adc51d10d43514b742f7be85505d0161caeadbb9
[safe/jmp/linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/namei.h>
38 #include <linux/swap.h>
39 #include <linux/sunrpc/svcauth_gss.h>
40 #include <linux/sunrpc/clnt.h>
41 #include "xdr4.h"
42 #include "vfs.h"
43
44 #define NFSDDBG_FACILITY                NFSDDBG_PROC
45
46 /* Globals */
47 time_t nfsd4_lease = 90;     /* default lease time */
48 time_t nfsd4_grace = 90;
49 static time_t boot_time;
50 static u32 current_ownerid = 1;
51 static u32 current_fileid = 1;
52 static u32 current_delegid = 1;
53 static u32 nfs4_init;
54 static stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static struct nfs4_delegation *
166 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
167 {
168         struct nfs4_delegation *dp;
169         struct nfs4_file *fp = stp->st_file;
170         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
171
172         dprintk("NFSD alloc_init_deleg\n");
173         if (fp->fi_had_conflict)
174                 return NULL;
175         if (num_delegations > max_delegations)
176                 return NULL;
177         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
178         if (dp == NULL)
179                 return dp;
180         num_delegations++;
181         INIT_LIST_HEAD(&dp->dl_perfile);
182         INIT_LIST_HEAD(&dp->dl_perclnt);
183         INIT_LIST_HEAD(&dp->dl_recall_lru);
184         dp->dl_client = clp;
185         get_nfs4_file(fp);
186         dp->dl_file = fp;
187         dp->dl_flock = NULL;
188         get_file(stp->st_vfs_file);
189         dp->dl_vfs_file = stp->st_vfs_file;
190         dp->dl_type = type;
191         dp->dl_ident = cb->cb_ident;
192         dp->dl_stateid.si_boot = get_seconds();
193         dp->dl_stateid.si_stateownerid = current_delegid++;
194         dp->dl_stateid.si_fileid = 0;
195         dp->dl_stateid.si_generation = 0;
196         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
197         dp->dl_time = 0;
198         atomic_set(&dp->dl_count, 1);
199         list_add(&dp->dl_perfile, &fp->fi_delegations);
200         list_add(&dp->dl_perclnt, &clp->cl_delegations);
201         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
202         return dp;
203 }
204
205 void
206 nfs4_put_delegation(struct nfs4_delegation *dp)
207 {
208         if (atomic_dec_and_test(&dp->dl_count)) {
209                 dprintk("NFSD: freeing dp %p\n",dp);
210                 put_nfs4_file(dp->dl_file);
211                 kmem_cache_free(deleg_slab, dp);
212                 num_delegations--;
213         }
214 }
215
216 /* Remove the associated file_lock first, then remove the delegation.
217  * lease_modify() is called to remove the FS_LEASE file_lock from
218  * the i_flock list, eventually calling nfsd's lock_manager
219  * fl_release_callback.
220  */
221 static void
222 nfs4_close_delegation(struct nfs4_delegation *dp)
223 {
224         struct file *filp = dp->dl_vfs_file;
225
226         dprintk("NFSD: close_delegation dp %p\n",dp);
227         dp->dl_vfs_file = NULL;
228         /* The following nfsd_close may not actually close the file,
229          * but we want to remove the lease in any case. */
230         if (dp->dl_flock)
231                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
232         nfsd_close(filp);
233 }
234
235 /* Called under the state lock. */
236 static void
237 unhash_delegation(struct nfs4_delegation *dp)
238 {
239         list_del_init(&dp->dl_perfile);
240         list_del_init(&dp->dl_perclnt);
241         spin_lock(&recall_lock);
242         list_del_init(&dp->dl_recall_lru);
243         spin_unlock(&recall_lock);
244         nfs4_close_delegation(dp);
245         nfs4_put_delegation(dp);
246 }
247
248 /* 
249  * SETCLIENTID state 
250  */
251
252 /* Hash tables for nfs4_clientid state */
253 #define CLIENT_HASH_BITS                 4
254 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
255 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
256
257 #define clientid_hashval(id) \
258         ((id) & CLIENT_HASH_MASK)
259 #define clientstr_hashval(name) \
260         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
261 /*
262  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
263  * used in reboot/reset lease grace period processing
264  *
265  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
266  * setclientid_confirmed info. 
267  *
268  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
269  * setclientid info.
270  *
271  * client_lru holds client queue ordered by nfs4_client.cl_time
272  * for lease renewal.
273  *
274  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
275  * for last close replay.
276  */
277 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
278 static int reclaim_str_hashtbl_size = 0;
279 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
280 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
283 static struct list_head client_lru;
284 static struct list_head close_lru;
285
286 static void unhash_generic_stateid(struct nfs4_stateid *stp)
287 {
288         list_del(&stp->st_hash);
289         list_del(&stp->st_perfile);
290         list_del(&stp->st_perstateowner);
291 }
292
293 static void free_generic_stateid(struct nfs4_stateid *stp)
294 {
295         put_nfs4_file(stp->st_file);
296         kmem_cache_free(stateid_slab, stp);
297 }
298
299 static void release_lock_stateid(struct nfs4_stateid *stp)
300 {
301         unhash_generic_stateid(stp);
302         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
303         free_generic_stateid(stp);
304 }
305
306 static void unhash_lockowner(struct nfs4_stateowner *sop)
307 {
308         struct nfs4_stateid *stp;
309
310         list_del(&sop->so_idhash);
311         list_del(&sop->so_strhash);
312         list_del(&sop->so_perstateid);
313         while (!list_empty(&sop->so_stateids)) {
314                 stp = list_first_entry(&sop->so_stateids,
315                                 struct nfs4_stateid, st_perstateowner);
316                 release_lock_stateid(stp);
317         }
318 }
319
320 static void release_lockowner(struct nfs4_stateowner *sop)
321 {
322         unhash_lockowner(sop);
323         nfs4_put_stateowner(sop);
324 }
325
326 static void
327 release_stateid_lockowners(struct nfs4_stateid *open_stp)
328 {
329         struct nfs4_stateowner *lock_sop;
330
331         while (!list_empty(&open_stp->st_lockowners)) {
332                 lock_sop = list_entry(open_stp->st_lockowners.next,
333                                 struct nfs4_stateowner, so_perstateid);
334                 /* list_del(&open_stp->st_lockowners);  */
335                 BUG_ON(lock_sop->so_is_open_owner);
336                 release_lockowner(lock_sop);
337         }
338 }
339
340 static void release_open_stateid(struct nfs4_stateid *stp)
341 {
342         unhash_generic_stateid(stp);
343         release_stateid_lockowners(stp);
344         nfsd_close(stp->st_vfs_file);
345         free_generic_stateid(stp);
346 }
347
348 static void unhash_openowner(struct nfs4_stateowner *sop)
349 {
350         struct nfs4_stateid *stp;
351
352         list_del(&sop->so_idhash);
353         list_del(&sop->so_strhash);
354         list_del(&sop->so_perclient);
355         list_del(&sop->so_perstateid); /* XXX: necessary? */
356         while (!list_empty(&sop->so_stateids)) {
357                 stp = list_first_entry(&sop->so_stateids,
358                                 struct nfs4_stateid, st_perstateowner);
359                 release_open_stateid(stp);
360         }
361 }
362
363 static void release_openowner(struct nfs4_stateowner *sop)
364 {
365         unhash_openowner(sop);
366         list_del(&sop->so_close_lru);
367         nfs4_put_stateowner(sop);
368 }
369
370 static DEFINE_SPINLOCK(sessionid_lock);
371 #define SESSION_HASH_SIZE       512
372 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
373
374 static inline int
375 hash_sessionid(struct nfs4_sessionid *sessionid)
376 {
377         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
378
379         return sid->sequence % SESSION_HASH_SIZE;
380 }
381
382 static inline void
383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
384 {
385         u32 *ptr = (u32 *)(&sessionid->data[0]);
386         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
387 }
388
389 static void
390 gen_sessionid(struct nfsd4_session *ses)
391 {
392         struct nfs4_client *clp = ses->se_client;
393         struct nfsd4_sessionid *sid;
394
395         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
396         sid->clientid = clp->cl_clientid;
397         sid->sequence = current_sessionid++;
398         sid->reserved = 0;
399 }
400
401 /*
402  * The protocol defines ca_maxresponssize_cached to include the size of
403  * the rpc header, but all we need to cache is the data starting after
404  * the end of the initial SEQUENCE operation--the rest we regenerate
405  * each time.  Therefore we can advertise a ca_maxresponssize_cached
406  * value that is the number of bytes in our cache plus a few additional
407  * bytes.  In order to stay on the safe side, and not promise more than
408  * we can cache, those additional bytes must be the minimum possible: 24
409  * bytes of rpc header (xid through accept state, with AUTH_NULL
410  * verifier), 12 for the compound header (with zero-length tag), and 44
411  * for the SEQUENCE op response:
412  */
413 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
414
415 /*
416  * Give the client the number of ca_maxresponsesize_cached slots it
417  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
418  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
419  * than NFSD_MAX_SLOTS_PER_SESSION.
420  *
421  * If we run out of reserved DRC memory we should (up to a point)
422  * re-negotiate active sessions and reduce their slot usage to make
423  * rooom for new connections. For now we just fail the create session.
424  */
425 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
426 {
427         int mem, size = fchan->maxresp_cached;
428
429         if (fchan->maxreqs < 1)
430                 return nfserr_inval;
431
432         if (size < NFSD_MIN_HDR_SEQ_SZ)
433                 size = NFSD_MIN_HDR_SEQ_SZ;
434         size -= NFSD_MIN_HDR_SEQ_SZ;
435         if (size > NFSD_SLOT_CACHE_SIZE)
436                 size = NFSD_SLOT_CACHE_SIZE;
437
438         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
439         mem = fchan->maxreqs * size;
440         if (mem > NFSD_MAX_MEM_PER_SESSION) {
441                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
442                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
443                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
444                 mem = fchan->maxreqs * size;
445         }
446
447         spin_lock(&nfsd_drc_lock);
448         /* bound the total session drc memory ussage */
449         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
450                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
451                 mem = fchan->maxreqs * size;
452         }
453         nfsd_drc_mem_used += mem;
454         spin_unlock(&nfsd_drc_lock);
455
456         if (fchan->maxreqs == 0)
457                 return nfserr_serverfault;
458
459         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
460         return 0;
461 }
462
463 /*
464  * fchan holds the client values on input, and the server values on output
465  * sv_max_mesg is the maximum payload plus one page for overhead.
466  */
467 static int init_forechannel_attrs(struct svc_rqst *rqstp,
468                                   struct nfsd4_channel_attrs *session_fchan,
469                                   struct nfsd4_channel_attrs *fchan)
470 {
471         int status = 0;
472         __u32   maxcount = nfsd_serv->sv_max_mesg;
473
474         /* headerpadsz set to zero in encode routine */
475
476         /* Use the client's max request and max response size if possible */
477         if (fchan->maxreq_sz > maxcount)
478                 fchan->maxreq_sz = maxcount;
479         session_fchan->maxreq_sz = fchan->maxreq_sz;
480
481         if (fchan->maxresp_sz > maxcount)
482                 fchan->maxresp_sz = maxcount;
483         session_fchan->maxresp_sz = fchan->maxresp_sz;
484
485         /* Use the client's maxops if possible */
486         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
487                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
488         session_fchan->maxops = fchan->maxops;
489
490         /* FIXME: Error means no more DRC pages so the server should
491          * recover pages from existing sessions. For now fail session
492          * creation.
493          */
494         status = set_forechannel_drc_size(fchan);
495
496         session_fchan->maxresp_cached = fchan->maxresp_cached;
497         session_fchan->maxreqs = fchan->maxreqs;
498
499         dprintk("%s status %d\n", __func__, status);
500         return status;
501 }
502
503 static void
504 free_session_slots(struct nfsd4_session *ses)
505 {
506         int i;
507
508         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
509                 kfree(ses->se_slots[i]);
510 }
511
512 /*
513  * We don't actually need to cache the rpc and session headers, so we
514  * can allocate a little less for each slot:
515  */
516 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
517 {
518         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
519 }
520
521 static int
522 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
523                    struct nfsd4_create_session *cses)
524 {
525         struct nfsd4_session *new, tmp;
526         struct nfsd4_slot *sp;
527         int idx, slotsize, cachesize, i;
528         int status;
529
530         memset(&tmp, 0, sizeof(tmp));
531
532         /* FIXME: For now, we just accept the client back channel attributes. */
533         tmp.se_bchannel = cses->back_channel;
534         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
535                                         &cses->fore_channel);
536         if (status)
537                 goto out;
538
539         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
540                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
541
542         status = nfserr_serverfault;
543         /* allocate struct nfsd4_session and slot table pointers in one piece */
544         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
545         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
546         if (!new)
547                 goto out;
548
549         memcpy(new, &tmp, sizeof(*new));
550
551         /* allocate each struct nfsd4_slot and data cache in one piece */
552         cachesize = slot_bytes(&new->se_fchannel);
553         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
554                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
555                 if (!sp)
556                         goto out_free;
557                 new->se_slots[i] = sp;
558         }
559
560         new->se_client = clp;
561         gen_sessionid(new);
562         idx = hash_sessionid(&new->se_sessionid);
563         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
564                NFS4_MAX_SESSIONID_LEN);
565
566         new->se_flags = cses->flags;
567         kref_init(&new->se_ref);
568         spin_lock(&sessionid_lock);
569         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
570         list_add(&new->se_perclnt, &clp->cl_sessions);
571         spin_unlock(&sessionid_lock);
572
573         status = nfs_ok;
574 out:
575         return status;
576 out_free:
577         free_session_slots(new);
578         kfree(new);
579         goto out;
580 }
581
582 /* caller must hold sessionid_lock */
583 static struct nfsd4_session *
584 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
585 {
586         struct nfsd4_session *elem;
587         int idx;
588
589         dump_sessionid(__func__, sessionid);
590         idx = hash_sessionid(sessionid);
591         dprintk("%s: idx is %d\n", __func__, idx);
592         /* Search in the appropriate list */
593         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
594                 dump_sessionid("list traversal", &elem->se_sessionid);
595                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
596                             NFS4_MAX_SESSIONID_LEN)) {
597                         return elem;
598                 }
599         }
600
601         dprintk("%s: session not found\n", __func__);
602         return NULL;
603 }
604
605 /* caller must hold sessionid_lock */
606 static void
607 unhash_session(struct nfsd4_session *ses)
608 {
609         list_del(&ses->se_hash);
610         list_del(&ses->se_perclnt);
611 }
612
613 static void
614 release_session(struct nfsd4_session *ses)
615 {
616         spin_lock(&sessionid_lock);
617         unhash_session(ses);
618         spin_unlock(&sessionid_lock);
619         nfsd4_put_session(ses);
620 }
621
622 void
623 free_session(struct kref *kref)
624 {
625         struct nfsd4_session *ses;
626         int mem;
627
628         ses = container_of(kref, struct nfsd4_session, se_ref);
629         spin_lock(&nfsd_drc_lock);
630         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
631         nfsd_drc_mem_used -= mem;
632         spin_unlock(&nfsd_drc_lock);
633         free_session_slots(ses);
634         kfree(ses);
635 }
636
637 static inline void
638 renew_client(struct nfs4_client *clp)
639 {
640         /*
641         * Move client to the end to the LRU list.
642         */
643         dprintk("renewing client (clientid %08x/%08x)\n", 
644                         clp->cl_clientid.cl_boot, 
645                         clp->cl_clientid.cl_id);
646         list_move_tail(&clp->cl_lru, &client_lru);
647         clp->cl_time = get_seconds();
648 }
649
650 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
651 static int
652 STALE_CLIENTID(clientid_t *clid)
653 {
654         if (clid->cl_boot == boot_time)
655                 return 0;
656         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
657                 clid->cl_boot, clid->cl_id, boot_time);
658         return 1;
659 }
660
661 /* 
662  * XXX Should we use a slab cache ?
663  * This type of memory management is somewhat inefficient, but we use it
664  * anyway since SETCLIENTID is not a common operation.
665  */
666 static struct nfs4_client *alloc_client(struct xdr_netobj name)
667 {
668         struct nfs4_client *clp;
669
670         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
671         if (clp == NULL)
672                 return NULL;
673         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
674         if (clp->cl_name.data == NULL) {
675                 kfree(clp);
676                 return NULL;
677         }
678         memcpy(clp->cl_name.data, name.data, name.len);
679         clp->cl_name.len = name.len;
680         return clp;
681 }
682
683 static inline void
684 free_client(struct nfs4_client *clp)
685 {
686         if (clp->cl_cred.cr_group_info)
687                 put_group_info(clp->cl_cred.cr_group_info);
688         kfree(clp->cl_principal);
689         kfree(clp->cl_name.data);
690         kfree(clp);
691 }
692
693 void
694 put_nfs4_client(struct nfs4_client *clp)
695 {
696         if (atomic_dec_and_test(&clp->cl_count))
697                 free_client(clp);
698 }
699
700 static void
701 expire_client(struct nfs4_client *clp)
702 {
703         struct nfs4_stateowner *sop;
704         struct nfs4_delegation *dp;
705         struct list_head reaplist;
706
707         INIT_LIST_HEAD(&reaplist);
708         spin_lock(&recall_lock);
709         while (!list_empty(&clp->cl_delegations)) {
710                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
711                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
712                                 dp->dl_flock);
713                 list_del_init(&dp->dl_perclnt);
714                 list_move(&dp->dl_recall_lru, &reaplist);
715         }
716         spin_unlock(&recall_lock);
717         while (!list_empty(&reaplist)) {
718                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
719                 list_del_init(&dp->dl_recall_lru);
720                 unhash_delegation(dp);
721         }
722         list_del(&clp->cl_idhash);
723         list_del(&clp->cl_strhash);
724         list_del(&clp->cl_lru);
725         while (!list_empty(&clp->cl_openowners)) {
726                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
727                 release_openowner(sop);
728         }
729         while (!list_empty(&clp->cl_sessions)) {
730                 struct nfsd4_session  *ses;
731                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
732                                  se_perclnt);
733                 release_session(ses);
734         }
735         nfsd4_set_callback_client(clp, NULL);
736         if (clp->cl_cb_xprt)
737                 svc_xprt_put(clp->cl_cb_xprt);
738         put_nfs4_client(clp);
739 }
740
741 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
742 {
743         memcpy(target->cl_verifier.data, source->data,
744                         sizeof(target->cl_verifier.data));
745 }
746
747 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
748 {
749         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
750         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
751 }
752
753 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
754 {
755         target->cr_uid = source->cr_uid;
756         target->cr_gid = source->cr_gid;
757         target->cr_group_info = source->cr_group_info;
758         get_group_info(target->cr_group_info);
759 }
760
761 static int same_name(const char *n1, const char *n2)
762 {
763         return 0 == memcmp(n1, n2, HEXDIR_LEN);
764 }
765
766 static int
767 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
768 {
769         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
770 }
771
772 static int
773 same_clid(clientid_t *cl1, clientid_t *cl2)
774 {
775         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
776 }
777
778 /* XXX what about NGROUP */
779 static int
780 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
781 {
782         return cr1->cr_uid == cr2->cr_uid;
783 }
784
785 static void gen_clid(struct nfs4_client *clp)
786 {
787         static u32 current_clientid = 1;
788
789         clp->cl_clientid.cl_boot = boot_time;
790         clp->cl_clientid.cl_id = current_clientid++; 
791 }
792
793 static void gen_confirm(struct nfs4_client *clp)
794 {
795         static u32 i;
796         u32 *p;
797
798         p = (u32 *)clp->cl_confirm.data;
799         *p++ = get_seconds();
800         *p++ = i++;
801 }
802
803 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
804                 struct svc_rqst *rqstp, nfs4_verifier *verf)
805 {
806         struct nfs4_client *clp;
807         struct sockaddr *sa = svc_addr(rqstp);
808         char *princ;
809
810         clp = alloc_client(name);
811         if (clp == NULL)
812                 return NULL;
813
814         princ = svc_gss_principal(rqstp);
815         if (princ) {
816                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
817                 if (clp->cl_principal == NULL) {
818                         free_client(clp);
819                         return NULL;
820                 }
821         }
822
823         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
824         atomic_set(&clp->cl_count, 1);
825         atomic_set(&clp->cl_cb_conn.cb_set, 0);
826         INIT_LIST_HEAD(&clp->cl_idhash);
827         INIT_LIST_HEAD(&clp->cl_strhash);
828         INIT_LIST_HEAD(&clp->cl_openowners);
829         INIT_LIST_HEAD(&clp->cl_delegations);
830         INIT_LIST_HEAD(&clp->cl_sessions);
831         INIT_LIST_HEAD(&clp->cl_lru);
832         clear_bit(0, &clp->cl_cb_slot_busy);
833         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
834         copy_verf(clp, verf);
835         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
836         clp->cl_flavor = rqstp->rq_flavor;
837         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
838         gen_confirm(clp);
839
840         return clp;
841 }
842
843 static int check_name(struct xdr_netobj name)
844 {
845         if (name.len == 0) 
846                 return 0;
847         if (name.len > NFS4_OPAQUE_LIMIT) {
848                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
849                 return 0;
850         }
851         return 1;
852 }
853
854 static void
855 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
856 {
857         unsigned int idhashval;
858
859         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
860         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
861         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
862         list_add_tail(&clp->cl_lru, &client_lru);
863         clp->cl_time = get_seconds();
864 }
865
866 static void
867 move_to_confirmed(struct nfs4_client *clp)
868 {
869         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
870         unsigned int strhashval;
871
872         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
873         list_del_init(&clp->cl_strhash);
874         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
875         strhashval = clientstr_hashval(clp->cl_recdir);
876         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
877         renew_client(clp);
878 }
879
880 static struct nfs4_client *
881 find_confirmed_client(clientid_t *clid)
882 {
883         struct nfs4_client *clp;
884         unsigned int idhashval = clientid_hashval(clid->cl_id);
885
886         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
887                 if (same_clid(&clp->cl_clientid, clid))
888                         return clp;
889         }
890         return NULL;
891 }
892
893 static struct nfs4_client *
894 find_unconfirmed_client(clientid_t *clid)
895 {
896         struct nfs4_client *clp;
897         unsigned int idhashval = clientid_hashval(clid->cl_id);
898
899         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
900                 if (same_clid(&clp->cl_clientid, clid))
901                         return clp;
902         }
903         return NULL;
904 }
905
906 /*
907  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
908  * parameter. Matching is based on the fact the at least one of the
909  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
910  *
911  * FIXME: we need to unify the clientid namespaces for nfsv4.x
912  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
913  * and SET_CLIENTID{,_CONFIRM}
914  */
915 static inline int
916 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
917 {
918         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
919         return use_exchange_id == has_exchange_flags;
920 }
921
922 static struct nfs4_client *
923 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
924                              bool use_exchange_id)
925 {
926         struct nfs4_client *clp;
927
928         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
929                 if (same_name(clp->cl_recdir, dname) &&
930                     match_clientid_establishment(clp, use_exchange_id))
931                         return clp;
932         }
933         return NULL;
934 }
935
936 static struct nfs4_client *
937 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
938                                bool use_exchange_id)
939 {
940         struct nfs4_client *clp;
941
942         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
943                 if (same_name(clp->cl_recdir, dname) &&
944                     match_clientid_establishment(clp, use_exchange_id))
945                         return clp;
946         }
947         return NULL;
948 }
949
950 static void
951 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
952 {
953         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
954         unsigned short expected_family;
955
956         /* Currently, we only support tcp and tcp6 for the callback channel */
957         if (se->se_callback_netid_len == 3 &&
958             !memcmp(se->se_callback_netid_val, "tcp", 3))
959                 expected_family = AF_INET;
960         else if (se->se_callback_netid_len == 4 &&
961                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
962                 expected_family = AF_INET6;
963         else
964                 goto out_err;
965
966         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
967                                             se->se_callback_addr_len,
968                                             (struct sockaddr *) &cb->cb_addr,
969                                             sizeof(cb->cb_addr));
970
971         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
972                 goto out_err;
973
974         if (cb->cb_addr.ss_family == AF_INET6)
975                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
976
977         cb->cb_minorversion = 0;
978         cb->cb_prog = se->se_callback_prog;
979         cb->cb_ident = se->se_callback_ident;
980         return;
981 out_err:
982         cb->cb_addr.ss_family = AF_UNSPEC;
983         cb->cb_addrlen = 0;
984         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
985                 "will not receive delegations\n",
986                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
987
988         return;
989 }
990
991 /*
992  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
993  */
994 void
995 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
996 {
997         struct nfsd4_slot *slot = resp->cstate.slot;
998         unsigned int base;
999
1000         dprintk("--> %s slot %p\n", __func__, slot);
1001
1002         slot->sl_opcnt = resp->opcnt;
1003         slot->sl_status = resp->cstate.status;
1004
1005         if (nfsd4_not_cached(resp)) {
1006                 slot->sl_datalen = 0;
1007                 return;
1008         }
1009         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1010         base = (char *)resp->cstate.datap -
1011                                         (char *)resp->xbuf->head[0].iov_base;
1012         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1013                                     slot->sl_datalen))
1014                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1015         return;
1016 }
1017
1018 /*
1019  * Encode the replay sequence operation from the slot values.
1020  * If cachethis is FALSE encode the uncached rep error on the next
1021  * operation which sets resp->p and increments resp->opcnt for
1022  * nfs4svc_encode_compoundres.
1023  *
1024  */
1025 static __be32
1026 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1027                           struct nfsd4_compoundres *resp)
1028 {
1029         struct nfsd4_op *op;
1030         struct nfsd4_slot *slot = resp->cstate.slot;
1031
1032         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1033                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1034
1035         /* Encode the replayed sequence operation */
1036         op = &args->ops[resp->opcnt - 1];
1037         nfsd4_encode_operation(resp, op);
1038
1039         /* Return nfserr_retry_uncached_rep in next operation. */
1040         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1041                 op = &args->ops[resp->opcnt++];
1042                 op->status = nfserr_retry_uncached_rep;
1043                 nfsd4_encode_operation(resp, op);
1044         }
1045         return op->status;
1046 }
1047
1048 /*
1049  * The sequence operation is not cached because we can use the slot and
1050  * session values.
1051  */
1052 __be32
1053 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1054                          struct nfsd4_sequence *seq)
1055 {
1056         struct nfsd4_slot *slot = resp->cstate.slot;
1057         __be32 status;
1058
1059         dprintk("--> %s slot %p\n", __func__, slot);
1060
1061         /* Either returns 0 or nfserr_retry_uncached */
1062         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1063         if (status == nfserr_retry_uncached_rep)
1064                 return status;
1065
1066         /* The sequence operation has been encoded, cstate->datap set. */
1067         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1068
1069         resp->opcnt = slot->sl_opcnt;
1070         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1071         status = slot->sl_status;
1072
1073         return status;
1074 }
1075
1076 /*
1077  * Set the exchange_id flags returned by the server.
1078  */
1079 static void
1080 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1081 {
1082         /* pNFS is not supported */
1083         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1084
1085         /* Referrals are supported, Migration is not. */
1086         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1087
1088         /* set the wire flags to return to client. */
1089         clid->flags = new->cl_exchange_flags;
1090 }
1091
1092 __be32
1093 nfsd4_exchange_id(struct svc_rqst *rqstp,
1094                   struct nfsd4_compound_state *cstate,
1095                   struct nfsd4_exchange_id *exid)
1096 {
1097         struct nfs4_client *unconf, *conf, *new;
1098         int status;
1099         unsigned int            strhashval;
1100         char                    dname[HEXDIR_LEN];
1101         char                    addr_str[INET6_ADDRSTRLEN];
1102         nfs4_verifier           verf = exid->verifier;
1103         struct sockaddr         *sa = svc_addr(rqstp);
1104
1105         rpc_ntop(sa, addr_str, sizeof(addr_str));
1106         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1107                 "ip_addr=%s flags %x, spa_how %d\n",
1108                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1109                 addr_str, exid->flags, exid->spa_how);
1110
1111         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1112                 return nfserr_inval;
1113
1114         /* Currently only support SP4_NONE */
1115         switch (exid->spa_how) {
1116         case SP4_NONE:
1117                 break;
1118         case SP4_SSV:
1119                 return nfserr_encr_alg_unsupp;
1120         default:
1121                 BUG();                          /* checked by xdr code */
1122         case SP4_MACH_CRED:
1123                 return nfserr_serverfault;      /* no excuse :-/ */
1124         }
1125
1126         status = nfs4_make_rec_clidname(dname, &exid->clname);
1127
1128         if (status)
1129                 goto error;
1130
1131         strhashval = clientstr_hashval(dname);
1132
1133         nfs4_lock_state();
1134         status = nfs_ok;
1135
1136         conf = find_confirmed_client_by_str(dname, strhashval, true);
1137         if (conf) {
1138                 if (!same_verf(&verf, &conf->cl_verifier)) {
1139                         /* 18.35.4 case 8 */
1140                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1141                                 status = nfserr_not_same;
1142                                 goto out;
1143                         }
1144                         /* Client reboot: destroy old state */
1145                         expire_client(conf);
1146                         goto out_new;
1147                 }
1148                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1149                         /* 18.35.4 case 9 */
1150                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1151                                 status = nfserr_perm;
1152                                 goto out;
1153                         }
1154                         expire_client(conf);
1155                         goto out_new;
1156                 }
1157                 /*
1158                  * Set bit when the owner id and verifier map to an already
1159                  * confirmed client id (18.35.3).
1160                  */
1161                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1162
1163                 /*
1164                  * Falling into 18.35.4 case 2, possible router replay.
1165                  * Leave confirmed record intact and return same result.
1166                  */
1167                 copy_verf(conf, &verf);
1168                 new = conf;
1169                 goto out_copy;
1170         }
1171
1172         /* 18.35.4 case 7 */
1173         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1174                 status = nfserr_noent;
1175                 goto out;
1176         }
1177
1178         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1179         if (unconf) {
1180                 /*
1181                  * Possible retry or client restart.  Per 18.35.4 case 4,
1182                  * a new unconfirmed record should be generated regardless
1183                  * of whether any properties have changed.
1184                  */
1185                 expire_client(unconf);
1186         }
1187
1188 out_new:
1189         /* Normal case */
1190         new = create_client(exid->clname, dname, rqstp, &verf);
1191         if (new == NULL) {
1192                 status = nfserr_serverfault;
1193                 goto out;
1194         }
1195
1196         gen_clid(new);
1197         add_to_unconfirmed(new, strhashval);
1198 out_copy:
1199         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1200         exid->clientid.cl_id = new->cl_clientid.cl_id;
1201
1202         exid->seqid = 1;
1203         nfsd4_set_ex_flags(new, exid);
1204
1205         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1206                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1207         status = nfs_ok;
1208
1209 out:
1210         nfs4_unlock_state();
1211 error:
1212         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1213         return status;
1214 }
1215
1216 static int
1217 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1218 {
1219         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1220                 slot_seqid);
1221
1222         /* The slot is in use, and no response has been sent. */
1223         if (slot_inuse) {
1224                 if (seqid == slot_seqid)
1225                         return nfserr_jukebox;
1226                 else
1227                         return nfserr_seq_misordered;
1228         }
1229         /* Normal */
1230         if (likely(seqid == slot_seqid + 1))
1231                 return nfs_ok;
1232         /* Replay */
1233         if (seqid == slot_seqid)
1234                 return nfserr_replay_cache;
1235         /* Wraparound */
1236         if (seqid == 1 && (slot_seqid + 1) == 0)
1237                 return nfs_ok;
1238         /* Misordered replay or misordered new request */
1239         return nfserr_seq_misordered;
1240 }
1241
1242 /*
1243  * Cache the create session result into the create session single DRC
1244  * slot cache by saving the xdr structure. sl_seqid has been set.
1245  * Do this for solo or embedded create session operations.
1246  */
1247 static void
1248 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1249                            struct nfsd4_clid_slot *slot, int nfserr)
1250 {
1251         slot->sl_status = nfserr;
1252         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1253 }
1254
1255 static __be32
1256 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1257                             struct nfsd4_clid_slot *slot)
1258 {
1259         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1260         return slot->sl_status;
1261 }
1262
1263 __be32
1264 nfsd4_create_session(struct svc_rqst *rqstp,
1265                      struct nfsd4_compound_state *cstate,
1266                      struct nfsd4_create_session *cr_ses)
1267 {
1268         struct sockaddr *sa = svc_addr(rqstp);
1269         struct nfs4_client *conf, *unconf;
1270         struct nfsd4_clid_slot *cs_slot = NULL;
1271         int status = 0;
1272
1273         nfs4_lock_state();
1274         unconf = find_unconfirmed_client(&cr_ses->clientid);
1275         conf = find_confirmed_client(&cr_ses->clientid);
1276
1277         if (conf) {
1278                 cs_slot = &conf->cl_cs_slot;
1279                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1280                 if (status == nfserr_replay_cache) {
1281                         dprintk("Got a create_session replay! seqid= %d\n",
1282                                 cs_slot->sl_seqid);
1283                         /* Return the cached reply status */
1284                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1285                         goto out;
1286                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1287                         status = nfserr_seq_misordered;
1288                         dprintk("Sequence misordered!\n");
1289                         dprintk("Expected seqid= %d but got seqid= %d\n",
1290                                 cs_slot->sl_seqid, cr_ses->seqid);
1291                         goto out;
1292                 }
1293                 cs_slot->sl_seqid++;
1294         } else if (unconf) {
1295                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1296                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1297                         status = nfserr_clid_inuse;
1298                         goto out;
1299                 }
1300
1301                 cs_slot = &unconf->cl_cs_slot;
1302                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1303                 if (status) {
1304                         /* an unconfirmed replay returns misordered */
1305                         status = nfserr_seq_misordered;
1306                         goto out_cache;
1307                 }
1308
1309                 cs_slot->sl_seqid++; /* from 0 to 1 */
1310                 move_to_confirmed(unconf);
1311
1312                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1313                         unconf->cl_cb_xprt = rqstp->rq_xprt;
1314                         svc_xprt_get(unconf->cl_cb_xprt);
1315                         rpc_copy_addr(
1316                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1317                                 sa);
1318                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1319                         unconf->cl_cb_conn.cb_minorversion =
1320                                 cstate->minorversion;
1321                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1322                         unconf->cl_cb_seq_nr = 1;
1323                         nfsd4_probe_callback(unconf);
1324                 }
1325                 conf = unconf;
1326         } else {
1327                 status = nfserr_stale_clientid;
1328                 goto out;
1329         }
1330
1331         /*
1332          * We do not support RDMA or persistent sessions
1333          */
1334         cr_ses->flags &= ~SESSION4_PERSIST;
1335         cr_ses->flags &= ~SESSION4_RDMA;
1336
1337         status = alloc_init_session(rqstp, conf, cr_ses);
1338         if (status)
1339                 goto out;
1340
1341         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1342                NFS4_MAX_SESSIONID_LEN);
1343         cr_ses->seqid = cs_slot->sl_seqid;
1344
1345 out_cache:
1346         /* cache solo and embedded create sessions under the state lock */
1347         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1348 out:
1349         nfs4_unlock_state();
1350         dprintk("%s returns %d\n", __func__, ntohl(status));
1351         return status;
1352 }
1353
1354 __be32
1355 nfsd4_destroy_session(struct svc_rqst *r,
1356                       struct nfsd4_compound_state *cstate,
1357                       struct nfsd4_destroy_session *sessionid)
1358 {
1359         struct nfsd4_session *ses;
1360         u32 status = nfserr_badsession;
1361
1362         /* Notes:
1363          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1364          * - Should we return nfserr_back_chan_busy if waiting for
1365          *   callbacks on to-be-destroyed session?
1366          * - Do we need to clear any callback info from previous session?
1367          */
1368
1369         dump_sessionid(__func__, &sessionid->sessionid);
1370         spin_lock(&sessionid_lock);
1371         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1372         if (!ses) {
1373                 spin_unlock(&sessionid_lock);
1374                 goto out;
1375         }
1376
1377         unhash_session(ses);
1378         spin_unlock(&sessionid_lock);
1379
1380         /* wait for callbacks */
1381         nfsd4_set_callback_client(ses->se_client, NULL);
1382         nfsd4_put_session(ses);
1383         status = nfs_ok;
1384 out:
1385         dprintk("%s returns %d\n", __func__, ntohl(status));
1386         return status;
1387 }
1388
1389 __be32
1390 nfsd4_sequence(struct svc_rqst *rqstp,
1391                struct nfsd4_compound_state *cstate,
1392                struct nfsd4_sequence *seq)
1393 {
1394         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1395         struct nfsd4_session *session;
1396         struct nfsd4_slot *slot;
1397         int status;
1398
1399         if (resp->opcnt != 1)
1400                 return nfserr_sequence_pos;
1401
1402         spin_lock(&sessionid_lock);
1403         status = nfserr_badsession;
1404         session = find_in_sessionid_hashtbl(&seq->sessionid);
1405         if (!session)
1406                 goto out;
1407
1408         status = nfserr_badslot;
1409         if (seq->slotid >= session->se_fchannel.maxreqs)
1410                 goto out;
1411
1412         slot = session->se_slots[seq->slotid];
1413         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1414
1415         /* We do not negotiate the number of slots yet, so set the
1416          * maxslots to the session maxreqs which is used to encode
1417          * sr_highest_slotid and the sr_target_slot id to maxslots */
1418         seq->maxslots = session->se_fchannel.maxreqs;
1419
1420         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1421         if (status == nfserr_replay_cache) {
1422                 cstate->slot = slot;
1423                 cstate->session = session;
1424                 /* Return the cached reply status and set cstate->status
1425                  * for nfsd4_proc_compound processing */
1426                 status = nfsd4_replay_cache_entry(resp, seq);
1427                 cstate->status = nfserr_replay_cache;
1428                 goto out;
1429         }
1430         if (status)
1431                 goto out;
1432
1433         /* Success! bump slot seqid */
1434         slot->sl_inuse = true;
1435         slot->sl_seqid = seq->seqid;
1436         slot->sl_cachethis = seq->cachethis;
1437
1438         cstate->slot = slot;
1439         cstate->session = session;
1440
1441         /* Hold a session reference until done processing the compound:
1442          * nfsd4_put_session called only if the cstate slot is set.
1443          */
1444         nfsd4_get_session(session);
1445 out:
1446         spin_unlock(&sessionid_lock);
1447         /* Renew the clientid on success and on replay */
1448         if (cstate->session) {
1449                 nfs4_lock_state();
1450                 renew_client(session->se_client);
1451                 nfs4_unlock_state();
1452         }
1453         dprintk("%s: return %d\n", __func__, ntohl(status));
1454         return status;
1455 }
1456
1457 __be32
1458 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1459                   struct nfsd4_setclientid *setclid)
1460 {
1461         struct sockaddr         *sa = svc_addr(rqstp);
1462         struct xdr_netobj       clname = { 
1463                 .len = setclid->se_namelen,
1464                 .data = setclid->se_name,
1465         };
1466         nfs4_verifier           clverifier = setclid->se_verf;
1467         unsigned int            strhashval;
1468         struct nfs4_client      *conf, *unconf, *new;
1469         __be32                  status;
1470         char                    dname[HEXDIR_LEN];
1471         
1472         if (!check_name(clname))
1473                 return nfserr_inval;
1474
1475         status = nfs4_make_rec_clidname(dname, &clname);
1476         if (status)
1477                 return status;
1478
1479         /* 
1480          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1481          * We get here on a DRC miss.
1482          */
1483
1484         strhashval = clientstr_hashval(dname);
1485
1486         nfs4_lock_state();
1487         conf = find_confirmed_client_by_str(dname, strhashval, false);
1488         if (conf) {
1489                 /* RFC 3530 14.2.33 CASE 0: */
1490                 status = nfserr_clid_inuse;
1491                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1492                         char addr_str[INET6_ADDRSTRLEN];
1493                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1494                                  sizeof(addr_str));
1495                         dprintk("NFSD: setclientid: string in use by client "
1496                                 "at %s\n", addr_str);
1497                         goto out;
1498                 }
1499         }
1500         /*
1501          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1502          * has a description of SETCLIENTID request processing consisting
1503          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1504          */
1505         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1506         status = nfserr_resource;
1507         if (!conf) {
1508                 /*
1509                  * RFC 3530 14.2.33 CASE 4:
1510                  * placed first, because it is the normal case
1511                  */
1512                 if (unconf)
1513                         expire_client(unconf);
1514                 new = create_client(clname, dname, rqstp, &clverifier);
1515                 if (new == NULL)
1516                         goto out;
1517                 gen_clid(new);
1518         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1519                 /*
1520                  * RFC 3530 14.2.33 CASE 1:
1521                  * probable callback update
1522                  */
1523                 if (unconf) {
1524                         /* Note this is removing unconfirmed {*x***},
1525                          * which is stronger than RFC recommended {vxc**}.
1526                          * This has the advantage that there is at most
1527                          * one {*x***} in either list at any time.
1528                          */
1529                         expire_client(unconf);
1530                 }
1531                 new = create_client(clname, dname, rqstp, &clverifier);
1532                 if (new == NULL)
1533                         goto out;
1534                 copy_clid(new, conf);
1535         } else if (!unconf) {
1536                 /*
1537                  * RFC 3530 14.2.33 CASE 2:
1538                  * probable client reboot; state will be removed if
1539                  * confirmed.
1540                  */
1541                 new = create_client(clname, dname, rqstp, &clverifier);
1542                 if (new == NULL)
1543                         goto out;
1544                 gen_clid(new);
1545         } else {
1546                 /*
1547                  * RFC 3530 14.2.33 CASE 3:
1548                  * probable client reboot; state will be removed if
1549                  * confirmed.
1550                  */
1551                 expire_client(unconf);
1552                 new = create_client(clname, dname, rqstp, &clverifier);
1553                 if (new == NULL)
1554                         goto out;
1555                 gen_clid(new);
1556         }
1557         gen_callback(new, setclid, rpc_get_scope_id(sa));
1558         add_to_unconfirmed(new, strhashval);
1559         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1560         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1561         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1562         status = nfs_ok;
1563 out:
1564         nfs4_unlock_state();
1565         return status;
1566 }
1567
1568
1569 /*
1570  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1571  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1572  * bullets, labeled as CASE1 - CASE4 below.
1573  */
1574 __be32
1575 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1576                          struct nfsd4_compound_state *cstate,
1577                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1578 {
1579         struct sockaddr *sa = svc_addr(rqstp);
1580         struct nfs4_client *conf, *unconf;
1581         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1582         clientid_t * clid = &setclientid_confirm->sc_clientid;
1583         __be32 status;
1584
1585         if (STALE_CLIENTID(clid))
1586                 return nfserr_stale_clientid;
1587         /* 
1588          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1589          * We get here on a DRC miss.
1590          */
1591
1592         nfs4_lock_state();
1593
1594         conf = find_confirmed_client(clid);
1595         unconf = find_unconfirmed_client(clid);
1596
1597         status = nfserr_clid_inuse;
1598         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1599                 goto out;
1600         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1601                 goto out;
1602
1603         /*
1604          * section 14.2.34 of RFC 3530 has a description of
1605          * SETCLIENTID_CONFIRM request processing consisting
1606          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1607          */
1608         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1609                 /*
1610                  * RFC 3530 14.2.34 CASE 1:
1611                  * callback update
1612                  */
1613                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1614                         status = nfserr_clid_inuse;
1615                 else {
1616                         /* XXX: We just turn off callbacks until we can handle
1617                           * change request correctly. */
1618                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1619                         expire_client(unconf);
1620                         status = nfs_ok;
1621
1622                 }
1623         } else if (conf && !unconf) {
1624                 /*
1625                  * RFC 3530 14.2.34 CASE 2:
1626                  * probable retransmitted request; play it safe and
1627                  * do nothing.
1628                  */
1629                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1630                         status = nfserr_clid_inuse;
1631                 else
1632                         status = nfs_ok;
1633         } else if (!conf && unconf
1634                         && same_verf(&unconf->cl_confirm, &confirm)) {
1635                 /*
1636                  * RFC 3530 14.2.34 CASE 3:
1637                  * Normal case; new or rebooted client:
1638                  */
1639                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1640                         status = nfserr_clid_inuse;
1641                 } else {
1642                         unsigned int hash =
1643                                 clientstr_hashval(unconf->cl_recdir);
1644                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1645                                                             hash, false);
1646                         if (conf) {
1647                                 nfsd4_remove_clid_dir(conf);
1648                                 expire_client(conf);
1649                         }
1650                         move_to_confirmed(unconf);
1651                         conf = unconf;
1652                         nfsd4_probe_callback(conf);
1653                         status = nfs_ok;
1654                 }
1655         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1656             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1657                                                                 &confirm)))) {
1658                 /*
1659                  * RFC 3530 14.2.34 CASE 4:
1660                  * Client probably hasn't noticed that we rebooted yet.
1661                  */
1662                 status = nfserr_stale_clientid;
1663         } else {
1664                 /* check that we have hit one of the cases...*/
1665                 status = nfserr_clid_inuse;
1666         }
1667 out:
1668         nfs4_unlock_state();
1669         return status;
1670 }
1671
1672 /* OPEN Share state helper functions */
1673 static inline struct nfs4_file *
1674 alloc_init_file(struct inode *ino)
1675 {
1676         struct nfs4_file *fp;
1677         unsigned int hashval = file_hashval(ino);
1678
1679         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1680         if (fp) {
1681                 atomic_set(&fp->fi_ref, 1);
1682                 INIT_LIST_HEAD(&fp->fi_hash);
1683                 INIT_LIST_HEAD(&fp->fi_stateids);
1684                 INIT_LIST_HEAD(&fp->fi_delegations);
1685                 spin_lock(&recall_lock);
1686                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1687                 spin_unlock(&recall_lock);
1688                 fp->fi_inode = igrab(ino);
1689                 fp->fi_id = current_fileid++;
1690                 fp->fi_had_conflict = false;
1691                 return fp;
1692         }
1693         return NULL;
1694 }
1695
1696 static void
1697 nfsd4_free_slab(struct kmem_cache **slab)
1698 {
1699         if (*slab == NULL)
1700                 return;
1701         kmem_cache_destroy(*slab);
1702         *slab = NULL;
1703 }
1704
1705 void
1706 nfsd4_free_slabs(void)
1707 {
1708         nfsd4_free_slab(&stateowner_slab);
1709         nfsd4_free_slab(&file_slab);
1710         nfsd4_free_slab(&stateid_slab);
1711         nfsd4_free_slab(&deleg_slab);
1712 }
1713
1714 static int
1715 nfsd4_init_slabs(void)
1716 {
1717         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1718                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1719         if (stateowner_slab == NULL)
1720                 goto out_nomem;
1721         file_slab = kmem_cache_create("nfsd4_files",
1722                         sizeof(struct nfs4_file), 0, 0, NULL);
1723         if (file_slab == NULL)
1724                 goto out_nomem;
1725         stateid_slab = kmem_cache_create("nfsd4_stateids",
1726                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1727         if (stateid_slab == NULL)
1728                 goto out_nomem;
1729         deleg_slab = kmem_cache_create("nfsd4_delegations",
1730                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1731         if (deleg_slab == NULL)
1732                 goto out_nomem;
1733         return 0;
1734 out_nomem:
1735         nfsd4_free_slabs();
1736         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1737         return -ENOMEM;
1738 }
1739
1740 void
1741 nfs4_free_stateowner(struct kref *kref)
1742 {
1743         struct nfs4_stateowner *sop =
1744                 container_of(kref, struct nfs4_stateowner, so_ref);
1745         kfree(sop->so_owner.data);
1746         kmem_cache_free(stateowner_slab, sop);
1747 }
1748
1749 static inline struct nfs4_stateowner *
1750 alloc_stateowner(struct xdr_netobj *owner)
1751 {
1752         struct nfs4_stateowner *sop;
1753
1754         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1755                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1756                         memcpy(sop->so_owner.data, owner->data, owner->len);
1757                         sop->so_owner.len = owner->len;
1758                         kref_init(&sop->so_ref);
1759                         return sop;
1760                 } 
1761                 kmem_cache_free(stateowner_slab, sop);
1762         }
1763         return NULL;
1764 }
1765
1766 static struct nfs4_stateowner *
1767 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1768         struct nfs4_stateowner *sop;
1769         struct nfs4_replay *rp;
1770         unsigned int idhashval;
1771
1772         if (!(sop = alloc_stateowner(&open->op_owner)))
1773                 return NULL;
1774         idhashval = ownerid_hashval(current_ownerid);
1775         INIT_LIST_HEAD(&sop->so_idhash);
1776         INIT_LIST_HEAD(&sop->so_strhash);
1777         INIT_LIST_HEAD(&sop->so_perclient);
1778         INIT_LIST_HEAD(&sop->so_stateids);
1779         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1780         INIT_LIST_HEAD(&sop->so_close_lru);
1781         sop->so_time = 0;
1782         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1783         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1784         list_add(&sop->so_perclient, &clp->cl_openowners);
1785         sop->so_is_open_owner = 1;
1786         sop->so_id = current_ownerid++;
1787         sop->so_client = clp;
1788         sop->so_seqid = open->op_seqid;
1789         sop->so_confirmed = 0;
1790         rp = &sop->so_replay;
1791         rp->rp_status = nfserr_serverfault;
1792         rp->rp_buflen = 0;
1793         rp->rp_buf = rp->rp_ibuf;
1794         return sop;
1795 }
1796
1797 static inline void
1798 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1799         struct nfs4_stateowner *sop = open->op_stateowner;
1800         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1801
1802         INIT_LIST_HEAD(&stp->st_hash);
1803         INIT_LIST_HEAD(&stp->st_perstateowner);
1804         INIT_LIST_HEAD(&stp->st_lockowners);
1805         INIT_LIST_HEAD(&stp->st_perfile);
1806         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1807         list_add(&stp->st_perstateowner, &sop->so_stateids);
1808         list_add(&stp->st_perfile, &fp->fi_stateids);
1809         stp->st_stateowner = sop;
1810         get_nfs4_file(fp);
1811         stp->st_file = fp;
1812         stp->st_stateid.si_boot = get_seconds();
1813         stp->st_stateid.si_stateownerid = sop->so_id;
1814         stp->st_stateid.si_fileid = fp->fi_id;
1815         stp->st_stateid.si_generation = 0;
1816         stp->st_access_bmap = 0;
1817         stp->st_deny_bmap = 0;
1818         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1819                   &stp->st_access_bmap);
1820         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1821         stp->st_openstp = NULL;
1822 }
1823
1824 static void
1825 move_to_close_lru(struct nfs4_stateowner *sop)
1826 {
1827         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1828
1829         list_move_tail(&sop->so_close_lru, &close_lru);
1830         sop->so_time = get_seconds();
1831 }
1832
1833 static int
1834 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1835                                                         clientid_t *clid)
1836 {
1837         return (sop->so_owner.len == owner->len) &&
1838                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1839                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1840 }
1841
1842 static struct nfs4_stateowner *
1843 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1844 {
1845         struct nfs4_stateowner *so = NULL;
1846
1847         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1848                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1849                         return so;
1850         }
1851         return NULL;
1852 }
1853
1854 /* search file_hashtbl[] for file */
1855 static struct nfs4_file *
1856 find_file(struct inode *ino)
1857 {
1858         unsigned int hashval = file_hashval(ino);
1859         struct nfs4_file *fp;
1860
1861         spin_lock(&recall_lock);
1862         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1863                 if (fp->fi_inode == ino) {
1864                         get_nfs4_file(fp);
1865                         spin_unlock(&recall_lock);
1866                         return fp;
1867                 }
1868         }
1869         spin_unlock(&recall_lock);
1870         return NULL;
1871 }
1872
1873 static inline int access_valid(u32 x, u32 minorversion)
1874 {
1875         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1876                 return 0;
1877         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1878                 return 0;
1879         x &= ~NFS4_SHARE_ACCESS_MASK;
1880         if (minorversion && x) {
1881                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1882                         return 0;
1883                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1884                         return 0;
1885                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1886         }
1887         if (x)
1888                 return 0;
1889         return 1;
1890 }
1891
1892 static inline int deny_valid(u32 x)
1893 {
1894         /* Note: unlike access bits, deny bits may be zero. */
1895         return x <= NFS4_SHARE_DENY_BOTH;
1896 }
1897
1898 /*
1899  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1900  * st_{access,deny}_bmap field of the stateid, in order to track not
1901  * only what share bits are currently in force, but also what
1902  * combinations of share bits previous opens have used.  This allows us
1903  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1904  * return an error if the client attempt to downgrade to a combination
1905  * of share bits not explicable by closing some of its previous opens.
1906  *
1907  * XXX: This enforcement is actually incomplete, since we don't keep
1908  * track of access/deny bit combinations; so, e.g., we allow:
1909  *
1910  *      OPEN allow read, deny write
1911  *      OPEN allow both, deny none
1912  *      DOWNGRADE allow read, deny none
1913  *
1914  * which we should reject.
1915  */
1916 static void
1917 set_access(unsigned int *access, unsigned long bmap) {
1918         int i;
1919
1920         *access = 0;
1921         for (i = 1; i < 4; i++) {
1922                 if (test_bit(i, &bmap))
1923                         *access |= i;
1924         }
1925 }
1926
1927 static void
1928 set_deny(unsigned int *deny, unsigned long bmap) {
1929         int i;
1930
1931         *deny = 0;
1932         for (i = 0; i < 4; i++) {
1933                 if (test_bit(i, &bmap))
1934                         *deny |= i ;
1935         }
1936 }
1937
1938 static int
1939 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1940         unsigned int access, deny;
1941
1942         set_access(&access, stp->st_access_bmap);
1943         set_deny(&deny, stp->st_deny_bmap);
1944         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1945                 return 0;
1946         return 1;
1947 }
1948
1949 /*
1950  * Called to check deny when READ with all zero stateid or
1951  * WRITE with all zero or all one stateid
1952  */
1953 static __be32
1954 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1955 {
1956         struct inode *ino = current_fh->fh_dentry->d_inode;
1957         struct nfs4_file *fp;
1958         struct nfs4_stateid *stp;
1959         __be32 ret;
1960
1961         dprintk("NFSD: nfs4_share_conflict\n");
1962
1963         fp = find_file(ino);
1964         if (!fp)
1965                 return nfs_ok;
1966         ret = nfserr_locked;
1967         /* Search for conflicting share reservations */
1968         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1969                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1970                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1971                         goto out;
1972         }
1973         ret = nfs_ok;
1974 out:
1975         put_nfs4_file(fp);
1976         return ret;
1977 }
1978
1979 static inline void
1980 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1981 {
1982         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1983                 drop_file_write_access(filp);
1984                 spin_lock(&filp->f_lock);
1985                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1986                 spin_unlock(&filp->f_lock);
1987         }
1988 }
1989
1990 /*
1991  * Spawn a thread to perform a recall on the delegation represented
1992  * by the lease (file_lock)
1993  *
1994  * Called from break_lease() with lock_kernel() held.
1995  * Note: we assume break_lease will only call this *once* for any given
1996  * lease.
1997  */
1998 static
1999 void nfsd_break_deleg_cb(struct file_lock *fl)
2000 {
2001         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2002
2003         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2004         if (!dp)
2005                 return;
2006
2007         /* We're assuming the state code never drops its reference
2008          * without first removing the lease.  Since we're in this lease
2009          * callback (and since the lease code is serialized by the kernel
2010          * lock) we know the server hasn't removed the lease yet, we know
2011          * it's safe to take a reference: */
2012         atomic_inc(&dp->dl_count);
2013         atomic_inc(&dp->dl_client->cl_count);
2014
2015         spin_lock(&recall_lock);
2016         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2017         spin_unlock(&recall_lock);
2018
2019         /* only place dl_time is set. protected by lock_kernel*/
2020         dp->dl_time = get_seconds();
2021
2022         /*
2023          * We don't want the locks code to timeout the lease for us;
2024          * we'll remove it ourself if the delegation isn't returned
2025          * in time.
2026          */
2027         fl->fl_break_time = 0;
2028
2029         dp->dl_file->fi_had_conflict = true;
2030         nfsd4_cb_recall(dp);
2031 }
2032
2033 /*
2034  * The file_lock is being reapd.
2035  *
2036  * Called by locks_free_lock() with lock_kernel() held.
2037  */
2038 static
2039 void nfsd_release_deleg_cb(struct file_lock *fl)
2040 {
2041         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2042
2043         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2044
2045         if (!(fl->fl_flags & FL_LEASE) || !dp)
2046                 return;
2047         dp->dl_flock = NULL;
2048 }
2049
2050 /*
2051  * Set the delegation file_lock back pointer.
2052  *
2053  * Called from setlease() with lock_kernel() held.
2054  */
2055 static
2056 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2057 {
2058         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2059
2060         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2061         if (!dp)
2062                 return;
2063         dp->dl_flock = new;
2064 }
2065
2066 /*
2067  * Called from setlease() with lock_kernel() held
2068  */
2069 static
2070 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2071 {
2072         struct nfs4_delegation *onlistd =
2073                 (struct nfs4_delegation *)onlist->fl_owner;
2074         struct nfs4_delegation *tryd =
2075                 (struct nfs4_delegation *)try->fl_owner;
2076
2077         if (onlist->fl_lmops != try->fl_lmops)
2078                 return 0;
2079
2080         return onlistd->dl_client == tryd->dl_client;
2081 }
2082
2083
2084 static
2085 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2086 {
2087         if (arg & F_UNLCK)
2088                 return lease_modify(onlist, arg);
2089         else
2090                 return -EAGAIN;
2091 }
2092
2093 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2094         .fl_break = nfsd_break_deleg_cb,
2095         .fl_release_private = nfsd_release_deleg_cb,
2096         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2097         .fl_mylease = nfsd_same_client_deleg_cb,
2098         .fl_change = nfsd_change_deleg_cb,
2099 };
2100
2101
2102 __be32
2103 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2104                     struct nfsd4_open *open)
2105 {
2106         clientid_t *clientid = &open->op_clientid;
2107         struct nfs4_client *clp = NULL;
2108         unsigned int strhashval;
2109         struct nfs4_stateowner *sop = NULL;
2110
2111         if (!check_name(open->op_owner))
2112                 return nfserr_inval;
2113
2114         if (STALE_CLIENTID(&open->op_clientid))
2115                 return nfserr_stale_clientid;
2116
2117         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2118         sop = find_openstateowner_str(strhashval, open);
2119         open->op_stateowner = sop;
2120         if (!sop) {
2121                 /* Make sure the client's lease hasn't expired. */
2122                 clp = find_confirmed_client(clientid);
2123                 if (clp == NULL)
2124                         return nfserr_expired;
2125                 goto renew;
2126         }
2127         /* When sessions are used, skip open sequenceid processing */
2128         if (nfsd4_has_session(cstate))
2129                 goto renew;
2130         if (!sop->so_confirmed) {
2131                 /* Replace unconfirmed owners without checking for replay. */
2132                 clp = sop->so_client;
2133                 release_openowner(sop);
2134                 open->op_stateowner = NULL;
2135                 goto renew;
2136         }
2137         if (open->op_seqid == sop->so_seqid - 1) {
2138                 if (sop->so_replay.rp_buflen)
2139                         return nfserr_replay_me;
2140                 /* The original OPEN failed so spectacularly
2141                  * that we don't even have replay data saved!
2142                  * Therefore, we have no choice but to continue
2143                  * processing this OPEN; presumably, we'll
2144                  * fail again for the same reason.
2145                  */
2146                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2147                 goto renew;
2148         }
2149         if (open->op_seqid != sop->so_seqid)
2150                 return nfserr_bad_seqid;
2151 renew:
2152         if (open->op_stateowner == NULL) {
2153                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2154                 if (sop == NULL)
2155                         return nfserr_resource;
2156                 open->op_stateowner = sop;
2157         }
2158         list_del_init(&sop->so_close_lru);
2159         renew_client(sop->so_client);
2160         return nfs_ok;
2161 }
2162
2163 static inline __be32
2164 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2165 {
2166         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2167                 return nfserr_openmode;
2168         else
2169                 return nfs_ok;
2170 }
2171
2172 static struct nfs4_delegation *
2173 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2174 {
2175         struct nfs4_delegation *dp;
2176
2177         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2178                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2179                         return dp;
2180         }
2181         return NULL;
2182 }
2183
2184 static __be32
2185 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2186                 struct nfs4_delegation **dp)
2187 {
2188         int flags;
2189         __be32 status = nfserr_bad_stateid;
2190
2191         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2192         if (*dp == NULL)
2193                 goto out;
2194         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2195                                                 RD_STATE : WR_STATE;
2196         status = nfs4_check_delegmode(*dp, flags);
2197         if (status)
2198                 *dp = NULL;
2199 out:
2200         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2201                 return nfs_ok;
2202         if (status)
2203                 return status;
2204         open->op_stateowner->so_confirmed = 1;
2205         return nfs_ok;
2206 }
2207
2208 static __be32
2209 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2210 {
2211         struct nfs4_stateid *local;
2212         __be32 status = nfserr_share_denied;
2213         struct nfs4_stateowner *sop = open->op_stateowner;
2214
2215         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2216                 /* ignore lock owners */
2217                 if (local->st_stateowner->so_is_open_owner == 0)
2218                         continue;
2219                 /* remember if we have seen this open owner */
2220                 if (local->st_stateowner == sop)
2221                         *stpp = local;
2222                 /* check for conflicting share reservations */
2223                 if (!test_share(local, open))
2224                         goto out;
2225         }
2226         status = 0;
2227 out:
2228         return status;
2229 }
2230
2231 static inline struct nfs4_stateid *
2232 nfs4_alloc_stateid(void)
2233 {
2234         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2235 }
2236
2237 static __be32
2238 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2239                 struct nfs4_delegation *dp,
2240                 struct svc_fh *cur_fh, int flags)
2241 {
2242         struct nfs4_stateid *stp;
2243
2244         stp = nfs4_alloc_stateid();
2245         if (stp == NULL)
2246                 return nfserr_resource;
2247
2248         if (dp) {
2249                 get_file(dp->dl_vfs_file);
2250                 stp->st_vfs_file = dp->dl_vfs_file;
2251         } else {
2252                 __be32 status;
2253                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2254                                 &stp->st_vfs_file);
2255                 if (status) {
2256                         if (status == nfserr_dropit)
2257                                 status = nfserr_jukebox;
2258                         kmem_cache_free(stateid_slab, stp);
2259                         return status;
2260                 }
2261         }
2262         *stpp = stp;
2263         return 0;
2264 }
2265
2266 static inline __be32
2267 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2268                 struct nfsd4_open *open)
2269 {
2270         struct iattr iattr = {
2271                 .ia_valid = ATTR_SIZE,
2272                 .ia_size = 0,
2273         };
2274         if (!open->op_truncate)
2275                 return 0;
2276         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2277                 return nfserr_inval;
2278         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2279 }
2280
2281 static __be32
2282 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2283 {
2284         struct file *filp = stp->st_vfs_file;
2285         struct inode *inode = filp->f_path.dentry->d_inode;
2286         unsigned int share_access, new_writer;
2287         __be32 status;
2288
2289         set_access(&share_access, stp->st_access_bmap);
2290         new_writer = (~share_access) & open->op_share_access
2291                         & NFS4_SHARE_ACCESS_WRITE;
2292
2293         if (new_writer) {
2294                 int err = get_write_access(inode);
2295                 if (err)
2296                         return nfserrno(err);
2297                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2298                 if (err)
2299                         return nfserrno(err);
2300                 file_take_write(filp);
2301         }
2302         status = nfsd4_truncate(rqstp, cur_fh, open);
2303         if (status) {
2304                 if (new_writer)
2305                         put_write_access(inode);
2306                 return status;
2307         }
2308         /* remember the open */
2309         filp->f_mode |= open->op_share_access;
2310         __set_bit(open->op_share_access, &stp->st_access_bmap);
2311         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2312
2313         return nfs_ok;
2314 }
2315
2316
2317 static void
2318 nfs4_set_claim_prev(struct nfsd4_open *open)
2319 {
2320         open->op_stateowner->so_confirmed = 1;
2321         open->op_stateowner->so_client->cl_firststate = 1;
2322 }
2323
2324 /*
2325  * Attempt to hand out a delegation.
2326  */
2327 static void
2328 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2329 {
2330         struct nfs4_delegation *dp;
2331         struct nfs4_stateowner *sop = stp->st_stateowner;
2332         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2333         struct file_lock fl, *flp = &fl;
2334         int status, flag = 0;
2335
2336         flag = NFS4_OPEN_DELEGATE_NONE;
2337         open->op_recall = 0;
2338         switch (open->op_claim_type) {
2339                 case NFS4_OPEN_CLAIM_PREVIOUS:
2340                         if (!atomic_read(&cb->cb_set))
2341                                 open->op_recall = 1;
2342                         flag = open->op_delegate_type;
2343                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2344                                 goto out;
2345                         break;
2346                 case NFS4_OPEN_CLAIM_NULL:
2347                         /* Let's not give out any delegations till everyone's
2348                          * had the chance to reclaim theirs.... */
2349                         if (locks_in_grace())
2350                                 goto out;
2351                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2352                                 goto out;
2353                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2354                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2355                         else
2356                                 flag = NFS4_OPEN_DELEGATE_READ;
2357                         break;
2358                 default:
2359                         goto out;
2360         }
2361
2362         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2363         if (dp == NULL) {
2364                 flag = NFS4_OPEN_DELEGATE_NONE;
2365                 goto out;
2366         }
2367         locks_init_lock(&fl);
2368         fl.fl_lmops = &nfsd_lease_mng_ops;
2369         fl.fl_flags = FL_LEASE;
2370         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2371         fl.fl_end = OFFSET_MAX;
2372         fl.fl_owner =  (fl_owner_t)dp;
2373         fl.fl_file = stp->st_vfs_file;
2374         fl.fl_pid = current->tgid;
2375
2376         /* vfs_setlease checks to see if delegation should be handed out.
2377          * the lock_manager callbacks fl_mylease and fl_change are used
2378          */
2379         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2380                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2381                 unhash_delegation(dp);
2382                 flag = NFS4_OPEN_DELEGATE_NONE;
2383                 goto out;
2384         }
2385
2386         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2387
2388         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2389                 STATEID_VAL(&dp->dl_stateid));
2390 out:
2391         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2392                         && flag == NFS4_OPEN_DELEGATE_NONE
2393                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2394                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2395         open->op_delegate_type = flag;
2396 }
2397
2398 /*
2399  * called with nfs4_lock_state() held.
2400  */
2401 __be32
2402 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2403 {
2404         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2405         struct nfs4_file *fp = NULL;
2406         struct inode *ino = current_fh->fh_dentry->d_inode;
2407         struct nfs4_stateid *stp = NULL;
2408         struct nfs4_delegation *dp = NULL;
2409         __be32 status;
2410
2411         status = nfserr_inval;
2412         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2413                         || !deny_valid(open->op_share_deny))
2414                 goto out;
2415         /*
2416          * Lookup file; if found, lookup stateid and check open request,
2417          * and check for delegations in the process of being recalled.
2418          * If not found, create the nfs4_file struct
2419          */
2420         fp = find_file(ino);
2421         if (fp) {
2422                 if ((status = nfs4_check_open(fp, open, &stp)))
2423                         goto out;
2424                 status = nfs4_check_deleg(fp, open, &dp);
2425                 if (status)
2426                         goto out;
2427         } else {
2428                 status = nfserr_bad_stateid;
2429                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2430                         goto out;
2431                 status = nfserr_resource;
2432                 fp = alloc_init_file(ino);
2433                 if (fp == NULL)
2434                         goto out;
2435         }
2436
2437         /*
2438          * OPEN the file, or upgrade an existing OPEN.
2439          * If truncate fails, the OPEN fails.
2440          */
2441         if (stp) {
2442                 /* Stateid was found, this is an OPEN upgrade */
2443                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2444                 if (status)
2445                         goto out;
2446                 update_stateid(&stp->st_stateid);
2447         } else {
2448                 /* Stateid was not found, this is a new OPEN */
2449                 int flags = 0;
2450                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2451                         flags |= NFSD_MAY_READ;
2452                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2453                         flags |= NFSD_MAY_WRITE;
2454                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2455                 if (status)
2456                         goto out;
2457                 init_stateid(stp, fp, open);
2458                 status = nfsd4_truncate(rqstp, current_fh, open);
2459                 if (status) {
2460                         release_open_stateid(stp);
2461                         goto out;
2462                 }
2463                 if (nfsd4_has_session(&resp->cstate))
2464                         update_stateid(&stp->st_stateid);
2465         }
2466         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2467
2468         if (nfsd4_has_session(&resp->cstate)) {
2469                 open->op_stateowner->so_confirmed = 1;
2470                 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2471         }
2472
2473         /*
2474         * Attempt to hand out a delegation. No error return, because the
2475         * OPEN succeeds even if we fail.
2476         */
2477         nfs4_open_delegation(current_fh, open, stp);
2478
2479         status = nfs_ok;
2480
2481         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2482                 STATEID_VAL(&stp->st_stateid));
2483 out:
2484         if (fp)
2485                 put_nfs4_file(fp);
2486         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2487                 nfs4_set_claim_prev(open);
2488         /*
2489         * To finish the open response, we just need to set the rflags.
2490         */
2491         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2492         if (!open->op_stateowner->so_confirmed &&
2493             !nfsd4_has_session(&resp->cstate))
2494                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2495
2496         return status;
2497 }
2498
2499 __be32
2500 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2501             clientid_t *clid)
2502 {
2503         struct nfs4_client *clp;
2504         __be32 status;
2505
2506         nfs4_lock_state();
2507         dprintk("process_renew(%08x/%08x): starting\n", 
2508                         clid->cl_boot, clid->cl_id);
2509         status = nfserr_stale_clientid;
2510         if (STALE_CLIENTID(clid))
2511                 goto out;
2512         clp = find_confirmed_client(clid);
2513         status = nfserr_expired;
2514         if (clp == NULL) {
2515                 /* We assume the client took too long to RENEW. */
2516                 dprintk("nfsd4_renew: clientid not found!\n");
2517                 goto out;
2518         }
2519         renew_client(clp);
2520         status = nfserr_cb_path_down;
2521         if (!list_empty(&clp->cl_delegations)
2522                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2523                 goto out;
2524         status = nfs_ok;
2525 out:
2526         nfs4_unlock_state();
2527         return status;
2528 }
2529
2530 struct lock_manager nfsd4_manager = {
2531 };
2532
2533 static void
2534 nfsd4_end_grace(void)
2535 {
2536         dprintk("NFSD: end of grace period\n");
2537         nfsd4_recdir_purge_old();
2538         locks_end_grace(&nfsd4_manager);
2539         /*
2540          * Now that every NFSv4 client has had the chance to recover and
2541          * to see the (possibly new, possibly shorter) lease time, we
2542          * can safely set the next grace time to the current lease time:
2543          */
2544         nfsd4_grace = nfsd4_lease;
2545 }
2546
2547 static time_t
2548 nfs4_laundromat(void)
2549 {
2550         struct nfs4_client *clp;
2551         struct nfs4_stateowner *sop;
2552         struct nfs4_delegation *dp;
2553         struct list_head *pos, *next, reaplist;
2554         time_t cutoff = get_seconds() - nfsd4_lease;
2555         time_t t, clientid_val = nfsd4_lease;
2556         time_t u, test_val = nfsd4_lease;
2557
2558         nfs4_lock_state();
2559
2560         dprintk("NFSD: laundromat service - starting\n");
2561         if (locks_in_grace())
2562                 nfsd4_end_grace();
2563         list_for_each_safe(pos, next, &client_lru) {
2564                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2565                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2566                         t = clp->cl_time - cutoff;
2567                         if (clientid_val > t)
2568                                 clientid_val = t;
2569                         break;
2570                 }
2571                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2572                         clp->cl_clientid.cl_id);
2573                 nfsd4_remove_clid_dir(clp);
2574                 expire_client(clp);
2575         }
2576         INIT_LIST_HEAD(&reaplist);
2577         spin_lock(&recall_lock);
2578         list_for_each_safe(pos, next, &del_recall_lru) {
2579                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2580                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2581                         u = dp->dl_time - cutoff;
2582                         if (test_val > u)
2583                                 test_val = u;
2584                         break;
2585                 }
2586                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2587                                     dp, dp->dl_flock);
2588                 list_move(&dp->dl_recall_lru, &reaplist);
2589         }
2590         spin_unlock(&recall_lock);
2591         list_for_each_safe(pos, next, &reaplist) {
2592                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2593                 list_del_init(&dp->dl_recall_lru);
2594                 unhash_delegation(dp);
2595         }
2596         test_val = nfsd4_lease;
2597         list_for_each_safe(pos, next, &close_lru) {
2598                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2599                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2600                         u = sop->so_time - cutoff;
2601                         if (test_val > u)
2602                                 test_val = u;
2603                         break;
2604                 }
2605                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2606                         sop->so_id);
2607                 release_openowner(sop);
2608         }
2609         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2610                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2611         nfs4_unlock_state();
2612         return clientid_val;
2613 }
2614
2615 static struct workqueue_struct *laundry_wq;
2616 static void laundromat_main(struct work_struct *);
2617 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2618
2619 static void
2620 laundromat_main(struct work_struct *not_used)
2621 {
2622         time_t t;
2623
2624         t = nfs4_laundromat();
2625         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2626         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2627 }
2628
2629 static struct nfs4_stateowner *
2630 search_close_lru(u32 st_id, int flags)
2631 {
2632         struct nfs4_stateowner *local = NULL;
2633
2634         if (flags & CLOSE_STATE) {
2635                 list_for_each_entry(local, &close_lru, so_close_lru) {
2636                         if (local->so_id == st_id)
2637                                 return local;
2638                 }
2639         }
2640         return NULL;
2641 }
2642
2643 static inline int
2644 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2645 {
2646         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2647 }
2648
2649 static int
2650 STALE_STATEID(stateid_t *stateid)
2651 {
2652         if (time_after((unsigned long)boot_time,
2653                         (unsigned long)stateid->si_boot)) {
2654                 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2655                         STATEID_VAL(stateid));
2656                 return 1;
2657         }
2658         return 0;
2659 }
2660
2661 static int
2662 EXPIRED_STATEID(stateid_t *stateid)
2663 {
2664         if (time_before((unsigned long)boot_time,
2665                         ((unsigned long)stateid->si_boot)) &&
2666             time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2667                 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2668                         STATEID_VAL(stateid));
2669                 return 1;
2670         }
2671         return 0;
2672 }
2673
2674 static __be32
2675 stateid_error_map(stateid_t *stateid)
2676 {
2677         if (STALE_STATEID(stateid))
2678                 return nfserr_stale_stateid;
2679         if (EXPIRED_STATEID(stateid))
2680                 return nfserr_expired;
2681
2682         dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2683                 STATEID_VAL(stateid));
2684         return nfserr_bad_stateid;
2685 }
2686
2687 static inline int
2688 access_permit_read(unsigned long access_bmap)
2689 {
2690         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2691                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2692                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2693 }
2694
2695 static inline int
2696 access_permit_write(unsigned long access_bmap)
2697 {
2698         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2699                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2700 }
2701
2702 static
2703 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2704 {
2705         __be32 status = nfserr_openmode;
2706
2707         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2708                 goto out;
2709         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2710                 goto out;
2711         status = nfs_ok;
2712 out:
2713         return status;
2714 }
2715
2716 static inline __be32
2717 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2718 {
2719         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2720                 return nfs_ok;
2721         else if (locks_in_grace()) {
2722                 /* Answer in remaining cases depends on existance of
2723                  * conflicting state; so we must wait out the grace period. */
2724                 return nfserr_grace;
2725         } else if (flags & WR_STATE)
2726                 return nfs4_share_conflict(current_fh,
2727                                 NFS4_SHARE_DENY_WRITE);
2728         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2729                 return nfs4_share_conflict(current_fh,
2730                                 NFS4_SHARE_DENY_READ);
2731 }
2732
2733 /*
2734  * Allow READ/WRITE during grace period on recovered state only for files
2735  * that are not able to provide mandatory locking.
2736  */
2737 static inline int
2738 grace_disallows_io(struct inode *inode)
2739 {
2740         return locks_in_grace() && mandatory_lock(inode);
2741 }
2742
2743 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2744 {
2745         /*
2746          * When sessions are used the stateid generation number is ignored
2747          * when it is zero.
2748          */
2749         if ((flags & HAS_SESSION) && in->si_generation == 0)
2750                 goto out;
2751
2752         /* If the client sends us a stateid from the future, it's buggy: */
2753         if (in->si_generation > ref->si_generation)
2754                 return nfserr_bad_stateid;
2755         /*
2756          * The following, however, can happen.  For example, if the
2757          * client sends an open and some IO at the same time, the open
2758          * may bump si_generation while the IO is still in flight.
2759          * Thanks to hard links and renames, the client never knows what
2760          * file an open will affect.  So it could avoid that situation
2761          * only by serializing all opens and IO from the same open
2762          * owner.  To recover from the old_stateid error, the client
2763          * will just have to retry the IO:
2764          */
2765         if (in->si_generation < ref->si_generation)
2766                 return nfserr_old_stateid;
2767 out:
2768         return nfs_ok;
2769 }
2770
2771 static int is_delegation_stateid(stateid_t *stateid)
2772 {
2773         return stateid->si_fileid == 0;
2774 }
2775
2776 /*
2777 * Checks for stateid operations
2778 */
2779 __be32
2780 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2781                            stateid_t *stateid, int flags, struct file **filpp)
2782 {
2783         struct nfs4_stateid *stp = NULL;
2784         struct nfs4_delegation *dp = NULL;
2785         struct svc_fh *current_fh = &cstate->current_fh;
2786         struct inode *ino = current_fh->fh_dentry->d_inode;
2787         __be32 status;
2788
2789         if (filpp)
2790                 *filpp = NULL;
2791
2792         if (grace_disallows_io(ino))
2793                 return nfserr_grace;
2794
2795         if (nfsd4_has_session(cstate))
2796                 flags |= HAS_SESSION;
2797
2798         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2799                 return check_special_stateids(current_fh, stateid, flags);
2800
2801         status = nfserr_stale_stateid;
2802         if (STALE_STATEID(stateid)) 
2803                 goto out;
2804
2805         status = nfserr_bad_stateid;
2806         if (is_delegation_stateid(stateid)) {
2807                 dp = find_delegation_stateid(ino, stateid);
2808                 if (!dp) {
2809                         status = stateid_error_map(stateid);
2810                         goto out;
2811                 }
2812                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2813                                                   flags);
2814                 if (status)
2815                         goto out;
2816                 status = nfs4_check_delegmode(dp, flags);
2817                 if (status)
2818                         goto out;
2819                 renew_client(dp->dl_client);
2820                 if (filpp)
2821                         *filpp = dp->dl_vfs_file;
2822         } else { /* open or lock stateid */
2823                 stp = find_stateid(stateid, flags);
2824                 if (!stp) {
2825                         status = stateid_error_map(stateid);
2826                         goto out;
2827                 }
2828                 if (nfs4_check_fh(current_fh, stp))
2829                         goto out;
2830                 if (!stp->st_stateowner->so_confirmed)
2831                         goto out;
2832                 status = check_stateid_generation(stateid, &stp->st_stateid,
2833                                                   flags);
2834                 if (status)
2835                         goto out;
2836                 status = nfs4_check_openmode(stp, flags);
2837                 if (status)
2838                         goto out;
2839                 renew_client(stp->st_stateowner->so_client);
2840                 if (filpp)
2841                         *filpp = stp->st_vfs_file;
2842         }
2843         status = nfs_ok;
2844 out:
2845         return status;
2846 }
2847
2848 static inline int
2849 setlkflg (int type)
2850 {
2851         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2852                 RD_STATE : WR_STATE;
2853 }
2854
2855 /* 
2856  * Checks for sequence id mutating operations. 
2857  */
2858 static __be32
2859 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2860                          stateid_t *stateid, int flags,
2861                          struct nfs4_stateowner **sopp,
2862                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2863 {
2864         struct nfs4_stateid *stp;
2865         struct nfs4_stateowner *sop;
2866         struct svc_fh *current_fh = &cstate->current_fh;
2867         __be32 status;
2868
2869         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2870                 seqid, STATEID_VAL(stateid));
2871
2872         *stpp = NULL;
2873         *sopp = NULL;
2874
2875         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2876                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2877                 return nfserr_bad_stateid;
2878         }
2879
2880         if (STALE_STATEID(stateid))
2881                 return nfserr_stale_stateid;
2882
2883         if (nfsd4_has_session(cstate))
2884                 flags |= HAS_SESSION;
2885
2886         /*
2887         * We return BAD_STATEID if filehandle doesn't match stateid, 
2888         * the confirmed flag is incorrecly set, or the generation 
2889         * number is incorrect.  
2890         */
2891         stp = find_stateid(stateid, flags);
2892         if (stp == NULL) {
2893                 /*
2894                  * Also, we should make sure this isn't just the result of
2895                  * a replayed close:
2896                  */
2897                 sop = search_close_lru(stateid->si_stateownerid, flags);
2898                 if (sop == NULL)
2899                         return stateid_error_map(stateid);
2900                 *sopp = sop;
2901                 goto check_replay;
2902         }
2903
2904         *stpp = stp;
2905         *sopp = sop = stp->st_stateowner;
2906
2907         if (lock) {
2908                 clientid_t *lockclid = &lock->v.new.clientid;
2909                 struct nfs4_client *clp = sop->so_client;
2910                 int lkflg = 0;
2911                 __be32 status;
2912
2913                 lkflg = setlkflg(lock->lk_type);
2914
2915                 if (lock->lk_is_new) {
2916                         if (!sop->so_is_open_owner)
2917                                 return nfserr_bad_stateid;
2918                         if (!(flags & HAS_SESSION) &&
2919                             !same_clid(&clp->cl_clientid, lockclid))
2920                                 return nfserr_bad_stateid;
2921                         /* stp is the open stateid */
2922                         status = nfs4_check_openmode(stp, lkflg);
2923                         if (status)
2924                                 return status;
2925                 } else {
2926                         /* stp is the lock stateid */
2927                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2928                         if (status)
2929                                 return status;
2930                }
2931         }
2932
2933         if (nfs4_check_fh(current_fh, stp)) {
2934                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2935                 return nfserr_bad_stateid;
2936         }
2937
2938         /*
2939         *  We now validate the seqid and stateid generation numbers.
2940         *  For the moment, we ignore the possibility of 
2941         *  generation number wraparound.
2942         */
2943         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2944                 goto check_replay;
2945
2946         if (sop->so_confirmed && flags & CONFIRM) {
2947                 dprintk("NFSD: preprocess_seqid_op: expected"
2948                                 " unconfirmed stateowner!\n");
2949                 return nfserr_bad_stateid;
2950         }
2951         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2952                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2953                                 " confirmed yet!\n");
2954                 return nfserr_bad_stateid;
2955         }
2956         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2957         if (status)
2958                 return status;
2959         renew_client(sop->so_client);
2960         return nfs_ok;
2961
2962 check_replay:
2963         if (seqid == sop->so_seqid - 1) {
2964                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2965                 /* indicate replay to calling function */
2966                 return nfserr_replay_me;
2967         }
2968         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2969                         sop->so_seqid, seqid);
2970         *sopp = NULL;
2971         return nfserr_bad_seqid;
2972 }
2973
2974 __be32
2975 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2976                    struct nfsd4_open_confirm *oc)
2977 {
2978         __be32 status;
2979         struct nfs4_stateowner *sop;
2980         struct nfs4_stateid *stp;
2981
2982         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2983                         (int)cstate->current_fh.fh_dentry->d_name.len,
2984                         cstate->current_fh.fh_dentry->d_name.name);
2985
2986         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2987         if (status)
2988                 return status;
2989
2990         nfs4_lock_state();
2991
2992         if ((status = nfs4_preprocess_seqid_op(cstate,
2993                                         oc->oc_seqid, &oc->oc_req_stateid,
2994                                         CONFIRM | OPEN_STATE,
2995                                         &oc->oc_stateowner, &stp, NULL)))
2996                 goto out; 
2997
2998         sop = oc->oc_stateowner;
2999         sop->so_confirmed = 1;
3000         update_stateid(&stp->st_stateid);
3001         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3002         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3003                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3004
3005         nfsd4_create_clid_dir(sop->so_client);
3006 out:
3007         if (oc->oc_stateowner) {
3008                 nfs4_get_stateowner(oc->oc_stateowner);
3009                 cstate->replay_owner = oc->oc_stateowner;
3010         }
3011         nfs4_unlock_state();
3012         return status;
3013 }
3014
3015
3016 /*
3017  * unset all bits in union bitmap (bmap) that
3018  * do not exist in share (from successful OPEN_DOWNGRADE)
3019  */
3020 static void
3021 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3022 {
3023         int i;
3024         for (i = 1; i < 4; i++) {
3025                 if ((i & access) != i)
3026                         __clear_bit(i, bmap);
3027         }
3028 }
3029
3030 static void
3031 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3032 {
3033         int i;
3034         for (i = 0; i < 4; i++) {
3035                 if ((i & deny) != i)
3036                         __clear_bit(i, bmap);
3037         }
3038 }
3039
3040 __be32
3041 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3042                      struct nfsd4_compound_state *cstate,
3043                      struct nfsd4_open_downgrade *od)
3044 {
3045         __be32 status;
3046         struct nfs4_stateid *stp;
3047         unsigned int share_access;
3048
3049         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3050                         (int)cstate->current_fh.fh_dentry->d_name.len,
3051                         cstate->current_fh.fh_dentry->d_name.name);
3052
3053         if (!access_valid(od->od_share_access, cstate->minorversion)
3054                         || !deny_valid(od->od_share_deny))
3055                 return nfserr_inval;
3056
3057         nfs4_lock_state();
3058         if ((status = nfs4_preprocess_seqid_op(cstate,
3059                                         od->od_seqid,
3060                                         &od->od_stateid, 
3061                                         OPEN_STATE,
3062                                         &od->od_stateowner, &stp, NULL)))
3063                 goto out; 
3064
3065         status = nfserr_inval;
3066         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3067                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3068                         stp->st_access_bmap, od->od_share_access);
3069                 goto out;
3070         }
3071         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3072                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3073                         stp->st_deny_bmap, od->od_share_deny);
3074                 goto out;
3075         }
3076         set_access(&share_access, stp->st_access_bmap);
3077         nfs4_file_downgrade(stp->st_vfs_file,
3078                             share_access & ~od->od_share_access);
3079
3080         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3081         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3082
3083         update_stateid(&stp->st_stateid);
3084         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3085         status = nfs_ok;
3086 out:
3087         if (od->od_stateowner) {
3088                 nfs4_get_stateowner(od->od_stateowner);
3089                 cstate->replay_owner = od->od_stateowner;
3090         }
3091         nfs4_unlock_state();
3092         return status;
3093 }
3094
3095 /*
3096  * nfs4_unlock_state() called after encode
3097  */
3098 __be32
3099 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3100             struct nfsd4_close *close)
3101 {
3102         __be32 status;
3103         struct nfs4_stateid *stp;
3104
3105         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3106                         (int)cstate->current_fh.fh_dentry->d_name.len,
3107                         cstate->current_fh.fh_dentry->d_name.name);
3108
3109         nfs4_lock_state();
3110         /* check close_lru for replay */
3111         if ((status = nfs4_preprocess_seqid_op(cstate,
3112                                         close->cl_seqid,
3113                                         &close->cl_stateid, 
3114                                         OPEN_STATE | CLOSE_STATE,
3115                                         &close->cl_stateowner, &stp, NULL)))
3116                 goto out; 
3117         status = nfs_ok;
3118         update_stateid(&stp->st_stateid);
3119         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3120
3121         /* release_stateid() calls nfsd_close() if needed */
3122         release_open_stateid(stp);
3123
3124         /* place unused nfs4_stateowners on so_close_lru list to be
3125          * released by the laundromat service after the lease period
3126          * to enable us to handle CLOSE replay
3127          */
3128         if (list_empty(&close->cl_stateowner->so_stateids))
3129                 move_to_close_lru(close->cl_stateowner);
3130 out:
3131         if (close->cl_stateowner) {
3132                 nfs4_get_stateowner(close->cl_stateowner);
3133                 cstate->replay_owner = close->cl_stateowner;
3134         }
3135         nfs4_unlock_state();
3136         return status;
3137 }
3138
3139 __be32
3140 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3141                   struct nfsd4_delegreturn *dr)
3142 {
3143         struct nfs4_delegation *dp;
3144         stateid_t *stateid = &dr->dr_stateid;
3145         struct inode *inode;
3146         __be32 status;
3147         int flags = 0;
3148
3149         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3150                 return status;
3151         inode = cstate->current_fh.fh_dentry->d_inode;
3152
3153         if (nfsd4_has_session(cstate))
3154                 flags |= HAS_SESSION;
3155         nfs4_lock_state();
3156         status = nfserr_bad_stateid;
3157         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3158                 goto out;
3159         status = nfserr_stale_stateid;
3160         if (STALE_STATEID(stateid))
3161                 goto out;
3162         status = nfserr_bad_stateid;
3163         if (!is_delegation_stateid(stateid))
3164                 goto out;
3165         dp = find_delegation_stateid(inode, stateid);
3166         if (!dp) {
3167                 status = stateid_error_map(stateid);
3168                 goto out;
3169         }
3170         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3171         if (status)
3172                 goto out;
3173         renew_client(dp->dl_client);
3174
3175         unhash_delegation(dp);
3176 out:
3177         nfs4_unlock_state();
3178
3179         return status;
3180 }
3181
3182
3183 /* 
3184  * Lock owner state (byte-range locks)
3185  */
3186 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3187 #define LOCK_HASH_BITS              8
3188 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3189 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3190
3191 static inline u64
3192 end_offset(u64 start, u64 len)
3193 {
3194         u64 end;
3195
3196         end = start + len;
3197         return end >= start ? end: NFS4_MAX_UINT64;
3198 }
3199
3200 /* last octet in a range */
3201 static inline u64
3202 last_byte_offset(u64 start, u64 len)
3203 {
3204         u64 end;
3205
3206         BUG_ON(!len);
3207         end = start + len;
3208         return end > start ? end - 1: NFS4_MAX_UINT64;
3209 }
3210
3211 #define lockownerid_hashval(id) \
3212         ((id) & LOCK_HASH_MASK)
3213
3214 static inline unsigned int
3215 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3216                 struct xdr_netobj *ownername)
3217 {
3218         return (file_hashval(inode) + cl_id
3219                         + opaque_hashval(ownername->data, ownername->len))
3220                 & LOCK_HASH_MASK;
3221 }
3222
3223 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3224 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3225 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3226
3227 static struct nfs4_stateid *
3228 find_stateid(stateid_t *stid, int flags)
3229 {
3230         struct nfs4_stateid *local;
3231         u32 st_id = stid->si_stateownerid;
3232         u32 f_id = stid->si_fileid;
3233         unsigned int hashval;
3234
3235         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3236         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3237                 hashval = stateid_hashval(st_id, f_id);
3238                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3239                         if ((local->st_stateid.si_stateownerid == st_id) &&
3240                             (local->st_stateid.si_fileid == f_id))
3241                                 return local;
3242                 }
3243         } 
3244
3245         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3246                 hashval = stateid_hashval(st_id, f_id);
3247                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3248                         if ((local->st_stateid.si_stateownerid == st_id) &&
3249                             (local->st_stateid.si_fileid == f_id))
3250                                 return local;
3251                 }
3252         }
3253         return NULL;
3254 }
3255
3256 static struct nfs4_delegation *
3257 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3258 {
3259         struct nfs4_file *fp;
3260         struct nfs4_delegation *dl;
3261
3262         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3263                 STATEID_VAL(stid));
3264
3265         fp = find_file(ino);
3266         if (!fp)
3267                 return NULL;
3268         dl = find_delegation_file(fp, stid);
3269         put_nfs4_file(fp);
3270         return dl;
3271 }
3272
3273 /*
3274  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3275  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3276  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3277  * locking, this prevents us from being completely protocol-compliant.  The
3278  * real solution to this problem is to start using unsigned file offsets in
3279  * the VFS, but this is a very deep change!
3280  */
3281 static inline void
3282 nfs4_transform_lock_offset(struct file_lock *lock)
3283 {
3284         if (lock->fl_start < 0)
3285                 lock->fl_start = OFFSET_MAX;
3286         if (lock->fl_end < 0)
3287                 lock->fl_end = OFFSET_MAX;
3288 }
3289
3290 /* Hack!: For now, we're defining this just so we can use a pointer to it
3291  * as a unique cookie to identify our (NFSv4's) posix locks. */
3292 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3293 };
3294
3295 static inline void
3296 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3297 {
3298         struct nfs4_stateowner *sop;
3299         unsigned int hval;
3300
3301         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3302                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3303                 hval = lockownerid_hashval(sop->so_id);
3304                 kref_get(&sop->so_ref);
3305                 deny->ld_sop = sop;
3306                 deny->ld_clientid = sop->so_client->cl_clientid;
3307         } else {
3308                 deny->ld_sop = NULL;
3309                 deny->ld_clientid.cl_boot = 0;
3310                 deny->ld_clientid.cl_id = 0;
3311         }
3312         deny->ld_start = fl->fl_start;
3313         deny->ld_length = NFS4_MAX_UINT64;
3314         if (fl->fl_end != NFS4_MAX_UINT64)
3315                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3316         deny->ld_type = NFS4_READ_LT;
3317         if (fl->fl_type != F_RDLCK)
3318                 deny->ld_type = NFS4_WRITE_LT;
3319 }
3320
3321 static struct nfs4_stateowner *
3322 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3323                 struct xdr_netobj *owner)
3324 {
3325         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3326         struct nfs4_stateowner *op;
3327
3328         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3329                 if (same_owner_str(op, owner, clid))
3330                         return op;
3331         }
3332         return NULL;
3333 }
3334
3335 /*
3336  * Alloc a lock owner structure.
3337  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3338  * occured. 
3339  *
3340  * strhashval = lock_ownerstr_hashval 
3341  */
3342
3343 static struct nfs4_stateowner *
3344 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3345         struct nfs4_stateowner *sop;
3346         struct nfs4_replay *rp;
3347         unsigned int idhashval;
3348
3349         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3350                 return NULL;
3351         idhashval = lockownerid_hashval(current_ownerid);
3352         INIT_LIST_HEAD(&sop->so_idhash);
3353         INIT_LIST_HEAD(&sop->so_strhash);
3354         INIT_LIST_HEAD(&sop->so_perclient);
3355         INIT_LIST_HEAD(&sop->so_stateids);
3356         INIT_LIST_HEAD(&sop->so_perstateid);
3357         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3358         sop->so_time = 0;
3359         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3360         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3361         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3362         sop->so_is_open_owner = 0;
3363         sop->so_id = current_ownerid++;
3364         sop->so_client = clp;
3365         /* It is the openowner seqid that will be incremented in encode in the
3366          * case of new lockowners; so increment the lock seqid manually: */
3367         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3368         sop->so_confirmed = 1;
3369         rp = &sop->so_replay;
3370         rp->rp_status = nfserr_serverfault;
3371         rp->rp_buflen = 0;
3372         rp->rp_buf = rp->rp_ibuf;
3373         return sop;
3374 }
3375
3376 static struct nfs4_stateid *
3377 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3378 {
3379         struct nfs4_stateid *stp;
3380         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3381
3382         stp = nfs4_alloc_stateid();
3383         if (stp == NULL)
3384                 goto out;
3385         INIT_LIST_HEAD(&stp->st_hash);
3386         INIT_LIST_HEAD(&stp->st_perfile);
3387         INIT_LIST_HEAD(&stp->st_perstateowner);
3388         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3389         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3390         list_add(&stp->st_perfile, &fp->fi_stateids);
3391         list_add(&stp->st_perstateowner, &sop->so_stateids);
3392         stp->st_stateowner = sop;
3393         get_nfs4_file(fp);
3394         stp->st_file = fp;
3395         stp->st_stateid.si_boot = get_seconds();
3396         stp->st_stateid.si_stateownerid = sop->so_id;
3397         stp->st_stateid.si_fileid = fp->fi_id;
3398         stp->st_stateid.si_generation = 0;
3399         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3400         stp->st_access_bmap = open_stp->st_access_bmap;
3401         stp->st_deny_bmap = open_stp->st_deny_bmap;
3402         stp->st_openstp = open_stp;
3403
3404 out:
3405         return stp;
3406 }
3407
3408 static int
3409 check_lock_length(u64 offset, u64 length)
3410 {
3411         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3412              LOFF_OVERFLOW(offset, length)));
3413 }
3414
3415 /*
3416  *  LOCK operation 
3417  */
3418 __be32
3419 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3420            struct nfsd4_lock *lock)
3421 {
3422         struct nfs4_stateowner *open_sop = NULL;
3423         struct nfs4_stateowner *lock_sop = NULL;
3424         struct nfs4_stateid *lock_stp;
3425         struct file *filp;
3426         struct file_lock file_lock;
3427         struct file_lock conflock;
3428         __be32 status = 0;
3429         unsigned int strhashval;
3430         unsigned int cmd;
3431         int err;
3432
3433         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3434                 (long long) lock->lk_offset,
3435                 (long long) lock->lk_length);
3436
3437         if (check_lock_length(lock->lk_offset, lock->lk_length))
3438                  return nfserr_inval;
3439
3440         if ((status = fh_verify(rqstp, &cstate->current_fh,
3441                                 S_IFREG, NFSD_MAY_LOCK))) {
3442                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3443                 return status;
3444         }
3445
3446         nfs4_lock_state();
3447
3448         if (lock->lk_is_new) {
3449                 /*
3450                  * Client indicates that this is a new lockowner.
3451                  * Use open owner and open stateid to create lock owner and
3452                  * lock stateid.
3453                  */
3454                 struct nfs4_stateid *open_stp = NULL;
3455                 struct nfs4_file *fp;
3456                 
3457                 status = nfserr_stale_clientid;
3458                 if (!nfsd4_has_session(cstate) &&
3459                     STALE_CLIENTID(&lock->lk_new_clientid))
3460                         goto out;
3461
3462                 /* validate and update open stateid and open seqid */
3463                 status = nfs4_preprocess_seqid_op(cstate,
3464                                         lock->lk_new_open_seqid,
3465                                         &lock->lk_new_open_stateid,
3466                                         OPEN_STATE,
3467                                         &lock->lk_replay_owner, &open_stp,
3468                                         lock);
3469                 if (status)
3470                         goto out;
3471                 open_sop = lock->lk_replay_owner;
3472                 /* create lockowner and lock stateid */
3473                 fp = open_stp->st_file;
3474                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3475                                 open_sop->so_client->cl_clientid.cl_id, 
3476                                 &lock->v.new.owner);
3477                 /* XXX: Do we need to check for duplicate stateowners on
3478                  * the same file, or should they just be allowed (and
3479                  * create new stateids)? */
3480                 status = nfserr_resource;
3481                 lock_sop = alloc_init_lock_stateowner(strhashval,
3482                                 open_sop->so_client, open_stp, lock);
3483                 if (lock_sop == NULL)
3484                         goto out;
3485                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3486                 if (lock_stp == NULL)
3487                         goto out;
3488         } else {
3489                 /* lock (lock owner + lock stateid) already exists */
3490                 status = nfs4_preprocess_seqid_op(cstate,
3491                                        lock->lk_old_lock_seqid, 
3492                                        &lock->lk_old_lock_stateid, 
3493                                        LOCK_STATE,
3494                                        &lock->lk_replay_owner, &lock_stp, lock);
3495                 if (status)
3496                         goto out;
3497                 lock_sop = lock->lk_replay_owner;
3498         }
3499         /* lock->lk_replay_owner and lock_stp have been created or found */
3500         filp = lock_stp->st_vfs_file;
3501
3502         status = nfserr_grace;
3503         if (locks_in_grace() && !lock->lk_reclaim)
3504                 goto out;
3505         status = nfserr_no_grace;
3506         if (!locks_in_grace() && lock->lk_reclaim)
3507                 goto out;
3508
3509         locks_init_lock(&file_lock);
3510         switch (lock->lk_type) {
3511                 case NFS4_READ_LT:
3512                 case NFS4_READW_LT:
3513                         file_lock.fl_type = F_RDLCK;
3514                         cmd = F_SETLK;
3515                 break;
3516                 case NFS4_WRITE_LT:
3517                 case NFS4_WRITEW_LT:
3518                         file_lock.fl_type = F_WRLCK;
3519                         cmd = F_SETLK;
3520                 break;
3521                 default:
3522                         status = nfserr_inval;
3523                 goto out;
3524         }
3525         file_lock.fl_owner = (fl_owner_t)lock_sop;
3526         file_lock.fl_pid = current->tgid;
3527         file_lock.fl_file = filp;
3528         file_lock.fl_flags = FL_POSIX;
3529         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3530
3531         file_lock.fl_start = lock->lk_offset;
3532         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3533         nfs4_transform_lock_offset(&file_lock);
3534
3535         /*
3536         * Try to lock the file in the VFS.
3537         * Note: locks.c uses the BKL to protect the inode's lock list.
3538         */
3539
3540         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3541         switch (-err) {
3542         case 0: /* success! */
3543                 update_stateid(&lock_stp->st_stateid);
3544                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3545                                 sizeof(stateid_t));
3546                 status = 0;
3547                 break;
3548         case (EAGAIN):          /* conflock holds conflicting lock */
3549                 status = nfserr_denied;
3550                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3551                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3552                 break;
3553         case (EDEADLK):
3554                 status = nfserr_deadlock;
3555                 break;
3556         default:        
3557                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3558                 status = nfserr_resource;
3559                 break;
3560         }
3561 out:
3562         if (status && lock->lk_is_new && lock_sop)
3563                 release_lockowner(lock_sop);
3564         if (lock->lk_replay_owner) {
3565                 nfs4_get_stateowner(lock->lk_replay_owner);
3566                 cstate->replay_owner = lock->lk_replay_owner;
3567         }
3568         nfs4_unlock_state();
3569         return status;
3570 }
3571
3572 /*
3573  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3574  * so we do a temporary open here just to get an open file to pass to
3575  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3576  * inode operation.)
3577  */
3578 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3579 {
3580         struct file *file;
3581         int err;
3582
3583         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3584         if (err)
3585                 return err;
3586         err = vfs_test_lock(file, lock);
3587         nfsd_close(file);
3588         return err;
3589 }
3590
3591 /*
3592  * LOCKT operation
3593  */
3594 __be32
3595 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3596             struct nfsd4_lockt *lockt)
3597 {
3598         struct inode *inode;
3599         struct file_lock file_lock;
3600         int error;
3601         __be32 status;
3602
3603         if (locks_in_grace())
3604                 return nfserr_grace;
3605
3606         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3607                  return nfserr_inval;
3608
3609         lockt->lt_stateowner = NULL;
3610         nfs4_lock_state();
3611
3612         status = nfserr_stale_clientid;
3613         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3614                 goto out;
3615
3616         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3617                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3618                 if (status == nfserr_symlink)
3619                         status = nfserr_inval;
3620                 goto out;
3621         }
3622
3623         inode = cstate->current_fh.fh_dentry->d_inode;
3624         locks_init_lock(&file_lock);
3625         switch (lockt->lt_type) {
3626                 case NFS4_READ_LT:
3627                 case NFS4_READW_LT:
3628                         file_lock.fl_type = F_RDLCK;
3629                 break;
3630                 case NFS4_WRITE_LT:
3631                 case NFS4_WRITEW_LT:
3632                         file_lock.fl_type = F_WRLCK;
3633                 break;
3634                 default:
3635                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3636                         status = nfserr_inval;
3637                 goto out;
3638         }
3639
3640         lockt->lt_stateowner = find_lockstateowner_str(inode,
3641                         &lockt->lt_clientid, &lockt->lt_owner);
3642         if (lockt->lt_stateowner)
3643                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3644         file_lock.fl_pid = current->tgid;
3645         file_lock.fl_flags = FL_POSIX;
3646
3647         file_lock.fl_start = lockt->lt_offset;
3648         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3649
3650         nfs4_transform_lock_offset(&file_lock);
3651
3652         status = nfs_ok;
3653         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3654         if (error) {
3655                 status = nfserrno(error);
3656                 goto out;
3657         }
3658         if (file_lock.fl_type != F_UNLCK) {
3659                 status = nfserr_denied;
3660                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3661         }
3662 out:
3663         nfs4_unlock_state();
3664         return status;
3665 }
3666
3667 __be32
3668 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3669             struct nfsd4_locku *locku)
3670 {
3671         struct nfs4_stateid *stp;
3672         struct file *filp = NULL;
3673         struct file_lock file_lock;
3674         __be32 status;
3675         int err;
3676                                                         
3677         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3678                 (long long) locku->lu_offset,
3679                 (long long) locku->lu_length);
3680
3681         if (check_lock_length(locku->lu_offset, locku->lu_length))
3682                  return nfserr_inval;
3683
3684         nfs4_lock_state();
3685                                                                                 
3686         if ((status = nfs4_preprocess_seqid_op(cstate,
3687                                         locku->lu_seqid, 
3688                                         &locku->lu_stateid, 
3689                                         LOCK_STATE,
3690                                         &locku->lu_stateowner, &stp, NULL)))
3691                 goto out;
3692
3693         filp = stp->st_vfs_file;
3694         BUG_ON(!filp);
3695         locks_init_lock(&file_lock);
3696         file_lock.fl_type = F_UNLCK;
3697         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3698         file_lock.fl_pid = current->tgid;
3699         file_lock.fl_file = filp;
3700         file_lock.fl_flags = FL_POSIX; 
3701         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3702         file_lock.fl_start = locku->lu_offset;
3703
3704         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3705         nfs4_transform_lock_offset(&file_lock);
3706
3707         /*
3708         *  Try to unlock the file in the VFS.
3709         */
3710         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3711         if (err) {
3712                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3713                 goto out_nfserr;
3714         }
3715         /*
3716         * OK, unlock succeeded; the only thing left to do is update the stateid.
3717         */
3718         update_stateid(&stp->st_stateid);
3719         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3720
3721 out:
3722         if (locku->lu_stateowner) {
3723                 nfs4_get_stateowner(locku->lu_stateowner);
3724                 cstate->replay_owner = locku->lu_stateowner;
3725         }
3726         nfs4_unlock_state();
3727         return status;
3728
3729 out_nfserr:
3730         status = nfserrno(err);
3731         goto out;
3732 }
3733
3734 /*
3735  * returns
3736  *      1: locks held by lockowner
3737  *      0: no locks held by lockowner
3738  */
3739 static int
3740 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3741 {
3742         struct file_lock **flpp;
3743         struct inode *inode = filp->f_path.dentry->d_inode;
3744         int status = 0;
3745
3746         lock_kernel();
3747         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3748                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3749                         status = 1;
3750                         goto out;
3751                 }
3752         }
3753 out:
3754         unlock_kernel();
3755         return status;
3756 }
3757
3758 __be32
3759 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3760                         struct nfsd4_compound_state *cstate,
3761                         struct nfsd4_release_lockowner *rlockowner)
3762 {
3763         clientid_t *clid = &rlockowner->rl_clientid;
3764         struct nfs4_stateowner *sop;
3765         struct nfs4_stateid *stp;
3766         struct xdr_netobj *owner = &rlockowner->rl_owner;
3767         struct list_head matches;
3768         int i;
3769         __be32 status;
3770
3771         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3772                 clid->cl_boot, clid->cl_id);
3773
3774         /* XXX check for lease expiration */
3775
3776         status = nfserr_stale_clientid;
3777         if (STALE_CLIENTID(clid))
3778                 return status;
3779
3780         nfs4_lock_state();
3781
3782         status = nfserr_locks_held;
3783         /* XXX: we're doing a linear search through all the lockowners.
3784          * Yipes!  For now we'll just hope clients aren't really using
3785          * release_lockowner much, but eventually we have to fix these
3786          * data structures. */
3787         INIT_LIST_HEAD(&matches);
3788         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3789                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3790                         if (!same_owner_str(sop, owner, clid))
3791                                 continue;
3792                         list_for_each_entry(stp, &sop->so_stateids,
3793                                         st_perstateowner) {
3794                                 if (check_for_locks(stp->st_vfs_file, sop))
3795                                         goto out;
3796                                 /* Note: so_perclient unused for lockowners,
3797                                  * so it's OK to fool with here. */
3798                                 list_add(&sop->so_perclient, &matches);
3799                         }
3800                 }
3801         }
3802         /* Clients probably won't expect us to return with some (but not all)
3803          * of the lockowner state released; so don't release any until all
3804          * have been checked. */
3805         status = nfs_ok;
3806         while (!list_empty(&matches)) {
3807                 sop = list_entry(matches.next, struct nfs4_stateowner,
3808                                                                 so_perclient);
3809                 /* unhash_stateowner deletes so_perclient only
3810                  * for openowners. */
3811                 list_del(&sop->so_perclient);
3812                 release_lockowner(sop);
3813         }
3814 out:
3815         nfs4_unlock_state();
3816         return status;
3817 }
3818
3819 static inline struct nfs4_client_reclaim *
3820 alloc_reclaim(void)
3821 {
3822         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3823 }
3824
3825 int
3826 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3827 {
3828         unsigned int strhashval = clientstr_hashval(name);
3829         struct nfs4_client *clp;
3830
3831         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3832         return clp ? 1 : 0;
3833 }
3834
3835 /*
3836  * failure => all reset bets are off, nfserr_no_grace...
3837  */
3838 int
3839 nfs4_client_to_reclaim(const char *name)
3840 {
3841         unsigned int strhashval;
3842         struct nfs4_client_reclaim *crp = NULL;
3843
3844         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3845         crp = alloc_reclaim();
3846         if (!crp)
3847                 return 0;
3848         strhashval = clientstr_hashval(name);
3849         INIT_LIST_HEAD(&crp->cr_strhash);
3850         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3851         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3852         reclaim_str_hashtbl_size++;
3853         return 1;
3854 }
3855
3856 static void
3857 nfs4_release_reclaim(void)
3858 {
3859         struct nfs4_client_reclaim *crp = NULL;
3860         int i;
3861
3862         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3863                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3864                         crp = list_entry(reclaim_str_hashtbl[i].next,
3865                                         struct nfs4_client_reclaim, cr_strhash);
3866                         list_del(&crp->cr_strhash);
3867                         kfree(crp);
3868                         reclaim_str_hashtbl_size--;
3869                 }
3870         }
3871         BUG_ON(reclaim_str_hashtbl_size);
3872 }
3873
3874 /*
3875  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3876 static struct nfs4_client_reclaim *
3877 nfs4_find_reclaim_client(clientid_t *clid)
3878 {
3879         unsigned int strhashval;
3880         struct nfs4_client *clp;
3881         struct nfs4_client_reclaim *crp = NULL;
3882
3883
3884         /* find clientid in conf_id_hashtbl */
3885         clp = find_confirmed_client(clid);
3886         if (clp == NULL)
3887                 return NULL;
3888
3889         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3890                             clp->cl_name.len, clp->cl_name.data,
3891                             clp->cl_recdir);
3892
3893         /* find clp->cl_name in reclaim_str_hashtbl */
3894         strhashval = clientstr_hashval(clp->cl_recdir);
3895         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3896                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3897                         return crp;
3898                 }
3899         }
3900         return NULL;
3901 }
3902
3903 /*
3904 * Called from OPEN. Look for clientid in reclaim list.
3905 */
3906 __be32
3907 nfs4_check_open_reclaim(clientid_t *clid)
3908 {
3909         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3910 }
3911
3912 /* initialization to perform at module load time: */
3913
3914 int
3915 nfs4_state_init(void)
3916 {
3917         int i, status;
3918
3919         status = nfsd4_init_slabs();
3920         if (status)
3921                 return status;
3922         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3923                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3924                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3925                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3926                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3927                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3928         }
3929         for (i = 0; i < SESSION_HASH_SIZE; i++)
3930                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3931         for (i = 0; i < FILE_HASH_SIZE; i++) {
3932                 INIT_LIST_HEAD(&file_hashtbl[i]);
3933         }
3934         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3935                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3936                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3937         }
3938         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3939                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3940                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3941         }
3942         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3943                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3944                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3945         }
3946         memset(&onestateid, ~0, sizeof(stateid_t));
3947         INIT_LIST_HEAD(&close_lru);
3948         INIT_LIST_HEAD(&client_lru);
3949         INIT_LIST_HEAD(&del_recall_lru);
3950         reclaim_str_hashtbl_size = 0;
3951         return 0;
3952 }
3953
3954 static void
3955 nfsd4_load_reboot_recovery_data(void)
3956 {
3957         int status;
3958
3959         nfs4_lock_state();
3960         nfsd4_init_recdir(user_recovery_dirname);
3961         status = nfsd4_recdir_load();
3962         nfs4_unlock_state();
3963         if (status)
3964                 printk("NFSD: Failure reading reboot recovery data\n");
3965 }
3966
3967 /*
3968  * Since the lifetime of a delegation isn't limited to that of an open, a
3969  * client may quite reasonably hang on to a delegation as long as it has
3970  * the inode cached.  This becomes an obvious problem the first time a
3971  * client's inode cache approaches the size of the server's total memory.
3972  *
3973  * For now we avoid this problem by imposing a hard limit on the number
3974  * of delegations, which varies according to the server's memory size.
3975  */
3976 static void
3977 set_max_delegations(void)
3978 {
3979         /*
3980          * Allow at most 4 delegations per megabyte of RAM.  Quick
3981          * estimates suggest that in the worst case (where every delegation
3982          * is for a different inode), a delegation could take about 1.5K,
3983          * giving a worst case usage of about 6% of memory.
3984          */
3985         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3986 }
3987
3988 /* initialization to perform when the nfsd service is started: */
3989
3990 static int
3991 __nfs4_state_start(void)
3992 {
3993         int ret;
3994
3995         boot_time = get_seconds();
3996         locks_start_grace(&nfsd4_manager);
3997         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3998                nfsd4_grace);
3999         ret = set_callback_cred();
4000         if (ret)
4001                 return -ENOMEM;
4002         laundry_wq = create_singlethread_workqueue("nfsd4");
4003         if (laundry_wq == NULL)
4004                 return -ENOMEM;
4005         ret = nfsd4_create_callback_queue();
4006         if (ret)
4007                 goto out_free_laundry;
4008         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4009         set_max_delegations();
4010         return 0;
4011 out_free_laundry:
4012         destroy_workqueue(laundry_wq);
4013         return ret;
4014 }
4015
4016 int
4017 nfs4_state_start(void)
4018 {
4019         int ret;
4020
4021         if (nfs4_init)
4022                 return 0;
4023         nfsd4_load_reboot_recovery_data();
4024         ret = __nfs4_state_start();
4025         if (ret)
4026                 return ret;
4027         nfs4_init = 1;
4028         return 0;
4029 }
4030
4031 static void
4032 __nfs4_state_shutdown(void)
4033 {
4034         int i;
4035         struct nfs4_client *clp = NULL;
4036         struct nfs4_delegation *dp = NULL;
4037         struct list_head *pos, *next, reaplist;
4038
4039         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4040                 while (!list_empty(&conf_id_hashtbl[i])) {
4041                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4042                         expire_client(clp);
4043                 }
4044                 while (!list_empty(&unconf_str_hashtbl[i])) {
4045                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4046                         expire_client(clp);
4047                 }
4048         }
4049         INIT_LIST_HEAD(&reaplist);
4050         spin_lock(&recall_lock);
4051         list_for_each_safe(pos, next, &del_recall_lru) {
4052                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4053                 list_move(&dp->dl_recall_lru, &reaplist);
4054         }
4055         spin_unlock(&recall_lock);
4056         list_for_each_safe(pos, next, &reaplist) {
4057                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4058                 list_del_init(&dp->dl_recall_lru);
4059                 unhash_delegation(dp);
4060         }
4061
4062         nfsd4_shutdown_recdir();
4063         nfs4_init = 0;
4064 }
4065
4066 void
4067 nfs4_state_shutdown(void)
4068 {
4069         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4070         destroy_workqueue(laundry_wq);
4071         locks_end_grace(&nfsd4_manager);
4072         nfs4_lock_state();
4073         nfs4_release_reclaim();
4074         __nfs4_state_shutdown();
4075         nfsd4_destroy_callback_queue();
4076         nfs4_unlock_state();
4077 }
4078
4079 /*
4080  * user_recovery_dirname