
3180         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3181         if (status)
3182                 goto out;
3183         renew_client(dp->dl_client);
3184
3185         unhash_delegation(dp);
3186 out:
3187         nfs4_unlock_state();
3188
3189         return status;
3190 }
3191
3192
3193 /* 
3194  * Lock owner state (byte-range locks)
3195  */
3196 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3197 #define LOCK_HASH_BITS              8
3198 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3199 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3200
3201 static inline u64
3202 end_offset(u64 start, u64 len)
3203 {
3204         u64 end;
3205
3206         end = start + len;
3207         return end >= start ? end: NFS4_MAX_UINT64;
3208 }
3209
3210 /* last octet in a range */
3211 static inline u64
3212 last_byte_offset(u64 start, u64 len)
3213 {
3214         u64 end;
3215
3216         BUG_ON(!len);
3217         end = start + len;
3218         return end > start ? end - 1: NFS4_MAX_UINT64;
3219 }
3220
3221 #define lockownerid_hashval(id) \
3222         ((id) & LOCK_HASH_MASK)
3223
3224 static inline unsigned int
3225 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3226                 struct xdr_netobj *ownername)
3227 {
3228         return (file_hashval(inode) + cl_id
3229                         + opaque_hashval(ownername->data, ownername->len))
3230                 & LOCK_HASH_MASK;
3231 }
3232
3233 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3234 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3235 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3236
3237 static struct nfs4_stateid *
3238 find_stateid(stateid_t *stid, int flags)
3239 {
3240         struct nfs4_stateid *local;
3241         u32 st_id = stid->si_stateownerid;
3242         u32 f_id = stid->si_fileid;
3243         unsigned int hashval;
3244
3245         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3246         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3247                 hashval = stateid_hashval(st_id, f_id);
3248                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3249                         if ((local->st_stateid.si_stateownerid == st_id) &&
3250                             (local->st_stateid.si_fileid == f_id))
3251                                 return local;
3252                 }
3253         } 
3254
3255         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3256                 hashval = stateid_hashval(st_id, f_id);
3257                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3258                         if ((local->st_stateid.si_stateownerid == st_id) &&
3259                             (local->st_stateid.si_fileid == f_id))
3260                                 return local;
3261                 }
3262         }
3263         return NULL;
3264 }
3265
3266 static struct nfs4_delegation *
3267 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3268 {
3269         struct nfs4_file *fp;
3270         struct nfs4_delegation *dl;
3271
3272         dprintk("NFSD:find_delegation_stateid stateid=(%08x/%08x/%08x/%08x)\n",
3273                     stid->si_boot, stid->si_stateownerid,
3274                     stid->si_fileid, stid->si_generation);
3275
3276         fp = find_file(ino);
3277         if (!fp)
3278                 return NULL;
3279         dl = find_delegation_file(fp, stid);
3280         put_nfs4_file(fp);
3281         return dl;
3282 }
3283
3284 /*
3285  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3286  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3287  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3288  * locking, this prevents us from being completely protocol-compliant.  The
3289  * real solution to this problem is to start using unsigned file offsets in
3290  * the VFS, but this is a very deep change!
3291  */
3292 static inline void
3293 nfs4_transform_lock_offset(struct file_lock *lock)
3294 {
3295         if (lock->fl_start < 0)
3296                 lock->fl_start = OFFSET_MAX;
3297         if (lock->fl_end < 0)
3298                 lock->fl_end = OFFSET_MAX;
3299 }
3300
3301 /* Hack!: For now, we're defining this just so we can use a pointer to it
3302  * as a unique cookie to identify our (NFSv4's) posix locks. */
3303 static struct lock_manager_operations nfsd_posix_mng_ops  = {
3304 };
3305
3306 static inline void
3307 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3308 {
3309         struct nfs4_stateowner *sop;
3310         unsigned int hval;
3311
3312         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3313                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3314                 hval = lockownerid_hashval(sop->so_id);
3315                 kref_get(&sop->so_ref);
3316                 deny->ld_sop = sop;
3317                 deny->ld_clientid = sop->so_client->cl_clientid;
3318         } else {
3319                 deny->ld_sop = NULL;
3320                 deny->ld_clientid.cl_boot = 0;
3321                 deny->ld_clientid.cl_id = 0;
3322         }
3323         deny->ld_start = fl->fl_start;
3324         deny->ld_length = NFS4_MAX_UINT64;
3325         if (fl->fl_end != NFS4_MAX_UINT64)
3326                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3327         deny->ld_type = NFS4_READ_LT;
3328         if (fl->fl_type != F_RDLCK)
3329                 deny->ld_type = NFS4_WRITE_LT;
3330 }
3331
3332 static struct nfs4_stateowner *
3333 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3334                 struct xdr_netobj *owner)
3335 {
3336         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3337         struct nfs4_stateowner *op;
3338
3339         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3340                 if (same_owner_str(op, owner, clid))
3341                         return op;
3342         }
3343         return NULL;
3344 }
3345
3346 /*
3347  * Alloc a lock owner structure.
3348  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3349  * occured. 
3350  *
3351  * strhashval = lock_ownerstr_hashval 
3352  */
3353
3354 static struct nfs4_stateowner *
3355 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3356         struct nfs4_stateowner *sop;
3357         struct nfs4_replay *rp;
3358         unsigned int idhashval;
3359
3360         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3361                 return NULL;
3362         idhashval = lockownerid_hashval(current_ownerid);
3363         INIT_LIST_HEAD(&sop->so_idhash);
3364         INIT_LIST_HEAD(&sop->so_strhash);
3365         INIT_LIST_HEAD(&sop->so_perclient);
3366         INIT_LIST_HEAD(&sop->so_stateids);
3367         INIT_LIST_HEAD(&sop->so_perstateid);
3368         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3369         sop->so_time = 0;
3370         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3371         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3372         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3373         sop->so_is_open_owner = 0;
3374         sop->so_id = current_ownerid++;
3375         sop->so_client = clp;
3376         /* It is the openowner seqid that will be incremented in encode in the
3377          * case of new lockowners; so increment the lock seqid manually: */
3378         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3379         sop->so_confirmed = 1;
3380         rp = &sop->so_replay;
3381         rp->rp_status = nfserr_serverfault;
3382         rp->rp_buflen = 0;
3383         rp->rp_buf = rp->rp_ibuf;
3384         return sop;
3385 }
3386
3387 static struct nfs4_stateid *
3388 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3389 {
3390         struct nfs4_stateid *stp;
3391         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3392
3393         stp = nfs4_alloc_stateid();
3394         if (stp == NULL)
3395                 goto out;
3396         INIT_LIST_HEAD(&stp->st_hash);
3397         INIT_LIST_HEAD(&stp->st_perfile);
3398         INIT_LIST_HEAD(&stp->st_perstateowner);
3399         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3400         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3401         list_add(&stp->st_perfile, &fp->fi_stateids);
3402         list_add(&stp->st_perstateowner, &sop->so_stateids);
3403         stp->st_stateowner = sop;
3404         get_nfs4_file(fp);
3405         stp->st_file = fp;
3406         stp->st_stateid.si_boot = get_seconds();
3407         stp->st_stateid.si_stateownerid = sop->so_id;
3408         stp->st_stateid.si_fileid = fp->fi_id;
3409         stp->st_stateid.si_generation = 0;
3410         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3411         stp->st_access_bmap = open_stp->st_access_bmap;
3412         stp->st_deny_bmap = open_stp->st_deny_bmap;
3413         stp->st_openstp = open_stp;
3414
3415 out:
3416         return stp;
3417 }
3418
3419 static int
3420 check_lock_length(u64 offset, u64 length)
3421 {
3422         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3423              LOFF_OVERFLOW(offset, length)));
3424 }
3425
3426 /*
3427  *  LOCK operation 
3428  */
3429 __be32
3430 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3431            struct nfsd4_lock *lock)
3432 {
3433         struct nfs4_stateowner *open_sop = NULL;
3434         struct nfs4_stateowner *lock_sop = NULL;
3435         struct nfs4_stateid *lock_stp;
3436         struct file *filp;
3437         struct file_lock file_lock;
3438         struct file_lock conflock;
3439         __be32 status = 0;
3440         unsigned int strhashval;
3441         unsigned int cmd;
3442         int err;
3443
3444         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3445                 (long long) lock->lk_offset,
3446                 (long long) lock->lk_length);
3447
3448         if (check_lock_length(lock->lk_offset, lock->lk_length))
3449                  return nfserr_inval;
3450
3451         if ((status = fh_verify(rqstp, &cstate->current_fh,
3452                                 S_IFREG, NFSD_MAY_LOCK))) {
3453                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3454                 return status;
3455         }
3456
3457         nfs4_lock_state();
3458
3459         if (lock->lk_is_new) {
3460                 /*
3461                  * Client indicates that this is a new lockowner.
3462                  * Use open owner and open stateid to create lock owner and
3463                  * lock stateid.
3464                  */
3465                 struct nfs4_stateid *open_stp = NULL;
3466                 struct nfs4_file *fp;
3467                 
3468                 status = nfserr_stale_clientid;
3469                 if (!nfsd4_has_session(cstate) &&
3470                     STALE_CLIENTID(&lock->lk_new_clientid))
3471                         goto out;
3472
3473                 /* validate and update open stateid and open seqid */
3474                 status = nfs4_preprocess_seqid_op(cstate,
3475                                         lock->lk_new_open_seqid,
3476                                         &lock->lk_new_open_stateid,
3477                                         OPEN_STATE,
3478                                         &lock->lk_replay_owner, &open_stp,
3479                                         lock);
3480                 if (status)
3481                         goto out;
3482                 open_sop = lock->lk_replay_owner;
3483                 /* create lockowner and lock stateid */
3484                 fp = open_stp->st_file;
3485                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3486                                 open_sop->so_client->cl_clientid.cl_id, 
3487                                 &lock->v.new.owner);
3488                 /* XXX: Do we need to check for duplicate stateowners on
3489                  * the same file, or should they just be allowed (and
3490                  * create new stateids)? */
3491                 status = nfserr_resource;
3492                 lock_sop = alloc_init_lock_stateowner(strhashval,
3493                                 open_sop->so_client, open_stp, lock);
3494                 if (lock_sop == NULL)
3495                         goto out;
3496                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3497                 if (lock_stp == NULL)
3498                         goto out;
3499         } else {
3500                 /* lock (lock owner + lock stateid) already exists */
3501                 status = nfs4_preprocess_seqid_op(cstate,
3502                                        lock->lk_old_lock_seqid, 
3503                                        &lock->lk_old_lock_stateid, 
3504                                        LOCK_STATE,
3505                                        &lock->lk_replay_owner, &lock_stp, lock);
3506                 if (status)
3507                         goto out;
3508                 lock_sop = lock->lk_replay_owner;
3509         }
3510         /* lock->lk_replay_owner and lock_stp have been created or found */
3511         filp = lock_stp->st_vfs_file;
3512
3513         status = nfserr_grace;
3514         if (locks_in_grace() && !lock->lk_reclaim)
3515                 goto out;
3516         status = nfserr_no_grace;
3517         if (!locks_in_grace() && lock->lk_reclaim)
3518                 goto out;
3519
3520         locks_init_lock(&file_lock);
3521         switch (lock->lk_type) {
3522                 case NFS4_READ_LT:
3523                 case NFS4_READW_LT:
3524                         file_lock.fl_type = F_RDLCK;
3525                         cmd = F_SETLK;
3526                 break;
3527                 case NFS4_WRITE_LT:
3528                 case NFS4_WRITEW_LT:
3529                         file_lock.fl_type = F_WRLCK;
3530                         cmd = F_SETLK;
3531                 break;
3532                 default:
3533                         status = nfserr_inval;
3534                 goto out;
3535         }
3536         file_lock.fl_owner = (fl_owner_t)lock_sop;
3537         file_lock.fl_pid = current->tgid;
3538         file_lock.fl_file = filp;
3539         file_lock.fl_flags = FL_POSIX;
3540         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3541
3542         file_lock.fl_start = lock->lk_offset;
3543         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3544         nfs4_transform_lock_offset(&file_lock);
3545
3546         /*
3547         * Try to lock the file in the VFS.
3548         * Note: locks.c uses the BKL to protect the inode's lock list.
3549         */
3550
3551         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3552         switch (-err) {
3553         case 0: /* success! */
3554                 update_stateid(&lock_stp->st_stateid);
3555                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3556                                 sizeof(stateid_t));
3557                 status = 0;
3558                 break;
3559         case (EAGAIN):          /* conflock holds conflicting lock */
3560                 status = nfserr_denied;
3561                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3562                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3563                 break;
3564         case (EDEADLK):
3565                 status = nfserr_deadlock;
3566                 break;
3567         default:        
3568                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3569                 status = nfserr_resource;
3570                 break;
3571         }
3572 out:
3573         if (status && lock->lk_is_new && lock_sop)
3574                 release_lockowner(lock_sop);
3575         if (lock->lk_replay_owner) {
3576                 nfs4_get_stateowner(lock->lk_replay_owner);
3577                 cstate->replay_owner = lock->lk_replay_owner;
3578         }
3579         nfs4_unlock_state();
3580         return status;
3581 }
3582
3583 /*
3584  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3585  * so we do a temporary open here just to get an open file to pass to
3586  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3587  * inode operation.)
3588  */
3589 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3590 {
3591         struct file *file;
3592         int err;
3593
3594         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3595         if (err)
3596                 return err;
3597         err = vfs_test_lock(file, lock);
3598         nfsd_close(file);
3599         return err;
3600 }
3601
3602 /*
3603  * LOCKT operation
3604  */
3605 __be32
3606 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3607             struct nfsd4_lockt *lockt)
3608 {
3609         struct inode *inode;
3610         struct file_lock file_lock;
3611         int error;
3612         __be32 status;
3613
3614         if (locks_in_grace())
3615                 return nfserr_grace;
3616
3617         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3618                  return nfserr_inval;
3619
3620         lockt->lt_stateowner = NULL;
3621         nfs4_lock_state();
3622
3623         status = nfserr_stale_clientid;
3624         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3625                 goto out;
3626
3627         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3628                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3629                 if (status == nfserr_symlink)
3630                         status = nfserr_inval;
3631                 goto out;
3632         }
3633
3634         inode = cstate->current_fh.fh_dentry->d_inode;
3635         locks_init_lock(&file_lock);
3636         switch (lockt->lt_type) {
3637                 case NFS4_READ_LT:
3638                 case NFS4_READW_LT:
3639                         file_lock.fl_type = F_RDLCK;
3640                 break;
3641                 case NFS4_WRITE_LT:
3642                 case NFS4_WRITEW_LT:
3643                         file_lock.fl_type = F_WRLCK;
3644                 break;
3645                 default:
3646                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3647                         status = nfserr_inval;
3648                 goto out;
3649         }
3650
3651         lockt->lt_stateowner = find_lockstateowner_str(inode,
3652                         &lockt->lt_clientid, &lockt->lt_owner);
3653         if (lockt->lt_stateowner)
3654                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3655         file_lock.fl_pid = current->tgid;
3656         file_lock.fl_flags = FL_POSIX;
3657
3658         file_lock.fl_start = lockt->lt_offset;
3659         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3660
3661         nfs4_transform_lock_offset(&file_lock);
3662
3663         status = nfs_ok;
3664         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3665         if (error) {
3666                 status = nfserrno(error);
3667                 goto out;
3668         }
3669         if (file_lock.fl_type != F_UNLCK) {
3670                 status = nfserr_denied;
3671                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3672         }
3673 out:
3674         nfs4_unlock_state();
3675         return status;
3676 }
3677
3678 __be32
3679 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3680             struct nfsd4_locku *locku)
3681 {
3682         struct nfs4_stateid *stp;
3683         struct file *filp = NULL;
3684         struct file_lock file_lock;
3685         __be32 status;
3686         int err;
3687                                                         
3688         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3689                 (long long) locku->lu_offset,
3690                 (long long) locku->lu_length);
3691
3692         if (check_lock_length(locku->lu_offset, locku->lu_length))
3693                  return nfserr_inval;
3694
3695         nfs4_lock_state();
3696                                                                                 
3697         if ((status = nfs4_preprocess_seqid_op(cstate,
3698                                         locku->lu_seqid, 
3699                                         &locku->lu_stateid, 
3700                                         LOCK_STATE,
3701                                         &locku->lu_stateowner, &stp, NULL)))
3702                 goto out;
3703
3704         filp = stp->st_vfs_file;
3705         BUG_ON(!filp);
3706         locks_init_lock(&file_lock);
3707         file_lock.fl_type = F_UNLCK;
3708         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3709         file_lock.fl_pid = current->tgid;
3710         file_lock.fl_file = filp;
3711         file_lock.fl_flags = FL_POSIX; 
3712         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3713         file_lock.fl_start = locku->lu_offset;
3714
3715         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3716         nfs4_transform_lock_offset(&file_lock);
3717
3718         /*
3719         *  Try to unlock the file in the VFS.
3720         */
3721         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3722         if (err) {
3723                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3724                 goto out_nfserr;
3725         }
3726         /*
3727         * OK, unlock succeeded; the only thing left to do is update the stateid.
3728         */
3729         update_stateid(&stp->st_stateid);
3730         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3731
3732 out:
3733         if (locku->lu_stateowner) {
3734                 nfs4_get_stateowner(locku->lu_stateowner);
3735                 cstate->replay_owner = locku->lu_stateowner;
3736         }
3737         nfs4_unlock_state();
3738         return status;
3739
3740 out_nfserr:
3741         status = nfserrno(err);
3742         goto out;
3743 }
3744
3745 /*
3746  * returns
3747  *      1: locks held by lockowner
3748  *      0: no locks held by lockowner
3749  */
3750 static int
3751 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3752 {
3753         struct file_lock **flpp;
3754         struct inode *inode = filp->f_path.dentry->d_inode;
3755         int status = 0;
3756
3757         lock_kernel();
3758         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3759                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3760                         status = 1;
3761                         goto out;
3762                 }
3763         }
3764 out:
3765         unlock_kernel();
3766         return status;
3767 }
3768
3769 __be32
3770 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3771                         struct nfsd4_compound_state *cstate,
3772                         struct nfsd4_release_lockowner *rlockowner)
3773 {
3774         clientid_t *clid = &rlockowner->rl_clientid;
3775         struct nfs4_stateowner *sop;
3776         struct nfs4_stateid *stp;
3777         struct xdr_netobj *owner = &rlockowner->rl_owner;
3778         struct list_head matches;
3779         int i;
3780         __be32 status;
3781
3782         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3783                 clid->cl_boot, clid->cl_id);
3784
3785         /* XXX check for lease expiration */
3786
3787         status = nfserr_stale_clientid;
3788         if (STALE_CLIENTID(clid))
3789                 return status;
3790
3791         nfs4_lock_state();
3792
3793         status = nfserr_locks_held;
3794         /* XXX: we're doing a linear search through all the lockowners.
3795          * Yipes!  For now we'll just hope clients aren't really using
3796          * release_lockowner much, but eventually we have to fix these
3797          * data structures. */
3798         INIT_LIST_HEAD(&matches);
3799         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3800                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3801                         if (!same_owner_str(sop, owner, clid))
3802                                 continue;
3803                         list_for_each_entry(stp, &sop->so_stateids,
3804                                         st_perstateowner) {
3805                                 if (check_for_locks(stp->st_vfs_file, sop))
3806                                         goto out;
3807                                 /* Note: so_perclient unused for lockowners,
3808                                  * so it's OK to fool with here. */
3809                                 list_add(&sop->so_perclient, &matches);
3810                         }
3811                 }
3812         }
3813         /* Clients probably won't expect us to return with some (but not all)
3814          * of the lockowner state released; so don't release any until all
3815          * have been checked. */
3816         status = nfs_ok;
3817         while (!list_empty(&matches)) {
3818                 sop = list_entry(matches.next, struct nfs4_stateowner,
3819                                                                 so_perclient);
3820                 /* unhash_stateowner deletes so_perclient only
3821                  * for openowners. */
3822                 list_del(&sop->so_perclient);
3823                 release_lockowner(sop);
3824         }
3825 out:
3826         nfs4_unlock_state();
3827         return status;
3828 }
3829
3830 static inline struct nfs4_client_reclaim *
3831 alloc_reclaim(void)
3832 {
3833         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3834 }
3835
3836 int
3837 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3838 {
3839         unsigned int strhashval = clientstr_hashval(name);
3840         struct nfs4_client *clp;
3841
3842         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3843         return clp ? 1 : 0;
3844 }
3845
3846 /*
3847  * failure => all reset bets are off, nfserr_no_grace...
3848  */
3849 int
3850 nfs4_client_to_reclaim(const char *name)
3851 {
3852         unsigned int strhashval;
3853         struct nfs4_client_reclaim *crp = NULL;
3854
3855         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3856         crp = alloc_reclaim();
3857         if (!crp)
3858                 return 0;
3859         strhashval = clientstr_hashval(name);
3860         INIT_LIST_HEAD(&crp->cr_strhash);
3861         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3862         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3863         reclaim_str_hashtbl_size++;
3864         return 1;
3865 }
3866
3867 static void
3868 nfs4_release_reclaim(void)
3869 {
3870         struct nfs4_client_reclaim *crp = NULL;
3871         int i;
3872
3873         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3874                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3875                         crp = list_entry(reclaim_str_hashtbl[i].next,
3876                                         struct nfs4_client_reclaim, cr_strhash);
3877                         list_del(&crp->cr_strhash);
3878                         kfree(crp);
3879                         reclaim_str_hashtbl_size--;
3880                 }
3881         }
3882         BUG_ON(reclaim_str_hashtbl_size);
3883 }
3884
3885 /*
3886  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3887 static struct nfs4_client_reclaim *
3888 nfs4_find_reclaim_client(clientid_t *clid)
3889 {
3890         unsigned int strhashval;
3891         struct nfs4_client *clp;
3892         struct nfs4_client_reclaim *crp = NULL;
3893
3894
3895         /* find clientid in conf_id_hashtbl */
3896         clp = find_confirmed_client(clid);
3897         if (clp == NULL)
3898                 return NULL;
3899
3900         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3901                             clp->cl_name.len, clp->cl_name.data,
3902                             clp->cl_recdir);
3903
3904         /* find clp->cl_name in reclaim_str_hashtbl */
3905         strhashval = clientstr_hashval(clp->cl_recdir);
3906         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3907                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3908                         return crp;
3909                 }
3910         }
3911         return NULL;
3912 }
3913
3914 /*
3915 * Called from OPEN. Look for clientid in reclaim list.
3916 */
3917 __be32
3918 nfs4_check_open_reclaim(clientid_t *clid)
3919 {
3920         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3921 }
3922
3923 /* initialization to perform at module load time: */
3924
3925 int
3926 nfs4_state_init(void)
3927 {
3928         int i, status;
3929
3930         status = nfsd4_init_slabs();
3931         if (status)
3932                 return status;
3933         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3934                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3935                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3936                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3937                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3938                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3939         }
3940         for (i = 0; i < SESSION_HASH_SIZE; i++)
3941                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3942         for (i = 0; i < FILE_HASH_SIZE; i++) {
3943                 INIT_LIST_HEAD(&file_hashtbl[i]);
3944         }
3945         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3946                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3947                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3948         }
3949         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3950                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3951                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3952         }
3953         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3954                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3955                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3956         }
3957         memset(&onestateid, ~0, sizeof(stateid_t));
3958         INIT_LIST_HEAD(&close_lru);
3959         INIT_LIST_HEAD(&client_lru);
3960         INIT_LIST_HEAD(&del_recall_lru);
3961         reclaim_str_hashtbl_size = 0;
3962         return 0;
3963 }
3964
3965 static void
3966 nfsd4_load_reboot_recovery_data(void)
3967 {
3968         int status;
3969
3970         nfs4_lock_state();
3971         nfsd4_init_recdir(user_recovery_dirname);
3972         status = nfsd4_recdir_load();
3973         nfs4_unlock_state();
3974         if (status)
3975                 printk("NFSD: Failure reading reboot recovery data\n");
3976 }
3977
3978 unsigned long
3979 get_nfs4_grace_period(void)
3980 {
3981         return max(user_lease_time, lease_time) * HZ;
3982 }
3983
3984 /*
3985  * Since the lifetime of a delegation isn't limited to that of an open, a
3986  * client may quite reasonably hang on to a delegation as long as it has
3987  * the inode cached.  This becomes an obvious problem the first time a
3988  * client's inode cache approaches the size of the server's total memory.
3989  *
3990  * For now we avoid this problem by imposing a hard limit on the number
3991  * of delegations, which varies according to the server's memory size.
3992  */
3993 static void
3994 set_max_delegations(void)
3995 {
3996         /*
3997          * Allow at most 4 delegations per megabyte of RAM.  Quick
3998          * estimates suggest that in the worst case (where every delegation
3999          * is for a different inode), a delegation could take about 1.5K,
4000          * giving a worst case usage of about 6% of memory.
4001          */
4002         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4003 }
4004
4005 /* initialization to perform when the nfsd service is started: */
4006
4007 static void
4008 __nfs4_state_start(void)
4009 {
4010         unsigned long grace_time;
4011
4012         boot_time = get_seconds();
4013         grace_time = get_nfs4_grace_period();
4014         lease_time = user_lease_time;
4015         locks_start_grace(&nfsd4_manager);
4016         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4017                grace_time/HZ);
4018         laundry_wq = create_singlethread_workqueue("nfsd4");
4019         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4020         set_max_delegations();
4021 }
4022
4023 void
4024 nfs4_state_start(void)
4025 {
4026         if (nfs4_init)
4027                 return;
4028         nfsd4_load_reboot_recovery_data();
4029         __nfs4_state_start();
4030         nfs4_init = 1;
4031         return;
4032 }
4033
4034 time_t
4035 nfs4_lease_time(void)
4036 {
4037         return lease_time;
4038 }
4039
4040 static void
4041 __nfs4_state_shutdown(void)
4042 {
4043         int i;
4044         struct nfs4_client *clp = NULL;
4045         struct nfs4_delegation *dp = NULL;
4046         struct list_head *pos, *next, reaplist;
4047
4048         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4049                 while (!list_empty(&conf_id_hashtbl[i])) {
4050                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4051                         expire_client(clp);
4052                 }
4053                 while (!list_empty(&unconf_str_hashtbl[i])) {
4054                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4055                         expire_client(clp);
4056                 }
4057         }
4058         INIT_LIST_HEAD(&reaplist);
4059         spin_lock(&recall_lock);
4060         list_for_each_safe(pos, next, &del_recall_lru) {
4061                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4062                 list_move(&dp->dl_recall_lru, &reaplist);
4063         }
4064         spin_unlock(&recall_lock);
4065         list_for_each_safe(pos, next, &reaplist) {
4066                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4067                 list_del_init(&dp->dl_recall_lru);
4068                 unhash_delegation(dp);
4069         }
4070
4071         nfsd4_shutdown_recdir();
4072         nfs4_init = 0;
4073 }
4074
4075 void
4076 nfs4_state_shutdown(void)
4077 {
4078         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4079         destroy_workqueue(laundry_wq);
4080         locks_end_grace(&nfsd4_manager);
4081         nfs4_lock_state();
4082         nfs4_release_reclaim();
4083         __nfs4_