5051ade30dfb41c3c2a7e50502fe717eac2e51e2
[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         return dp;
202 }
203
204 void
205 nfs4_put_delegation(struct nfs4_delegation *dp)
206 {
207         if (atomic_dec_and_test(&dp->dl_count)) {
208                 dprintk("NFSD: freeing dp %p\n",dp);
209                 put_nfs4_file(dp->dl_file);
210                 kmem_cache_free(deleg_slab, dp);
211                 num_delegations--;
212         }
213 }
214
215 /* Remove the associated file_lock first, then remove the delegation.
216  * lease_modify() is called to remove the FS_LEASE file_lock from
217  * the i_flock list, eventually calling nfsd's lock_manager
218  * fl_release_callback.
219  */
220 static void
221 nfs4_close_delegation(struct nfs4_delegation *dp)
222 {
223         struct file *filp = dp->dl_vfs_file;
224
225         dprintk("NFSD: close_delegation dp %p\n",dp);
226         dp->dl_vfs_file = NULL;
227         /* The following nfsd_close may not actually close the file,
228          * but we want to remove the lease in any case. */
229         if (dp->dl_flock)
230                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
231         nfsd_close(filp);
232 }
233
234 /* Called under the state lock. */
235 static void
236 unhash_delegation(struct nfs4_delegation *dp)
237 {
238         list_del_init(&dp->dl_perfile);
239         list_del_init(&dp->dl_perclnt);
240         spin_lock(&recall_lock);
241         list_del_init(&dp->dl_recall_lru);
242         spin_unlock(&recall_lock);
243         nfs4_close_delegation(dp);
244         nfs4_put_delegation(dp);
245 }
246
247 /* 
248  * SETCLIENTID state 
249  */
250
251 /* Hash tables for nfs4_clientid state */
252 #define CLIENT_HASH_BITS                 4
253 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
254 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
255
256 #define clientid_hashval(id) \
257         ((id) & CLIENT_HASH_MASK)
258 #define clientstr_hashval(name) \
259         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
260 /*
261  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
262  * used in reboot/reset lease grace period processing
263  *
264  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
265  * setclientid_confirmed info. 
266  *
267  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
268  * setclientid info.
269  *
270  * client_lru holds client queue ordered by nfs4_client.cl_time
271  * for lease renewal.
272  *
273  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
274  * for last close replay.
275  */
276 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
277 static int reclaim_str_hashtbl_size = 0;
278 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
279 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
280 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head client_lru;
283 static struct list_head close_lru;
284
285 static void unhash_generic_stateid(struct nfs4_stateid *stp)
286 {
287         list_del(&stp->st_hash);
288         list_del(&stp->st_perfile);
289         list_del(&stp->st_perstateowner);
290 }
291
292 static void free_generic_stateid(struct nfs4_stateid *stp)
293 {
294         put_nfs4_file(stp->st_file);
295         kmem_cache_free(stateid_slab, stp);
296 }
297
298 static void release_lock_stateid(struct nfs4_stateid *stp)
299 {
300         unhash_generic_stateid(stp);
301         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
302         free_generic_stateid(stp);
303 }
304
305 static void unhash_lockowner(struct nfs4_stateowner *sop)
306 {
307         struct nfs4_stateid *stp;
308
309         list_del(&sop->so_idhash);
310         list_del(&sop->so_strhash);
311         list_del(&sop->so_perstateid);
312         while (!list_empty(&sop->so_stateids)) {
313                 stp = list_first_entry(&sop->so_stateids,
314                                 struct nfs4_stateid, st_perstateowner);
315                 release_lock_stateid(stp);
316         }
317 }
318
319 static void release_lockowner(struct nfs4_stateowner *sop)
320 {
321         unhash_lockowner(sop);
322         nfs4_put_stateowner(sop);
323 }
324
325 static void
326 release_stateid_lockowners(struct nfs4_stateid *open_stp)
327 {
328         struct nfs4_stateowner *lock_sop;
329
330         while (!list_empty(&open_stp->st_lockowners)) {
331                 lock_sop = list_entry(open_stp->st_lockowners.next,
332                                 struct nfs4_stateowner, so_perstateid);
333                 /* list_del(&open_stp->st_lockowners);  */
334                 BUG_ON(lock_sop->so_is_open_owner);
335                 release_lockowner(lock_sop);
336         }
337 }
338
339 static void release_open_stateid(struct nfs4_stateid *stp)
340 {
341         unhash_generic_stateid(stp);
342         release_stateid_lockowners(stp);
343         nfsd_close(stp->st_vfs_file);
344         free_generic_stateid(stp);
345 }
346
347 static void unhash_openowner(struct nfs4_stateowner *sop)
348 {
349         struct nfs4_stateid *stp;
350
351         list_del(&sop->so_idhash);
352         list_del(&sop->so_strhash);
353         list_del(&sop->so_perclient);
354         list_del(&sop->so_perstateid); /* XXX: necessary? */
355         while (!list_empty(&sop->so_stateids)) {
356                 stp = list_first_entry(&sop->so_stateids,
357                                 struct nfs4_stateid, st_perstateowner);
358                 release_open_stateid(stp);
359         }
360 }
361
362 static void release_openowner(struct nfs4_stateowner *sop)
363 {
364         unhash_openowner(sop);
365         list_del(&sop->so_close_lru);
366         nfs4_put_stateowner(sop);
367 }
368
369 static DEFINE_SPINLOCK(sessionid_lock);
370 #define SESSION_HASH_SIZE       512
371 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
372
373 static inline int
374 hash_sessionid(struct nfs4_sessionid *sessionid)
375 {
376         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
377
378         return sid->sequence % SESSION_HASH_SIZE;
379 }
380
381 static inline void
382 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
383 {
384         u32 *ptr = (u32 *)(&sessionid->data[0]);
385         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
386 }
387
388 static void
389 gen_sessionid(struct nfsd4_session *ses)
390 {
391         struct nfs4_client *clp = ses->se_client;
392         struct nfsd4_sessionid *sid;
393
394         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
395         sid->clientid = clp->cl_clientid;
396         sid->sequence = current_sessionid++;
397         sid->reserved = 0;
398 }
399
400 /*
401  * The protocol defines ca_maxresponssize_cached to include the size of
402  * the rpc header, but all we need to cache is the data starting after
403  * the end of the initial SEQUENCE operation--the rest we regenerate
404  * each time.  Therefore we can advertise a ca_maxresponssize_cached
405  * value that is the number of bytes in our cache plus a few additional
406  * bytes.  In order to stay on the safe side, and not promise more than
407  * we can cache, those additional bytes must be the minimum possible: 24
408  * bytes of rpc header (xid through accept state, with AUTH_NULL
409  * verifier), 12 for the compound header (with zero-length tag), and 44
410  * for the SEQUENCE op response:
411  */
412 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
413
414 /*
415  * Give the client the number of ca_maxresponsesize_cached slots it
416  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
417  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
418  * than NFSD_MAX_SLOTS_PER_SESSION.
419  *
420  * If we run out of reserved DRC memory we should (up to a point)
421  * re-negotiate active sessions and reduce their slot usage to make
422  * rooom for new connections. For now we just fail the create session.
423  */
424 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
425 {
426         int mem, size = fchan->maxresp_cached;
427
428         if (fchan->maxreqs < 1)
429                 return nfserr_inval;
430
431         if (size < NFSD_MIN_HDR_SEQ_SZ)
432                 size = NFSD_MIN_HDR_SEQ_SZ;
433         size -= NFSD_MIN_HDR_SEQ_SZ;
434         if (size > NFSD_SLOT_CACHE_SIZE)
435                 size = NFSD_SLOT_CACHE_SIZE;
436
437         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
438         mem = fchan->maxreqs * size;
439         if (mem > NFSD_MAX_MEM_PER_SESSION) {
440                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
441                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
442                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
443                 mem = fchan->maxreqs * size;
444         }
445
446         spin_lock(&nfsd_drc_lock);
447         /* bound the total session drc memory ussage */
448         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
449                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
450                 mem = fchan->maxreqs * size;
451         }
452         nfsd_drc_mem_used += mem;
453         spin_unlock(&nfsd_drc_lock);
454
455         if (fchan->maxreqs == 0)
456                 return nfserr_serverfault;
457
458         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
459         return 0;
460 }
461
462 /*
463  * fchan holds the client values on input, and the server values on output
464  * sv_max_mesg is the maximum payload plus one page for overhead.
465  */
466 static int init_forechannel_attrs(struct svc_rqst *rqstp,
467                                   struct nfsd4_channel_attrs *session_fchan,
468                                   struct nfsd4_channel_attrs *fchan)
469 {
470         int status = 0;
471         __u32   maxcount = nfsd_serv->sv_max_mesg;
472
473         /* headerpadsz set to zero in encode routine */
474
475         /* Use the client's max request and max response size if possible */
476         if (fchan->maxreq_sz > maxcount)
477                 fchan->maxreq_sz = maxcount;
478         session_fchan->maxreq_sz = fchan->maxreq_sz;
479
480         if (fchan->maxresp_sz > maxcount)
481                 fchan->maxresp_sz = maxcount;
482         session_fchan->maxresp_sz = fchan->maxresp_sz;
483
484         /* Use the client's maxops if possible */
485         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
486                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
487         session_fchan->maxops = fchan->maxops;
488
489         /* FIXME: Error means no more DRC pages so the server should
490          * recover pages from existing sessions. For now fail session
491          * creation.
492          */
493         status = set_forechannel_drc_size(fchan);
494
495         session_fchan->maxresp_cached = fchan->maxresp_cached;
496         session_fchan->maxreqs = fchan->maxreqs;
497
498         dprintk("%s status %d\n", __func__, status);
499         return status;
500 }
501
502 static void
503 free_session_slots(struct nfsd4_session *ses)
504 {
505         int i;
506
507         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
508                 kfree(ses->se_slots[i]);
509 }
510
511 /*
512  * We don't actually need to cache the rpc and session headers, so we
513  * can allocate a little less for each slot:
514  */
515 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
516 {
517         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
518 }
519
520 static int
521 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
522                    struct nfsd4_create_session *cses)
523 {
524         struct nfsd4_session *new, tmp;
525         struct nfsd4_slot *sp;
526         int idx, slotsize, cachesize, i;
527         int status;
528
529         memset(&tmp, 0, sizeof(tmp));
530
531         /* FIXME: For now, we just accept the client back channel attributes. */
532         tmp.se_bchannel = cses->back_channel;
533         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
534                                         &cses->fore_channel);
535         if (status)
536                 goto out;
537
538         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
539                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
540
541         status = nfserr_serverfault;
542         /* allocate struct nfsd4_session and slot table pointers in one piece */
543         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
544         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
545         if (!new)
546                 goto out;
547
548         memcpy(new, &tmp, sizeof(*new));
549
550         /* allocate each struct nfsd4_slot and data cache in one piece */
551         cachesize = slot_bytes(&new->se_fchannel);
552         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
553                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
554                 if (!sp)
555                         goto out_free;
556                 new->se_slots[i] = sp;
557         }
558
559         new->se_client = clp;
560         gen_sessionid(new);
561         idx = hash_sessionid(&new->se_sessionid);
562         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
563                NFS4_MAX_SESSIONID_LEN);
564
565         new->se_flags = cses->flags;
566         kref_init(&new->se_ref);
567         spin_lock(&sessionid_lock);
568         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
569         list_add(&new->se_perclnt, &clp->cl_sessions);
570         spin_unlock(&sessionid_lock);
571
572         status = nfs_ok;
573 out:
574         return status;
575 out_free:
576         free_session_slots(new);
577         kfree(new);
578         goto out;
579 }
580
581 /* caller must hold sessionid_lock */
582 static struct nfsd4_session *
583 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
584 {
585         struct nfsd4_session *elem;
586         int idx;
587
588         dump_sessionid(__func__, sessionid);
589         idx = hash_sessionid(sessionid);
590         dprintk("%s: idx is %d\n", __func__, idx);
591         /* Search in the appropriate list */
592         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
593                 dump_sessionid("list traversal", &elem->se_sessionid);
594                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
595                             NFS4_MAX_SESSIONID_LEN)) {
596                         return elem;
597                 }
598         }
599
600         dprintk("%s: session not found\n", __func__);
601         return NULL;
602 }
603
604 /* caller must hold sessionid_lock */
605 static void
606 unhash_session(struct nfsd4_session *ses)
607 {
608         list_del(&ses->se_hash);
609         list_del(&ses->se_perclnt);
610 }
611
612 static void
613 release_session(struct nfsd4_session *ses)
614 {
615         spin_lock(&sessionid_lock);
616         unhash_session(ses);
617         spin_unlock(&sessionid_lock);
618         nfsd4_put_session(ses);
619 }
620
621 void
622 free_session(struct kref *kref)
623 {
624         struct nfsd4_session *ses;
625         int mem;
626
627         ses = container_of(kref, struct nfsd4_session, se_ref);
628         spin_lock(&nfsd_drc_lock);
629         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
630         nfsd_drc_mem_used -= mem;
631         spin_unlock(&nfsd_drc_lock);
632         free_session_slots(ses);
633         kfree(ses);
634 }
635
636 static inline void
637 renew_client(struct nfs4_client *clp)
638 {
639         /*
640         * Move client to the end to the LRU list.
641         */
642         dprintk("renewing client (clientid %08x/%08x)\n", 
643                         clp->cl_clientid.cl_boot, 
644                         clp->cl_clientid.cl_id);
645         list_move_tail(&clp->cl_lru, &client_lru);
646         clp->cl_time = get_seconds();
647 }
648
649 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
650 static int
651 STALE_CLIENTID(clientid_t *clid)
652 {
653         if (clid->cl_boot == boot_time)
654                 return 0;
655         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
656                 clid->cl_boot, clid->cl_id, boot_time);
657         return 1;
658 }
659
660 /* 
661  * XXX Should we use a slab cache ?
662  * This type of memory management is somewhat inefficient, but we use it
663  * anyway since SETCLIENTID is not a common operation.
664  */
665 static struct nfs4_client *alloc_client(struct xdr_netobj name)
666 {
667         struct nfs4_client *clp;
668
669         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
670         if (clp == NULL)
671                 return NULL;
672         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
673         if (clp->cl_name.data == NULL) {
674                 kfree(clp);
675                 return NULL;
676         }
677         memcpy(clp->cl_name.data, name.data, name.len);
678         clp->cl_name.len = name.len;
679         return clp;
680 }
681
682 static void
683 shutdown_callback_client(struct nfs4_client *clp)
684 {
685         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
686
687         if (clnt) {
688                 /*
689                  * Callback threads take a reference on the client, so there
690                  * should be no outstanding callbacks at this point.
691                  */
692                 clp->cl_cb_conn.cb_client = NULL;
693                 rpc_shutdown_client(clnt);
694         }
695 }
696
697 static inline void
698 free_client(struct nfs4_client *clp)
699 {
700         if (clp->cl_cred.cr_group_info)
701                 put_group_info(clp->cl_cred.cr_group_info);
702         kfree(clp->cl_principal);
703         kfree(clp->cl_name.data);
704         kfree(clp);
705 }
706
707 void
708 put_nfs4_client(struct nfs4_client *clp)
709 {
710         if (atomic_dec_and_test(&clp->cl_count))
711                 free_client(clp);
712 }
713
714 static void
715 expire_client(struct nfs4_client *clp)
716 {
717         struct nfs4_stateowner *sop;
718         struct nfs4_delegation *dp;
719         struct list_head reaplist;
720
721         INIT_LIST_HEAD(&reaplist);
722         spin_lock(&recall_lock);
723         while (!list_empty(&clp->cl_delegations)) {
724                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
725                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
726                                 dp->dl_flock);
727                 list_del_init(&dp->dl_perclnt);
728                 list_move(&dp->dl_recall_lru, &reaplist);
729         }
730         spin_unlock(&recall_lock);
731         while (!list_empty(&reaplist)) {
732                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
733                 list_del_init(&dp->dl_recall_lru);
734                 unhash_delegation(dp);
735         }
736         list_del(&clp->cl_idhash);
737         list_del(&clp->cl_strhash);
738         list_del(&clp->cl_lru);
739         while (!list_empty(&clp->cl_openowners)) {
740                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
741                 release_openowner(sop);
742         }
743         while (!list_empty(&clp->cl_sessions)) {
744                 struct nfsd4_session  *ses;
745                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
746                                  se_perclnt);
747                 release_session(ses);
748         }
749         shutdown_callback_client(clp);
750         if (clp->cl_cb_xprt)
751                 svc_xprt_put(clp->cl_cb_xprt);
752         put_nfs4_client(clp);
753 }
754
755 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
756 {
757         memcpy(target->cl_verifier.data, source->data,
758                         sizeof(target->cl_verifier.data));
759 }
760
761 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
762 {
763         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
764         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
765 }
766
767 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
768 {
769         target->cr_uid = source->cr_uid;
770         target->cr_gid = source->cr_gid;
771         target->cr_group_info = source->cr_group_info;
772         get_group_info(target->cr_group_info);
773 }
774
775 static int same_name(const char *n1, const char *n2)
776 {
777         return 0 == memcmp(n1, n2, HEXDIR_LEN);
778 }
779
780 static int
781 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
782 {
783         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
784 }
785
786 static int
787 same_clid(clientid_t *cl1, clientid_t *cl2)
788 {
789         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
790 }
791
792 /* XXX what about NGROUP */
793 static int
794 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
795 {
796         return cr1->cr_uid == cr2->cr_uid;
797 }
798
799 static void gen_clid(struct nfs4_client *clp)
800 {
801         static u32 current_clientid = 1;
802
803         clp->cl_clientid.cl_boot = boot_time;
804         clp->cl_clientid.cl_id = current_clientid++; 
805 }
806
807 static void gen_confirm(struct nfs4_client *clp)
808 {
809         static u32 i;
810         u32 *p;
811
812         p = (u32 *)clp->cl_confirm.data;
813         *p++ = get_seconds();
814         *p++ = i++;
815 }
816
817 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
818                 struct svc_rqst *rqstp, nfs4_verifier *verf)
819 {
820         struct nfs4_client *clp;
821         struct sockaddr *sa = svc_addr(rqstp);
822         char *princ;
823
824         clp = alloc_client(name);
825         if (clp == NULL)
826                 return NULL;
827
828         princ = svc_gss_principal(rqstp);
829         if (princ) {
830                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
831                 if (clp->cl_principal == NULL) {
832                         free_client(clp);
833                         return NULL;
834                 }
835         }
836
837         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
838         atomic_set(&clp->cl_count, 1);
839         atomic_set(&clp->cl_cb_conn.cb_set, 0);
840         INIT_LIST_HEAD(&clp->cl_idhash);
841         INIT_LIST_HEAD(&clp->cl_strhash);
842         INIT_LIST_HEAD(&clp->cl_openowners);
843         INIT_LIST_HEAD(&clp->cl_delegations);
844         INIT_LIST_HEAD(&clp->cl_sessions);
845         INIT_LIST_HEAD(&clp->cl_lru);
846         clear_bit(0, &clp->cl_cb_slot_busy);
847         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
848         copy_verf(clp, verf);
849         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
850         clp->cl_flavor = rqstp->rq_flavor;
851         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
852         gen_confirm(clp);
853
854         return clp;
855 }
856
857 static int check_name(struct xdr_netobj name)
858 {
859         if (name.len == 0) 
860                 return 0;
861         if (name.len > NFS4_OPAQUE_LIMIT) {
862                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
863                 return 0;
864         }
865         return 1;
866 }
867
868 static void
869 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
870 {
871         unsigned int idhashval;
872
873         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
874         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
875         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
876         list_add_tail(&clp->cl_lru, &client_lru);
877         clp->cl_time = get_seconds();
878 }
879
880 static void
881 move_to_confirmed(struct nfs4_client *clp)
882 {
883         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
884         unsigned int strhashval;
885
886         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
887         list_del_init(&clp->cl_strhash);
888         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
889         strhashval = clientstr_hashval(clp->cl_recdir);
890         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
891         renew_client(clp);
892 }
893
894 static struct nfs4_client *
895 find_confirmed_client(clientid_t *clid)
896 {
897         struct nfs4_client *clp;
898         unsigned int idhashval = clientid_hashval(clid->cl_id);
899
900         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
901                 if (same_clid(&clp->cl_clientid, clid))
902                         return clp;
903         }
904         return NULL;
905 }
906
907 static struct nfs4_client *
908 find_unconfirmed_client(clientid_t *clid)
909 {
910         struct nfs4_client *clp;
911         unsigned int idhashval = clientid_hashval(clid->cl_id);
912
913         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
914                 if (same_clid(&clp->cl_clientid, clid))
915                         return clp;
916         }
917         return NULL;
918 }
919
920 /*
921  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
922  * parameter. Matching is based on the fact the at least one of the
923  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
924  *
925  * FIXME: we need to unify the clientid namespaces for nfsv4.x
926  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
927  * and SET_CLIENTID{,_CONFIRM}
928  */
929 static inline int
930 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
931 {
932         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
933         return use_exchange_id == has_exchange_flags;
934 }
935
936 static struct nfs4_client *
937 find_confirmed_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, &conf_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 struct nfs4_client *
951 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
952                                bool use_exchange_id)
953 {
954         struct nfs4_client *clp;
955
956         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
957                 if (same_name(clp->cl_recdir, dname) &&
958                     match_clientid_establishment(clp, use_exchange_id))
959                         return clp;
960         }
961         return NULL;
962 }
963
964 static void
965 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
966 {
967         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
968         unsigned short expected_family;
969
970         /* Currently, we only support tcp and tcp6 for the callback channel */
971         if (se->se_callback_netid_len == 3 &&
972             !memcmp(se->se_callback_netid_val, "tcp", 3))
973                 expected_family = AF_INET;
974         else if (se->se_callback_netid_len == 4 &&
975                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
976                 expected_family = AF_INET6;
977         else
978                 goto out_err;
979
980         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
981                                             se->se_callback_addr_len,
982                                             (struct sockaddr *) &cb->cb_addr,
983                                             sizeof(cb->cb_addr));
984
985         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
986                 goto out_err;
987
988         if (cb->cb_addr.ss_family == AF_INET6)
989                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
990
991         cb->cb_minorversion = 0;
992         cb->cb_prog = se->se_callback_prog;
993         cb->cb_ident = se->se_callback_ident;
994         return;
995 out_err:
996         cb->cb_addr.ss_family = AF_UNSPEC;
997         cb->cb_addrlen = 0;
998         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
999                 "will not receive delegations\n",
1000                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1001
1002         return;
1003 }
1004
1005 /*
1006  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1007  */
1008 void
1009 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1010 {
1011         struct nfsd4_slot *slot = resp->cstate.slot;
1012         unsigned int base;
1013
1014         dprintk("--> %s slot %p\n", __func__, slot);
1015
1016         slot->sl_opcnt = resp->opcnt;
1017         slot->sl_status = resp->cstate.status;
1018
1019         if (nfsd4_not_cached(resp)) {
1020                 slot->sl_datalen = 0;
1021                 return;
1022         }
1023         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1024         base = (char *)resp->cstate.datap -
1025                                         (char *)resp->xbuf->head[0].iov_base;
1026         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1027                                     slot->sl_datalen))
1028                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1029         return;
1030 }
1031
1032 /*
1033  * Encode the replay sequence operation from the slot values.
1034  * If cachethis is FALSE encode the uncached rep error on the next
1035  * operation which sets resp->p and increments resp->opcnt for
1036  * nfs4svc_encode_compoundres.
1037  *
1038  */
1039 static __be32
1040 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1041                           struct nfsd4_compoundres *resp)
1042 {
1043         struct nfsd4_op *op;
1044         struct nfsd4_slot *slot = resp->cstate.slot;
1045
1046         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1047                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1048
1049         /* Encode the replayed sequence operation */
1050         op = &args->ops[resp->opcnt - 1];
1051         nfsd4_encode_operation(resp, op);
1052
1053         /* Return nfserr_retry_uncached_rep in next operation. */
1054         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1055                 op = &args->ops[resp->opcnt++];
1056                 op->status = nfserr_retry_uncached_rep;
1057                 nfsd4_encode_operation(resp, op);
1058         }
1059         return op->status;
1060 }
1061
1062 /*
1063  * The sequence operation is not cached because we can use the slot and
1064  * session values.
1065  */
1066 __be32
1067 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1068                          struct nfsd4_sequence *seq)
1069 {
1070         struct nfsd4_slot *slot = resp->cstate.slot;
1071         __be32 status;
1072
1073         dprintk("--> %s slot %p\n", __func__, slot);
1074
1075         /* Either returns 0 or nfserr_retry_uncached */
1076         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1077         if (status == nfserr_retry_uncached_rep)
1078                 return status;
1079
1080         /* The sequence operation has been encoded, cstate->datap set. */
1081         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1082
1083         resp->opcnt = slot->sl_opcnt;
1084         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1085         status = slot->sl_status;
1086
1087         return status;
1088 }
1089
1090 /*
1091  * Set the exchange_id flags returned by the server.
1092  */
1093 static void
1094 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1095 {
1096         /* pNFS is not supported */
1097         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1098
1099         /* Referrals are supported, Migration is not. */
1100         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1101
1102         /* set the wire flags to return to client. */
1103         clid->flags = new->cl_exchange_flags;
1104 }
1105
1106 __be32
1107 nfsd4_exchange_id(struct svc_rqst *rqstp,
1108                   struct nfsd4_compound_state *cstate,
1109                   struct nfsd4_exchange_id *exid)
1110 {
1111         struct nfs4_client *unconf, *conf, *new;
1112         int status;
1113         unsigned int            strhashval;
1114         char                    dname[HEXDIR_LEN];
1115         char                    addr_str[INET6_ADDRSTRLEN];
1116         nfs4_verifier           verf = exid->verifier;
1117         struct sockaddr         *sa = svc_addr(rqstp);
1118
1119         rpc_ntop(sa, addr_str, sizeof(addr_str));
1120         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1121                 "ip_addr=%s flags %x, spa_how %d\n",
1122                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1123                 addr_str, exid->flags, exid->spa_how);
1124
1125         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1126                 return nfserr_inval;
1127
1128         /* Currently only support SP4_NONE */
1129         switch (exid->spa_how) {
1130         case SP4_NONE:
1131                 break;
1132         case SP4_SSV:
1133                 return nfserr_encr_alg_unsupp;
1134         default:
1135                 BUG();                          /* checked by xdr code */
1136         case SP4_MACH_CRED:
1137                 return nfserr_serverfault;      /* no excuse :-/ */
1138         }
1139
1140         status = nfs4_make_rec_clidname(dname, &exid->clname);
1141
1142         if (status)
1143                 goto error;
1144
1145         strhashval = clientstr_hashval(dname);
1146
1147         nfs4_lock_state();
1148         status = nfs_ok;
1149
1150         conf = find_confirmed_client_by_str(dname, strhashval, true);
1151         if (conf) {
1152                 if (!same_verf(&verf, &conf->cl_verifier)) {
1153                         /* 18.35.4 case 8 */
1154                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1155                                 status = nfserr_not_same;
1156                                 goto out;
1157                         }
1158                         /* Client reboot: destroy old state */
1159                         expire_client(conf);
1160                         goto out_new;
1161                 }
1162                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1163                         /* 18.35.4 case 9 */
1164                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1165                                 status = nfserr_perm;
1166                                 goto out;
1167                         }
1168                         expire_client(conf);
1169                         goto out_new;
1170                 }
1171                 /*
1172                  * Set bit when the owner id and verifier map to an already
1173                  * confirmed client id (18.35.3).
1174                  */
1175                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1176
1177                 /*
1178                  * Falling into 18.35.4 case 2, possible router replay.
1179                  * Leave confirmed record intact and return same result.
1180                  */
1181                 copy_verf(conf, &verf);
1182                 new = conf;
1183                 goto out_copy;
1184         }
1185
1186         /* 18.35.4 case 7 */
1187         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1188                 status = nfserr_noent;
1189                 goto out;
1190         }
1191
1192         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1193         if (unconf) {
1194                 /*
1195                  * Possible retry or client restart.  Per 18.35.4 case 4,
1196                  * a new unconfirmed record should be generated regardless
1197                  * of whether any properties have changed.
1198                  */
1199                 expire_client(unconf);
1200         }
1201
1202 out_new:
1203         /* Normal case */
1204         new = create_client(exid->clname, dname, rqstp, &verf);
1205         if (new == NULL) {
1206                 status = nfserr_serverfault;
1207                 goto out;
1208         }
1209
1210         gen_clid(new);
1211         add_to_unconfirmed(new, strhashval);
1212 out_copy:
1213         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1214         exid->clientid.cl_id = new->cl_clientid.cl_id;
1215
1216         exid->seqid = 1;
1217         nfsd4_set_ex_flags(new, exid);
1218
1219         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1220                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1221         status = nfs_ok;
1222
1223 out:
1224         nfs4_unlock_state();
1225 error:
1226         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1227         return status;
1228 }
1229
1230 static int
1231 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1232 {
1233         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1234                 slot_seqid);
1235
1236         /* The slot is in use, and no response has been sent. */
1237         if (slot_inuse) {
1238                 if (seqid == slot_seqid)
1239                         return nfserr_jukebox;
1240                 else
1241                         return nfserr_seq_misordered;
1242         }
1243         /* Normal */
1244         if (likely(seqid == slot_seqid + 1))
1245                 return nfs_ok;
1246         /* Replay */
1247         if (seqid == slot_seqid)
1248                 return nfserr_replay_cache;
1249         /* Wraparound */
1250         if (seqid == 1 && (slot_seqid + 1) == 0)
1251                 return nfs_ok;
1252         /* Misordered replay or misordered new request */
1253         return nfserr_seq_misordered;
1254 }
1255
1256 /*
1257  * Cache the create session result into the create session single DRC
1258  * slot cache by saving the xdr structure. sl_seqid has been set.
1259  * Do this for solo or embedded create session operations.
1260  */
1261 static void
1262 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1263                            struct nfsd4_clid_slot *slot, int nfserr)
1264 {
1265         slot->sl_status = nfserr;
1266         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1267 }
1268
1269 static __be32
1270 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1271                             struct nfsd4_clid_slot *slot)
1272 {
1273         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1274         return slot->sl_status;
1275 }
1276
1277 __be32
1278 nfsd4_create_session(struct svc_rqst *rqstp,
1279                      struct nfsd4_compound_state *cstate,
1280                      struct nfsd4_create_session *cr_ses)
1281 {
1282         struct sockaddr *sa = svc_addr(rqstp);
1283         struct nfs4_client *conf, *unconf;
1284         struct nfsd4_clid_slot *cs_slot = NULL;
1285         int status = 0;
1286
1287         nfs4_lock_state();
1288         unconf = find_unconfirmed_client(&cr_ses->clientid);
1289         conf = find_confirmed_client(&cr_ses->clientid);
1290
1291         if (conf) {
1292                 cs_slot = &conf->cl_cs_slot;
1293                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1294                 if (status == nfserr_replay_cache) {
1295                         dprintk("Got a create_session replay! seqid= %d\n",
1296                                 cs_slot->sl_seqid);
1297                         /* Return the cached reply status */
1298                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1299                         goto out;
1300                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1301                         status = nfserr_seq_misordered;
1302                         dprintk("Sequence misordered!\n");
1303                         dprintk("Expected seqid= %d but got seqid= %d\n",
1304                                 cs_slot->sl_seqid, cr_ses->seqid);
1305                         goto out;
1306                 }
1307                 cs_slot->sl_seqid++;
1308         } else if (unconf) {
1309                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1310                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1311                         status = nfserr_clid_inuse;
1312                         goto out;
1313                 }
1314
1315                 cs_slot = &unconf->cl_cs_slot;
1316                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1317                 if (status) {
1318                         /* an unconfirmed replay returns misordered */
1319                         status = nfserr_seq_misordered;
1320                         goto out_cache;
1321                 }
1322
1323                 cs_slot->sl_seqid++; /* from 0 to 1 */
1324                 move_to_confirmed(unconf);
1325
1326                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1327                         unconf->cl_cb_xprt = rqstp->rq_xprt;
1328                         svc_xprt_get(unconf->cl_cb_xprt);
1329                         rpc_copy_addr(
1330                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1331                                 sa);
1332                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1333                         unconf->cl_cb_conn.cb_minorversion =
1334                                 cstate->minorversion;
1335                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1336                         unconf->cl_cb_seq_nr = 1;
1337                         nfsd4_probe_callback(unconf);
1338                 }
1339                 conf = unconf;
1340         } else {
1341                 status = nfserr_stale_clientid;
1342                 goto out;
1343         }
1344
1345         /*
1346          * We do not support RDMA or persistent sessions
1347          */
1348         cr_ses->flags &= ~SESSION4_PERSIST;
1349         cr_ses->flags &= ~SESSION4_RDMA;
1350
1351         status = alloc_init_session(rqstp, conf, cr_ses);
1352         if (status)
1353                 goto out;
1354
1355         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1356                NFS4_MAX_SESSIONID_LEN);
1357         cr_ses->seqid = cs_slot->sl_seqid;
1358
1359 out_cache:
1360         /* cache solo and embedded create sessions under the state lock */
1361         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1362 out:
1363         nfs4_unlock_state();
1364         dprintk("%s returns %d\n", __func__, ntohl(status));
1365         return status;
1366 }
1367
1368 __be32
1369 nfsd4_destroy_session(struct svc_rqst *r,
1370                       struct nfsd4_compound_state *cstate,
1371                       struct nfsd4_destroy_session *sessionid)
1372 {
1373         struct nfsd4_session *ses;
1374         u32 status = nfserr_badsession;
1375
1376         /* Notes:
1377          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1378          * - Should we return nfserr_back_chan_busy if waiting for
1379          *   callbacks on to-be-destroyed session?
1380          * - Do we need to clear any callback info from previous session?
1381          */
1382
1383         dump_sessionid(__func__, &sessionid->sessionid);
1384         spin_lock(&sessionid_lock);
1385         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1386         if (!ses) {
1387                 spin_unlock(&sessionid_lock);
1388                 goto out;
1389         }
1390
1391         unhash_session(ses);
1392         spin_unlock(&sessionid_lock);
1393
1394         /* wait for callbacks */
1395         shutdown_callback_client(ses->se_client);
1396         nfsd4_put_session(ses);
1397         status = nfs_ok;
1398 out:
1399         dprintk("%s returns %d\n", __func__, ntohl(status));
1400         return status;
1401 }
1402
1403 __be32
1404 nfsd4_sequence(struct svc_rqst *rqstp,
1405                struct nfsd4_compound_state *cstate,
1406                struct nfsd4_sequence *seq)
1407 {
1408         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1409         struct nfsd4_session *session;
1410         struct nfsd4_slot *slot;
1411         int status;
1412
1413         if (resp->opcnt != 1)
1414                 return nfserr_sequence_pos;
1415
1416         spin_lock(&sessionid_lock);
1417         status = nfserr_badsession;
1418         session = find_in_sessionid_hashtbl(&seq->sessionid);
1419         if (!session)
1420                 goto out;
1421
1422         status = nfserr_badslot;
1423         if (seq->slotid >= session->se_fchannel.maxreqs)
1424                 goto out;
1425
1426         slot = session->se_slots[seq->slotid];
1427         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1428
1429         /* We do not negotiate the number of slots yet, so set the
1430          * maxslots to the session maxreqs which is used to encode
1431          * sr_highest_slotid and the sr_target_slot id to maxslots */
1432         seq->maxslots = session->se_fchannel.maxreqs;
1433
1434         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1435         if (status == nfserr_replay_cache) {
1436                 cstate->slot = slot;
1437                 cstate->session = session;
1438                 /* Return the cached reply status and set cstate->status
1439                  * for nfsd4_proc_compound processing */
1440                 status = nfsd4_replay_cache_entry(resp, seq);
1441                 cstate->status = nfserr_replay_cache;
1442                 goto out;
1443         }
1444         if (status)
1445                 goto out;
1446
1447         /* Success! bump slot seqid */
1448         slot->sl_inuse = true;
1449         slot->sl_seqid = seq->seqid;
1450         slot->sl_cachethis = seq->cachethis;
1451
1452         cstate->slot = slot;
1453         cstate->session = session;
1454
1455         /* Hold a session reference until done processing the compound:
1456          * nfsd4_put_session called only if the cstate slot is set.
1457          */
1458         nfsd4_get_session(session);
1459 out:
1460         spin_unlock(&sessionid_lock);
1461         /* Renew the clientid on success and on replay */
1462         if (cstate->session) {
1463                 nfs4_lock_state();
1464                 renew_client(session->se_client);
1465                 nfs4_unlock_state();
1466         }
1467         dprintk("%s: return %d\n", __func__, ntohl(status));
1468         return status;
1469 }
1470
1471 __be32
1472 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1473                   struct nfsd4_setclientid *setclid)
1474 {
1475         struct sockaddr         *sa = svc_addr(rqstp);
1476         struct xdr_netobj       clname = { 
1477                 .len = setclid->se_namelen,
1478                 .data = setclid->se_name,
1479         };
1480         nfs4_verifier           clverifier = setclid->se_verf;
1481         unsigned int            strhashval;
1482         struct nfs4_client      *conf, *unconf, *new;
1483         __be32                  status;
1484         char                    dname[HEXDIR_LEN];
1485         
1486         if (!check_name(clname))
1487                 return nfserr_inval;
1488
1489         status = nfs4_make_rec_clidname(dname, &clname);
1490         if (status)
1491                 return status;
1492
1493         /* 
1494          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1495          * We get here on a DRC miss.
1496          */
1497
1498         strhashval = clientstr_hashval(dname);
1499
1500         nfs4_lock_state();
1501         conf = find_confirmed_client_by_str(dname, strhashval, false);
1502         if (conf) {
1503                 /* RFC 3530 14.2.33 CASE 0: */
1504                 status = nfserr_clid_inuse;
1505                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1506                         char addr_str[INET6_ADDRSTRLEN];
1507                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1508                                  sizeof(addr_str));
1509                         dprintk("NFSD: setclientid: string in use by client "
1510                                 "at %s\n", addr_str);
1511                         goto out;
1512                 }
1513         }
1514         /*
1515          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1516          * has a description of SETCLIENTID request processing consisting
1517          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1518          */
1519         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1520         status = nfserr_resource;
1521         if (!conf) {
1522                 /*
1523                  * RFC 3530 14.2.33 CASE 4:
1524                  * placed first, because it is the normal case
1525                  */
1526                 if (unconf)
1527                         expire_client(unconf);
1528                 new = create_client(clname, dname, rqstp, &clverifier);
1529                 if (new == NULL)
1530                         goto out;
1531                 gen_clid(new);
1532         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1533                 /*
1534                  * RFC 3530 14.2.33 CASE 1:
1535                  * probable callback update
1536                  */
1537                 if (unconf) {
1538                         /* Note this is removing unconfirmed {*x***},
1539                          * which is stronger than RFC recommended {vxc**}.
1540                          * This has the advantage that there is at most
1541                          * one {*x***} in either list at any time.
1542                          */
1543                         expire_client(unconf);
1544                 }
1545                 new = create_client(clname, dname, rqstp, &clverifier);
1546                 if (new == NULL)
1547                         goto out;
1548                 copy_clid(new, conf);
1549         } else if (!unconf) {
1550                 /*
1551                  * RFC 3530 14.2.33 CASE 2:
1552                  * probable client reboot; state will be removed if
1553                  * confirmed.
1554                  */
1555                 new = create_client(clname, dname, rqstp, &clverifier);
1556                 if (new == NULL)
1557                         goto out;
1558                 gen_clid(new);
1559         } else {
1560                 /*
1561                  * RFC 3530 14.2.33 CASE 3:
1562                  * probable client reboot; state will be removed if
1563                  * confirmed.
1564                  */
1565                 expire_client(unconf);
1566                 new = create_client(clname, dname, rqstp, &clverifier);
1567                 if (new == NULL)
1568                         goto out;
1569                 gen_clid(new);
1570         }
1571         gen_callback(new, setclid, rpc_get_scope_id(sa));
1572         add_to_unconfirmed(new, strhashval);
1573         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1574         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1575         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1576         status = nfs_ok;
1577 out:
1578         nfs4_unlock_state();
1579         return status;
1580 }
1581
1582
1583 /*
1584  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1585  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1586  * bullets, labeled as CASE1 - CASE4 below.
1587  */
1588 __be32
1589 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1590                          struct nfsd4_compound_state *cstate,
1591                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1592 {
1593         struct sockaddr *sa = svc_addr(rqstp);
1594         struct nfs4_client *conf, *unconf;
1595         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1596         clientid_t * clid = &setclientid_confirm->sc_clientid;
1597         __be32 status;
1598
1599         if (STALE_CLIENTID(clid))
1600                 return nfserr_stale_clientid;
1601         /* 
1602          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1603          * We get here on a DRC miss.
1604          */
1605
1606         nfs4_lock_state();
1607
1608         conf = find_confirmed_client(clid);
1609         unconf = find_unconfirmed_client(clid);
1610
1611         status = nfserr_clid_inuse;
1612         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1613                 goto out;
1614         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1615                 goto out;
1616
1617         /*
1618          * section 14.2.34 of RFC 3530 has a description of
1619          * SETCLIENTID_CONFIRM request processing consisting
1620          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1621          */
1622         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1623                 /*
1624                  * RFC 3530 14.2.34 CASE 1:
1625                  * callback update
1626                  */
1627                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1628                         status = nfserr_clid_inuse;
1629                 else {
1630                         /* XXX: We just turn off callbacks until we can handle
1631                           * change request correctly. */
1632                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1633                         expire_client(unconf);
1634                         status = nfs_ok;
1635
1636                 }
1637         } else if (conf && !unconf) {
1638                 /*
1639                  * RFC 3530 14.2.34 CASE 2:
1640                  * probable retransmitted request; play it safe and
1641                  * do nothing.
1642                  */
1643                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1644                         status = nfserr_clid_inuse;
1645                 else
1646                         status = nfs_ok;
1647         } else if (!conf && unconf
1648                         && same_verf(&unconf->cl_confirm, &confirm)) {
1649                 /*
1650                  * RFC 3530 14.2.34 CASE 3:
1651                  * Normal case; new or rebooted client:
1652                  */
1653                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1654                         status = nfserr_clid_inuse;
1655                 } else {
1656                         unsigned int hash =
1657                                 clientstr_hashval(unconf->cl_recdir);
1658                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1659                                                             hash, false);
1660                         if (conf) {
1661                                 nfsd4_remove_clid_dir(conf);
1662                                 expire_client(conf);
1663                         }
1664                         move_to_confirmed(unconf);
1665                         conf = unconf;
1666                         nfsd4_probe_callback(conf);
1667                         status = nfs_ok;
1668                 }
1669         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1670             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1671                                                                 &confirm)))) {
1672                 /*
1673                  * RFC 3530 14.2.34 CASE 4:
1674                  * Client probably hasn't noticed that we rebooted yet.
1675                  */
1676                 status = nfserr_stale_clientid;
1677         } else {
1678                 /* check that we have hit one of the cases...*/
1679                 status = nfserr_clid_inuse;
1680         }
1681 out:
1682         nfs4_unlock_state();
1683         return status;
1684 }
1685
1686 /* OPEN Share state helper functions */
1687 static inline struct nfs4_file *
1688 alloc_init_file(struct inode *ino)
1689 {
1690         struct nfs4_file *fp;
1691         unsigned int hashval = file_hashval(ino);
1692
1693         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1694         if (fp) {
1695                 atomic_set(&fp->fi_ref, 1);
1696                 INIT_LIST_HEAD(&fp->fi_hash);
1697                 INIT_LIST_HEAD(&fp->fi_stateids);
1698                 INIT_LIST_HEAD(&fp->fi_delegations);
1699                 spin_lock(&recall_lock);
1700                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1701                 spin_unlock(&recall_lock);
1702                 fp->fi_inode = igrab(ino);
1703                 fp->fi_id = current_fileid++;
1704                 fp->fi_had_conflict = false;
1705                 return fp;
1706         }
1707         return NULL;
1708 }
1709
1710 static void
1711 nfsd4_free_slab(struct kmem_cache **slab)
1712 {
1713         if (*slab == NULL)
1714                 return;
1715         kmem_cache_destroy(*slab);
1716         *slab = NULL;
1717 }
1718
1719 void
1720 nfsd4_free_slabs(void)
1721 {
1722         nfsd4_free_slab(&stateowner_slab);
1723         nfsd4_free_slab(&file_slab);
1724         nfsd4_free_slab(&stateid_slab);
1725         nfsd4_free_slab(&deleg_slab);
1726 }
1727
1728 static int
1729 nfsd4_init_slabs(void)
1730 {
1731         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1732                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1733         if (stateowner_slab == NULL)
1734                 goto out_nomem;
1735         file_slab = kmem_cache_create("nfsd4_files",
1736                         sizeof(struct nfs4_file), 0, 0, NULL);
1737         if (file_slab == NULL)
1738                 goto out_nomem;
1739         stateid_slab = kmem_cache_create("nfsd4_stateids",
1740                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1741         if (stateid_slab == NULL)
1742                 goto out_nomem;
1743         deleg_slab = kmem_cache_create("nfsd4_delegations",
1744                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1745         if (deleg_slab == NULL)
1746                 goto out_nomem;
1747         return 0;
1748 out_nomem:
1749         nfsd4_free_slabs();
1750         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1751         return -ENOMEM;
1752 }
1753
1754 void
1755 nfs4_free_stateowner(struct kref *kref)
1756 {
1757         struct nfs4_stateowner *sop =
1758                 container_of(kref, struct nfs4_stateowner, so_ref);
1759         kfree(sop->so_owner.data);
1760         kmem_cache_free(stateowner_slab, sop);
1761 }
1762
1763 static inline struct nfs4_stateowner *
1764 alloc_stateowner(struct xdr_netobj *owner)
1765 {
1766         struct nfs4_stateowner *sop;
1767
1768         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1769                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1770                         memcpy(sop->so_owner.data, owner->data, owner->len);
1771                         sop->so_owner.len = owner->len;
1772                         kref_init(&sop->so_ref);
1773                         return sop;
1774                 } 
1775                 kmem_cache_free(stateowner_slab, sop);
1776         }
1777         return NULL;
1778 }
1779
1780 static struct nfs4_stateowner *
1781 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1782         struct nfs4_stateowner *sop;
1783         struct nfs4_replay *rp;
1784         unsigned int idhashval;
1785
1786         if (!(sop = alloc_stateowner(&open->op_owner)))
1787                 return NULL;
1788         idhashval = ownerid_hashval(current_ownerid);
1789         INIT_LIST_HEAD(&sop->so_idhash);
1790         INIT_LIST_HEAD(&sop->so_strhash);
1791         INIT_LIST_HEAD(&sop->so_perclient);
1792         INIT_LIST_HEAD(&sop->so_stateids);
1793         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1794         INIT_LIST_HEAD(&sop->so_close_lru);
1795         sop->so_time = 0;
1796         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1797         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1798         list_add(&sop->so_perclient, &clp->cl_openowners);
1799         sop->so_is_open_owner = 1;
1800         sop->so_id = current_ownerid++;
1801         sop->so_client = clp;
1802         sop->so_seqid = open->op_seqid;
1803         sop->so_confirmed = 0;
1804         rp = &sop->so_replay;
1805         rp->rp_status = nfserr_serverfault;
1806         rp->rp_buflen = 0;
1807         rp->rp_buf = rp->rp_ibuf;
1808         return sop;
1809 }
1810
1811 static inline void
1812 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1813         struct nfs4_stateowner *sop = open->op_stateowner;
1814         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1815
1816         INIT_LIST_HEAD(&stp->st_hash);
1817         INIT_LIST_HEAD(&stp->st_perstateowner);
1818         INIT_LIST_HEAD(&stp->st_lockowners);
1819         INIT_LIST_HEAD(&stp->st_perfile);
1820         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1821         list_add(&stp->st_perstateowner, &sop->so_stateids);
1822         list_add(&stp->st_perfile, &fp->fi_stateids);
1823         stp->st_stateowner = sop;
1824         get_nfs4_file(fp);
1825         stp->st_file = fp;
1826         stp->st_stateid.si_boot = get_seconds();
1827         stp->st_stateid.si_stateownerid = sop->so_id;
1828         stp->st_stateid.si_fileid = fp->fi_id;
1829         stp->st_stateid.si_generation = 0;
1830         stp->st_access_bmap = 0;
1831         stp->st_deny_bmap = 0;
1832         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1833                   &stp->st_access_bmap);
1834         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1835         stp->st_openstp = NULL;
1836 }
1837
1838 static void
1839 move_to_close_lru(struct nfs4_stateowner *sop)
1840 {
1841         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1842
1843         list_move_tail(&sop->so_close_lru, &close_lru);
1844         sop->so_time = get_seconds();
1845 }
1846
1847 static int
1848 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1849                                                         clientid_t *clid)
1850 {
1851         return (sop->so_owner.len == owner->len) &&
1852                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1853                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1854 }
1855
1856 static struct nfs4_stateowner *
1857 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1858 {
1859         struct nfs4_stateowner *so = NULL;
1860
1861         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1862                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1863                         return so;
1864         }
1865         return NULL;
1866 }
1867
1868 /* search file_hashtbl[] for file */
1869 static struct nfs4_file *
1870 find_file(struct inode *ino)
1871 {
1872         unsigned int hashval = file_hashval(ino);
1873         struct nfs4_file *fp;
1874
1875         spin_lock(&recall_lock);
1876         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1877                 if (fp->fi_inode == ino) {
1878                         get_nfs4_file(fp);
1879                         spin_unlock(&recall_lock);
1880                         return fp;
1881                 }
1882         }
1883         spin_unlock(&recall_lock);
1884         return NULL;
1885 }
1886
1887 static inline int access_valid(u32 x, u32 minorversion)
1888 {
1889         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1890                 return 0;
1891         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1892                 return 0;
1893         x &= ~NFS4_SHARE_ACCESS_MASK;
1894         if (minorversion && x) {
1895                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1896                         return 0;
1897                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1898                         return 0;
1899                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1900         }
1901         if (x)
1902                 return 0;
1903         return 1;
1904 }
1905
1906 static inline int deny_valid(u32 x)
1907 {
1908         /* Note: unlike access bits, deny bits may be zero. */
1909         return x <= NFS4_SHARE_DENY_BOTH;
1910 }
1911
1912 /*
1913  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1914  * st_{access,deny}_bmap field of the stateid, in order to track not
1915  * only what share bits are currently in force, but also what
1916  * combinations of share bits previous opens have used.  This allows us
1917  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1918  * return an error if the client attempt to downgrade to a combination
1919  * of share bits not explicable by closing some of its previous opens.
1920  *
1921  * XXX: This enforcement is actually incomplete, since we don't keep
1922  * track of access/deny bit combinations; so, e.g., we allow:
1923  *
1924  *      OPEN allow read, deny write
1925  *      OPEN allow both, deny none
1926  *      DOWNGRADE allow read, deny none
1927  *
1928  * which we should reject.
1929  */
1930 static void
1931 set_access(unsigned int *access, unsigned long bmap) {
1932         int i;
1933
1934         *access = 0;
1935         for (i = 1; i < 4; i++) {
1936                 if (test_bit(i, &bmap))
1937                         *access |= i;
1938         }
1939 }
1940
1941 static void
1942 set_deny(unsigned int *deny, unsigned long bmap) {
1943         int i;
1944
1945         *deny = 0;
1946         for (i = 0; i < 4; i++) {
1947                 if (test_bit(i, &bmap))
1948                         *deny |= i ;
1949         }
1950 }
1951
1952 static int
1953 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1954         unsigned int access, deny;
1955
1956         set_access(&access, stp->st_access_bmap);
1957         set_deny(&deny, stp->st_deny_bmap);
1958         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1959                 return 0;
1960         return 1;
1961 }
1962
1963 /*
1964  * Called to check deny when READ with all zero stateid or
1965  * WRITE with all zero or all one stateid
1966  */
1967 static __be32
1968 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1969 {
1970         struct inode *ino = current_fh->fh_dentry->d_inode;
1971         struct nfs4_file *fp;
1972         struct nfs4_stateid *stp;
1973         __be32 ret;
1974
1975         dprintk("NFSD: nfs4_share_conflict\n");
1976
1977         fp = find_file(ino);
1978         if (!fp)
1979                 return nfs_ok;
1980         ret = nfserr_locked;
1981         /* Search for conflicting share reservations */
1982         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1983                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1984                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1985                         goto out;
1986         }
1987         ret = nfs_ok;
1988 out:
1989         put_nfs4_file(fp);
1990         return ret;
1991 }
1992
1993 static inline void
1994 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1995 {
1996         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1997                 drop_file_write_access(filp);
1998                 spin_lock(&filp->f_lock);
1999                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2000                 spin_unlock(&filp->f_lock);
2001         }
2002 }
2003
2004 /*
2005  * Spawn a thread to perform a recall on the delegation represented
2006  * by the lease (file_lock)
2007  *
2008  * Called from break_lease() with lock_kernel() held.
2009  * Note: we assume break_lease will only call this *once* for any given
2010  * lease.
2011  */
2012 static
2013 void nfsd_break_deleg_cb(struct file_lock *fl)
2014 {
2015         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2016
2017         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2018         if (!dp)
2019                 return;
2020
2021         /* We're assuming the state code never drops its reference
2022          * without first removing the lease.  Since we're in this lease
2023          * callback (and since the lease code is serialized by the kernel
2024          * lock) we know the server hasn't removed the lease yet, we know
2025          * it's safe to take a reference: */
2026         atomic_inc(&dp->dl_count);
2027         atomic_inc(&dp->dl_client->cl_count);
2028
2029         spin_lock(&recall_lock);
2030         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2031         spin_unlock(&recall_lock);
2032
2033         /* only place dl_time is set. protected by lock_kernel*/
2034         dp->dl_time = get_seconds();
2035
2036         /*
2037          * We don't want the locks code to timeout the lease for us;
2038          * we'll remove it ourself if the delegation isn't returned
2039          * in time.
2040          */
2041         fl->fl_break_time = 0;
2042
2043         dp->dl_file->fi_had_conflict = true;
2044         nfsd4_cb_recall(dp);
2045 }
2046
2047 /*
2048  * The file_lock is being reapd.
2049  *
2050  * Called by locks_free_lock() with lock_kernel() held.
2051  */
2052 static
2053 void nfsd_release_deleg_cb(struct file_lock *fl)
2054 {
2055         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2056
2057         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2058
2059         if (!(fl->fl_flags & FL_LEASE) || !dp)
2060                 return;
2061         dp->dl_flock = NULL;
2062 }
2063
2064 /*
2065  * Set the delegation file_lock back pointer.
2066  *
2067  * Called from setlease() with lock_kernel() held.
2068  */
2069 static
2070 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2071 {
2072         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2073
2074         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2075         if (!dp)
2076                 return;
2077         dp->dl_flock = new;
2078 }
2079
2080 /*
2081  * Called from setlease() with lock_kernel() held
2082  */
2083 static
2084 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2085 {
2086         struct nfs4_delegation *onlistd =
2087                 (struct nfs4_delegation *)onlist->fl_owner;
2088         struct nfs4_delegation *tryd =
2089                 (struct nfs4_delegation *)try->fl_owner;
2090
2091         if (onlist->fl_lmops != try->fl_lmops)
2092                 return 0;
2093
2094         return onlistd->dl_client == tryd->dl_client;
2095 }
2096
2097
2098 static
2099 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2100 {
2101         if (arg & F_UNLCK)
2102                 return lease_modify(onlist, arg);
2103         else
2104                 return -EAGAIN;
2105 }
2106
2107 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2108         .fl_break = nfsd_break_deleg_cb,
2109         .fl_release_private = nfsd_release_deleg_cb,
2110         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2111         .fl_mylease = nfsd_same_client_deleg_cb,
2112         .fl_change = nfsd_change_deleg_cb,
2113 };
2114
2115
2116 __be32
2117 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2118                     struct nfsd4_open *open)
2119 {
2120         clientid_t *clientid = &open->op_clientid;
2121         struct nfs4_client *clp = NULL;
2122         unsigned int strhashval;
2123         struct nfs4_stateowner *sop = NULL;
2124
2125         if (!check_name(open->op_owner))
2126                 return nfserr_inval;
2127
2128         if (STALE_CLIENTID(&open->op_clientid))
2129                 return nfserr_stale_clientid;
2130
2131         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2132         sop = find_openstateowner_str(strhashval, open);
2133         open->op_stateowner = sop;
2134         if (!sop) {
2135                 /* Make sure the client's lease hasn't expired. */
2136                 clp = find_confirmed_client(clientid);
2137                 if (clp == NULL)
2138                         return nfserr_expired;
2139                 goto renew;
2140         }
2141         /* When sessions are used, skip open sequenceid processing */
2142         if (nfsd4_has_session(cstate))
2143                 goto renew;
2144         if (!sop->so_confirmed) {
2145                 /* Replace unconfirmed owners without checking for replay. */
2146                 clp = sop->so_client;
2147                 release_openowner(sop);
2148                 open->op_stateowner = NULL;
2149                 goto renew;
2150         }
2151         if (open->op_seqid == sop->so_seqid - 1) {
2152                 if (sop->so_replay.rp_buflen)
2153                         return nfserr_replay_me;
2154                 /* The original OPEN failed so spectacularly
2155                  * that we don't even have replay data saved!
2156                  * Therefore, we have no choice but to continue
2157                  * processing this OPEN; presumably, we'll
2158                  * fail again for the same reason.
2159                  */
2160                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2161                 goto renew;
2162         }
2163         if (open->op_seqid != sop->so_seqid)
2164                 return nfserr_bad_seqid;
2165 renew:
2166         if (open->op_stateowner == NULL) {
2167                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2168                 if (sop == NULL)
2169                         return nfserr_resource;
2170                 open->op_stateowner = sop;
2171         }
2172         list_del_init(&sop->so_close_lru);
2173         renew_client(sop->so_client);
2174         return nfs_ok;
2175 }
2176
2177 static inline __be32
2178 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2179 {
2180         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2181                 return nfserr_openmode;
2182         else
2183                 return nfs_ok;
2184 }
2185
2186 static struct nfs4_delegation *
2187 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2188 {
2189         struct nfs4_delegation *dp;
2190
2191         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2192                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2193                         return dp;
2194         }
2195         return NULL;
2196 }
2197
2198 static __be32
2199 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2200                 struct nfs4_delegation **dp)
2201 {
2202         int flags;
2203         __be32 status = nfserr_bad_stateid;
2204
2205         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2206         if (*dp == NULL)
2207                 goto out;
2208         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2209                                                 RD_STATE : WR_STATE;
2210         status = nfs4_check_delegmode(*dp, flags);
2211         if (status)
2212                 *dp = NULL;
2213 out:
2214         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2215                 return nfs_ok;
2216         if (status)
2217                 return status;
2218         open->op_stateowner->so_confirmed = 1;
2219         return nfs_ok;
2220 }
2221
2222 static __be32
2223 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2224 {
2225         struct nfs4_stateid *local;
2226         __be32 status = nfserr_share_denied;
2227         struct nfs4_stateowner *sop = open->op_stateowner;
2228
2229         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2230                 /* ignore lock owners */
2231                 if (local->st_stateowner->so_is_open_owner == 0)
2232                         continue;
2233                 /* remember if we have seen this open owner */
2234                 if (local->st_stateowner == sop)
2235                         *stpp = local;
2236                 /* check for conflicting share reservations */
2237                 if (!test_share(local, open))
2238                         goto out;
2239         }
2240         status = 0;
2241 out:
2242         return status;
2243 }
2244
2245 static inline struct nfs4_stateid *
2246 nfs4_alloc_stateid(void)
2247 {
2248         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2249 }
2250
2251 static __be32
2252 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2253                 struct nfs4_delegation *dp,
2254                 struct svc_fh *cur_fh, int flags)
2255 {
2256         struct nfs4_stateid *stp;
2257
2258         stp = nfs4_alloc_stateid();
2259         if (stp == NULL)
2260                 return nfserr_resource;
2261
2262         if (dp) {
2263                 get_file(dp->dl_vfs_file);
2264                 stp->st_vfs_file = dp->dl_vfs_file;
2265         } else {
2266                 __be32 status;
2267                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2268                                 &stp->st_vfs_file);
2269                 if (status) {
2270                         if (status == nfserr_dropit)
2271                                 status = nfserr_jukebox;
2272                         kmem_cache_free(stateid_slab, stp);
2273                         return status;
2274                 }
2275         }
2276         *stpp = stp;
2277         return 0;
2278 }
2279
2280 static inline __be32
2281 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2282                 struct nfsd4_open *open)
2283 {
2284         struct iattr iattr = {
2285                 .ia_valid = ATTR_SIZE,
2286                 .ia_size = 0,
2287         };
2288         if (!open->op_truncate)
2289                 return 0;
2290         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2291                 return nfserr_inval;
2292         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2293 }
2294
2295 static __be32
2296 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2297 {
2298         struct file *filp = stp->st_vfs_file;
2299         struct inode *inode = filp->f_path.dentry->d_inode;
2300         unsigned int share_access, new_writer;
2301         __be32 status;
2302
2303         set_access(&share_access, stp->st_access_bmap);
2304         new_writer = (~share_access) & open->op_share_access
2305                         & NFS4_SHARE_ACCESS_WRITE;
2306
2307         if (new_writer) {
2308                 int err = get_write_access(inode);
2309                 if (err)
2310                         return nfserrno(err);
2311                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2312                 if (err)
2313                         return nfserrno(err);
2314                 file_take_write(filp);
2315         }
2316         status = nfsd4_truncate(rqstp, cur_fh, open);
2317         if (status) {
2318                 if (new_writer)
2319                         put_write_access(inode);
2320                 return status;
2321         }
2322         /* remember the open */
2323         filp->f_mode |= open->op_share_access;
2324         __set_bit(open->op_share_access, &stp->st_access_bmap);
2325         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2326
2327         return nfs_ok;
2328 }
2329
2330
2331 static void
2332 nfs4_set_claim_prev(struct nfsd4_open *open)
2333 {
2334         open->op_stateowner->so_confirmed = 1;
2335         open->op_stateowner->so_client->cl_firststate = 1;
2336 }
2337
2338 /*
2339  * Attempt to hand out a delegation.
2340  */
2341 static void
2342 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2343 {
2344         struct nfs4_delegation *dp;
2345         struct nfs4_stateowner *sop = stp->st_stateowner;
2346         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2347         struct file_lock fl, *flp = &fl;
2348         int status, flag = 0;
2349
2350         flag = NFS4_OPEN_DELEGATE_NONE;
2351         open->op_recall = 0;
2352         switch (open->op_claim_type) {
2353                 case NFS4_OPEN_CLAIM_PREVIOUS:
2354                         if (!atomic_read(&cb->cb_set))
2355                                 open->op_recall = 1;
2356                         flag = open->op_delegate_type;
2357                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2358                                 goto out;
2359                         break;
2360                 case NFS4_OPEN_CLAIM_NULL:
2361                         /* Let's not give out any delegations till everyone's
2362                          * had the chance to reclaim theirs.... */
2363                         if (locks_in_grace())
2364                                 goto out;
2365                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2366                                 goto out;
2367                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2368                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2369                         else
2370                                 flag = NFS4_OPEN_DELEGATE_READ;
2371                         break;
2372                 default:
2373                         goto out;
2374         }
2375
2376         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2377         if (dp == NULL) {
2378                 flag = NFS4_OPEN_DELEGATE_NONE;
2379                 goto out;
2380         }
2381         locks_init_lock(&fl);
2382         fl.fl_lmops = &nfsd_lease_mng_ops;
2383         fl.fl_flags = FL_LEASE;
2384         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2385         fl.fl_end = OFFSET_MAX;
2386         fl.fl_owner =  (fl_owner_t)dp;
2387         fl.fl_file = stp->st_vfs_file;
2388         fl.fl_pid = current->tgid;
2389
2390         /* vfs_setlease checks to see if delegation should be handed out.
2391          * the lock_manager callbacks fl_mylease and fl_change are used
2392          */
2393         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2394                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2395                 unhash_delegation(dp);
2396                 flag = NFS4_OPEN_DELEGATE_NONE;
2397                 goto out;
2398         }
2399
2400         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2401
2402         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2403                 STATEID_VAL(&dp->dl_stateid));
2404 out:
2405         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2406                         && flag == NFS4_OPEN_DELEGATE_NONE
2407                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2408                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2409         open->op_delegate_type = flag;
2410 }
2411
2412 /*
2413  * called with nfs4_lock_state() held.
2414  */
2415 __be32
2416 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2417 {
2418         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2419         struct nfs4_file *fp = NULL;
2420         struct inode *ino = current_fh->fh_dentry->d_inode;
2421         struct nfs4_stateid *stp = NULL;
2422         struct nfs4_delegation *dp = NULL;
2423         __be32 status;
2424
2425         status = nfserr_inval;
2426         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2427                         || !deny_valid(open->op_share_deny))
2428                 goto out;
2429         /*
2430          * Lookup file; if found, lookup stateid and check open request,
2431          * and check for delegations in the process of being recalled.
2432          * If not found, create the nfs4_file struct
2433          */
2434         fp = find_file(ino);
2435         if (fp) {
2436                 if ((status = nfs4_check_open(fp, open, &stp)))
2437                         goto out;
2438                 status = nfs4_check_deleg(fp, open, &dp);
2439                 if (status)
2440                         goto out;
2441         } else {
2442                 status = nfserr_bad_stateid;
2443                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2444                         goto out;
2445                 status = nfserr_resource;
2446                 fp = alloc_init_file(ino);
2447                 if (fp == NULL)
2448                         goto out;
2449         }
2450
2451         /*
2452          * OPEN the file, or upgrade an existing OPEN.
2453          * If truncate fails, the OPEN fails.
2454          */
2455         if (stp) {
2456                 /* Stateid was found, this is an OPEN upgrade */
2457                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2458                 if (status)
2459                         goto out;
2460                 update_stateid(&stp->st_stateid);
2461         } else {
2462                 /* Stateid was not found, this is a new OPEN */
2463                 int flags = 0;
2464                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2465                         flags |= NFSD_MAY_READ;
2466                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2467                         flags |= NFSD_MAY_WRITE;
2468                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2469                 if (status)
2470                         goto out;
2471                 init_stateid(stp, fp, open);
2472                 status = nfsd4_truncate(rqstp, current_fh, open);
2473                 if (status) {
2474                         release_open_stateid(stp);
2475                         goto out;
2476                 }
2477                 if (nfsd4_has_session(&resp->cstate))
2478                         update_stateid(&stp->st_stateid);
2479         }
2480         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2481
2482         if (nfsd4_has_session(&resp->cstate)) {
2483                 open->op_stateowner->so_confirmed = 1;
2484                 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2485         }
2486
2487         /*
2488         * Attempt to hand out a delegation. No error return, because the
2489         * OPEN succeeds even if we fail.
2490         */
2491         nfs4_open_delegation(current_fh, open, stp);
2492
2493         status = nfs_ok;
2494
2495         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2496                 STATEID_VAL(&stp->st_stateid));
2497 out:
2498         if (fp)
2499                 put_nfs4_file(fp);
2500         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2501                 nfs4_set_claim_prev(open);
2502         /*
2503         * To finish the open response, we just need to set the rflags.
2504         */
2505         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2506         if (!open->op_stateowner->so_confirmed &&
2507             !nfsd4_has_session(&resp->cstate))
2508                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2509
2510         return status;
2511 }
2512
2513 __be32
2514 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2515             clientid_t *clid)
2516 {
2517         struct nfs4_client *clp;
2518         __be32 status;
2519
2520         nfs4_lock_state();
2521         dprintk("process_renew(%08x/%08x): starting\n", 
2522                         clid->cl_boot, clid->cl_id);
2523         status = nfserr_stale_clientid;
2524         if (STALE_CLIENTID(clid))
2525                 goto out;
2526         clp = find_confirmed_client(clid);
2527         status = nfserr_expired;
2528         if (clp == NULL) {
2529                 /* We assume the client took too long to RENEW. */
2530                 dprintk("nfsd4_renew: clientid not found!\n");
2531                 goto out;
2532         }
2533         renew_client(clp);
2534         status = nfserr_cb_path_down;
2535         if (!list_empty(&clp->cl_delegations)
2536                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2537                 goto out;
2538         status = nfs_ok;
2539 out:
2540         nfs4_unlock_state();
2541         return status;
2542 }
2543
2544 struct lock_manager nfsd4_manager = {
2545 };
2546
2547 static void
2548 nfsd4_end_grace(void)
2549 {
2550         dprintk("NFSD: end of grace period\n");
2551         nfsd4_recdir_purge_old();
2552         locks_end_grace(&nfsd4_manager);
2553         /*
2554          * Now that every NFSv4 client has had the chance to recover and
2555          * to see the (possibly new, possibly shorter) lease time, we
2556          * can safely set the next grace time to the current lease time:
2557          */
2558         nfsd4_grace = nfsd4_lease;
2559 }
2560
2561 static time_t
2562 nfs4_laundromat(void)
2563 {
2564         struct nfs4_client *clp;
2565         struct nfs4_stateowner *sop;
2566         struct nfs4_delegation *dp;
2567         struct list_head *pos, *next, reaplist;
2568         time_t cutoff = get_seconds() - nfsd4_lease;
2569         time_t t, clientid_val = nfsd4_lease;
2570         time_t u, test_val = nfsd4_lease;
2571
2572         nfs4_lock_state();
2573
2574         dprintk("NFSD: laundromat service - starting\n");
2575         if (locks_in_grace())
2576                 nfsd4_end_grace();
2577         list_for_each_safe(pos, next, &client_lru) {
2578                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2579                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2580                         t = clp->cl_time - cutoff;
2581                         if (clientid_val > t)
2582                                 clientid_val = t;
2583                         break;
2584                 }
2585                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2586                         clp->cl_clientid.cl_id);
2587                 nfsd4_remove_clid_dir(clp);
2588                 expire_client(clp);
2589         }
2590         INIT_LIST_HEAD(&reaplist);
2591         spin_lock(&recall_lock);
2592         list_for_each_safe(pos, next, &del_recall_lru) {
2593                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2594                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2595                         u = dp->dl_time - cutoff;
2596                         if (test_val > u)
2597                                 test_val = u;
2598                         break;
2599                 }
2600                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2601                                     dp, dp->dl_flock);
2602                 list_move(&dp->dl_recall_lru, &reaplist);
2603         }
2604         spin_unlock(&recall_lock);
2605         list_for_each_safe(pos, next, &reaplist) {
2606                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2607                 list_del_init(&dp->dl_recall_lru);
2608                 unhash_delegation(dp);
2609         }
2610         test_val = nfsd4_lease;
2611         list_for_each_safe(pos, next, &close_lru) {
2612                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2613                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2614                         u = sop->so_time - cutoff;
2615                         if (test_val > u)
2616                                 test_val = u;
2617                         break;
2618                 }
2619                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2620                         sop->so_id);
2621                 release_openowner(sop);
2622         }
2623         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2624                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2625         nfs4_unlock_state();
2626         return clientid_val;
2627 }
2628
2629 static struct workqueue_struct *laundry_wq;
2630 static void laundromat_main(struct work_struct *);
2631 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2632
2633 static void
2634 laundromat_main(struct work_struct *not_used)
2635 {
2636         time_t t;
2637
2638         t = nfs4_laundromat();
2639         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2640         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2641 }
2642
2643 static struct nfs4_stateowner *
2644 search_close_lru(u32 st_id, int flags)
2645 {
2646         struct nfs4_stateowner *local = NULL;
2647
2648         if (flags & CLOSE_STATE) {
2649                 list_for_each_entry(local, &close_lru, so_close_lru) {
2650                         if (local->so_id == st_id)
2651                                 return local;
2652                 }
2653         }
2654         return NULL;
2655 }
2656
2657 static inline int
2658 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2659 {
2660         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2661 }
2662
2663 static int
2664 STALE_STATEID(stateid_t *stateid)
2665 {
2666         if (time_after((unsigned long)boot_time,
2667                         (unsigned long)stateid->si_boot)) {
2668                 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2669                         STATEID_VAL(stateid));
2670                 return 1;
2671         }
2672         return 0;
2673 }
2674
2675 static int
2676 EXPIRED_STATEID(stateid_t *stateid)
2677 {
2678         if (time_before((unsigned long)boot_time,
2679                         ((unsigned long)stateid->si_boot)) &&
2680             time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2681                 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2682                         STATEID_VAL(stateid));
2683                 return 1;
2684         }
2685         return 0;
2686 }
2687
2688 static __be32
2689 stateid_error_map(stateid_t *stateid)
2690 {
2691         if (STALE_STATEID(stateid))
2692                 return nfserr_stale_stateid;
2693         if (EXPIRED_STATEID(stateid))
2694                 return nfserr_expired;
2695
2696         dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2697                 STATEID_VAL(stateid));
2698         return nfserr_bad_stateid;
2699 }
2700
2701 static inline int
2702 access_permit_read(unsigned long access_bmap)
2703 {
2704         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2705                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2706                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2707 }
2708
2709 static inline int
2710 access_permit_write(unsigned long access_bmap)
2711 {
2712         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2713                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2714 }
2715
2716 static
2717 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2718 {
2719         __be32 status = nfserr_openmode;
2720
2721         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2722                 goto out;
2723         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2724                 goto out;
2725         status = nfs_ok;
2726 out:
2727         return status;
2728 }
2729
2730 static inline __be32
2731 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2732 {
2733         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2734                 return nfs_ok;
2735         else if (locks_in_grace()) {
2736                 /* Answer in remaining cases depends on existance of
2737                  * conflicting state; so we must wait out the grace period. */
2738                 return nfserr_grace;
2739         } else if (flags & WR_STATE)
2740                 return nfs4_share_conflict(current_fh,
2741                                 NFS4_SHARE_DENY_WRITE);
2742         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2743                 return nfs4_share_conflict(current_fh,
2744                                 NFS4_SHARE_DENY_READ);
2745 }
2746
2747 /*
2748  * Allow READ/WRITE during grace period on recovered state only for files
2749  * that are not able to provide mandatory locking.
2750  */
2751 static inline int
2752 grace_disallows_io(struct inode *inode)
2753 {
2754         return locks_in_grace() && mandatory_lock(inode);
2755 }
2756
2757 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2758 {
2759         /*
2760          * When sessions are used the stateid generation number is ignored
2761          * when it is zero.
2762          */
2763         if ((flags & HAS_SESSION) && in->si_generation == 0)
2764                 goto out;
2765
2766         /* If the client sends us a stateid from the future, it's buggy: */
2767         if (in->si_generation > ref->si_generation)
2768                 return nfserr_bad_stateid;
2769         /*
2770          * The following, however, can happen.  For example, if the
2771          * client sends an open and some IO at the same time, the open
2772          * may bump si_generation while the IO is still in flight.
2773          * Thanks to hard links and renames, the client never knows what
2774          * file an open will affect.  So it could avoid that situation
2775          * only by serializing all opens and IO from the same open
2776          * owner.  To recover from the old_stateid error, the client
2777          * will just have to retry the IO:
2778          */
2779         if (in->si_generation < ref->si_generation)
2780                 return nfserr_old_stateid;
2781 out:
2782         return nfs_ok;
2783 }
2784
2785 static int is_delegation_stateid(stateid_t *stateid)
2786 {
2787         return stateid->si_fileid == 0;
2788 }
2789
2790 /*
2791 * Checks for stateid operations
2792 */
2793 __be32
2794 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2795                            stateid_t *stateid, int flags, struct file **filpp)
2796 {
2797         struct nfs4_stateid *stp = NULL;
2798         struct nfs4_delegation *dp = NULL;
2799         struct svc_fh *current_fh = &cstate->current_fh;
2800         struct inode *ino = current_fh->fh_dentry->d_inode;
2801         __be32 status;
2802
2803         if (filpp)
2804                 *filpp = NULL;
2805
2806         if (grace_disallows_io(ino))
2807                 return nfserr_grace;
2808
2809         if (nfsd4_has_session(cstate))
2810                 flags |= HAS_SESSION;
2811
2812         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2813                 return check_special_stateids(current_fh, stateid, flags);
2814
2815         status = nfserr_stale_stateid;
2816         if (STALE_STATEID(stateid)) 
2817                 goto out;
2818
2819         status = nfserr_bad_stateid;
2820         if (is_delegation_stateid(stateid)) {
2821                 dp = find_delegation_stateid(ino, stateid);
2822                 if (!dp) {
2823                         status = stateid_error_map(stateid);
2824                         goto out;
2825                 }
2826                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2827                                                   flags);
2828                 if (status)
2829                         goto out;
2830                 status = nfs4_check_delegmode(dp, flags);
2831                 if (status)
2832                         goto out;
2833                 renew_client(dp->dl_client);
2834                 if (filpp)
2835                         *filpp = dp->dl_vfs_file;
2836         } else { /* open or lock stateid */
2837                 stp = find_stateid(stateid, flags);
2838                 if (!stp) {
2839                         status = stateid_error_map(stateid);
2840                         goto out;
2841                 }
2842                 if (nfs4_check_fh(current_fh, stp))
2843                         goto out;
2844                 if (!stp->st_stateowner->so_confirmed)
2845                         goto out;
2846                 status = check_stateid_generation(stateid, &stp->st_stateid,
2847                                                   flags);
2848                 if (status)
2849                         goto out;
2850                 status = nfs4_check_openmode(stp, flags);
2851                 if (status)
2852                         goto out;
2853                 renew_client(stp->st_stateowner->so_client);
2854                 if (filpp)
2855                         *filpp = stp->st_vfs_file;
2856         }
2857         status = nfs_ok;
2858 out:
2859         return status;
2860 }
2861
2862 static inline int
2863 setlkflg (int type)
2864 {
2865         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2866                 RD_STATE : WR_STATE;
2867 }
2868
2869 /* 
2870  * Checks for sequence id mutating operations. 
2871  */
2872 static __be32
2873 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2874                          stateid_t *stateid, int flags,
2875                          struct nfs4_stateowner **sopp,
2876                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2877 {
2878         struct nfs4_stateid *stp;
2879         struct nfs4_stateowner *sop;
2880         struct svc_fh *current_fh = &cstate->current_fh;
2881         __be32 status;
2882
2883         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2884                 seqid, STATEID_VAL(stateid));
2885
2886         *stpp = NULL;
2887         *sopp = NULL;
2888
2889         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2890                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2891                 return nfserr_bad_stateid;
2892         }
2893
2894         if (STALE_STATEID(stateid))
2895                 return nfserr_stale_stateid;
2896
2897         if (nfsd4_has_session(cstate))
2898                 flags |= HAS_SESSION;
2899
2900         /*
2901         * We return BAD_STATEID if filehandle doesn't match stateid, 
2902         * the confirmed flag is incorrecly set, or the generation 
2903         * number is incorrect.  
2904         */
2905         stp = find_stateid(stateid, flags);
2906         if (stp == NULL) {
2907                 /*
2908                  * Also, we should make sure this isn't just the result of
2909                  * a replayed close:
2910                  */
2911                 sop = search_close_lru(stateid->si_stateownerid, flags);
2912                 if (sop == NULL)
2913                         return stateid_error_map(stateid);
2914                 *sopp = sop;
2915                 goto check_replay;
2916         }
2917
2918         *stpp = stp;
2919         *sopp = sop = stp->st_stateowner;
2920
2921         if (lock) {
2922                 clientid_t *lockclid = &lock->v.new.clientid;
2923                 struct nfs4_client *clp = sop->so_client;
2924                 int lkflg = 0;
2925                 __be32 status;
2926
2927                 lkflg = setlkflg(lock->lk_type);
2928
2929                 if (lock->lk_is_new) {
2930                         if (!sop->so_is_open_owner)
2931                                 return nfserr_bad_stateid;
2932                         if (!(flags & HAS_SESSION) &&
2933                             !same_clid(&clp->cl_clientid, lockclid))
2934                                 return nfserr_bad_stateid;
2935                         /* stp is the open stateid */
2936                         status = nfs4_check_openmode(stp, lkflg);
2937                         if (status)
2938                                 return status;
2939                 } else {
2940                         /* stp is the lock stateid */
2941                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2942                         if (status)
2943                                 return status;
2944                }
2945         }
2946
2947         if (nfs4_check_fh(current_fh, stp)) {
2948                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2949                 return nfserr_bad_stateid;
2950         }
2951
2952         /*
2953         *  We now validate the seqid and stateid generation numbers.
2954         *  For the moment, we ignore the possibility of 
2955         *  generation number wraparound.
2956         */
2957         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2958                 goto check_replay;
2959
2960         if (sop->so_confirmed && flags & CONFIRM) {
2961                 dprintk("NFSD: preprocess_seqid_op: expected"
2962                                 " unconfirmed stateowner!\n");
2963                 return nfserr_bad_stateid;
2964         }
2965         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2966                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2967                                 " confirmed yet!\n");
2968                 return nfserr_bad_stateid;
2969         }
2970         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2971         if (status)
2972                 return status;
2973         renew_client(sop->so_client);
2974         return nfs_ok;
2975
2976 check_replay:
2977         if (seqid == sop->so_seqid - 1) {
2978                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2979                 /* indicate replay to calling function */
2980                 return nfserr_replay_me;
2981         }
2982         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2983                         sop->so_seqid, seqid);
2984         *sopp = NULL;
2985         return nfserr_bad_seqid;
2986 }
2987
2988 __be32
2989 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2990                    struct nfsd4_open_confirm *oc)
2991 {
2992         __be32 status;
2993         struct nfs4_stateowner *sop;
2994         struct nfs4_stateid *stp;
2995
2996         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2997                         (int)cstate->current_fh.fh_dentry->d_name.len,
2998                         cstate->current_fh.fh_dentry->d_name.name);
2999
3000         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3001         if (status)
3002                 return status;
3003
3004         nfs4_lock_state();
3005
3006         if ((status = nfs4_preprocess_seqid_op(cstate,
3007                                         oc->oc_seqid, &oc->oc_req_stateid,
3008                                         CONFIRM | OPEN_STATE,
3009                                         &oc->oc_stateowner, &stp, NULL)))
3010                 goto out; 
3011
3012         sop = oc->oc_stateowner;
3013         sop->so_confirmed = 1;
3014         update_stateid(&stp->st_stateid);
3015         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3016         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3017                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3018
3019         nfsd4_create_clid_dir(sop->so_client);
3020 out:
3021         if (oc->oc_stateowner) {
3022                 nfs4_get_stateowner(oc->oc_stateowner);
3023                 cstate->replay_owner = oc->oc_stateowner;
3024         }
3025         nfs4_unlock_state();
3026         return status;
3027 }
3028
3029
3030 /*
3031  * unset all bits in union bitmap (bmap) that
3032  * do not exist in share (from successful OPEN_DOWNGRADE)
3033  */
3034 static void
3035 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3036 {
3037         int i;
3038         for (i = 1; i < 4; i++) {
3039                 if ((i & access) != i)
3040                         __clear_bit(i, bmap);
3041         }
3042 }
3043
3044 static void
3045 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3046 {
3047         int i;
3048         for (i = 0; i < 4; i++) {
3049                 if ((i & deny) != i)
3050                         __clear_bit(i, bmap);
3051         }
3052 }
3053
3054 __be32
3055 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3056                      struct nfsd4_compound_state *cstate,
3057                      struct nfsd4_open_downgrade *od)
3058 {
3059         __be32 status;
3060         struct nfs4_stateid *stp;
3061         unsigned int share_access;
3062
3063         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3064                         (int)cstate->current_fh.fh_dentry->d_name.len,
3065                         cstate->current_fh.fh_dentry->d_name.name);
3066
3067         if (!access_valid(od->od_share_access, cstate->minorversion)
3068                         || !deny_valid(od->od_share_deny))
3069                 return nfserr_inval;
3070
3071         nfs4_lock_state();
3072         if ((status = nfs4_preprocess_seqid_op(cstate,
3073                                         od->od_seqid,
3074                                         &od->od_stateid, 
3075                                         OPEN_STATE,
3076                                         &od->od_stateowner, &stp, NULL)))
3077                 goto out; 
3078
3079         status = nfserr_inval;
3080         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3081                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3082                         stp->st_access_bmap, od->od_share_access);
3083                 goto out;
3084         }
3085         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3086                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3087                         stp->st_deny_bmap, od->od_share_deny);
3088                 goto out;
3089         }
3090         set_access(&share_access, stp->st_access_bmap);
3091         nfs4_file_downgrade(stp->st_vfs_file,
3092                             share_access & ~od->od_share_access);
3093
3094         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3095         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3096
3097         update_stateid(&stp->st_stateid);
3098         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3099         status = nfs_ok;
3100 out:
3101         if (od->od_stateowner) {
3102                 nfs4_get_stateowner(od->od_stateowner);
3103                 cstate->replay_owner = od->od_stateowner;
3104         }
3105         nfs4_unlock_state();
3106         return status;
3107 }
3108
3109 /*
3110  * nfs4_unlock_state() called after encode
3111  */
3112 __be32
3113 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3114             struct nfsd4_close *close)
3115 {
3116         __be32 status;
3117         struct nfs4_stateid *stp;
3118
3119         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3120                         (int)cstate->current_fh.fh_dentry->d_name.len,
3121                         cstate->current_fh.fh_dentry->d_name.name);
3122
3123         nfs4_lock_state();
3124         /* check close_lru for replay */
3125         if ((status = nfs4_preprocess_seqid_op(cstate,
3126                                         close->cl_seqid,
3127                                         &close->cl_stateid, 
3128                                         OPEN_STATE | CLOSE_STATE,
3129                                         &close->cl_stateowner, &stp, NULL)))
3130                 goto out; 
3131         status = nfs_ok;
3132         update_stateid(&stp->st_stateid);
3133         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3134
3135         /* release_stateid() calls nfsd_close() if needed */
3136         release_open_stateid(stp);
3137
3138         /* place unused nfs4_stateowners on so_close_lru list to be
3139          * released by the laundromat service after the lease period
3140          * to enable us to handle CLOSE replay
3141          */
3142         if (list_empty(&close->cl_stateowner->so_stateids))
3143                 move_to_close_lru(close->cl_stateowner);
3144 out:
3145         if (close->cl_stateowner) {
3146                 nfs4_get_stateowner(close->cl_stateowner);
3147                 cstate->replay_owner = close->cl_stateowner;
3148         }
3149         nfs4_unlock_state();
3150         return status;
3151 }
3152
3153 __be32
3154 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3155                   struct nfsd4_delegreturn *dr)
3156 {
3157         struct nfs4_delegation *dp;
3158         stateid_t *stateid = &dr->dr_stateid;
3159         struct inode *inode;
3160         __be32 status;
3161         int flags = 0;
3162
3163         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3164                 return status;
3165         inode = cstate->current_fh.fh_dentry->d_inode;
3166
3167         if (nfsd4_has_session(cstate))
3168                 flags |= HAS_SESSION;
3169         nfs4_lock_state();
3170         status = nfserr_bad_stateid;
3171         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3172                 goto out;
3173         status = nfserr_stale_stateid;
3174         if (STALE_STATEID(stateid))
3175                 goto out;
3176         status = nfserr_bad_stateid;
3177         if (!is_delegation_stateid(stateid))
3178                 goto out;
3179         dp = find_delegation_stateid(inode, stateid);
3180         if (!dp) {
3181                 status = stateid_error_map(stateid);
3182                 goto out;
3183         }
3184         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3185         if (status)
3186                 goto out;
3187         renew_client(dp->dl_client);
3188
3189         unhash_delegation(dp);
3190 out:
3191         nfs4_unlock_state();
3192
3193         return status;
3194 }
3195
3196
3197 /* 
3198  * Lock owner state (byte-range locks)
3199  */
3200 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3201 #define LOCK_HASH_BITS              8
3202 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3203 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3204
3205 static inline u64
3206 end_offset(u64 start, u64 len)
3207 {
3208         u64 end;
3209
3210         end = start + len;
3211         return end >= start ? end: NFS4_MAX_UINT64;
3212 }
3213
3214 /* last octet in a range */
3215 static inline u64
3216 last_byte_offset(u64 start, u64 len)
3217 {
3218         u64 end;
3219
3220         BUG_ON(!len);
3221         end = start + len;
3222         return end > start ? end - 1: NFS4_MAX_UINT64;
3223 }
3224
3225 #define lockownerid_hashval(id) \
3226         ((id) & LOCK_HASH_MASK)
3227
3228 static inline unsigned int
3229 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3230                 struct xdr_netobj *ownername)
3231 {
3232         return (file_hashval(inode) + cl_id
3233                         + opaque_hashval(ownername->data, ownername->len))
3234                 & LOCK_HASH_MASK;
3235 }
3236
3237 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3238 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3239 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3240
3241 static struct nfs4_stateid *
3242 find_stateid(stateid_t *stid, int flags)
3243 {
3244         struct nfs4_stateid *local;
3245         u32 st_id = stid->si_stateownerid;
3246         u32 f_id = stid->si_fileid;
3247         unsigned int hashval;
3248
3249         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3250         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3251                 hashval = stateid_hashval(st_id, f_id);
3252                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3253                         if ((local->st_stateid.si_stateownerid == st_id) &&
3254                             (local->st_stateid.si_fileid == f_id))
3255                                 return local;
3256                 }
3257         } 
3258
3259         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3260                 hashval = stateid_hashval(st_id, f_id);
3261                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3262                         if ((local->st_stateid.si_stateownerid == st_id) &&
3263                             (local->st_stateid.si_fileid == f_id))
3264                                 return local;
3265                 }
3266         }
3267         return NULL;
3268 }
3269
3270 static struct nfs4_delegation *
3271 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3272 {
3273         struct nfs4_file *fp;
3274         struct nfs4_delegation *dl;
3275
3276         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3277                 STATEID_VAL(stid));
3278
3279         fp = find_file(ino);
3280         if (!fp)
3281                 return NULL;
3282         dl = find_delegation_file(fp, stid);
3283         put_nfs4_file(fp);
3284         return dl;
3285 }
3286
3287 /*
3288  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3289  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3290  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3291  * locking, this prevents us from being completely protocol-compliant.  The
3292  * real solution to this problem is to start using unsigned file offsets in
3293  * the VFS, but this is a very deep change!
3294  */
3295 static inline void
3296 nfs4_transform_lock_offset(struct file_lock *lock)
3297 {
3298         if (lock->fl_start < 0)
3299                 lock->fl_start = OFFSET_MAX;
3300         if (lock->fl_end < 0)
3301                 lock->fl_end = OFFSET_MAX;
3302 }
3303
3304 /* Hack!: For now, we're defining this just so we can use a pointer to it
3305  * as a unique cookie to identify our (NFSv4's) posix locks. */
3306 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3307 };
3308
3309 static inline void
3310 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3311 {
3312         struct nfs4_stateowner *sop;
3313         unsigned int hval;
3314
3315         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3316                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3317                 hval = lockownerid_hashval(sop->so_id);
3318                 kref_get(&sop->so_ref);
3319                 deny->ld_sop = sop;
3320                 deny->ld_clientid = sop->so_client->cl_clientid;
3321         } else {
3322                 deny->ld_sop = NULL;
3323                 deny->ld_clientid.cl_boot = 0;
3324                 deny->ld_clientid.cl_id = 0;
3325         }
3326         deny->ld_start = fl->fl_start;
3327         deny->ld_length = NFS4_MAX_UINT64;
3328         if (fl->fl_end != NFS4_MAX_UINT64)
3329                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3330         deny->ld_type = NFS4_READ_LT;
3331         if (fl->fl_type != F_RDLCK)
3332                 deny->ld_type = NFS4_WRITE_LT;
3333 }
3334
3335 static struct nfs4_stateowner *
3336 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3337                 struct xdr_netobj *owner)
3338 {
3339         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3340         struct nfs4_stateowner *op;
3341
3342         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3343                 if (same_owner_str(op, owner, clid))
3344                         return op;
3345         }
3346         return NULL;
3347 }
3348
3349 /*
3350  * Alloc a lock owner structure.
3351  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3352  * occured. 
3353  *
3354  * strhashval = lock_ownerstr_hashval 
3355  */
3356
3357 static struct nfs4_stateowner *
3358 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3359         struct nfs4_stateowner *sop;
3360         struct nfs4_replay *rp;
3361         unsigned int idhashval;
3362
3363         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3364                 return NULL;
3365         idhashval = lockownerid_hashval(current_ownerid);
3366         INIT_LIST_HEAD(&sop->so_idhash);
3367         INIT_LIST_HEAD(&sop->so_strhash);
3368         INIT_LIST_HEAD(&sop->so_perclient);
3369         INIT_LIST_HEAD(&sop->so_stateids);
3370         INIT_LIST_HEAD(&sop->so_perstateid);
3371         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3372         sop->so_time = 0;
3373         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3374         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3375         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3376         sop->so_is_open_owner = 0;
3377         sop->so_id = current_ownerid++;
3378         sop->so_client = clp;
3379         /* It is the openowner seqid that will be incremented in encode in the
3380          * case of new lockowners; so increment the lock seqid manually: */
3381         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3382         sop->so_confirmed = 1;
3383         rp = &sop->so_replay;
3384         rp->rp_status = nfserr_serverfault;
3385         rp->rp_buflen = 0;
3386         rp->rp_buf = rp->rp_ibuf;
3387         return sop;
3388 }
3389
3390 static struct nfs4_stateid *
3391 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3392 {
3393         struct nfs4_stateid *stp;
3394         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3395
3396         stp = nfs4_alloc_stateid();
3397         if (stp == NULL)
3398                 goto out;
3399         INIT_LIST_HEAD(&stp->st_hash);
3400         INIT_LIST_HEAD(&stp->st_perfile);
3401         INIT_LIST_HEAD(&stp->st_perstateowner);
3402         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3403         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3404         list_add(&stp->st_perfile, &fp->fi_stateids);
3405         list_add(&stp->st_perstateowner, &sop->so_stateids);
3406         stp->st_stateowner = sop;
3407         get_nfs4_file(fp);
3408         stp->st_file = fp;
3409         stp->st_stateid.si_boot = get_seconds();
3410         stp->st_stateid.si_stateownerid = sop->so_id;
3411         stp->st_stateid.si_fileid = fp->fi_id;
3412         stp->st_stateid.si_generation = 0;
3413         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3414         stp->st_access_bmap = open_stp->st_access_bmap;
3415         stp->st_deny_bmap = open_stp->st_deny_bmap;
3416         stp->st_openstp = open_stp;
3417
3418 out:
3419         return stp;
3420 }
3421
3422 static int
3423 check_lock_length(u64 offset, u64 length)
3424 {
3425         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3426              LOFF_OVERFLOW(offset, length)));
3427 }
3428
3429 /*
3430  *  LOCK operation 
3431  */
3432 __be32
3433 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3434            struct nfsd4_lock *lock)
3435 {
3436         struct nfs4_stateowner *open_sop = NULL;
3437         struct nfs4_stateowner *lock_sop = NULL;
3438         struct nfs4_stateid *lock_stp;
3439         struct file *filp;
3440         struct file_lock file_lock;
3441         struct file_lock conflock;
3442         __be32 status = 0;
3443         unsigned int strhashval;
3444         unsigned int cmd;
3445         int err;
3446
3447         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3448                 (long long) lock->lk_offset,
3449                 (long long) lock->lk_length);
3450
3451         if (check_lock_length(lock->lk_offset, lock->lk_length))
3452                  return nfserr_inval;
3453
3454         if ((status = fh_verify(rqstp, &cstate->current_fh,
3455                                 S_IFREG, NFSD_MAY_LOCK))) {
3456                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3457                 return status;
3458         }
3459
3460         nfs4_lock_state();
3461
3462         if (lock->lk_is_new) {
3463                 /*
3464                  * Client indicates that this is a new lockowner.
3465                  * Use open owner and open stateid to create lock owner and
3466                  * lock stateid.
3467                  */
3468                 struct nfs4_stateid *open_stp = NULL;
3469                 struct nfs4_file *fp;
3470                 
3471                 status = nfserr_stale_clientid;
3472                 if (!nfsd4_has_session(cstate) &&
3473                     STALE_CLIENTID(&lock->lk_new_clientid))
3474                         goto out;
3475
3476                 /* validate and update open stateid and open seqid */
3477                 status = nfs4_preprocess_seqid_op(cstate,
3478                                         lock->lk_new_open_seqid,
3479                                         &lock->lk_new_open_stateid,
3480                                         OPEN_STATE,
3481                                         &lock->lk_replay_owner, &open_stp,
3482                                         lock);
3483                 if (status)
3484                         goto out;
3485                 open_sop = lock->lk_replay_owner;
3486                 /* create lockowner and lock stateid */
3487                 fp = open_stp->st_file;
3488                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3489                                 open_sop->so_client->cl_clientid.cl_id, 
3490                                 &lock->v.new.owner);
3491                 /* XXX: Do we need to check for duplicate stateowners on
3492                  * the same file, or should they just be allowed (and
3493                  * create new stateids)? */
3494                 status = nfserr_resource;
3495                 lock_sop = alloc_init_lock_stateowner(strhashval,
3496                                 open_sop->so_client, open_stp, lock);
3497                 if (lock_sop == NULL)
3498                         goto out;
3499                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3500                 if (lock_stp == NULL)
3501                         goto out;
3502         } else {
3503                 /* lock (lock owner + lock stateid) already exists */
3504                 status = nfs4_preprocess_seqid_op(cstate,
3505                                        lock->lk_old_lock_seqid, 
3506                                        &lock->lk_old_lock_stateid, 
3507                                        LOCK_STATE,
3508                                        &lock->lk_replay_owner, &lock_stp, lock);
3509                 if (status)
3510                         goto out;
3511                 lock_sop = lock->lk_replay_owner;
3512         }
3513         /* lock->lk_replay_owner and lock_stp have been created or found */
3514         filp = lock_stp->st_vfs_file;
3515
3516         status = nfserr_grace;
3517         if (locks_in_grace() && !lock->lk_reclaim)
3518                 goto out;
3519         status = nfserr_no_grace;
3520         if (!locks_in_grace() && lock->lk_reclaim)
3521                 goto out;
3522
3523         locks_init_lock(&file_lock);
3524         switch (lock->lk_type) {
3525                 case NFS4_READ_LT:
3526                 case NFS4_READW_LT:
3527                         file_lock.fl_type = F_RDLCK;
3528                         cmd = F_SETLK;
3529                 break;
3530                 case NFS4_WRITE_LT:
3531                 case NFS4_WRITEW_LT:
3532                         file_lock.fl_type = F_WRLCK;
3533                         cmd = F_SETLK;
3534                 break;
3535                 default:
3536                         status = nfserr_inval;
3537                 goto out;
3538         }
3539         file_lock.fl_owner = (fl_owner_t)lock_sop;
3540         file_lock.fl_pid = current->tgid;
3541         file_lock.fl_file = filp;
3542         file_lock.fl_flags = FL_POSIX;
3543         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3544
3545         file_lock.fl_start = lock->lk_offset;
3546         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3547         nfs4_transform_lock_offset(&file_lock);
3548
3549         /*
3550         * Try to lock the file in the VFS.
3551         * Note: locks.c uses the BKL to protect the inode's lock list.
3552         */
3553
3554         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3555         switch (-err) {
3556         case 0: /* success! */
3557                 update_stateid(&lock_stp->st_stateid);
3558                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3559                                 sizeof(stateid_t));
3560                 status = 0;
3561                 break;
3562         case (EAGAIN):          /* conflock holds conflicting lock */
3563                 status = nfserr_denied;
3564                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3565                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3566                 break;
3567         case (EDEADLK):
3568                 status = nfserr_deadlock;
3569                 break;
3570         default:        
3571                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3572                 status = nfserr_resource;
3573                 break;
3574         }
3575 out:
3576         if (status && lock->lk_is_new && lock_sop)
3577                 release_lockowner(lock_sop);
3578         if (lock->lk_replay_owner) {
3579                 nfs4_get_stateowner(lock->lk_replay_owner);
3580                 cstate->replay_owner = lock->lk_replay_owner;
3581         }
3582         nfs4_unlock_state();
3583         return status;
3584 }
3585
3586 /*
3587  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3588  * so we do a temporary open here just to get an open file to pass to
3589  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3590  * inode operation.)
3591  */
3592 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3593 {
3594         struct file *file;
3595         int err;
3596
3597         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3598         if (err)
3599                 return err;
3600         err = vfs_test_lock(file, lock);
3601         nfsd_close(file);
3602         return err;
3603 }
3604
3605 /*
3606  * LOCKT operation
3607  */
3608 __be32
3609 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3610             struct nfsd4_lockt *lockt)
3611 {
3612         struct inode *inode;
3613         struct file_lock file_lock;
3614         int error;
3615         __be32 status;
3616
3617         if (locks_in_grace())
3618                 return nfserr_grace;
3619
3620         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3621                  return nfserr_inval;
3622
3623         lockt->lt_stateowner = NULL;
3624         nfs4_lock_state();
3625
3626         status = nfserr_stale_clientid;
3627         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3628                 goto out;
3629
3630         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3631                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3632                 if (status == nfserr_symlink)
3633                         status = nfserr_inval;
3634                 goto out;
3635         }
3636
3637         inode = cstate->current_fh.fh_dentry->d_inode;
3638         locks_init_lock(&file_lock);
3639         switch (lockt->lt_type) {
3640                 case NFS4_READ_LT:
3641                 case NFS4_READW_LT:
3642                         file_lock.fl_type = F_RDLCK;
3643                 break;
3644                 case NFS4_WRITE_LT:
3645                 case NFS4_WRITEW_LT:
3646                         file_lock.fl_type = F_WRLCK;
3647                 break;
3648                 default:
3649                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3650                         status = nfserr_inval;
3651                 goto out;
3652         }
3653
3654         lockt->lt_stateowner = find_lockstateowner_str(inode,
3655                         &lockt->lt_clientid, &lockt->lt_owner);
3656         if (lockt->lt_stateowner)
3657                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3658         file_lock.fl_pid = current->tgid;
3659         file_lock.fl_flags = FL_POSIX;
3660
3661         file_lock.fl_start = lockt->lt_offset;
3662         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3663
3664         nfs4_transform_lock_offset(&file_lock);
3665
3666         status = nfs_ok;
3667         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3668         if (error) {
3669                 status = nfserrno(error);
3670                 goto out;
3671         }
3672         if (file_lock.fl_type != F_UNLCK) {
3673                 status = nfserr_denied;
3674                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3675         }
3676 out:
3677         nfs4_unlock_state();
3678         return status;
3679 }
3680
3681 __be32
3682 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3683             struct nfsd4_locku *locku)
3684 {
3685         struct nfs4_stateid *stp;
3686         struct file *filp = NULL;
3687         struct file_lock file_lock;
3688         __be32 status;
3689         int err;
3690                                                         
3691         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3692                 (long long) locku->lu_offset,
3693                 (long long) locku->lu_length);
3694
3695         if (check_lock_length(locku->lu_offset, locku->lu_length))
3696                  return nfserr_inval;
3697
3698         nfs4_lock_state();
3699                                                                                 
3700         if ((status = nfs4_preprocess_seqid_op(cstate,
3701                                         locku->lu_seqid, 
3702                                         &locku->lu_stateid, 
3703                                         LOCK_STATE,
3704                                         &locku->lu_stateowner, &stp, NULL)))
3705                 goto out;
3706
3707         filp = stp->st_vfs_file;
3708         BUG_ON(!filp);
3709         locks_init_lock(&file_lock);
3710         file_lock.fl_type = F_UNLCK;
3711         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3712         file_lock.fl_pid = current->tgid;
3713         file_lock.fl_file = filp;
3714         file_lock.fl_flags = FL_POSIX; 
3715         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3716         file_lock.fl_start = locku->lu_offset;
3717
3718         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3719         nfs4_transform_lock_offset(&file_lock);
3720
3721         /*
3722         *  Try to unlock the file in the VFS.
3723         */
3724         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3725         if (err) {
3726                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3727                 goto out_nfserr;
3728         }
3729         /*
3730         * OK, unlock succeeded; the only thing left to do is update the stateid.
3731         */
3732         update_stateid(&stp->st_stateid);
3733         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3734
3735 out:
3736         if (locku->lu_stateowner) {
3737                 nfs4_get_stateowner(locku->lu_stateowner);
3738                 cstate->replay_owner = locku->lu_stateowner;
3739         }
3740         nfs4_unlock_state();
3741         return status;
3742
3743 out_nfserr:
3744         status = nfserrno(err);
3745         goto out;
3746 }
3747
3748 /*
3749  * returns
3750  *      1: locks held by lockowner
3751  *      0: no locks held by lockowner
3752  */
3753 static int
3754 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3755 {
3756         struct file_lock **flpp;
3757         struct inode *inode = filp->f_path.dentry->d_inode;
3758         int status = 0;
3759
3760         lock_kernel();
3761         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3762                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3763                         status = 1;
3764                         goto out;
3765                 }
3766         }
3767 out:
3768         unlock_kernel();
3769         return status;
3770 }
3771
3772 __be32
3773 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3774                         struct nfsd4_compound_state *cstate,
3775                         struct nfsd4_release_lockowner *rlockowner)
3776 {
3777         clientid_t *clid = &rlockowner->rl_clientid;
3778         struct nfs4_stateowner *sop;
3779         struct nfs4_stateid *stp;
3780         struct xdr_netobj *owner = &rlockowner->rl_owner;
3781         struct list_head matches;
3782         int i;
3783         __be32 status;
3784
3785         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3786                 clid->cl_boot, clid->cl_id);
3787
3788         /* XXX check for lease expiration */
3789
3790         status = nfserr_stale_clientid;
3791         if (STALE_CLIENTID(clid))
3792                 return status;
3793
3794         nfs4_lock_state();
3795
3796         status = nfserr_locks_held;
3797         /* XXX: we're doing a linear search through all the lockowners.
3798          * Yipes!  For now we'll just hope clients aren't really using
3799          * release_lockowner much, but eventually we have to fix these
3800          * data structures. */
3801         INIT_LIST_HEAD(&matches);
3802         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3803                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3804                         if (!same_owner_str(sop, owner, clid))
3805                                 continue;
3806                         list_for_each_entry(stp, &sop->so_stateids,
3807                                         st_perstateowner) {
3808                                 if (check_for_locks(stp->st_vfs_file, sop))
3809                                         goto out;
3810                                 /* Note: so_perclient unused for lockowners,
3811                                  * so it's OK to fool with here. */
3812                                 list_add(&sop->so_perclient, &matches);
3813                         }
3814                 }
3815         }
3816         /* Clients probably won't expect us to return with some (but not all)
3817          * of the lockowner state released; so don't release any until all
3818          * have been checked. */
3819         status = nfs_ok;
3820         while (!list_empty(&matches)) {
3821                 sop = list_entry(matches.next, struct nfs4_stateowner,
3822                                                                 so_perclient);
3823                 /* unhash_stateowner deletes so_perclient only
3824                  * for openowners. */
3825                 list_del(&sop->so_perclient);
3826                 release_lockowner(sop);
3827         }
3828 out:
3829         nfs4_unlock_state();
3830         return status;
3831 }
3832
3833 static inline struct nfs4_client_reclaim *
3834 alloc_reclaim(void)
3835 {
3836         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3837 }
3838
3839 int
3840 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3841 {
3842         unsigned int strhashval = clientstr_hashval(name);
3843         struct nfs4_client *clp;
3844
3845         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3846         return clp ? 1 : 0;
3847 }
3848
3849 /*
3850  * failure => all reset bets are off, nfserr_no_grace...
3851  */
3852 int
3853 nfs4_client_to_reclaim(const char *name)
3854 {
3855         unsigned int strhashval;
3856         struct nfs4_client_reclaim *crp = NULL;
3857
3858         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3859         crp = alloc_reclaim();
3860         if (!crp)
3861                 return 0;
3862         strhashval = clientstr_hashval(name);
3863         INIT_LIST_HEAD(&crp->cr_strhash);
3864         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3865         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3866         reclaim_str_hashtbl_size++;
3867         return 1;
3868 }
3869
3870 static void
3871 nfs4_release_reclaim(void)
3872 {
3873         struct nfs4_client_reclaim *crp = NULL;
3874         int i;
3875
3876         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3877                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3878                         crp = list_entry(reclaim_str_hashtbl[i].next,
3879                                         struct nfs4_client_reclaim, cr_strhash);
3880                         list_del(&crp->cr_strhash);
3881                         kfree(crp);
3882                         reclaim_str_hashtbl_size--;
3883                 }
3884         }
3885         BUG_ON(reclaim_str_hashtbl_size);
3886 }
3887
3888 /*
3889  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3890 static struct nfs4_client_reclaim *
3891 nfs4_find_reclaim_client(clientid_t *clid)
3892 {
3893         unsigned int strhashval;
3894         struct nfs4_client *clp;
3895         struct nfs4_client_reclaim *crp = NULL;
3896
3897
3898         /* find clientid in conf_id_hashtbl */
3899         clp = find_confirmed_client(clid);
3900         if (clp == NULL)
3901                 return NULL;
3902
3903         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3904                             clp->cl_name.len, clp->cl_name.data,
3905                             clp->cl_recdir);
3906
3907         /* find clp->cl_name in reclaim_str_hashtbl */
3908         strhashval = clientstr_hashval(clp->cl_recdir);
3909         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3910                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3911                         return crp;
3912                 }
3913         }
3914         return NULL;
3915 }
3916
3917 /*
3918 * Called from OPEN. Look for clientid in reclaim list.
3919 */
3920 __be32
3921 nfs4_check_open_reclaim(clientid_t *clid)
3922 {
3923         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3924 }
3925
3926 /* initialization to perform at module load time: */
3927
3928 int
3929 nfs4_state_init(void)
3930 {
3931         int i, status;
3932
3933         status = nfsd4_init_slabs();
3934         if (status)
3935                 return status;
3936         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3937                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3938                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3939                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3940                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3941                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3942         }
3943         for (i = 0; i < SESSION_HASH_SIZE; i++)
3944                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3945         for (i = 0; i < FILE_HASH_SIZE; i++) {
3946                 INIT_LIST_HEAD(&file_hashtbl[i]);
3947         }
3948         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3949                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3950                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3951         }
3952         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3953                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3954                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3955         }
3956         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3957                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3958                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3959         }
3960         memset(&onestateid, ~0, sizeof(stateid_t));
3961         INIT_LIST_HEAD(&close_lru);
3962         INIT_LIST_HEAD(&client_lru);
3963         INIT_LIST_HEAD(&del_recall_lru);
3964         reclaim_str_hashtbl_size = 0;
3965         return 0;
3966 }
3967
3968 static void
3969 nfsd4_load_reboot_recovery_data(void)
3970 {
3971         int status;
3972
3973         nfs4_lock_state();
3974         nfsd4_init_recdir(user_recovery_dirname);
3975         status = nfsd4_recdir_load();
3976         nfs4_unlock_state();
3977         if (status)
3978                 printk("NFSD: Failure reading reboot recovery data\n");
3979 }
3980
3981 /*
3982  * Since the lifetime of a delegation isn't limited to that of an open, a
3983  * client may quite reasonably hang on to a delegation as long as it has
3984  * the inode cached.  This becomes an obvious problem the first time a
3985  * client's inode cache approaches the size of the server's total memory.
3986  *
3987  * For now we avoid this problem by imposing a hard limit on the number
3988  * of delegations, which varies according to the server's memory size.
3989  */
3990 static void
3991 set_max_delegations(void)
3992 {
3993         /*
3994          * Allow at most 4 delegations per megabyte of RAM.  Quick
3995          * estimates suggest that in the worst case (where every delegation
3996          * is for a different inode), a delegation could take about 1.5K,
3997          * giving a worst case usage of about 6% of memory.
3998          */
3999         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4000 }
4001
4002 /* initialization to perform when the nfsd service is started: */
4003
4004 static int
4005 __nfs4_state_start(void)
4006 {
4007         boot_time = get_seconds();
4008         locks_start_grace(&nfsd4_manager);
4009         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4010                nfsd4_grace);
4011         laundry_wq = create_singlethread_workqueue("nfsd4");
4012         if (laundry_wq == NULL)
4013                 return -ENOMEM;
4014         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4015         set_max_delegations();
4016         return set_callback_cred();
4017 }
4018
4019 int
4020 nfs4_state_start(void)
4021 {
4022         int ret;
4023
4024         if (nfs4_init)
4025                 return 0;
4026         nfsd4_load_reboot_recovery_data();
4027         ret = __nfs4_state_start();
4028         if (ret)
4029                 return ret;
4030         nfs4_init = 1;
4031         return 0;
4032 }
4033
4034 static void
4035 __nfs4_state_shutdown(void)
4036 {
4037         int i;
4038         struct nfs4_client *clp = NULL;
4039         struct nfs4_delegation *dp = NULL;
4040         struct list_head *pos, *next, reaplist;
4041
4042         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4043                 while (!list_empty(&conf_id_hashtbl[i])) {
4044                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4045                         expire_client(clp);
4046                 }
4047                 while (!list_empty(&unconf_str_hashtbl[i])) {
4048                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4049                         expire_client(clp);
4050                 }
4051         }
4052         INIT_LIST_HEAD(&reaplist);
4053         spin_lock(&recall_lock);
4054         list_for_each_safe(pos, next, &del_recall_lru) {
4055                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4056                 list_move(&dp->dl_recall_lru, &reaplist);
4057         }
4058         spin_unlock(&recall_lock);
4059         list_for_each_safe(pos, next, &reaplist) {
4060                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4061                 list_del_init(&dp->dl_recall_lru);
4062                 unhash_delegation(dp);
4063         }
4064
4065         nfsd4_shutdown_recdir();
4066         nfs4_init = 0;
4067 }
4068
4069 void
4070 nfs4_state_shutdown(void)
4071 {
4072         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4073         destroy_workqueue(laundry_wq);
4074         locks_end_grace(&nfsd4_manager);
4075         nfs4_lock_state();
4076         nfs4_release_reclaim();
4077         __nfs4_state_shutdown();
4078         nfs4_unlock_state();