nfsd4: allow 4.0 clients to change callback path
[safe/jmp/linux-2.6] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/smp_lock.h>
37 #include <linux/namei.h>
38 #include <linux/swap.h>
39 #include <linux/sunrpc/svcauth_gss.h>
40 #include <linux/sunrpc/clnt.h>
41 #include "xdr4.h"
42 #include "vfs.h"
43
44 #define NFSDDBG_FACILITY                NFSDDBG_PROC
45
46 /* Globals */
47 time_t nfsd4_lease = 90;     /* default lease time */
48 time_t nfsd4_grace = 90;
49 static time_t boot_time;
50 static u32 current_ownerid = 1;
51 static u32 current_fileid = 1;
52 static u32 current_delegid = 1;
53 static u32 nfs4_init;
54 static stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static struct nfs4_delegation *
166 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
167 {
168         struct nfs4_delegation *dp;
169         struct nfs4_file *fp = stp->st_file;
170         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
171
172         dprintk("NFSD alloc_init_deleg\n");
173         if (fp->fi_had_conflict)
174                 return NULL;
175         if (num_delegations > max_delegations)
176                 return NULL;
177         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
178         if (dp == NULL)
179                 return dp;
180         num_delegations++;
181         INIT_LIST_HEAD(&dp->dl_perfile);
182         INIT_LIST_HEAD(&dp->dl_perclnt);
183         INIT_LIST_HEAD(&dp->dl_recall_lru);
184         dp->dl_client = clp;
185         get_nfs4_file(fp);
186         dp->dl_file = fp;
187         dp->dl_flock = NULL;
188         get_file(stp->st_vfs_file);
189         dp->dl_vfs_file = stp->st_vfs_file;
190         dp->dl_type = type;
191         dp->dl_ident = cb->cb_ident;
192         dp->dl_stateid.si_boot = get_seconds();
193         dp->dl_stateid.si_stateownerid = current_delegid++;
194         dp->dl_stateid.si_fileid = 0;
195         dp->dl_stateid.si_generation = 0;
196         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
197         dp->dl_time = 0;
198         atomic_set(&dp->dl_count, 1);
199         list_add(&dp->dl_perfile, &fp->fi_delegations);
200         list_add(&dp->dl_perclnt, &clp->cl_delegations);
201         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
202         return dp;
203 }
204
205 void
206 nfs4_put_delegation(struct nfs4_delegation *dp)
207 {
208         if (atomic_dec_and_test(&dp->dl_count)) {
209                 dprintk("NFSD: freeing dp %p\n",dp);
210                 put_nfs4_file(dp->dl_file);
211                 kmem_cache_free(deleg_slab, dp);
212                 num_delegations--;
213         }
214 }
215
216 /* Remove the associated file_lock first, then remove the delegation.
217  * lease_modify() is called to remove the FS_LEASE file_lock from
218  * the i_flock list, eventually calling nfsd's lock_manager
219  * fl_release_callback.
220  */
221 static void
222 nfs4_close_delegation(struct nfs4_delegation *dp)
223 {
224         struct file *filp = dp->dl_vfs_file;
225
226         dprintk("NFSD: close_delegation dp %p\n",dp);
227         dp->dl_vfs_file = NULL;
228         /* The following nfsd_close may not actually close the file,
229          * but we want to remove the lease in any case. */
230         if (dp->dl_flock)
231                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
232         nfsd_close(filp);
233 }
234
235 /* Called under the state lock. */
236 static void
237 unhash_delegation(struct nfs4_delegation *dp)
238 {
239         list_del_init(&dp->dl_perfile);
240         list_del_init(&dp->dl_perclnt);
241         spin_lock(&recall_lock);
242         list_del_init(&dp->dl_recall_lru);
243         spin_unlock(&recall_lock);
244         nfs4_close_delegation(dp);
245         nfs4_put_delegation(dp);
246 }
247
248 /* 
249  * SETCLIENTID state 
250  */
251
252 /* Hash tables for nfs4_clientid state */
253 #define CLIENT_HASH_BITS                 4
254 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
255 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
256
257 #define clientid_hashval(id) \
258         ((id) & CLIENT_HASH_MASK)
259 #define clientstr_hashval(name) \
260         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
261 /*
262  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
263  * used in reboot/reset lease grace period processing
264  *
265  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
266  * setclientid_confirmed info. 
267  *
268  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
269  * setclientid info.
270  *
271  * client_lru holds client queue ordered by nfs4_client.cl_time
272  * for lease renewal.
273  *
274  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
275  * for last close replay.
276  */
277 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
278 static int reclaim_str_hashtbl_size = 0;
279 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
280 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
283 static struct list_head client_lru;
284 static struct list_head close_lru;
285
286 static void unhash_generic_stateid(struct nfs4_stateid *stp)
287 {
288         list_del(&stp->st_hash);
289         list_del(&stp->st_perfile);
290         list_del(&stp->st_perstateowner);
291 }
292
293 static void free_generic_stateid(struct nfs4_stateid *stp)
294 {
295         put_nfs4_file(stp->st_file);
296         kmem_cache_free(stateid_slab, stp);
297 }
298
299 static void release_lock_stateid(struct nfs4_stateid *stp)
300 {
301         unhash_generic_stateid(stp);
302         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
303         free_generic_stateid(stp);
304 }
305
306 static void unhash_lockowner(struct nfs4_stateowner *sop)
307 {
308         struct nfs4_stateid *stp;
309
310         list_del(&sop->so_idhash);
311         list_del(&sop->so_strhash);
312         list_del(&sop->so_perstateid);
313         while (!list_empty(&sop->so_stateids)) {
314                 stp = list_first_entry(&sop->so_stateids,
315                                 struct nfs4_stateid, st_perstateowner);
316                 release_lock_stateid(stp);
317         }
318 }
319
320 static void release_lockowner(struct nfs4_stateowner *sop)
321 {
322         unhash_lockowner(sop);
323         nfs4_put_stateowner(sop);
324 }
325
326 static void
327 release_stateid_lockowners(struct nfs4_stateid *open_stp)
328 {
329         struct nfs4_stateowner *lock_sop;
330
331         while (!list_empty(&open_stp->st_lockowners)) {
332                 lock_sop = list_entry(open_stp->st_lockowners.next,
333                                 struct nfs4_stateowner, so_perstateid);
334                 /* list_del(&open_stp->st_lockowners);  */
335                 BUG_ON(lock_sop->so_is_open_owner);
336                 release_lockowner(lock_sop);
337         }
338 }
339
340 static void release_open_stateid(struct nfs4_stateid *stp)
341 {
342         unhash_generic_stateid(stp);
343         release_stateid_lockowners(stp);
344         nfsd_close(stp->st_vfs_file);
345         free_generic_stateid(stp);
346 }
347
348 static void unhash_openowner(struct nfs4_stateowner *sop)
349 {
350         struct nfs4_stateid *stp;
351
352         list_del(&sop->so_idhash);
353         list_del(&sop->so_strhash);
354         list_del(&sop->so_perclient);
355         list_del(&sop->so_perstateid); /* XXX: necessary? */
356         while (!list_empty(&sop->so_stateids)) {
357                 stp = list_first_entry(&sop->so_stateids,
358                                 struct nfs4_stateid, st_perstateowner);
359                 release_open_stateid(stp);
360         }
361 }
362
363 static void release_openowner(struct nfs4_stateowner *sop)
364 {
365         unhash_openowner(sop);
366         list_del(&sop->so_close_lru);
367         nfs4_put_stateowner(sop);
368 }
369
370 static DEFINE_SPINLOCK(sessionid_lock);
371 #define SESSION_HASH_SIZE       512
372 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
373
374 static inline int
375 hash_sessionid(struct nfs4_sessionid *sessionid)
376 {
377         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
378
379         return sid->sequence % SESSION_HASH_SIZE;
380 }
381
382 static inline void
383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
384 {
385         u32 *ptr = (u32 *)(&sessionid->data[0]);
386         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
387 }
388
389 static void
390 gen_sessionid(struct nfsd4_session *ses)
391 {
392         struct nfs4_client *clp = ses->se_client;
393         struct nfsd4_sessionid *sid;
394
395         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
396         sid->clientid = clp->cl_clientid;
397         sid->sequence = current_sessionid++;
398         sid->reserved = 0;
399 }
400
401 /*
402  * The protocol defines ca_maxresponssize_cached to include the size of
403  * the rpc header, but all we need to cache is the data starting after
404  * the end of the initial SEQUENCE operation--the rest we regenerate
405  * each time.  Therefore we can advertise a ca_maxresponssize_cached
406  * value that is the number of bytes in our cache plus a few additional
407  * bytes.  In order to stay on the safe side, and not promise more than
408  * we can cache, those additional bytes must be the minimum possible: 24
409  * bytes of rpc header (xid through accept state, with AUTH_NULL
410  * verifier), 12 for the compound header (with zero-length tag), and 44
411  * for the SEQUENCE op response:
412  */
413 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
414
415 /*
416  * Give the client the number of ca_maxresponsesize_cached slots it
417  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
418  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
419  * than NFSD_MAX_SLOTS_PER_SESSION.
420  *
421  * If we run out of reserved DRC memory we should (up to a point)
422  * re-negotiate active sessions and reduce their slot usage to make
423  * rooom for new connections. For now we just fail the create session.
424  */
425 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
426 {
427         int mem, size = fchan->maxresp_cached;
428
429         if (fchan->maxreqs < 1)
430                 return nfserr_inval;
431
432         if (size < NFSD_MIN_HDR_SEQ_SZ)
433                 size = NFSD_MIN_HDR_SEQ_SZ;
434         size -= NFSD_MIN_HDR_SEQ_SZ;
435         if (size > NFSD_SLOT_CACHE_SIZE)
436                 size = NFSD_SLOT_CACHE_SIZE;
437
438         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
439         mem = fchan->maxreqs * size;
440         if (mem > NFSD_MAX_MEM_PER_SESSION) {
441                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
442                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
443                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
444                 mem = fchan->maxreqs * size;
445         }
446
447         spin_lock(&nfsd_drc_lock);
448         /* bound the total session drc memory ussage */
449         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
450                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
451                 mem = fchan->maxreqs * size;
452         }
453         nfsd_drc_mem_used += mem;
454         spin_unlock(&nfsd_drc_lock);
455
456         if (fchan->maxreqs == 0)
457                 return nfserr_serverfault;
458
459         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
460         return 0;
461 }
462
463 /*
464  * fchan holds the client values on input, and the server values on output
465  * sv_max_mesg is the maximum payload plus one page for overhead.
466  */
467 static int init_forechannel_attrs(struct svc_rqst *rqstp,
468                                   struct nfsd4_channel_attrs *session_fchan,
469                                   struct nfsd4_channel_attrs *fchan)
470 {
471         int status = 0;
472         __u32   maxcount = nfsd_serv->sv_max_mesg;
473
474         /* headerpadsz set to zero in encode routine */
475
476         /* Use the client's max request and max response size if possible */
477         if (fchan->maxreq_sz > maxcount)
478                 fchan->maxreq_sz = maxcount;
479         session_fchan->maxreq_sz = fchan->maxreq_sz;
480
481         if (fchan->maxresp_sz > maxcount)
482                 fchan->maxresp_sz = maxcount;
483         session_fchan->maxresp_sz = fchan->maxresp_sz;
484
485         /* Use the client's maxops if possible */
486         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
487                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
488         session_fchan->maxops = fchan->maxops;
489
490         /* FIXME: Error means no more DRC pages so the server should
491          * recover pages from existing sessions. For now fail session
492          * creation.
493          */
494         status = set_forechannel_drc_size(fchan);
495
496         session_fchan->maxresp_cached = fchan->maxresp_cached;
497         session_fchan->maxreqs = fchan->maxreqs;
498
499         dprintk("%s status %d\n", __func__, status);
500         return status;
501 }
502
503 static void
504 free_session_slots(struct nfsd4_session *ses)
505 {
506         int i;
507
508         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
509                 kfree(ses->se_slots[i]);
510 }
511
512 /*
513  * We don't actually need to cache the rpc and session headers, so we
514  * can allocate a little less for each slot:
515  */
516 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
517 {
518         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
519 }
520
521 static int
522 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
523                    struct nfsd4_create_session *cses)
524 {
525         struct nfsd4_session *new, tmp;
526         struct nfsd4_slot *sp;
527         int idx, slotsize, cachesize, i;
528         int status;
529
530         memset(&tmp, 0, sizeof(tmp));
531
532         /* FIXME: For now, we just accept the client back channel attributes. */
533         tmp.se_bchannel = cses->back_channel;
534         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
535                                         &cses->fore_channel);
536         if (status)
537                 goto out;
538
539         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
540                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
541
542         status = nfserr_serverfault;
543         /* allocate struct nfsd4_session and slot table pointers in one piece */
544         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
545         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
546         if (!new)
547                 goto out;
548
549         memcpy(new, &tmp, sizeof(*new));
550
551         /* allocate each struct nfsd4_slot and data cache in one piece */
552         cachesize = slot_bytes(&new->se_fchannel);
553         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
554                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
555                 if (!sp)
556                         goto out_free;
557                 new->se_slots[i] = sp;
558         }
559
560         new->se_client = clp;
561         gen_sessionid(new);
562         idx = hash_sessionid(&new->se_sessionid);
563         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
564                NFS4_MAX_SESSIONID_LEN);
565
566         new->se_flags = cses->flags;
567         kref_init(&new->se_ref);
568         spin_lock(&sessionid_lock);
569         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
570         list_add(&new->se_perclnt, &clp->cl_sessions);
571         spin_unlock(&sessionid_lock);
572
573         status = nfs_ok;
574 out:
575         return status;
576 out_free:
577         free_session_slots(new);
578         kfree(new);
579         goto out;
580 }
581
582 /* caller must hold sessionid_lock */
583 static struct nfsd4_session *
584 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
585 {
586         struct nfsd4_session *elem;
587         int idx;
588
589         dump_sessionid(__func__, sessionid);
590         idx = hash_sessionid(sessionid);
591         dprintk("%s: idx is %d\n", __func__, idx);
592         /* Search in the appropriate list */
593         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
594                 dump_sessionid("list traversal", &elem->se_sessionid);
595                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
596                             NFS4_MAX_SESSIONID_LEN)) {
597                         return elem;
598                 }
599         }
600
601         dprintk("%s: session not found\n", __func__);
602         return NULL;
603 }
604
605 /* caller must hold sessionid_lock */
606 static void
607 unhash_session(struct nfsd4_session *ses)
608 {
609         list_del(&ses->se_hash);
610         list_del(&ses->se_perclnt);
611 }
612
613 static void
614 release_session(struct nfsd4_session *ses)
615 {
616         spin_lock(&sessionid_lock);
617         unhash_session(ses);
618         spin_unlock(&sessionid_lock);
619         nfsd4_put_session(ses);
620 }
621
622 void
623 free_session(struct kref *kref)
624 {
625         struct nfsd4_session *ses;
626         int mem;
627
628         ses = container_of(kref, struct nfsd4_session, se_ref);
629         spin_lock(&nfsd_drc_lock);
630         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
631         nfsd_drc_mem_used -= mem;
632         spin_unlock(&nfsd_drc_lock);
633         free_session_slots(ses);
634         kfree(ses);
635 }
636
637 static inline void
638 renew_client(struct nfs4_client *clp)
639 {
640         /*
641         * Move client to the end to the LRU list.
642         */
643         dprintk("renewing client (clientid %08x/%08x)\n", 
644                         clp->cl_clientid.cl_boot, 
645                         clp->cl_clientid.cl_id);
646         list_move_tail(&clp->cl_lru, &client_lru);
647         clp->cl_time = get_seconds();
648 }
649
650 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
651 static int
652 STALE_CLIENTID(clientid_t *clid)
653 {
654         if (clid->cl_boot == boot_time)
655                 return 0;
656         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
657                 clid->cl_boot, clid->cl_id, boot_time);
658         return 1;
659 }
660
661 /* 
662  * XXX Should we use a slab cache ?
663  * This type of memory management is somewhat inefficient, but we use it
664  * anyway since SETCLIENTID is not a common operation.
665  */
666 static struct nfs4_client *alloc_client(struct xdr_netobj name)
667 {
668         struct nfs4_client *clp;
669
670         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
671         if (clp == NULL)
672                 return NULL;
673         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
674         if (clp->cl_name.data == NULL) {
675                 kfree(clp);
676                 return NULL;
677         }
678         memcpy(clp->cl_name.data, name.data, name.len);
679         clp->cl_name.len = name.len;
680         return clp;
681 }
682
683 static inline void
684 free_client(struct nfs4_client *clp)
685 {
686         if (clp->cl_cred.cr_group_info)
687                 put_group_info(clp->cl_cred.cr_group_info);
688         kfree(clp->cl_principal);
689         kfree(clp->cl_name.data);
690         kfree(clp);
691 }
692
693 static void
694 expire_client(struct nfs4_client *clp)
695 {
696         struct nfs4_stateowner *sop;
697         struct nfs4_delegation *dp;
698         struct list_head reaplist;
699
700         INIT_LIST_HEAD(&reaplist);
701         spin_lock(&recall_lock);
702         while (!list_empty(&clp->cl_delegations)) {
703                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
704                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
705                                 dp->dl_flock);
706                 list_del_init(&dp->dl_perclnt);
707                 list_move(&dp->dl_recall_lru, &reaplist);
708         }
709         spin_unlock(&recall_lock);
710         while (!list_empty(&reaplist)) {
711                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
712                 list_del_init(&dp->dl_recall_lru);
713                 unhash_delegation(dp);
714         }
715         list_del(&clp->cl_idhash);
716         list_del(&clp->cl_strhash);
717         list_del(&clp->cl_lru);
718         while (!list_empty(&clp->cl_openowners)) {
719                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
720                 release_openowner(sop);
721         }
722         while (!list_empty(&clp->cl_sessions)) {
723                 struct nfsd4_session  *ses;
724                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
725                                  se_perclnt);
726                 release_session(ses);
727         }
728         nfsd4_set_callback_client(clp, NULL);
729         if (clp->cl_cb_conn.cb_xprt)
730                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
731         free_client(clp);
732 }
733
734 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
735 {
736         memcpy(target->cl_verifier.data, source->data,
737                         sizeof(target->cl_verifier.data));
738 }
739
740 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
741 {
742         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
743         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
744 }
745
746 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
747 {
748         target->cr_uid = source->cr_uid;
749         target->cr_gid = source->cr_gid;
750         target->cr_group_info = source->cr_group_info;
751         get_group_info(target->cr_group_info);
752 }
753
754 static int same_name(const char *n1, const char *n2)
755 {
756         return 0 == memcmp(n1, n2, HEXDIR_LEN);
757 }
758
759 static int
760 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
761 {
762         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
763 }
764
765 static int
766 same_clid(clientid_t *cl1, clientid_t *cl2)
767 {
768         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
769 }
770
771 /* XXX what about NGROUP */
772 static int
773 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
774 {
775         return cr1->cr_uid == cr2->cr_uid;
776 }
777
778 static void gen_clid(struct nfs4_client *clp)
779 {
780         static u32 current_clientid = 1;
781
782         clp->cl_clientid.cl_boot = boot_time;
783         clp->cl_clientid.cl_id = current_clientid++; 
784 }
785
786 static void gen_confirm(struct nfs4_client *clp)
787 {
788         static u32 i;
789         u32 *p;
790
791         p = (u32 *)clp->cl_confirm.data;
792         *p++ = get_seconds();
793         *p++ = i++;
794 }
795
796 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
797                 struct svc_rqst *rqstp, nfs4_verifier *verf)
798 {
799         struct nfs4_client *clp;
800         struct sockaddr *sa = svc_addr(rqstp);
801         char *princ;
802
803         clp = alloc_client(name);
804         if (clp == NULL)
805                 return NULL;
806
807         princ = svc_gss_principal(rqstp);
808         if (princ) {
809                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
810                 if (clp->cl_principal == NULL) {
811                         free_client(clp);
812                         return NULL;
813                 }
814         }
815
816         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
817         atomic_set(&clp->cl_cb_set, 0);
818         INIT_LIST_HEAD(&clp->cl_idhash);
819         INIT_LIST_HEAD(&clp->cl_strhash);
820         INIT_LIST_HEAD(&clp->cl_openowners);
821         INIT_LIST_HEAD(&clp->cl_delegations);
822         INIT_LIST_HEAD(&clp->cl_sessions);
823         INIT_LIST_HEAD(&clp->cl_lru);
824         clear_bit(0, &clp->cl_cb_slot_busy);
825         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
826         copy_verf(clp, verf);
827         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
828         clp->cl_flavor = rqstp->rq_flavor;
829         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
830         gen_confirm(clp);
831
832         return clp;
833 }
834
835 static int check_name(struct xdr_netobj name)
836 {
837         if (name.len == 0) 
838                 return 0;
839         if (name.len > NFS4_OPAQUE_LIMIT) {
840                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
841                 return 0;
842         }
843         return 1;
844 }
845
846 static void
847 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
848 {
849         unsigned int idhashval;
850
851         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
852         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
853         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
854         list_add_tail(&clp->cl_lru, &client_lru);
855         clp->cl_time = get_seconds();
856 }
857
858 static void
859 move_to_confirmed(struct nfs4_client *clp)
860 {
861         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
862         unsigned int strhashval;
863
864         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
865         list_del_init(&clp->cl_strhash);
866         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
867         strhashval = clientstr_hashval(clp->cl_recdir);
868         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
869         renew_client(clp);
870 }
871
872 static struct nfs4_client *
873 find_confirmed_client(clientid_t *clid)
874 {
875         struct nfs4_client *clp;
876         unsigned int idhashval = clientid_hashval(clid->cl_id);
877
878         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
879                 if (same_clid(&clp->cl_clientid, clid))
880                         return clp;
881         }
882         return NULL;
883 }
884
885 static struct nfs4_client *
886 find_unconfirmed_client(clientid_t *clid)
887 {
888         struct nfs4_client *clp;
889         unsigned int idhashval = clientid_hashval(clid->cl_id);
890
891         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
892                 if (same_clid(&clp->cl_clientid, clid))
893                         return clp;
894         }
895         return NULL;
896 }
897
898 /*
899  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
900  * parameter. Matching is based on the fact the at least one of the
901  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
902  *
903  * FIXME: we need to unify the clientid namespaces for nfsv4.x
904  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
905  * and SET_CLIENTID{,_CONFIRM}
906  */
907 static inline int
908 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
909 {
910         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
911         return use_exchange_id == has_exchange_flags;
912 }
913
914 static struct nfs4_client *
915 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
916                              bool use_exchange_id)
917 {
918         struct nfs4_client *clp;
919
920         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
921                 if (same_name(clp->cl_recdir, dname) &&
922                     match_clientid_establishment(clp, use_exchange_id))
923                         return clp;
924         }
925         return NULL;
926 }
927
928 static struct nfs4_client *
929 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
930                                bool use_exchange_id)
931 {
932         struct nfs4_client *clp;
933
934         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
935                 if (same_name(clp->cl_recdir, dname) &&
936                     match_clientid_establishment(clp, use_exchange_id))
937                         return clp;
938         }
939         return NULL;
940 }
941
942 static void
943 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
944 {
945         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
946         unsigned short expected_family;
947
948         /* Currently, we only support tcp and tcp6 for the callback channel */
949         if (se->se_callback_netid_len == 3 &&
950             !memcmp(se->se_callback_netid_val, "tcp", 3))
951                 expected_family = AF_INET;
952         else if (se->se_callback_netid_len == 4 &&
953                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
954                 expected_family = AF_INET6;
955         else
956                 goto out_err;
957
958         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
959                                             se->se_callback_addr_len,
960                                             (struct sockaddr *) &cb->cb_addr,
961                                             sizeof(cb->cb_addr));
962
963         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
964                 goto out_err;
965
966         if (cb->cb_addr.ss_family == AF_INET6)
967                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
968
969         cb->cb_minorversion = 0;
970         cb->cb_prog = se->se_callback_prog;
971         cb->cb_ident = se->se_callback_ident;
972         return;
973 out_err:
974         cb->cb_addr.ss_family = AF_UNSPEC;
975         cb->cb_addrlen = 0;
976         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
977                 "will not receive delegations\n",
978                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
979
980         return;
981 }
982
983 /*
984  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
985  */
986 void
987 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
988 {
989         struct nfsd4_slot *slot = resp->cstate.slot;
990         unsigned int base;
991
992         dprintk("--> %s slot %p\n", __func__, slot);
993
994         slot->sl_opcnt = resp->opcnt;
995         slot->sl_status = resp->cstate.status;
996
997         if (nfsd4_not_cached(resp)) {
998                 slot->sl_datalen = 0;
999                 return;
1000         }
1001         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1002         base = (char *)resp->cstate.datap -
1003                                         (char *)resp->xbuf->head[0].iov_base;
1004         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1005                                     slot->sl_datalen))
1006                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1007         return;
1008 }
1009
1010 /*
1011  * Encode the replay sequence operation from the slot values.
1012  * If cachethis is FALSE encode the uncached rep error on the next
1013  * operation which sets resp->p and increments resp->opcnt for
1014  * nfs4svc_encode_compoundres.
1015  *
1016  */
1017 static __be32
1018 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1019                           struct nfsd4_compoundres *resp)
1020 {
1021         struct nfsd4_op *op;
1022         struct nfsd4_slot *slot = resp->cstate.slot;
1023
1024         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1025                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1026
1027         /* Encode the replayed sequence operation */
1028         op = &args->ops[resp->opcnt - 1];
1029         nfsd4_encode_operation(resp, op);
1030
1031         /* Return nfserr_retry_uncached_rep in next operation. */
1032         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1033                 op = &args->ops[resp->opcnt++];
1034                 op->status = nfserr_retry_uncached_rep;
1035                 nfsd4_encode_operation(resp, op);
1036         }
1037         return op->status;
1038 }
1039
1040 /*
1041  * The sequence operation is not cached because we can use the slot and
1042  * session values.
1043  */
1044 __be32
1045 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1046                          struct nfsd4_sequence *seq)
1047 {
1048         struct nfsd4_slot *slot = resp->cstate.slot;
1049         __be32 status;
1050
1051         dprintk("--> %s slot %p\n", __func__, slot);
1052
1053         /* Either returns 0 or nfserr_retry_uncached */
1054         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1055         if (status == nfserr_retry_uncached_rep)
1056                 return status;
1057
1058         /* The sequence operation has been encoded, cstate->datap set. */
1059         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1060
1061         resp->opcnt = slot->sl_opcnt;
1062         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1063         status = slot->sl_status;
1064
1065         return status;
1066 }
1067
1068 /*
1069  * Set the exchange_id flags returned by the server.
1070  */
1071 static void
1072 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1073 {
1074         /* pNFS is not supported */
1075         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1076
1077         /* Referrals are supported, Migration is not. */
1078         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1079
1080         /* set the wire flags to return to client. */
1081         clid->flags = new->cl_exchange_flags;
1082 }
1083
1084 __be32
1085 nfsd4_exchange_id(struct svc_rqst *rqstp,
1086                   struct nfsd4_compound_state *cstate,
1087                   struct nfsd4_exchange_id *exid)
1088 {
1089         struct nfs4_client *unconf, *conf, *new;
1090         int status;
1091         unsigned int            strhashval;
1092         char                    dname[HEXDIR_LEN];
1093         char                    addr_str[INET6_ADDRSTRLEN];
1094         nfs4_verifier           verf = exid->verifier;
1095         struct sockaddr         *sa = svc_addr(rqstp);
1096
1097         rpc_ntop(sa, addr_str, sizeof(addr_str));
1098         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1099                 "ip_addr=%s flags %x, spa_how %d\n",
1100                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1101                 addr_str, exid->flags, exid->spa_how);
1102
1103         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1104                 return nfserr_inval;
1105
1106         /* Currently only support SP4_NONE */
1107         switch (exid->spa_how) {
1108         case SP4_NONE:
1109                 break;
1110         case SP4_SSV:
1111                 return nfserr_encr_alg_unsupp;
1112         default:
1113                 BUG();                          /* checked by xdr code */
1114         case SP4_MACH_CRED:
1115                 return nfserr_serverfault;      /* no excuse :-/ */
1116         }
1117
1118         status = nfs4_make_rec_clidname(dname, &exid->clname);
1119
1120         if (status)
1121                 goto error;
1122
1123         strhashval = clientstr_hashval(dname);
1124
1125         nfs4_lock_state();
1126         status = nfs_ok;
1127
1128         conf = find_confirmed_client_by_str(dname, strhashval, true);
1129         if (conf) {
1130                 if (!same_verf(&verf, &conf->cl_verifier)) {
1131                         /* 18.35.4 case 8 */
1132                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1133                                 status = nfserr_not_same;
1134                                 goto out;
1135                         }
1136                         /* Client reboot: destroy old state */
1137                         expire_client(conf);
1138                         goto out_new;
1139                 }
1140                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1141                         /* 18.35.4 case 9 */
1142                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1143                                 status = nfserr_perm;
1144                                 goto out;
1145                         }
1146                         expire_client(conf);
1147                         goto out_new;
1148                 }
1149                 /*
1150                  * Set bit when the owner id and verifier map to an already
1151                  * confirmed client id (18.35.3).
1152                  */
1153                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1154
1155                 /*
1156                  * Falling into 18.35.4 case 2, possible router replay.
1157                  * Leave confirmed record intact and return same result.
1158                  */
1159                 copy_verf(conf, &verf);
1160                 new = conf;
1161                 goto out_copy;
1162         }
1163
1164         /* 18.35.4 case 7 */
1165         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1166                 status = nfserr_noent;
1167                 goto out;
1168         }
1169
1170         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1171         if (unconf) {
1172                 /*
1173                  * Possible retry or client restart.  Per 18.35.4 case 4,
1174                  * a new unconfirmed record should be generated regardless
1175                  * of whether any properties have changed.
1176                  */
1177                 expire_client(unconf);
1178         }
1179
1180 out_new:
1181         /* Normal case */
1182         new = create_client(exid->clname, dname, rqstp, &verf);
1183         if (new == NULL) {
1184                 status = nfserr_serverfault;
1185                 goto out;
1186         }
1187
1188         gen_clid(new);
1189         add_to_unconfirmed(new, strhashval);
1190 out_copy:
1191         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1192         exid->clientid.cl_id = new->cl_clientid.cl_id;
1193
1194         exid->seqid = 1;
1195         nfsd4_set_ex_flags(new, exid);
1196
1197         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1198                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1199         status = nfs_ok;
1200
1201 out:
1202         nfs4_unlock_state();
1203 error:
1204         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1205         return status;
1206 }
1207
1208 static int
1209 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1210 {
1211         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1212                 slot_seqid);
1213
1214         /* The slot is in use, and no response has been sent. */
1215         if (slot_inuse) {
1216                 if (seqid == slot_seqid)
1217                         return nfserr_jukebox;
1218                 else
1219                         return nfserr_seq_misordered;
1220         }
1221         /* Normal */
1222         if (likely(seqid == slot_seqid + 1))
1223                 return nfs_ok;
1224         /* Replay */
1225         if (seqid == slot_seqid)
1226                 return nfserr_replay_cache;
1227         /* Wraparound */
1228         if (seqid == 1 && (slot_seqid + 1) == 0)
1229                 return nfs_ok;
1230         /* Misordered replay or misordered new request */
1231         return nfserr_seq_misordered;
1232 }
1233
1234 /*
1235  * Cache the create session result into the create session single DRC
1236  * slot cache by saving the xdr structure. sl_seqid has been set.
1237  * Do this for solo or embedded create session operations.
1238  */
1239 static void
1240 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1241                            struct nfsd4_clid_slot *slot, int nfserr)
1242 {
1243         slot->sl_status = nfserr;
1244         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1245 }
1246
1247 static __be32
1248 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1249                             struct nfsd4_clid_slot *slot)
1250 {
1251         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1252         return slot->sl_status;
1253 }
1254
1255 __be32
1256 nfsd4_create_session(struct svc_rqst *rqstp,
1257                      struct nfsd4_compound_state *cstate,
1258                      struct nfsd4_create_session *cr_ses)
1259 {
1260         struct sockaddr *sa = svc_addr(rqstp);
1261         struct nfs4_client *conf, *unconf;
1262         struct nfsd4_clid_slot *cs_slot = NULL;
1263         int status = 0;
1264
1265         nfs4_lock_state();
1266         unconf = find_unconfirmed_client(&cr_ses->clientid);
1267         conf = find_confirmed_client(&cr_ses->clientid);
1268
1269         if (conf) {
1270                 cs_slot = &conf->cl_cs_slot;
1271                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1272                 if (status == nfserr_replay_cache) {
1273                         dprintk("Got a create_session replay! seqid= %d\n",
1274                                 cs_slot->sl_seqid);
1275                         /* Return the cached reply status */
1276                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1277                         goto out;
1278                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1279                         status = nfserr_seq_misordered;
1280                         dprintk("Sequence misordered!\n");
1281                         dprintk("Expected seqid= %d but got seqid= %d\n",
1282                                 cs_slot->sl_seqid, cr_ses->seqid);
1283                         goto out;
1284                 }
1285                 cs_slot->sl_seqid++;
1286         } else if (unconf) {
1287                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1288                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1289                         status = nfserr_clid_inuse;
1290                         goto out;
1291                 }
1292
1293                 cs_slot = &unconf->cl_cs_slot;
1294                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1295                 if (status) {
1296                         /* an unconfirmed replay returns misordered */
1297                         status = nfserr_seq_misordered;
1298                         goto out_cache;
1299                 }
1300
1301                 cs_slot->sl_seqid++; /* from 0 to 1 */
1302                 move_to_confirmed(unconf);
1303
1304                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1305                         unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1306                         svc_xprt_get(rqstp->rq_xprt);
1307                         rpc_copy_addr(
1308                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1309                                 sa);
1310                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1311                         unconf->cl_cb_conn.cb_minorversion =
1312                                 cstate->minorversion;
1313                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1314                         unconf->cl_cb_seq_nr = 1;
1315                         nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1316                 }
1317                 conf = unconf;
1318         } else {
1319                 status = nfserr_stale_clientid;
1320                 goto out;
1321         }
1322
1323         /*
1324          * We do not support RDMA or persistent sessions
1325          */
1326         cr_ses->flags &= ~SESSION4_PERSIST;
1327         cr_ses->flags &= ~SESSION4_RDMA;
1328
1329         status = alloc_init_session(rqstp, conf, cr_ses);
1330         if (status)
1331                 goto out;
1332
1333         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1334                NFS4_MAX_SESSIONID_LEN);
1335         cr_ses->seqid = cs_slot->sl_seqid;
1336
1337 out_cache:
1338         /* cache solo and embedded create sessions under the state lock */
1339         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1340 out:
1341         nfs4_unlock_state();
1342         dprintk("%s returns %d\n", __func__, ntohl(status));
1343         return status;
1344 }
1345
1346 __be32
1347 nfsd4_destroy_session(struct svc_rqst *r,
1348                       struct nfsd4_compound_state *cstate,
1349                       struct nfsd4_destroy_session *sessionid)
1350 {
1351         struct nfsd4_session *ses;
1352         u32 status = nfserr_badsession;
1353
1354         /* Notes:
1355          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1356          * - Should we return nfserr_back_chan_busy if waiting for
1357          *   callbacks on to-be-destroyed session?
1358          * - Do we need to clear any callback info from previous session?
1359          */
1360
1361         dump_sessionid(__func__, &sessionid->sessionid);
1362         spin_lock(&sessionid_lock);
1363         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1364         if (!ses) {
1365                 spin_unlock(&sessionid_lock);
1366                 goto out;
1367         }
1368
1369         unhash_session(ses);
1370         spin_unlock(&sessionid_lock);
1371
1372         /* wait for callbacks */
1373         nfsd4_set_callback_client(ses->se_client, NULL);
1374         nfsd4_put_session(ses);
1375         status = nfs_ok;
1376 out:
1377         dprintk("%s returns %d\n", __func__, ntohl(status));
1378         return status;
1379 }
1380
1381 __be32
1382 nfsd4_sequence(struct svc_rqst *rqstp,
1383                struct nfsd4_compound_state *cstate,
1384                struct nfsd4_sequence *seq)
1385 {
1386         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1387         struct nfsd4_session *session;
1388         struct nfsd4_slot *slot;
1389         int status;
1390
1391         if (resp->opcnt != 1)
1392                 return nfserr_sequence_pos;
1393
1394         spin_lock(&sessionid_lock);
1395         status = nfserr_badsession;
1396         session = find_in_sessionid_hashtbl(&seq->sessionid);
1397         if (!session)
1398                 goto out;
1399
1400         status = nfserr_badslot;
1401         if (seq->slotid >= session->se_fchannel.maxreqs)
1402                 goto out;
1403
1404         slot = session->se_slots[seq->slotid];
1405         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1406
1407         /* We do not negotiate the number of slots yet, so set the
1408          * maxslots to the session maxreqs which is used to encode
1409          * sr_highest_slotid and the sr_target_slot id to maxslots */
1410         seq->maxslots = session->se_fchannel.maxreqs;
1411
1412         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1413         if (status == nfserr_replay_cache) {
1414                 cstate->slot = slot;
1415                 cstate->session = session;
1416                 /* Return the cached reply status and set cstate->status
1417                  * for nfsd4_proc_compound processing */
1418                 status = nfsd4_replay_cache_entry(resp, seq);
1419                 cstate->status = nfserr_replay_cache;
1420                 goto out;
1421         }
1422         if (status)
1423                 goto out;
1424
1425         /* Success! bump slot seqid */
1426         slot->sl_inuse = true;
1427         slot->sl_seqid = seq->seqid;
1428         slot->sl_cachethis = seq->cachethis;
1429
1430         cstate->slot = slot;
1431         cstate->session = session;
1432
1433         /* Hold a session reference until done processing the compound:
1434          * nfsd4_put_session called only if the cstate slot is set.
1435          */
1436         nfsd4_get_session(session);
1437 out:
1438         spin_unlock(&sessionid_lock);
1439         /* Renew the clientid on success and on replay */
1440         if (cstate->session) {
1441                 nfs4_lock_state();
1442                 renew_client(session->se_client);
1443                 nfs4_unlock_state();
1444         }
1445         dprintk("%s: return %d\n", __func__, ntohl(status));
1446         return status;
1447 }
1448
1449 __be32
1450 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1451                   struct nfsd4_setclientid *setclid)
1452 {
1453         struct sockaddr         *sa = svc_addr(rqstp);
1454         struct xdr_netobj       clname = { 
1455                 .len = setclid->se_namelen,
1456                 .data = setclid->se_name,
1457         };
1458         nfs4_verifier           clverifier = setclid->se_verf;
1459         unsigned int            strhashval;
1460         struct nfs4_client      *conf, *unconf, *new;
1461         __be32                  status;
1462         char                    dname[HEXDIR_LEN];
1463         
1464         if (!check_name(clname))
1465                 return nfserr_inval;
1466
1467         status = nfs4_make_rec_clidname(dname, &clname);
1468         if (status)
1469                 return status;
1470
1471         /* 
1472          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1473          * We get here on a DRC miss.
1474          */
1475
1476         strhashval = clientstr_hashval(dname);
1477
1478         nfs4_lock_state();
1479         conf = find_confirmed_client_by_str(dname, strhashval, false);
1480         if (conf) {
1481                 /* RFC 3530 14.2.33 CASE 0: */
1482                 status = nfserr_clid_inuse;
1483                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1484                         char addr_str[INET6_ADDRSTRLEN];
1485                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1486                                  sizeof(addr_str));
1487                         dprintk("NFSD: setclientid: string in use by client "
1488                                 "at %s\n", addr_str);
1489                         goto out;
1490                 }
1491         }
1492         /*
1493          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1494          * has a description of SETCLIENTID request processing consisting
1495          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1496          */
1497         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1498         status = nfserr_resource;
1499         if (!conf) {
1500                 /*
1501                  * RFC 3530 14.2.33 CASE 4:
1502                  * placed first, because it is the normal case
1503                  */
1504                 if (unconf)
1505                         expire_client(unconf);
1506                 new = create_client(clname, dname, rqstp, &clverifier);
1507                 if (new == NULL)
1508                         goto out;
1509                 gen_clid(new);
1510         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1511                 /*
1512                  * RFC 3530 14.2.33 CASE 1:
1513                  * probable callback update
1514                  */
1515                 if (unconf) {
1516                         /* Note this is removing unconfirmed {*x***},
1517                          * which is stronger than RFC recommended {vxc**}.
1518                          * This has the advantage that there is at most
1519                          * one {*x***} in either list at any time.
1520                          */
1521                         expire_client(unconf);
1522                 }
1523                 new = create_client(clname, dname, rqstp, &clverifier);
1524                 if (new == NULL)
1525                         goto out;
1526                 copy_clid(new, conf);
1527         } else if (!unconf) {
1528                 /*
1529                  * RFC 3530 14.2.33 CASE 2:
1530                  * probable client reboot; state will be removed if
1531                  * confirmed.
1532                  */
1533                 new = create_client(clname, dname, rqstp, &clverifier);
1534                 if (new == NULL)
1535                         goto out;
1536                 gen_clid(new);
1537         } else {
1538                 /*
1539                  * RFC 3530 14.2.33 CASE 3:
1540                  * probable client reboot; state will be removed if
1541                  * confirmed.
1542                  */
1543                 expire_client(unconf);
1544                 new = create_client(clname, dname, rqstp, &clverifier);
1545                 if (new == NULL)
1546                         goto out;
1547                 gen_clid(new);
1548         }
1549         gen_callback(new, setclid, rpc_get_scope_id(sa));
1550         add_to_unconfirmed(new, strhashval);
1551         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1552         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1553         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1554         status = nfs_ok;
1555 out:
1556         nfs4_unlock_state();
1557         return status;
1558 }
1559
1560
1561 /*
1562  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1563  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1564  * bullets, labeled as CASE1 - CASE4 below.
1565  */
1566 __be32
1567 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1568                          struct nfsd4_compound_state *cstate,
1569                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1570 {
1571         struct sockaddr *sa = svc_addr(rqstp);
1572         struct nfs4_client *conf, *unconf;
1573         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1574         clientid_t * clid = &setclientid_confirm->sc_clientid;
1575         __be32 status;
1576
1577         if (STALE_CLIENTID(clid))
1578                 return nfserr_stale_clientid;
1579         /* 
1580          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1581          * We get here on a DRC miss.
1582          */
1583
1584         nfs4_lock_state();
1585
1586         conf = find_confirmed_client(clid);
1587         unconf = find_unconfirmed_client(clid);
1588
1589         status = nfserr_clid_inuse;
1590         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1591                 goto out;
1592         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1593                 goto out;
1594
1595         /*
1596          * section 14.2.34 of RFC 3530 has a description of
1597          * SETCLIENTID_CONFIRM request processing consisting
1598          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1599          */
1600         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1601                 /*
1602                  * RFC 3530 14.2.34 CASE 1:
1603                  * callback update
1604                  */
1605                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1606                         status = nfserr_clid_inuse;
1607                 else {
1608                         atomic_set(&conf->cl_cb_set, 0);
1609                         nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1610                         expire_client(unconf);
1611                         status = nfs_ok;
1612
1613                 }
1614         } else if (conf && !unconf) {
1615                 /*
1616                  * RFC 3530 14.2.34 CASE 2:
1617                  * probable retransmitted request; play it safe and
1618                  * do nothing.
1619                  */
1620                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1621                         status = nfserr_clid_inuse;
1622                 else
1623                         status = nfs_ok;
1624         } else if (!conf && unconf
1625                         && same_verf(&unconf->cl_confirm, &confirm)) {
1626                 /*
1627                  * RFC 3530 14.2.34 CASE 3:
1628                  * Normal case; new or rebooted client:
1629                  */
1630                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1631                         status = nfserr_clid_inuse;
1632                 } else {
1633                         unsigned int hash =
1634                                 clientstr_hashval(unconf->cl_recdir);
1635                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1636                                                             hash, false);
1637                         if (conf) {
1638                                 nfsd4_remove_clid_dir(conf);
1639                                 expire_client(conf);
1640                         }
1641                         move_to_confirmed(unconf);
1642                         conf = unconf;
1643                         nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1644                         status = nfs_ok;
1645                 }
1646         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1647             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1648                                                                 &confirm)))) {
1649                 /*
1650                  * RFC 3530 14.2.34 CASE 4:
1651                  * Client probably hasn't noticed that we rebooted yet.
1652                  */
1653                 status = nfserr_stale_clientid;
1654         } else {
1655                 /* check that we have hit one of the cases...*/
1656                 status = nfserr_clid_inuse;
1657         }
1658 out:
1659         nfs4_unlock_state();
1660         return status;
1661 }
1662
1663 /* OPEN Share state helper functions */
1664 static inline struct nfs4_file *
1665 alloc_init_file(struct inode *ino)
1666 {
1667         struct nfs4_file *fp;
1668         unsigned int hashval = file_hashval(ino);
1669
1670         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1671         if (fp) {
1672                 atomic_set(&fp->fi_ref, 1);
1673                 INIT_LIST_HEAD(&fp->fi_hash);
1674                 INIT_LIST_HEAD(&fp->fi_stateids);
1675                 INIT_LIST_HEAD(&fp->fi_delegations);
1676                 spin_lock(&recall_lock);
1677                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1678                 spin_unlock(&recall_lock);
1679                 fp->fi_inode = igrab(ino);
1680                 fp->fi_id = current_fileid++;
1681                 fp->fi_had_conflict = false;
1682                 return fp;
1683         }
1684         return NULL;
1685 }
1686
1687 static void
1688 nfsd4_free_slab(struct kmem_cache **slab)
1689 {
1690         if (*slab == NULL)
1691                 return;
1692         kmem_cache_destroy(*slab);
1693         *slab = NULL;
1694 }
1695
1696 void
1697 nfsd4_free_slabs(void)
1698 {
1699         nfsd4_free_slab(&stateowner_slab);
1700         nfsd4_free_slab(&file_slab);
1701         nfsd4_free_slab(&stateid_slab);
1702         nfsd4_free_slab(&deleg_slab);
1703 }
1704
1705 static int
1706 nfsd4_init_slabs(void)
1707 {
1708         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1709                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1710         if (stateowner_slab == NULL)
1711                 goto out_nomem;
1712         file_slab = kmem_cache_create("nfsd4_files",
1713                         sizeof(struct nfs4_file), 0, 0, NULL);
1714         if (file_slab == NULL)
1715                 goto out_nomem;
1716         stateid_slab = kmem_cache_create("nfsd4_stateids",
1717                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1718         if (stateid_slab == NULL)
1719                 goto out_nomem;
1720         deleg_slab = kmem_cache_create("nfsd4_delegations",
1721                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1722         if (deleg_slab == NULL)
1723                 goto out_nomem;
1724         return 0;
1725 out_nomem:
1726         nfsd4_free_slabs();
1727         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1728         return -ENOMEM;
1729 }
1730
1731 void
1732 nfs4_free_stateowner(struct kref *kref)
1733 {
1734         struct nfs4_stateowner *sop =
1735                 container_of(kref, struct nfs4_stateowner, so_ref);
1736         kfree(sop->so_owner.data);
1737         kmem_cache_free(stateowner_slab, sop);
1738 }
1739
1740 static inline struct nfs4_stateowner *
1741 alloc_stateowner(struct xdr_netobj *owner)
1742 {
1743         struct nfs4_stateowner *sop;
1744
1745         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1746                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1747                         memcpy(sop->so_owner.data, owner->data, owner->len);
1748                         sop->so_owner.len = owner->len;
1749                         kref_init(&sop->so_ref);
1750                         return sop;
1751                 } 
1752                 kmem_cache_free(stateowner_slab, sop);
1753         }
1754         return NULL;
1755 }
1756
1757 static struct nfs4_stateowner *
1758 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1759         struct nfs4_stateowner *sop;
1760         struct nfs4_replay *rp;
1761         unsigned int idhashval;
1762
1763         if (!(sop = alloc_stateowner(&open->op_owner)))
1764                 return NULL;
1765         idhashval = ownerid_hashval(current_ownerid);
1766         INIT_LIST_HEAD(&sop->so_idhash);
1767         INIT_LIST_HEAD(&sop->so_strhash);
1768         INIT_LIST_HEAD(&sop->so_perclient);
1769         INIT_LIST_HEAD(&sop->so_stateids);
1770         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1771         INIT_LIST_HEAD(&sop->so_close_lru);
1772         sop->so_time = 0;
1773         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1774         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1775         list_add(&sop->so_perclient, &clp->cl_openowners);
1776         sop->so_is_open_owner = 1;
1777         sop->so_id = current_ownerid++;
1778         sop->so_client = clp;
1779         sop->so_seqid = open->op_seqid;
1780         sop->so_confirmed = 0;
1781         rp = &sop->so_replay;
1782         rp->rp_status = nfserr_serverfault;
1783         rp->rp_buflen = 0;
1784         rp->rp_buf = rp->rp_ibuf;
1785         return sop;
1786 }
1787
1788 static inline void
1789 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1790         struct nfs4_stateowner *sop = open->op_stateowner;
1791         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1792
1793         INIT_LIST_HEAD(&stp->st_hash);
1794         INIT_LIST_HEAD(&stp->st_perstateowner);
1795         INIT_LIST_HEAD(&stp->st_lockowners);
1796         INIT_LIST_HEAD(&stp->st_perfile);
1797         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1798         list_add(&stp->st_perstateowner, &sop->so_stateids);
1799         list_add(&stp->st_perfile, &fp->fi_stateids);
1800         stp->st_stateowner = sop;
1801         get_nfs4_file(fp);
1802         stp->st_file = fp;
1803         stp->st_stateid.si_boot = get_seconds();
1804         stp->st_stateid.si_stateownerid = sop->so_id;
1805         stp->st_stateid.si_fileid = fp->fi_id;
1806         stp->st_stateid.si_generation = 0;
1807         stp->st_access_bmap = 0;
1808         stp->st_deny_bmap = 0;
1809         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1810                   &stp->st_access_bmap);
1811         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1812         stp->st_openstp = NULL;
1813 }
1814
1815 static void
1816 move_to_close_lru(struct nfs4_stateowner *sop)
1817 {
1818         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1819
1820         list_move_tail(&sop->so_close_lru, &close_lru);
1821         sop->so_time = get_seconds();
1822 }
1823
1824 static int
1825 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1826                                                         clientid_t *clid)
1827 {
1828         return (sop->so_owner.len == owner->len) &&
1829                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1830                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1831 }
1832
1833 static struct nfs4_stateowner *
1834 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1835 {
1836         struct nfs4_stateowner *so = NULL;
1837
1838         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1839                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1840                         return so;
1841         }
1842         return NULL;
1843 }
1844
1845 /* search file_hashtbl[] for file */
1846 static struct nfs4_file *
1847 find_file(struct inode *ino)
1848 {
1849         unsigned int hashval = file_hashval(ino);
1850         struct nfs4_file *fp;
1851
1852         spin_lock(&recall_lock);
1853         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1854                 if (fp->fi_inode == ino) {
1855                         get_nfs4_file(fp);
1856                         spin_unlock(&recall_lock);
1857                         return fp;
1858                 }
1859         }
1860         spin_unlock(&recall_lock);
1861         return NULL;
1862 }
1863
1864 static inline int access_valid(u32 x, u32 minorversion)
1865 {
1866         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1867                 return 0;
1868         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1869                 return 0;
1870         x &= ~NFS4_SHARE_ACCESS_MASK;
1871         if (minorversion && x) {
1872                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1873                         return 0;
1874                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1875                         return 0;
1876                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1877         }
1878         if (x)
1879                 return 0;
1880         return 1;
1881 }
1882
1883 static inline int deny_valid(u32 x)
1884 {
1885         /* Note: unlike access bits, deny bits may be zero. */
1886         return x <= NFS4_SHARE_DENY_BOTH;
1887 }
1888
1889 /*
1890  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1891  * st_{access,deny}_bmap field of the stateid, in order to track not
1892  * only what share bits are currently in force, but also what
1893  * combinations of share bits previous opens have used.  This allows us
1894  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1895  * return an error if the client attempt to downgrade to a combination
1896  * of share bits not explicable by closing some of its previous opens.
1897  *
1898  * XXX: This enforcement is actually incomplete, since we don't keep
1899  * track of access/deny bit combinations; so, e.g., we allow:
1900  *
1901  *      OPEN allow read, deny write
1902  *      OPEN allow both, deny none
1903  *      DOWNGRADE allow read, deny none
1904  *
1905  * which we should reject.
1906  */
1907 static void
1908 set_access(unsigned int *access, unsigned long bmap) {
1909         int i;
1910
1911         *access = 0;
1912         for (i = 1; i < 4; i++) {
1913                 if (test_bit(i, &bmap))
1914                         *access |= i;
1915         }
1916 }
1917
1918 static void
1919 set_deny(unsigned int *deny, unsigned long bmap) {
1920         int i;
1921
1922         *deny = 0;
1923         for (i = 0; i < 4; i++) {
1924                 if (test_bit(i, &bmap))
1925                         *deny |= i ;
1926         }
1927 }
1928
1929 static int
1930 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1931         unsigned int access, deny;
1932
1933         set_access(&access, stp->st_access_bmap);
1934         set_deny(&deny, stp->st_deny_bmap);
1935         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1936                 return 0;
1937         return 1;
1938 }
1939
1940 /*
1941  * Called to check deny when READ with all zero stateid or
1942  * WRITE with all zero or all one stateid
1943  */
1944 static __be32
1945 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1946 {
1947         struct inode *ino = current_fh->fh_dentry->d_inode;
1948         struct nfs4_file *fp;
1949         struct nfs4_stateid *stp;
1950         __be32 ret;
1951
1952         dprintk("NFSD: nfs4_share_conflict\n");
1953
1954         fp = find_file(ino);
1955         if (!fp)
1956                 return nfs_ok;
1957         ret = nfserr_locked;
1958         /* Search for conflicting share reservations */
1959         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1960                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1961                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1962                         goto out;
1963         }
1964         ret = nfs_ok;
1965 out:
1966         put_nfs4_file(fp);
1967         return ret;
1968 }
1969
1970 static inline void
1971 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1972 {
1973         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
1974                 drop_file_write_access(filp);
1975                 spin_lock(&filp->f_lock);
1976                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
1977                 spin_unlock(&filp->f_lock);
1978         }
1979 }
1980
1981 /*
1982  * Spawn a thread to perform a recall on the delegation represented
1983  * by the lease (file_lock)
1984  *
1985  * Called from break_lease() with lock_kernel() held.
1986  * Note: we assume break_lease will only call this *once* for any given
1987  * lease.
1988  */
1989 static
1990 void nfsd_break_deleg_cb(struct file_lock *fl)
1991 {
1992         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
1993
1994         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
1995         if (!dp)
1996                 return;
1997
1998         /* We're assuming the state code never drops its reference
1999          * without first removing the lease.  Since we're in this lease
2000          * callback (and since the lease code is serialized by the kernel
2001          * lock) we know the server hasn't removed the lease yet, we know
2002          * it's safe to take a reference: */
2003         atomic_inc(&dp->dl_count);
2004
2005         spin_lock(&recall_lock);
2006         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2007         spin_unlock(&recall_lock);
2008
2009         /* only place dl_time is set. protected by lock_kernel*/
2010         dp->dl_time = get_seconds();
2011
2012         /*
2013          * We don't want the locks code to timeout the lease for us;
2014          * we'll remove it ourself if the delegation isn't returned
2015          * in time.
2016          */
2017         fl->fl_break_time = 0;
2018
2019         dp->dl_file->fi_had_conflict = true;
2020         nfsd4_cb_recall(dp);
2021 }
2022
2023 /*
2024  * The file_lock is being reapd.
2025  *
2026  * Called by locks_free_lock() with lock_kernel() held.
2027  */
2028 static
2029 void nfsd_release_deleg_cb(struct file_lock *fl)
2030 {
2031         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2032
2033         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2034
2035         if (!(fl->fl_flags & FL_LEASE) || !dp)
2036                 return;
2037         dp->dl_flock = NULL;
2038 }
2039
2040 /*
2041  * Set the delegation file_lock back pointer.
2042  *
2043  * Called from setlease() with lock_kernel() held.
2044  */
2045 static
2046 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2047 {
2048         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2049
2050         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2051         if (!dp)
2052                 return;
2053         dp->dl_flock = new;
2054 }
2055
2056 /*
2057  * Called from setlease() with lock_kernel() held
2058  */
2059 static
2060 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2061 {
2062         struct nfs4_delegation *onlistd =
2063                 (struct nfs4_delegation *)onlist->fl_owner;
2064         struct nfs4_delegation *tryd =
2065                 (struct nfs4_delegation *)try->fl_owner;
2066
2067         if (onlist->fl_lmops != try->fl_lmops)
2068                 return 0;
2069
2070         return onlistd->dl_client == tryd->dl_client;
2071 }
2072
2073
2074 static
2075 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2076 {
2077         if (arg & F_UNLCK)
2078                 return lease_modify(onlist, arg);
2079         else
2080                 return -EAGAIN;
2081 }
2082
2083 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2084         .fl_break = nfsd_break_deleg_cb,
2085         .fl_release_private = nfsd_release_deleg_cb,
2086         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2087         .fl_mylease = nfsd_same_client_deleg_cb,
2088         .fl_change = nfsd_change_deleg_cb,
2089 };
2090
2091
2092 __be32
2093 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2094                     struct nfsd4_open *open)
2095 {
2096         clientid_t *clientid = &open->op_clientid;
2097         struct nfs4_client *clp = NULL;
2098         unsigned int strhashval;
2099         struct nfs4_stateowner *sop = NULL;
2100
2101         if (!check_name(open->op_owner))
2102                 return nfserr_inval;
2103
2104         if (STALE_CLIENTID(&open->op_clientid))
2105                 return nfserr_stale_clientid;
2106
2107         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2108         sop = find_openstateowner_str(strhashval, open);
2109         open->op_stateowner = sop;
2110         if (!sop) {
2111                 /* Make sure the client's lease hasn't expired. */
2112                 clp = find_confirmed_client(clientid);
2113                 if (clp == NULL)
2114                         return nfserr_expired;
2115                 goto renew;
2116         }
2117         /* When sessions are used, skip open sequenceid processing */
2118         if (nfsd4_has_session(cstate))
2119                 goto renew;
2120         if (!sop->so_confirmed) {
2121                 /* Replace unconfirmed owners without checking for replay. */
2122                 clp = sop->so_client;
2123                 release_openowner(sop);
2124                 open->op_stateowner = NULL;
2125                 goto renew;
2126         }
2127         if (open->op_seqid == sop->so_seqid - 1) {
2128                 if (sop->so_replay.rp_buflen)
2129                         return nfserr_replay_me;
2130                 /* The original OPEN failed so spectacularly
2131                  * that we don't even have replay data saved!
2132                  * Therefore, we have no choice but to continue
2133                  * processing this OPEN; presumably, we'll
2134                  * fail again for the same reason.
2135                  */
2136                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2137                 goto renew;
2138         }
2139         if (open->op_seqid != sop->so_seqid)
2140                 return nfserr_bad_seqid;
2141 renew:
2142         if (open->op_stateowner == NULL) {
2143                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2144                 if (sop == NULL)
2145                         return nfserr_resource;
2146                 open->op_stateowner = sop;
2147         }
2148         list_del_init(&sop->so_close_lru);
2149         renew_client(sop->so_client);
2150         return nfs_ok;
2151 }
2152
2153 static inline __be32
2154 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2155 {
2156         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2157                 return nfserr_openmode;
2158         else
2159                 return nfs_ok;
2160 }
2161
2162 static struct nfs4_delegation *
2163 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2164 {
2165         struct nfs4_delegation *dp;
2166
2167         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2168                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2169                         return dp;
2170         }
2171         return NULL;
2172 }
2173
2174 static __be32
2175 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2176                 struct nfs4_delegation **dp)
2177 {
2178         int flags;
2179         __be32 status = nfserr_bad_stateid;
2180
2181         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2182         if (*dp == NULL)
2183                 goto out;
2184         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2185                                                 RD_STATE : WR_STATE;
2186         status = nfs4_check_delegmode(*dp, flags);
2187         if (status)
2188                 *dp = NULL;
2189 out:
2190         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2191                 return nfs_ok;
2192         if (status)
2193                 return status;
2194         open->op_stateowner->so_confirmed = 1;
2195         return nfs_ok;
2196 }
2197
2198 static __be32
2199 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2200 {
2201         struct nfs4_stateid *local;
2202         __be32 status = nfserr_share_denied;
2203         struct nfs4_stateowner *sop = open->op_stateowner;
2204
2205         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2206                 /* ignore lock owners */
2207                 if (local->st_stateowner->so_is_open_owner == 0)
2208                         continue;
2209                 /* remember if we have seen this open owner */
2210                 if (local->st_stateowner == sop)
2211                         *stpp = local;
2212                 /* check for conflicting share reservations */
2213                 if (!test_share(local, open))
2214                         goto out;
2215         }
2216         status = 0;
2217 out:
2218         return status;
2219 }
2220
2221 static inline struct nfs4_stateid *
2222 nfs4_alloc_stateid(void)
2223 {
2224         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2225 }
2226
2227 static __be32
2228 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2229                 struct nfs4_delegation *dp,
2230                 struct svc_fh *cur_fh, int flags)
2231 {
2232         struct nfs4_stateid *stp;
2233
2234         stp = nfs4_alloc_stateid();
2235         if (stp == NULL)
2236                 return nfserr_resource;
2237
2238         if (dp) {
2239                 get_file(dp->dl_vfs_file);
2240                 stp->st_vfs_file = dp->dl_vfs_file;
2241         } else {
2242                 __be32 status;
2243                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2244                                 &stp->st_vfs_file);
2245                 if (status) {
2246                         if (status == nfserr_dropit)
2247                                 status = nfserr_jukebox;
2248                         kmem_cache_free(stateid_slab, stp);
2249                         return status;
2250                 }
2251         }
2252         *stpp = stp;
2253         return 0;
2254 }
2255
2256 static inline __be32
2257 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2258                 struct nfsd4_open *open)
2259 {
2260         struct iattr iattr = {
2261                 .ia_valid = ATTR_SIZE,
2262                 .ia_size = 0,
2263         };
2264         if (!open->op_truncate)
2265                 return 0;
2266         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2267                 return nfserr_inval;
2268         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2269 }
2270
2271 static __be32
2272 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2273 {
2274         struct file *filp = stp->st_vfs_file;
2275         struct inode *inode = filp->f_path.dentry->d_inode;
2276         unsigned int share_access, new_writer;
2277         __be32 status;
2278
2279         set_access(&share_access, stp->st_access_bmap);
2280         new_writer = (~share_access) & open->op_share_access
2281                         & NFS4_SHARE_ACCESS_WRITE;
2282
2283         if (new_writer) {
2284                 int err = get_write_access(inode);
2285                 if (err)
2286                         return nfserrno(err);
2287                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2288                 if (err)
2289                         return nfserrno(err);
2290                 file_take_write(filp);
2291         }
2292         status = nfsd4_truncate(rqstp, cur_fh, open);
2293         if (status) {
2294                 if (new_writer)
2295                         put_write_access(inode);
2296                 return status;
2297         }
2298         /* remember the open */
2299         filp->f_mode |= open->op_share_access;
2300         __set_bit(open->op_share_access, &stp->st_access_bmap);
2301         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2302
2303         return nfs_ok;
2304 }
2305
2306
2307 static void
2308 nfs4_set_claim_prev(struct nfsd4_open *open)
2309 {
2310         open->op_stateowner->so_confirmed = 1;
2311         open->op_stateowner->so_client->cl_firststate = 1;
2312 }
2313
2314 /*
2315  * Attempt to hand out a delegation.
2316  */
2317 static void
2318 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2319 {
2320         struct nfs4_delegation *dp;
2321         struct nfs4_stateowner *sop = stp->st_stateowner;
2322         int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2323         struct file_lock fl, *flp = &fl;
2324         int status, flag = 0;
2325
2326         flag = NFS4_OPEN_DELEGATE_NONE;
2327         open->op_recall = 0;
2328         switch (open->op_claim_type) {
2329                 case NFS4_OPEN_CLAIM_PREVIOUS:
2330                         if (!cb_up)
2331                                 open->op_recall = 1;
2332                         flag = open->op_delegate_type;
2333                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2334                                 goto out;
2335                         break;
2336                 case NFS4_OPEN_CLAIM_NULL:
2337                         /* Let's not give out any delegations till everyone's
2338                          * had the chance to reclaim theirs.... */
2339                         if (locks_in_grace())
2340                                 goto out;
2341                         if (!cb_up || !sop->so_confirmed)
2342                                 goto out;
2343                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2344                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2345                         else
2346                                 flag = NFS4_OPEN_DELEGATE_READ;
2347                         break;
2348                 default:
2349                         goto out;
2350         }
2351
2352         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2353         if (dp == NULL) {
2354                 flag = NFS4_OPEN_DELEGATE_NONE;
2355                 goto out;
2356         }
2357         locks_init_lock(&fl);
2358         fl.fl_lmops = &nfsd_lease_mng_ops;
2359         fl.fl_flags = FL_LEASE;
2360         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2361         fl.fl_end = OFFSET_MAX;
2362         fl.fl_owner =  (fl_owner_t)dp;
2363         fl.fl_file = stp->st_vfs_file;
2364         fl.fl_pid = current->tgid;
2365
2366         /* vfs_setlease checks to see if delegation should be handed out.
2367          * the lock_manager callbacks fl_mylease and fl_change are used
2368          */
2369         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2370                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2371                 unhash_delegation(dp);
2372                 flag = NFS4_OPEN_DELEGATE_NONE;
2373                 goto out;
2374         }
2375
2376         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2377
2378         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2379                 STATEID_VAL(&dp->dl_stateid));
2380 out:
2381         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2382                         && flag == NFS4_OPEN_DELEGATE_NONE
2383                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2384                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2385         open->op_delegate_type = flag;
2386 }
2387
2388 /*
2389  * called with nfs4_lock_state() held.
2390  */
2391 __be32
2392 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2393 {
2394         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2395         struct nfs4_file *fp = NULL;
2396         struct inode *ino = current_fh->fh_dentry->d_inode;
2397         struct nfs4_stateid *stp = NULL;
2398         struct nfs4_delegation *dp = NULL;
2399         __be32 status;
2400
2401         status = nfserr_inval;
2402         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2403                         || !deny_valid(open->op_share_deny))
2404                 goto out;
2405         /*
2406          * Lookup file; if found, lookup stateid and check open request,
2407          * and check for delegations in the process of being recalled.
2408          * If not found, create the nfs4_file struct
2409          */
2410         fp = find_file(ino);
2411         if (fp) {
2412                 if ((status = nfs4_check_open(fp, open, &stp)))
2413                         goto out;
2414                 status = nfs4_check_deleg(fp, open, &dp);
2415                 if (status)
2416                         goto out;
2417         } else {
2418                 status = nfserr_bad_stateid;
2419                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2420                         goto out;
2421                 status = nfserr_resource;
2422                 fp = alloc_init_file(ino);
2423                 if (fp == NULL)
2424                         goto out;
2425         }
2426
2427         /*
2428          * OPEN the file, or upgrade an existing OPEN.
2429          * If truncate fails, the OPEN fails.
2430          */
2431         if (stp) {
2432                 /* Stateid was found, this is an OPEN upgrade */
2433                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2434                 if (status)
2435                         goto out;
2436                 update_stateid(&stp->st_stateid);
2437         } else {
2438                 /* Stateid was not found, this is a new OPEN */
2439                 int flags = 0;
2440                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2441                         flags |= NFSD_MAY_READ;
2442                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2443                         flags |= NFSD_MAY_WRITE;
2444                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2445                 if (status)
2446                         goto out;
2447                 init_stateid(stp, fp, open);
2448                 status = nfsd4_truncate(rqstp, current_fh, open);
2449                 if (status) {
2450                         release_open_stateid(stp);
2451                         goto out;
2452                 }
2453                 if (nfsd4_has_session(&resp->cstate))
2454                         update_stateid(&stp->st_stateid);
2455         }
2456         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2457
2458         if (nfsd4_has_session(&resp->cstate)) {
2459                 open->op_stateowner->so_confirmed = 1;
2460                 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2461         }
2462
2463         /*
2464         * Attempt to hand out a delegation. No error return, because the
2465         * OPEN succeeds even if we fail.
2466         */
2467         nfs4_open_delegation(current_fh, open, stp);
2468
2469         status = nfs_ok;
2470
2471         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2472                 STATEID_VAL(&stp->st_stateid));
2473 out:
2474         if (fp)
2475                 put_nfs4_file(fp);
2476         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2477                 nfs4_set_claim_prev(open);
2478         /*
2479         * To finish the open response, we just need to set the rflags.
2480         */
2481         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2482         if (!open->op_stateowner->so_confirmed &&
2483             !nfsd4_has_session(&resp->cstate))
2484                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2485
2486         return status;
2487 }
2488
2489 __be32
2490 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2491             clientid_t *clid)
2492 {
2493         struct nfs4_client *clp;
2494         __be32 status;
2495
2496         nfs4_lock_state();
2497         dprintk("process_renew(%08x/%08x): starting\n", 
2498                         clid->cl_boot, clid->cl_id);
2499         status = nfserr_stale_clientid;
2500         if (STALE_CLIENTID(clid))
2501                 goto out;
2502         clp = find_confirmed_client(clid);
2503         status = nfserr_expired;
2504         if (clp == NULL) {
2505                 /* We assume the client took too long to RENEW. */
2506                 dprintk("nfsd4_renew: clientid not found!\n");
2507                 goto out;
2508         }
2509         renew_client(clp);
2510         status = nfserr_cb_path_down;
2511         if (!list_empty(&clp->cl_delegations)
2512                         && !atomic_read(&clp->cl_cb_set))
2513                 goto out;
2514         status = nfs_ok;
2515 out:
2516         nfs4_unlock_state();
2517         return status;
2518 }
2519
2520 struct lock_manager nfsd4_manager = {
2521 };
2522
2523 static void
2524 nfsd4_end_grace(void)
2525 {
2526         dprintk("NFSD: end of grace period\n");
2527         nfsd4_recdir_purge_old();
2528         locks_end_grace(&nfsd4_manager);
2529         /*
2530          * Now that every NFSv4 client has had the chance to recover and
2531          * to see the (possibly new, possibly shorter) lease time, we
2532          * can safely set the next grace time to the current lease time:
2533          */
2534         nfsd4_grace = nfsd4_lease;
2535 }
2536
2537 static time_t
2538 nfs4_laundromat(void)
2539 {
2540         struct nfs4_client *clp;
2541         struct nfs4_stateowner *sop;
2542         struct nfs4_delegation *dp;
2543         struct list_head *pos, *next, reaplist;
2544         time_t cutoff = get_seconds() - nfsd4_lease;
2545         time_t t, clientid_val = nfsd4_lease;
2546         time_t u, test_val = nfsd4_lease;
2547
2548         nfs4_lock_state();
2549
2550         dprintk("NFSD: laundromat service - starting\n");
2551         if (locks_in_grace())
2552                 nfsd4_end_grace();
2553         list_for_each_safe(pos, next, &client_lru) {
2554                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2555                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2556                         t = clp->cl_time - cutoff;
2557                         if (clientid_val > t)
2558                                 clientid_val = t;
2559                         break;
2560                 }
2561                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2562                         clp->cl_clientid.cl_id);
2563                 nfsd4_remove_clid_dir(clp);
2564                 expire_client(clp);
2565         }
2566         INIT_LIST_HEAD(&reaplist);
2567         spin_lock(&recall_lock);
2568         list_for_each_safe(pos, next, &del_recall_lru) {
2569                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2570                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2571                         u = dp->dl_time - cutoff;
2572                         if (test_val > u)
2573                                 test_val = u;
2574                         break;
2575                 }
2576                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2577                                     dp, dp->dl_flock);
2578                 list_move(&dp->dl_recall_lru, &reaplist);
2579         }
2580         spin_unlock(&recall_lock);
2581         list_for_each_safe(pos, next, &reaplist) {
2582                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2583                 list_del_init(&dp->dl_recall_lru);
2584                 unhash_delegation(dp);
2585         }
2586         test_val = nfsd4_lease;
2587         list_for_each_safe(pos, next, &close_lru) {
2588                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2589                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2590                         u = sop->so_time - cutoff;
2591                         if (test_val > u)
2592                                 test_val = u;
2593                         break;
2594                 }
2595                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2596                         sop->so_id);
2597                 release_openowner(sop);
2598         }
2599         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2600                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2601         nfs4_unlock_state();
2602         return clientid_val;
2603 }
2604
2605 static struct workqueue_struct *laundry_wq;
2606 static void laundromat_main(struct work_struct *);
2607 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2608
2609 static void
2610 laundromat_main(struct work_struct *not_used)
2611 {
2612         time_t t;
2613
2614         t = nfs4_laundromat();
2615         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2616         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2617 }
2618
2619 static struct nfs4_stateowner *
2620 search_close_lru(u32 st_id, int flags)
2621 {
2622         struct nfs4_stateowner *local = NULL;
2623
2624         if (flags & CLOSE_STATE) {
2625                 list_for_each_entry(local, &close_lru, so_close_lru) {
2626                         if (local->so_id == st_id)
2627                                 return local;
2628                 }
2629         }
2630         return NULL;
2631 }
2632
2633 static inline int
2634 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2635 {
2636         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2637 }
2638
2639 static int
2640 STALE_STATEID(stateid_t *stateid)
2641 {
2642         if (time_after((unsigned long)boot_time,
2643                         (unsigned long)stateid->si_boot)) {
2644                 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2645                         STATEID_VAL(stateid));
2646                 return 1;
2647         }
2648         return 0;
2649 }
2650
2651 static int
2652 EXPIRED_STATEID(stateid_t *stateid)
2653 {
2654         if (time_before((unsigned long)boot_time,
2655                         ((unsigned long)stateid->si_boot)) &&
2656             time_before((unsigned long)(stateid->si_boot + nfsd4_lease), get_seconds())) {
2657                 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2658                         STATEID_VAL(stateid));
2659                 return 1;
2660         }
2661         return 0;
2662 }
2663
2664 static __be32
2665 stateid_error_map(stateid_t *stateid)
2666 {
2667         if (STALE_STATEID(stateid))
2668                 return nfserr_stale_stateid;
2669         if (EXPIRED_STATEID(stateid))
2670                 return nfserr_expired;
2671
2672         dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2673                 STATEID_VAL(stateid));
2674         return nfserr_bad_stateid;
2675 }
2676
2677 static inline int
2678 access_permit_read(unsigned long access_bmap)
2679 {
2680         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2681                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2682                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2683 }
2684
2685 static inline int
2686 access_permit_write(unsigned long access_bmap)
2687 {
2688         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2689                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2690 }
2691
2692 static
2693 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2694 {
2695         __be32 status = nfserr_openmode;
2696
2697         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2698                 goto out;
2699         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2700                 goto out;
2701         status = nfs_ok;
2702 out:
2703         return status;
2704 }
2705
2706 static inline __be32
2707 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2708 {
2709         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2710                 return nfs_ok;
2711         else if (locks_in_grace()) {
2712                 /* Answer in remaining cases depends on existance of
2713                  * conflicting state; so we must wait out the grace period. */
2714                 return nfserr_grace;
2715         } else if (flags & WR_STATE)
2716                 return nfs4_share_conflict(current_fh,
2717                                 NFS4_SHARE_DENY_WRITE);
2718         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2719                 return nfs4_share_conflict(current_fh,
2720                                 NFS4_SHARE_DENY_READ);
2721 }
2722
2723 /*
2724  * Allow READ/WRITE during grace period on recovered state only for files
2725  * that are not able to provide mandatory locking.
2726  */
2727 static inline int
2728 grace_disallows_io(struct inode *inode)
2729 {
2730         return locks_in_grace() && mandatory_lock(inode);
2731 }
2732
2733 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2734 {
2735         /*
2736          * When sessions are used the stateid generation number is ignored
2737          * when it is zero.
2738          */
2739         if ((flags & HAS_SESSION) && in->si_generation == 0)
2740                 goto out;
2741
2742         /* If the client sends us a stateid from the future, it's buggy: */
2743         if (in->si_generation > ref->si_generation)
2744                 return nfserr_bad_stateid;
2745         /*
2746          * The following, however, can happen.  For example, if the
2747          * client sends an open and some IO at the same time, the open
2748          * may bump si_generation while the IO is still in flight.
2749          * Thanks to hard links and renames, the client never knows what
2750          * file an open will affect.  So it could avoid that situation
2751          * only by serializing all opens and IO from the same open
2752          * owner.  To recover from the old_stateid error, the client
2753          * will just have to retry the IO:
2754          */
2755         if (in->si_generation < ref->si_generation)
2756                 return nfserr_old_stateid;
2757 out:
2758         return nfs_ok;
2759 }
2760
2761 static int is_delegation_stateid(stateid_t *stateid)
2762 {
2763         return stateid->si_fileid == 0;
2764 }
2765
2766 /*
2767 * Checks for stateid operations
2768 */
2769 __be32
2770 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2771                            stateid_t *stateid, int flags, struct file **filpp)
2772 {
2773         struct nfs4_stateid *stp = NULL;
2774         struct nfs4_delegation *dp = NULL;
2775         struct svc_fh *current_fh = &cstate->current_fh;
2776         struct inode *ino = current_fh->fh_dentry->d_inode;
2777         __be32 status;
2778
2779         if (filpp)
2780                 *filpp = NULL;
2781
2782         if (grace_disallows_io(ino))
2783                 return nfserr_grace;
2784
2785         if (nfsd4_has_session(cstate))
2786                 flags |= HAS_SESSION;
2787
2788         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2789                 return check_special_stateids(current_fh, stateid, flags);
2790
2791         status = nfserr_stale_stateid;
2792         if (STALE_STATEID(stateid)) 
2793                 goto out;
2794
2795         status = nfserr_bad_stateid;
2796         if (is_delegation_stateid(stateid)) {
2797                 dp = find_delegation_stateid(ino, stateid);
2798                 if (!dp) {
2799                         status = stateid_error_map(stateid);
2800                         goto out;
2801                 }
2802                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2803                                                   flags);
2804                 if (status)
2805                         goto out;
2806                 status = nfs4_check_delegmode(dp, flags);
2807                 if (status)
2808                         goto out;
2809                 renew_client(dp->dl_client);
2810                 if (filpp)
2811                         *filpp = dp->dl_vfs_file;
2812         } else { /* open or lock stateid */
2813                 stp = find_stateid(stateid, flags);
2814                 if (!stp) {
2815                         status = stateid_error_map(stateid);
2816                         goto out;
2817                 }
2818                 if (nfs4_check_fh(current_fh, stp))
2819                         goto out;
2820                 if (!stp->st_stateowner->so_confirmed)
2821                         goto out;
2822                 status = check_stateid_generation(stateid, &stp->st_stateid,
2823                                                   flags);
2824                 if (status)
2825                         goto out;
2826                 status = nfs4_check_openmode(stp, flags);
2827                 if (status)
2828                         goto out;
2829                 renew_client(stp->st_stateowner->so_client);
2830                 if (filpp)
2831                         *filpp = stp->st_vfs_file;
2832         }
2833         status = nfs_ok;
2834 out:
2835         return status;
2836 }
2837
2838 static inline int
2839 setlkflg (int type)
2840 {
2841         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2842                 RD_STATE : WR_STATE;
2843 }
2844
2845 /* 
2846  * Checks for sequence id mutating operations. 
2847  */
2848 static __be32
2849 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2850                          stateid_t *stateid, int flags,
2851                          struct nfs4_stateowner **sopp,
2852                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2853 {
2854         struct nfs4_stateid *stp;
2855         struct nfs4_stateowner *sop;
2856         struct svc_fh *current_fh = &cstate->current_fh;
2857         __be32 status;
2858
2859         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2860                 seqid, STATEID_VAL(stateid));
2861
2862         *stpp = NULL;
2863         *sopp = NULL;
2864
2865         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2866                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2867                 return nfserr_bad_stateid;
2868         }
2869
2870         if (STALE_STATEID(stateid))
2871                 return nfserr_stale_stateid;
2872
2873         if (nfsd4_has_session(cstate))
2874                 flags |= HAS_SESSION;
2875
2876         /*
2877         * We return BAD_STATEID if filehandle doesn't match stateid, 
2878         * the confirmed flag is incorrecly set, or the generation 
2879         * number is incorrect.  
2880         */
2881         stp = find_stateid(stateid, flags);
2882         if (stp == NULL) {
2883                 /*
2884                  * Also, we should make sure this isn't just the result of
2885                  * a replayed close:
2886                  */
2887                 sop = search_close_lru(stateid->si_stateownerid, flags);
2888                 if (sop == NULL)
2889                         return stateid_error_map(stateid);
2890                 *sopp = sop;
2891                 goto check_replay;
2892         }
2893
2894         *stpp = stp;
2895         *sopp = sop = stp->st_stateowner;
2896
2897         if (lock) {
2898                 clientid_t *lockclid = &lock->v.new.clientid;
2899                 struct nfs4_client *clp = sop->so_client;
2900                 int lkflg = 0;
2901                 __be32 status;
2902
2903                 lkflg = setlkflg(lock->lk_type);
2904
2905                 if (lock->lk_is_new) {
2906                         if (!sop->so_is_open_owner)
2907                                 return nfserr_bad_stateid;
2908                         if (!(flags & HAS_SESSION) &&
2909                             !same_clid(&clp->cl_clientid, lockclid))
2910                                 return nfserr_bad_stateid;
2911                         /* stp is the open stateid */
2912                         status = nfs4_check_openmode(stp, lkflg);
2913                         if (status)
2914                                 return status;
2915                 } else {
2916                         /* stp is the lock stateid */
2917                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2918                         if (status)
2919                                 return status;
2920                }
2921         }
2922
2923         if (nfs4_check_fh(current_fh, stp)) {
2924                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2925                 return nfserr_bad_stateid;
2926         }
2927
2928         /*
2929         *  We now validate the seqid and stateid generation numbers.
2930         *  For the moment, we ignore the possibility of 
2931         *  generation number wraparound.
2932         */
2933         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2934                 goto check_replay;
2935
2936         if (sop->so_confirmed && flags & CONFIRM) {
2937                 dprintk("NFSD: preprocess_seqid_op: expected"
2938                                 " unconfirmed stateowner!\n");
2939                 return nfserr_bad_stateid;
2940         }
2941         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2942                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2943                                 " confirmed yet!\n");
2944                 return nfserr_bad_stateid;
2945         }
2946         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2947         if (status)
2948                 return status;
2949         renew_client(sop->so_client);
2950         return nfs_ok;
2951
2952 check_replay:
2953         if (seqid == sop->so_seqid - 1) {
2954                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2955                 /* indicate replay to calling function */
2956                 return nfserr_replay_me;
2957         }
2958         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2959                         sop->so_seqid, seqid);
2960         *sopp = NULL;
2961         return nfserr_bad_seqid;
2962 }
2963
2964 __be32
2965 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2966                    struct nfsd4_open_confirm *oc)
2967 {
2968         __be32 status;
2969         struct nfs4_stateowner *sop;
2970         struct nfs4_stateid *stp;
2971
2972         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2973                         (int)cstate->current_fh.fh_dentry->d_name.len,
2974                         cstate->current_fh.fh_dentry->d_name.name);
2975
2976         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2977         if (status)
2978                 return status;
2979
2980         nfs4_lock_state();
2981
2982         if ((status = nfs4_preprocess_seqid_op(cstate,
2983                                         oc->oc_seqid, &oc->oc_req_stateid,
2984                                         CONFIRM | OPEN_STATE,
2985                                         &oc->oc_stateowner, &stp, NULL)))
2986                 goto out; 
2987
2988         sop = oc->oc_stateowner;
2989         sop->so_confirmed = 1;
2990         update_stateid(&stp->st_stateid);
2991         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
2992         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
2993                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
2994
2995         nfsd4_create_clid_dir(sop->so_client);
2996 out:
2997         if (oc->oc_stateowner) {
2998                 nfs4_get_stateowner(oc->oc_stateowner);
2999                 cstate->replay_owner = oc->oc_stateowner;
3000         }
3001         nfs4_unlock_state();
3002         return status;
3003 }
3004
3005
3006 /*
3007  * unset all bits in union bitmap (bmap) that
3008  * do not exist in share (from successful OPEN_DOWNGRADE)
3009  */
3010 static void
3011 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3012 {
3013         int i;
3014         for (i = 1; i < 4; i++) {
3015                 if ((i & access) != i)
3016                         __clear_bit(i, bmap);
3017         }
3018 }
3019
3020 static void
3021 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3022 {
3023         int i;
3024         for (i = 0; i < 4; i++) {
3025                 if ((i & deny) != i)
3026                         __clear_bit(i, bmap);
3027         }
3028 }
3029
3030 __be32
3031 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3032                      struct nfsd4_compound_state *cstate,
3033                      struct nfsd4_open_downgrade *od)
3034 {
3035         __be32 status;
3036         struct nfs4_stateid *stp;
3037         unsigned int share_access;
3038
3039         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3040                         (int)cstate->current_fh.fh_dentry->d_name.len,
3041                         cstate->current_fh.fh_dentry->d_name.name);
3042
3043         if (!access_valid(od->od_share_access, cstate->minorversion)
3044                         || !deny_valid(od->od_share_deny))
3045                 return nfserr_inval;
3046
3047         nfs4_lock_state();
3048         if ((status = nfs4_preprocess_seqid_op(cstate,
3049                                         od->od_seqid,
3050                                         &od->od_stateid, 
3051                                         OPEN_STATE,
3052                                         &od->od_stateowner, &stp, NULL)))
3053                 goto out; 
3054
3055         status = nfserr_inval;
3056         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3057                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3058                         stp->st_access_bmap, od->od_share_access);
3059                 goto out;
3060         }
3061         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3062                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3063                         stp->st_deny_bmap, od->od_share_deny);
3064                 goto out;
3065         }
3066         set_access(&share_access, stp->st_access_bmap);
3067         nfs4_file_downgrade(stp->st_vfs_file,
3068                             share_access & ~od->od_share_access);
3069
3070         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3071         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3072
3073         update_stateid(&stp->st_stateid);
3074         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3075         status = nfs_ok;
3076 out:
3077         if (od->od_stateowner) {
3078                 nfs4_get_stateowner(od->od_stateowner);
3079                 cstate->replay_owner = od->od_stateowner;
3080         }
3081         nfs4_unlock_state();
3082         return status;
3083 }
3084
3085 /*
3086  * nfs4_unlock_state() called after encode
3087  */
3088 __be32
3089 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3090             struct nfsd4_close *close)
3091 {
3092         __be32 status;
3093         struct nfs4_stateid *stp;
3094
3095         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3096                         (int)cstate->current_fh.fh_dentry->d_name.len,
3097                         cstate->current_fh.fh_dentry->d_name.name);
3098
3099         nfs4_lock_state();
3100         /* check close_lru for replay */
3101         if ((status = nfs4_preprocess_seqid_op(cstate,
3102                                         close->cl_seqid,
3103                                         &close->cl_stateid, 
3104                                         OPEN_STATE | CLOSE_STATE,
3105                                         &close->cl_stateowner, &stp, NULL)))
3106                 goto out; 
3107         status = nfs_ok;
3108         update_stateid(&stp->st_stateid);
3109         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3110
3111         /* release_stateid() calls nfsd_close() if needed */
3112         release_open_stateid(stp);
3113
3114         /* place unused nfs4_stateowners on so_close_lru list to be
3115          * released by the laundromat service after the lease period
3116          * to enable us to handle CLOSE replay
3117          */
3118         if (list_empty(&close->cl_stateowner->so_stateids))
3119                 move_to_close_lru(close->cl_stateowner);
3120 out:
3121         if (close->cl_stateowner) {
3122                 nfs4_get_stateowner(close->cl_stateowner);
3123                 cstate->replay_owner = close->cl_stateowner;
3124         }
3125         nfs4_unlock_state();
3126         return status;
3127 }
3128
3129 __be32
3130 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3131                   struct nfsd4_delegreturn *dr)
3132 {
3133         struct nfs4_delegation *dp;
3134         stateid_t *stateid = &dr->dr_stateid;
3135         struct inode *inode;
3136         __be32 status;
3137         int flags = 0;
3138
3139         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3140                 return status;
3141         inode = cstate->current_fh.fh_dentry->d_inode;
3142
3143         if (nfsd4_has_session(cstate))
3144                 flags |= HAS_SESSION;
3145         nfs4_lock_state();
3146         status = nfserr_bad_stateid;
3147         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3148                 goto out;
3149         status = nfserr_stale_stateid;
3150         if (STALE_STATEID(stateid))
3151                 goto out;
3152         status = nfserr_bad_stateid;
3153         if (!is_delegation_stateid(stateid))
3154                 goto out;
3155         dp = find_delegation_stateid(inode, stateid);
3156         if (!dp) {
3157                 status = stateid_error_map(stateid);
3158                 goto out;
3159         }
3160         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3161         if (status)
3162                 goto out;
3163         renew_client(dp->dl_client);
3164
3165         unhash_delegation(dp);
3166 out:
3167         nfs4_unlock_state();
3168
3169         return status;
3170 }
3171
3172
3173 /* 
3174  * Lock owner state (byte-range locks)
3175  */
3176 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3177 #define LOCK_HASH_BITS              8
3178 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3179 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3180
3181 static inline u64
3182 end_offset(u64 start, u64 len)
3183 {
3184         u64 end;
3185
3186         end = start + len;
3187         return end >= start ? end: NFS4_MAX_UINT64;
3188 }
3189
3190 /* last octet in a range */
3191 static inline u64
3192 last_byte_offset(u64 start, u64 len)
3193 {
3194         u64 end;
3195
3196         BUG_ON(!len);
3197         end = start + len;
3198         return end > start ? end - 1: NFS4_MAX_UINT64;
3199 }
3200
3201 #define lockownerid_hashval(id) \
3202         ((id) & LOCK_HASH_MASK)
3203
3204 static inline unsigned int
3205 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3206                 struct xdr_netobj *ownername)
3207 {
3208         return (file_hashval(inode) + cl_id
3209                         + opaque_hashval(ownername->data, ownername->len))
3210                 & LOCK_HASH_MASK;
3211 }
3212
3213 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3214 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3215 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3216
3217 static struct nfs4_stateid *
3218 find_stateid(stateid_t *stid, int flags)
3219 {
3220         struct nfs4_stateid *local;
3221         u32 st_id = stid->si_stateownerid;
3222         u32 f_id = stid->si_fileid;
3223         unsigned int hashval;
3224
3225         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3226         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3227                 hashval = stateid_hashval(st_id, f_id);
3228                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3229                         if ((local->st_stateid.si_stateownerid == st_id) &&
3230                             (local->st_stateid.si_fileid == f_id))
3231                                 return local;
3232                 }
3233         } 
3234
3235         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3236                 hashval = stateid_hashval(st_id, f_id);
3237                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3238                         if ((local->st_stateid.si_stateownerid == st_id) &&
3239                             (local->st_stateid.si_fileid == f_id))
3240                                 return local;
3241                 }
3242         }
3243         return NULL;
3244 }
3245
3246 static struct nfs4_delegation *
3247 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3248 {
3249         struct nfs4_file *fp;
3250         struct nfs4_delegation *dl;
3251
3252         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3253                 STATEID_VAL(stid));
3254
3255         fp = find_file(ino);
3256         if (!fp)
3257                 return NULL;
3258         dl = find_delegation_file(fp, stid);
3259         put_nfs4_file(fp);
3260         return dl;
3261 }
3262
3263 /*
3264  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3265  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3266  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3267  * locking, this prevents us from being completely protocol-compliant.  The
3268  * real solution to this problem is to start using unsigned file offsets in
3269  * the VFS, but this is a very deep change!
3270  */
3271 static inline void
3272 nfs4_transform_lock_offset(struct file_lock *lock)
3273 {
3274         if (lock->fl_start < 0)
3275                 lock->fl_start = OFFSET_MAX;
3276         if (lock->fl_end < 0)
3277                 lock->fl_end = OFFSET_MAX;
3278 }
3279
3280 /* Hack!: For now, we're defining this just so we can use a pointer to it
3281  * as a unique cookie to identify our (NFSv4's) posix locks. */
3282 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3283 };
3284
3285 static inline void
3286 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3287 {
3288         struct nfs4_stateowner *sop;
3289         unsigned int hval;
3290
3291         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3292                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3293                 hval = lockownerid_hashval(sop->so_id);
3294                 kref_get(&sop->so_ref);
3295                 deny->ld_sop = sop;
3296                 deny->ld_clientid = sop->so_client->cl_clientid;
3297         } else {
3298                 deny->ld_sop = NULL;
3299                 deny->ld_clientid.cl_boot = 0;
3300                 deny->ld_clientid.cl_id = 0;
3301         }
3302         deny->ld_start = fl->fl_start;
3303         deny->ld_length = NFS4_MAX_UINT64;
3304         if (fl->fl_end != NFS4_MAX_UINT64)
3305                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3306         deny->ld_type = NFS4_READ_LT;
3307         if (fl->fl_type != F_RDLCK)
3308                 deny->ld_type = NFS4_WRITE_LT;
3309 }
3310
3311 static struct nfs4_stateowner *
3312 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3313                 struct xdr_netobj *owner)
3314 {
3315         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3316         struct nfs4_stateowner *op;
3317
3318         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3319                 if (same_owner_str(op, owner, clid))
3320                         return op;
3321         }
3322         return NULL;
3323 }
3324
3325 /*
3326  * Alloc a lock owner structure.
3327  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3328  * occured. 
3329  *
3330  * strhashval = lock_ownerstr_hashval 
3331  */
3332
3333 static struct nfs4_stateowner *
3334 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3335         struct nfs4_stateowner *sop;
3336         struct nfs4_replay *rp;
3337         unsigned int idhashval;
3338
3339         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3340                 return NULL;
3341         idhashval = lockownerid_hashval(current_ownerid);
3342         INIT_LIST_HEAD(&sop->so_idhash);
3343         INIT_LIST_HEAD(&sop->so_strhash);
3344         INIT_LIST_HEAD(&sop->so_perclient);
3345         INIT_LIST_HEAD(&sop->so_stateids);
3346         INIT_LIST_HEAD(&sop->so_perstateid);
3347         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3348         sop->so_time = 0;
3349         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3350         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3351         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3352         sop->so_is_open_owner = 0;
3353         sop->so_id = current_ownerid++;
3354         sop->so_client = clp;
3355         /* It is the openowner seqid that will be incremented in encode in the
3356          * case of new lockowners; so increment the lock seqid manually: */
3357         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3358         sop->so_confirmed = 1;
3359         rp = &sop->so_replay;
3360         rp->rp_status = nfserr_serverfault;
3361         rp->rp_buflen = 0;
3362         rp->rp_buf = rp->rp_ibuf;
3363         return sop;
3364 }
3365
3366 static struct nfs4_stateid *
3367 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3368 {
3369         struct nfs4_stateid *stp;
3370         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3371
3372         stp = nfs4_alloc_stateid();
3373         if (stp == NULL)
3374                 goto out;
3375         INIT_LIST_HEAD(&stp->st_hash);
3376         INIT_LIST_HEAD(&stp->st_perfile);
3377         INIT_LIST_HEAD(&stp->st_perstateowner);
3378         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3379         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3380         list_add(&stp->st_perfile, &fp->fi_stateids);
3381         list_add(&stp->st_perstateowner, &sop->so_stateids);
3382         stp->st_stateowner = sop;
3383         get_nfs4_file(fp);
3384         stp->st_file = fp;
3385         stp->st_stateid.si_boot = get_seconds();
3386         stp->st_stateid.si_stateownerid = sop->so_id;
3387         stp->st_stateid.si_fileid = fp->fi_id;
3388         stp->st_stateid.si_generation = 0;
3389         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3390         stp->st_access_bmap = open_stp->st_access_bmap;
3391         stp->st_deny_bmap = open_stp->st_deny_bmap;
3392         stp->st_openstp = open_stp;
3393
3394 out:
3395         return stp;
3396 }
3397
3398 static int
3399 check_lock_length(u64 offset, u64 length)
3400 {
3401         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3402              LOFF_OVERFLOW(offset, length)));
3403 }
3404
3405 /*
3406  *  LOCK operation 
3407  */
3408 __be32
3409 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3410            struct nfsd4_lock *lock)
3411 {
3412         struct nfs4_stateowner *open_sop = NULL;
3413         struct nfs4_stateowner *lock_sop = NULL;
3414         struct nfs4_stateid *lock_stp;
3415         struct file *filp;
3416         struct file_lock file_lock;
3417         struct file_lock conflock;
3418         __be32 status = 0;
3419         unsigned int strhashval;
3420         unsigned int cmd;
3421         int err;
3422
3423         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3424                 (long long) lock->lk_offset,
3425                 (long long) lock->lk_length);
3426
3427         if (check_lock_length(lock->lk_offset, lock->lk_length))
3428                  return nfserr_inval;
3429
3430         if ((status = fh_verify(rqstp, &cstate->current_fh,
3431                                 S_IFREG, NFSD_MAY_LOCK))) {
3432                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3433                 return status;
3434         }
3435
3436         nfs4_lock_state();
3437
3438         if (lock->lk_is_new) {
3439                 /*
3440                  * Client indicates that this is a new lockowner.
3441                  * Use open owner and open stateid to create lock owner and
3442                  * lock stateid.
3443                  */
3444                 struct nfs4_stateid *open_stp = NULL;
3445                 struct nfs4_file *fp;
3446                 
3447                 status = nfserr_stale_clientid;
3448                 if (!nfsd4_has_session(cstate) &&
3449                     STALE_CLIENTID(&lock->lk_new_clientid))
3450                         goto out;
3451
3452                 /* validate and update open stateid and open seqid */
3453                 status = nfs4_preprocess_seqid_op(cstate,
3454                                         lock->lk_new_open_seqid,
3455                                         &lock->lk_new_open_stateid,
3456                                         OPEN_STATE,
3457                                         &lock->lk_replay_owner, &open_stp,
3458                                         lock);
3459                 if (status)
3460                         goto out;
3461                 open_sop = lock->lk_replay_owner;
3462                 /* create lockowner and lock stateid */
3463                 fp = open_stp->st_file;
3464                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3465                                 open_sop->so_client->cl_clientid.cl_id, 
3466                                 &lock->v.new.owner);
3467                 /* XXX: Do we need to check for duplicate stateowners on
3468                  * the same file, or should they just be allowed (and
3469                  * create new stateids)? */
3470                 status = nfserr_resource;
3471                 lock_sop = alloc_init_lock_stateowner(strhashval,
3472                                 open_sop->so_client, open_stp, lock);
3473                 if (lock_sop == NULL)
3474                         goto out;
3475                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3476                 if (lock_stp == NULL)
3477                         goto out;
3478         } else {
3479                 /* lock (lock owner + lock stateid) already exists */
3480                 status = nfs4_preprocess_seqid_op(cstate,
3481                                        lock->lk_old_lock_seqid, 
3482                                        &lock->lk_old_lock_stateid, 
3483                                        LOCK_STATE,
3484                                        &lock->lk_replay_owner, &lock_stp, lock);
3485                 if (status)
3486                         goto out;
3487                 lock_sop = lock->lk_replay_owner;
3488         }
3489         /* lock->lk_replay_owner and lock_stp have been created or found */
3490         filp = lock_stp->st_vfs_file;
3491
3492         status = nfserr_grace;
3493         if (locks_in_grace() && !lock->lk_reclaim)
3494                 goto out;
3495         status = nfserr_no_grace;
3496         if (!locks_in_grace() && lock->lk_reclaim)
3497                 goto out;
3498
3499         locks_init_lock(&file_lock);
3500         switch (lock->lk_type) {
3501                 case NFS4_READ_LT:
3502                 case NFS4_READW_LT:
3503                         file_lock.fl_type = F_RDLCK;
3504                         cmd = F_SETLK;
3505                 break;
3506                 case NFS4_WRITE_LT:
3507                 case NFS4_WRITEW_LT:
3508                         file_lock.fl_type = F_WRLCK;
3509                         cmd = F_SETLK;
3510                 break;
3511                 default:
3512                         status = nfserr_inval;
3513                 goto out;
3514         }
3515         file_lock.fl_owner = (fl_owner_t)lock_sop;
3516         file_lock.fl_pid = current->tgid;
3517         file_lock.fl_file = filp;
3518         file_lock.fl_flags = FL_POSIX;
3519         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3520
3521         file_lock.fl_start = lock->lk_offset;
3522         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3523         nfs4_transform_lock_offset(&file_lock);
3524
3525         /*
3526         * Try to lock the file in the VFS.
3527         * Note: locks.c uses the BKL to protect the inode's lock list.
3528         */
3529
3530         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3531         switch (-err) {
3532         case 0: /* success! */
3533                 update_stateid(&lock_stp->st_stateid);
3534                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3535                                 sizeof(stateid_t));
3536                 status = 0;
3537                 break;
3538         case (EAGAIN):          /* conflock holds conflicting lock */
3539                 status = nfserr_denied;
3540                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3541                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3542                 break;
3543         case (EDEADLK):
3544                 status = nfserr_deadlock;
3545                 break;
3546         default:        
3547                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3548                 status = nfserr_resource;
3549                 break;
3550         }
3551 out:
3552         if (status && lock->lk_is_new && lock_sop)
3553                 release_lockowner(lock_sop);
3554         if (lock->lk_replay_owner) {
3555                 nfs4_get_stateowner(lock->lk_replay_owner);
3556                 cstate->replay_owner = lock->lk_replay_owner;
3557         }
3558         nfs4_unlock_state();
3559         return status;
3560 }
3561
3562 /*
3563  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3564  * so we do a temporary open here just to get an open file to pass to
3565  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3566  * inode operation.)
3567  */
3568 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3569 {
3570         struct file *file;
3571         int err;
3572
3573         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3574         if (err)
3575                 return err;
3576         err = vfs_test_lock(file, lock);
3577         nfsd_close(file);
3578         return err;
3579 }
3580
3581 /*
3582  * LOCKT operation
3583  */
3584 __be32
3585 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3586             struct nfsd4_lockt *lockt)
3587 {
3588         struct inode *inode;
3589         struct file_lock file_lock;
3590         int error;
3591         __be32 status;
3592
3593         if (locks_in_grace())
3594                 return nfserr_grace;
3595
3596         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3597                  return nfserr_inval;
3598
3599         lockt->lt_stateowner = NULL;
3600         nfs4_lock_state();
3601
3602         status = nfserr_stale_clientid;
3603         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3604                 goto out;
3605
3606         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3607                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3608                 if (status == nfserr_symlink)
3609                         status = nfserr_inval;
3610                 goto out;
3611         }
3612
3613         inode = cstate->current_fh.fh_dentry->d_inode;
3614         locks_init_lock(&file_lock);
3615         switch (lockt->lt_type) {
3616                 case NFS4_READ_LT:
3617                 case NFS4_READW_LT:
3618                         file_lock.fl_type = F_RDLCK;
3619                 break;
3620                 case NFS4_WRITE_LT:
3621                 case NFS4_WRITEW_LT:
3622                         file_lock.fl_type = F_WRLCK;
3623                 break;
3624                 default:
3625                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3626                         status = nfserr_inval;
3627                 goto out;
3628         }
3629
3630         lockt->lt_stateowner = find_lockstateowner_str(inode,
3631                         &lockt->lt_clientid, &lockt->lt_owner);
3632         if (lockt->lt_stateowner)
3633                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3634         file_lock.fl_pid = current->tgid;
3635         file_lock.fl_flags = FL_POSIX;
3636
3637         file_lock.fl_start = lockt->lt_offset;
3638         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3639
3640         nfs4_transform_lock_offset(&file_lock);
3641
3642         status = nfs_ok;
3643         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3644         if (error) {
3645                 status = nfserrno(error);
3646                 goto out;
3647         }
3648         if (file_lock.fl_type != F_UNLCK) {
3649                 status = nfserr_denied;
3650                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3651         }
3652 out:
3653         nfs4_unlock_state();
3654         return status;
3655 }
3656
3657 __be32
3658 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3659             struct nfsd4_locku *locku)
3660 {
3661         struct nfs4_stateid *stp;
3662         struct file *filp = NULL;
3663         struct file_lock file_lock;
3664         __be32 status;
3665         int err;
3666                                                         
3667         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3668                 (long long) locku->lu_offset,
3669                 (long long) locku->lu_length);
3670
3671         if (check_lock_length(locku->lu_offset, locku->lu_length))
3672                  return nfserr_inval;
3673
3674         nfs4_lock_state();
3675                                                                                 
3676         if ((status = nfs4_preprocess_seqid_op(cstate,
3677                                         locku->lu_seqid, 
3678                                         &locku->lu_stateid, 
3679                                         LOCK_STATE,
3680                                         &locku->lu_stateowner, &stp, NULL)))
3681                 goto out;
3682
3683         filp = stp->st_vfs_file;
3684         BUG_ON(!filp);
3685         locks_init_lock(&file_lock);
3686         file_lock.fl_type = F_UNLCK;
3687         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3688         file_lock.fl_pid = current->tgid;
3689         file_lock.fl_file = filp;
3690         file_lock.fl_flags = FL_POSIX; 
3691         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3692         file_lock.fl_start = locku->lu_offset;
3693
3694         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3695         nfs4_transform_lock_offset(&file_lock);
3696
3697         /*
3698         *  Try to unlock the file in the VFS.
3699         */
3700         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3701         if (err) {
3702                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3703                 goto out_nfserr;
3704         }
3705         /*
3706         * OK, unlock succeeded; the only thing left to do is update the stateid.
3707         */
3708         update_stateid(&stp->st_stateid);
3709         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3710
3711 out:
3712         if (locku->lu_stateowner) {
3713                 nfs4_get_stateowner(locku->lu_stateowner);
3714                 cstate->replay_owner = locku->lu_stateowner;
3715         }
3716         nfs4_unlock_state();
3717         return status;
3718
3719 out_nfserr:
3720         status = nfserrno(err);
3721         goto out;
3722 }
3723
3724 /*
3725  * returns
3726  *      1: locks held by lockowner
3727  *      0: no locks held by lockowner
3728  */
3729 static int
3730 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3731 {
3732         struct file_lock **flpp;
3733         struct inode *inode = filp->f_path.dentry->d_inode;
3734         int status = 0;
3735
3736         lock_kernel();
3737         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3738                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3739                         status = 1;
3740                         goto out;
3741                 }
3742         }
3743 out:
3744         unlock_kernel();
3745         return status;
3746 }
3747
3748 __be32
3749 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3750                         struct nfsd4_compound_state *cstate,
3751                         struct nfsd4_release_lockowner *rlockowner)
3752 {
3753         clientid_t *clid = &rlockowner->rl_clientid;
3754         struct nfs4_stateowner *sop;
3755         struct nfs4_stateid *stp;
3756         struct xdr_netobj *owner = &rlockowner->rl_owner;
3757         struct list_head matches;
3758         int i;
3759         __be32 status;
3760
3761         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3762                 clid->cl_boot, clid->cl_id);
3763
3764         /* XXX check for lease expiration */
3765
3766         status = nfserr_stale_clientid;
3767         if (STALE_CLIENTID(clid))
3768                 return status;
3769
3770         nfs4_lock_state();
3771
3772         status = nfserr_locks_held;
3773         /* XXX: we're doing a linear search through all the lockowners.
3774          * Yipes!  For now we'll just hope clients aren't really using
3775          * release_lockowner much, but eventually we have to fix these
3776          * data structures. */
3777         INIT_LIST_HEAD(&matches);
3778         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3779                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3780                         if (!same_owner_str(sop, owner, clid))
3781                                 continue;
3782                         list_for_each_entry(stp, &sop->so_stateids,
3783                                         st_perstateowner) {
3784                                 if (check_for_locks(stp->st_vfs_file, sop))
3785                                         goto out;
3786                                 /* Note: so_perclient unused for lockowners,
3787                                  * so it's OK to fool with here. */
3788                                 list_add(&sop->so_perclient, &matches);
3789                         }
3790                 }
3791         }
3792         /* Clients probably won't expect us to return with some (but not all)
3793          * of the lockowner state released; so don't release any until all
3794          * have been checked. */
3795         status = nfs_ok;
3796         while (!list_empty(&matches)) {
3797                 sop = list_entry(matches.next, struct nfs4_stateowner,
3798                                                                 so_perclient);
3799                 /* unhash_stateowner deletes so_perclient only
3800                  * for openowners. */
3801                 list_del(&sop->so_perclient);
3802                 release_lockowner(sop);
3803         }
3804 out:
3805         nfs4_unlock_state();
3806         return status;
3807 }
3808
3809 static inline struct nfs4_client_reclaim *
3810 alloc_reclaim(void)
3811 {
3812         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3813 }
3814
3815 int
3816 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3817 {
3818         unsigned int strhashval = clientstr_hashval(name);
3819         struct nfs4_client *clp;
3820
3821         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3822         return clp ? 1 : 0;
3823 }
3824
3825 /*
3826  * failure => all reset bets are off, nfserr_no_grace...
3827  */
3828 int
3829 nfs4_client_to_reclaim(const char *name)
3830 {
3831         unsigned int strhashval;
3832         struct nfs4_client_reclaim *crp = NULL;
3833
3834         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3835         crp = alloc_reclaim();
3836         if (!crp)
3837                 return 0;
3838         strhashval = clientstr_hashval(name);
3839         INIT_LIST_HEAD(&crp->cr_strhash);
3840         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3841         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3842         reclaim_str_hashtbl_size++;
3843         return 1;
3844 }
3845
3846 static void
3847 nfs4_release_reclaim(void)
3848 {
3849         struct nfs4_client_reclaim *crp = NULL;
3850         int i;
3851
3852         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3853                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3854                         crp = list_entry(reclaim_str_hashtbl[i].next,
3855                                         struct nfs4_client_reclaim, cr_strhash);
3856                         list_del(&crp->cr_strhash);
3857                         kfree(crp);
3858                         reclaim_str_hashtbl_size--;
3859                 }
3860         }
3861         BUG_ON(reclaim_str_hashtbl_size);
3862 }
3863
3864 /*
3865  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3866 static struct nfs4_client_reclaim *
3867 nfs4_find_reclaim_client(clientid_t *clid)
3868 {
3869         unsigned int strhashval;
3870         struct nfs4_client *clp;
3871         struct nfs4_client_reclaim *crp = NULL;
3872
3873
3874         /* find clientid in conf_id_hashtbl */
3875         clp = find_confirmed_client(clid);
3876         if (clp == NULL)
3877                 return NULL;
3878
3879         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3880                             clp->cl_name.len, clp->cl_name.data,
3881                             clp->cl_recdir);
3882
3883         /* find clp->cl_name in reclaim_str_hashtbl */
3884         strhashval = clientstr_hashval(clp->cl_recdir);
3885         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3886                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3887                         return crp;
3888                 }
3889         }
3890         return NULL;
3891 }
3892
3893 /*
3894 * Called from OPEN. Look for clientid in reclaim list.
3895 */
3896 __be32
3897 nfs4_check_open_reclaim(clientid_t *clid)
3898 {
3899         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3900 }
3901
3902 /* initialization to perform at module load time: */
3903
3904 int
3905 nfs4_state_init(void)
3906 {
3907         int i, status;
3908
3909         status = nfsd4_init_slabs();
3910         if (status)
3911                 return status;
3912         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3913                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3914                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3915                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3916                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3917                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3918         }
3919         for (i = 0; i < SESSION_HASH_SIZE; i++)
3920                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3921         for (i = 0; i < FILE_HASH_SIZE; i++) {
3922                 INIT_LIST_HEAD(&file_hashtbl[i]);
3923         }
3924         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3925                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3926                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3927         }
3928         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3929                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3930                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3931         }
3932         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3933                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3934                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3935         }
3936         memset(&onestateid, ~0, sizeof(stateid_t));
3937         INIT_LIST_HEAD(&close_lru);
3938         INIT_LIST_HEAD(&client_lru);
3939         INIT_LIST_HEAD(&del_recall_lru);
3940         reclaim_str_hashtbl_size = 0;
3941         return 0;
3942 }
3943
3944 static void
3945 nfsd4_load_reboot_recovery_data(void)
3946 {
3947         int status;
3948
3949         nfs4_lock_state();
3950         nfsd4_init_recdir(user_recovery_dirname);
3951         status = nfsd4_recdir_load();
3952         nfs4_unlock_state();
3953         if (status)
3954                 printk("NFSD: Failure reading reboot recovery data\n");
3955 }
3956
3957 /*
3958  * Since the lifetime of a delegation isn't limited to that of an open, a
3959  * client may quite reasonably hang on to a delegation as long as it has
3960  * the inode cached.  This becomes an obvious problem the first time a
3961  * client's inode cache approaches the size of the server's total memory.
3962  *
3963  * For now we avoid this problem by imposing a hard limit on the number
3964  * of delegations, which varies according to the server's memory size.
3965  */
3966 static void
3967 set_max_delegations(void)
3968 {
3969         /*
3970          * Allow at most 4 delegations per megabyte of RAM.  Quick
3971          * estimates suggest that in the worst case (where every delegation
3972          * is for a different inode), a delegation could take about 1.5K,
3973          * giving a worst case usage of about 6% of memory.
3974          */
3975         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
3976 }
3977
3978 /* initialization to perform when the nfsd service is started: */
3979
3980 static int
3981 __nfs4_state_start(void)
3982 {
3983         int ret;
3984
3985         boot_time = get_seconds();
3986         locks_start_grace(&nfsd4_manager);
3987         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
3988                nfsd4_grace);
3989         ret = set_callback_cred();
3990         if (ret)
3991                 return -ENOMEM;
3992         laundry_wq = create_singlethread_workqueue("nfsd4");
3993         if (laundry_wq == NULL)
3994                 return -ENOMEM;
3995         ret = nfsd4_create_callback_queue();
3996         if (ret)
3997                 goto out_free_laundry;
3998         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
3999         set_max_delegations();
4000         return 0;
4001 out_free_laundry:
4002         destroy_workqueue(laundry_wq);
4003         return ret;
4004 }
4005
4006 int
4007 nfs4_state_start(void)
4008 {
4009         int ret;
4010
4011         if (nfs4_init)
4012                 return 0;
4013         nfsd4_load_reboot_recovery_data();
4014         ret = __nfs4_state_start();
4015         if (ret)
4016                 return ret;
4017         nfs4_init = 1;
4018         return 0;
4019 }
4020
4021 static void
4022 __nfs4_state_shutdown(void)
4023 {
4024         int i;
4025         struct nfs4_client *clp = NULL;
4026         struct nfs4_delegation *dp = NULL;
4027         struct list_head *pos, *next, reaplist;
4028
4029         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4030                 while (!list_empty(&conf_id_hashtbl[i])) {
4031                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4032                         expire_client(clp);
4033                 }
4034                 while (!list_empty(&unconf_str_hashtbl[i])) {
4035                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4036                         expire_client(clp);
4037                 }
4038         }
4039         INIT_LIST_HEAD(&reaplist);
4040         spin_lock(&recall_lock);
4041         list_for_each_safe(pos, next, &del_recall_lru) {
4042                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4043                 list_move(&dp->dl_recall_lru, &reaplist);
4044         }
4045         spin_unlock(&recall_lock);
4046         list_for_each_safe(pos, next, &reaplist) {
4047                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4048                 list_del_init(&dp->dl_recall_lru);
4049                 unhash_delegation(dp);
4050         }
4051
4052         nfsd4_shutdown_recdir();
4053         nfs4_init = 0;
4054 }
4055
4056 void
4057 nfs4_state_shutdown(void)
4058 {
4059         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4060         destroy_workqueue(laundry_wq);
4061         locks_end_grace(&nfsd4_manager);
4062         nfs4_lock_state();
4063         nfs4_release_reclaim();
4064         __nfs4_state_shutdown();
4065         nfsd4_destroy_callback_queue();
4066         nfs4_unlock_state();
4067 }
4068
4069 /*
4070  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4071  * accessed when nfsd is starting.
4072  */
4073 static void
4074 nfs4_set_recdir(char *recdir)
4075 {
4076         strcpy(user_recovery_dirname, recdir);
4077 }
4078
4079 /*
4080  * Change the NFSv4 recovery directory to recdir.
4081  */
4082 int
4083 nfs4_reset_recoverydir(char *recdir)
4084 {
4085         int status;
4086         struct path path;
4087
4088         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4089         if (status)
4090                 return status;
4091         status = -ENOTDIR;
4092         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4093                 nfs4_set_recdir(recdir);
4094                 status = 0;
4095         }
4096         path_put(&path);
4097         return status;
4098 }
4099
4100 char *
4101 nfs4_recoverydir(void)
4102 {
4103         return user_recovery_dirname;
4104 }