include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit...
[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/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
42 #include "xdr4.h"
43 #include "vfs.h"
44
45 #define NFSDDBG_FACILITY                NFSDDBG_PROC
46
47 /* Globals */
48 static time_t lease_time = 90;     /* default lease time */
49 static time_t user_lease_time = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static u32 nfs4_init;
55 static stateid_t zerostateid;             /* bits all 0 */
56 static stateid_t onestateid;              /* bits all 1 */
57 static u64 current_sessionid = 1;
58
59 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
60 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
61
62 /* forward declarations */
63 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
64 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
65 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
66 static void nfs4_set_recdir(char *recdir);
67
68 /* Locking: */
69
70 /* Currently used for almost all code touching nfsv4 state: */
71 static DEFINE_MUTEX(client_mutex);
72
73 /*
74  * Currently used for the del_recall_lru and file hash table.  In an
75  * effort to decrease the scope of the client_mutex, this spinlock may
76  * eventually cover more:
77  */
78 static DEFINE_SPINLOCK(recall_lock);
79
80 static struct kmem_cache *stateowner_slab = NULL;
81 static struct kmem_cache *file_slab = NULL;
82 static struct kmem_cache *stateid_slab = NULL;
83 static struct kmem_cache *deleg_slab = NULL;
84
85 void
86 nfs4_lock_state(void)
87 {
88         mutex_lock(&client_mutex);
89 }
90
91 void
92 nfs4_unlock_state(void)
93 {
94         mutex_unlock(&client_mutex);
95 }
96
97 static inline u32
98 opaque_hashval(const void *ptr, int nbytes)
99 {
100         unsigned char *cptr = (unsigned char *) ptr;
101
102         u32 x = 0;
103         while (nbytes--) {
104                 x *= 37;
105                 x += *cptr++;
106         }
107         return x;
108 }
109
110 static struct list_head del_recall_lru;
111
112 static inline void
113 put_nfs4_file(struct nfs4_file *fi)
114 {
115         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
116                 list_del(&fi->fi_hash);
117                 spin_unlock(&recall_lock);
118                 iput(fi->fi_inode);
119                 kmem_cache_free(file_slab, fi);
120         }
121 }
122
123 static inline void
124 get_nfs4_file(struct nfs4_file *fi)
125 {
126         atomic_inc(&fi->fi_ref);
127 }
128
129 static int num_delegations;
130 unsigned int max_delegations;
131
132 /*
133  * Open owner state (share locks)
134  */
135
136 /* hash tables for nfs4_stateowner */
137 #define OWNER_HASH_BITS              8
138 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
139 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
140
141 #define ownerid_hashval(id) \
142         ((id) & OWNER_HASH_MASK)
143 #define ownerstr_hashval(clientid, ownername) \
144         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
145
146 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
147 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
148
149 /* hash table for nfs4_file */
150 #define FILE_HASH_BITS                   8
151 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
152 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
153 /* hash table for (open)nfs4_stateid */
154 #define STATEID_HASH_BITS              10
155 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
156 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
157
158 #define file_hashval(x) \
159         hash_ptr(x, FILE_HASH_BITS)
160 #define stateid_hashval(owner_id, file_id)  \
161         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
162
163 static struct list_head file_hashtbl[FILE_HASH_SIZE];
164 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
165
166 static struct nfs4_delegation *
167 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
168 {
169         struct nfs4_delegation *dp;
170         struct nfs4_file *fp = stp->st_file;
171         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
172
173         dprintk("NFSD alloc_init_deleg\n");
174         if (fp->fi_had_conflict)
175                 return NULL;
176         if (num_delegations > max_delegations)
177                 return NULL;
178         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
179         if (dp == NULL)
180                 return dp;
181         num_delegations++;
182         INIT_LIST_HEAD(&dp->dl_perfile);
183         INIT_LIST_HEAD(&dp->dl_perclnt);
184         INIT_LIST_HEAD(&dp->dl_recall_lru);
185         dp->dl_client = clp;
186         get_nfs4_file(fp);
187         dp->dl_file = fp;
188         dp->dl_flock = NULL;
189         get_file(stp->st_vfs_file);
190         dp->dl_vfs_file = stp->st_vfs_file;
191         dp->dl_type = type;
192         dp->dl_ident = cb->cb_ident;
193         dp->dl_stateid.si_boot = get_seconds();
194         dp->dl_stateid.si_stateownerid = current_delegid++;
195         dp->dl_stateid.si_fileid = 0;
196         dp->dl_stateid.si_generation = 0;
197         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
198         dp->dl_time = 0;
199         atomic_set(&dp->dl_count, 1);
200         list_add(&dp->dl_perfile, &fp->fi_delegations);
201         list_add(&dp->dl_perclnt, &clp->cl_delegations);
202         return dp;
203 }
204
205 void
206 nfs4_put_delegation(struct nfs4_delegation *dp)
207 {
208         if (atomic_dec_and_test(&dp->dl_count)) {
209                 dprintk("NFSD: freeing dp %p\n",dp);
210                 put_nfs4_file(dp->dl_file);
211                 kmem_cache_free(deleg_slab, dp);
212                 num_delegations--;
213         }
214 }
215
216 /* Remove the associated file_lock first, then remove the delegation.
217  * lease_modify() is called to remove the FS_LEASE file_lock from
218  * the i_flock list, eventually calling nfsd's lock_manager
219  * fl_release_callback.
220  */
221 static void
222 nfs4_close_delegation(struct nfs4_delegation *dp)
223 {
224         struct file *filp = dp->dl_vfs_file;
225
226         dprintk("NFSD: close_delegation dp %p\n",dp);
227         dp->dl_vfs_file = NULL;
228         /* The following nfsd_close may not actually close the file,
229          * but we want to remove the lease in any case. */
230         if (dp->dl_flock)
231                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
232         nfsd_close(filp);
233 }
234
235 /* Called under the state lock. */
236 static void
237 unhash_delegation(struct nfs4_delegation *dp)
238 {
239         list_del_init(&dp->dl_perfile);
240         list_del_init(&dp->dl_perclnt);
241         spin_lock(&recall_lock);
242         list_del_init(&dp->dl_recall_lru);
243         spin_unlock(&recall_lock);
244         nfs4_close_delegation(dp);
245         nfs4_put_delegation(dp);
246 }
247
248 /* 
249  * SETCLIENTID state 
250  */
251
252 /* Hash tables for nfs4_clientid state */
253 #define CLIENT_HASH_BITS                 4
254 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
255 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
256
257 #define clientid_hashval(id) \
258         ((id) & CLIENT_HASH_MASK)
259 #define clientstr_hashval(name) \
260         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
261 /*
262  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
263  * used in reboot/reset lease grace period processing
264  *
265  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
266  * setclientid_confirmed info. 
267  *
268  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
269  * setclientid info.
270  *
271  * client_lru holds client queue ordered by nfs4_client.cl_time
272  * for lease renewal.
273  *
274  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
275  * for last close replay.
276  */
277 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
278 static int reclaim_str_hashtbl_size = 0;
279 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
280 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
281 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
282 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
283 static struct list_head client_lru;
284 static struct list_head close_lru;
285
286 static void unhash_generic_stateid(struct nfs4_stateid *stp)
287 {
288         list_del(&stp->st_hash);
289         list_del(&stp->st_perfile);
290         list_del(&stp->st_perstateowner);
291 }
292
293 static void free_generic_stateid(struct nfs4_stateid *stp)
294 {
295         put_nfs4_file(stp->st_file);
296         kmem_cache_free(stateid_slab, stp);
297 }
298
299 static void release_lock_stateid(struct nfs4_stateid *stp)
300 {
301         unhash_generic_stateid(stp);
302         locks_remove_posix(stp->st_vfs_file, (fl_owner_t)stp->st_stateowner);
303         free_generic_stateid(stp);
304 }
305
306 static void unhash_lockowner(struct nfs4_stateowner *sop)
307 {
308         struct nfs4_stateid *stp;
309
310         list_del(&sop->so_idhash);
311         list_del(&sop->so_strhash);
312         list_del(&sop->so_perstateid);
313         while (!list_empty(&sop->so_stateids)) {
314                 stp = list_first_entry(&sop->so_stateids,
315                                 struct nfs4_stateid, st_perstateowner);
316                 release_lock_stateid(stp);
317         }
318 }
319
320 static void release_lockowner(struct nfs4_stateowner *sop)
321 {
322         unhash_lockowner(sop);
323         nfs4_put_stateowner(sop);
324 }
325
326 static void
327 release_stateid_lockowners(struct nfs4_stateid *open_stp)
328 {
329         struct nfs4_stateowner *lock_sop;
330
331         while (!list_empty(&open_stp->st_lockowners)) {
332                 lock_sop = list_entry(open_stp->st_lockowners.next,
333                                 struct nfs4_stateowner, so_perstateid);
334                 /* list_del(&open_stp->st_lockowners);  */
335                 BUG_ON(lock_sop->so_is_open_owner);
336                 release_lockowner(lock_sop);
337         }
338 }
339
340 static void release_open_stateid(struct nfs4_stateid *stp)
341 {
342         unhash_generic_stateid(stp);
343         release_stateid_lockowners(stp);
344         nfsd_close(stp->st_vfs_file);
345         free_generic_stateid(stp);
346 }
347
348 static void unhash_openowner(struct nfs4_stateowner *sop)
349 {
350         struct nfs4_stateid *stp;
351
352         list_del(&sop->so_idhash);
353         list_del(&sop->so_strhash);
354         list_del(&sop->so_perclient);
355         list_del(&sop->so_perstateid); /* XXX: necessary? */
356         while (!list_empty(&sop->so_stateids)) {
357                 stp = list_first_entry(&sop->so_stateids,
358                                 struct nfs4_stateid, st_perstateowner);
359                 release_open_stateid(stp);
360         }
361 }
362
363 static void release_openowner(struct nfs4_stateowner *sop)
364 {
365         unhash_openowner(sop);
366         list_del(&sop->so_close_lru);
367         nfs4_put_stateowner(sop);
368 }
369
370 static DEFINE_SPINLOCK(sessionid_lock);
371 #define SESSION_HASH_SIZE       512
372 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
373
374 static inline int
375 hash_sessionid(struct nfs4_sessionid *sessionid)
376 {
377         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
378
379         return sid->sequence % SESSION_HASH_SIZE;
380 }
381
382 static inline void
383 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
384 {
385         u32 *ptr = (u32 *)(&sessionid->data[0]);
386         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
387 }
388
389 static void
390 gen_sessionid(struct nfsd4_session *ses)
391 {
392         struct nfs4_client *clp = ses->se_client;
393         struct nfsd4_sessionid *sid;
394
395         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
396         sid->clientid = clp->cl_clientid;
397         sid->sequence = current_sessionid++;
398         sid->reserved = 0;
399 }
400
401 /*
402  * The protocol defines ca_maxresponssize_cached to include the size of
403  * the rpc header, but all we need to cache is the data starting after
404  * the end of the initial SEQUENCE operation--the rest we regenerate
405  * each time.  Therefore we can advertise a ca_maxresponssize_cached
406  * value that is the number of bytes in our cache plus a few additional
407  * bytes.  In order to stay on the safe side, and not promise more than
408  * we can cache, those additional bytes must be the minimum possible: 24
409  * bytes of rpc header (xid through accept state, with AUTH_NULL
410  * verifier), 12 for the compound header (with zero-length tag), and 44
411  * for the SEQUENCE op response:
412  */
413 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
414
415 /*
416  * Give the client the number of ca_maxresponsesize_cached slots it
417  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
418  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
419  * than NFSD_MAX_SLOTS_PER_SESSION.
420  *
421  * If we run out of reserved DRC memory we should (up to a point)
422  * re-negotiate active sessions and reduce their slot usage to make
423  * rooom for new connections. For now we just fail the create session.
424  */
425 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
426 {
427         int mem, size = fchan->maxresp_cached;
428
429         if (fchan->maxreqs < 1)
430                 return nfserr_inval;
431
432         if (size < NFSD_MIN_HDR_SEQ_SZ)
433                 size = NFSD_MIN_HDR_SEQ_SZ;
434         size -= NFSD_MIN_HDR_SEQ_SZ;
435         if (size > NFSD_SLOT_CACHE_SIZE)
436                 size = NFSD_SLOT_CACHE_SIZE;
437
438         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
439         mem = fchan->maxreqs * size;
440         if (mem > NFSD_MAX_MEM_PER_SESSION) {
441                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
442                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
443                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
444                 mem = fchan->maxreqs * size;
445         }
446
447         spin_lock(&nfsd_drc_lock);
448         /* bound the total session drc memory ussage */
449         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
450                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
451                 mem = fchan->maxreqs * size;
452         }
453         nfsd_drc_mem_used += mem;
454         spin_unlock(&nfsd_drc_lock);
455
456         if (fchan->maxreqs == 0)
457                 return nfserr_serverfault;
458
459         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
460         return 0;
461 }
462
463 /*
464  * fchan holds the client values on input, and the server values on output
465  * sv_max_mesg is the maximum payload plus one page for overhead.
466  */
467 static int init_forechannel_attrs(struct svc_rqst *rqstp,
468                                   struct nfsd4_channel_attrs *session_fchan,
469                                   struct nfsd4_channel_attrs *fchan)
470 {
471         int status = 0;
472         __u32   maxcount = nfsd_serv->sv_max_mesg;
473
474         /* headerpadsz set to zero in encode routine */
475
476         /* Use the client's max request and max response size if possible */
477         if (fchan->maxreq_sz > maxcount)
478                 fchan->maxreq_sz = maxcount;
479         session_fchan->maxreq_sz = fchan->maxreq_sz;
480
481         if (fchan->maxresp_sz > maxcount)
482                 fchan->maxresp_sz = maxcount;
483         session_fchan->maxresp_sz = fchan->maxresp_sz;
484
485         /* Use the client's maxops if possible */
486         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
487                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
488         session_fchan->maxops = fchan->maxops;
489
490         /* FIXME: Error means no more DRC pages so the server should
491          * recover pages from existing sessions. For now fail session
492          * creation.
493          */
494         status = set_forechannel_drc_size(fchan);
495
496         session_fchan->maxresp_cached = fchan->maxresp_cached;
497         session_fchan->maxreqs = fchan->maxreqs;
498
499         dprintk("%s status %d\n", __func__, status);
500         return status;
501 }
502
503 static void
504 free_session_slots(struct nfsd4_session *ses)
505 {
506         int i;
507
508         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
509                 kfree(ses->se_slots[i]);
510 }
511
512 /*
513  * We don't actually need to cache the rpc and session headers, so we
514  * can allocate a little less for each slot:
515  */
516 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
517 {
518         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
519 }
520
521 static int
522 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
523                    struct nfsd4_create_session *cses)
524 {
525         struct nfsd4_session *new, tmp;
526         struct nfsd4_slot *sp;
527         int idx, slotsize, cachesize, i;
528         int status;
529
530         memset(&tmp, 0, sizeof(tmp));
531
532         /* FIXME: For now, we just accept the client back channel attributes. */
533         tmp.se_bchannel = cses->back_channel;
534         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
535                                         &cses->fore_channel);
536         if (status)
537                 goto out;
538
539         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
540                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
541
542         status = nfserr_serverfault;
543         /* allocate struct nfsd4_session and slot table pointers in one piece */
544         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
545         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
546         if (!new)
547                 goto out;
548
549         memcpy(new, &tmp, sizeof(*new));
550
551         /* allocate each struct nfsd4_slot and data cache in one piece */
552         cachesize = slot_bytes(&new->se_fchannel);
553         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
554                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
555                 if (!sp)
556                         goto out_free;
557                 new->se_slots[i] = sp;
558         }
559
560         new->se_client = clp;
561         gen_sessionid(new);
562         idx = hash_sessionid(&new->se_sessionid);
563         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
564                NFS4_MAX_SESSIONID_LEN);
565
566         new->se_flags = cses->flags;
567         kref_init(&new->se_ref);
568         spin_lock(&sessionid_lock);
569         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
570         list_add(&new->se_perclnt, &clp->cl_sessions);
571         spin_unlock(&sessionid_lock);
572
573         status = nfs_ok;
574 out:
575         return status;
576 out_free:
577         free_session_slots(new);
578         kfree(new);
579         goto out;
580 }
581
582 /* caller must hold sessionid_lock */
583 static struct nfsd4_session *
584 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
585 {
586         struct nfsd4_session *elem;
587         int idx;
588
589         dump_sessionid(__func__, sessionid);
590         idx = hash_sessionid(sessionid);
591         dprintk("%s: idx is %d\n", __func__, idx);
592         /* Search in the appropriate list */
593         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
594                 dump_sessionid("list traversal", &elem->se_sessionid);
595                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
596                             NFS4_MAX_SESSIONID_LEN)) {
597                         return elem;
598                 }
599         }
600
601         dprintk("%s: session not found\n", __func__);
602         return NULL;
603 }
604
605 /* caller must hold sessionid_lock */
606 static void
607 unhash_session(struct nfsd4_session *ses)
608 {
609         list_del(&ses->se_hash);
610         list_del(&ses->se_perclnt);
611 }
612
613 static void
614 release_session(struct nfsd4_session *ses)
615 {
616         spin_lock(&sessionid_lock);
617         unhash_session(ses);
618         spin_unlock(&sessionid_lock);
619         nfsd4_put_session(ses);
620 }
621
622 void
623 free_session(struct kref *kref)
624 {
625         struct nfsd4_session *ses;
626         int mem;
627
628         ses = container_of(kref, struct nfsd4_session, se_ref);
629         spin_lock(&nfsd_drc_lock);
630         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
631         nfsd_drc_mem_used -= mem;
632         spin_unlock(&nfsd_drc_lock);
633         free_session_slots(ses);
634         kfree(ses);
635 }
636
637 static inline void
638 renew_client(struct nfs4_client *clp)
639 {
640         /*
641         * Move client to the end to the LRU list.
642         */
643         dprintk("renewing client (clientid %08x/%08x)\n", 
644                         clp->cl_clientid.cl_boot, 
645                         clp->cl_clientid.cl_id);
646         list_move_tail(&clp->cl_lru, &client_lru);
647         clp->cl_time = get_seconds();
648 }
649
650 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
651 static int
652 STALE_CLIENTID(clientid_t *clid)
653 {
654         if (clid->cl_boot == boot_time)
655                 return 0;
656         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
657                 clid->cl_boot, clid->cl_id, boot_time);
658         return 1;
659 }
660
661 /* 
662  * XXX Should we use a slab cache ?
663  * This type of memory management is somewhat inefficient, but we use it
664  * anyway since SETCLIENTID is not a common operation.
665  */
666 static struct nfs4_client *alloc_client(struct xdr_netobj name)
667 {
668         struct nfs4_client *clp;
669
670         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
671         if (clp == NULL)
672                 return NULL;
673         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
674         if (clp->cl_name.data == NULL) {
675                 kfree(clp);
676                 return NULL;
677         }
678         memcpy(clp->cl_name.data, name.data, name.len);
679         clp->cl_name.len = name.len;
680         return clp;
681 }
682
683 static void
684 shutdown_callback_client(struct nfs4_client *clp)
685 {
686         struct rpc_clnt *clnt = clp->cl_cb_conn.cb_client;
687
688         if (clnt) {
689                 /*
690                  * Callback threads take a reference on the client, so there
691                  * should be no outstanding callbacks at this point.
692                  */
693                 clp->cl_cb_conn.cb_client = NULL;
694                 rpc_shutdown_client(clnt);
695         }
696 }
697
698 static inline void
699 free_client(struct nfs4_client *clp)
700 {
701         shutdown_callback_client(clp);
702         if (clp->cl_cb_xprt)
703                 svc_xprt_put(clp->cl_cb_xprt);
704         if (clp->cl_cred.cr_group_info)
705                 put_group_info(clp->cl_cred.cr_group_info);
706         kfree(clp->cl_principal);
707         kfree(clp->cl_name.data);
708         kfree(clp);
709 }
710
711 void
712 put_nfs4_client(struct nfs4_client *clp)
713 {
714         if (atomic_dec_and_test(&clp->cl_count))
715                 free_client(clp);
716 }
717
718 static void
719 expire_client(struct nfs4_client *clp)
720 {
721         struct nfs4_stateowner *sop;
722         struct nfs4_delegation *dp;
723         struct list_head reaplist;
724
725         dprintk("NFSD: expire_client cl_count %d\n",
726                             atomic_read(&clp->cl_count));
727
728         INIT_LIST_HEAD(&reaplist);
729         spin_lock(&recall_lock);
730         while (!list_empty(&clp->cl_delegations)) {
731                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
732                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
733                                 dp->dl_flock);
734                 list_del_init(&dp->dl_perclnt);
735                 list_move(&dp->dl_recall_lru, &reaplist);
736         }
737         spin_unlock(&recall_lock);
738         while (!list_empty(&reaplist)) {
739                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
740                 list_del_init(&dp->dl_recall_lru);
741                 unhash_delegation(dp);
742         }
743         list_del(&clp->cl_idhash);
744         list_del(&clp->cl_strhash);
745         list_del(&clp->cl_lru);
746         while (!list_empty(&clp->cl_openowners)) {
747                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
748                 release_openowner(sop);
749         }
750         while (!list_empty(&clp->cl_sessions)) {
751                 struct nfsd4_session  *ses;
752                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
753                                  se_perclnt);
754                 release_session(ses);
755         }
756         put_nfs4_client(clp);
757 }
758
759 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
760 {
761         memcpy(target->cl_verifier.data, source->data,
762                         sizeof(target->cl_verifier.data));
763 }
764
765 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
766 {
767         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
768         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
769 }
770
771 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
772 {
773         target->cr_uid = source->cr_uid;
774         target->cr_gid = source->cr_gid;
775         target->cr_group_info = source->cr_group_info;
776         get_group_info(target->cr_group_info);
777 }
778
779 static int same_name(const char *n1, const char *n2)
780 {
781         return 0 == memcmp(n1, n2, HEXDIR_LEN);
782 }
783
784 static int
785 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
786 {
787         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
788 }
789
790 static int
791 same_clid(clientid_t *cl1, clientid_t *cl2)
792 {
793         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
794 }
795
796 /* XXX what about NGROUP */
797 static int
798 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
799 {
800         return cr1->cr_uid == cr2->cr_uid;
801 }
802
803 static void gen_clid(struct nfs4_client *clp)
804 {
805         static u32 current_clientid = 1;
806
807         clp->cl_clientid.cl_boot = boot_time;
808         clp->cl_clientid.cl_id = current_clientid++; 
809 }
810
811 static void gen_confirm(struct nfs4_client *clp)
812 {
813         static u32 i;
814         u32 *p;
815
816         p = (u32 *)clp->cl_confirm.data;
817         *p++ = get_seconds();
818         *p++ = i++;
819 }
820
821 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
822                 struct svc_rqst *rqstp, nfs4_verifier *verf)
823 {
824         struct nfs4_client *clp;
825         struct sockaddr *sa = svc_addr(rqstp);
826         char *princ;
827
828         clp = alloc_client(name);
829         if (clp == NULL)
830                 return NULL;
831
832         princ = svc_gss_principal(rqstp);
833         if (princ) {
834                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
835                 if (clp->cl_principal == NULL) {
836                         free_client(clp);
837                         return NULL;
838                 }
839         }
840
841         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
842         atomic_set(&clp->cl_count, 1);
843         atomic_set(&clp->cl_cb_conn.cb_set, 0);
844         INIT_LIST_HEAD(&clp->cl_idhash);
845         INIT_LIST_HEAD(&clp->cl_strhash);
846         INIT_LIST_HEAD(&clp->cl_openowners);
847         INIT_LIST_HEAD(&clp->cl_delegations);
848         INIT_LIST_HEAD(&clp->cl_sessions);
849         INIT_LIST_HEAD(&clp->cl_lru);
850         clear_bit(0, &clp->cl_cb_slot_busy);
851         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
852         copy_verf(clp, verf);
853         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
854         clp->cl_flavor = rqstp->rq_flavor;
855         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
856         gen_confirm(clp);
857
858         return clp;
859 }
860
861 static int check_name(struct xdr_netobj name)
862 {
863         if (name.len == 0) 
864                 return 0;
865         if (name.len > NFS4_OPAQUE_LIMIT) {
866                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
867                 return 0;
868         }
869         return 1;
870 }
871
872 static void
873 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
874 {
875         unsigned int idhashval;
876
877         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
878         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
879         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
880         list_add_tail(&clp->cl_lru, &client_lru);
881         clp->cl_time = get_seconds();
882 }
883
884 static void
885 move_to_confirmed(struct nfs4_client *clp)
886 {
887         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
888         unsigned int strhashval;
889
890         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
891         list_del_init(&clp->cl_strhash);
892         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
893         strhashval = clientstr_hashval(clp->cl_recdir);
894         list_add(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
895         renew_client(clp);
896 }
897
898 static struct nfs4_client *
899 find_confirmed_client(clientid_t *clid)
900 {
901         struct nfs4_client *clp;
902         unsigned int idhashval = clientid_hashval(clid->cl_id);
903
904         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
905                 if (same_clid(&clp->cl_clientid, clid))
906                         return clp;
907         }
908         return NULL;
909 }
910
911 static struct nfs4_client *
912 find_unconfirmed_client(clientid_t *clid)
913 {
914         struct nfs4_client *clp;
915         unsigned int idhashval = clientid_hashval(clid->cl_id);
916
917         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
918                 if (same_clid(&clp->cl_clientid, clid))
919                         return clp;
920         }
921         return NULL;
922 }
923
924 /*
925  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
926  * parameter. Matching is based on the fact the at least one of the
927  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
928  *
929  * FIXME: we need to unify the clientid namespaces for nfsv4.x
930  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
931  * and SET_CLIENTID{,_CONFIRM}
932  */
933 static inline int
934 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
935 {
936         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
937         return use_exchange_id == has_exchange_flags;
938 }
939
940 static struct nfs4_client *
941 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
942                              bool use_exchange_id)
943 {
944         struct nfs4_client *clp;
945
946         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
947                 if (same_name(clp->cl_recdir, dname) &&
948                     match_clientid_establishment(clp, use_exchange_id))
949                         return clp;
950         }
951         return NULL;
952 }
953
954 static struct nfs4_client *
955 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
956                                bool use_exchange_id)
957 {
958         struct nfs4_client *clp;
959
960         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
961                 if (same_name(clp->cl_recdir, dname) &&
962                     match_clientid_establishment(clp, use_exchange_id))
963                         return clp;
964         }
965         return NULL;
966 }
967
968 static void
969 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
970 {
971         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
972         unsigned short expected_family;
973
974         /* Currently, we only support tcp and tcp6 for the callback channel */
975         if (se->se_callback_netid_len == 3 &&
976             !memcmp(se->se_callback_netid_val, "tcp", 3))
977                 expected_family = AF_INET;
978         else if (se->se_callback_netid_len == 4 &&
979                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
980                 expected_family = AF_INET6;
981         else
982                 goto out_err;
983
984         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
985                                             se->se_callback_addr_len,
986                                             (struct sockaddr *) &cb->cb_addr,
987                                             sizeof(cb->cb_addr));
988
989         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
990                 goto out_err;
991
992         if (cb->cb_addr.ss_family == AF_INET6)
993                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
994
995         cb->cb_minorversion = 0;
996         cb->cb_prog = se->se_callback_prog;
997         cb->cb_ident = se->se_callback_ident;
998         return;
999 out_err:
1000         cb->cb_addr.ss_family = AF_UNSPEC;
1001         cb->cb_addrlen = 0;
1002         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1003                 "will not receive delegations\n",
1004                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1005
1006         return;
1007 }
1008
1009 /*
1010  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1011  */
1012 void
1013 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1014 {
1015         struct nfsd4_slot *slot = resp->cstate.slot;
1016         unsigned int base;
1017
1018         dprintk("--> %s slot %p\n", __func__, slot);
1019
1020         slot->sl_opcnt = resp->opcnt;
1021         slot->sl_status = resp->cstate.status;
1022
1023         if (nfsd4_not_cached(resp)) {
1024                 slot->sl_datalen = 0;
1025                 return;
1026         }
1027         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1028         base = (char *)resp->cstate.datap -
1029                                         (char *)resp->xbuf->head[0].iov_base;
1030         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1031                                     slot->sl_datalen))
1032                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1033         return;
1034 }
1035
1036 /*
1037  * Encode the replay sequence operation from the slot values.
1038  * If cachethis is FALSE encode the uncached rep error on the next
1039  * operation which sets resp->p and increments resp->opcnt for
1040  * nfs4svc_encode_compoundres.
1041  *
1042  */
1043 static __be32
1044 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1045                           struct nfsd4_compoundres *resp)
1046 {
1047         struct nfsd4_op *op;
1048         struct nfsd4_slot *slot = resp->cstate.slot;
1049
1050         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1051                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1052
1053         /* Encode the replayed sequence operation */
1054         op = &args->ops[resp->opcnt - 1];
1055         nfsd4_encode_operation(resp, op);
1056
1057         /* Return nfserr_retry_uncached_rep in next operation. */
1058         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1059                 op = &args->ops[resp->opcnt++];
1060                 op->status = nfserr_retry_uncached_rep;
1061                 nfsd4_encode_operation(resp, op);
1062         }
1063         return op->status;
1064 }
1065
1066 /*
1067  * The sequence operation is not cached because we can use the slot and
1068  * session values.
1069  */
1070 __be32
1071 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1072                          struct nfsd4_sequence *seq)
1073 {
1074         struct nfsd4_slot *slot = resp->cstate.slot;
1075         __be32 status;
1076
1077         dprintk("--> %s slot %p\n", __func__, slot);
1078
1079         /* Either returns 0 or nfserr_retry_uncached */
1080         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1081         if (status == nfserr_retry_uncached_rep)
1082                 return status;
1083
1084         /* The sequence operation has been encoded, cstate->datap set. */
1085         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1086
1087         resp->opcnt = slot->sl_opcnt;
1088         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1089         status = slot->sl_status;
1090
1091         return status;
1092 }
1093
1094 /*
1095  * Set the exchange_id flags returned by the server.
1096  */
1097 static void
1098 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1099 {
1100         /* pNFS is not supported */
1101         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1102
1103         /* Referrals are supported, Migration is not. */
1104         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1105
1106         /* set the wire flags to return to client. */
1107         clid->flags = new->cl_exchange_flags;
1108 }
1109
1110 __be32
1111 nfsd4_exchange_id(struct svc_rqst *rqstp,
1112                   struct nfsd4_compound_state *cstate,
1113                   struct nfsd4_exchange_id *exid)
1114 {
1115         struct nfs4_client *unconf, *conf, *new;
1116         int status;
1117         unsigned int            strhashval;
1118         char                    dname[HEXDIR_LEN];
1119         char                    addr_str[INET6_ADDRSTRLEN];
1120         nfs4_verifier           verf = exid->verifier;
1121         struct sockaddr         *sa = svc_addr(rqstp);
1122
1123         rpc_ntop(sa, addr_str, sizeof(addr_str));
1124         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1125                 "ip_addr=%s flags %x, spa_how %d\n",
1126                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1127                 addr_str, exid->flags, exid->spa_how);
1128
1129         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1130                 return nfserr_inval;
1131
1132         /* Currently only support SP4_NONE */
1133         switch (exid->spa_how) {
1134         case SP4_NONE:
1135                 break;
1136         case SP4_SSV:
1137                 return nfserr_encr_alg_unsupp;
1138         default:
1139                 BUG();                          /* checked by xdr code */
1140         case SP4_MACH_CRED:
1141                 return nfserr_serverfault;      /* no excuse :-/ */
1142         }
1143
1144         status = nfs4_make_rec_clidname(dname, &exid->clname);
1145
1146         if (status)
1147                 goto error;
1148
1149         strhashval = clientstr_hashval(dname);
1150
1151         nfs4_lock_state();
1152         status = nfs_ok;
1153
1154         conf = find_confirmed_client_by_str(dname, strhashval, true);
1155         if (conf) {
1156                 if (!same_verf(&verf, &conf->cl_verifier)) {
1157                         /* 18.35.4 case 8 */
1158                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1159                                 status = nfserr_not_same;
1160                                 goto out;
1161                         }
1162                         /* Client reboot: destroy old state */
1163                         expire_client(conf);
1164                         goto out_new;
1165                 }
1166                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1167                         /* 18.35.4 case 9 */
1168                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1169                                 status = nfserr_perm;
1170                                 goto out;
1171                         }
1172                         expire_client(conf);
1173                         goto out_new;
1174                 }
1175                 /*
1176                  * Set bit when the owner id and verifier map to an already
1177                  * confirmed client id (18.35.3).
1178                  */
1179                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1180
1181                 /*
1182                  * Falling into 18.35.4 case 2, possible router replay.
1183                  * Leave confirmed record intact and return same result.
1184                  */
1185                 copy_verf(conf, &verf);
1186                 new = conf;
1187                 goto out_copy;
1188         }
1189
1190         /* 18.35.4 case 7 */
1191         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1192                 status = nfserr_noent;
1193                 goto out;
1194         }
1195
1196         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1197         if (unconf) {
1198                 /*
1199                  * Possible retry or client restart.  Per 18.35.4 case 4,
1200                  * a new unconfirmed record should be generated regardless
1201                  * of whether any properties have changed.
1202                  */
1203                 expire_client(unconf);
1204         }
1205
1206 out_new:
1207         /* Normal case */
1208         new = create_client(exid->clname, dname, rqstp, &verf);
1209         if (new == NULL) {
1210                 status = nfserr_serverfault;
1211                 goto out;
1212         }
1213
1214         gen_clid(new);
1215         add_to_unconfirmed(new, strhashval);
1216 out_copy:
1217         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1218         exid->clientid.cl_id = new->cl_clientid.cl_id;
1219
1220         exid->seqid = 1;
1221         nfsd4_set_ex_flags(new, exid);
1222
1223         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1224                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1225         status = nfs_ok;
1226
1227 out:
1228         nfs4_unlock_state();
1229 error:
1230         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1231         return status;
1232 }
1233
1234 static int
1235 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1236 {
1237         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1238                 slot_seqid);
1239
1240         /* The slot is in use, and no response has been sent. */
1241         if (slot_inuse) {
1242                 if (seqid == slot_seqid)
1243                         return nfserr_jukebox;
1244                 else
1245                         return nfserr_seq_misordered;
1246         }
1247         /* Normal */
1248         if (likely(seqid == slot_seqid + 1))
1249                 return nfs_ok;
1250         /* Replay */
1251         if (seqid == slot_seqid)
1252                 return nfserr_replay_cache;
1253         /* Wraparound */
1254         if (seqid == 1 && (slot_seqid + 1) == 0)
1255                 return nfs_ok;
1256         /* Misordered replay or misordered new request */
1257         return nfserr_seq_misordered;
1258 }
1259
1260 /*
1261  * Cache the create session result into the create session single DRC
1262  * slot cache by saving the xdr structure. sl_seqid has been set.
1263  * Do this for solo or embedded create session operations.
1264  */
1265 static void
1266 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1267                            struct nfsd4_clid_slot *slot, int nfserr)
1268 {
1269         slot->sl_status = nfserr;
1270         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1271 }
1272
1273 static __be32
1274 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1275                             struct nfsd4_clid_slot *slot)
1276 {
1277         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1278         return slot->sl_status;
1279 }
1280
1281 __be32
1282 nfsd4_create_session(struct svc_rqst *rqstp,
1283                      struct nfsd4_compound_state *cstate,
1284                      struct nfsd4_create_session *cr_ses)
1285 {
1286         struct sockaddr *sa = svc_addr(rqstp);
1287         struct nfs4_client *conf, *unconf;
1288         struct nfsd4_clid_slot *cs_slot = NULL;
1289         int status = 0;
1290
1291         nfs4_lock_state();
1292         unconf = find_unconfirmed_client(&cr_ses->clientid);
1293         conf = find_confirmed_client(&cr_ses->clientid);
1294
1295         if (conf) {
1296                 cs_slot = &conf->cl_cs_slot;
1297                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1298                 if (status == nfserr_replay_cache) {
1299                         dprintk("Got a create_session replay! seqid= %d\n",
1300                                 cs_slot->sl_seqid);
1301                         /* Return the cached reply status */
1302                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1303                         goto out;
1304                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1305                         status = nfserr_seq_misordered;
1306                         dprintk("Sequence misordered!\n");
1307                         dprintk("Expected seqid= %d but got seqid= %d\n",
1308                                 cs_slot->sl_seqid, cr_ses->seqid);
1309                         goto out;
1310                 }
1311                 cs_slot->sl_seqid++;
1312         } else if (unconf) {
1313                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1314                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1315                         status = nfserr_clid_inuse;
1316                         goto out;
1317                 }
1318
1319                 cs_slot = &unconf->cl_cs_slot;
1320                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1321                 if (status) {
1322                         /* an unconfirmed replay returns misordered */
1323                         status = nfserr_seq_misordered;
1324                         goto out_cache;
1325                 }
1326
1327                 cs_slot->sl_seqid++; /* from 0 to 1 */
1328                 move_to_confirmed(unconf);
1329
1330                 /*
1331                  * We do not support RDMA or persistent sessions
1332                  */
1333                 cr_ses->flags &= ~SESSION4_PERSIST;
1334                 cr_ses->flags &= ~SESSION4_RDMA;
1335
1336                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1337                         unconf->cl_cb_xprt = rqstp->rq_xprt;
1338                         svc_xprt_get(unconf->cl_cb_xprt);
1339                         rpc_copy_addr(
1340                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1341                                 sa);
1342                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1343                         unconf->cl_cb_conn.cb_minorversion =
1344                                 cstate->minorversion;
1345                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1346                         unconf->cl_cb_seq_nr = 1;
1347                         nfsd4_probe_callback(unconf);
1348                 }
1349                 conf = unconf;
1350         } else {
1351                 status = nfserr_stale_clientid;
1352                 goto out;
1353         }
1354
1355         status = alloc_init_session(rqstp, conf, cr_ses);
1356         if (status)
1357                 goto out;
1358
1359         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1360                NFS4_MAX_SESSIONID_LEN);
1361         cr_ses->seqid = cs_slot->sl_seqid;
1362
1363 out_cache:
1364         /* cache solo and embedded create sessions under the state lock */
1365         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1366 out:
1367         nfs4_unlock_state();
1368         dprintk("%s returns %d\n", __func__, ntohl(status));
1369         return status;
1370 }
1371
1372 __be32
1373 nfsd4_destroy_session(struct svc_rqst *r,
1374                       struct nfsd4_compound_state *cstate,
1375                       struct nfsd4_destroy_session *sessionid)
1376 {
1377         struct nfsd4_session *ses;
1378         u32 status = nfserr_badsession;
1379
1380         /* Notes:
1381          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1382          * - Should we return nfserr_back_chan_busy if waiting for
1383          *   callbacks on to-be-destroyed session?
1384          * - Do we need to clear any callback info from previous session?
1385          */
1386
1387         dump_sessionid(__func__, &sessionid->sessionid);
1388         spin_lock(&sessionid_lock);
1389         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1390         if (!ses) {
1391                 spin_unlock(&sessionid_lock);
1392                 goto out;
1393         }
1394
1395         unhash_session(ses);
1396         spin_unlock(&sessionid_lock);
1397
1398         /* wait for callbacks */
1399         shutdown_callback_client(ses->se_client);
1400         nfsd4_put_session(ses);
1401         status = nfs_ok;
1402 out:
1403         dprintk("%s returns %d\n", __func__, ntohl(status));
1404         return status;
1405 }
1406
1407 __be32
1408 nfsd4_sequence(struct svc_rqst *rqstp,
1409                struct nfsd4_compound_state *cstate,
1410                struct nfsd4_sequence *seq)
1411 {
1412         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1413         struct nfsd4_session *session;
1414         struct nfsd4_slot *slot;
1415         int status;
1416
1417         if (resp->opcnt != 1)
1418                 return nfserr_sequence_pos;
1419
1420         spin_lock(&sessionid_lock);
1421         status = nfserr_badsession;
1422         session = find_in_sessionid_hashtbl(&seq->sessionid);
1423         if (!session)
1424                 goto out;
1425
1426         status = nfserr_badslot;
1427         if (seq->slotid >= session->se_fchannel.maxreqs)
1428                 goto out;
1429
1430         slot = session->se_slots[seq->slotid];
1431         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1432
1433         /* We do not negotiate the number of slots yet, so set the
1434          * maxslots to the session maxreqs which is used to encode
1435          * sr_highest_slotid and the sr_target_slot id to maxslots */
1436         seq->maxslots = session->se_fchannel.maxreqs;
1437
1438         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1439         if (status == nfserr_replay_cache) {
1440                 cstate->slot = slot;
1441                 cstate->session = session;
1442                 /* Return the cached reply status and set cstate->status
1443                  * for nfsd4_proc_compound processing */
1444                 status = nfsd4_replay_cache_entry(resp, seq);
1445                 cstate->status = nfserr_replay_cache;
1446                 goto out;
1447         }
1448         if (status)
1449                 goto out;
1450
1451         /* Success! bump slot seqid */
1452         slot->sl_inuse = true;
1453         slot->sl_seqid = seq->seqid;
1454         slot->sl_cachethis = seq->cachethis;
1455
1456         cstate->slot = slot;
1457         cstate->session = session;
1458
1459         /* Hold a session reference until done processing the compound:
1460          * nfsd4_put_session called only if the cstate slot is set.
1461          */
1462         nfsd4_get_session(session);
1463 out:
1464         spin_unlock(&sessionid_lock);
1465         /* Renew the clientid on success and on replay */
1466         if (cstate->session) {
1467                 nfs4_lock_state();
1468                 renew_client(session->se_client);
1469                 nfs4_unlock_state();
1470         }
1471         dprintk("%s: return %d\n", __func__, ntohl(status));
1472         return status;
1473 }
1474
1475 __be32
1476 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1477                   struct nfsd4_setclientid *setclid)
1478 {
1479         struct sockaddr         *sa = svc_addr(rqstp);
1480         struct xdr_netobj       clname = { 
1481                 .len = setclid->se_namelen,
1482                 .data = setclid->se_name,
1483         };
1484         nfs4_verifier           clverifier = setclid->se_verf;
1485         unsigned int            strhashval;
1486         struct nfs4_client      *conf, *unconf, *new;
1487         __be32                  status;
1488         char                    dname[HEXDIR_LEN];
1489         
1490         if (!check_name(clname))
1491                 return nfserr_inval;
1492
1493         status = nfs4_make_rec_clidname(dname, &clname);
1494         if (status)
1495                 return status;
1496
1497         /* 
1498          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1499          * We get here on a DRC miss.
1500          */
1501
1502         strhashval = clientstr_hashval(dname);
1503
1504         nfs4_lock_state();
1505         conf = find_confirmed_client_by_str(dname, strhashval, false);
1506         if (conf) {
1507                 /* RFC 3530 14.2.33 CASE 0: */
1508                 status = nfserr_clid_inuse;
1509                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1510                         char addr_str[INET6_ADDRSTRLEN];
1511                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1512                                  sizeof(addr_str));
1513                         dprintk("NFSD: setclientid: string in use by client "
1514                                 "at %s\n", addr_str);
1515                         goto out;
1516                 }
1517         }
1518         /*
1519          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1520          * has a description of SETCLIENTID request processing consisting
1521          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1522          */
1523         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1524         status = nfserr_resource;
1525         if (!conf) {
1526                 /*
1527                  * RFC 3530 14.2.33 CASE 4:
1528                  * placed first, because it is the normal case
1529                  */
1530                 if (unconf)
1531                         expire_client(unconf);
1532                 new = create_client(clname, dname, rqstp, &clverifier);
1533                 if (new == NULL)
1534                         goto out;
1535                 gen_clid(new);
1536         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1537                 /*
1538                  * RFC 3530 14.2.33 CASE 1:
1539                  * probable callback update
1540                  */
1541                 if (unconf) {
1542                         /* Note this is removing unconfirmed {*x***},
1543                          * which is stronger than RFC recommended {vxc**}.
1544                          * This has the advantage that there is at most
1545                          * one {*x***} in either list at any time.
1546                          */
1547                         expire_client(unconf);
1548                 }
1549                 new = create_client(clname, dname, rqstp, &clverifier);
1550                 if (new == NULL)
1551                         goto out;
1552                 copy_clid(new, conf);
1553         } else if (!unconf) {
1554                 /*
1555                  * RFC 3530 14.2.33 CASE 2:
1556                  * probable client reboot; state will be removed if
1557                  * confirmed.
1558                  */
1559                 new = create_client(clname, dname, rqstp, &clverifier);
1560                 if (new == NULL)
1561                         goto out;
1562                 gen_clid(new);
1563         } else {
1564                 /*
1565                  * RFC 3530 14.2.33 CASE 3:
1566                  * probable client reboot; state will be removed if
1567                  * confirmed.
1568                  */
1569                 expire_client(unconf);
1570                 new = create_client(clname, dname, rqstp, &clverifier);
1571                 if (new == NULL)
1572                         goto out;
1573                 gen_clid(new);
1574         }
1575         gen_callback(new, setclid, rpc_get_scope_id(sa));
1576         add_to_unconfirmed(new, strhashval);
1577         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1578         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1579         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1580         status = nfs_ok;
1581 out:
1582         nfs4_unlock_state();
1583         return status;
1584 }
1585
1586
1587 /*
1588  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1589  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1590  * bullets, labeled as CASE1 - CASE4 below.
1591  */
1592 __be32
1593 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1594                          struct nfsd4_compound_state *cstate,
1595                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1596 {
1597         struct sockaddr *sa = svc_addr(rqstp);
1598         struct nfs4_client *conf, *unconf;
1599         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1600         clientid_t * clid = &setclientid_confirm->sc_clientid;
1601         __be32 status;
1602
1603         if (STALE_CLIENTID(clid))
1604                 return nfserr_stale_clientid;
1605         /* 
1606          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1607          * We get here on a DRC miss.
1608          */
1609
1610         nfs4_lock_state();
1611
1612         conf = find_confirmed_client(clid);
1613         unconf = find_unconfirmed_client(clid);
1614
1615         status = nfserr_clid_inuse;
1616         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1617                 goto out;
1618         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1619                 goto out;
1620
1621         /*
1622          * section 14.2.34 of RFC 3530 has a description of
1623          * SETCLIENTID_CONFIRM request processing consisting
1624          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1625          */
1626         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1627                 /*
1628                  * RFC 3530 14.2.34 CASE 1:
1629                  * callback update
1630                  */
1631                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1632                         status = nfserr_clid_inuse;
1633                 else {
1634                         /* XXX: We just turn off callbacks until we can handle
1635                           * change request correctly. */
1636                         atomic_set(&conf->cl_cb_conn.cb_set, 0);
1637                         expire_client(unconf);
1638                         status = nfs_ok;
1639
1640                 }
1641         } else if (conf && !unconf) {
1642                 /*
1643                  * RFC 3530 14.2.34 CASE 2:
1644                  * probable retransmitted request; play it safe and
1645                  * do nothing.
1646                  */
1647                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1648                         status = nfserr_clid_inuse;
1649                 else
1650                         status = nfs_ok;
1651         } else if (!conf && unconf
1652                         && same_verf(&unconf->cl_confirm, &confirm)) {
1653                 /*
1654                  * RFC 3530 14.2.34 CASE 3:
1655                  * Normal case; new or rebooted client:
1656                  */
1657                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1658                         status = nfserr_clid_inuse;
1659                 } else {
1660                         unsigned int hash =
1661                                 clientstr_hashval(unconf->cl_recdir);
1662                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1663                                                             hash, false);
1664                         if (conf) {
1665                                 nfsd4_remove_clid_dir(conf);
1666                                 expire_client(conf);
1667                         }
1668                         move_to_confirmed(unconf);
1669                         conf = unconf;
1670                         nfsd4_probe_callback(conf);
1671                         status = nfs_ok;
1672                 }
1673         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1674             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1675                                                                 &confirm)))) {
1676                 /*
1677                  * RFC 3530 14.2.34 CASE 4:
1678                  * Client probably hasn't noticed that we rebooted yet.
1679                  */
1680                 status = nfserr_stale_clientid;
1681         } else {
1682                 /* check that we have hit one of the cases...*/
1683                 status = nfserr_clid_inuse;
1684         }
1685 out:
1686         nfs4_unlock_state();
1687         return status;
1688 }
1689
1690 /* OPEN Share state helper functions */
1691 static inline struct nfs4_file *
1692 alloc_init_file(struct inode *ino)
1693 {
1694         struct nfs4_file *fp;
1695         unsigned int hashval = file_hashval(ino);
1696
1697         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1698         if (fp) {
1699                 atomic_set(&fp->fi_ref, 1);
1700                 INIT_LIST_HEAD(&fp->fi_hash);
1701                 INIT_LIST_HEAD(&fp->fi_stateids);
1702                 INIT_LIST_HEAD(&fp->fi_delegations);
1703                 spin_lock(&recall_lock);
1704                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1705                 spin_unlock(&recall_lock);
1706                 fp->fi_inode = igrab(ino);
1707                 fp->fi_id = current_fileid++;
1708                 fp->fi_had_conflict = false;
1709                 return fp;
1710         }
1711         return NULL;
1712 }
1713
1714 static void
1715 nfsd4_free_slab(struct kmem_cache **slab)
1716 {
1717         if (*slab == NULL)
1718                 return;
1719         kmem_cache_destroy(*slab);
1720         *slab = NULL;
1721 }
1722
1723 void
1724 nfsd4_free_slabs(void)
1725 {
1726         nfsd4_free_slab(&stateowner_slab);
1727         nfsd4_free_slab(&file_slab);
1728         nfsd4_free_slab(&stateid_slab);
1729         nfsd4_free_slab(&deleg_slab);
1730 }
1731
1732 static int
1733 nfsd4_init_slabs(void)
1734 {
1735         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1736                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1737         if (stateowner_slab == NULL)
1738                 goto out_nomem;
1739         file_slab = kmem_cache_create("nfsd4_files",
1740                         sizeof(struct nfs4_file), 0, 0, NULL);
1741         if (file_slab == NULL)
1742                 goto out_nomem;
1743         stateid_slab = kmem_cache_create("nfsd4_stateids",
1744                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1745         if (stateid_slab == NULL)
1746                 goto out_nomem;
1747         deleg_slab = kmem_cache_create("nfsd4_delegations",
1748                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1749         if (deleg_slab == NULL)
1750                 goto out_nomem;
1751         return 0;
1752 out_nomem:
1753         nfsd4_free_slabs();
1754         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1755         return -ENOMEM;
1756 }
1757
1758 void
1759 nfs4_free_stateowner(struct kref *kref)
1760 {
1761         struct nfs4_stateowner *sop =
1762                 container_of(kref, struct nfs4_stateowner, so_ref);
1763         kfree(sop->so_owner.data);
1764         kmem_cache_free(stateowner_slab, sop);
1765 }
1766
1767 static inline struct nfs4_stateowner *
1768 alloc_stateowner(struct xdr_netobj *owner)
1769 {
1770         struct nfs4_stateowner *sop;
1771
1772         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1773                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1774                         memcpy(sop->so_owner.data, owner->data, owner->len);
1775                         sop->so_owner.len = owner->len;
1776                         kref_init(&sop->so_ref);
1777                         return sop;
1778                 } 
1779                 kmem_cache_free(stateowner_slab, sop);
1780         }
1781         return NULL;
1782 }
1783
1784 static struct nfs4_stateowner *
1785 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1786         struct nfs4_stateowner *sop;
1787         struct nfs4_replay *rp;
1788         unsigned int idhashval;
1789
1790         if (!(sop = alloc_stateowner(&open->op_owner)))
1791                 return NULL;
1792         idhashval = ownerid_hashval(current_ownerid);
1793         INIT_LIST_HEAD(&sop->so_idhash);
1794         INIT_LIST_HEAD(&sop->so_strhash);
1795         INIT_LIST_HEAD(&sop->so_perclient);
1796         INIT_LIST_HEAD(&sop->so_stateids);
1797         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1798         INIT_LIST_HEAD(&sop->so_close_lru);
1799         sop->so_time = 0;
1800         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1801         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1802         list_add(&sop->so_perclient, &clp->cl_openowners);
1803         sop->so_is_open_owner = 1;
1804         sop->so_id = current_ownerid++;
1805         sop->so_client = clp;
1806         sop->so_seqid = open->op_seqid;
1807         sop->so_confirmed = 0;
1808         rp = &sop->so_replay;
1809         rp->rp_status = nfserr_serverfault;
1810         rp->rp_buflen = 0;
1811         rp->rp_buf = rp->rp_ibuf;
1812         return sop;
1813 }
1814
1815 static inline void
1816 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1817         struct nfs4_stateowner *sop = open->op_stateowner;
1818         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1819
1820         INIT_LIST_HEAD(&stp->st_hash);
1821         INIT_LIST_HEAD(&stp->st_perstateowner);
1822         INIT_LIST_HEAD(&stp->st_lockowners);
1823         INIT_LIST_HEAD(&stp->st_perfile);
1824         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
1825         list_add(&stp->st_perstateowner, &sop->so_stateids);
1826         list_add(&stp->st_perfile, &fp->fi_stateids);
1827         stp->st_stateowner = sop;
1828         get_nfs4_file(fp);
1829         stp->st_file = fp;
1830         stp->st_stateid.si_boot = get_seconds();
1831         stp->st_stateid.si_stateownerid = sop->so_id;
1832         stp->st_stateid.si_fileid = fp->fi_id;
1833         stp->st_stateid.si_generation = 0;
1834         stp->st_access_bmap = 0;
1835         stp->st_deny_bmap = 0;
1836         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
1837                   &stp->st_access_bmap);
1838         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
1839         stp->st_openstp = NULL;
1840 }
1841
1842 static void
1843 move_to_close_lru(struct nfs4_stateowner *sop)
1844 {
1845         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
1846
1847         list_move_tail(&sop->so_close_lru, &close_lru);
1848         sop->so_time = get_seconds();
1849 }
1850
1851 static int
1852 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
1853                                                         clientid_t *clid)
1854 {
1855         return (sop->so_owner.len == owner->len) &&
1856                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
1857                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
1858 }
1859
1860 static struct nfs4_stateowner *
1861 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
1862 {
1863         struct nfs4_stateowner *so = NULL;
1864
1865         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
1866                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
1867                         return so;
1868         }
1869         return NULL;
1870 }
1871
1872 /* search file_hashtbl[] for file */
1873 static struct nfs4_file *
1874 find_file(struct inode *ino)
1875 {
1876         unsigned int hashval = file_hashval(ino);
1877         struct nfs4_file *fp;
1878
1879         spin_lock(&recall_lock);
1880         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
1881                 if (fp->fi_inode == ino) {
1882                         get_nfs4_file(fp);
1883                         spin_unlock(&recall_lock);
1884                         return fp;
1885                 }
1886         }
1887         spin_unlock(&recall_lock);
1888         return NULL;
1889 }
1890
1891 static inline int access_valid(u32 x, u32 minorversion)
1892 {
1893         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
1894                 return 0;
1895         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
1896                 return 0;
1897         x &= ~NFS4_SHARE_ACCESS_MASK;
1898         if (minorversion && x) {
1899                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
1900                         return 0;
1901                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
1902                         return 0;
1903                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
1904         }
1905         if (x)
1906                 return 0;
1907         return 1;
1908 }
1909
1910 static inline int deny_valid(u32 x)
1911 {
1912         /* Note: unlike access bits, deny bits may be zero. */
1913         return x <= NFS4_SHARE_DENY_BOTH;
1914 }
1915
1916 /*
1917  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1918  * st_{access,deny}_bmap field of the stateid, in order to track not
1919  * only what share bits are currently in force, but also what
1920  * combinations of share bits previous opens have used.  This allows us
1921  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1922  * return an error if the client attempt to downgrade to a combination
1923  * of share bits not explicable by closing some of its previous opens.
1924  *
1925  * XXX: This enforcement is actually incomplete, since we don't keep
1926  * track of access/deny bit combinations; so, e.g., we allow:
1927  *
1928  *      OPEN allow read, deny write
1929  *      OPEN allow both, deny none
1930  *      DOWNGRADE allow read, deny none
1931  *
1932  * which we should reject.
1933  */
1934 static void
1935 set_access(unsigned int *access, unsigned long bmap) {
1936         int i;
1937
1938         *access = 0;
1939         for (i = 1; i < 4; i++) {
1940                 if (test_bit(i, &bmap))
1941                         *access |= i;
1942         }
1943 }
1944
1945 static void
1946 set_deny(unsigned int *deny, unsigned long bmap) {
1947         int i;
1948
1949         *deny = 0;
1950         for (i = 0; i < 4; i++) {
1951                 if (test_bit(i, &bmap))
1952                         *deny |= i ;
1953         }
1954 }
1955
1956 static int
1957 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
1958         unsigned int access, deny;
1959
1960         set_access(&access, stp->st_access_bmap);
1961         set_deny(&deny, stp->st_deny_bmap);
1962         if ((access & open->op_share_deny) || (deny & open->op_share_access))
1963                 return 0;
1964         return 1;
1965 }
1966
1967 /*
1968  * Called to check deny when READ with all zero stateid or
1969  * WRITE with all zero or all one stateid
1970  */
1971 static __be32
1972 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
1973 {
1974         struct inode *ino = current_fh->fh_dentry->d_inode;
1975         struct nfs4_file *fp;
1976         struct nfs4_stateid *stp;
1977         __be32 ret;
1978
1979         dprintk("NFSD: nfs4_share_conflict\n");
1980
1981         fp = find_file(ino);
1982         if (!fp)
1983                 return nfs_ok;
1984         ret = nfserr_locked;
1985         /* Search for conflicting share reservations */
1986         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
1987                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
1988                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
1989                         goto out;
1990         }
1991         ret = nfs_ok;
1992 out:
1993         put_nfs4_file(fp);
1994         return ret;
1995 }
1996
1997 static inline void
1998 nfs4_file_downgrade(struct file *filp, unsigned int share_access)
1999 {
2000         if (share_access & NFS4_SHARE_ACCESS_WRITE) {
2001                 drop_file_write_access(filp);
2002                 spin_lock(&filp->f_lock);
2003                 filp->f_mode = (filp->f_mode | FMODE_READ) & ~FMODE_WRITE;
2004                 spin_unlock(&filp->f_lock);
2005         }
2006 }
2007
2008 /*
2009  * Spawn a thread to perform a recall on the delegation represented
2010  * by the lease (file_lock)
2011  *
2012  * Called from break_lease() with lock_kernel() held.
2013  * Note: we assume break_lease will only call this *once* for any given
2014  * lease.
2015  */
2016 static
2017 void nfsd_break_deleg_cb(struct file_lock *fl)
2018 {
2019         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2020
2021         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2022         if (!dp)
2023                 return;
2024
2025         /* We're assuming the state code never drops its reference
2026          * without first removing the lease.  Since we're in this lease
2027          * callback (and since the lease code is serialized by the kernel
2028          * lock) we know the server hasn't removed the lease yet, we know
2029          * it's safe to take a reference: */
2030         atomic_inc(&dp->dl_count);
2031         atomic_inc(&dp->dl_client->cl_count);
2032
2033         spin_lock(&recall_lock);
2034         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2035         spin_unlock(&recall_lock);
2036
2037         /* only place dl_time is set. protected by lock_kernel*/
2038         dp->dl_time = get_seconds();
2039
2040         /*
2041          * We don't want the locks code to timeout the lease for us;
2042          * we'll remove it ourself if the delegation isn't returned
2043          * in time.
2044          */
2045         fl->fl_break_time = 0;
2046
2047         dp->dl_file->fi_had_conflict = true;
2048         nfsd4_cb_recall(dp);
2049 }
2050
2051 /*
2052  * The file_lock is being reapd.
2053  *
2054  * Called by locks_free_lock() with lock_kernel() held.
2055  */
2056 static
2057 void nfsd_release_deleg_cb(struct file_lock *fl)
2058 {
2059         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2060
2061         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2062
2063         if (!(fl->fl_flags & FL_LEASE) || !dp)
2064                 return;
2065         dp->dl_flock = NULL;
2066 }
2067
2068 /*
2069  * Set the delegation file_lock back pointer.
2070  *
2071  * Called from setlease() with lock_kernel() held.
2072  */
2073 static
2074 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2075 {
2076         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2077
2078         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2079         if (!dp)
2080                 return;
2081         dp->dl_flock = new;
2082 }
2083
2084 /*
2085  * Called from setlease() with lock_kernel() held
2086  */
2087 static
2088 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2089 {
2090         struct nfs4_delegation *onlistd =
2091                 (struct nfs4_delegation *)onlist->fl_owner;
2092         struct nfs4_delegation *tryd =
2093                 (struct nfs4_delegation *)try->fl_owner;
2094
2095         if (onlist->fl_lmops != try->fl_lmops)
2096                 return 0;
2097
2098         return onlistd->dl_client == tryd->dl_client;
2099 }
2100
2101
2102 static
2103 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2104 {
2105         if (arg & F_UNLCK)
2106                 return lease_modify(onlist, arg);
2107         else
2108                 return -EAGAIN;
2109 }
2110
2111 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2112         .fl_break = nfsd_break_deleg_cb,
2113         .fl_release_private = nfsd_release_deleg_cb,
2114         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2115         .fl_mylease = nfsd_same_client_deleg_cb,
2116         .fl_change = nfsd_change_deleg_cb,
2117 };
2118
2119
2120 __be32
2121 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2122                     struct nfsd4_open *open)
2123 {
2124         clientid_t *clientid = &open->op_clientid;
2125         struct nfs4_client *clp = NULL;
2126         unsigned int strhashval;
2127         struct nfs4_stateowner *sop = NULL;
2128
2129         if (!check_name(open->op_owner))
2130                 return nfserr_inval;
2131
2132         if (STALE_CLIENTID(&open->op_clientid))
2133                 return nfserr_stale_clientid;
2134
2135         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2136         sop = find_openstateowner_str(strhashval, open);
2137         open->op_stateowner = sop;
2138         if (!sop) {
2139                 /* Make sure the client's lease hasn't expired. */
2140                 clp = find_confirmed_client(clientid);
2141                 if (clp == NULL)
2142                         return nfserr_expired;
2143                 goto renew;
2144         }
2145         /* When sessions are used, skip open sequenceid processing */
2146         if (nfsd4_has_session(cstate))
2147                 goto renew;
2148         if (!sop->so_confirmed) {
2149                 /* Replace unconfirmed owners without checking for replay. */
2150                 clp = sop->so_client;
2151                 release_openowner(sop);
2152                 open->op_stateowner = NULL;
2153                 goto renew;
2154         }
2155         if (open->op_seqid == sop->so_seqid - 1) {
2156                 if (sop->so_replay.rp_buflen)
2157                         return nfserr_replay_me;
2158                 /* The original OPEN failed so spectacularly
2159                  * that we don't even have replay data saved!
2160                  * Therefore, we have no choice but to continue
2161                  * processing this OPEN; presumably, we'll
2162                  * fail again for the same reason.
2163                  */
2164                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2165                 goto renew;
2166         }
2167         if (open->op_seqid != sop->so_seqid)
2168                 return nfserr_bad_seqid;
2169 renew:
2170         if (open->op_stateowner == NULL) {
2171                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2172                 if (sop == NULL)
2173                         return nfserr_resource;
2174                 open->op_stateowner = sop;
2175         }
2176         list_del_init(&sop->so_close_lru);
2177         renew_client(sop->so_client);
2178         return nfs_ok;
2179 }
2180
2181 static inline __be32
2182 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2183 {
2184         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2185                 return nfserr_openmode;
2186         else
2187                 return nfs_ok;
2188 }
2189
2190 static struct nfs4_delegation *
2191 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2192 {
2193         struct nfs4_delegation *dp;
2194
2195         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2196                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2197                         return dp;
2198         }
2199         return NULL;
2200 }
2201
2202 static __be32
2203 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2204                 struct nfs4_delegation **dp)
2205 {
2206         int flags;
2207         __be32 status = nfserr_bad_stateid;
2208
2209         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2210         if (*dp == NULL)
2211                 goto out;
2212         flags = open->op_share_access == NFS4_SHARE_ACCESS_READ ?
2213                                                 RD_STATE : WR_STATE;
2214         status = nfs4_check_delegmode(*dp, flags);
2215         if (status)
2216                 *dp = NULL;
2217 out:
2218         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2219                 return nfs_ok;
2220         if (status)
2221                 return status;
2222         open->op_stateowner->so_confirmed = 1;
2223         return nfs_ok;
2224 }
2225
2226 static __be32
2227 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2228 {
2229         struct nfs4_stateid *local;
2230         __be32 status = nfserr_share_denied;
2231         struct nfs4_stateowner *sop = open->op_stateowner;
2232
2233         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2234                 /* ignore lock owners */
2235                 if (local->st_stateowner->so_is_open_owner == 0)
2236                         continue;
2237                 /* remember if we have seen this open owner */
2238                 if (local->st_stateowner == sop)
2239                         *stpp = local;
2240                 /* check for conflicting share reservations */
2241                 if (!test_share(local, open))
2242                         goto out;
2243         }
2244         status = 0;
2245 out:
2246         return status;
2247 }
2248
2249 static inline struct nfs4_stateid *
2250 nfs4_alloc_stateid(void)
2251 {
2252         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2253 }
2254
2255 static __be32
2256 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2257                 struct nfs4_delegation *dp,
2258                 struct svc_fh *cur_fh, int flags)
2259 {
2260         struct nfs4_stateid *stp;
2261
2262         stp = nfs4_alloc_stateid();
2263         if (stp == NULL)
2264                 return nfserr_resource;
2265
2266         if (dp) {
2267                 get_file(dp->dl_vfs_file);
2268                 stp->st_vfs_file = dp->dl_vfs_file;
2269         } else {
2270                 __be32 status;
2271                 status = nfsd_open(rqstp, cur_fh, S_IFREG, flags,
2272                                 &stp->st_vfs_file);
2273                 if (status) {
2274                         if (status == nfserr_dropit)
2275                                 status = nfserr_jukebox;
2276                         kmem_cache_free(stateid_slab, stp);
2277                         return status;
2278                 }
2279         }
2280         *stpp = stp;
2281         return 0;
2282 }
2283
2284 static inline __be32
2285 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2286                 struct nfsd4_open *open)
2287 {
2288         struct iattr iattr = {
2289                 .ia_valid = ATTR_SIZE,
2290                 .ia_size = 0,
2291         };
2292         if (!open->op_truncate)
2293                 return 0;
2294         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2295                 return nfserr_inval;
2296         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2297 }
2298
2299 static __be32
2300 nfs4_upgrade_open(struct svc_rqst *rqstp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2301 {
2302         struct file *filp = stp->st_vfs_file;
2303         struct inode *inode = filp->f_path.dentry->d_inode;
2304         unsigned int share_access, new_writer;
2305         __be32 status;
2306
2307         set_access(&share_access, stp->st_access_bmap);
2308         new_writer = (~share_access) & open->op_share_access
2309                         & NFS4_SHARE_ACCESS_WRITE;
2310
2311         if (new_writer) {
2312                 int err = get_write_access(inode);
2313                 if (err)
2314                         return nfserrno(err);
2315                 err = mnt_want_write(cur_fh->fh_export->ex_path.mnt);
2316                 if (err)
2317                         return nfserrno(err);
2318                 file_take_write(filp);
2319         }
2320         status = nfsd4_truncate(rqstp, cur_fh, open);
2321         if (status) {
2322                 if (new_writer)
2323                         put_write_access(inode);
2324                 return status;
2325         }
2326         /* remember the open */
2327         filp->f_mode |= open->op_share_access;
2328         __set_bit(open->op_share_access, &stp->st_access_bmap);
2329         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2330
2331         return nfs_ok;
2332 }
2333
2334
2335 static void
2336 nfs4_set_claim_prev(struct nfsd4_open *open)
2337 {
2338         open->op_stateowner->so_confirmed = 1;
2339         open->op_stateowner->so_client->cl_firststate = 1;
2340 }
2341
2342 /*
2343  * Attempt to hand out a delegation.
2344  */
2345 static void
2346 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2347 {
2348         struct nfs4_delegation *dp;
2349         struct nfs4_stateowner *sop = stp->st_stateowner;
2350         struct nfs4_cb_conn *cb = &sop->so_client->cl_cb_conn;
2351         struct file_lock fl, *flp = &fl;
2352         int status, flag = 0;
2353
2354         flag = NFS4_OPEN_DELEGATE_NONE;
2355         open->op_recall = 0;
2356         switch (open->op_claim_type) {
2357                 case NFS4_OPEN_CLAIM_PREVIOUS:
2358                         if (!atomic_read(&cb->cb_set))
2359                                 open->op_recall = 1;
2360                         flag = open->op_delegate_type;
2361                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2362                                 goto out;
2363                         break;
2364                 case NFS4_OPEN_CLAIM_NULL:
2365                         /* Let's not give out any delegations till everyone's
2366                          * had the chance to reclaim theirs.... */
2367                         if (locks_in_grace())
2368                                 goto out;
2369                         if (!atomic_read(&cb->cb_set) || !sop->so_confirmed)
2370                                 goto out;
2371                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2372                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2373                         else
2374                                 flag = NFS4_OPEN_DELEGATE_READ;
2375                         break;
2376                 default:
2377                         goto out;
2378         }
2379
2380         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2381         if (dp == NULL) {
2382                 flag = NFS4_OPEN_DELEGATE_NONE;
2383                 goto out;
2384         }
2385         locks_init_lock(&fl);
2386         fl.fl_lmops = &nfsd_lease_mng_ops;
2387         fl.fl_flags = FL_LEASE;
2388         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2389         fl.fl_end = OFFSET_MAX;
2390         fl.fl_owner =  (fl_owner_t)dp;
2391         fl.fl_file = stp->st_vfs_file;
2392         fl.fl_pid = current->tgid;
2393
2394         /* vfs_setlease checks to see if delegation should be handed out.
2395          * the lock_manager callbacks fl_mylease and fl_change are used
2396          */
2397         if ((status = vfs_setlease(stp->st_vfs_file, fl.fl_type, &flp))) {
2398                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2399                 unhash_delegation(dp);
2400                 flag = NFS4_OPEN_DELEGATE_NONE;
2401                 goto out;
2402         }
2403
2404         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2405
2406         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2407                 STATEID_VAL(&dp->dl_stateid));
2408 out:
2409         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2410                         && flag == NFS4_OPEN_DELEGATE_NONE
2411                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2412                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2413         open->op_delegate_type = flag;
2414 }
2415
2416 /*
2417  * called with nfs4_lock_state() held.
2418  */
2419 __be32
2420 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2421 {
2422         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2423         struct nfs4_file *fp = NULL;
2424         struct inode *ino = current_fh->fh_dentry->d_inode;
2425         struct nfs4_stateid *stp = NULL;
2426         struct nfs4_delegation *dp = NULL;
2427         __be32 status;
2428
2429         status = nfserr_inval;
2430         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2431                         || !deny_valid(open->op_share_deny))
2432                 goto out;
2433         /*
2434          * Lookup file; if found, lookup stateid and check open request,
2435          * and check for delegations in the process of being recalled.
2436          * If not found, create the nfs4_file struct
2437          */
2438         fp = find_file(ino);
2439         if (fp) {
2440                 if ((status = nfs4_check_open(fp, open, &stp)))
2441                         goto out;
2442                 status = nfs4_check_deleg(fp, open, &dp);
2443                 if (status)
2444                         goto out;
2445         } else {
2446                 status = nfserr_bad_stateid;
2447                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2448                         goto out;
2449                 status = nfserr_resource;
2450                 fp = alloc_init_file(ino);
2451                 if (fp == NULL)
2452                         goto out;
2453         }
2454
2455         /*
2456          * OPEN the file, or upgrade an existing OPEN.
2457          * If truncate fails, the OPEN fails.
2458          */
2459         if (stp) {
2460                 /* Stateid was found, this is an OPEN upgrade */
2461                 status = nfs4_upgrade_open(rqstp, current_fh, stp, open);
2462                 if (status)
2463                         goto out;
2464                 update_stateid(&stp->st_stateid);
2465         } else {
2466                 /* Stateid was not found, this is a new OPEN */
2467                 int flags = 0;
2468                 if (open->op_share_access & NFS4_SHARE_ACCESS_READ)
2469                         flags |= NFSD_MAY_READ;
2470                 if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2471                         flags |= NFSD_MAY_WRITE;
2472                 status = nfs4_new_open(rqstp, &stp, dp, current_fh, flags);
2473                 if (status)
2474                         goto out;
2475                 init_stateid(stp, fp, open);
2476                 status = nfsd4_truncate(rqstp, current_fh, open);
2477                 if (status) {
2478                         release_open_stateid(stp);
2479                         goto out;
2480                 }
2481                 if (nfsd4_has_session(&resp->cstate))
2482                         update_stateid(&stp->st_stateid);
2483         }
2484         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2485
2486         if (nfsd4_has_session(&resp->cstate)) {
2487                 open->op_stateowner->so_confirmed = 1;
2488                 nfsd4_create_clid_dir(open->op_stateowner->so_client);
2489         }
2490
2491         /*
2492         * Attempt to hand out a delegation. No error return, because the
2493         * OPEN succeeds even if we fail.
2494         */
2495         nfs4_open_delegation(current_fh, open, stp);
2496
2497         status = nfs_ok;
2498
2499         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2500                 STATEID_VAL(&stp->st_stateid));
2501 out:
2502         if (fp)
2503                 put_nfs4_file(fp);
2504         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2505                 nfs4_set_claim_prev(open);
2506         /*
2507         * To finish the open response, we just need to set the rflags.
2508         */
2509         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2510         if (!open->op_stateowner->so_confirmed &&
2511             !nfsd4_has_session(&resp->cstate))
2512                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2513
2514         return status;
2515 }
2516
2517 __be32
2518 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2519             clientid_t *clid)
2520 {
2521         struct nfs4_client *clp;
2522         __be32 status;
2523
2524         nfs4_lock_state();
2525         dprintk("process_renew(%08x/%08x): starting\n", 
2526                         clid->cl_boot, clid->cl_id);
2527         status = nfserr_stale_clientid;
2528         if (STALE_CLIENTID(clid))
2529                 goto out;
2530         clp = find_confirmed_client(clid);
2531         status = nfserr_expired;
2532         if (clp == NULL) {
2533                 /* We assume the client took too long to RENEW. */
2534                 dprintk("nfsd4_renew: clientid not found!\n");
2535                 goto out;
2536         }
2537         renew_client(clp);
2538         status = nfserr_cb_path_down;
2539         if (!list_empty(&clp->cl_delegations)
2540                         && !atomic_read(&clp->cl_cb_conn.cb_set))
2541                 goto out;
2542         status = nfs_ok;
2543 out:
2544         nfs4_unlock_state();
2545         return status;
2546 }
2547
2548 struct lock_manager nfsd4_manager = {
2549 };
2550
2551 static void
2552 nfsd4_end_grace(void)
2553 {
2554         dprintk("NFSD: end of grace period\n");
2555         nfsd4_recdir_purge_old();
2556         locks_end_grace(&nfsd4_manager);
2557 }
2558
2559 static time_t
2560 nfs4_laundromat(void)
2561 {
2562         struct nfs4_client *clp;
2563         struct nfs4_stateowner *sop;
2564         struct nfs4_delegation *dp;
2565         struct list_head *pos, *next, reaplist;
2566         time_t cutoff = get_seconds() - NFSD_LEASE_TIME;
2567         time_t t, clientid_val = NFSD_LEASE_TIME;
2568         time_t u, test_val = NFSD_LEASE_TIME;
2569
2570         nfs4_lock_state();
2571
2572         dprintk("NFSD: laundromat service - starting\n");
2573         if (locks_in_grace())
2574                 nfsd4_end_grace();
2575         list_for_each_safe(pos, next, &client_lru) {
2576                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2577                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2578                         t = clp->cl_time - cutoff;
2579                         if (clientid_val > t)
2580                                 clientid_val = t;
2581                         break;
2582                 }
2583                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2584                         clp->cl_clientid.cl_id);
2585                 nfsd4_remove_clid_dir(clp);
2586                 expire_client(clp);
2587         }
2588         INIT_LIST_HEAD(&reaplist);
2589         spin_lock(&recall_lock);
2590         list_for_each_safe(pos, next, &del_recall_lru) {
2591                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2592                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2593                         u = dp->dl_time - cutoff;
2594                         if (test_val > u)
2595                                 test_val = u;
2596                         break;
2597                 }
2598                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2599                                     dp, dp->dl_flock);
2600                 list_move(&dp->dl_recall_lru, &reaplist);
2601         }
2602         spin_unlock(&recall_lock);
2603         list_for_each_safe(pos, next, &reaplist) {
2604                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2605                 list_del_init(&dp->dl_recall_lru);
2606                 unhash_delegation(dp);
2607         }
2608         test_val = NFSD_LEASE_TIME;
2609         list_for_each_safe(pos, next, &close_lru) {
2610                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2611                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2612                         u = sop->so_time - cutoff;
2613                         if (test_val > u)
2614                                 test_val = u;
2615                         break;
2616                 }
2617                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2618                         sop->so_id);
2619                 release_openowner(sop);
2620         }
2621         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2622                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2623         nfs4_unlock_state();
2624         return clientid_val;
2625 }
2626
2627 static struct workqueue_struct *laundry_wq;
2628 static void laundromat_main(struct work_struct *);
2629 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2630
2631 static void
2632 laundromat_main(struct work_struct *not_used)
2633 {
2634         time_t t;
2635
2636         t = nfs4_laundromat();
2637         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2638         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2639 }
2640
2641 static struct nfs4_stateowner *
2642 search_close_lru(u32 st_id, int flags)
2643 {
2644         struct nfs4_stateowner *local = NULL;
2645
2646         if (flags & CLOSE_STATE) {
2647                 list_for_each_entry(local, &close_lru, so_close_lru) {
2648                         if (local->so_id == st_id)
2649                                 return local;
2650                 }
2651         }
2652         return NULL;
2653 }
2654
2655 static inline int
2656 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2657 {
2658         return fhp->fh_dentry->d_inode != stp->st_vfs_file->f_path.dentry->d_inode;
2659 }
2660
2661 static int
2662 STALE_STATEID(stateid_t *stateid)
2663 {
2664         if (time_after((unsigned long)boot_time,
2665                         (unsigned long)stateid->si_boot)) {
2666                 dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2667                         STATEID_VAL(stateid));
2668                 return 1;
2669         }
2670         return 0;
2671 }
2672
2673 static int
2674 EXPIRED_STATEID(stateid_t *stateid)
2675 {
2676         if (time_before((unsigned long)boot_time,
2677                         ((unsigned long)stateid->si_boot)) &&
2678             time_before((unsigned long)(stateid->si_boot + lease_time), get_seconds())) {
2679                 dprintk("NFSD: expired stateid " STATEID_FMT "!\n",
2680                         STATEID_VAL(stateid));
2681                 return 1;
2682         }
2683         return 0;
2684 }
2685
2686 static __be32
2687 stateid_error_map(stateid_t *stateid)
2688 {
2689         if (STALE_STATEID(stateid))
2690                 return nfserr_stale_stateid;
2691         if (EXPIRED_STATEID(stateid))
2692                 return nfserr_expired;
2693
2694         dprintk("NFSD: bad stateid " STATEID_FMT "!\n",
2695                 STATEID_VAL(stateid));
2696         return nfserr_bad_stateid;
2697 }
2698
2699 static inline int
2700 access_permit_read(unsigned long access_bmap)
2701 {
2702         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2703                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2704                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2705 }
2706
2707 static inline int
2708 access_permit_write(unsigned long access_bmap)
2709 {
2710         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2711                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2712 }
2713
2714 static
2715 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2716 {
2717         __be32 status = nfserr_openmode;
2718
2719         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2720                 goto out;
2721         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2722                 goto out;
2723         status = nfs_ok;
2724 out:
2725         return status;
2726 }
2727
2728 static inline __be32
2729 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2730 {
2731         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2732                 return nfs_ok;
2733         else if (locks_in_grace()) {
2734                 /* Answer in remaining cases depends on existance of
2735                  * conflicting state; so we must wait out the grace period. */
2736                 return nfserr_grace;
2737         } else if (flags & WR_STATE)
2738                 return nfs4_share_conflict(current_fh,
2739                                 NFS4_SHARE_DENY_WRITE);
2740         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2741                 return nfs4_share_conflict(current_fh,
2742                                 NFS4_SHARE_DENY_READ);
2743 }
2744
2745 /*
2746  * Allow READ/WRITE during grace period on recovered state only for files
2747  * that are not able to provide mandatory locking.
2748  */
2749 static inline int
2750 grace_disallows_io(struct inode *inode)
2751 {
2752         return locks_in_grace() && mandatory_lock(inode);
2753 }
2754
2755 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2756 {
2757         /*
2758          * When sessions are used the stateid generation number is ignored
2759          * when it is zero.
2760          */
2761         if ((flags & HAS_SESSION) && in->si_generation == 0)
2762                 goto out;
2763
2764         /* If the client sends us a stateid from the future, it's buggy: */
2765         if (in->si_generation > ref->si_generation)
2766                 return nfserr_bad_stateid;
2767         /*
2768          * The following, however, can happen.  For example, if the
2769          * client sends an open and some IO at the same time, the open
2770          * may bump si_generation while the IO is still in flight.
2771          * Thanks to hard links and renames, the client never knows what
2772          * file an open will affect.  So it could avoid that situation
2773          * only by serializing all opens and IO from the same open
2774          * owner.  To recover from the old_stateid error, the client
2775          * will just have to retry the IO:
2776          */
2777         if (in->si_generation < ref->si_generation)
2778                 return nfserr_old_stateid;
2779 out:
2780         return nfs_ok;
2781 }
2782
2783 static int is_delegation_stateid(stateid_t *stateid)
2784 {
2785         return stateid->si_fileid == 0;
2786 }
2787
2788 /*
2789 * Checks for stateid operations
2790 */
2791 __be32
2792 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2793                            stateid_t *stateid, int flags, struct file **filpp)
2794 {
2795         struct nfs4_stateid *stp = NULL;
2796         struct nfs4_delegation *dp = NULL;
2797         struct svc_fh *current_fh = &cstate->current_fh;
2798         struct inode *ino = current_fh->fh_dentry->d_inode;
2799         __be32 status;
2800
2801         if (filpp)
2802                 *filpp = NULL;
2803
2804         if (grace_disallows_io(ino))
2805                 return nfserr_grace;
2806
2807         if (nfsd4_has_session(cstate))
2808                 flags |= HAS_SESSION;
2809
2810         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2811                 return check_special_stateids(current_fh, stateid, flags);
2812
2813         status = nfserr_stale_stateid;
2814         if (STALE_STATEID(stateid)) 
2815                 goto out;
2816
2817         status = nfserr_bad_stateid;
2818         if (is_delegation_stateid(stateid)) {
2819                 dp = find_delegation_stateid(ino, stateid);
2820                 if (!dp) {
2821                         status = stateid_error_map(stateid);
2822                         goto out;
2823                 }
2824                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2825                                                   flags);
2826                 if (status)
2827                         goto out;
2828                 status = nfs4_check_delegmode(dp, flags);
2829                 if (status)
2830                         goto out;
2831                 renew_client(dp->dl_client);
2832                 if (filpp)
2833                         *filpp = dp->dl_vfs_file;
2834         } else { /* open or lock stateid */
2835                 stp = find_stateid(stateid, flags);
2836                 if (!stp) {
2837                         status = stateid_error_map(stateid);
2838                         goto out;
2839                 }
2840                 if (nfs4_check_fh(current_fh, stp))
2841                         goto out;
2842                 if (!stp->st_stateowner->so_confirmed)
2843                         goto out;
2844                 status = check_stateid_generation(stateid, &stp->st_stateid,
2845                                                   flags);
2846                 if (status)
2847                         goto out;
2848                 status = nfs4_check_openmode(stp, flags);
2849                 if (status)
2850                         goto out;
2851                 renew_client(stp->st_stateowner->so_client);
2852                 if (filpp)
2853                         *filpp = stp->st_vfs_file;
2854         }
2855         status = nfs_ok;
2856 out:
2857         return status;
2858 }
2859
2860 static inline int
2861 setlkflg (int type)
2862 {
2863         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2864                 RD_STATE : WR_STATE;
2865 }
2866
2867 /* 
2868  * Checks for sequence id mutating operations. 
2869  */
2870 static __be32
2871 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
2872                          stateid_t *stateid, int flags,
2873                          struct nfs4_stateowner **sopp,
2874                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
2875 {
2876         struct nfs4_stateid *stp;
2877         struct nfs4_stateowner *sop;
2878         struct svc_fh *current_fh = &cstate->current_fh;
2879         __be32 status;
2880
2881         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
2882                 seqid, STATEID_VAL(stateid));
2883
2884         *stpp = NULL;
2885         *sopp = NULL;
2886
2887         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
2888                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
2889                 return nfserr_bad_stateid;
2890         }
2891
2892         if (STALE_STATEID(stateid))
2893                 return nfserr_stale_stateid;
2894
2895         if (nfsd4_has_session(cstate))
2896                 flags |= HAS_SESSION;
2897
2898         /*
2899         * We return BAD_STATEID if filehandle doesn't match stateid, 
2900         * the confirmed flag is incorrecly set, or the generation 
2901         * number is incorrect.  
2902         */
2903         stp = find_stateid(stateid, flags);
2904         if (stp == NULL) {
2905                 /*
2906                  * Also, we should make sure this isn't just the result of
2907                  * a replayed close:
2908                  */
2909                 sop = search_close_lru(stateid->si_stateownerid, flags);
2910                 if (sop == NULL)
2911                         return stateid_error_map(stateid);
2912                 *sopp = sop;
2913                 goto check_replay;
2914         }
2915
2916         *stpp = stp;
2917         *sopp = sop = stp->st_stateowner;
2918
2919         if (lock) {
2920                 clientid_t *lockclid = &lock->v.new.clientid;
2921                 struct nfs4_client *clp = sop->so_client;
2922                 int lkflg = 0;
2923                 __be32 status;
2924
2925                 lkflg = setlkflg(lock->lk_type);
2926
2927                 if (lock->lk_is_new) {
2928                         if (!sop->so_is_open_owner)
2929                                 return nfserr_bad_stateid;
2930                         if (!(flags & HAS_SESSION) &&
2931                             !same_clid(&clp->cl_clientid, lockclid))
2932                                 return nfserr_bad_stateid;
2933                         /* stp is the open stateid */
2934                         status = nfs4_check_openmode(stp, lkflg);
2935                         if (status)
2936                                 return status;
2937                 } else {
2938                         /* stp is the lock stateid */
2939                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
2940                         if (status)
2941                                 return status;
2942                }
2943         }
2944
2945         if (nfs4_check_fh(current_fh, stp)) {
2946                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
2947                 return nfserr_bad_stateid;
2948         }
2949
2950         /*
2951         *  We now validate the seqid and stateid generation numbers.
2952         *  For the moment, we ignore the possibility of 
2953         *  generation number wraparound.
2954         */
2955         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
2956                 goto check_replay;
2957
2958         if (sop->so_confirmed && flags & CONFIRM) {
2959                 dprintk("NFSD: preprocess_seqid_op: expected"
2960                                 " unconfirmed stateowner!\n");
2961                 return nfserr_bad_stateid;
2962         }
2963         if (!sop->so_confirmed && !(flags & CONFIRM)) {
2964                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
2965                                 " confirmed yet!\n");
2966                 return nfserr_bad_stateid;
2967         }
2968         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
2969         if (status)
2970                 return status;
2971         renew_client(sop->so_client);
2972         return nfs_ok;
2973
2974 check_replay:
2975         if (seqid == sop->so_seqid - 1) {
2976                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
2977                 /* indicate replay to calling function */
2978                 return nfserr_replay_me;
2979         }
2980         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
2981                         sop->so_seqid, seqid);
2982         *sopp = NULL;
2983         return nfserr_bad_seqid;
2984 }
2985
2986 __be32
2987 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2988                    struct nfsd4_open_confirm *oc)
2989 {
2990         __be32 status;
2991         struct nfs4_stateowner *sop;
2992         struct nfs4_stateid *stp;
2993
2994         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
2995                         (int)cstate->current_fh.fh_dentry->d_name.len,
2996                         cstate->current_fh.fh_dentry->d_name.name);
2997
2998         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
2999         if (status)
3000                 return status;
3001
3002         nfs4_lock_state();
3003
3004         if ((status = nfs4_preprocess_seqid_op(cstate,
3005                                         oc->oc_seqid, &oc->oc_req_stateid,
3006                                         CONFIRM | OPEN_STATE,
3007                                         &oc->oc_stateowner, &stp, NULL)))
3008                 goto out; 
3009
3010         sop = oc->oc_stateowner;
3011         sop->so_confirmed = 1;
3012         update_stateid(&stp->st_stateid);
3013         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3014         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3015                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3016
3017         nfsd4_create_clid_dir(sop->so_client);
3018 out:
3019         if (oc->oc_stateowner) {
3020                 nfs4_get_stateowner(oc->oc_stateowner);
3021                 cstate->replay_owner = oc->oc_stateowner;
3022         }
3023         nfs4_unlock_state();
3024         return status;
3025 }
3026
3027
3028 /*
3029  * unset all bits in union bitmap (bmap) that
3030  * do not exist in share (from successful OPEN_DOWNGRADE)
3031  */
3032 static void
3033 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3034 {
3035         int i;
3036         for (i = 1; i < 4; i++) {
3037                 if ((i & access) != i)
3038                         __clear_bit(i, bmap);
3039         }
3040 }
3041
3042 static void
3043 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3044 {
3045         int i;
3046         for (i = 0; i < 4; i++) {
3047                 if ((i & deny) != i)
3048                         __clear_bit(i, bmap);
3049         }
3050 }
3051
3052 __be32
3053 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3054                      struct nfsd4_compound_state *cstate,
3055                      struct nfsd4_open_downgrade *od)
3056 {
3057         __be32 status;
3058         struct nfs4_stateid *stp;
3059         unsigned int share_access;
3060
3061         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3062                         (int)cstate->current_fh.fh_dentry->d_name.len,
3063                         cstate->current_fh.fh_dentry->d_name.name);
3064
3065         if (!access_valid(od->od_share_access, cstate->minorversion)
3066                         || !deny_valid(od->od_share_deny))
3067                 return nfserr_inval;
3068
3069         nfs4_lock_state();
3070         if ((status = nfs4_preprocess_seqid_op(cstate,
3071                                         od->od_seqid,
3072                                         &od->od_stateid, 
3073                                         OPEN_STATE,
3074                                         &od->od_stateowner, &stp, NULL)))
3075                 goto out; 
3076
3077         status = nfserr_inval;
3078         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3079                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3080                         stp->st_access_bmap, od->od_share_access);
3081                 goto out;
3082         }
3083         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3084                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3085                         stp->st_deny_bmap, od->od_share_deny);
3086                 goto out;
3087         }
3088         set_access(&share_access, stp->st_access_bmap);
3089         nfs4_file_downgrade(stp->st_vfs_file,
3090                             share_access & ~od->od_share_access);
3091
3092         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3093         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3094
3095         update_stateid(&stp->st_stateid);
3096         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3097         status = nfs_ok;
3098 out:
3099         if (od->od_stateowner) {
3100                 nfs4_get_stateowner(od->od_stateowner);
3101                 cstate->replay_owner = od->od_stateowner;
3102         }
3103         nfs4_unlock_state();
3104         return status;
3105 }
3106
3107 /*
3108  * nfs4_unlock_state() called after encode
3109  */
3110 __be32
3111 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3112             struct nfsd4_close *close)
3113 {
3114         __be32 status;
3115         struct nfs4_stateid *stp;
3116
3117         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3118                         (int)cstate->current_fh.fh_dentry->d_name.len,
3119                         cstate->current_fh.fh_dentry->d_name.name);
3120
3121         nfs4_lock_state();
3122         /* check close_lru for replay */
3123         if ((status = nfs4_preprocess_seqid_op(cstate,
3124                                         close->cl_seqid,
3125                                         &close->cl_stateid, 
3126                                         OPEN_STATE | CLOSE_STATE,
3127                                         &close->cl_stateowner, &stp, NULL)))
3128                 goto out; 
3129         status = nfs_ok;
3130         update_stateid(&stp->st_stateid);
3131         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3132
3133         /* release_stateid() calls nfsd_close() if needed */
3134         release_open_stateid(stp);
3135
3136         /* place unused nfs4_stateowners on so_close_lru list to be
3137          * released by the laundromat service after the lease period
3138          * to enable us to handle CLOSE replay
3139          */
3140         if (list_empty(&close->cl_stateowner->so_stateids))
3141                 move_to_close_lru(close->cl_stateowner);
3142 out:
3143         if (close->cl_stateowner) {
3144                 nfs4_get_stateowner(close->cl_stateowner);
3145                 cstate->replay_owner = close->cl_stateowner;
3146         }
3147         nfs4_unlock_state();
3148         return status;
3149 }
3150
3151 __be32
3152 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3153                   struct nfsd4_delegreturn *dr)
3154 {
3155         struct nfs4_delegation *dp;
3156         stateid_t *stateid = &dr->dr_stateid;
3157         struct inode *inode;
3158         __be32 status;
3159         int flags = 0;
3160
3161         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3162                 return status;
3163         inode = cstate->current_fh.fh_dentry->d_inode;
3164
3165         if (nfsd4_has_session(cstate))
3166                 flags |= HAS_SESSION;
3167         nfs4_lock_state();
3168         status = nfserr_bad_stateid;
3169         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3170                 goto out;
3171         status = nfserr_stale_stateid;
3172         if (STALE_STATEID(stateid))
3173                 goto out;
3174         status = nfserr_bad_stateid;
3175         if (!is_delegation_stateid(stateid))
3176                 goto out;
3177         dp = find_delegation_stateid(inode, stateid);
3178         if (!dp) {
3179                 status = stateid_error_map(stateid);
3180                 goto out;
3181         }
3182         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3183         if (status)
3184                 goto out;
3185         renew_client(dp->dl_client);
3186
3187         unhash_delegation(dp);
3188 out:
3189         nfs4_unlock_state();
3190
3191         return status;
3192 }
3193
3194
3195 /* 
3196  * Lock owner state (byte-range locks)
3197  */
3198 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3199 #define LOCK_HASH_BITS              8
3200 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3201 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3202
3203 static inline u64
3204 end_offset(u64 start, u64 len)
3205 {
3206         u64 end;
3207
3208         end = start + len;
3209         return end >= start ? end: NFS4_MAX_UINT64;
3210 }
3211
3212 /* last octet in a range */
3213 static inline u64
3214 last_byte_offset(u64 start, u64 len)
3215 {
3216         u64 end;
3217
3218         BUG_ON(!len);
3219         end = start + len;
3220         return end > start ? end - 1: NFS4_MAX_UINT64;
3221 }
3222
3223 #define lockownerid_hashval(id) \
3224         ((id) & LOCK_HASH_MASK)
3225
3226 static inline unsigned int
3227 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3228                 struct xdr_netobj *ownername)
3229 {
3230         return (file_hashval(inode) + cl_id
3231                         + opaque_hashval(ownername->data, ownername->len))
3232                 & LOCK_HASH_MASK;
3233 }
3234
3235 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3236 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3237 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3238
3239 static struct nfs4_stateid *
3240 find_stateid(stateid_t *stid, int flags)
3241 {
3242         struct nfs4_stateid *local;
3243         u32 st_id = stid->si_stateownerid;
3244         u32 f_id = stid->si_fileid;
3245         unsigned int hashval;
3246
3247         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3248         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3249                 hashval = stateid_hashval(st_id, f_id);
3250                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3251                         if ((local->st_stateid.si_stateownerid == st_id) &&
3252                             (local->st_stateid.si_fileid == f_id))
3253                                 return local;
3254                 }
3255         } 
3256
3257         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3258                 hashval = stateid_hashval(st_id, f_id);
3259                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3260                         if ((local->st_stateid.si_stateownerid == st_id) &&
3261                             (local->st_stateid.si_fileid == f_id))
3262                                 return local;
3263                 }
3264         }
3265         return NULL;
3266 }
3267
3268 static struct nfs4_delegation *
3269 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3270 {
3271         struct nfs4_file *fp;
3272         struct nfs4_delegation *dl;
3273
3274         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3275                 STATEID_VAL(stid));
3276
3277         fp = find_file(ino);
3278         if (!fp)
3279                 return NULL;
3280         dl = find_delegation_file(fp, stid);
3281         put_nfs4_file(fp);
3282         return dl;
3283 }
3284
3285 /*
3286  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3287  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3288  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3289  * locking, this prevents us from being completely protocol-compliant.  The
3290  * real solution to this problem is to start using unsigned file offsets in
3291  * the VFS, but this is a very deep change!
3292  */
3293 static inline void
3294 nfs4_transform_lock_offset(struct file_lock *lock)
3295 {
3296         if (lock->fl_start < 0)
3297                 lock->fl_start = OFFSET_MAX;
3298         if (lock->fl_end < 0)
3299                 lock->fl_end = OFFSET_MAX;
3300 }
3301
3302 /* Hack!: For now, we're defining this just so we can use a pointer to it
3303  * as a unique cookie to identify our (NFSv4's) posix locks. */
3304 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3305 };
3306
3307 static inline void
3308 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3309 {
3310         struct nfs4_stateowner *sop;
3311         unsigned int hval;
3312
3313         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3314                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3315                 hval = lockownerid_hashval(sop->so_id);
3316                 kref_get(&sop->so_ref);
3317                 deny->ld_sop = sop;
3318                 deny->ld_clientid = sop->so_client->cl_clientid;
3319         } else {
3320                 deny->ld_sop = NULL;
3321                 deny->ld_clientid.cl_boot = 0;
3322                 deny->ld_clientid.cl_id = 0;
3323         }
3324         deny->ld_start = fl->fl_start;
3325         deny->ld_length = NFS4_MAX_UINT64;
3326         if (fl->fl_end != NFS4_MAX_UINT64)
3327                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3328         deny->ld_type = NFS4_READ_LT;
3329         if (fl->fl_type != F_RDLCK)
3330                 deny->ld_type = NFS4_WRITE_LT;
3331 }
3332
3333 static struct nfs4_stateowner *
3334 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3335                 struct xdr_netobj *owner)
3336 {
3337         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3338         struct nfs4_stateowner *op;
3339
3340         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3341                 if (same_owner_str(op, owner, clid))
3342                         return op;
3343         }
3344         return NULL;
3345 }
3346
3347 /*
3348  * Alloc a lock owner structure.
3349  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3350  * occured. 
3351  *
3352  * strhashval = lock_ownerstr_hashval 
3353  */
3354
3355 static struct nfs4_stateowner *
3356 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3357         struct nfs4_stateowner *sop;
3358         struct nfs4_replay *rp;
3359         unsigned int idhashval;
3360
3361         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3362                 return NULL;
3363         idhashval = lockownerid_hashval(current_ownerid);
3364         INIT_LIST_HEAD(&sop->so_idhash);
3365         INIT_LIST_HEAD(&sop->so_strhash);
3366         INIT_LIST_HEAD(&sop->so_perclient);
3367         INIT_LIST_HEAD(&sop->so_stateids);
3368         INIT_LIST_HEAD(&sop->so_perstateid);
3369         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3370         sop->so_time = 0;
3371         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3372         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3373         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3374         sop->so_is_open_owner = 0;
3375         sop->so_id = current_ownerid++;
3376         sop->so_client = clp;
3377         /* It is the openowner seqid that will be incremented in encode in the
3378          * case of new lockowners; so increment the lock seqid manually: */
3379         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3380         sop->so_confirmed = 1;
3381         rp = &sop->so_replay;
3382         rp->rp_status = nfserr_serverfault;
3383         rp->rp_buflen = 0;
3384         rp->rp_buf = rp->rp_ibuf;
3385         return sop;
3386 }
3387
3388 static struct nfs4_stateid *
3389 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3390 {
3391         struct nfs4_stateid *stp;
3392         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3393
3394         stp = nfs4_alloc_stateid();
3395         if (stp == NULL)
3396                 goto out;
3397         INIT_LIST_HEAD(&stp->st_hash);
3398         INIT_LIST_HEAD(&stp->st_perfile);
3399         INIT_LIST_HEAD(&stp->st_perstateowner);
3400         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3401         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3402         list_add(&stp->st_perfile, &fp->fi_stateids);
3403         list_add(&stp->st_perstateowner, &sop->so_stateids);
3404         stp->st_stateowner = sop;
3405         get_nfs4_file(fp);
3406         stp->st_file = fp;
3407         stp->st_stateid.si_boot = get_seconds();
3408         stp->st_stateid.si_stateownerid = sop->so_id;
3409         stp->st_stateid.si_fileid = fp->fi_id;
3410         stp->st_stateid.si_generation = 0;
3411         stp->st_vfs_file = open_stp->st_vfs_file; /* FIXME refcount?? */
3412         stp->st_access_bmap = open_stp->st_access_bmap;
3413         stp->st_deny_bmap = open_stp->st_deny_bmap;
3414         stp->st_openstp = open_stp;
3415
3416 out:
3417         return stp;
3418 }
3419
3420 static int
3421 check_lock_length(u64 offset, u64 length)
3422 {
3423         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3424              LOFF_OVERFLOW(offset, length)));
3425 }
3426
3427 /*
3428  *  LOCK operation 
3429  */
3430 __be32
3431 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3432            struct nfsd4_lock *lock)
3433 {
3434         struct nfs4_stateowner *open_sop = NULL;
3435         struct nfs4_stateowner *lock_sop = NULL;
3436         struct nfs4_stateid *lock_stp;
3437         struct file *filp;
3438         struct file_lock file_lock;
3439         struct file_lock conflock;
3440         __be32 status = 0;
3441         unsigned int strhashval;
3442         unsigned int cmd;
3443         int err;
3444
3445         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3446                 (long long) lock->lk_offset,
3447                 (long long) lock->lk_length);
3448
3449         if (check_lock_length(lock->lk_offset, lock->lk_length))
3450                  return nfserr_inval;
3451
3452         if ((status = fh_verify(rqstp, &cstate->current_fh,
3453                                 S_IFREG, NFSD_MAY_LOCK))) {
3454                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3455                 return status;
3456         }
3457
3458         nfs4_lock_state();
3459
3460         if (lock->lk_is_new) {
3461                 /*
3462                  * Client indicates that this is a new lockowner.
3463                  * Use open owner and open stateid to create lock owner and
3464                  * lock stateid.
3465                  */
3466                 struct nfs4_stateid *open_stp = NULL;
3467                 struct nfs4_file *fp;
3468                 
3469                 status = nfserr_stale_clientid;
3470                 if (!nfsd4_has_session(cstate) &&
3471                     STALE_CLIENTID(&lock->lk_new_clientid))
3472                         goto out;
3473
3474                 /* validate and update open stateid and open seqid */
3475                 status = nfs4_preprocess_seqid_op(cstate,
3476                                         lock->lk_new_open_seqid,
3477                                         &lock->lk_new_open_stateid,
3478                                         OPEN_STATE,
3479                                         &lock->lk_replay_owner, &open_stp,
3480                                         lock);
3481                 if (status)
3482                         goto out;
3483                 open_sop = lock->lk_replay_owner;
3484                 /* create lockowner and lock stateid */
3485                 fp = open_stp->st_file;
3486                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3487                                 open_sop->so_client->cl_clientid.cl_id, 
3488                                 &lock->v.new.owner);
3489                 /* XXX: Do we need to check for duplicate stateowners on
3490                  * the same file, or should they just be allowed (and
3491                  * create new stateids)? */
3492                 status = nfserr_resource;
3493                 lock_sop = alloc_init_lock_stateowner(strhashval,
3494                                 open_sop->so_client, open_stp, lock);
3495                 if (lock_sop == NULL)
3496                         goto out;
3497                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3498                 if (lock_stp == NULL)
3499                         goto out;
3500         } else {
3501                 /* lock (lock owner + lock stateid) already exists */
3502                 status = nfs4_preprocess_seqid_op(cstate,
3503                                        lock->lk_old_lock_seqid, 
3504                                        &lock->lk_old_lock_stateid, 
3505                                        LOCK_STATE,
3506                                        &lock->lk_replay_owner, &lock_stp, lock);
3507                 if (status)
3508                         goto out;
3509                 lock_sop = lock->lk_replay_owner;
3510         }
3511         /* lock->lk_replay_owner and lock_stp have been created or found */
3512         filp = lock_stp->st_vfs_file;
3513
3514         status = nfserr_grace;
3515         if (locks_in_grace() && !lock->lk_reclaim)
3516                 goto out;
3517         status = nfserr_no_grace;
3518         if (!locks_in_grace() && lock->lk_reclaim)
3519                 goto out;
3520
3521         locks_init_lock(&file_lock);
3522         switch (lock->lk_type) {
3523                 case NFS4_READ_LT:
3524                 case NFS4_READW_LT:
3525                         file_lock.fl_type = F_RDLCK;
3526                         cmd = F_SETLK;
3527                 break;
3528                 case NFS4_WRITE_LT:
3529                 case NFS4_WRITEW_LT:
3530                         file_lock.fl_type = F_WRLCK;
3531                         cmd = F_SETLK;
3532                 break;
3533                 default:
3534                         status = nfserr_inval;
3535                 goto out;
3536         }
3537         file_lock.fl_owner = (fl_owner_t)lock_sop;
3538         file_lock.fl_pid = current->tgid;
3539         file_lock.fl_file = filp;
3540         file_lock.fl_flags = FL_POSIX;
3541         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3542
3543         file_lock.fl_start = lock->lk_offset;
3544         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3545         nfs4_transform_lock_offset(&file_lock);
3546
3547         /*
3548         * Try to lock the file in the VFS.
3549         * Note: locks.c uses the BKL to protect the inode's lock list.
3550         */
3551
3552         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3553         switch (-err) {
3554         case 0: /* success! */
3555                 update_stateid(&lock_stp->st_stateid);
3556                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3557                                 sizeof(stateid_t));
3558                 status = 0;
3559                 break;
3560         case (EAGAIN):          /* conflock holds conflicting lock */
3561                 status = nfserr_denied;
3562                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3563                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3564                 break;
3565         case (EDEADLK):
3566                 status = nfserr_deadlock;
3567                 break;
3568         default:        
3569                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3570                 status = nfserr_resource;
3571                 break;
3572         }
3573 out:
3574         if (status && lock->lk_is_new && lock_sop)
3575                 release_lockowner(lock_sop);
3576         if (lock->lk_replay_owner) {
3577                 nfs4_get_stateowner(lock->lk_replay_owner);
3578                 cstate->replay_owner = lock->lk_replay_owner;
3579         }
3580         nfs4_unlock_state();
3581         return status;
3582 }
3583
3584 /*
3585  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3586  * so we do a temporary open here just to get an open file to pass to
3587  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3588  * inode operation.)
3589  */
3590 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3591 {
3592         struct file *file;
3593         int err;
3594
3595         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3596         if (err)
3597                 return err;
3598         err = vfs_test_lock(file, lock);
3599         nfsd_close(file);
3600         return err;
3601 }
3602
3603 /*
3604  * LOCKT operation
3605  */
3606 __be32
3607 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3608             struct nfsd4_lockt *lockt)
3609 {
3610         struct inode *inode;
3611         struct file_lock file_lock;
3612         int error;
3613         __be32 status;
3614
3615         if (locks_in_grace())
3616                 return nfserr_grace;
3617
3618         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3619                  return nfserr_inval;
3620
3621         lockt->lt_stateowner = NULL;
3622         nfs4_lock_state();
3623
3624         status = nfserr_stale_clientid;
3625         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3626                 goto out;
3627
3628         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3629                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3630                 if (status == nfserr_symlink)
3631                         status = nfserr_inval;
3632                 goto out;
3633         }
3634
3635         inode = cstate->current_fh.fh_dentry->d_inode;
3636         locks_init_lock(&file_lock);
3637         switch (lockt->lt_type) {
3638                 case NFS4_READ_LT:
3639                 case NFS4_READW_LT:
3640                         file_lock.fl_type = F_RDLCK;
3641                 break;
3642                 case NFS4_WRITE_LT:
3643                 case NFS4_WRITEW_LT:
3644                         file_lock.fl_type = F_WRLCK;
3645                 break;
3646                 default:
3647                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3648                         status = nfserr_inval;
3649                 goto out;
3650         }
3651
3652         lockt->lt_stateowner = find_lockstateowner_str(inode,
3653                         &lockt->lt_clientid, &lockt->lt_owner);
3654         if (lockt->lt_stateowner)
3655                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3656         file_lock.fl_pid = current->tgid;
3657         file_lock.fl_flags = FL_POSIX;
3658
3659         file_lock.fl_start = lockt->lt_offset;
3660         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3661
3662         nfs4_transform_lock_offset(&file_lock);
3663
3664         status = nfs_ok;
3665         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3666         if (error) {
3667                 status = nfserrno(error);
3668                 goto out;
3669         }
3670         if (file_lock.fl_type != F_UNLCK) {
3671                 status = nfserr_denied;
3672                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3673         }
3674 out:
3675         nfs4_unlock_state();
3676         return status;
3677 }
3678
3679 __be32
3680 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3681             struct nfsd4_locku *locku)
3682 {
3683         struct nfs4_stateid *stp;
3684         struct file *filp = NULL;
3685         struct file_lock file_lock;
3686         __be32 status;
3687         int err;
3688                                                         
3689         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3690                 (long long) locku->lu_offset,
3691                 (long long) locku->lu_length);
3692
3693         if (check_lock_length(locku->lu_offset, locku->lu_length))
3694                  return nfserr_inval;
3695
3696         nfs4_lock_state();
3697                                                                                 
3698         if ((status = nfs4_preprocess_seqid_op(cstate,
3699                                         locku->lu_seqid, 
3700                                         &locku->lu_stateid, 
3701                                         LOCK_STATE,
3702                                         &locku->lu_stateowner, &stp, NULL)))
3703                 goto out;
3704
3705         filp = stp->st_vfs_file;
3706         BUG_ON(!filp);
3707         locks_init_lock(&file_lock);
3708         file_lock.fl_type = F_UNLCK;
3709         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3710         file_lock.fl_pid = current->tgid;
3711         file_lock.fl_file = filp;
3712         file_lock.fl_flags = FL_POSIX; 
3713         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3714         file_lock.fl_start = locku->lu_offset;
3715
3716         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3717         nfs4_transform_lock_offset(&file_lock);
3718
3719         /*
3720         *  Try to unlock the file in the VFS.
3721         */
3722         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3723         if (err) {
3724                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3725                 goto out_nfserr;
3726         }
3727         /*
3728         * OK, unlock succeeded; the only thing left to do is update the stateid.
3729         */
3730         update_stateid(&stp->st_stateid);
3731         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3732
3733 out:
3734         if (locku->lu_stateowner) {
3735                 nfs4_get_stateowner(locku->lu_stateowner);
3736                 cstate->replay_owner = locku->lu_stateowner;
3737         }
3738         nfs4_unlock_state();
3739         return status;
3740
3741 out_nfserr:
3742         status = nfserrno(err);
3743         goto out;
3744 }
3745
3746 /*
3747  * returns
3748  *      1: locks held by lockowner
3749  *      0: no locks held by lockowner
3750  */
3751 static int
3752 check_for_locks(struct file *filp, struct nfs4_stateowner *lowner)
3753 {
3754         struct file_lock **flpp;
3755         struct inode *inode = filp->f_path.dentry->d_inode;
3756         int status = 0;
3757
3758         lock_kernel();
3759         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3760                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3761                         status = 1;
3762                         goto out;
3763                 }
3764         }
3765 out:
3766         unlock_kernel();
3767         return status;
3768 }
3769
3770 __be32
3771 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3772                         struct nfsd4_compound_state *cstate,
3773                         struct nfsd4_release_lockowner *rlockowner)
3774 {
3775         clientid_t *clid = &rlockowner->rl_clientid;
3776         struct nfs4_stateowner *sop;
3777         struct nfs4_stateid *stp;
3778         struct xdr_netobj *owner = &rlockowner->rl_owner;
3779         struct list_head matches;
3780         int i;
3781         __be32 status;
3782
3783         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3784                 clid->cl_boot, clid->cl_id);
3785
3786         /* XXX check for lease expiration */
3787
3788         status = nfserr_stale_clientid;
3789         if (STALE_CLIENTID(clid))
3790                 return status;
3791
3792         nfs4_lock_state();
3793
3794         status = nfserr_locks_held;
3795         /* XXX: we're doing a linear search through all the lockowners.
3796          * Yipes!  For now we'll just hope clients aren't really using
3797          * release_lockowner much, but eventually we have to fix these
3798          * data structures. */
3799         INIT_LIST_HEAD(&matches);
3800         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3801                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3802                         if (!same_owner_str(sop, owner, clid))
3803                                 continue;
3804                         list_for_each_entry(stp, &sop->so_stateids,
3805                                         st_perstateowner) {
3806                                 if (check_for_locks(stp->st_vfs_file, sop))
3807                                         goto out;
3808                                 /* Note: so_perclient unused for lockowners,
3809                                  * so it's OK to fool with here. */
3810                                 list_add(&sop->so_perclient, &matches);
3811                         }
3812                 }
3813         }
3814         /* Clients probably won't expect us to return with some (but not all)
3815          * of the lockowner state released; so don't release any until all
3816          * have been checked. */
3817         status = nfs_ok;
3818         while (!list_empty(&matches)) {
3819                 sop = list_entry(matches.next, struct nfs4_stateowner,
3820                                                                 so_perclient);
3821                 /* unhash_stateowner deletes so_perclient only
3822                  * for openowners. */
3823                 list_del(&sop->so_perclient);
3824                 release_lockowner(sop);
3825         }
3826 out:
3827         nfs4_unlock_state();
3828         return status;
3829 }
3830
3831 static inline struct nfs4_client_reclaim *
3832 alloc_reclaim(void)
3833 {
3834         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3835 }
3836
3837 int
3838 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3839 {
3840         unsigned int strhashval = clientstr_hashval(name);
3841         struct nfs4_client *clp;
3842
3843         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3844         return clp ? 1 : 0;
3845 }
3846
3847 /*
3848  * failure => all reset bets are off, nfserr_no_grace...
3849  */
3850 int
3851 nfs4_client_to_reclaim(const char *name)
3852 {
3853         unsigned int strhashval;
3854         struct nfs4_client_reclaim *crp = NULL;
3855
3856         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3857         crp = alloc_reclaim();
3858         if (!crp)
3859                 return 0;
3860         strhashval = clientstr_hashval(name);
3861         INIT_LIST_HEAD(&crp->cr_strhash);
3862         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3863         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
3864         reclaim_str_hashtbl_size++;
3865         return 1;
3866 }
3867
3868 static void
3869 nfs4_release_reclaim(void)
3870 {
3871         struct nfs4_client_reclaim *crp = NULL;
3872         int i;
3873
3874         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3875                 while (!list_empty(&reclaim_str_hashtbl[i])) {
3876                         crp = list_entry(reclaim_str_hashtbl[i].next,
3877                                         struct nfs4_client_reclaim, cr_strhash);
3878                         list_del(&crp->cr_strhash);
3879                         kfree(crp);
3880                         reclaim_str_hashtbl_size--;
3881                 }
3882         }
3883         BUG_ON(reclaim_str_hashtbl_size);
3884 }
3885
3886 /*
3887  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
3888 static struct nfs4_client_reclaim *
3889 nfs4_find_reclaim_client(clientid_t *clid)
3890 {
3891         unsigned int strhashval;
3892         struct nfs4_client *clp;
3893         struct nfs4_client_reclaim *crp = NULL;
3894
3895
3896         /* find clientid in conf_id_hashtbl */
3897         clp = find_confirmed_client(clid);
3898         if (clp == NULL)
3899                 return NULL;
3900
3901         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
3902                             clp->cl_name.len, clp->cl_name.data,
3903                             clp->cl_recdir);
3904
3905         /* find clp->cl_name in reclaim_str_hashtbl */
3906         strhashval = clientstr_hashval(clp->cl_recdir);
3907         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
3908                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
3909                         return crp;
3910                 }
3911         }
3912         return NULL;
3913 }
3914
3915 /*
3916 * Called from OPEN. Look for clientid in reclaim list.
3917 */
3918 __be32
3919 nfs4_check_open_reclaim(clientid_t *clid)
3920 {
3921         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
3922 }
3923
3924 /* initialization to perform at module load time: */
3925
3926 int
3927 nfs4_state_init(void)
3928 {
3929         int i, status;
3930
3931         status = nfsd4_init_slabs();
3932         if (status)
3933                 return status;
3934         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
3935                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
3936                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
3937                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
3938                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
3939                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
3940         }
3941         for (i = 0; i < SESSION_HASH_SIZE; i++)
3942                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
3943         for (i = 0; i < FILE_HASH_SIZE; i++) {
3944                 INIT_LIST_HEAD(&file_hashtbl[i]);
3945         }
3946         for (i = 0; i < OWNER_HASH_SIZE; i++) {
3947                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
3948                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
3949         }
3950         for (i = 0; i < STATEID_HASH_SIZE; i++) {
3951                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
3952                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
3953         }
3954         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3955                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
3956                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
3957         }
3958         memset(&onestateid, ~0, sizeof(stateid_t));
3959         INIT_LIST_HEAD(&close_lru);
3960         INIT_LIST_HEAD(&client_lru);
3961         INIT_LIST_HEAD(&del_recall_lru);
3962         reclaim_str_hashtbl_size = 0;
3963         return 0;
3964 }
3965
3966 static void
3967 nfsd4_load_reboot_recovery_data(void)
3968 {
3969         int status;
3970
3971         nfs4_lock_state();
3972         nfsd4_init_recdir(user_recovery_dirname);
3973         status = nfsd4_recdir_load();
3974         nfs4_unlock_state();
3975         if (status)
3976                 printk("NFSD: Failure reading reboot recovery data\n");
3977 }
3978
3979 unsigned long
3980 get_nfs4_grace_period(void)
3981 {
3982         return max(user_lease_time, lease_time) * HZ;
3983 }
3984
3985 /*
3986  * Since the lifetime of a delegation isn't limited to that of an open, a
3987  * client may quite reasonably hang on to a delegation as long as it has
3988  * the inode cached.  This becomes an obvious problem the first time a
3989  * client's inode cache approaches the size of the server's total memory.
3990  *
3991  * For now we avoid this problem by imposing a hard limit on the number
3992  * of delegations, which varies according to the server's memory size.
3993  */
3994 static void
3995 set_max_delegations(void)
3996 {
3997         /*
3998          * Allow at most 4 delegations per megabyte of RAM.  Quick
3999          * estimates suggest that in the worst case (where every delegation
4000          * is for a different inode), a delegation could take about 1.5K,
4001          * giving a worst case usage of about 6% of memory.
4002          */
4003         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4004 }
4005
4006 /* initialization to perform when the nfsd service is started: */
4007
4008 static int
4009 __nfs4_state_start(void)
4010 {
4011         unsigned long grace_time;
4012
4013         boot_time = get_seconds();
4014         grace_time = get_nfs4_grace_period();
4015         lease_time = user_lease_time;
4016         locks_start_grace(&nfsd4_manager);
4017         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4018                grace_time/HZ);
4019         laundry_wq = create_singlethread_workqueue("nfsd4");
4020         if (laundry_wq == NULL)
4021                 return -ENOMEM;
4022         queue_delayed_work(laundry_wq, &laundromat_work, grace_time);
4023         set_max_delegations();
4024         return set_callback_cred();
4025 }
4026
4027 int
4028 nfs4_state_start(void)
4029 {
4030         int ret;
4031
4032         if (nfs4_init)
4033                 return 0;
4034         nfsd4_load_reboot_recovery_data();
4035         ret = __nfs4_state_start();
4036         if (ret)
4037                 return ret;
4038         nfs4_init = 1;
4039         return 0;
4040 }
4041
4042 time_t
4043 nfs4_lease_time(void)
4044 {
4045         return lease_time;
4046 }
4047
4048 static void
4049 __nfs4_state_shutdown(void)
4050 {
4051         int i;
4052         struct nfs4_client *clp = NULL;
4053         struct nfs4_delegation *dp = NULL;
4054         struct list_head *pos, *next, reaplist;
4055
4056         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4057                 while (!list_empty(&conf_id_hashtbl[i])) {
4058                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4059                         expire_client(clp);
4060                 }
4061                 while (!list_empty(&unconf_str_hashtbl[i])) {
4062                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4063                         expire_client(clp);
4064                 }
4065         }
4066         INIT_LIST_HEAD(&reaplist);
4067         spin_lock(&recall_lock);
4068         list_for_each_safe(pos, next, &del_recall_lru) {
4069                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4070                 list_move(&dp->dl_recall_lru, &reaplist);
4071         }
4072         spin_unlock(&recall_lock);
4073         list_for_each_safe(pos, next, &reaplist) {
4074                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4075                 list_del_init(&dp->dl_recall_lru);
4076                 unhash_delegation(dp);
4077         }
4078
4079         nfsd4_shutdown_recdir();
4080         nfs4_init = 0;
4081 }
4082
4083 void
4084 nfs4_state_shutdown(void)
4085 {
4086         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4087         destroy_workqueue(laundry_wq);
4088         locks_end_grace(&nfsd4_manager);
4089         nfs4_lock_state();
4090         nfs4_release_reclaim();
4091         __nfs4_state_shutdown();
4092         nfs4_unlock_state();
4093 }
4094
4095 /*
4096  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4097  * accessed when nfsd is starting.
4098  */
4099 static void
4100 nfs4_set_recdir(char *recdir)
4101 {
4102         strcpy(user_recovery_dirname, recdir);
4103 }
4104
4105 /*
4106  * Change the NFSv4 recovery directory to recdir.
4107  */
4108 int
4109 nfs4_reset_recoverydir(char *recdir)
4110 {
4111         int status;
4112         struct path path;
4113
4114         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4115         if (status)
4116                 return status;
4117         status = -ENOTDIR;
4118         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4119                 nfs4_set_recdir(recdir);
4120                 status = 0;
4121         }
4122         path_put(&path);
4123         return status;
4124 }
4125
4126 char *
4127 nfs4_recoverydir(void)
4128 {
4129         return user_recovery_dirname;
4130 }
4131
4132 /*
4133  * Called when leasetime is changed.
4134  *
4135  * The only way the protocol gives us to handle on-the-fly lease changes is to
4136  * simulate a reboot.  Instead of doing that, we just wait till the next time
4137  * we start to register any changes in lease time.  If the administrator
4138  * really wants to change the lease time *now*, they can go ahead and bring
4139  * nfsd down and then back up again after changing the lease time.
4140  *
4141  * user_lease_time is protected by nfsd_mutex since it's only really accessed
4142  * when nfsd is starting
4143  */
4144 void
4145 nfs4_reset_lease(time_t leasetime)
4146 {
4147         user_lease_time = leasetime;
4148 }