nfsd41: Create the recovery entry for the NFSv4.1 client
[linux-2.6.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/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 static time_t lease_time = 90;     /* default lease time */
48 static time_t user_lease_time = 90;
49 static time_t boot_time;
50 static u32 current_ownerid = 1;
51 static u32 current_fileid = 1;
52 static u32 current_delegid = 1;
53 static u32 nfs4_init;
54 static stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static struct nfs4_delegation *
166 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
167 {
168         struct nfs4_delegation *dp;
169         struct nfs4_file *fp = stp->st_file;
170         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
171
172         dprintk("NFSD alloc_init_deleg\n");
173         if (fp->fi_had_conflict)
174                 return NULL;
175         if (num_delegations > max_delegations)
176                 return NULL;
177         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
178         if (dp == NULL)
179                 return dp;
180         num_delegations++;
181         INIT_LIST_HEAD(&dp->dl_perfile);
182         INIT_LIST_HEAD(&dp->dl_perclnt);
183         INIT_LIST_HEAD(&dp->dl_recall_lru);
184         dp->dl_client = clp;
185         get_nfs4_file(fp);
186         dp->dl_file = fp;
187         dp->dl_flock = NULL;
188         get_file(stp->st_vfs_file);
189         dp->dl_vfs_file = stp->st_vfs_file;
190         dp->dl_type = type;
191         dp->dl_ident = cb->cb_ident;
192         dp->dl_stateid.si_boot = get_seconds();
193         dp->dl_stateid.si_stateownerid = current_delegid++;
194         dp->dl_stateid.si_fileid = 0;
195         dp->dl_stateid.si_generation = 0;
196         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
197         dp->dl_time = 0;
198         atomic_set(&dp->dl_count, 1);
199         list_add(&dp->dl_perfile, &fp->fi_delegations);
200         list_add(&dp->dl_perclnt, &clp->cl_delegations);
201         return dp;
202 }
203
204 void
205 nfs4_put_delegation(struct nfs4_delegation *dp)
206 {
207         if (atomic_dec_and_test(&dp->dl_count)) {
208                 dprintk("NFSD: freeing dp %p\n",dp);
209                 put_nfs4_file(dp->dl_file);
210                 kmem_cache_free(deleg_slab, dp);
211                 num_delegations--;
212         }
213 }
214
215 /* Remove the associated file_lock first, then remove the delegation.
216  * lease_modify() is called to remove the FS_LEASE file_lock from
217  * the i_flock list, eventually calling nfsd's lock_manager
218  * fl_release_callback.
219  */
220 static void
221 nfs4_close_delegation(struct nfs4_delegation *dp)
222 {
223         struct file *filp = dp->dl_vfs_file;
224
225         dprintk("NFSD: close_delegation dp %p\n",dp);
226         dp->dl_vfs_file = NULL;
227         /* The following nfsd_close may not actually close the file,
228          * but we want to remove the lease in any case. */
229         if (dp->dl_flock)
230                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
231         nfsd_close(filp);
232 }
233
234 /* Called under the state lock. */
235 static void
236 unhash_delegation(struct nfs4_delegation *dp)
237 {
238         list_del_init(&dp->dl_perfile);
239         list_del_init(&dp->dl_perclnt);
240         spin_lock(&recall_lock);
241         list_del_init(&dp->dl_recall_lru);
242         spin_unlock(&recall_lock);
243         nfs4_close_delegation(dp);
244         nfs4_put_delegation(dp);
245 }
246
247 /* 
248  * SETCLIENTID state 
249  */
250
251 /* Hash tables for nfs4_clientid state */
252 #define CLIENT_HASH_BITS                 4
253 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
254 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
255
256 #define clientid_hashval(id) \
257         ((id) & CLIENT_HASH_MASK)
258 #define clientstr_hashval(name) \
259         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
260 /*
261  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
262  * used in reboot/reset lease grace period processing
263  *
264  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
265  * setclientid_confirmed info. 
266  *
267  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
268  * setclientid info.
269  *
270  * client_lru holds client queue ordered by nfs4_client.cl_time
271  * for lease renewal.
272  *
273  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
274  * for last close replay.
275  */
276 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
277 static int reclaim_str_hashtbl_size = 0;
278 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
279 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
280 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head client_lru;
283 static struct list_head close_lru;
284
285 static void unhash_generic_stateid(struct nfs4_stateid *stp)
286 {
287         list_del(&stp->st_hash);
288         list_del(&stp->st_perfile);
289         list_del(&stp->st_perstateowner);
290 }
291
292 static void free_generic_stateid(struct nfs4_stateid *stp)
293 {
294         put_nfs4_file(stp->st_file);
295         kmem_cache_free(stateid_slab, stp);
296 }
297
298 static void release_lock_stateid(struct nfs4_stateid *stp)
299 {
300         unhash_generic_stateid(stp);
301         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
302         free_generic_stateid(stp);
303 }
304
305 static void unhash_lockowner(struct nfs4_stateowner *sop)
306 {
307         struct nfs4_stateid *stp;
308
309         list_del(&sop->so_idhash);
310         list_del(&sop->so_strhash);
311         list_del(&sop->so_perstateid);
312         while (!list_empty(&sop->so_stateids)) {
313                 stp = list_first_entry(&sop->so_stateids,
314                                 struct nfs4_stateid, st_perstateowner);
315                 release_lock_stateid(stp);
316         }
317 }
318
319 static void release_lockowner(struct nfs4_stateowner *sop)
320 {
321         unhash_lockowner(sop);
322         nfs4_put_stateowner(sop);
323 }
324
325 static void
326 release_stateid_lockowners(struct nfs4_stateid *open_stp)
327 {
328         struct nfs4_stateowner *lock_sop;
329
330         while (!list_empty(&open_stp->st_lockowners)) {
331                 lock_sop = list_entry(open_stp->st_lockowners.next,
332                                 struct nfs4_stateowner, so_perstateid);
333                 /* list_del(&open_stp->st_lockowners);  */
334                 BUG_ON(lock_sop->so_is_open_owner);
335                 release_lockowner(lock_sop);
336         }
337 }
338
339 static void release_open_stateid(struct nfs4_stateid *stp)
340 {
341         unhash_generic_stateid(stp);
342         release_stateid_lockowners(stp);
343         nfsd_close(stp->st_vfs_file);
344         free_generic_stateid(stp);
345 }
346
347 static void unhash_openowner(struct nfs4_stateowner *sop)
348 {
349         struct nfs4_stateid *stp;
350
351         list_del(&sop->so_idhash);
352         list_del(&sop->so_strhash);
353         list_del(&sop->so_perclient);
354         list_del(&sop->so_perstateid); /* XXX: necessary? */
355         while (!list_empty(&sop->so_stateids)) {
356                 stp = list_first_entry(&sop->so_stateids,
357                                 struct nfs4_stateid, st_perstateowner);
358                 release_open_stateid(stp);
359         }
360 }
361
362 static void release_openowner(struct nfs4_stateowner *sop)
363 {
364         unhash_openowner(sop);
365         list_del(&sop->so_close_lru);
366         nfs4_put_stateowner(sop);
367 }
368
369 static DEFINE_SPINLOCK(sessionid_lock);
370 #define SESSION_HASH_SIZE       512
371 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
372
373 static inline int
374 hash_sessionid(struct nfs4_sessionid *sessionid)
375 {
376         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
377
378         return sid->sequence % SESSION_HASH_SIZE;
379 }
380
381 static inline void
382 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
383 {
384         u32 *ptr = (u32 *)(&sessionid->data[0]);
385         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
386 }
387
388 static void
389 gen_sessionid(struct nfsd4_session *ses)
390 {
391         struct nfs4_client *clp = ses->se_client;
392         struct nfsd4_sessionid *sid;
393
394         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
395         sid->clientid = clp->cl_clientid;
396         sid->sequence = current_sessionid++;
397         sid->reserved = 0;
398 }
399
400 /*
401  * The protocol defines ca_maxresponssize_cached to include the size of
402  * the rpc header, but all we need to cache is the data starting after
403  * the end of the initial SEQUENCE operation--the rest we regenerate
404  * each time.  Therefore we can advertise a ca_maxresponssize_cached
405  * value that is the number of bytes in our cache plus a few additional
406  * bytes.  In order to stay on the safe side, and not promise more than
407  * we can cache, those additional bytes must be the minimum possible: 24
408  * bytes of rpc header (xid through accept state, with AUTH_NULL
409  * verifier), 12 for the compound header (with zero-length tag), and 44
410  * for the SEQUENCE op response:
411  */
412 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
413
414 /*
415  * Give the client the number of ca_maxresponsesize_cached slots it
416  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
417  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
418  * than NFSD_MAX_SLOTS_PER_SESSION.
419  *
420  * If we run out of reserved DRC memory we should (up to a point)
421  * re-negotiate active sessions and reduce their slot usage to make
422  * rooom for new connections. For now we just fail the create session.
423  */
424 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
425 {
426         int mem, size = fchan->maxresp_cached;
427
428         if (fchan->maxreqs < 1)
429                 return nfserr_inval;
430
431         if (size < NFSD_MIN_HDR_SEQ_SZ)
432                 size = NFSD_MIN_HDR_SEQ_SZ;
433         size -= NFSD_MIN_HDR_SEQ_SZ;
434         if (size > NFSD_SLOT_CACHE_SIZE)
435                 size = NFSD_SLOT_CACHE_SIZE;
436
437         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
438         mem = fchan->maxreqs * size;
439         if (mem > NFSD_MAX_MEM_PER_SESSION) {
440                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
441                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
442                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
443                 mem = fchan->maxreqs * size;
444         }
445
446         spin_lock(&nfsd_drc_lock);
447         /* bound the total session drc memory ussage */
448         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
449                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
450                 mem = fchan->maxreqs * size;
451         }
452         nfsd_drc_mem_used += mem;
453         spin_unlock(&nfsd_drc_lock);
454
455         if (fchan->maxreqs == 0)
456                 return nfserr_serverfault;
457
458         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
459         return 0;
460 }
461
462 /*
463  * fchan holds the client values on input, and the server values on output
464  * sv_max_mesg is the maximum payload plus one page for overhead.
465  */
466 static int init_forechannel_attrs(struct svc_rqst *rqstp,
467                                   struct nfsd4_channel_attrs *session_fchan,
468                                   struct nfsd4_channel_attrs *fchan)
469 {
470         int status = 0;
471         __u32   maxcount = nfsd_serv->sv_max_mesg;
472
473         /* headerpadsz set to zero in encode routine */
474
475         /* Use the client's max request and max response size if possible */
476         if (fchan->maxreq_sz > maxcount)
477                 fchan->maxreq_sz = maxcount;
478         session_fchan->maxreq_sz = fchan->maxreq_sz;
479
480         if (fchan->maxresp_sz > maxcount)
481                 fchan->maxresp_sz = maxcount;
482         session_fchan->maxresp_sz = fchan->maxresp_sz;
483
484         /* Use the client's maxops if possible */
485         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
486                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
487         session_fchan->maxops = fchan->maxops;
488
489         /* FIXME: Error means no more DRC pages so the server should
490          * recover pages from existing sessions. For now fail session
491          * creation.
492          */
493         status = set_forechannel_drc_size(fchan);
494
495         session_fchan->maxresp_cached = fchan->maxresp_cached;
496         session_fchan->maxreqs = fchan->maxreqs;
497
498         dprintk("%s status %d\n", __func__, status);
499         return status;
500 }
501
502 static void
503 free_session_slots(struct nfsd4_session *ses)
504 {
505         int i;
506
507         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
508                 kfree(ses->se_slots[i]);
509 }
510
511 /*
512  * We don't actually need to cache the rpc and session headers, so we
513  * can allocate a little less for each slot:
514  */
515 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
516 {
517         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
518 }
519
520 static int
521 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
522                    struct nfsd4_create_session *cses)
523 {
524         struct nfsd4_session *new, tmp;
525         struct nfsd4_slot *sp;
526         int idx, slotsize, cachesize, i;
527         int status;
528
529         memset(&tmp, 0, sizeof(tmp));
530
531         /* FIXME: For now, we just accept the client back channel attributes. */
532         tmp.se_bchannel = cses->back_channel;
533         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
534                                         &cses->fore_channel);
535         if (status)
536                 goto out;
537
538         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
539                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
540
541         status = nfserr_serverfault;
542         /* allocate struct nfsd4_session and slot table pointers in one piece */
543         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
544         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
545         if (!new)
546                 goto out;
547
548         memcpy(new, &tmp, sizeof(*new));
549
550         /* allocate each struct nfsd4_slot and data cache in one piece */
551         cachesize = slot_bytes(&new->se_fchannel);
552         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
553                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
554                 if (!sp)
555                         goto out_free;
556                 new->se_slots[i] = sp;
557         }
558
559         new->se_client = clp;
560         gen_sessionid(new);
561         idx = hash_sessionid(&new->se_sessionid);
562         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
563                NFS4_MAX_SESSIONID_LEN);
564
565         new->se_flags = cses->flags;
566         kref_init(&new->se_ref);
567         spin_lock(&sessionid_lock);
568         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
569         list_add(&new->se_perclnt, &clp->cl_sessions);
570         spin_unlock(&sessionid_lock);
571
572         status = nfs_ok;
573 out:
574         return status;
575 out_free:
576         free_session_slots(new);
577         kfree(new);
578         goto out;
579 }
580
581 /* caller must hold sessionid_lock */
582 static struct nfsd4_session *
583 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
584 {
585         struct nfsd4_session *elem;
586         int idx;
587
588         dump_sessionid(__func__, sessionid);
589         idx = hash_sessionid(sessionid);
590         dprintk("%s: idx is %d\n", __func__, idx);
591         /* Search in the appropriate list */
592         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
593                 dump_sessionid("list traversal", &elem->se_sessionid);
594                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
595                             NFS4_MAX_SESSIONID_LEN)) {
596                         return elem;
597                 }
598         }
599
600         dprintk("%s: session not found\n", __func__);
601         return NULL;
602 }
603
604 /* caller must hold sessionid_lock */
605 static void
606 unhash_session(struct nfsd4_session *ses)
607 {
608         list_del(&ses->se_hash);
609         list_del(&ses->se_perclnt);
610 }
611
612 static void
613 release_session(struct nfsd4_session *ses)
614 {
615         spin_lock(&sessionid_lock);
616         unhash_session(ses);
617         spin_unlock(&sessionid_lock);
618         nfsd4_put_session(ses);
619 }
620
621 void
622 free_session(struct kref *kref)
623 {
624         struct nfsd4_session *ses;
625         int mem;
626
627         ses = container_of(kref, struct nfsd4_session, se_ref);
628         spin_lock(&nfsd_drc_lock);
629         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
630         nfsd_drc_mem_used -= mem;
631         spin_unlock(&nfsd_drc_lock);
632         free_session_slots(ses);
633         kfree(ses);
634 }
635
636 static inline void
637 renew_client(struct nfs4_client *clp)
638 {
639         /*
640         * Move client to the end to the LRU list.
641         */
642         dprintk("renewing client (clientid %08x/%08x)\n", 
643                         clp->cl_clientid.cl_boot, 
644                         clp->cl_clientid.cl_id);
645         list_move_tail(&clp->cl_lru, &client_lru);
646         clp->cl_time = get_seconds();
647 }
648
649 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
650 static int
651 STALE_CLIENTID(clientid_t *clid)
652 {
653         if (clid->cl_boot == boot_time)
654                 return 0;
655         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
656                 clid->cl_boot, clid->cl_id, boot_time);
657         return 1;
658 }
659
660 /* 
661  * XXX Should we use a slab cache ?
662  * This type of memory management is somewhat inefficient, but we use it
663  * anyway since SETCLIENTID is not a common operation.
664  */
665 static struct nfs4_client *alloc_client(struct xdr_netobj name)
666 {
667         struct nfs4_client *clp;
668
669         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
670         if (clp == NULL)
671                 return NULL;
672         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
673         if (clp->cl_name.data == NULL) {
674                 kfree(clp);
675                 return NULL;
676         }
677         memcpy(clp->cl_name.data, name.data, name.len);
678         clp->cl_name.len = name.len;
679         return clp;
680 }
681
682 static void
683 shutdown_callback_client(struct nfs4_client *clp)
684 {
685         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
686
687         if (clnt) {
688                 /*
689                  * Callback threads take a reference on the client, so there
690                  * should be no outstanding callbacks at this point.
691                  */
692                 clp->cl_cb_conn.cb_client = NULL;
693                 rpc_shutdown_client(clnt);
694         }
695 }
696
697 static inline void
698 free_client(struct nfs4_client *clp)
699 {
700         shutdown_callback_client(clp);
701         if (clp->cl_cb_xprt)
702                 svc_xprt_put(clp->cl_cb_xprt);
703         if (clp->cl_cred.cr_group_info)
704                 put_group_info(clp->cl_cred.cr_group_info);
705         kfree(clp->cl_principal);
706         kfree(clp->cl_name.data);
707         kfree(clp);
708 }
709
710 void
711 put_nfs4_client(struct nfs4_client *clp)
712 {
713         if (atomic_dec_and_test(&clp->cl_count))
714                 free_client(clp);
715 }
716
717 static void
718 expire_client(struct nfs4_client *clp)
719 {
720         struct nfs4_stateowner *sop;
721         struct nfs4_delegation *dp;
722         struct list_head reaplist;
723
724         dprintk("NFSD: expire_client cl_count %d\n",
725                             atomic_read(&clp->cl_count));
726
727         INIT_LIST_HEAD(&reaplist);
728         spin_lock(&recall_lock);
729         while (!list_empty(&clp->cl_delegations)) {
730                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
731                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
732                                 dp->dl_flock);
733                 list_del_init(&dp->dl_perclnt);
734                 list_move(&dp->dl_recall_lru, &reaplist);
735         }
736         spin_unlock(&recall_lock);
737         while (!list_empty(&reaplist)) {
738                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
739                 list_del_init(&dp->dl_recall_lru);
740                 unhash_delegation(dp);
741         }
742         list_del(&clp->cl_idhash);
743         list_del(&clp->cl_strhash);
744         list_del(&clp->cl_lru);
745         while (!list_empty(&clp->cl_openowners)) {
746                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
747                 release_openowner(sop);
748         }
749         while (!list_empty(&clp->cl_sessions)) {
750                 struct nfsd4_session  *ses;
751                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
752                                  se_perclnt);
753                 release_session(ses);
754         }
755         put_nfs4_client(clp);
756 }
757
758 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
759 {
760         memcpy(target->cl_verifier.data, source->data,
761                         sizeof(target->cl_verifier.data));
762 }
763
764 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
765 {
766         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
767         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
768 }
769
770 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
771 {
772         target->cr_uid = source->cr_uid;
773         target->cr_gid = source->cr_gid;
774         target->cr_group_info = source->cr_group_info;
775         get_group_info(target->cr_group_info);
776 }
777
778 static int same_name(const char *n1, const char *n2)
779 {
780         return 0 == memcmp(n1, n2, HEXDIR_LEN);
781 }
782
783 static int
784 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
785 {
786         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
787 }
788
789 static int
790 same_clid(clientid_t *cl1, clientid_t *cl2)
791 {
792         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
793 }
794
795 /* XXX what about NGROUP */
796 static int
797 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
798 {
799         return cr1->cr_uid == cr2->cr_uid;
800 }
801
802 static void gen_clid(struct nfs4_client *clp)
803 {
804         static u32 current_clientid = 1;
805
806         clp->cl_clientid.cl_boot = boot_time;
807         clp->cl_clientid.cl_id = current_clientid++; 
808 }
809
810 static void gen_confirm(struct nfs4_client *clp)
811 {
812         static u32 i;
813         u32 *p;
814
815         p = (u32 *)clp->cl_confirm.data;
816         *p++ = get_seconds();
817         *p++ = i++;
818 }
819
820 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
821                 struct svc_rqst *rqstp, nfs4_verifier *verf)
822 {
823         struct nfs4_client *clp;
824         struct sockaddr *sa = svc_addr(rqstp);
825         char *princ;
826
827         clp = alloc_client(name);
828         if (clp == NULL)
829                 return NULL;
830
831         princ = svc_gss_principal(rqstp);
832         if (princ) {
833                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
834                 if (clp->cl_principal == NULL) {
835                         free_client(clp);
836                         return NULL;
837                 }
838         }
839
840         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
841         atomic_set(&clp->cl_count, 1);
842         atomic_set(&clp->cl_cb_conn.cb_set, 0);
843         INIT_LIST_HEAD(&clp->cl_idhash);
844         INIT_LIST_HEAD(&clp->cl_strhash);
845         INIT_LIST_HEAD(&clp->cl_openowners);
846         INIT_LIST_HEAD(&clp->cl_delegations);
847         INIT_LIST_HEAD(&clp->cl_sessions);
848         INIT_LIST_HEAD(&clp->cl_lru);
849         clear_bit(0, &clp->cl_cb_slot_busy);
850         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
851         copy_verf(clp, verf);
852         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
853         clp->cl_flavor = rqstp->rq_flavor;
854         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
855         gen_confirm(clp);
856
857         return clp;
858 }
859
860 static int check_name(struct xdr_netobj name)
861 {
862         if (name.len == 0) 
863                 return 0;
864         if (name.len > NFS4_OPAQUE_LIMIT) {
865                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
866                 return 0;
867         }
868         return 1;
869 }
870
871 static void
872 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
873 {
874         unsigned int idhashval;
875
876         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
877         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
878         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
879         list_add_tail(&clp->cl_lru, &client_lru);
880         clp->cl_time = get_seconds();
881 }
882
883 static void
884 move_to_confirmed(struct nfs4_client *clp)
885 {
886         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
887         unsigned int strhashval;
888
889         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
890         list_del_init(&clp->cl_strhash);
891         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
892         strhashval = clientstr_hashval(clp->cl_recdir);
893         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
894         renew_client(clp);
895 }
896
897 static struct nfs4_client *
898 find_confirmed_client(clientid_t *clid)
899 {
900         struct nfs4_client *clp;
901         unsigned int idhashval = clientid_hashval(clid->cl_id);
902
903         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
904                 if (same_clid(&clp->cl_clientid, clid))
905                         return clp;
906         }
907         return NULL;
908 }
909
910 static struct nfs4_client *
911 find_unconfirmed_client(clientid_t *clid)
912 {
913         struct nfs4_client *clp;
914         unsigned int idhashval = clientid_hashval(clid->cl_id);
915
916         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
917                 if (same_clid(&clp->cl_clientid, clid))
918                         return clp;
919         }
920         return NULL;
921 }
922
923 /*
924  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
925  * parameter. Matching is based on the fact the at least one of the
926  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
927  *
928  * FIXME: we need to unify the clientid namespaces for nfsv4.x
929  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
930  * and SET_CLIENTID{,_CONFIRM}
931  */
932 static inline int
933 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
934 {
935         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
936         return use_exchange_id == has_exchange_flags;
937 }
938
939 static struct nfs4_client *
940 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
941                              bool use_exchange_id)
942 {
943         struct nfs4_client *clp;
944
945         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
946                 if (same_name(clp->cl_recdir, dname) &&
947                     match_clientid_establishment(clp, use_exchange_id))
948                         return clp;
949         }
950         return NULL;
951 }
952
953 static struct nfs4_client *
954 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
955                                bool use_exchange_id)
956 {
957         struct nfs4_client *clp;
958
959         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
960                 if (same_name(clp->cl_recdir, dname) &&
961                     match_clientid_establishment(clp, use_exchange_id))
962                         return clp;
963         }
964         return NULL;
965 }
966
967 static void
968 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
969 {
970         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
971         unsigned short expected_family;
972
973         /* Currently, we only support tcp and tcp6 for the callback channel */
974         if (se->se_callback_netid_len == 3 &&
975             !memcmp(se->se_callback_netid_val, "tcp", 3))
976                 expected_family = AF_INET;
977         else if (se->se_callback_netid_len == 4 &&
978                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
979                 expected_family = AF_INET6;
980         else
981                 goto out_err;
982
983         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
984                                             se->se_callback_addr_len,
985                                             (struct sockaddr *) &cb->cb_addr,
986                                             sizeof(cb->cb_addr));
987
988         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
989                 goto out_err;
990
991         if (cb->cb_addr.ss_family == AF_INET6)
992                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
993
994         cb->cb_minorversion = 0;
995         cb->cb_prog = se->se_callback_prog;
996         cb->cb_ident = se->se_callback_ident;
997         return;
998 out_err:
999         cb->cb_addr.ss_family = AF_UNSPEC;
1000         cb->cb_addrlen = 0;
1001         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1002                 "will not receive delegations\n",
1003                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1004
1005         return;
1006 }
1007
1008 /*
1009  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1010  */
1011 void
1012 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1013 {
1014         struct nfsd4_slot *slot = resp->cstate.slot;
1015         unsigned int base;
1016
1017         dprintk("--> %s slot %p\n", __func__, slot);
1018
1019         slot->sl_opcnt = resp->opcnt;
1020         slot->sl_status = resp->cstate.status;
1021
1022         if (nfsd4_not_cached(resp)) {
1023                 slot->sl_datalen = 0;
1024                 return;
1025         }
1026         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1027         base = (char *)resp->cstate.datap -
1028                                         (char *)resp->xbuf->head[0].iov_base;
1029         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1030                                     slot->sl_datalen))
1031                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1032         return;
1033 }
1034
1035 /*
1036  * Encode the replay sequence operation from the slot values.
1037  * If cachethis is FALSE encode the uncached rep error on the next
1038  * operation which sets resp->p and increments resp->opcnt for
1039  * nfs4svc_encode_compoundres.
1040  *
1041  */
1042 static __be32
1043 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1044                           struct nfsd4_compoundres *resp)
1045 {
1046         struct nfsd4_op *op;
1047         struct nfsd4_slot *slot = resp->cstate.slot;
1048
1049         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1050                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1051
1052         /* Encode the replayed sequence operation */
1053         op = &args->ops[resp->opcnt - 1];
1054         nfsd4_encode_operation(resp, op);
1055
1056         /* Return nfserr_retry_uncached_rep in next operation. */
1057         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1058                 op = &args->ops[resp->opcnt++];
1059                 op->status = nfserr_retry_uncached_rep;
1060                 nfsd4_encode_operation(resp, op);
1061         }
1062         return op->status;
1063 }
1064
1065 /*
1066  * The sequence operation is not cached because we can use the slot and
1067  * session values.
1068  */
1069 __be32
1070 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1071                          struct nfsd4_sequence *seq)
1072 {
1073         struct nfsd4_slot *slot = resp->cstate.slot;
1074         __be32 status;
1075
1076         dprintk("--> %s slot %p\n", __func__, slot);
1077
1078         /* Either returns 0 or nfserr_retry_uncached */
1079         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1080         if (status == nfserr_retry_uncached_rep)
1081                 return status;
1082
1083         /* The sequence operation has been encoded, cstate->datap set. */
1084         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1085
1086         resp->opcnt = slot->sl_opcnt;
1087         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1088         status = slot->sl_status;
1089
1090         return status;
1091 }
1092
1093 /*
1094  * Set the exchange_id flags returned by the server.
1095  */
1096 static void
1097 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1098 {
1099         /* pNFS is not supported */
1100         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1101
1102         /* Referrals are supported, Migration is not. */
1103         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1104
1105         /* set the wire flags to return to client. */
1106         clid->flags = new->cl_exchange_flags;
1107 }
1108
1109 __be32
1110 nfsd4_exchange_id(struct svc_rqst *rqstp,
1111                   struct nfsd4_compound_state *cstate,
1112                   struct nfsd4_exchange_id *exid)
1113 {
1114         struct nfs4_client *unconf, *conf, *new;
1115         int status;
1116         unsigned int            strhashval;
1117         char                    dname[HEXDIR_LEN];
1118         char                    addr_str[INET6_ADDRSTRLEN];
1119         nfs4_verifier           verf = exid->verifier;
1120         struct sockaddr         *sa = svc_addr(rqstp);
1121
1122         rpc_ntop(sa, addr_str, sizeof(addr_str));
1123         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1124                 "ip_addr=%s flags %x, spa_how %d\n",
1125                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1126                 addr_str, exid->flags, exid->spa_how);
1127
1128         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1129                 return nfserr_inval;
1130
1131         /* Currently only support SP4_NONE */
1132         switch (exid->spa_how) {
1133         case SP4_NONE:
1134                 break;
1135         case SP4_SSV:
1136                 return nfserr_encr_alg_unsupp;
1137         default:
1138                 BUG();                          /* checked by xdr code */
1139         case SP4_MACH_CRED:
1140                 return nfserr_serverfault;      /* no excuse :-/ */
1141         }
1142
1143         status = nfs4_make_rec_clidname(dname, &exid->clname);
1144
1145         if (status)
1146                 goto error;
1147
1148         strhashval = clientstr_hashval(dname);
1149
1150         nfs4_lock_state();
1151         status = nfs_ok;
1152
1153         conf = find_confirmed_client_by_str(dname, strhashval, true);
1154         if (conf) {
1155                 if (!same_verf(&verf, &conf->cl_verifier)) {
1156                         /* 18.35.4 case 8 */
1157                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1158                                 status = nfserr_not_same;
1159                                 goto out;
1160                         }
1161                         /* Client reboot: destroy old state */
1162                         expire_client(conf);
1163                         goto out_new;
1164                 }
1165                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1166                         /* 18.35.4 case 9 */
1167                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1168                                 status = nfserr_perm;
1169                                 goto out;
1170                         }
1171                         expire_client(conf);
1172                         goto out_new;
1173                 }
1174                 /*
1175                  * Set bit when the owner id and verifier map to an already
1176                  * confirmed client id (18.35.3).
1177                  */
1178                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1179
1180                 /*
1181                  * Falling into 18.35.4 case 2, possible router replay.
1182                  * Leave confirmed record intact and return same result.
1183                  */
1184                 copy_verf(conf, &verf);
1185                 new = conf;
1186                 goto out_copy;
1187         }
1188
1189         /* 18.35.4 case 7 */
1190         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1191                 status = nfserr_noent;
1192                 goto out;
1193         }
1194
1195         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1196         if (unconf) {
1197                 /*
1198                  * Possible retry or client restart.  Per 18.35.4 case 4,
1199                  * a new unconfirmed record should be generated regardless
1200                  * of whether any properties have changed.
1201                  */
1202                 expire_client(unconf);
1203         }
1204
1205 out_new:
1206         /* Normal case */
1207         new = create_client(exid->clname, dname, rqstp, &verf);
1208         if (new == NULL) {
1209                 status = nfserr_serverfault;
1210                 goto out;
1211         }
1212
1213         gen_clid(new);
1214         add_to_unconfirmed(new, strhashval);
1215 out_copy:
1216         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1217         exid->clientid.cl_id = new->cl_clientid.cl_id;
1218
1219         exid->seqid = 1;
1220         nfsd4_set_ex_flags(new, exid);
1221
1222         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1223                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1224         status = nfs_ok;
1225
1226 out:
1227         nfs4_unlock_state();
1228 error:
1229         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1230         return status;
1231 }
1232
1233 static int
1234 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1235 {
1236         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1237                 slot_seqid);
1238
1239         /* The slot is in use, and no response has been sent. */
1240         if (slot_inuse) {
1241                 if (seqid == slot_seqid)
1242                         return nfserr_jukebox;
1243                 else
1244                         return nfserr_seq_misordered;
1245         }
1246         /* Normal */
1247         if (likely(seqid == slot_seqid + 1))
1248                 return nfs_ok;
1249         /* Replay */
1250         if (seqid == slot_seqid)
1251                 return nfserr_replay_cache;
1252         /* Wraparound */
1253         if (seqid == 1 && (slot_seqid + 1) == 0)
1254                 return nfs_ok;
1255         /* Misordered replay or misordered new request */
1256         return nfserr_seq_misordered;
1257 }
1258
1259 /*
1260  * Cache the create session result into the create session single DRC
1261  * slot cache by saving the xdr structure. sl_seqid has been set.
1262  * Do this for solo or embedded create session operations.
1263  */
1264 static void
1265 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1266                            struct nfsd4_clid_slot *slot, int nfserr)
1267 {
1268         slot->sl_status = nfserr;
1269         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1270 }
1271
1272 static __be32
1273 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1274                             struct nfsd4_clid_slot *slot)
1275 {
1276         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1277         return slot->sl_status;
1278 }
1279
1280 __be32
1281 nfsd4_create_session(struct svc_rqst *rqstp,
1282                      struct nfsd4_compound_state *cstate,
1283                      struct nfsd4_create_session *cr_ses)
1284 {
1285         struct sockaddr *sa = svc_addr(rqstp);
1286         struct nfs4_client *conf, *unconf;
1287         struct nfsd4_clid_slot *cs_slot = NULL;
1288         int status = 0;
1289
1290         nfs4_lock_state();
1291         unconf = find_unconfirmed_client(&cr_ses->clientid);
1292         conf = find_confirmed_client(&cr_ses->clientid);
1293
1294         if (conf) {
1295                 cs_slot = &conf->cl_cs_slot;
1296                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1297                 if (status == nfserr_replay_cache) {
1298                         dprintk("Got a create_session replay! seqid= %d\n",
1299                                 cs_slot->sl_seqid);
1300                         /* Return the cached reply status */
1301                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1302                         goto out;
1303                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1304                         status = nfserr_seq_misordered;
1305                         dprintk("Sequence misordered!\n");
1306                         dprintk("Expected seqid= %d but got seqid= %d\n",
1307                                 cs_slot->sl_seqid, cr_ses->seqid);
1308                         goto out;
1309                 }
1310                 cs_slot->sl_seqid++;
1311         } else if (unconf) {
1312                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1313                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1314                         status = nfserr_clid_inuse;
1315                         goto out;
1316                 }
1317
1318                 cs_slot = &unconf->cl_cs_slot;
1319                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1320                 if (status) {
1321                         /* an unconfirmed replay returns misordered */
1322                         status = nfserr_seq_misordered;
1323                         goto out_cache;
1324                 }
1325
1326                 cs_slot->sl_seqid++; /* from 0 to 1 */
1327                 move_to_confirmed(unconf);
1328
1329                 /*
1330                  * We do not support RDMA or persistent sessions
1331                  */
1332                 cr_ses->flags &= ~SESSION4_PERSIST;
1333                 cr_ses->flags &= ~SESSION4_RDMA;
1334
1335                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1336                         unconf->cl_cb_xprt = rqstp->rq_xprt;
1337                         svc_xprt_get(unconf->cl_cb_xprt);
1338                         rpc_copy_addr(
1339                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1340                                 sa);
1341                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1342                         unconf->cl_cb_conn.cb_minorversion =
1343                                 cstate->minorversion;
1344                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1345                         unconf->cl_cb_seq_nr = 1;
1346                         nfsd4_probe_callback(unconf);
1347                 }
1348                 conf = unconf;
1349         } else {
1350                 status = nfserr_stale_clientid;
1351                 goto out;
1352         }
1353
1354         status = alloc_init_session(rqstp, conf, cr_ses);
1355         if (status)
1356                 goto out;
1357
1358         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1359                NFS4_MAX_SESSIONID_LEN);
1360         cr_ses->seqid = cs_slot->sl_seqid;
1361
1362 out_cache:
1363         /* cache solo and embedded create sessions under the state lock */
1364         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1365 out:
1366         nfs4_unlock_state();
1367         dprintk("%s returns %d\n", __func__, ntohl(status));
1368         return status;
1369 }
1370
1371 __be32
1372 nfsd4_destroy_session(struct svc_rqst *r,
1373                       struct nfsd4_compound_state *cstate,
1374                       struct nfsd4_destroy_session *sessionid)
1375 {
1376         struct nfsd4_session *ses;
1377         u32 status = nfserr_badsession;
1378
1379         /* Notes:
1380          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1381          * - Should we return nfserr_back_chan_busy if waiting for
1382          *   callbacks on to-be-destroyed session?
1383          * - Do we need to clear any callback info from previous session?
1384          */
1385
1386         dump_sessionid(__func__, &sessionid->sessionid);
1387         spin_lock(&sessionid_lock);
1388         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1389         if (!ses) {
1390                 spin_unlock(&sessionid_lock);
1391                 goto out;
1392         }
1393
1394         unhash_session(ses);
1395         spin_unlock(&sessionid_lock);
1396
1397         /* wait for callbacks */
1398         shutdown_callback_client(ses->se_client);
1399         nfsd4_put_session(ses);
1400         status = nfs_ok;
1401 out:
1402         dprintk("%s returns %d\n", __func__, ntohl(status));
1403         return status;
1404 }
1405
1406 __be32
1407 nfsd4_sequence(struct svc_rqst *rqstp,
1408                struct nfsd4_compound_state *cstate,
1409                struct nfsd4_sequence *seq)
1410 {
1411         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1412         struct nfsd4_session *session;
1413         struct nfsd4_slot *slot;
1414         int status;
1415
1416         if (resp->opcnt != 1)
1417                 return nfserr_sequence_pos;
1418
1419         spin_lock(&sessionid_lock);
1420         status = nfserr_badsession;
1421         session = find_in_sessionid_hashtbl(&seq->sessionid);
1422         if (!session)
1423                 goto out;
1424
1425         status = nfserr_badslot;
1426         if (seq->slotid >= session->se_fchannel.maxreqs)
1427                 goto out;
1428
1429         slot = session->se_slots[seq->slotid];
1430         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1431
1432         /* We do not negotiate the number of slots yet, so set the
1433          * maxslots to the session maxreqs which is used to encode
1434          * sr_highest_slotid and the sr_target_slot id to maxslots */
1435         seq->maxslots = session->se_fchannel.maxreqs;
1436
1437         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1438         if (status == nfserr_replay_cache) {
1439                 cstate->slot = slot;
1440                 cstate->session = session;
1441                 /* Return the cached reply status and set cstate->status
1442                  * for nfsd4_proc_compound processing */
1443                 status = nfsd4_replay_cache_entry(resp, seq);
1444                 cstate->status = nfserr_replay_cache;
1445                 goto out;
1446         }
1447         if (status)
1448                 goto out;
1449
1450         /* Success! bump slot seqid */
1451         slot->sl_inuse = true;
1452         slot->sl_seqid = seq->seqid;
1453         slot->sl_cachethis = seq->cachethis;
1454
1455         cstate->slot = slot;
1456         cstate->session = session;
1457
1458         /* Hold a session reference until done processing the compound:
1459          * nfsd4_put_session called only if the cstate slot is set.
1460          */
1461         nfsd4_get_session(session);
1462 out:
1463         spin_unlock(&sessionid_lock);
1464         /* Renew the clientid on success and on replay */
1465         if (cstate->session) {
1466                 nfs4_lock_state();
1467                 renew_client(session->se_client);
1468                 nfs4_unlock_state();
1469         }
1470         dprintk("%s: return %d\n", __func__, ntohl(status));
1471         return status;
1472 }
1473
1474 __be32
1475 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1476                   struct nfsd4_setclientid *setclid)
1477 {
1478         struct sockaddr         *sa = svc_addr(rqstp);
1479         struct xdr_netobj       clname = { 
1480                 .len = setclid->se_namelen,
1481                 .data = setclid->se_name,
1482         };
1483         nfs4_verifier           clverifier = setclid->se_verf;
1484         unsigned int            strhashval;
1485         struct nfs4_client      *conf, *unconf, *new;
1486         __be32                  status;
1487         char                    dname[HEXDIR_LEN];
1488         
1489         if (!check_name(clname))
1490                 return nfserr_inval;
1491
1492         status = nfs4_make_rec_clidname(dname, &clname);
1493         if (status)
1494                 return status;
1495
1496         /* 
1497          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1498          * We get here on a DRC miss.
1499          */
1500
1501         strhashval = clientstr_hashval(dname);
1502
1503         nfs4_lock_state();
1504         conf = find_confirmed_client_by_str(dname, strhashval, false);
1505         if (conf) {
1506                 /* RFC 3530 14.2.33 CASE 0: */
1507                 status = nfserr_clid_inuse;
1508                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1509                         char addr_str[INET6_ADDRSTRLEN];
1510                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1511                                  sizeof(addr_str));
1512                         dprintk("NFSD: setclientid: string in use by client "
1513                                 "at %s\n", addr_str);
1514                         goto out;
1515                 }
1516         }
1517         /*
1518          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1519          * has a description of SETCLIENTID request processing consisting
1520          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1521          */
1522         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1523         status = nfserr_resource;
1524         if (!conf) {
1525                 /*
1526                  * RFC 3530 14.2.33 CASE 4:
1527                  * placed first, because it is the normal case
1528                  */
1529                 if (unconf)
1530                         expire_client(unconf);
1531                 new = create_client(clname, dname, rqstp, &clverifier);
1532                 if (new == NULL)
1533                         goto out;
1534                 gen_clid(new);
1535         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1536                 /*
1537                  * RFC 3530 14.2.33 CASE 1:
1538                  * probable callback update
1539                  */
1540                 if (unconf) {
1541                         /* Note this is removing unconfirmed {*x***},
1542                          * which is stronger than RFC recommended {vxc**}.
1543                          * This has the advantage that there is at most
1544                          * one {*x***} in either list at any time.
1545                          */
1546                         expire_client(unconf);
1547                 }
1548                 new = create_client(clname, dname, rqstp, &clverifier);
1549                 if (new == NULL)
1550                         goto out;
1551                 copy_clid(new, conf);
1552         } else if (!unconf) {
1553                 /*
1554                  * RFC 3530 14.2.33 CASE 2:
1555                  * probable client reboot; state will be removed if
1556                  * confirmed.
1557                  */
1558                 new = create_client(clname, dname, rqstp, &clverifier);
1559                 if (new == NULL)
1560                         goto out;
1561                 gen_clid(new);
1562         } else {
1563                 /*
1564                  * RFC 3530 14.2.33 CASE 3:
1565                  * probable client reboot; state will be removed if
1566                  * confirmed.
1567                  */
1568                 expire_client(unconf);
1569                 new = create_client(clname, dname, rqstp, &clverifier);
1570                 if (new == NULL)
1571                         goto out;
1572                 gen_clid(new);
1573         }
1574         gen_callback(new, setclid, rpc_get_scope_id(sa));
1575         add_to_unconfirmed(new, strhashval);
1576         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1577         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1578         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1579         status = nfs_ok;
1580 out:
1581         nfs4_unlock_state();
1582         return status;
1583 }
1584
1585
1586 /*
1587  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1588  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1589  * bullets, labeled as CASE1 - CASE4 below.
1590  */
1591 __be32
1592 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1593                          struct nfsd4_compound_state *cstate,
1594                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1595 {
1596         struct sockaddr *sa = svc_addr(rqstp);
1597         struct nfs4_client *conf, *unconf;
1598         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1599         clientid_t * clid = &setclientid_confirm->sc_clientid;
1600         __be32 status;
1601
1602         if (STALE_CLIENTID(clid))
1603                 return nfserr_stale_clientid;
1604         /* 
1605          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1606          * We get here on a DRC miss.
1607          */
1608
1609         nfs4_lock_state();
1610
1611         conf = find_confirmed_client(clid);
1612         unconf = find_unconfirmed_client(clid);
1613
1614         status = nfserr_clid_inuse;
1615         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1616                 goto out;
1617         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1618                 goto out;
1619
1620         /*
1621          * section 14.2.34 of RFC 3530 has a description of
1622          * SETCLIENTID_CONFIRM request processing consisting
1623          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1624          */
1625         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1626                 /*
1627                  * RFC 3530 14.2.34 CASE 1:
1628                  * callback update
1629                  */
1630                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1631                         status = nfserr_clid_inuse;
1632                 else {
1633                         /* XXX: We just turn off callbacks until we can handle
1634                           * change request correctly. */
1635                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1636                         expire_client(unconf);
1637                         status = nfs_ok;
1638
1639                 }
1640         } else if (conf && !unconf) {
1641                 /*
1642                  * RFC 3530 14.2.34 CASE 2:
1643                  * probable retransmitted request; play it safe and
1644                  * do nothing.
1645                  */
1646                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1647                         status = nfserr_clid_inuse;
1648                 else
1649                         status = nfs_ok;
1650         } else if (!conf && unconf
1651                         && same_verf(&unconf->cl_confirm, &confirm)) {
1652                 /*
1653                  * RFC 3530 14.2.34 CASE 3:
1654                  * Normal case; new or rebooted client:
1655                  */
1656                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1657                         status = nfserr_clid_inuse;
1658                 } else {
1659                         unsigned int hash =
1660                                 clientstr_hashval(unconf->cl_recdir);
1661                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1662                                                             hash, false);
1663                         if (conf) {
1664                                 nfsd4_remove_clid_dir(conf);
1665                                 expire_client(conf);
1666                         }
1667                         move_to_confirmed(unconf);
1668                         conf = unconf;
1669                         nfsd4_probe_callback(conf);
1670                         status = nfs_ok;
1671                 }
1672         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1673             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1674                                                                 &confirm)))) {
1675                 /*
1676                  * RFC 3530 14.2.34 CASE 4:
1677                  * Client probably hasn't noticed that we rebooted yet.
1678                  */
1679                 status = nfserr_stale_clientid;
1680         } else {
1681                 /* check that we have hit one of the cases...*/
1682                 status = nfserr_clid_inuse;
1683         }
1684 out:
1685         nfs4_unlock_state();
1686         return status;
1687 }
1688
1689 /* OPEN Share state helper functions */
1690 static inline struct nfs4_file *
1691 alloc_init_file(struct inode *ino)
1692 {
1693         struct nfs4_file *fp;
1694         unsigned int hashval = file_hashval(ino);
1695
1696         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1697         if (fp) {
1698                 atomic_set(&fp->fi_ref, 1);
1699                 INIT_LIST_HEAD(&fp->fi_hash);
1700                 INIT_LIST_HEAD(&fp->fi_stateids);
1701                 INIT_LIST_HEAD(&fp->fi_delegations);
1702                 spin_lock(&recall_lock);
1703                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1704                 spin_unlock(&recall_lock);
1705                 fp->fi_inode = igrab(ino);
1706                 fp->fi_id = current_fileid++;
1707                 fp->fi_had_conflict = false;
1708                 return fp;
1709         }
1710         return NULL;
1711 }
1712
1713 static void
1714 nfsd4_free_slab(struct kmem_cache **slab)
1715 {
1716         if (*slab == NULL)
1717                 return;
1718         kmem_cache_destroy(*slab);
1719         *slab = NULL;
1720 }
1721
1722 void
1723 nfsd4_free_slabs(void)
1724 {
1725         nfsd4_free_slab(&stateowner_slab);
1726         nfsd4_free_slab(&file_slab);
1727         nfsd4_free_slab(&stateid_slab);
1728         nfsd4_free_slab(&deleg_slab);
1729 }
1730
1731 static int
1732 nfsd4_init_slabs(void)
1733 {
1734         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1735                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1736         if (stateowner_slab == NULL)
1737                 goto out_nomem;
1738         file_slab = kmem_cache_create("nfsd4_files",
1739                         sizeof(struct nfs4_file), 0, 0, NULL);
1740         if (file_slab == NULL)
1741                 goto out_nomem;
1742         stateid_slab = kmem_cache_create("nfsd4_stateids",
1743                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1744         if (stateid_slab == NULL)
1745                 goto out_nomem;
1746         deleg_slab = kmem_cache_create("nfsd4_delegations",
1747                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1748         if (deleg_slab == NULL)
1749                 goto out_nomem;
1750         return 0;
1751 out_nomem:
1752         nfsd4_free_slabs();
1753         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1754         return -ENOMEM;
1755 }
1756
1757 void
1758 nfs4_free_stateowner(struct kref *kref)
1759 {
1760         struct nfs4_stateowner *sop =
1761                 container_of(kref, struct nfs4_stateowner, so_ref);
1762         kfree(sop->so_owner.data);
1763         kmem_cache_free(stateowner_slab, sop);
1764 }
1765
1766 static inline struct nfs4_stateowner *
1767 alloc_stateowner(struct xdr_netobj *owner)
1768 {
1769         struct nfs4_stateowner *sop;
1770
1771         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1772                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1773                         memcpy(sop->so_owner.data, owner->data, owner->len);
1774                         sop->so_owner.len = owner->len;
1775                         kref_init(&sop->so_ref);
1776                         return sop;
1777                 } 
1778                 kmem_cache_free(stateowner_slab, sop);
1779         }
1780         return NULL;
1781 }
1782
1783 static struct nfs4_stateowner *
1784 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1785         struct nfs4_stateowner *sop;
1786         struct nfs4_replay *rp;
1787         unsigned int idhashval;
1788
1789         if (!(sop = alloc_stateowner(&open->op_owner)))
1790                 return NULL;
1791         idhashval = ownerid_hashval(current_ownerid);
1792         INIT_LIST_HEAD(&sop->so_idhash);
1793         INIT_LIST_HEAD(&sop->so_strhash);
1794         INIT_LIST_HEAD(&sop->so_perclient);
1795         INIT_LIST_HEAD(&sop->so_stateids);
1796         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1797         INIT_LIST_HEAD(&sop->so_close_lru);
1798         sop->so_time = 0;
1799         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1800         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1801         list_add(&sop->so_perclient, &clp->cl_openowners);
1802         sop->so_is_open_owner = 1;
1803         sop->so_id = current_ownerid++;
1804         sop->so_client = clp;
1805         sop->so_seqid = open->op_seqid;
1806         sop->so_confirmed = 0;
1807         rp = &sop->so_replay;
1808         rp->rp_status = nfserr_serverfault;
1809         rp->rp_buflen = 0;
1810         rp->rp_buf = rp->rp_ibuf;
1811         return sop;
1812 }
1813
1814 static inline void
1815 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1816         struct nfs4_stateowner *sop = open->op_stateowner;
1817         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1818
1819         INIT_LIST_HEAD(&stp->st_hash);
1820         INIT_LIST_HEAD(&stp->st_perstateowner);
1821         INIT_LIST_HEAD(&stp->st_lockowners);
1822         INIT_LIST_HEAD(&stp->st_perfile);
1823         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1824         list_add(&stp->st_perstateowner, &sop->so_stateids);
1825         list_add(&stp->st_perfile, &fp->fi_stateids);
1826         stp->st_stateowner = sop;
1827         get_nfs4_file(fp);
1828         stp->st_file = fp;
1829         stp->st_stateid.si_boot = get_seconds();
1830         stp->st_stateid.si_stateownerid = sop->so_id;
1831         stp->st_stateid.si_fileid = fp->fi_id;
1832         stp->st_stateid.si_generation = 0;
1833         stp->st_access_bmap = 0;
1834         stp->st_deny_bmap = 0;
1835         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1836                   &stp->st_access_bmap);
1837         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1838         stp->st_openstp = NULL;
1839 }
1840
1841 static void
1842 move_to_close_lru(struct nfs4_stateowner *sop)
1843 {
1844         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1845
1846         list_move_tail(&sop->so_close_lru, &close_lru);
1847         sop->so_time = get_seconds();
1848 }
1849
1850 static int
1851 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1852                                                         clientid_t *clid)
1853 {
1854         return (sop->so_owner.len == owner->len) &&
1855                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1856                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1857 }
1858
1859 static struct nfs4_stateowner *
1860 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1861 {
1862         struct nfs4_stateowner *so = NULL;
1863
1864         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1865                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1866                         return so;
1867         }
1868         return NULL;
1869 }
1870
1871 /* search file_hashtbl[] for file */
1872 static struct nfs4_file *
1873 find_file(struct inode *ino)
1874 {
1875         unsigned int hashval = file_hashval(ino);
1876         struct nfs4_file *fp;
1877
1878         spin_lock(&recall_lock);
1879         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1880                 if (fp->fi_inode == ino) {
1881                         get_nfs4_file(fp);
1882                         spin_unlock(&recall_lock);
1883                         return fp;
1884                 }
1885         }
1886         spin_unlock(&recall_lock);
1887         return NULL;
1888 }
1889
1890 static inline int access_valid(u32 x, u32 minorversion)
1891 {
1892         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1893                 return 0;
1894         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1895                 return 0;
1896         x &= ~NFS4_SHARE_ACCESS_MASK;
1897         if (minorversion && x) {
1898                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1899                         return 0;
1900                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1901                         return 0;
1902                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1903         }
1904         if (x)
1905                 return 0;
1906         return 1;
1907 }
1908
1909 static inline int deny_valid(u32 x)
1910 {
1911         /* Note: unlike access bits, deny bits may be zero. */
1912         return x <= NFS4_SHARE_DENY_BOTH;
1913 }
1914
1915 /*
1916  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1917  * st_{access,deny}_bmap field of the stateid, in order to track not
1918  * only what share bits are currently in force, but also what
1919  * combinations of share bits previous opens have used.  This allows us
1920  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1921  * return an error if the client attempt to downgrade to a combination
1922  * of share bits not explicable by closing some of its previous opens.
1923  *
1924  * XXX: This enforcement is actually incomplete, since we don't keep
1925  * track of access/deny bit combinations; so, e.g., we allow:
1926  *
1927  *      OPEN allow read, deny write
1928  *      OPEN allow both, deny none
1929  *      DOWNGRADE allow read, deny none
1930  *
1931  * which we should reject.
1932  */
1933 static void
1934 set_access(unsigned int *access, unsigned long bmap) {
1935         int i;
1936
1937         *access = 0;
1938         for (i = 1; i < 4; i++) {
1939                 if (test_bit(i, &bmap))
1940                         *access |= i;
1941         }
1942 }
1943
1944 static void
1945 set_deny(unsigned int *deny, unsigned long bmap) {
1946         int i;
1947
1948         *deny = 0;
1949         for (i = 0; i < 4; i++) {
1950                 if (test_bit(i, &bmap))
1951                         *deny |= i ;
1952         }
1953 }
1954
1955 static int
1956 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1957         unsigned int access, deny;
1958
1959         set_access(&access, stp->st_access_bmap);
1960         set_deny(&deny, stp->st_deny_bmap);
1961         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1962                 return 0;
1963         return 1;
1964 }
1965
1966 /*
1967  * Called to check deny when READ with all zero stateid or
1968  * WRITE with all zero or all one stateid
1969  */
1970 static __be32
1971 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1972 {
1973         struct inode *ino = current_fh->fh_dentry->d_inode;
1974         struct nfs4_file *fp;
1975         struct nfs4_stateid *stp;
1976         __be32 ret;
1977
1978         dprintk("NFSD: nfs4_share_conflict\n");
1979
1980         fp = find_file(ino);
1981         if (!fp)
1982                 return nfs_ok;
1983         ret = nfserr_locked;
1984         /* Search for conflicting share reservations */
1985         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1986                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1987                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1988                         goto out;
1989         }
1990         ret = nfs_ok;
1991 out:
1992         put_nfs4_file(fp);
1993         return ret;
1994 }
1995
1996 static inline void
1997 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1998 {
1999         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2000                 drop_file_write_access(filp);
2001                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2002         }
2003 }
2004
2005 /*
2006  * Spawn a thread to perform a recall on the delegation represented
2007  * by the lease (file_lock)
2008  *
2009  * Called from break_lease() with lock_kernel() held.
2010  * Note: we assume break_lease will only call this *once* for any given
2011  * lease.
2012  */
2013 static
2014 void nfsd_break_deleg_cb(struct file_lock *fl)
2015 {
2016         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2017
2018         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2019         if (!dp)
2020                 return;
2021
2022         /* We're assuming the state code never drops its reference
2023          * without first removing the lease.  Since we're in this lease
2024          * callback (and since the lease code is serialized by the kernel
2025          * lock) we know the server hasn't removed the lease yet, we know
2026          * it's safe to take a reference: */
2027         atomic_inc(&dp->dl_count);
2028         atomic_inc(&dp->dl_client->cl_count);
2029
2030         spin_lock(&recall_lock);
2031         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2032         spin_unlock(&recall_lock);
2033
2034         /* only place dl_time is set. protected by lock_kernel*/
2035         dp->dl_time = get_seconds();
2036
2037         /*
2038          * We don't want the locks code to timeout the lease for us;
2039          * we'll remove it ourself if the delegation isn't returned
2040          * in time.
2041          */
2042         fl->fl_break_time = 0;
2043
2044         dp->dl_file->fi_had_conflict = true;
2045         nfsd4_cb_recall(dp);
2046 }
2047
2048 /*
2049  * The file_lock is being reapd.
2050  *
2051  * Called by locks_free_lock() with lock_kernel() held.
2052  */
2053 static
2054 void nfsd_release_deleg_cb(struct file_lock *fl)
2055 {
2056         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2057
2058         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2059
2060         if (!(fl->fl_flags & FL_LEASE) || !dp)
2061                 return;
2062         dp->dl_flock = NULL;
2063 }
2064
2065 /*
2066  * Set the delegation file_lock back pointer.
2067  *
2068  * Called from setlease() with lock_kernel() held.
2069  */
2070 static
2071 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2072 {
2073         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2074
2075         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2076         if (!dp)
2077                 return;
2078         dp->dl_flock = new;
2079 }
2080
2081 /*
2082  * Called from setlease() with lock_kernel() held
2083  */
2084 static
2085 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2086 {
2087         struct nfs4_delegation *onlistd =
2088                 (struct nfs4_delegation *)onlist->fl_owner;
2089         struct nfs4_delegation *tryd =
2090                 (struct nfs4_delegation *)try->fl_owner;
2091
2092         if (onlist->fl_lmops != try->fl_lmops)
2093                 return 0;
2094
2095         return onlistd->dl_client == tryd->dl_client;
2096 }
2097
2098
2099 static
2100 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2101 {
2102         if (arg & F_UNLCK)
2103                 return lease_modify(onlist, arg);
2104         else
2105                 return -EAGAIN;
2106 }
2107
2108 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2109         .fl_break = nfsd_break_deleg_cb,
2110         .fl_release_private = nfsd_release_deleg_cb,
2111         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2112         .fl_mylease = nfsd_same_client_deleg_cb,
2113         .fl_change = nfsd_change_deleg_cb,
2114 };
2115
2116
2117 __be32
2118 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2119                     struct nfsd4_open *open)
2120 {
2121         clientid_t *clientid = &open->op_clientid;
2122         struct nfs4_client *clp = NULL;
2123         unsigned int strhashval;
2124         struct nfs4_stateowner *sop = NULL;
2125
2126         if (!check_name(open->op_owner))
2127                 return nfserr_inval;
2128
2129         if (STALE_CLIENTID(&open->op_clientid))
2130                 return nfserr_stale_clientid;
2131
2132         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2133         sop = find_openstateowner_str(strhashval, open);
2134         open->op_stateowner = sop;
2135         if (!sop) {
2136                 /* Make sure the client's lease hasn't expired. */
2137                 clp = find_confirmed_client(clientid);
2138                 if (clp == NULL)
2139                         return nfserr_expired;
2140                 goto renew;
2141         }
2142         /* When sessions are used, skip open sequenceid processing */
2143         if (nfsd4_has_session(cstate))
2144                 goto renew;
2145         if (!sop->so_confirmed) {
2146                 /* Replace unconfirmed owners without checking for replay. */
2147                 clp = sop->so_client;
2148                 release_openowner(sop);
2149                 open->op_stateowner = NULL;
2150                 goto renew;
2151         }
2152         if (open->op_seqid == sop->so_seqid - 1) {
2153                 if (sop->so_replay.rp_buflen)
2154                         return nfserr_replay_me;
2155                 /* The original OPEN failed so spectacularly
2156                  * that we don't even have replay data saved!
2157                  * Therefore, we have no choice but to continue
2158                  * processing this OPEN; presumably, we'll
2159                  * fail again for the same reason.
2160                  */
2161                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2162                 goto renew;
2163         }
2164         if (open->op_seqid != sop->so_seqid)
2165                 return nfserr_bad_seqid;
2166 renew:
2167         if (open->op_stateowner == NULL) {
2168                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2169                 if (sop == NULL)
2170                         return nfserr_resource;
2171                 open->op_stateowner = sop;
2172         }
2173         list_del_init(&sop->so_close_lru);
2174         renew_client(sop->so_client);
2175         return nfs_ok;
2176 }
2177
2178 static inline __be32
2179 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2180 {
2181         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2182                 return nfserr_openmode;
2183         else
2184                 return nfs_ok;
2185 }
2186
2187 static struct nfs4_delegation *
2188 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2189 {
2190         struct nfs4_delegation *dp;
2191
2192         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2193                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2194                         return dp;
2195         }
2196         return NULL;
2197 }
2198
2199 static __be32
2200 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2201                 struct nfs4_delegation **dp)
2202 {
2203         int flags;
2204         __be32 status = nfserr_bad_stateid;
2205
2206         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2207         if (*dp == NULL)
2208                 goto out;
2209         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2210                                                 RD_STATE : WR_STATE;
2211         status = nfs4_check_delegmode(*dp, flags);
2212         if (status)
2213                 *dp = NULL;
2214 out:
2215         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2216                 return nfs_ok;
2217         if (status)
2218                 return status;
2219         open->op_stateowner->so_confirmed = 1;
2220         return nfs_ok;
2221 }
2222
2223 static __be32
2224 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2225 {
2226         struct nfs4_stateid *local;
2227         __be32 status = nfserr_share_denied;
2228         struct nfs4_stateowner *sop = open->op_stateowner;
2229
2230         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2231                 /* ignore lock owners */
2232                 if (local->st_stateowner->so_is_open_owner == 0)
2233                         continue;
2234                 /* remember if we have seen this open owner */
2235                 if (local->st_stateowner == sop)
2236                         *stpp = local;
2237                 /* check for conflicting share reservations */
2238                 if (!test_share(local, open))
2239                         goto out;
2240         }
2241         status = 0;
2242 out:
2243         return status;
2244 }
2245
2246 static inline struct nfs4_stateid *
2247 nfs4_alloc_stateid(void)
2248 {
2249         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2250 }
2251
2252 static __be32
2253 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2254                 struct nfs4_delegation *dp,
2255                 struct svc_fh *cur_fh, int flags)
2256 {
2257         struct nfs4_stateid *stp;
2258
2259         stp = nfs4_alloc_stateid();
2260         if (stp == NULL)
2261                 return nfserr_resource;
2262
2263         if (dp) {
2264                 get_file(dp->dl_vfs_file);
2265                 stp->st_vfs_file = dp->dl_vfs_file;
2266         } else {
2267                 __be32 status;
2268                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2269                                 &stp->st_vfs_file);
2270                 if (status) {
2271                         if (status == nfserr_dropit)
2272                                 status = nfserr_jukebox;
2273                         kmem_cache_free(stateid_slab, stp);
2274                         return status;
2275                 }
2276         }
2277         *stpp = stp;
2278         return 0;
2279 }
2280
2281 static inline __be32
2282 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2283                 struct nfsd4_open *open)
2284 {
2285         struct iattr iattr = {
2286                 .ia_valid = ATTR_SIZE,
2287                 .ia_size = 0,
2288         };
2289         if (!open->op_truncate)
2290                 return 0;
2291         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2292                 return nfserr_inval;
2293         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2294 }
2295
2296 static __be32
2297 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2298 {
2299         struct file *filp = stp->st_vfs_file;
2300         struct inode *inode = filp->f_path.dentry->d_inode;
2301         unsigned int share_access, new_writer;
2302         __be32 status;
2303
2304         set_access(&share_access, stp->st_access_bmap);
2305         new_writer = (~share_access) & open->op_share_access
2306                         & NFS4_SHARE_ACCESS_WRITE;
2307
2308         if (new_writer) {
2309                 int err = get_write_access(inode);
2310                 if (err)
2311                         return nfserrno(err);
2312                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2313                 if (err)
2314                         return nfserrno(err);
2315                 file_take_write(filp);
2316         }
2317         status = nfsd4_truncate(rqstp, cur_fh, open);
2318         if (status) {
2319                 if (new_writer)
2320                         put_write_access(inode);
2321                 return status;
2322         }
2323         /* remember the open */
2324         filp->f_mode |= open->op_share_access;
2325         __set_bit(open->op_share_access, &stp->st_access_bmap);
2326         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2327
2328         return nfs_ok;
2329 }
2330
2331
2332 static void
2333 nfs4_set_claim_prev(struct nfsd4_open *open)
2334 {
2335         open->op_stateowner->so_confirmed = 1;
2336         open->op_stateowner->so_client->cl_firststate = 1;
2337 }
2338
2339 /*
2340  * Attempt to hand out a delegation.
2341  */
2342 static void
2343 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2344 {
2345         struct nfs4_delegation *dp;
2346         struct nfs4_stateowner *sop = stp->st_stateowner;
2347         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2348         struct file_lock fl, *flp = &fl;
2349         int status, flag = 0;
2350
2351         flag = NFS4_OPEN_DELEGATE_NONE;
2352         open->op_recall = 0;
2353         switch (open->op_claim_type) {
2354                 case NFS4_OPEN_CLAIM_PREVIOUS:
2355                         if (!atomic_read(&cb->cb_set))
2356                                 open->op_recall = 1;
2357                         flag = open->op_delegate_type;
2358                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2359                                 goto out;
2360                         break;
2361                 case NFS4_OPEN_CLAIM_NULL:
2362                         /* Let's not give out any delegations till everyone's
2363                          * had the chance to reclaim theirs.... */
2364                         if (locks_in_grace())
2365                                 goto out;
2366                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2367                                 goto out;
2368                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2369                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2370                         else
2371                                 flag = NFS4_OPEN_DELEGATE_READ;
2372                         break;
2373                 default:
2374                         goto out;
2375         }
2376
2377         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2378         if (dp == NULL) {
2379                 flag = NFS4_OPEN_DELEGATE_NONE;
2380                 goto out;
2381         }
2382         locks_init_lock(&fl);
2383         fl.fl_lmops = &nfsd_lease_mng_ops;
2384         fl.fl_flags = FL_LEASE;
2385         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2386         fl.fl_end = OFFSET_MAX;
2387         fl.fl_owner =  (fl_owner_t)dp;
2388         fl.fl_file = stp->st_vfs_file;
2389         fl.fl_pid = current->tgid;
2390
2391         /* vfs_setlease checks to see if delegation should be handed out.
2392          * the lock_manager callbacks fl_mylease and fl_change are used
2393          */
2394         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2395                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2396                 unhash_delegation(dp);
2397                 flag = NFS4_OPEN_DELEGATE_NONE;
2398                 goto out;
2399         }
2400
2401         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2402
2403         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2404                 STATEID_VAL(&dp->dl_stateid));
2405 out:
2406         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2407                         && flag == NFS4_OPEN_DELEGATE_NONE
2408                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2409                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2410         open->op_delegate_type = flag;
2411 }
2412
2413 /*
2414  * called with nfs4_lock_state() held.
2415  */
2416 __be32
2417 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2418 {
2419         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2420         struct nfs4_file *fp = NULL;
2421         struct inode *ino = current_fh->fh_dentry->d_inode;
2422         struct nfs4_stateid *stp = NULL;
2423         struct nfs4_delegation *dp = NULL;
2424         __be32 status;
2425
2426         status = nfserr_inval;
2427         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2428                         || !deny_valid(open->op_share_deny))
2429                 goto out;
2430         /*
2431          * Lookup file; if found, lookup stateid and check open request,
2432          * and check for delegations in the process of being recalled.
2433          * If not found, create the nfs4_file struct
2434          */
2435         fp = find_file(ino);
2436         if (fp) {
2437                 if ((status = nfs4_check_open(fp, open, &stp)))
2438                         goto out;
2439                 status = nfs4_check_deleg(fp, open, &dp);
2440                 if (status)
2441                         goto out;
2442         } else {
2443                 status = nfserr_bad_stateid;
2444                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2445                         goto out;
2446                 status = nfserr_resource;
2447                 fp = alloc_init_file(ino);
2448                 if (fp == NULL)
2449                         goto out;
2450         }
2451
2452         /*
2453          * OPEN the file, or upgrade an existing OPEN.
2454          * If truncate fails, the OPEN fails.
2455          */
2456         if (stp) {
2457                 /* Stateid was found, this is an OPEN upgrade */
2458                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2459                 if (status)
2460                         goto out;
2461                 update_stateid(&stp->st_stateid);
2462         } else {
2463                 /* Stateid was not found, this is a new OPEN */
2464                 int flags = 0;
2465                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2466                         flags |= NFSD_MAY_READ;
2467                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2468                         flags |= NFSD_MAY_WRITE;
2469                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2470                 if (status)
2471                         goto out;
2472                 init_stateid(stp, fp, open);
2473                 status = nfsd4_truncate(rqstp, current_fh, open);
2474                 if (status) {
2475                         release_open_stateid(stp);
2476                         goto out;
2477                 }
2478                 if (nfsd4_has_session(&resp->cstate))
2479                         update_stateid(&stp->st_stateid);
2480         }
2481         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2482
2483         if (nfsd4_has_session(&resp->cstate)) {
2484                 open->op_stateowner->so_confirmed = 1;
2485                 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2486         }
2487
2488         /*
2489         * Attempt to hand out a delegation. No error return, because the
2490         * OPEN succeeds even if we fail.
2491         */
2492         nfs4_open_delegation(current_fh, open, stp);
2493
2494         status = nfs_ok;
2495
2496         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2497                 STATEID_VAL(&stp->st_stateid));
2498 out:
2499         if (fp)
2500                 put_nfs4_file(fp);
2501         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2502                 nfs4_set_claim_prev(open);
2503         /*
2504         * To finish the open response, we just need to set the rflags.
2505         */
2506         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2507         if (!open->op_stateowner->so_confirmed &&
2508             !nfsd4_has_session(&resp->cstate))
2509                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2510
2511         return status;
2512 }
2513
2514 __be32
2515 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2516             clientid_t *clid)
2517 {
2518         struct nfs4_client *clp;
2519         __be32 status;
2520
2521         nfs4_lock_state();
2522         dprintk("process_renew(%08x/%08x): starting\n", 
2523                         clid->cl_boot, clid->cl_id);
2524         status = nfserr_stale_clientid;
2525         if (STALE_CLIENTID(clid))
2526                 goto out;
2527         clp = find_confirmed_client(clid);
2528         status = nfserr_expired;
2529         if (clp == NULL) {
2530                 /* We assume the client took too long to RENEW. */
2531                 dprintk("nfsd4_renew: clientid not found!\n");
2532                 goto out;
2533         }
2534         renew_client(clp);
2535         status = nfserr_cb_path_down;
2536         if (!list_empty(&clp->cl_delegations)
2537                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2538                 goto out;
2539         status = nfs_ok;
2540 out:
2541         nfs4_unlock_state();
2542         return status;
2543 }
2544
2545 struct lock_manager nfsd4_manager = {
2546 };
2547
2548 static void
2549 nfsd4_end_grace(void)
2550 {
2551         dprintk("NFSD: end of grace period\n");
2552         nfsd4_recdir_purge_old();
2553         locks_end_grace(&nfsd4_manager);
2554 }
2555
2556 static time_t
2557 nfs4_laundromat(void)
2558 {
2559         struct nfs4_client *clp;
2560         struct nfs4_stateowner *sop;
2561         struct nfs4_delegation *dp;
2562         struct list_head *pos, *next, reaplist;
2563         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2564         time_t t, clientid_val = NFSD_LEASE_TIME;
2565         time_t u, test_val = NFSD_LEASE_TIME;
2566
2567         nfs4_lock_state();
2568
2569         dprintk("NFSD: laundromat service - starting\n");
2570         if (locks_in_grace())
2571                 nfsd4_end_grace();
2572         list_for_each_safe(pos, next, &client_lru) {
2573                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2574                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2575                         t = clp->cl_time - cutoff;
2576                         if (clientid_val > t)
2577                                 clientid_val = t;
2578                         break;
2579                 }
2580                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2581                         clp->cl_clientid.cl_id);
2582                 nfsd4_remove_clid_dir(clp);
2583                 expire_client(clp);
2584         }
2585         INIT_LIST_HEAD(&reaplist);
2586         spin_lock(&recall_lock);
2587         list_for_each_safe(pos, next, &del_recall_lru) {
2588                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2589                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2590                         u = dp->dl_time - cutoff;
2591                         if (test_val > u)
2592                                 test_val = u;
2593                         break;
2594                 }
2595                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2596                                     dp, dp->dl_flock);
2597                 list_move(&dp->dl_recall_lru, &reaplist);
2598         }
2599         spin_unlock(&recall_lock);
2600         list_for_each_safe(pos, next, &reaplist) {
2601                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2602                 list_del_init(&dp->dl_recall_lru);
2603                 unhash_delegation(dp);
2604         }
2605         test_val = NFSD_LEASE_TIME;
2606         list_for_each_safe(pos, next, &close_lru) {
2607                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2608                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2609                         u = sop->so_time - cutoff;
2610                         if (test_val > u)
2611                                 test_val = u;
2612                         break;
2613                 }
2614                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2615                         sop->so_id);
2616                 release_openowner(sop);
2617         }
2618         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2619                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2620         nfs4_unlock_state();
2621         return clientid_val;
2622 }
2623
2624 static struct workqueue_struct *laundry_wq;
2625 static void laundromat_main(struct work_struct *);
2626 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2627
2628 static void
2629 laundromat_main(struct work_struct *not_used)
2630 {
2631         time_t t;
2632
2633         t = nfs4_laundromat();
2634         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2635         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2636 }
2637
2638 static struct nfs4_stateowner *
2639 search_close_lru(u32 st_id, int flags)
2640 {
2641         struct nfs4_stateowner *local = NULL;
2642
2643         if (flags & CLOSE_STATE) {
2644                 list_for_each_entry(local, &close_lru, so_close_lru) {
2645                         if (local->so_id == st_id)
2646                                 return local;
2647                 }
2648         }
2649         return NULL;
2650 }
2651
2652 static inline int
2653 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2654 {
2655         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2656 }
2657
2658 static int
2659 STALE_STATEID(stateid_t *stateid)
2660 {
2661         if (time_after((unsigned long)boot_time,
2662                         (unsigned long)stateid->si_boot)) {
2663                 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2664                         STATEID_VAL(stateid));
2665                 return 1;
2666         }
2667         return 0;
2668 }
2669
2670 static int
2671 EXPIRED_STATEID(stateid_t *stateid)
2672 {
2673         if (time_before((unsigned long)boot_time,
2674                         ((unsigned long)stateid->si_boot)) &&
2675             time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2676                 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2677                         STATEID_VAL(stateid));
2678                 return 1;
2679         }
2680         return 0;
2681 }
2682
2683 static __be32
2684 stateid_error_map(stateid_t *stateid)
2685 {
2686         if (STALE_STATEID(stateid))
2687                 return nfserr_stale_stateid;
2688         if (EXPIRED_STATEID(stateid))
2689                 return nfserr_expired;
2690
2691         dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2692                 STATEID_VAL(stateid));
2693         return nfserr_bad_stateid;
2694 }
2695
2696 static inline int
2697 access_permit_read(unsigned long access_bmap)
2698 {
2699         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2700                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2701                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2702 }
2703
2704 static inline int
2705 access_permit_write(unsigned long access_bmap)
2706 {
2707         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2708                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2709 }
2710
2711 static
2712 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2713 {
2714         __be32 status = nfserr_openmode;
2715
2716         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2717                 goto out;
2718         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2719                 goto out;
2720         status = nfs_ok;
2721 out:
2722         return status;
2723 }
2724
2725 static inline __be32
2726 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2727 {
2728         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2729                 return nfs_ok;
2730         else if (locks_in_grace()) {
2731                 /* Answer in remaining cases depends on existance of
2732                  * conflicting state; so we must wait out the grace period. */
2733                 return nfserr_grace;
2734         } else if (flags & WR_STATE)
2735                 return nfs4_share_conflict(current_fh,
2736                                 NFS4_SHARE_DENY_WRITE);
2737         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2738                 return nfs4_share_conflict(current_fh,
2739                                 NFS4_SHARE_DENY_READ);
2740 }
2741
2742 /*
2743  * Allow READ/WRITE during grace period on recovered state only for files
2744  * that are not able to provide mandatory locking.
2745  */
2746 static inline int
2747 grace_disallows_io(struct inode *inode)
2748 {
2749         return locks_in_grace() && mandatory_lock(inode);
2750 }
2751
2752 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2753 {
2754         /*
2755          * When sessions are used the stateid generation number is ignored
2756          * when it is zero.
2757          */
2758         if ((flags & HAS_SESSION) && in->si_generation == 0)
2759                 goto out;
2760
2761         /* If the client sends us a stateid from the future, it's buggy: */
2762         if (in->si_generation > ref->si_generation)
2763                 return nfserr_bad_stateid;
2764         /*
2765          * The following, however, can happen.  For example, if the
2766          * client sends an open and some IO at the same time, the open
2767          * may bump si_generation while the IO is still in flight.
2768          * Thanks to hard links and renames, the client never knows what
2769          * file an open will affect.  So it could avoid that situation
2770          * only by serializing all opens and IO from the same open
2771          * owner.  To recover from the old_stateid error, the client
2772          * will just have to retry the IO:
2773          */
2774         if (in->si_generation < ref->si_generation)
2775                 return nfserr_old_stateid;
2776 out:
2777         return nfs_ok;
2778 }
2779
2780 static int is_delegation_stateid(stateid_t *stateid)
2781 {
2782         return stateid->si_fileid == 0;
2783 }
2784
2785 /*
2786 * Checks for stateid operations
2787 */
2788 __be32
2789 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2790                            stateid_t *stateid, int flags, struct file **filpp)
2791 {
2792         struct nfs4_stateid *stp = NULL;
2793         struct nfs4_delegation *dp = NULL;
2794         struct svc_fh *current_fh = &cstate->current_fh;
2795         struct inode *ino = current_fh->fh_dentry->d_inode;
2796         __be32 status;
2797
2798         if (filpp)
2799                 *filpp = NULL;
2800
2801         if (grace_disallows_io(ino))
2802                 return nfserr_grace;
2803
2804         if (nfsd4_has_session(cstate))
2805                 flags |= HAS_SESSION;
2806
2807         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2808                 return check_special_stateids(current_fh, stateid, flags);
2809
2810         status = nfserr_stale_stateid;
2811         if (STALE_STATEID(stateid)) 
2812                 goto out;
2813
2814         status = nfserr_bad_stateid;
2815         if (is_delegation_stateid(stateid)) {
2816                 dp = find_delegation_stateid(ino, stateid);
2817                 if (!dp) {
2818                         status = stateid_error_map(stateid);
2819                         goto out;
2820                 }
2821                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2822                                                   flags);
2823                 if (status)
2824                         goto out;
2825                 status = nfs4_check_delegmode(dp, flags);
2826                 if (status)
2827                         goto out;
2828                 renew_client(dp->dl_client);
2829                 if (filpp)
2830                         *filpp = dp->dl_vfs_file;
2831         } else { /* open or lock stateid */
2832                 stp = find_stateid(stateid, flags);
2833                 if (!stp) {
2834                         status = stateid_error_map(stateid);
2835                         goto out;
2836                 }
2837                 if (nfs4_check_fh(current_fh, stp))
2838                         goto out;
2839                 if (!stp->st_stateowner->so_confirmed)
2840                         goto out;
2841                 status = check_stateid_generation(stateid, &stp->st_stateid,
2842                                                   flags);
2843                 if (status)
2844                         goto out;
2845                 status = nfs4_check_openmode(stp, flags);
2846                 if (status)
2847                         goto out;
2848                 renew_client(stp->st_stateowner->so_client);
2849                 if (filpp)
2850                         *filpp = stp->st_vfs_file;
2851         }
2852         status = nfs_ok;
2853 out:
2854         return status;
2855 }
2856
2857 static inline int
2858 setlkflg (int type)
2859 {
2860         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2861                 RD_STATE : WR_STATE;
2862 }
2863
2864 /* 
2865  * Checks for sequence id mutating operations. 
2866  */
2867 static __be32
2868 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2869                          stateid_t *stateid, int flags,
2870                          struct nfs4_stateowner **sopp,
2871                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2872 {
2873         struct nfs4_stateid *stp;
2874         struct nfs4_stateowner *sop;
2875         struct svc_fh *current_fh = &cstate->current_fh;
2876         __be32 status;
2877
2878         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2879                 seqid, STATEID_VAL(stateid));
2880
2881         *stpp = NULL;
2882         *sopp = NULL;
2883
2884         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2885                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2886                 return nfserr_bad_stateid;
2887         }
2888
2889         if (STALE_STATEID(stateid))
2890                 return nfserr_stale_stateid;
2891
2892         if (nfsd4_has_session(cstate))
2893                 flags |= HAS_SESSION;
2894
2895         /*
2896         * We return BAD_STATEID if filehandle doesn't match stateid, 
2897         * the confirmed flag is incorrecly set, or the generation 
2898         * number is incorrect.  
2899         */
2900         stp = find_stateid(stateid, flags);
2901         if (stp == NULL) {
2902                 /*
2903                  * Also, we should make sure this isn't just the result of
2904                  * a replayed close:
2905                  */
2906                 sop = search_close_lru(stateid->si_stateownerid, flags);
2907                 if (sop == NULL)
2908                         return stateid_error_map(stateid);
2909                 *sopp = sop;
2910                 goto check_replay;
2911         }
2912
2913         *stpp = stp;
2914         *sopp = sop = stp->st_stateowner;
2915
2916         if (lock) {
2917                 clientid_t *lockclid = &lock->v.new.clientid;
2918                 struct nfs4_client *clp = sop->so_client;
2919                 int lkflg = 0;
2920                 __be32 status;
2921
2922                 lkflg = setlkflg(lock->lk_type);
2923
2924                 if (lock->lk_is_new) {
2925                         if (!sop->so_is_open_owner)
2926                                 return nfserr_bad_stateid;
2927                         if (!(flags & HAS_SESSION) &&
2928                             !same_clid(&clp->cl_clientid, lockclid))
2929                                 return nfserr_bad_stateid;
2930                         /* stp is the open stateid */
2931                         status = nfs4_check_openmode(stp, lkflg);
2932                         if (status)
2933                                 return status;
2934                 } else {
2935                         /* stp is the lock stateid */
2936                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2937                         if (status)
2938                                 return status;
2939                }
2940         }
2941
2942         if (nfs4_check_fh(current_fh, stp)) {
2943                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2944                 return nfserr_bad_stateid;
2945         }
2946
2947         /*
2948         *  We now validate the seqid and stateid generation numbers.
2949         *  For the moment, we ignore the possibility of 
2950         *  generation number wraparound.
2951         */
2952         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2953                 goto check_replay;
2954
2955         if (sop->so_confirmed && flags & CONFIRM) {
2956                 dprintk("NFSD: preprocess_seqid_op: expected"
2957                                 " unconfirmed stateowner!\n");
2958                 return nfserr_bad_stateid;
2959         }
2960         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2961                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2962                                 " confirmed yet!\n");
2963                 return nfserr_bad_stateid;
2964         }
2965         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2966         if (status)
2967                 return status;
2968         renew_client(sop->so_client);
2969         return nfs_ok;
2970
2971 check_replay:
2972         if (seqid == sop->so_seqid - 1) {
2973                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2974                 /* indicate replay to calling function */
2975                 return nfserr_replay_me;
2976         }
2977         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2978                         sop->so_seqid, seqid);
2979         *sopp = NULL;
2980         return nfserr_bad_seqid;
2981 }
2982
2983 __be32
2984 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2985                    struct nfsd4_open_confirm *oc)
2986 {
2987         __be32 status;
2988         struct nfs4_stateowner *sop;
2989         struct nfs4_stateid *stp;
2990
2991         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2992                         (int)cstate->current_fh.fh_dentry->d_name.len,
2993                         cstate->current_fh.fh_dentry->d_name.name);
2994
2995         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2996         if (status)
2997                 return status;
2998
2999         nfs4_lock_state();
3000
3001         if ((status = nfs4_preprocess_seqid_op(cstate,
3002                                         oc->oc_seqid, &oc->oc_req_stateid,
3003                                         CONFIRM | OPEN_STATE,
3004                                         &oc->oc_stateowner, &stp, NULL)))
3005                 goto out; 
3006
3007         sop = oc->oc_stateowner;
3008         sop->so_confirmed = 1;
3009         update_stateid(&stp->st_stateid);
3010         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3011         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3012                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3013
3014         nfsd4_create_clid_dir(sop->so_client);
3015 out:
3016         if (oc->oc_stateowner) {
3017                 nfs4_get_stateowner(oc->oc_stateowner);
3018                 cstate->replay_owner = oc->oc_stateowner;
3019         }
3020         nfs4_unlock_state();
3021         return status;
3022 }
3023
3024
3025 /*
3026  * unset all bits in union bitmap (bmap) that
3027  * do not exist in share (from successful OPEN_DOWNGRADE)
3028  */
3029 static void
3030 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3031 {
3032         int i;
3033         for (i = 1; i < 4; i++) {
3034                 if ((i & access) != i)
3035                         __clear_bit(i, bmap);
3036         }
3037 }
3038
3039 static void
3040 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3041 {
3042         int i;
3043         for (i = 0; i < 4; i++) {
3044                 if ((i & deny) != i)
3045                         __clear_bit(i, bmap);
3046         }
3047 }
3048
3049 __be32
3050 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3051                      struct nfsd4_compound_state *cstate,
3052                      struct nfsd4_open_downgrade *od)
3053 {
3054         __be32 status;
3055         struct nfs4_stateid *stp;
3056         unsigned int share_access;
3057
3058         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3059                         (int)cstate->current_fh.fh_dentry->d_name.len,
3060                         cstate->current_fh.fh_dentry->d_name.name);
3061
3062         if (!access_valid(od->od_share_access, cstate->minorversion)
3063                         || !deny_valid(od->od_share_deny))
3064                 return nfserr_inval;
3065
3066         nfs4_lock_state();
3067         if ((status = nfs4_preprocess_seqid_op(cstate,
3068                                         od->od_seqid,
3069                                         &od->od_stateid, 
3070                                         OPEN_STATE,
3071                                         &od->od_stateowner, &stp, NULL)))
3072                 goto out; 
3073
3074         status = nfserr_inval;
3075         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3076                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3077                         stp->st_access_bmap, od->od_share_access);
3078                 goto out;
3079         }
3080         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3081                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3082                         stp->st_deny_bmap, od->od_share_deny);
3083                 goto out;
3084         }
3085         set_access(&share_access, stp->st_access_bmap);
3086         nfs4_file_downgrade(stp->st_vfs_file,
3087                             share_access & ~od->od_share_access);
3088
3089         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3090         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3091
3092         update_stateid(&stp->st_stateid);
3093         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3094         status = nfs_ok;
3095 out:
3096         if (od->od_stateowner) {
3097                 nfs4_get_stateowner(od->od_stateowner);
3098                 cstate->replay_owner = od->od_stateowner;
3099         }
3100         nfs4_unlock_state();
3101         return status;
3102 }
3103
3104 /*
3105  * nfs4_unlock_state() called after encode
3106  */
3107 __be32
3108 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3109             struct nfsd4_close *close)
3110 {
3111         __be32 status;
3112         struct nfs4_stateid *stp;
3113
3114         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3115                         (int)cstate->current_fh.fh_dentry->d_name.len,
3116                         cstate->current_fh.fh_dentry->d_name.name);
3117
3118         nfs4_lock_state();
3119         /* check close_lru for replay */
3120         if ((status = nfs4_preprocess_seqid_op(cstate,
3121                                         close->cl_seqid,
3122                                         &close->cl_stateid, 
3123                                         OPEN_STATE | CLOSE_STATE,
3124                                         &close->cl_stateowner, &stp, NULL)))
3125                 goto out; 
3126         status = nfs_ok;
3127         update_stateid(&stp->st_stateid);
3128         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3129
3130         /* release_stateid() calls nfsd_close() if needed */
3131         release_open_stateid(stp);
3132
3133         /* place unused nfs4_stateowners on so_close_lru list to be
3134          * released by the laundromat service after the lease period
3135          * to enable us to handle CLOSE replay
3136          */
3137         if (list_empty(&close->cl_stateowner->so_stateids))
3138                 move_to_close_lru(close->cl_stateowner);
3139 out:
3140         if (close->cl_stateowner) {
3141                 nfs4_get_stateowner(close->cl_stateowner);
3142                 cstate->replay_owner = close->cl_stateowner;
3143         }
3144         nfs4_unlock_state();
3145         return status;
3146 }
3147
3148 __be32
3149 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3150                   struct nfsd4_delegreturn *dr)
3151 {
3152         struct nfs4_delegation *dp;
3153         stateid_t *stateid = &dr->dr_stateid;
3154         struct inode *inode;
3155         __be32 status;
3156         int flags = 0;
3157
3158         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3159                 return status;
3160         inode = cstate->current_fh.fh_dentry->d_inode;
3161
3162         if (nfsd4_has_session(cstate))
3163                 flags |= HAS_SESSION;
3164         nfs4_lock_state();
3165         status = nfserr_bad_stateid;
3166         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3167                 goto out;
3168         status = nfserr_stale_stateid;
3169         if (STALE_STATEID(stateid))
3170                 goto out;
3171         status = nfserr_bad_stateid;
3172         if (!is_delegation_stateid(stateid))
3173                 goto out;
3174         dp = find_delegation_stateid(inode, stateid);
3175         if (!dp) {
3176                 status = stateid_error_map(stateid);
3177                 goto out;
3178         }
3179         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3180         if (status)
3181                 goto out;
3182         renew_client(dp->dl_client);
3183
3184         unhash_delegation(dp);
3185 out:
3186         nfs4_unlock_state();
3187
3188         return status;
3189 }
3190
3191
3192 /* 
3193  * Lock owner state (byte-range locks)
3194  */
3195 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3196 #define LOCK_HASH_BITS              8
3197 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3198 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3199
3200 static inline u64
3201 end_offset(u64 start, u64 len)
3202 {
3203         u64 end;
3204
3205         end = start + len;
3206         return end >= start ? end: NFS4_MAX_UINT64;
3207 }
3208
3209 /* last octet in a range */
3210 static inline u64
3211 last_byte_offset(u64 start, u64 len)
3212 {
3213         u64 end;
3214
3215         BUG_ON(!len);
3216         end = start + len;
3217         return end > start ? end - 1: NFS4_MAX_UINT64;
3218 }
3219
3220 #define lockownerid_hashval(id) \
3221         ((id) & LOCK_HASH_MASK)
3222
3223 static inline unsigned int
3224 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3225                 struct xdr_netobj *ownername)
3226 {
3227         return (file_hashval(inode) + cl_id
3228                         + opaque_hashval(ownername->data, ownername->len))
3229                 & LOCK_HASH_MASK;
3230 }
3231
3232 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3233 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3234 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3235
3236 static struct nfs4_stateid *
3237 find_stateid(stateid_t *stid, int flags)
3238 {
3239         struct nfs4_stateid *local;
3240         u32 st_id = stid->si_stateownerid;
3241         u32 f_id = stid->si_fileid;
3242         unsigned int hashval;
3243
3244         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3245         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3246                 hashval = stateid_hashval(st_id, f_id);
3247                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3248                         if ((local->st_stateid.si_stateownerid == st_id) &&
3249                             (local->st_stateid.si_fileid == f_id))
3250                                 return local;
3251                 }
3252         } 
3253
3254         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3255                 hashval = stateid_hashval(st_id, f_id);
3256                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3257                         if ((local->st_stateid.si_stateownerid == st_id) &&
3258                             (local->st_stateid.si_fileid == f_id))
3259                                 return local;
3260                 }
3261         }
3262         return NULL;
3263 }
3264
3265 static struct nfs4_delegation *
3266 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3267 {
3268         struct nfs4_file *fp;
3269         struct nfs4_delegation *dl;
3270
3271         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3272                 STATEID_VAL(stid));
3273
3274         fp = find_file(ino);
3275         if (!fp)
3276                 return NULL;
3277         dl = find_delegation_file(fp, stid);
3278         put_nfs4_file(fp);
3279         return dl;
3280 }
3281
3282 /*
3283  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3284  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3285  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3286  * locking, this prevents us from being completely protocol-compliant.  The
3287  * real solution to this problem is to start using unsigned file offsets in
3288  * the VFS, but this is a very deep change!
3289  */
3290 static inline void
3291 nfs4_transform_lock_offset(struct file_lock *lock)
3292 {
3293         if (lock->fl_start < 0)
3294                 lock->fl_start = OFFSET_MAX;
3295         if (lock->fl_end < 0)
3296                 lock->fl_end = OFFSET_MAX;
3297 }
3298
3299 /* Hack!: For now, we're defining this just so we can use a pointer to it
3300  * as a unique cookie to identify our (NFSv4's) posix locks. */
3301 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3302 };
3303
3304 static inline void
3305 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3306 {
3307         struct nfs4_stateowner *sop;
3308         unsigned int hval;
3309
3310         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3311                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3312                 hval = lockownerid_hashval(sop->so_id);
3313                 kref_get(&sop->so_ref);
3314                 deny->ld_sop = sop;
3315                 deny->ld_clientid = sop->so_client->cl_clientid;
3316         } else {
3317                 deny->ld_sop = NULL;
3318                 deny->ld_clientid.cl_boot = 0;
3319                 deny->ld_clientid.cl_id = 0;
3320         }
3321         deny->ld_start = fl->fl_start;
3322         deny->ld_length = NFS4_MAX_UINT64;
3323         if (fl->fl_end != NFS4_MAX_UINT64)
3324                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3325         deny->ld_type = NFS4_READ_LT;
3326         if (fl->fl_type != F_RDLCK)
3327                 deny->ld_type = NFS4_WRITE_LT;
3328 }
3329
3330 static struct nfs4_stateowner *
3331 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3332                 struct xdr_netobj *owner)
3333 {
3334         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3335         struct nfs4_stateowner *op;
3336
3337         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3338                 if (same_owner_str(op, owner, clid))
3339                         return op;
3340         }
3341         return NULL;
3342 }
3343
3344 /*
3345  * Alloc a lock owner structure.
3346  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3347  * occured. 
3348  *
3349  * strhashval = lock_ownerstr_hashval 
3350  */
3351
3352 static struct nfs4_stateowner *
3353 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3354         struct nfs4_stateowner *sop;
3355         struct nfs4_replay *rp;
3356         unsigned int idhashval;
3357
3358         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3359                 return NULL;
3360         idhashval = lockownerid_hashval(current_ownerid);
3361         INIT_LIST_HEAD(&sop->so_idhash);
3362         INIT_LIST_HEAD(&sop->so_strhash);
3363         INIT_LIST_HEAD(&sop->so_perclient);
3364         INIT_LIST_HEAD(&sop->so_stateids);
3365         INIT_LIST_HEAD(&sop->so_perstateid);
3366         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3367         sop->so_time = 0;
3368         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3369         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3370         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3371         sop->so_is_open_owner = 0;
3372         sop->so_id = current_ownerid++;
3373         sop->so_client = clp;
3374         /* It is the openowner seqid that will be incremented in encode in the
3375          * case of new lockowners; so increment the lock seqid manually: */
3376         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3377         sop->so_confirmed = 1;
3378         rp = &sop->so_replay;
3379         rp->rp_status = nfserr_serverfault;
3380         rp->rp_buflen = 0;
3381         rp->rp_buf = rp->rp_ibuf;
3382         return sop;
3383 }
3384
3385 static struct nfs4_stateid *
3386 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3387 {
3388         struct nfs4_stateid *stp;
3389         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3390
3391         stp = nfs4_alloc_stateid();
3392         if (stp == NULL)
3393                 goto out;
3394         INIT_LIST_HEAD(&stp->st_hash);
3395         INIT_LIST_HEAD(&stp->st_perfile);
3396         INIT_LIST_HEAD(&stp->st_perstateowner);
3397         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3398         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3399         list_add(&stp->st_perfile, &fp->fi_stateids);
3400         list_add(&stp->st_perstateowner, &sop->so_stateids);
3401         stp->st_stateowner = sop;
3402         get_nfs4_file(fp);
3403         stp->st_file = fp;
3404         stp->st_stateid.si_boot = get_seconds();
3405         stp->st_stateid.si_stateownerid = sop->so_id;
3406         stp->st_stateid.si_fileid = fp->fi_id;
3407         stp->st_stateid.si_generation = 0;
3408         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3409         stp->st_access_bmap = open_stp->st_access_bmap;
3410         stp->st_deny_bmap = open_stp->st_deny_bmap;
3411         stp->st_openstp = open_stp;
3412
3413 out:
3414         return stp;
3415 }
3416
3417 static int
3418 check_lock_length(u64 offset, u64 length)
3419 {
3420         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3421              LOFF_OVERFLOW(offset, length)));
3422 }
3423
3424 /*
3425  *  LOCK operation 
3426  */
3427 __be32
3428 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3429            struct nfsd4_lock *lock)
3430 {
3431         struct nfs4_stateowner *open_sop = NULL;
3432         struct nfs4_stateowner *lock_sop = NULL;
3433         struct nfs4_stateid *lock_stp;
3434         struct file *filp;
3435         struct file_lock file_lock;
3436         struct file_lock conflock;
3437         __be32 status = 0;
3438         unsigned int strhashval;
3439         unsigned int cmd;
3440         int err;
3441
3442         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3443                 (long long) lock->lk_offset,
3444                 (long long) lock->lk_length);
3445
3446         if (check_lock_length(lock->lk_offset, lock->lk_length))
3447                  return nfserr_inval;
3448
3449         if ((status = fh_verify(rqstp, &cstate->current_fh,
3450                                 S_IFREG, NFSD_MAY_LOCK))) {
3451                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3452                 return status;
3453         }
3454
3455         nfs4_lock_state();
3456
3457         if (lock->lk_is_new) {
3458                 /*
3459                  * Client indicates that this is a new lockowner.
3460                  * Use open owner and open stateid to create lock owner and
3461                  * lock stateid.
3462                  */
3463                 struct nfs4_stateid *open_stp = NULL;
3464                 struct nfs4_file *fp;
3465                 
3466                 status = nfserr_stale_clientid;
3467                 if (!nfsd4_has_session(cstate) &&
3468                     STALE_CLIENTID(&lock->lk_new_clientid))
3469                         goto out;
3470
3471                 /* validate and update open stateid and open seqid */
3472                 status = nfs4_preprocess_seqid_op(cstate,
3473                                         lock->lk_new_open_seqid,
3474                                         &lock->lk_new_open_stateid,
3475                                         OPEN_STATE,
3476                                         &lock->lk_replay_owner, &open_stp,
3477                                         lock);
3478                 if (status)
3479                         goto out;
3480                 open_sop = lock->lk_replay_owner;
3481                 /* create lockowner and lock stateid */
3482                 fp = open_stp->st_file;
3483                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3484                                 open_sop->so_client->cl_clientid.cl_id, 
3485                                 &lock->v.new.owner);
3486                 /* XXX: Do we need to check for duplicate stateowners on
3487                  * the same file, or should they just be allowed (and
3488                  * create new stateids)? */
3489                 status = nfserr_resource;
3490                 lock_sop = alloc_init_lock_stateowner(strhashval,
3491                                 open_sop->so_client, open_stp, lock);
3492                 if (lock_sop == NULL)
3493                         goto out;
3494                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3495                 if (lock_stp == NULL)
3496                         goto out;
3497         } else {
3498                 /* lock (lock owner + lock stateid) already exists */
3499                 status = nfs4_preprocess_seqid_op(cstate,
3500                                        lock->lk_old_lock_seqid, 
3501                                        &lock->lk_old_lock_stateid, 
3502                                        LOCK_STATE,
3503                                        &lock->lk_replay_owner, &lock_stp, lock);
3504                 if (status)
3505                         goto out;
3506                 lock_sop = lock->lk_replay_owner;
3507         }
3508         /* lock->lk_replay_owner and lock_stp have been created or found */
3509         filp = lock_stp->st_vfs_file;
3510
3511         status = nfserr_grace;
3512         if (locks_in_grace() && !lock->lk_reclaim)
3513                 goto out;
3514         status = nfserr_no_grace;
3515         if (!locks_in_grace() && lock->lk_reclaim)
3516                 goto out;
3517
3518         locks_init_lock(&file_lock);
3519         switch (lock->lk_type) {
3520                 case NFS4_READ_LT:
3521                 case NFS4_READW_LT:
3522                         file_lock.fl_type = F_RDLCK;
3523                         cmd = F_SETLK;
3524                 break;
3525                 case NFS4_WRITE_LT:
3526                 case NFS4_WRITEW_LT:
3527                         file_lock.fl_type = F_WRLCK;
3528                         cmd = F_SETLK;
3529                 break;
3530                 default:
3531                         status = nfserr_inval;
3532                 goto out;
3533         }
3534         file_lock.fl_owner = (fl_owner_t)lock_sop;
3535         file_lock.fl_pid = current->tgid;
3536         file_lock.fl_file = filp;
3537         file_lock.fl_flags = FL_POSIX;
3538         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3539
3540         file_lock.fl_start = lock->lk_offset;
3541         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3542         nfs4_transform_lock_offset(&file_lock);
3543
3544         /*
3545         * Try to lock the file in the VFS.
3546         * Note: locks.c uses the BKL to protect the inode's lock list.
3547         */
3548
3549         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3550         switch (-err) {
3551         case 0: /* success! */
3552                 update_stateid(&lock_stp->st_stateid);
3553                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3554                                 sizeof(stateid_t));
3555                 status = 0;
3556                 break;
3557         case (EAGAIN):          /* conflock holds conflicting lock */
3558                 status = nfserr_denied;
3559                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3560                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3561                 break;
3562         case (EDEADLK):
3563                 status = nfserr_deadlock;
3564                 break;
3565         default:        
3566                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3567                 status = nfserr_resource;
3568                 break;
3569         }
3570 out:
3571         if (status && lock->lk_is_new && lock_sop)
3572                 release_lockowner(lock_sop);
3573         if (lock->lk_replay_owner) {
3574                 nfs4_get_stateowner(lock->lk_replay_owner);
3575                 cstate->replay_owner = lock->lk_replay_owner;
3576         }
3577         nfs4_unlock_state();
3578         return status;
3579 }
3580
3581 /*
3582  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3583  * so we do a temporary open here just to get an open file to pass to
3584  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3585  * inode operation.)
3586  */
3587 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3588 {
3589         struct file *file;
3590         int err;
3591
3592         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3593         if (err)
3594                 return err;
3595         err = vfs_test_lock(file, lock);
3596         nfsd_close(file);
3597         return err;
3598 }
3599
3600 /*
3601  * LOCKT operation
3602  */
3603 __be32
3604 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3605             struct nfsd4_lockt *lockt)
3606 {
3607         struct inode *inode;
3608         struct file_lock file_lock;
3609         int error;
3610         __be32 status;
3611
3612         if (locks_in_grace())
3613                 return nfserr_grace;
3614
3615         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3616                  return nfserr_inval;
3617
3618         lockt->lt_stateowner = NULL;
3619         nfs4_lock_state();
3620
3621         status = nfserr_stale_clientid;
3622         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3623                 goto out;
3624
3625         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3626                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3627                 if (status == nfserr_symlink)
3628                         status = nfserr_inval;
3629                 goto out;
3630         }
3631
3632         inode = cstate->current_fh.fh_dentry->d_inode;
3633         locks_init_lock(&file_lock);
3634         switch (lockt->lt_type) {
3635                 case NFS4_READ_LT:
3636                 case NFS4_READW_LT:
3637                         file_lock.fl_type = F_RDLCK;
3638                 break;
3639                 case NFS4_WRITE_LT:
3640                 case NFS4_WRITEW_LT:
3641                         file_lock.fl_type = F_WRLCK;
3642                 break;
3643                 default:
3644                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3645                         status = nfserr_inval;
3646                 goto out;
3647         }
3648
3649         lockt->lt_stateowner = find_lockstateowner_str(inode,
3650                         &lockt->lt_clientid, &lockt->lt_owner);
3651         if (lockt->lt_stateowner)
3652                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3653         file_lock.fl_pid = current->tgid;
3654         file_lock.fl_flags = FL_POSIX;
3655
3656         file_lock.fl_start = lockt->lt_offset;
3657         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3658
3659         nfs4_transform_lock_offset(&file_lock);
3660
3661         status = nfs_ok;
3662         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3663         if (error) {
3664                 status = nfserrno(error);
3665                 goto out;
3666         }
3667         if (file_lock.fl_type != F_UNLCK) {
3668                 status = nfserr_denied;
3669                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3670         }
3671 out:
3672         nfs4_unlock_state();
3673         return status;
3674 }
3675
3676 __be32
3677 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3678             struct nfsd4_locku *locku)
3679 {
3680         struct nfs4_stateid *stp;
3681         struct file *filp = NULL;
3682         struct file_lock file_lock;
3683         __be32 status;
3684         int err;
3685                                                         
3686         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3687                 (long long) locku->lu_offset,
3688                 (long long) locku->lu_length);
3689
3690         if (check_lock_length(locku->lu_offset, locku->lu_length))
3691                  return nfserr_inval;
3692
3693         nfs4_lock_state();
3694                                                                                 
3695         if ((status = nfs4_preprocess_seqid_op(cstate,
3696                                         locku->lu_seqid, 
3697                                         &locku->lu_stateid, 
3698                                         LOCK_STATE,
3699                                         &locku->lu_stateowner, &stp, NULL)))
3700                 goto out;
3701
3702         filp = stp->st_vfs_file;
3703         BUG_ON(!filp);
3704         locks_init_lock(&file_lock);
3705         file_lock.fl_type = F_UNLCK;
3706         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3707         file_lock.fl_pid = current->tgid;
3708         file_lock.fl_file = filp;
3709         file_lock.fl_flags = FL_POSIX; 
3710         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3711         file_lock.fl_start = locku->lu_offset;
3712
3713         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3714         nfs4_transform_lock_offset(&file_lock);
3715
3716         /*
3717         *  Try to unlock the file in the VFS.
3718         */
3719         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3720         if (err) {
3721                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3722                 goto out_nfserr;
3723         }
3724         /*
3725         * OK, unlock succeeded; the only thing left to do is update the stateid.
3726         */
3727         update_stateid(&stp->st_stateid);
3728         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3729
3730 out:
3731         if (locku->lu_stateowner) {
3732                 nfs4_get_stateowner(locku->lu_stateowner);
3733                 cstate->replay_owner = locku->lu_stateowner;
3734         }
3735         nfs4_unlock_state();
3736         return status;
3737
3738 out_nfserr:
3739         status = nfserrno(err);
3740         goto out;
3741 }
3742
3743 /*
3744  * returns
3745  *      1: locks held by lockowner
3746  *      0: no locks held by lockowner
3747  */
3748 static int
3749 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3750 {
3751         struct file_lock **flpp;
3752         struct inode *inode = filp->f_path.dentry->d_inode;
3753         int status = 0;
3754
3755         lock_kernel();
3756         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3757                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3758                         status = 1;
3759                         goto out;
3760                 }
3761         }
3762 out:
3763         unlock_kernel();
3764         return status;
3765 }
3766
3767 __be32
3768 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3769                         struct nfsd4_compound_state *cstate,
3770                         struct nfsd4_release_lockowner *rlockowner)
3771 {
3772         clientid_t *clid = &rlockowner->rl_clientid;
3773         struct nfs4_stateowner *sop;
3774         struct nfs4_stateid *stp;
3775         struct xdr_netobj *owner = &rlockowner->rl_owner;
3776         struct list_head matches;
3777         int i;
3778         __be32 status;
3779
3780         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3781                 clid->cl_boot, clid->cl_id);
3782
3783         /* XXX check for lease expiration */
3784
3785         status = nfserr_stale_clientid;
3786         if (STALE_CLIENTID(clid))
3787                 return status;
3788
3789         nfs4_lock_state();
3790
3791         status = nfserr_locks_held;
3792         /* XXX: we're doing a linear search through all the lockowners.
3793          * Yipes!  For now we'll just hope clients aren't really using
3794          * release_lockowner much, but eventually we have to fix these
3795          * data structures. */
3796         INIT_LIST_HEAD(&matches);
3797         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3798                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3799                         if (!same_owner_str(sop, owner, clid))
3800                                 continue;
3801                         list_for_each_entry(stp, &sop->so_stateids,
3802                                         st_perstateowner) {
3803                                 if (check_for_locks(stp->st_vfs_file, sop))
3804                                         goto out;
3805                                 /* Note: so_perclient unused for lockowners,
3806                                  * so it's OK to fool with here. */
3807                                 list_add(&sop->so_perclient, &matches);
3808                         }
3809                 }
3810         }
3811         /* Clients probably won't expect us to return with some (but not all)
3812          * of the lockowner state released; so don't release any until all
3813          * have been checked. */
3814         status = nfs_ok;
3815         while (!list_empty(&matches)) {
3816                 sop = list_entry(matches.next, struct nfs4_stateowner,
3817                                                                 so_perclient);
3818                 /* unhash_stateowner deletes so_perclient only
3819                  * for openowners. */
3820                 list_del(&sop->so_perclient);
3821                 release_lockowner(sop);
3822         }
3823 out:
3824         nfs4_unlock_state();
3825         return status;
3826 }
3827
3828 static inline struct nfs4_client_reclaim *
3829 alloc_reclaim(void)
3830 {
3831         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3832 }
3833
3834 int
3835 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3836 {
3837         unsigned int strhashval = clientstr_hashval(name);
3838         struct nfs4_client *clp;
3839
3840         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3841         return clp ? 1 : 0;
3842 }
3843
3844 /*
3845  * failure => all reset bets are off, nfserr_no_grace...
3846  */
3847 int
3848 nfs4_client_to_reclaim(const char *name)
3849 {
3850         unsigned int strhashval;
3851         struct nfs4_client_reclaim *crp = NULL;
3852
3853         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3854         crp = alloc_reclaim();
3855         if (!crp)
3856                 return 0;
3857         strhashval = clientstr_hashval(name);
3858         INIT_LIST_HEAD(&crp->cr_strhash);
3859         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3860         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3861         reclaim_str_hashtbl_size++;
3862         return 1;
3863 }
3864
3865 static void
3866 nfs4_release_reclaim(void)
3867 {
3868         struct nfs4_client_reclaim *crp = NULL;
3869         int i;
3870
3871         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3872                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3873                         crp = list_entry(reclaim_str_hashtbl[i].next,
3874                                         struct nfs4_client_reclaim, cr_strhash);
3875                         list_del(&crp->cr_strhash);
3876                         kfree(crp);
3877                         reclaim_str_hashtbl_size--;
3878                 }
3879         }
3880         BUG_ON(reclaim_str_hashtbl_size);
3881 }
3882
3883 /*
3884  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3885 static struct nfs4_client_reclaim *
3886 nfs4_find_reclaim_client(clientid_t *clid)
3887 {
3888         unsigned int strhashval;
3889         struct nfs4_client *clp;
3890         struct nfs4_client_reclaim *crp = NULL;
3891
3892
3893         /* find clientid in conf_id_hashtbl */
3894         clp = find_confirmed_client(clid);
3895         if (clp == NULL)
3896                 return NULL;
3897
3898         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3899                             clp->cl_name.len, clp->cl_name.data,
3900                             clp->cl_recdir);
3901
3902         /* find clp->cl_name in reclaim_str_hashtbl */
3903         strhashval = clientstr_hashval(clp->cl_recdir);
3904         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3905                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3906                         return crp;
3907                 }
3908         }
3909         return NULL;
3910 }
3911
3912 /*
3913 * Called from OPEN. Look for clientid in reclaim list.
3914 */
3915 __be32
3916 nfs4_check_open_reclaim(clientid_t *clid)
3917 {
3918         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3919 }
3920
3921 /* initialization to perform at module load time: */
3922
3923 int
3924 nfs4_state_init(void)
3925 {
3926         int i, status;
3927
3928         status = nfsd4_init_slabs();
3929         if (status)
3930                 return status;
3931         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3932                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3933                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3934                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3935                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3936                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3937         }
3938         for (i = 0; i < SESSION_HASH_SIZE; i++)
3939                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3940         for (i = 0; i < FILE_HASH_SIZE; i++) {
3941                 INIT_LIST_HEAD(&file_hashtbl[i]);
3942         }
3943         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3944                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3945                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3946         }
3947         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3948                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3949                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3950         }
3951         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3952                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3953                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3954         }
3955         memset(&onestateid, ~0, sizeof(stateid_t));
3956         INIT_LIST_HEAD(&close_lru);
3957         INIT_LIST_HEAD(&client_lru);
3958         INIT_LIST_HEAD(&del_recall_lru);
3959         reclaim_str_hashtbl_size = 0;
3960         return 0;
3961 }
3962
3963 static void
3964 nfsd4_load_reboot_recovery_data(void)
3965 {
3966         int status;
3967
3968         nfs4_lock_state();
3969         nfsd4_init_recdir(user_recovery_dirname);
3970         status = nfsd4_recdir_load();
3971         nfs4_unlock_state();
3972         if (status)
3973                 printk("NFSD: Failure reading reboot recovery data\n");
3974 }
3975
3976 unsigned long
3977 get_nfs4_grace_period(void)
3978 {
3979         return max(user_lease_time, lease_time) * HZ;
3980 }
3981
3982 /*
3983  * Since the lifetime of a delegation isn't limited to that of an open, a
3984  * client may quite reasonably hang on to a delegation as long as it has
3985  * the inode cached.  This becomes an obvious problem the first time a
3986  * client's inode cache approaches the size of the server's total memory.
3987  *
3988  * For now we avoid this problem by imposing a hard limit on the number
3989  * of delegations, which varies according to the server's memory size.
3990  */
3991 static void
3992 set_max_delegations(void)
3993 {
3994         /*
3995          * Allow at most 4 delegations per megabyte of RAM.  Quick
3996          * estimates suggest that in the worst case (where every delegation
3997          * is for a different inode), a delegation could take about 1.5K,
3998          * giving a worst case usage of about 6% of memory.
3999          */
4000         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4001 }
4002
4003 /* initialization to perform when the nfsd service is started: */
4004
4005 static int
4006 __nfs4_state_start(void)
4007 {
4008         unsigned long grace_time;
4009
4010         boot_time = get_seconds();
4011         grace_time = get_nfs4_grace_period();
4012         lease_time = user_lease_time;
4013         locks_start_grace(&nfsd4_manager);
4014         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4015                grace_time/HZ);
4016         laundry_wq = create_singlethread_workqueue("nfsd4");
4017         if (laundry_wq == NULL)
4018                 return -ENOMEM;
4019         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4020         set_max_delegations();
4021         return set_callback_cred();
4022 }
4023
4024 int
4025 nfs4_state_start(void)
4026 {
4027         int ret;
4028
4029         if (nfs4_init)
4030                 return 0;
4031         nfsd4_load_reboot_recovery_data();
4032         ret = __nfs4_state_start();
4033         if (ret)
4034                 return ret;
4035         nfs4_init = 1;
4036         return 0;
4037 }
4038
4039 time_t
4040 nfs4_lease_time(void)
4041 {
4042         return lease_time;
4043 }
4044
4045 static void
4046 __nfs4_state_shutdown(void)
4047 {
4048         int i;
4049         struct nfs4_client *clp = NULL;
4050         struct nfs4_delegation *dp = NULL;
4051         struct list_head *pos, *next, reaplist;
4052
4053         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4054                 while (!list_empty(&conf_id_hashtbl[i])) {
4055                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4056                         expire_client(clp);
4057                 }
4058                 while (!list_empty(&unconf_str_hashtbl[i])) {
4059                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4060                         expire_client(clp);
4061                 }
4062         }
4063         INIT_LIST_HEAD(&reaplist);
4064         spin_lock(&recall_lock);
4065         list_for_each_safe(pos, next, &del_recall_lru) {
4066                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4067                 list_move(&dp->dl_recall_lru, &reaplist);
4068         }
4069         spin_unlock(&recall_lock);
4070         list_for_each_safe(pos, next, &reaplist) {
4071                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4072                 list_del_init(&dp->dl_recall_lru);
4073                 unhash_delegation(dp);
4074         }
4075
4076         nfsd4_shutdown_recdir();
4077         nfs4_init = 0;
4078 }
4079
4080 void
4081 nfs4_state_shutdown(void)
4082 {
4083         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4084         destroy_workqueue(laundry_wq);
4085         locks_end_grace(&nfsd4_manager);
4086         nfs4_lock_state();
4087         nfs4_release_reclaim();
4088         __nfs4_state_shutdown();
4089         nfs4_unlock_state();
4090 }
4091
4092 /*
4093  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4094  * accessed when nfsd is starting.
4095  */
4096 static void
4097 nfs4_set_recdir(char *recdir)
4098 {
4099         strcpy(user_recovery_dirname, recdir);
4100 }
4101
4102 /*
4103  * Change the NFSv4 recovery directory to recdir.
4104  */
4105 int
4106 nfs4_reset_recoverydir(char *recdir)
4107 {
4108         int status;
4109         struct path path;
4110
4111         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4112         if (status)
4113                 return status;
4114         status = -ENOTDIR;
4115         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4116                 nfs4_set_recdir(recdir);
4117                 status = 0;
4118         }
4119         path_put(&path);
4120         return status;
4121 }
4122
4123 char *
4124 nfs4_recoverydir(void)
4125 {
4126         return user_recovery_dirname;
4127 }
4128
4129 /*
4130  * Called when leasetime is changed.
4131  *
4132  * The only way the protocol gives us to handle on-the-fly lease changes is to
4133  * simulate a reboot.  Instead of doing that, we just wait till the next time
4134  * we start to register any changes in lease time.  If the administrator
4135  * really wants to change the lease time *now*, they can go ahead and bring
4136  * nfsd down and then back up again after changing the lease time.
4137  *
4138  * user_lease_time is protected by nfsd_mutex since it's only really accessed
4139  * when nfsd is starting
4140  */
4141 void
4142 nfs4_reset_lease(time_t leasetime)
4143 {
4144         user_lease_time = leasetime;
4145 }