nfsd4: fix file open accounting for RDWR opens
[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 time_t nfsd4_lease = 90;     /* default lease time */
49 time_t nfsd4_grace = 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 stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151 #define FILE_HASH_MASK                  (FILE_HASH_SIZE - 1)
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
166 {
167         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168         atomic_inc(&fp->fi_access[oflag]);
169 }
170
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
172 {
173         if (oflag == O_RDWR) {
174                 __nfs4_file_get_access(fp, O_RDONLY);
175                 __nfs4_file_get_access(fp, O_WRONLY);
176         } else
177                 __nfs4_file_get_access(fp, oflag);
178 }
179
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
181 {
182         if (fp->fi_fds[oflag]) {
183                 fput(fp->fi_fds[oflag]);
184                 fp->fi_fds[oflag] = NULL;
185         }
186 }
187
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
189 {
190         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191                 nfs4_file_put_fd(fp, O_RDWR);
192                 nfs4_file_put_fd(fp, oflag);
193         }
194 }
195
196 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
197 {
198         if (oflag == O_RDWR) {
199                 __nfs4_file_put_access(fp, O_RDONLY);
200                 __nfs4_file_put_access(fp, O_WRONLY);
201         } else
202                 __nfs4_file_put_access(fp, oflag);
203 }
204
205 static struct nfs4_delegation *
206 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
207 {
208         struct nfs4_delegation *dp;
209         struct nfs4_file *fp = stp->st_file;
210         struct nfs4_cb_conn *cb = &stp->st_stateowner->so_client->cl_cb_conn;
211
212         dprintk("NFSD alloc_init_deleg\n");
213         /*
214          * Major work on the lease subsystem (for example, to support
215          * calbacks on stat) will be required before we can support
216          * write delegations properly.
217          */
218         if (type != NFS4_OPEN_DELEGATE_READ)
219                 return NULL;
220         if (fp->fi_had_conflict)
221                 return NULL;
222         if (num_delegations > max_delegations)
223                 return NULL;
224         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
225         if (dp == NULL)
226                 return dp;
227         num_delegations++;
228         INIT_LIST_HEAD(&dp->dl_perfile);
229         INIT_LIST_HEAD(&dp->dl_perclnt);
230         INIT_LIST_HEAD(&dp->dl_recall_lru);
231         dp->dl_client = clp;
232         get_nfs4_file(fp);
233         dp->dl_file = fp;
234         nfs4_file_get_access(fp, O_RDONLY);
235         dp->dl_flock = NULL;
236         dp->dl_type = type;
237         dp->dl_ident = cb->cb_ident;
238         dp->dl_stateid.si_boot = boot_time;
239         dp->dl_stateid.si_stateownerid = current_delegid++;
240         dp->dl_stateid.si_fileid = 0;
241         dp->dl_stateid.si_generation = 0;
242         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
243         dp->dl_time = 0;
244         atomic_set(&dp->dl_count, 1);
245         list_add(&dp->dl_perfile, &fp->fi_delegations);
246         list_add(&dp->dl_perclnt, &clp->cl_delegations);
247         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
248         return dp;
249 }
250
251 void
252 nfs4_put_delegation(struct nfs4_delegation *dp)
253 {
254         if (atomic_dec_and_test(&dp->dl_count)) {
255                 dprintk("NFSD: freeing dp %p\n",dp);
256                 put_nfs4_file(dp->dl_file);
257                 kmem_cache_free(deleg_slab, dp);
258                 num_delegations--;
259         }
260 }
261
262 /* Remove the associated file_lock first, then remove the delegation.
263  * lease_modify() is called to remove the FS_LEASE file_lock from
264  * the i_flock list, eventually calling nfsd's lock_manager
265  * fl_release_callback.
266  */
267 static void
268 nfs4_close_delegation(struct nfs4_delegation *dp)
269 {
270         struct file *filp = find_readable_file(dp->dl_file);
271
272         dprintk("NFSD: close_delegation dp %p\n",dp);
273         if (dp->dl_flock)
274                 vfs_setlease(filp, F_UNLCK, &dp->dl_flock);
275         nfs4_file_put_access(dp->dl_file, O_RDONLY);
276 }
277
278 /* Called under the state lock. */
279 static void
280 unhash_delegation(struct nfs4_delegation *dp)
281 {
282         list_del_init(&dp->dl_perfile);
283         list_del_init(&dp->dl_perclnt);
284         spin_lock(&recall_lock);
285         list_del_init(&dp->dl_recall_lru);
286         spin_unlock(&recall_lock);
287         nfs4_close_delegation(dp);
288         nfs4_put_delegation(dp);
289 }
290
291 /* 
292  * SETCLIENTID state 
293  */
294
295 /* client_lock protects the client lru list and session hash table */
296 static DEFINE_SPINLOCK(client_lock);
297
298 /* Hash tables for nfs4_clientid state */
299 #define CLIENT_HASH_BITS                 4
300 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
301 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
302
303 #define clientid_hashval(id) \
304         ((id) & CLIENT_HASH_MASK)
305 #define clientstr_hashval(name) \
306         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
307 /*
308  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
309  * used in reboot/reset lease grace period processing
310  *
311  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
312  * setclientid_confirmed info. 
313  *
314  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
315  * setclientid info.
316  *
317  * client_lru holds client queue ordered by nfs4_client.cl_time
318  * for lease renewal.
319  *
320  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
321  * for last close replay.
322  */
323 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
324 static int reclaim_str_hashtbl_size = 0;
325 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
326 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
327 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
328 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
329 static struct list_head client_lru;
330 static struct list_head close_lru;
331
332 static void unhash_generic_stateid(struct nfs4_stateid *stp)
333 {
334         list_del(&stp->st_hash);
335         list_del(&stp->st_perfile);
336         list_del(&stp->st_perstateowner);
337 }
338
339 static void free_generic_stateid(struct nfs4_stateid *stp)
340 {
341         put_nfs4_file(stp->st_file);
342         kmem_cache_free(stateid_slab, stp);
343 }
344
345 static void release_lock_stateid(struct nfs4_stateid *stp)
346 {
347         struct file *file;
348
349         unhash_generic_stateid(stp);
350         file = find_any_file(stp->st_file);
351         if (file)
352                 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
353         free_generic_stateid(stp);
354 }
355
356 static void unhash_lockowner(struct nfs4_stateowner *sop)
357 {
358         struct nfs4_stateid *stp;
359
360         list_del(&sop->so_idhash);
361         list_del(&sop->so_strhash);
362         list_del(&sop->so_perstateid);
363         while (!list_empty(&sop->so_stateids)) {
364                 stp = list_first_entry(&sop->so_stateids,
365                                 struct nfs4_stateid, st_perstateowner);
366                 release_lock_stateid(stp);
367         }
368 }
369
370 static void release_lockowner(struct nfs4_stateowner *sop)
371 {
372         unhash_lockowner(sop);
373         nfs4_put_stateowner(sop);
374 }
375
376 static void
377 release_stateid_lockowners(struct nfs4_stateid *open_stp)
378 {
379         struct nfs4_stateowner *lock_sop;
380
381         while (!list_empty(&open_stp->st_lockowners)) {
382                 lock_sop = list_entry(open_stp->st_lockowners.next,
383                                 struct nfs4_stateowner, so_perstateid);
384                 /* list_del(&open_stp->st_lockowners);  */
385                 BUG_ON(lock_sop->so_is_open_owner);
386                 release_lockowner(lock_sop);
387         }
388 }
389
390 /*
391  * We store the NONE, READ, WRITE, and BOTH bits separately in the
392  * st_{access,deny}_bmap field of the stateid, in order to track not
393  * only what share bits are currently in force, but also what
394  * combinations of share bits previous opens have used.  This allows us
395  * to enforce the recommendation of rfc 3530 14.2.19 that the server
396  * return an error if the client attempt to downgrade to a combination
397  * of share bits not explicable by closing some of its previous opens.
398  *
399  * XXX: This enforcement is actually incomplete, since we don't keep
400  * track of access/deny bit combinations; so, e.g., we allow:
401  *
402  *      OPEN allow read, deny write
403  *      OPEN allow both, deny none
404  *      DOWNGRADE allow read, deny none
405  *
406  * which we should reject.
407  */
408 static void
409 set_access(unsigned int *access, unsigned long bmap) {
410         int i;
411
412         *access = 0;
413         for (i = 1; i < 4; i++) {
414                 if (test_bit(i, &bmap))
415                         *access |= i;
416         }
417 }
418
419 static void
420 set_deny(unsigned int *deny, unsigned long bmap) {
421         int i;
422
423         *deny = 0;
424         for (i = 0; i < 4; i++) {
425                 if (test_bit(i, &bmap))
426                         *deny |= i ;
427         }
428 }
429
430 static int
431 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
432         unsigned int access, deny;
433
434         set_access(&access, stp->st_access_bmap);
435         set_deny(&deny, stp->st_deny_bmap);
436         if ((access & open->op_share_deny) || (deny & open->op_share_access))
437                 return 0;
438         return 1;
439 }
440
441 static int nfs4_access_to_omode(u32 access)
442 {
443         switch (access) {
444         case NFS4_SHARE_ACCESS_READ:
445                 return O_RDONLY;
446         case NFS4_SHARE_ACCESS_WRITE:
447                 return O_WRONLY;
448         case NFS4_SHARE_ACCESS_BOTH:
449                 return O_RDWR;
450         }
451         BUG();
452 }
453
454 static int nfs4_access_bmap_to_omode(struct nfs4_stateid *stp)
455 {
456         unsigned int access;
457
458         set_access(&access, stp->st_access_bmap);
459         return nfs4_access_to_omode(access);
460 }
461
462 static void release_open_stateid(struct nfs4_stateid *stp)
463 {
464         int oflag = nfs4_access_bmap_to_omode(stp);
465
466         unhash_generic_stateid(stp);
467         release_stateid_lockowners(stp);
468         nfs4_file_put_access(stp->st_file, oflag);
469         free_generic_stateid(stp);
470 }
471
472 static void unhash_openowner(struct nfs4_stateowner *sop)
473 {
474         struct nfs4_stateid *stp;
475
476         list_del(&sop->so_idhash);
477         list_del(&sop->so_strhash);
478         list_del(&sop->so_perclient);
479         list_del(&sop->so_perstateid); /* XXX: necessary? */
480         while (!list_empty(&sop->so_stateids)) {
481                 stp = list_first_entry(&sop->so_stateids,
482                                 struct nfs4_stateid, st_perstateowner);
483                 release_open_stateid(stp);
484         }
485 }
486
487 static void release_openowner(struct nfs4_stateowner *sop)
488 {
489         unhash_openowner(sop);
490         list_del(&sop->so_close_lru);
491         nfs4_put_stateowner(sop);
492 }
493
494 #define SESSION_HASH_SIZE       512
495 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
496
497 static inline int
498 hash_sessionid(struct nfs4_sessionid *sessionid)
499 {
500         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
501
502         return sid->sequence % SESSION_HASH_SIZE;
503 }
504
505 static inline void
506 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
507 {
508         u32 *ptr = (u32 *)(&sessionid->data[0]);
509         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
510 }
511
512 static void
513 gen_sessionid(struct nfsd4_session *ses)
514 {
515         struct nfs4_client *clp = ses->se_client;
516         struct nfsd4_sessionid *sid;
517
518         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
519         sid->clientid = clp->cl_clientid;
520         sid->sequence = current_sessionid++;
521         sid->reserved = 0;
522 }
523
524 /*
525  * The protocol defines ca_maxresponssize_cached to include the size of
526  * the rpc header, but all we need to cache is the data starting after
527  * the end of the initial SEQUENCE operation--the rest we regenerate
528  * each time.  Therefore we can advertise a ca_maxresponssize_cached
529  * value that is the number of bytes in our cache plus a few additional
530  * bytes.  In order to stay on the safe side, and not promise more than
531  * we can cache, those additional bytes must be the minimum possible: 24
532  * bytes of rpc header (xid through accept state, with AUTH_NULL
533  * verifier), 12 for the compound header (with zero-length tag), and 44
534  * for the SEQUENCE op response:
535  */
536 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
537
538 /*
539  * Give the client the number of ca_maxresponsesize_cached slots it
540  * requests, of size bounded by NFSD_SLOT_CACHE_SIZE,
541  * NFSD_MAX_MEM_PER_SESSION, and nfsd_drc_max_mem. Do not allow more
542  * than NFSD_MAX_SLOTS_PER_SESSION.
543  *
544  * If we run out of reserved DRC memory we should (up to a point)
545  * re-negotiate active sessions and reduce their slot usage to make
546  * rooom for new connections. For now we just fail the create session.
547  */
548 static int set_forechannel_drc_size(struct nfsd4_channel_attrs *fchan)
549 {
550         int mem, size = fchan->maxresp_cached;
551
552         if (fchan->maxreqs < 1)
553                 return nfserr_inval;
554
555         if (size < NFSD_MIN_HDR_SEQ_SZ)
556                 size = NFSD_MIN_HDR_SEQ_SZ;
557         size -= NFSD_MIN_HDR_SEQ_SZ;
558         if (size > NFSD_SLOT_CACHE_SIZE)
559                 size = NFSD_SLOT_CACHE_SIZE;
560
561         /* bound the maxreqs by NFSD_MAX_MEM_PER_SESSION */
562         mem = fchan->maxreqs * size;
563         if (mem > NFSD_MAX_MEM_PER_SESSION) {
564                 fchan->maxreqs = NFSD_MAX_MEM_PER_SESSION / size;
565                 if (fchan->maxreqs > NFSD_MAX_SLOTS_PER_SESSION)
566                         fchan->maxreqs = NFSD_MAX_SLOTS_PER_SESSION;
567                 mem = fchan->maxreqs * size;
568         }
569
570         spin_lock(&nfsd_drc_lock);
571         /* bound the total session drc memory ussage */
572         if (mem + nfsd_drc_mem_used > nfsd_drc_max_mem) {
573                 fchan->maxreqs = (nfsd_drc_max_mem - nfsd_drc_mem_used) / size;
574                 mem = fchan->maxreqs * size;
575         }
576         nfsd_drc_mem_used += mem;
577         spin_unlock(&nfsd_drc_lock);
578
579         if (fchan->maxreqs == 0)
580                 return nfserr_jukebox;
581
582         fchan->maxresp_cached = size + NFSD_MIN_HDR_SEQ_SZ;
583         return 0;
584 }
585
586 /*
587  * fchan holds the client values on input, and the server values on output
588  * sv_max_mesg is the maximum payload plus one page for overhead.
589  */
590 static int init_forechannel_attrs(struct svc_rqst *rqstp,
591                                   struct nfsd4_channel_attrs *session_fchan,
592                                   struct nfsd4_channel_attrs *fchan)
593 {
594         int status = 0;
595         __u32   maxcount = nfsd_serv->sv_max_mesg;
596
597         /* headerpadsz set to zero in encode routine */
598
599         /* Use the client's max request and max response size if possible */
600         if (fchan->maxreq_sz > maxcount)
601                 fchan->maxreq_sz = maxcount;
602         session_fchan->maxreq_sz = fchan->maxreq_sz;
603
604         if (fchan->maxresp_sz > maxcount)
605                 fchan->maxresp_sz = maxcount;
606         session_fchan->maxresp_sz = fchan->maxresp_sz;
607
608         /* Use the client's maxops if possible */
609         if (fchan->maxops > NFSD_MAX_OPS_PER_COMPOUND)
610                 fchan->maxops = NFSD_MAX_OPS_PER_COMPOUND;
611         session_fchan->maxops = fchan->maxops;
612
613         /* FIXME: Error means no more DRC pages so the server should
614          * recover pages from existing sessions. For now fail session
615          * creation.
616          */
617         status = set_forechannel_drc_size(fchan);
618
619         session_fchan->maxresp_cached = fchan->maxresp_cached;
620         session_fchan->maxreqs = fchan->maxreqs;
621
622         dprintk("%s status %d\n", __func__, status);
623         return status;
624 }
625
626 static void
627 free_session_slots(struct nfsd4_session *ses)
628 {
629         int i;
630
631         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
632                 kfree(ses->se_slots[i]);
633 }
634
635 /*
636  * We don't actually need to cache the rpc and session headers, so we
637  * can allocate a little less for each slot:
638  */
639 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
640 {
641         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
642 }
643
644 static int
645 alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp,
646                    struct nfsd4_create_session *cses)
647 {
648         struct nfsd4_session *new, tmp;
649         struct nfsd4_slot *sp;
650         int idx, slotsize, cachesize, i;
651         int status;
652
653         memset(&tmp, 0, sizeof(tmp));
654
655         /* FIXME: For now, we just accept the client back channel attributes. */
656         tmp.se_bchannel = cses->back_channel;
657         status = init_forechannel_attrs(rqstp, &tmp.se_fchannel,
658                                         &cses->fore_channel);
659         if (status)
660                 goto out;
661
662         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot)
663                      + sizeof(struct nfsd4_session) > PAGE_SIZE);
664
665         status = nfserr_jukebox;
666         /* allocate struct nfsd4_session and slot table pointers in one piece */
667         slotsize = tmp.se_fchannel.maxreqs * sizeof(struct nfsd4_slot *);
668         new = kzalloc(sizeof(*new) + slotsize, GFP_KERNEL);
669         if (!new)
670                 goto out;
671
672         memcpy(new, &tmp, sizeof(*new));
673
674         /* allocate each struct nfsd4_slot and data cache in one piece */
675         cachesize = slot_bytes(&new->se_fchannel);
676         for (i = 0; i < new->se_fchannel.maxreqs; i++) {
677                 sp = kzalloc(sizeof(*sp) + cachesize, GFP_KERNEL);
678                 if (!sp)
679                         goto out_free;
680                 new->se_slots[i] = sp;
681         }
682
683         new->se_client = clp;
684         gen_sessionid(new);
685         idx = hash_sessionid(&new->se_sessionid);
686         memcpy(clp->cl_sessionid.data, new->se_sessionid.data,
687                NFS4_MAX_SESSIONID_LEN);
688
689         new->se_flags = cses->flags;
690         kref_init(&new->se_ref);
691         spin_lock(&client_lock);
692         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
693         list_add(&new->se_perclnt, &clp->cl_sessions);
694         spin_unlock(&client_lock);
695
696         status = nfs_ok;
697 out:
698         return status;
699 out_free:
700         free_session_slots(new);
701         kfree(new);
702         goto out;
703 }
704
705 /* caller must hold client_lock */
706 static struct nfsd4_session *
707 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
708 {
709         struct nfsd4_session *elem;
710         int idx;
711
712         dump_sessionid(__func__, sessionid);
713         idx = hash_sessionid(sessionid);
714         /* Search in the appropriate list */
715         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
716                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
717                             NFS4_MAX_SESSIONID_LEN)) {
718                         return elem;
719                 }
720         }
721
722         dprintk("%s: session not found\n", __func__);
723         return NULL;
724 }
725
726 /* caller must hold client_lock */
727 static void
728 unhash_session(struct nfsd4_session *ses)
729 {
730         list_del(&ses->se_hash);
731         list_del(&ses->se_perclnt);
732 }
733
734 void
735 free_session(struct kref *kref)
736 {
737         struct nfsd4_session *ses;
738         int mem;
739
740         ses = container_of(kref, struct nfsd4_session, se_ref);
741         spin_lock(&nfsd_drc_lock);
742         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
743         nfsd_drc_mem_used -= mem;
744         spin_unlock(&nfsd_drc_lock);
745         free_session_slots(ses);
746         kfree(ses);
747 }
748
749 /* must be called under the client_lock */
750 static inline void
751 renew_client_locked(struct nfs4_client *clp)
752 {
753         if (is_client_expired(clp)) {
754                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
755                         __func__,
756                         clp->cl_clientid.cl_boot,
757                         clp->cl_clientid.cl_id);
758                 return;
759         }
760
761         /*
762         * Move client to the end to the LRU list.
763         */
764         dprintk("renewing client (clientid %08x/%08x)\n", 
765                         clp->cl_clientid.cl_boot, 
766                         clp->cl_clientid.cl_id);
767         list_move_tail(&clp->cl_lru, &client_lru);
768         clp->cl_time = get_seconds();
769 }
770
771 static inline void
772 renew_client(struct nfs4_client *clp)
773 {
774         spin_lock(&client_lock);
775         renew_client_locked(clp);
776         spin_unlock(&client_lock);
777 }
778
779 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
780 static int
781 STALE_CLIENTID(clientid_t *clid)
782 {
783         if (clid->cl_boot == boot_time)
784                 return 0;
785         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
786                 clid->cl_boot, clid->cl_id, boot_time);
787         return 1;
788 }
789
790 /* 
791  * XXX Should we use a slab cache ?
792  * This type of memory management is somewhat inefficient, but we use it
793  * anyway since SETCLIENTID is not a common operation.
794  */
795 static struct nfs4_client *alloc_client(struct xdr_netobj name)
796 {
797         struct nfs4_client *clp;
798
799         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
800         if (clp == NULL)
801                 return NULL;
802         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
803         if (clp->cl_name.data == NULL) {
804                 kfree(clp);
805                 return NULL;
806         }
807         memcpy(clp->cl_name.data, name.data, name.len);
808         clp->cl_name.len = name.len;
809         return clp;
810 }
811
812 static inline void
813 free_client(struct nfs4_client *clp)
814 {
815         if (clp->cl_cred.cr_group_info)
816                 put_group_info(clp->cl_cred.cr_group_info);
817         kfree(clp->cl_principal);
818         kfree(clp->cl_name.data);
819         kfree(clp);
820 }
821
822 void
823 release_session_client(struct nfsd4_session *session)
824 {
825         struct nfs4_client *clp = session->se_client;
826
827         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
828                 return;
829         if (is_client_expired(clp)) {
830                 free_client(clp);
831                 session->se_client = NULL;
832         } else
833                 renew_client_locked(clp);
834         spin_unlock(&client_lock);
835 }
836
837 /* must be called under the client_lock */
838 static inline void
839 unhash_client_locked(struct nfs4_client *clp)
840 {
841         mark_client_expired(clp);
842         list_del(&clp->cl_lru);
843         while (!list_empty(&clp->cl_sessions)) {
844                 struct nfsd4_session  *ses;
845                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
846                                  se_perclnt);
847                 unhash_session(ses);
848                 nfsd4_put_session(ses);
849         }
850 }
851
852 static void
853 expire_client(struct nfs4_client *clp)
854 {
855         struct nfs4_stateowner *sop;
856         struct nfs4_delegation *dp;
857         struct list_head reaplist;
858
859         INIT_LIST_HEAD(&reaplist);
860         spin_lock(&recall_lock);
861         while (!list_empty(&clp->cl_delegations)) {
862                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
863                 dprintk("NFSD: expire client. dp %p, fp %p\n", dp,
864                                 dp->dl_flock);
865                 list_del_init(&dp->dl_perclnt);
866                 list_move(&dp->dl_recall_lru, &reaplist);
867         }
868         spin_unlock(&recall_lock);
869         while (!list_empty(&reaplist)) {
870                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
871                 list_del_init(&dp->dl_recall_lru);
872                 unhash_delegation(dp);
873         }
874         while (!list_empty(&clp->cl_openowners)) {
875                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
876                 release_openowner(sop);
877         }
878         nfsd4_set_callback_client(clp, NULL);
879         if (clp->cl_cb_conn.cb_xprt)
880                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
881         list_del(&clp->cl_idhash);
882         list_del(&clp->cl_strhash);
883         spin_lock(&client_lock);
884         unhash_client_locked(clp);
885         if (atomic_read(&clp->cl_refcount) == 0)
886                 free_client(clp);
887         spin_unlock(&client_lock);
888 }
889
890 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
891 {
892         memcpy(target->cl_verifier.data, source->data,
893                         sizeof(target->cl_verifier.data));
894 }
895
896 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
897 {
898         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
899         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
900 }
901
902 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
903 {
904         target->cr_uid = source->cr_uid;
905         target->cr_gid = source->cr_gid;
906         target->cr_group_info = source->cr_group_info;
907         get_group_info(target->cr_group_info);
908 }
909
910 static int same_name(const char *n1, const char *n2)
911 {
912         return 0 == memcmp(n1, n2, HEXDIR_LEN);
913 }
914
915 static int
916 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
917 {
918         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
919 }
920
921 static int
922 same_clid(clientid_t *cl1, clientid_t *cl2)
923 {
924         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
925 }
926
927 /* XXX what about NGROUP */
928 static int
929 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
930 {
931         return cr1->cr_uid == cr2->cr_uid;
932 }
933
934 static void gen_clid(struct nfs4_client *clp)
935 {
936         static u32 current_clientid = 1;
937
938         clp->cl_clientid.cl_boot = boot_time;
939         clp->cl_clientid.cl_id = current_clientid++; 
940 }
941
942 static void gen_confirm(struct nfs4_client *clp)
943 {
944         static u32 i;
945         u32 *p;
946
947         p = (u32 *)clp->cl_confirm.data;
948         *p++ = get_seconds();
949         *p++ = i++;
950 }
951
952 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
953                 struct svc_rqst *rqstp, nfs4_verifier *verf)
954 {
955         struct nfs4_client *clp;
956         struct sockaddr *sa = svc_addr(rqstp);
957         char *princ;
958
959         clp = alloc_client(name);
960         if (clp == NULL)
961                 return NULL;
962
963         princ = svc_gss_principal(rqstp);
964         if (princ) {
965                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
966                 if (clp->cl_principal == NULL) {
967                         free_client(clp);
968                         return NULL;
969                 }
970         }
971
972         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
973         atomic_set(&clp->cl_refcount, 0);
974         atomic_set(&clp->cl_cb_set, 0);
975         INIT_LIST_HEAD(&clp->cl_idhash);
976         INIT_LIST_HEAD(&clp->cl_strhash);
977         INIT_LIST_HEAD(&clp->cl_openowners);
978         INIT_LIST_HEAD(&clp->cl_delegations);
979         INIT_LIST_HEAD(&clp->cl_sessions);
980         INIT_LIST_HEAD(&clp->cl_lru);
981         clp->cl_time = get_seconds();
982         clear_bit(0, &clp->cl_cb_slot_busy);
983         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
984         copy_verf(clp, verf);
985         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
986         clp->cl_flavor = rqstp->rq_flavor;
987         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
988         gen_confirm(clp);
989
990         return clp;
991 }
992
993 static int check_name(struct xdr_netobj name)
994 {
995         if (name.len == 0) 
996                 return 0;
997         if (name.len > NFS4_OPAQUE_LIMIT) {
998                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
999                 return 0;
1000         }
1001         return 1;
1002 }
1003
1004 static void
1005 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1006 {
1007         unsigned int idhashval;
1008
1009         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1010         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1011         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1012         renew_client(clp);
1013 }
1014
1015 static void
1016 move_to_confirmed(struct nfs4_client *clp)
1017 {
1018         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1019         unsigned int strhashval;
1020
1021         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1022         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1023         strhashval = clientstr_hashval(clp->cl_recdir);
1024         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1025         renew_client(clp);
1026 }
1027
1028 static struct nfs4_client *
1029 find_confirmed_client(clientid_t *clid)
1030 {
1031         struct nfs4_client *clp;
1032         unsigned int idhashval = clientid_hashval(clid->cl_id);
1033
1034         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1035                 if (same_clid(&clp->cl_clientid, clid))
1036                         return clp;
1037         }
1038         return NULL;
1039 }
1040
1041 static struct nfs4_client *
1042 find_unconfirmed_client(clientid_t *clid)
1043 {
1044         struct nfs4_client *clp;
1045         unsigned int idhashval = clientid_hashval(clid->cl_id);
1046
1047         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1048                 if (same_clid(&clp->cl_clientid, clid))
1049                         return clp;
1050         }
1051         return NULL;
1052 }
1053
1054 /*
1055  * Return 1 iff clp's clientid establishment method matches the use_exchange_id
1056  * parameter. Matching is based on the fact the at least one of the
1057  * EXCHGID4_FLAG_USE_{NON_PNFS,PNFS_MDS,PNFS_DS} flags must be set for v4.1
1058  *
1059  * FIXME: we need to unify the clientid namespaces for nfsv4.x
1060  * and correctly deal with client upgrade/downgrade in EXCHANGE_ID
1061  * and SET_CLIENTID{,_CONFIRM}
1062  */
1063 static inline int
1064 match_clientid_establishment(struct nfs4_client *clp, bool use_exchange_id)
1065 {
1066         bool has_exchange_flags = (clp->cl_exchange_flags != 0);
1067         return use_exchange_id == has_exchange_flags;
1068 }
1069
1070 static struct nfs4_client *
1071 find_confirmed_client_by_str(const char *dname, unsigned int hashval,
1072                              bool use_exchange_id)
1073 {
1074         struct nfs4_client *clp;
1075
1076         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1077                 if (same_name(clp->cl_recdir, dname) &&
1078                     match_clientid_establishment(clp, use_exchange_id))
1079                         return clp;
1080         }
1081         return NULL;
1082 }
1083
1084 static struct nfs4_client *
1085 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval,
1086                                bool use_exchange_id)
1087 {
1088         struct nfs4_client *clp;
1089
1090         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1091                 if (same_name(clp->cl_recdir, dname) &&
1092                     match_clientid_establishment(clp, use_exchange_id))
1093                         return clp;
1094         }
1095         return NULL;
1096 }
1097
1098 static void
1099 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, u32 scopeid)
1100 {
1101         struct nfs4_cb_conn *cb = &clp->cl_cb_conn;
1102         unsigned short expected_family;
1103
1104         /* Currently, we only support tcp and tcp6 for the callback channel */
1105         if (se->se_callback_netid_len == 3 &&
1106             !memcmp(se->se_callback_netid_val, "tcp", 3))
1107                 expected_family = AF_INET;
1108         else if (se->se_callback_netid_len == 4 &&
1109                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1110                 expected_family = AF_INET6;
1111         else
1112                 goto out_err;
1113
1114         cb->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1115                                             se->se_callback_addr_len,
1116                                             (struct sockaddr *) &cb->cb_addr,
1117                                             sizeof(cb->cb_addr));
1118
1119         if (!cb->cb_addrlen || cb->cb_addr.ss_family != expected_family)
1120                 goto out_err;
1121
1122         if (cb->cb_addr.ss_family == AF_INET6)
1123                 ((struct sockaddr_in6 *) &cb->cb_addr)->sin6_scope_id = scopeid;
1124
1125         cb->cb_minorversion = 0;
1126         cb->cb_prog = se->se_callback_prog;
1127         cb->cb_ident = se->se_callback_ident;
1128         return;
1129 out_err:
1130         cb->cb_addr.ss_family = AF_UNSPEC;
1131         cb->cb_addrlen = 0;
1132         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1133                 "will not receive delegations\n",
1134                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1135
1136         return;
1137 }
1138
1139 /*
1140  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1141  */
1142 void
1143 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1144 {
1145         struct nfsd4_slot *slot = resp->cstate.slot;
1146         unsigned int base;
1147
1148         dprintk("--> %s slot %p\n", __func__, slot);
1149
1150         slot->sl_opcnt = resp->opcnt;
1151         slot->sl_status = resp->cstate.status;
1152
1153         if (nfsd4_not_cached(resp)) {
1154                 slot->sl_datalen = 0;
1155                 return;
1156         }
1157         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1158         base = (char *)resp->cstate.datap -
1159                                         (char *)resp->xbuf->head[0].iov_base;
1160         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1161                                     slot->sl_datalen))
1162                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1163         return;
1164 }
1165
1166 /*
1167  * Encode the replay sequence operation from the slot values.
1168  * If cachethis is FALSE encode the uncached rep error on the next
1169  * operation which sets resp->p and increments resp->opcnt for
1170  * nfs4svc_encode_compoundres.
1171  *
1172  */
1173 static __be32
1174 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1175                           struct nfsd4_compoundres *resp)
1176 {
1177         struct nfsd4_op *op;
1178         struct nfsd4_slot *slot = resp->cstate.slot;
1179
1180         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1181                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1182
1183         /* Encode the replayed sequence operation */
1184         op = &args->ops[resp->opcnt - 1];
1185         nfsd4_encode_operation(resp, op);
1186
1187         /* Return nfserr_retry_uncached_rep in next operation. */
1188         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1189                 op = &args->ops[resp->opcnt++];
1190                 op->status = nfserr_retry_uncached_rep;
1191                 nfsd4_encode_operation(resp, op);
1192         }
1193         return op->status;
1194 }
1195
1196 /*
1197  * The sequence operation is not cached because we can use the slot and
1198  * session values.
1199  */
1200 __be32
1201 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1202                          struct nfsd4_sequence *seq)
1203 {
1204         struct nfsd4_slot *slot = resp->cstate.slot;
1205         __be32 status;
1206
1207         dprintk("--> %s slot %p\n", __func__, slot);
1208
1209         /* Either returns 0 or nfserr_retry_uncached */
1210         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1211         if (status == nfserr_retry_uncached_rep)
1212                 return status;
1213
1214         /* The sequence operation has been encoded, cstate->datap set. */
1215         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1216
1217         resp->opcnt = slot->sl_opcnt;
1218         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1219         status = slot->sl_status;
1220
1221         return status;
1222 }
1223
1224 /*
1225  * Set the exchange_id flags returned by the server.
1226  */
1227 static void
1228 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1229 {
1230         /* pNFS is not supported */
1231         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1232
1233         /* Referrals are supported, Migration is not. */
1234         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1235
1236         /* set the wire flags to return to client. */
1237         clid->flags = new->cl_exchange_flags;
1238 }
1239
1240 __be32
1241 nfsd4_exchange_id(struct svc_rqst *rqstp,
1242                   struct nfsd4_compound_state *cstate,
1243                   struct nfsd4_exchange_id *exid)
1244 {
1245         struct nfs4_client *unconf, *conf, *new;
1246         int status;
1247         unsigned int            strhashval;
1248         char                    dname[HEXDIR_LEN];
1249         char                    addr_str[INET6_ADDRSTRLEN];
1250         nfs4_verifier           verf = exid->verifier;
1251         struct sockaddr         *sa = svc_addr(rqstp);
1252
1253         rpc_ntop(sa, addr_str, sizeof(addr_str));
1254         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1255                 "ip_addr=%s flags %x, spa_how %d\n",
1256                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1257                 addr_str, exid->flags, exid->spa_how);
1258
1259         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1260                 return nfserr_inval;
1261
1262         /* Currently only support SP4_NONE */
1263         switch (exid->spa_how) {
1264         case SP4_NONE:
1265                 break;
1266         case SP4_SSV:
1267                 return nfserr_encr_alg_unsupp;
1268         default:
1269                 BUG();                          /* checked by xdr code */
1270         case SP4_MACH_CRED:
1271                 return nfserr_serverfault;      /* no excuse :-/ */
1272         }
1273
1274         status = nfs4_make_rec_clidname(dname, &exid->clname);
1275
1276         if (status)
1277                 goto error;
1278
1279         strhashval = clientstr_hashval(dname);
1280
1281         nfs4_lock_state();
1282         status = nfs_ok;
1283
1284         conf = find_confirmed_client_by_str(dname, strhashval, true);
1285         if (conf) {
1286                 if (!same_verf(&verf, &conf->cl_verifier)) {
1287                         /* 18.35.4 case 8 */
1288                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1289                                 status = nfserr_not_same;
1290                                 goto out;
1291                         }
1292                         /* Client reboot: destroy old state */
1293                         expire_client(conf);
1294                         goto out_new;
1295                 }
1296                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1297                         /* 18.35.4 case 9 */
1298                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1299                                 status = nfserr_perm;
1300                                 goto out;
1301                         }
1302                         expire_client(conf);
1303                         goto out_new;
1304                 }
1305                 /*
1306                  * Set bit when the owner id and verifier map to an already
1307                  * confirmed client id (18.35.3).
1308                  */
1309                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1310
1311                 /*
1312                  * Falling into 18.35.4 case 2, possible router replay.
1313                  * Leave confirmed record intact and return same result.
1314                  */
1315                 copy_verf(conf, &verf);
1316                 new = conf;
1317                 goto out_copy;
1318         }
1319
1320         /* 18.35.4 case 7 */
1321         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1322                 status = nfserr_noent;
1323                 goto out;
1324         }
1325
1326         unconf  = find_unconfirmed_client_by_str(dname, strhashval, true);
1327         if (unconf) {
1328                 /*
1329                  * Possible retry or client restart.  Per 18.35.4 case 4,
1330                  * a new unconfirmed record should be generated regardless
1331                  * of whether any properties have changed.
1332                  */
1333                 expire_client(unconf);
1334         }
1335
1336 out_new:
1337         /* Normal case */
1338         new = create_client(exid->clname, dname, rqstp, &verf);
1339         if (new == NULL) {
1340                 status = nfserr_jukebox;
1341                 goto out;
1342         }
1343
1344         gen_clid(new);
1345         add_to_unconfirmed(new, strhashval);
1346 out_copy:
1347         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1348         exid->clientid.cl_id = new->cl_clientid.cl_id;
1349
1350         exid->seqid = 1;
1351         nfsd4_set_ex_flags(new, exid);
1352
1353         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1354                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1355         status = nfs_ok;
1356
1357 out:
1358         nfs4_unlock_state();
1359 error:
1360         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1361         return status;
1362 }
1363
1364 static int
1365 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1366 {
1367         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1368                 slot_seqid);
1369
1370         /* The slot is in use, and no response has been sent. */
1371         if (slot_inuse) {
1372                 if (seqid == slot_seqid)
1373                         return nfserr_jukebox;
1374                 else
1375                         return nfserr_seq_misordered;
1376         }
1377         /* Normal */
1378         if (likely(seqid == slot_seqid + 1))
1379                 return nfs_ok;
1380         /* Replay */
1381         if (seqid == slot_seqid)
1382                 return nfserr_replay_cache;
1383         /* Wraparound */
1384         if (seqid == 1 && (slot_seqid + 1) == 0)
1385                 return nfs_ok;
1386         /* Misordered replay or misordered new request */
1387         return nfserr_seq_misordered;
1388 }
1389
1390 /*
1391  * Cache the create session result into the create session single DRC
1392  * slot cache by saving the xdr structure. sl_seqid has been set.
1393  * Do this for solo or embedded create session operations.
1394  */
1395 static void
1396 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1397                            struct nfsd4_clid_slot *slot, int nfserr)
1398 {
1399         slot->sl_status = nfserr;
1400         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1401 }
1402
1403 static __be32
1404 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1405                             struct nfsd4_clid_slot *slot)
1406 {
1407         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1408         return slot->sl_status;
1409 }
1410
1411 __be32
1412 nfsd4_create_session(struct svc_rqst *rqstp,
1413                      struct nfsd4_compound_state *cstate,
1414                      struct nfsd4_create_session *cr_ses)
1415 {
1416         struct sockaddr *sa = svc_addr(rqstp);
1417         struct nfs4_client *conf, *unconf;
1418         struct nfsd4_clid_slot *cs_slot = NULL;
1419         int status = 0;
1420
1421         nfs4_lock_state();
1422         unconf = find_unconfirmed_client(&cr_ses->clientid);
1423         conf = find_confirmed_client(&cr_ses->clientid);
1424
1425         if (conf) {
1426                 cs_slot = &conf->cl_cs_slot;
1427                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1428                 if (status == nfserr_replay_cache) {
1429                         dprintk("Got a create_session replay! seqid= %d\n",
1430                                 cs_slot->sl_seqid);
1431                         /* Return the cached reply status */
1432                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1433                         goto out;
1434                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1435                         status = nfserr_seq_misordered;
1436                         dprintk("Sequence misordered!\n");
1437                         dprintk("Expected seqid= %d but got seqid= %d\n",
1438                                 cs_slot->sl_seqid, cr_ses->seqid);
1439                         goto out;
1440                 }
1441                 cs_slot->sl_seqid++;
1442         } else if (unconf) {
1443                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1444                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1445                         status = nfserr_clid_inuse;
1446                         goto out;
1447                 }
1448
1449                 cs_slot = &unconf->cl_cs_slot;
1450                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1451                 if (status) {
1452                         /* an unconfirmed replay returns misordered */
1453                         status = nfserr_seq_misordered;
1454                         goto out_cache;
1455                 }
1456
1457                 cs_slot->sl_seqid++; /* from 0 to 1 */
1458                 move_to_confirmed(unconf);
1459
1460                 if (cr_ses->flags & SESSION4_BACK_CHAN) {
1461                         unconf->cl_cb_conn.cb_xprt = rqstp->rq_xprt;
1462                         svc_xprt_get(rqstp->rq_xprt);
1463                         rpc_copy_addr(
1464                                 (struct sockaddr *)&unconf->cl_cb_conn.cb_addr,
1465                                 sa);
1466                         unconf->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1467                         unconf->cl_cb_conn.cb_minorversion =
1468                                 cstate->minorversion;
1469                         unconf->cl_cb_conn.cb_prog = cr_ses->callback_prog;
1470                         unconf->cl_cb_seq_nr = 1;
1471                         nfsd4_probe_callback(unconf, &unconf->cl_cb_conn);
1472                 }
1473                 conf = unconf;
1474         } else {
1475                 status = nfserr_stale_clientid;
1476                 goto out;
1477         }
1478
1479         /*
1480          * We do not support RDMA or persistent sessions
1481          */
1482         cr_ses->flags &= ~SESSION4_PERSIST;
1483         cr_ses->flags &= ~SESSION4_RDMA;
1484
1485         status = alloc_init_session(rqstp, conf, cr_ses);
1486         if (status)
1487                 goto out;
1488
1489         memcpy(cr_ses->sessionid.data, conf->cl_sessionid.data,
1490                NFS4_MAX_SESSIONID_LEN);
1491         cr_ses->seqid = cs_slot->sl_seqid;
1492
1493 out_cache:
1494         /* cache solo and embedded create sessions under the state lock */
1495         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1496 out:
1497         nfs4_unlock_state();
1498         dprintk("%s returns %d\n", __func__, ntohl(status));
1499         return status;
1500 }
1501
1502 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1503 {
1504         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1505         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1506
1507         return argp->opcnt == resp->opcnt;
1508 }
1509
1510 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1511 {
1512         if (!session)
1513                 return 0;
1514         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1515 }
1516
1517 __be32
1518 nfsd4_destroy_session(struct svc_rqst *r,
1519                       struct nfsd4_compound_state *cstate,
1520                       struct nfsd4_destroy_session *sessionid)
1521 {
1522         struct nfsd4_session *ses;
1523         u32 status = nfserr_badsession;
1524
1525         /* Notes:
1526          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1527          * - Should we return nfserr_back_chan_busy if waiting for
1528          *   callbacks on to-be-destroyed session?
1529          * - Do we need to clear any callback info from previous session?
1530          */
1531
1532         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1533                 if (!nfsd4_last_compound_op(r))
1534                         return nfserr_not_only_op;
1535         }
1536         dump_sessionid(__func__, &sessionid->sessionid);
1537         spin_lock(&client_lock);
1538         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1539         if (!ses) {
1540                 spin_unlock(&client_lock);
1541                 goto out;
1542         }
1543
1544         unhash_session(ses);
1545         spin_unlock(&client_lock);
1546
1547         nfs4_lock_state();
1548         /* wait for callbacks */
1549         nfsd4_set_callback_client(ses->se_client, NULL);
1550         nfs4_unlock_state();
1551         nfsd4_put_session(ses);
1552         status = nfs_ok;
1553 out:
1554         dprintk("%s returns %d\n", __func__, ntohl(status));
1555         return status;
1556 }
1557
1558 __be32
1559 nfsd4_sequence(struct svc_rqst *rqstp,
1560                struct nfsd4_compound_state *cstate,
1561                struct nfsd4_sequence *seq)
1562 {
1563         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1564         struct nfsd4_session *session;
1565         struct nfsd4_slot *slot;
1566         int status;
1567
1568         if (resp->opcnt != 1)
1569                 return nfserr_sequence_pos;
1570
1571         spin_lock(&client_lock);
1572         status = nfserr_badsession;
1573         session = find_in_sessionid_hashtbl(&seq->sessionid);
1574         if (!session)
1575                 goto out;
1576
1577         status = nfserr_badslot;
1578         if (seq->slotid >= session->se_fchannel.maxreqs)
1579                 goto out;
1580
1581         slot = session->se_slots[seq->slotid];
1582         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1583
1584         /* We do not negotiate the number of slots yet, so set the
1585          * maxslots to the session maxreqs which is used to encode
1586          * sr_highest_slotid and the sr_target_slot id to maxslots */
1587         seq->maxslots = session->se_fchannel.maxreqs;
1588
1589         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1590         if (status == nfserr_replay_cache) {
1591                 cstate->slot = slot;
1592                 cstate->session = session;
1593                 /* Return the cached reply status and set cstate->status
1594                  * for nfsd4_proc_compound processing */
1595                 status = nfsd4_replay_cache_entry(resp, seq);
1596                 cstate->status = nfserr_replay_cache;
1597                 goto out;
1598         }
1599         if (status)
1600                 goto out;
1601
1602         /* Success! bump slot seqid */
1603         slot->sl_inuse = true;
1604         slot->sl_seqid = seq->seqid;
1605         slot->sl_cachethis = seq->cachethis;
1606
1607         cstate->slot = slot;
1608         cstate->session = session;
1609
1610 out:
1611         /* Hold a session reference until done processing the compound. */
1612         if (cstate->session) {
1613                 nfsd4_get_session(cstate->session);
1614                 atomic_inc(&session->se_client->cl_refcount);
1615         }
1616         spin_unlock(&client_lock);
1617         dprintk("%s: return %d\n", __func__, ntohl(status));
1618         return status;
1619 }
1620
1621 __be32
1622 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1623 {
1624         if (rc->rca_one_fs) {
1625                 if (!cstate->current_fh.fh_dentry)
1626                         return nfserr_nofilehandle;
1627                 /*
1628                  * We don't take advantage of the rca_one_fs case.
1629                  * That's OK, it's optional, we can safely ignore it.
1630                  */
1631                  return nfs_ok;
1632         }
1633         nfs4_lock_state();
1634         if (is_client_expired(cstate->session->se_client)) {
1635                 nfs4_unlock_state();
1636                 /*
1637                  * The following error isn't really legal.
1638                  * But we only get here if the client just explicitly
1639                  * destroyed the client.  Surely it no longer cares what
1640                  * error it gets back on an operation for the dead
1641                  * client.
1642                  */
1643                 return nfserr_stale_clientid;
1644         }
1645         nfsd4_create_clid_dir(cstate->session->se_client);
1646         nfs4_unlock_state();
1647         return nfs_ok;
1648 }
1649
1650 __be32
1651 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1652                   struct nfsd4_setclientid *setclid)
1653 {
1654         struct sockaddr         *sa = svc_addr(rqstp);
1655         struct xdr_netobj       clname = { 
1656                 .len = setclid->se_namelen,
1657                 .data = setclid->se_name,
1658         };
1659         nfs4_verifier           clverifier = setclid->se_verf;
1660         unsigned int            strhashval;
1661         struct nfs4_client      *conf, *unconf, *new;
1662         __be32                  status;
1663         char                    dname[HEXDIR_LEN];
1664         
1665         if (!check_name(clname))
1666                 return nfserr_inval;
1667
1668         status = nfs4_make_rec_clidname(dname, &clname);
1669         if (status)
1670                 return status;
1671
1672         /* 
1673          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1674          * We get here on a DRC miss.
1675          */
1676
1677         strhashval = clientstr_hashval(dname);
1678
1679         nfs4_lock_state();
1680         conf = find_confirmed_client_by_str(dname, strhashval, false);
1681         if (conf) {
1682                 /* RFC 3530 14.2.33 CASE 0: */
1683                 status = nfserr_clid_inuse;
1684                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1685                         char addr_str[INET6_ADDRSTRLEN];
1686                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1687                                  sizeof(addr_str));
1688                         dprintk("NFSD: setclientid: string in use by client "
1689                                 "at %s\n", addr_str);
1690                         goto out;
1691                 }
1692         }
1693         /*
1694          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1695          * has a description of SETCLIENTID request processing consisting
1696          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1697          */
1698         unconf = find_unconfirmed_client_by_str(dname, strhashval, false);
1699         status = nfserr_resource;
1700         if (!conf) {
1701                 /*
1702                  * RFC 3530 14.2.33 CASE 4:
1703                  * placed first, because it is the normal case
1704                  */
1705                 if (unconf)
1706                         expire_client(unconf);
1707                 new = create_client(clname, dname, rqstp, &clverifier);
1708                 if (new == NULL)
1709                         goto out;
1710                 gen_clid(new);
1711         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1712                 /*
1713                  * RFC 3530 14.2.33 CASE 1:
1714                  * probable callback update
1715                  */
1716                 if (unconf) {
1717                         /* Note this is removing unconfirmed {*x***},
1718                          * which is stronger than RFC recommended {vxc**}.
1719                          * This has the advantage that there is at most
1720                          * one {*x***} in either list at any time.
1721                          */
1722                         expire_client(unconf);
1723                 }
1724                 new = create_client(clname, dname, rqstp, &clverifier);
1725                 if (new == NULL)
1726                         goto out;
1727                 copy_clid(new, conf);
1728         } else if (!unconf) {
1729                 /*
1730                  * RFC 3530 14.2.33 CASE 2:
1731                  * probable client reboot; state will be removed if
1732                  * confirmed.
1733                  */
1734                 new = create_client(clname, dname, rqstp, &clverifier);
1735                 if (new == NULL)
1736                         goto out;
1737                 gen_clid(new);
1738         } else {
1739                 /*
1740                  * RFC 3530 14.2.33 CASE 3:
1741                  * probable client reboot; state will be removed if
1742                  * confirmed.
1743                  */
1744                 expire_client(unconf);
1745                 new = create_client(clname, dname, rqstp, &clverifier);
1746                 if (new == NULL)
1747                         goto out;
1748                 gen_clid(new);
1749         }
1750         gen_callback(new, setclid, rpc_get_scope_id(sa));
1751         add_to_unconfirmed(new, strhashval);
1752         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1753         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1754         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1755         status = nfs_ok;
1756 out:
1757         nfs4_unlock_state();
1758         return status;
1759 }
1760
1761
1762 /*
1763  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1764  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1765  * bullets, labeled as CASE1 - CASE4 below.
1766  */
1767 __be32
1768 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1769                          struct nfsd4_compound_state *cstate,
1770                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1771 {
1772         struct sockaddr *sa = svc_addr(rqstp);
1773         struct nfs4_client *conf, *unconf;
1774         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1775         clientid_t * clid = &setclientid_confirm->sc_clientid;
1776         __be32 status;
1777
1778         if (STALE_CLIENTID(clid))
1779                 return nfserr_stale_clientid;
1780         /* 
1781          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1782          * We get here on a DRC miss.
1783          */
1784
1785         nfs4_lock_state();
1786
1787         conf = find_confirmed_client(clid);
1788         unconf = find_unconfirmed_client(clid);
1789
1790         status = nfserr_clid_inuse;
1791         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
1792                 goto out;
1793         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
1794                 goto out;
1795
1796         /*
1797          * section 14.2.34 of RFC 3530 has a description of
1798          * SETCLIENTID_CONFIRM request processing consisting
1799          * of 4 bullet points, labeled as CASE1 - CASE4 below.
1800          */
1801         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
1802                 /*
1803                  * RFC 3530 14.2.34 CASE 1:
1804                  * callback update
1805                  */
1806                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
1807                         status = nfserr_clid_inuse;
1808                 else {
1809                         atomic_set(&conf->cl_cb_set, 0);
1810                         nfsd4_probe_callback(conf, &unconf->cl_cb_conn);
1811                         expire_client(unconf);
1812                         status = nfs_ok;
1813
1814                 }
1815         } else if (conf && !unconf) {
1816                 /*
1817                  * RFC 3530 14.2.34 CASE 2:
1818                  * probable retransmitted request; play it safe and
1819                  * do nothing.
1820                  */
1821                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
1822                         status = nfserr_clid_inuse;
1823                 else
1824                         status = nfs_ok;
1825         } else if (!conf && unconf
1826                         && same_verf(&unconf->cl_confirm, &confirm)) {
1827                 /*
1828                  * RFC 3530 14.2.34 CASE 3:
1829                  * Normal case; new or rebooted client:
1830                  */
1831                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
1832                         status = nfserr_clid_inuse;
1833                 } else {
1834                         unsigned int hash =
1835                                 clientstr_hashval(unconf->cl_recdir);
1836                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
1837                                                             hash, false);
1838                         if (conf) {
1839                                 nfsd4_remove_clid_dir(conf);
1840                                 expire_client(conf);
1841                         }
1842                         move_to_confirmed(unconf);
1843                         conf = unconf;
1844                         nfsd4_probe_callback(conf, &conf->cl_cb_conn);
1845                         status = nfs_ok;
1846                 }
1847         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
1848             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
1849                                                                 &confirm)))) {
1850                 /*
1851                  * RFC 3530 14.2.34 CASE 4:
1852                  * Client probably hasn't noticed that we rebooted yet.
1853                  */
1854                 status = nfserr_stale_clientid;
1855         } else {
1856                 /* check that we have hit one of the cases...*/
1857                 status = nfserr_clid_inuse;
1858         }
1859 out:
1860         nfs4_unlock_state();
1861         return status;
1862 }
1863
1864 /* OPEN Share state helper functions */
1865 static inline struct nfs4_file *
1866 alloc_init_file(struct inode *ino)
1867 {
1868         struct nfs4_file *fp;
1869         unsigned int hashval = file_hashval(ino);
1870
1871         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
1872         if (fp) {
1873                 atomic_set(&fp->fi_ref, 1);
1874                 INIT_LIST_HEAD(&fp->fi_hash);
1875                 INIT_LIST_HEAD(&fp->fi_stateids);
1876                 INIT_LIST_HEAD(&fp->fi_delegations);
1877                 fp->fi_inode = igrab(ino);
1878                 fp->fi_id = current_fileid++;
1879                 fp->fi_had_conflict = false;
1880                 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
1881                 memset(fp->fi_access, 0, sizeof(fp->fi_access));
1882                 spin_lock(&recall_lock);
1883                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
1884                 spin_unlock(&recall_lock);
1885                 return fp;
1886         }
1887         return NULL;
1888 }
1889
1890 static void
1891 nfsd4_free_slab(struct kmem_cache **slab)
1892 {
1893         if (*slab == NULL)
1894                 return;
1895         kmem_cache_destroy(*slab);
1896         *slab = NULL;
1897 }
1898
1899 void
1900 nfsd4_free_slabs(void)
1901 {
1902         nfsd4_free_slab(&stateowner_slab);
1903         nfsd4_free_slab(&file_slab);
1904         nfsd4_free_slab(&stateid_slab);
1905         nfsd4_free_slab(&deleg_slab);
1906 }
1907
1908 static int
1909 nfsd4_init_slabs(void)
1910 {
1911         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
1912                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
1913         if (stateowner_slab == NULL)
1914                 goto out_nomem;
1915         file_slab = kmem_cache_create("nfsd4_files",
1916                         sizeof(struct nfs4_file), 0, 0, NULL);
1917         if (file_slab == NULL)
1918                 goto out_nomem;
1919         stateid_slab = kmem_cache_create("nfsd4_stateids",
1920                         sizeof(struct nfs4_stateid), 0, 0, NULL);
1921         if (stateid_slab == NULL)
1922                 goto out_nomem;
1923         deleg_slab = kmem_cache_create("nfsd4_delegations",
1924                         sizeof(struct nfs4_delegation), 0, 0, NULL);
1925         if (deleg_slab == NULL)
1926                 goto out_nomem;
1927         return 0;
1928 out_nomem:
1929         nfsd4_free_slabs();
1930         dprintk("nfsd4: out of memory while initializing nfsv4\n");
1931         return -ENOMEM;
1932 }
1933
1934 void
1935 nfs4_free_stateowner(struct kref *kref)
1936 {
1937         struct nfs4_stateowner *sop =
1938                 container_of(kref, struct nfs4_stateowner, so_ref);
1939         kfree(sop->so_owner.data);
1940         kmem_cache_free(stateowner_slab, sop);
1941 }
1942
1943 static inline struct nfs4_stateowner *
1944 alloc_stateowner(struct xdr_netobj *owner)
1945 {
1946         struct nfs4_stateowner *sop;
1947
1948         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
1949                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
1950                         memcpy(sop->so_owner.data, owner->data, owner->len);
1951                         sop->so_owner.len = owner->len;
1952                         kref_init(&sop->so_ref);
1953                         return sop;
1954                 } 
1955                 kmem_cache_free(stateowner_slab, sop);
1956         }
1957         return NULL;
1958 }
1959
1960 static struct nfs4_stateowner *
1961 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
1962         struct nfs4_stateowner *sop;
1963         struct nfs4_replay *rp;
1964         unsigned int idhashval;
1965
1966         if (!(sop = alloc_stateowner(&open->op_owner)))
1967                 return NULL;
1968         idhashval = ownerid_hashval(current_ownerid);
1969         INIT_LIST_HEAD(&sop->so_idhash);
1970         INIT_LIST_HEAD(&sop->so_strhash);
1971         INIT_LIST_HEAD(&sop->so_perclient);
1972         INIT_LIST_HEAD(&sop->so_stateids);
1973         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
1974         INIT_LIST_HEAD(&sop->so_close_lru);
1975         sop->so_time = 0;
1976         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
1977         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
1978         list_add(&sop->so_perclient, &clp->cl_openowners);
1979         sop->so_is_open_owner = 1;
1980         sop->so_id = current_ownerid++;
1981         sop->so_client = clp;
1982         sop->so_seqid = open->op_seqid;
1983         sop->so_confirmed = 0;
1984         rp = &sop->so_replay;
1985         rp->rp_status = nfserr_serverfault;
1986         rp->rp_buflen = 0;
1987         rp->rp_buf = rp->rp_ibuf;
1988         return sop;
1989 }
1990
1991 static inline void
1992 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
1993         struct nfs4_stateowner *sop = open->op_stateowner;
1994         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
1995
1996         INIT_LIST_HEAD(&stp->st_hash);
1997         INIT_LIST_HEAD(&stp->st_perstateowner);
1998         INIT_LIST_HEAD(&stp->st_lockowners);
1999         INIT_LIST_HEAD(&stp->st_perfile);
2000         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2001         list_add(&stp->st_perstateowner, &sop->so_stateids);
2002         list_add(&stp->st_perfile, &fp->fi_stateids);
2003         stp->st_stateowner = sop;
2004         get_nfs4_file(fp);
2005         stp->st_file = fp;
2006         stp->st_stateid.si_boot = boot_time;
2007         stp->st_stateid.si_stateownerid = sop->so_id;
2008         stp->st_stateid.si_fileid = fp->fi_id;
2009         stp->st_stateid.si_generation = 0;
2010         stp->st_access_bmap = 0;
2011         stp->st_deny_bmap = 0;
2012         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2013                   &stp->st_access_bmap);
2014         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2015         stp->st_openstp = NULL;
2016 }
2017
2018 static void
2019 move_to_close_lru(struct nfs4_stateowner *sop)
2020 {
2021         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2022
2023         list_move_tail(&sop->so_close_lru, &close_lru);
2024         sop->so_time = get_seconds();
2025 }
2026
2027 static int
2028 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2029                                                         clientid_t *clid)
2030 {
2031         return (sop->so_owner.len == owner->len) &&
2032                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2033                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2034 }
2035
2036 static struct nfs4_stateowner *
2037 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2038 {
2039         struct nfs4_stateowner *so = NULL;
2040
2041         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2042                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2043                         return so;
2044         }
2045         return NULL;
2046 }
2047
2048 /* search file_hashtbl[] for file */
2049 static struct nfs4_file *
2050 find_file(struct inode *ino)
2051 {
2052         unsigned int hashval = file_hashval(ino);
2053         struct nfs4_file *fp;
2054
2055         spin_lock(&recall_lock);
2056         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2057                 if (fp->fi_inode == ino) {
2058                         get_nfs4_file(fp);
2059                         spin_unlock(&recall_lock);
2060                         return fp;
2061                 }
2062         }
2063         spin_unlock(&recall_lock);
2064         return NULL;
2065 }
2066
2067 static inline int access_valid(u32 x, u32 minorversion)
2068 {
2069         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2070                 return 0;
2071         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2072                 return 0;
2073         x &= ~NFS4_SHARE_ACCESS_MASK;
2074         if (minorversion && x) {
2075                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2076                         return 0;
2077                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2078                         return 0;
2079                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2080         }
2081         if (x)
2082                 return 0;
2083         return 1;
2084 }
2085
2086 static inline int deny_valid(u32 x)
2087 {
2088         /* Note: unlike access bits, deny bits may be zero. */
2089         return x <= NFS4_SHARE_DENY_BOTH;
2090 }
2091
2092 /*
2093  * Called to check deny when READ with all zero stateid or
2094  * WRITE with all zero or all one stateid
2095  */
2096 static __be32
2097 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2098 {
2099         struct inode *ino = current_fh->fh_dentry->d_inode;
2100         struct nfs4_file *fp;
2101         struct nfs4_stateid *stp;
2102         __be32 ret;
2103
2104         dprintk("NFSD: nfs4_share_conflict\n");
2105
2106         fp = find_file(ino);
2107         if (!fp)
2108                 return nfs_ok;
2109         ret = nfserr_locked;
2110         /* Search for conflicting share reservations */
2111         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2112                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2113                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2114                         goto out;
2115         }
2116         ret = nfs_ok;
2117 out:
2118         put_nfs4_file(fp);
2119         return ret;
2120 }
2121
2122 static inline void
2123 nfs4_file_downgrade(struct nfs4_file *fp, unsigned int share_access)
2124 {
2125         if (share_access & NFS4_SHARE_ACCESS_WRITE)
2126                 nfs4_file_put_access(fp, O_WRONLY);
2127         if (share_access & NFS4_SHARE_ACCESS_READ)
2128                 nfs4_file_put_access(fp, O_RDONLY);
2129 }
2130
2131 /*
2132  * Spawn a thread to perform a recall on the delegation represented
2133  * by the lease (file_lock)
2134  *
2135  * Called from break_lease() with lock_kernel() held.
2136  * Note: we assume break_lease will only call this *once* for any given
2137  * lease.
2138  */
2139 static
2140 void nfsd_break_deleg_cb(struct file_lock *fl)
2141 {
2142         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2143
2144         dprintk("NFSD nfsd_break_deleg_cb: dp %p fl %p\n",dp,fl);
2145         if (!dp)
2146                 return;
2147
2148         /* We're assuming the state code never drops its reference
2149          * without first removing the lease.  Since we're in this lease
2150          * callback (and since the lease code is serialized by the kernel
2151          * lock) we know the server hasn't removed the lease yet, we know
2152          * it's safe to take a reference: */
2153         atomic_inc(&dp->dl_count);
2154
2155         spin_lock(&recall_lock);
2156         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2157         spin_unlock(&recall_lock);
2158
2159         /* only place dl_time is set. protected by lock_kernel*/
2160         dp->dl_time = get_seconds();
2161
2162         /*
2163          * We don't want the locks code to timeout the lease for us;
2164          * we'll remove it ourself if the delegation isn't returned
2165          * in time.
2166          */
2167         fl->fl_break_time = 0;
2168
2169         dp->dl_file->fi_had_conflict = true;
2170         nfsd4_cb_recall(dp);
2171 }
2172
2173 /*
2174  * The file_lock is being reapd.
2175  *
2176  * Called by locks_free_lock() with lock_kernel() held.
2177  */
2178 static
2179 void nfsd_release_deleg_cb(struct file_lock *fl)
2180 {
2181         struct nfs4_delegation *dp = (struct nfs4_delegation *)fl->fl_owner;
2182
2183         dprintk("NFSD nfsd_release_deleg_cb: fl %p dp %p dl_count %d\n", fl,dp, atomic_read(&dp->dl_count));
2184
2185         if (!(fl->fl_flags & FL_LEASE) || !dp)
2186                 return;
2187         dp->dl_flock = NULL;
2188 }
2189
2190 /*
2191  * Set the delegation file_lock back pointer.
2192  *
2193  * Called from setlease() with lock_kernel() held.
2194  */
2195 static
2196 void nfsd_copy_lock_deleg_cb(struct file_lock *new, struct file_lock *fl)
2197 {
2198         struct nfs4_delegation *dp = (struct nfs4_delegation *)new->fl_owner;
2199
2200         dprintk("NFSD: nfsd_copy_lock_deleg_cb: new fl %p dp %p\n", new, dp);
2201         if (!dp)
2202                 return;
2203         dp->dl_flock = new;
2204 }
2205
2206 /*
2207  * Called from setlease() with lock_kernel() held
2208  */
2209 static
2210 int nfsd_same_client_deleg_cb(struct file_lock *onlist, struct file_lock *try)
2211 {
2212         struct nfs4_delegation *onlistd =
2213                 (struct nfs4_delegation *)onlist->fl_owner;
2214         struct nfs4_delegation *tryd =
2215                 (struct nfs4_delegation *)try->fl_owner;
2216
2217         if (onlist->fl_lmops != try->fl_lmops)
2218                 return 0;
2219
2220         return onlistd->dl_client == tryd->dl_client;
2221 }
2222
2223
2224 static
2225 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2226 {
2227         if (arg & F_UNLCK)
2228                 return lease_modify(onlist, arg);
2229         else
2230                 return -EAGAIN;
2231 }
2232
2233 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2234         .fl_break = nfsd_break_deleg_cb,
2235         .fl_release_private = nfsd_release_deleg_cb,
2236         .fl_copy_lock = nfsd_copy_lock_deleg_cb,
2237         .fl_mylease = nfsd_same_client_deleg_cb,
2238         .fl_change = nfsd_change_deleg_cb,
2239 };
2240
2241
2242 __be32
2243 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2244                     struct nfsd4_open *open)
2245 {
2246         clientid_t *clientid = &open->op_clientid;
2247         struct nfs4_client *clp = NULL;
2248         unsigned int strhashval;
2249         struct nfs4_stateowner *sop = NULL;
2250
2251         if (!check_name(open->op_owner))
2252                 return nfserr_inval;
2253
2254         if (STALE_CLIENTID(&open->op_clientid))
2255                 return nfserr_stale_clientid;
2256
2257         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2258         sop = find_openstateowner_str(strhashval, open);
2259         open->op_stateowner = sop;
2260         if (!sop) {
2261                 /* Make sure the client's lease hasn't expired. */
2262                 clp = find_confirmed_client(clientid);
2263                 if (clp == NULL)
2264                         return nfserr_expired;
2265                 goto renew;
2266         }
2267         /* When sessions are used, skip open sequenceid processing */
2268         if (nfsd4_has_session(cstate))
2269                 goto renew;
2270         if (!sop->so_confirmed) {
2271                 /* Replace unconfirmed owners without checking for replay. */
2272                 clp = sop->so_client;
2273                 release_openowner(sop);
2274                 open->op_stateowner = NULL;
2275                 goto renew;
2276         }
2277         if (open->op_seqid == sop->so_seqid - 1) {
2278                 if (sop->so_replay.rp_buflen)
2279                         return nfserr_replay_me;
2280                 /* The original OPEN failed so spectacularly
2281                  * that we don't even have replay data saved!
2282                  * Therefore, we have no choice but to continue
2283                  * processing this OPEN; presumably, we'll
2284                  * fail again for the same reason.
2285                  */
2286                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2287                 goto renew;
2288         }
2289         if (open->op_seqid != sop->so_seqid)
2290                 return nfserr_bad_seqid;
2291 renew:
2292         if (open->op_stateowner == NULL) {
2293                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2294                 if (sop == NULL)
2295                         return nfserr_resource;
2296                 open->op_stateowner = sop;
2297         }
2298         list_del_init(&sop->so_close_lru);
2299         renew_client(sop->so_client);
2300         return nfs_ok;
2301 }
2302
2303 static inline __be32
2304 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2305 {
2306         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2307                 return nfserr_openmode;
2308         else
2309                 return nfs_ok;
2310 }
2311
2312 static struct nfs4_delegation *
2313 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2314 {
2315         struct nfs4_delegation *dp;
2316
2317         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile) {
2318                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid)
2319                         return dp;
2320         }
2321         return NULL;
2322 }
2323
2324 int share_access_to_flags(u32 share_access)
2325 {
2326         share_access &= ~NFS4_SHARE_WANT_MASK;
2327
2328         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2329 }
2330
2331 static __be32
2332 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2333                 struct nfs4_delegation **dp)
2334 {
2335         int flags;
2336         __be32 status = nfserr_bad_stateid;
2337
2338         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2339         if (*dp == NULL)
2340                 goto out;
2341         flags = share_access_to_flags(open->op_share_access);
2342         status = nfs4_check_delegmode(*dp, flags);
2343         if (status)
2344                 *dp = NULL;
2345 out:
2346         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2347                 return nfs_ok;
2348         if (status)
2349                 return status;
2350         open->op_stateowner->so_confirmed = 1;
2351         return nfs_ok;
2352 }
2353
2354 static __be32
2355 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2356 {
2357         struct nfs4_stateid *local;
2358         __be32 status = nfserr_share_denied;
2359         struct nfs4_stateowner *sop = open->op_stateowner;
2360
2361         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2362                 /* ignore lock owners */
2363                 if (local->st_stateowner->so_is_open_owner == 0)
2364                         continue;
2365                 /* remember if we have seen this open owner */
2366                 if (local->st_stateowner == sop)
2367                         *stpp = local;
2368                 /* check for conflicting share reservations */
2369                 if (!test_share(local, open))
2370                         goto out;
2371         }
2372         status = 0;
2373 out:
2374         return status;
2375 }
2376
2377 static inline struct nfs4_stateid *
2378 nfs4_alloc_stateid(void)
2379 {
2380         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2381 }
2382
2383 static inline int nfs4_access_to_access(u32 nfs4_access)
2384 {
2385         int flags = 0;
2386
2387         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2388                 flags |= NFSD_MAY_READ;
2389         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2390                 flags |= NFSD_MAY_WRITE;
2391         return flags;
2392 }
2393
2394 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file
2395 *fp, struct svc_fh *cur_fh, u32 nfs4_access)
2396 {
2397         __be32 status;
2398         int oflag = nfs4_access_to_omode(nfs4_access);
2399         int access = nfs4_access_to_access(nfs4_access);
2400
2401         if (!fp->fi_fds[oflag]) {
2402                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2403                         &fp->fi_fds[oflag]);
2404                 if (status == nfserr_dropit)
2405                         status = nfserr_jukebox;
2406                 if (status)
2407                         return status;
2408         }
2409         nfs4_file_get_access(fp, oflag);
2410
2411         return nfs_ok;
2412 }
2413
2414 static __be32
2415 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2416                 struct nfs4_file *fp, struct svc_fh *cur_fh,
2417                 struct nfsd4_open *open)
2418 {
2419         struct nfs4_stateid *stp;
2420         __be32 status;
2421
2422         stp = nfs4_alloc_stateid();
2423         if (stp == NULL)
2424                 return nfserr_resource;
2425
2426         status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open->op_share_access);
2427         if (status) {
2428                 kmem_cache_free(stateid_slab, stp);
2429                 return status;
2430         }
2431         *stpp = stp;
2432         return 0;
2433 }
2434
2435 static inline __be32
2436 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2437                 struct nfsd4_open *open)
2438 {
2439         struct iattr iattr = {
2440                 .ia_valid = ATTR_SIZE,
2441                 .ia_size = 0,
2442         };
2443         if (!open->op_truncate)
2444                 return 0;
2445         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2446                 return nfserr_inval;
2447         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2448 }
2449
2450 static __be32
2451 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2452 {
2453         u32 op_share_access, new_access;
2454         __be32 status;
2455
2456         set_access(&new_access, stp->st_access_bmap);
2457         new_access = (~new_access) & open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2458
2459         if (new_access) {
2460                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, new_access);
2461                 if (status)
2462                         return status;
2463         }
2464         status = nfsd4_truncate(rqstp, cur_fh, open);
2465         if (status) {
2466                 if (new_access) {
2467                         int oflag = nfs4_access_to_omode(new_access);
2468                         nfs4_file_put_access(fp, oflag);
2469                 }
2470                 return status;
2471         }
2472         /* remember the open */
2473         op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2474         __set_bit(op_share_access, &stp->st_access_bmap);
2475         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2476
2477         return nfs_ok;
2478 }
2479
2480
2481 static void
2482 nfs4_set_claim_prev(struct nfsd4_open *open)
2483 {
2484         open->op_stateowner->so_confirmed = 1;
2485         open->op_stateowner->so_client->cl_firststate = 1;
2486 }
2487
2488 /*
2489  * Attempt to hand out a delegation.
2490  */
2491 static void
2492 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2493 {
2494         struct nfs4_delegation *dp;
2495         struct nfs4_stateowner *sop = stp->st_stateowner;
2496         int cb_up = atomic_read(&sop->so_client->cl_cb_set);
2497         struct file_lock fl, *flp = &fl;
2498         int status, flag = 0;
2499
2500         flag = NFS4_OPEN_DELEGATE_NONE;
2501         open->op_recall = 0;
2502         switch (open->op_claim_type) {
2503                 case NFS4_OPEN_CLAIM_PREVIOUS:
2504                         if (!cb_up)
2505                                 open->op_recall = 1;
2506                         flag = open->op_delegate_type;
2507                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2508                                 goto out;
2509                         break;
2510                 case NFS4_OPEN_CLAIM_NULL:
2511                         /* Let's not give out any delegations till everyone's
2512                          * had the chance to reclaim theirs.... */
2513                         if (locks_in_grace())
2514                                 goto out;
2515                         if (!cb_up || !sop->so_confirmed)
2516                                 goto out;
2517                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2518                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2519                         else
2520                                 flag = NFS4_OPEN_DELEGATE_READ;
2521                         break;
2522                 default:
2523                         goto out;
2524         }
2525
2526         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2527         if (dp == NULL) {
2528                 flag = NFS4_OPEN_DELEGATE_NONE;
2529                 goto out;
2530         }
2531         locks_init_lock(&fl);
2532         fl.fl_lmops = &nfsd_lease_mng_ops;
2533         fl.fl_flags = FL_LEASE;
2534         fl.fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2535         fl.fl_end = OFFSET_MAX;
2536         fl.fl_owner =  (fl_owner_t)dp;
2537         fl.fl_file = find_readable_file(stp->st_file);
2538         BUG_ON(!fl.fl_file);
2539         fl.fl_pid = current->tgid;
2540
2541         /* vfs_setlease checks to see if delegation should be handed out.
2542          * the lock_manager callbacks fl_mylease and fl_change are used
2543          */
2544         if ((status = vfs_setlease(fl.fl_file, fl.fl_type, &flp))) {
2545                 dprintk("NFSD: setlease failed [%d], no delegation\n", status);
2546                 unhash_delegation(dp);
2547                 flag = NFS4_OPEN_DELEGATE_NONE;
2548                 goto out;
2549         }
2550
2551         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2552
2553         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2554                 STATEID_VAL(&dp->dl_stateid));
2555 out:
2556         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2557                         && flag == NFS4_OPEN_DELEGATE_NONE
2558                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2559                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2560         open->op_delegate_type = flag;
2561 }
2562
2563 /*
2564  * called with nfs4_lock_state() held.
2565  */
2566 __be32
2567 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2568 {
2569         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2570         struct nfs4_file *fp = NULL;
2571         struct inode *ino = current_fh->fh_dentry->d_inode;
2572         struct nfs4_stateid *stp = NULL;
2573         struct nfs4_delegation *dp = NULL;
2574         __be32 status;
2575
2576         status = nfserr_inval;
2577         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2578                         || !deny_valid(open->op_share_deny))
2579                 goto out;
2580         /*
2581          * Lookup file; if found, lookup stateid and check open request,
2582          * and check for delegations in the process of being recalled.
2583          * If not found, create the nfs4_file struct
2584          */
2585         fp = find_file(ino);
2586         if (fp) {
2587                 if ((status = nfs4_check_open(fp, open, &stp)))
2588                         goto out;
2589                 status = nfs4_check_deleg(fp, open, &dp);
2590                 if (status)
2591                         goto out;
2592         } else {
2593                 status = nfserr_bad_stateid;
2594                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2595                         goto out;
2596                 status = nfserr_resource;
2597                 fp = alloc_init_file(ino);
2598                 if (fp == NULL)
2599                         goto out;
2600         }
2601
2602         /*
2603          * OPEN the file, or upgrade an existing OPEN.
2604          * If truncate fails, the OPEN fails.
2605          */
2606         if (stp) {
2607                 /* Stateid was found, this is an OPEN upgrade */
2608                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2609                 if (status)
2610                         goto out;
2611                 update_stateid(&stp->st_stateid);
2612         } else {
2613                 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2614                 if (status)
2615                         goto out;
2616                 init_stateid(stp, fp, open);
2617                 status = nfsd4_truncate(rqstp, current_fh, open);
2618                 if (status) {
2619                         release_open_stateid(stp);
2620                         goto out;
2621                 }
2622                 if (nfsd4_has_session(&resp->cstate))
2623                         update_stateid(&stp->st_stateid);
2624         }
2625         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2626
2627         if (nfsd4_has_session(&resp->cstate))
2628                 open->op_stateowner->so_confirmed = 1;
2629
2630         /*
2631         * Attempt to hand out a delegation. No error return, because the
2632         * OPEN succeeds even if we fail.
2633         */
2634         nfs4_open_delegation(current_fh, open, stp);
2635
2636         status = nfs_ok;
2637
2638         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2639                 STATEID_VAL(&stp->st_stateid));
2640 out:
2641         if (fp)
2642                 put_nfs4_file(fp);
2643         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2644                 nfs4_set_claim_prev(open);
2645         /*
2646         * To finish the open response, we just need to set the rflags.
2647         */
2648         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2649         if (!open->op_stateowner->so_confirmed &&
2650             !nfsd4_has_session(&resp->cstate))
2651                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2652
2653         return status;
2654 }
2655
2656 __be32
2657 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2658             clientid_t *clid)
2659 {
2660         struct nfs4_client *clp;
2661         __be32 status;
2662
2663         nfs4_lock_state();
2664         dprintk("process_renew(%08x/%08x): starting\n", 
2665                         clid->cl_boot, clid->cl_id);
2666         status = nfserr_stale_clientid;
2667         if (STALE_CLIENTID(clid))
2668                 goto out;
2669         clp = find_confirmed_client(clid);
2670         status = nfserr_expired;
2671         if (clp == NULL) {
2672                 /* We assume the client took too long to RENEW. */
2673                 dprintk("nfsd4_renew: clientid not found!\n");
2674                 goto out;
2675         }
2676         renew_client(clp);
2677         status = nfserr_cb_path_down;
2678         if (!list_empty(&clp->cl_delegations)
2679                         && !atomic_read(&clp->cl_cb_set))
2680                 goto out;
2681         status = nfs_ok;
2682 out:
2683         nfs4_unlock_state();
2684         return status;
2685 }
2686
2687 struct lock_manager nfsd4_manager = {
2688 };
2689
2690 static void
2691 nfsd4_end_grace(void)
2692 {
2693         dprintk("NFSD: end of grace period\n");
2694         nfsd4_recdir_purge_old();
2695         locks_end_grace(&nfsd4_manager);
2696         /*
2697          * Now that every NFSv4 client has had the chance to recover and
2698          * to see the (possibly new, possibly shorter) lease time, we
2699          * can safely set the next grace time to the current lease time:
2700          */
2701         nfsd4_grace = nfsd4_lease;
2702 }
2703
2704 static time_t
2705 nfs4_laundromat(void)
2706 {
2707         struct nfs4_client *clp;
2708         struct nfs4_stateowner *sop;
2709         struct nfs4_delegation *dp;
2710         struct list_head *pos, *next, reaplist;
2711         time_t cutoff = get_seconds() - nfsd4_lease;
2712         time_t t, clientid_val = nfsd4_lease;
2713         time_t u, test_val = nfsd4_lease;
2714
2715         nfs4_lock_state();
2716
2717         dprintk("NFSD: laundromat service - starting\n");
2718         if (locks_in_grace())
2719                 nfsd4_end_grace();
2720         INIT_LIST_HEAD(&reaplist);
2721         spin_lock(&client_lock);
2722         list_for_each_safe(pos, next, &client_lru) {
2723                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2724                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
2725                         t = clp->cl_time - cutoff;
2726                         if (clientid_val > t)
2727                                 clientid_val = t;
2728                         break;
2729                 }
2730                 if (atomic_read(&clp->cl_refcount)) {
2731                         dprintk("NFSD: client in use (clientid %08x)\n",
2732                                 clp->cl_clientid.cl_id);
2733                         continue;
2734                 }
2735                 unhash_client_locked(clp);
2736                 list_add(&clp->cl_lru, &reaplist);
2737         }
2738         spin_unlock(&client_lock);
2739         list_for_each_safe(pos, next, &reaplist) {
2740                 clp = list_entry(pos, struct nfs4_client, cl_lru);
2741                 dprintk("NFSD: purging unused client (clientid %08x)\n",
2742                         clp->cl_clientid.cl_id);
2743                 nfsd4_remove_clid_dir(clp);
2744                 expire_client(clp);
2745         }
2746         spin_lock(&recall_lock);
2747         list_for_each_safe(pos, next, &del_recall_lru) {
2748                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2749                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
2750                         u = dp->dl_time - cutoff;
2751                         if (test_val > u)
2752                                 test_val = u;
2753                         break;
2754                 }
2755                 dprintk("NFSD: purging unused delegation dp %p, fp %p\n",
2756                                     dp, dp->dl_flock);
2757                 list_move(&dp->dl_recall_lru, &reaplist);
2758         }
2759         spin_unlock(&recall_lock);
2760         list_for_each_safe(pos, next, &reaplist) {
2761                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
2762                 list_del_init(&dp->dl_recall_lru);
2763                 unhash_delegation(dp);
2764         }
2765         test_val = nfsd4_lease;
2766         list_for_each_safe(pos, next, &close_lru) {
2767                 sop = list_entry(pos, struct nfs4_stateowner, so_close_lru);
2768                 if (time_after((unsigned long)sop->so_time, (unsigned long)cutoff)) {
2769                         u = sop->so_time - cutoff;
2770                         if (test_val > u)
2771                                 test_val = u;
2772                         break;
2773                 }
2774                 dprintk("NFSD: purging unused open stateowner (so_id %d)\n",
2775                         sop->so_id);
2776                 release_openowner(sop);
2777         }
2778         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
2779                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
2780         nfs4_unlock_state();
2781         return clientid_val;
2782 }
2783
2784 static struct workqueue_struct *laundry_wq;
2785 static void laundromat_main(struct work_struct *);
2786 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
2787
2788 static void
2789 laundromat_main(struct work_struct *not_used)
2790 {
2791         time_t t;
2792
2793         t = nfs4_laundromat();
2794         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
2795         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
2796 }
2797
2798 static struct nfs4_stateowner *
2799 search_close_lru(u32 st_id, int flags)
2800 {
2801         struct nfs4_stateowner *local = NULL;
2802
2803         if (flags & CLOSE_STATE) {
2804                 list_for_each_entry(local, &close_lru, so_close_lru) {
2805                         if (local->so_id == st_id)
2806                                 return local;
2807                 }
2808         }
2809         return NULL;
2810 }
2811
2812 static inline int
2813 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stateid *stp)
2814 {
2815         return fhp->fh_dentry->d_inode != stp->st_file->fi_inode;
2816 }
2817
2818 static int
2819 STALE_STATEID(stateid_t *stateid)
2820 {
2821         if (stateid->si_boot == boot_time)
2822                 return 0;
2823         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
2824                 STATEID_VAL(stateid));
2825         return 1;
2826 }
2827
2828 static inline int
2829 access_permit_read(unsigned long access_bmap)
2830 {
2831         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
2832                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
2833                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
2834 }
2835
2836 static inline int
2837 access_permit_write(unsigned long access_bmap)
2838 {
2839         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
2840                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
2841 }
2842
2843 static
2844 __be32 nfs4_check_openmode(struct nfs4_stateid *stp, int flags)
2845 {
2846         __be32 status = nfserr_openmode;
2847
2848         /* For lock stateid's, we test the parent open, not the lock: */
2849         if (stp->st_openstp)
2850                 stp = stp->st_openstp;
2851         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
2852                 goto out;
2853         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
2854                 goto out;
2855         status = nfs_ok;
2856 out:
2857         return status;
2858 }
2859
2860 static inline __be32
2861 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
2862 {
2863         if (ONE_STATEID(stateid) && (flags & RD_STATE))
2864                 return nfs_ok;
2865         else if (locks_in_grace()) {
2866                 /* Answer in remaining cases depends on existance of
2867                  * conflicting state; so we must wait out the grace period. */
2868                 return nfserr_grace;
2869         } else if (flags & WR_STATE)
2870                 return nfs4_share_conflict(current_fh,
2871                                 NFS4_SHARE_DENY_WRITE);
2872         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
2873                 return nfs4_share_conflict(current_fh,
2874                                 NFS4_SHARE_DENY_READ);
2875 }
2876
2877 /*
2878  * Allow READ/WRITE during grace period on recovered state only for files
2879  * that are not able to provide mandatory locking.
2880  */
2881 static inline int
2882 grace_disallows_io(struct inode *inode)
2883 {
2884         return locks_in_grace() && mandatory_lock(inode);
2885 }
2886
2887 static int check_stateid_generation(stateid_t *in, stateid_t *ref, int flags)
2888 {
2889         /*
2890          * When sessions are used the stateid generation number is ignored
2891          * when it is zero.
2892          */
2893         if ((flags & HAS_SESSION) && in->si_generation == 0)
2894                 goto out;
2895
2896         /* If the client sends us a stateid from the future, it's buggy: */
2897         if (in->si_generation > ref->si_generation)
2898                 return nfserr_bad_stateid;
2899         /*
2900          * The following, however, can happen.  For example, if the
2901          * client sends an open and some IO at the same time, the open
2902          * may bump si_generation while the IO is still in flight.
2903          * Thanks to hard links and renames, the client never knows what
2904          * file an open will affect.  So it could avoid that situation
2905          * only by serializing all opens and IO from the same open
2906          * owner.  To recover from the old_stateid error, the client
2907          * will just have to retry the IO:
2908          */
2909         if (in->si_generation < ref->si_generation)
2910                 return nfserr_old_stateid;
2911 out:
2912         return nfs_ok;
2913 }
2914
2915 static int is_delegation_stateid(stateid_t *stateid)
2916 {
2917         return stateid->si_fileid == 0;
2918 }
2919
2920 /*
2921 * Checks for stateid operations
2922 */
2923 __be32
2924 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
2925                            stateid_t *stateid, int flags, struct file **filpp)
2926 {
2927         struct nfs4_stateid *stp = NULL;
2928         struct nfs4_delegation *dp = NULL;
2929         struct svc_fh *current_fh = &cstate->current_fh;
2930         struct inode *ino = current_fh->fh_dentry->d_inode;
2931         __be32 status;
2932
2933         if (filpp)
2934                 *filpp = NULL;
2935
2936         if (grace_disallows_io(ino))
2937                 return nfserr_grace;
2938
2939         if (nfsd4_has_session(cstate))
2940                 flags |= HAS_SESSION;
2941
2942         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
2943                 return check_special_stateids(current_fh, stateid, flags);
2944
2945         status = nfserr_stale_stateid;
2946         if (STALE_STATEID(stateid)) 
2947                 goto out;
2948
2949         status = nfserr_bad_stateid;
2950         if (is_delegation_stateid(stateid)) {
2951                 dp = find_delegation_stateid(ino, stateid);
2952                 if (!dp)
2953                         goto out;
2954                 status = check_stateid_generation(stateid, &dp->dl_stateid,
2955                                                   flags);
2956                 if (status)
2957                         goto out;
2958                 status = nfs4_check_delegmode(dp, flags);
2959                 if (status)
2960                         goto out;
2961                 renew_client(dp->dl_client);
2962                 if (filpp)
2963                         *filpp = find_readable_file(dp->dl_file);
2964                 BUG_ON(!*filpp);
2965         } else { /* open or lock stateid */
2966                 stp = find_stateid(stateid, flags);
2967                 if (!stp)
2968                         goto out;
2969                 if (nfs4_check_fh(current_fh, stp))
2970                         goto out;
2971                 if (!stp->st_stateowner->so_confirmed)
2972                         goto out;
2973                 status = check_stateid_generation(stateid, &stp->st_stateid,
2974                                                   flags);
2975                 if (status)
2976                         goto out;
2977                 status = nfs4_check_openmode(stp, flags);
2978                 if (status)
2979                         goto out;
2980                 renew_client(stp->st_stateowner->so_client);
2981                 if (filpp) {
2982                         if (flags & RD_STATE)
2983                                 *filpp = find_readable_file(stp->st_file);
2984                         else
2985                                 *filpp = find_writeable_file(stp->st_file);
2986                         BUG_ON(!*filpp); /* assured by check_openmode */
2987                 }
2988         }
2989         status = nfs_ok;
2990 out:
2991         return status;
2992 }
2993
2994 static inline int
2995 setlkflg (int type)
2996 {
2997         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
2998                 RD_STATE : WR_STATE;
2999 }
3000
3001 /* 
3002  * Checks for sequence id mutating operations. 
3003  */
3004 static __be32
3005 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3006                          stateid_t *stateid, int flags,
3007                          struct nfs4_stateowner **sopp,
3008                          struct nfs4_stateid **stpp, struct nfsd4_lock *lock)
3009 {
3010         struct nfs4_stateid *stp;
3011         struct nfs4_stateowner *sop;
3012         struct svc_fh *current_fh = &cstate->current_fh;
3013         __be32 status;
3014
3015         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3016                 seqid, STATEID_VAL(stateid));
3017
3018         *stpp = NULL;
3019         *sopp = NULL;
3020
3021         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
3022                 dprintk("NFSD: preprocess_seqid_op: magic stateid!\n");
3023                 return nfserr_bad_stateid;
3024         }
3025
3026         if (STALE_STATEID(stateid))
3027                 return nfserr_stale_stateid;
3028
3029         if (nfsd4_has_session(cstate))
3030                 flags |= HAS_SESSION;
3031
3032         /*
3033         * We return BAD_STATEID if filehandle doesn't match stateid, 
3034         * the confirmed flag is incorrecly set, or the generation 
3035         * number is incorrect.  
3036         */
3037         stp = find_stateid(stateid, flags);
3038         if (stp == NULL) {
3039                 /*
3040                  * Also, we should make sure this isn't just the result of
3041                  * a replayed close:
3042                  */
3043                 sop = search_close_lru(stateid->si_stateownerid, flags);
3044                 if (sop == NULL)
3045                         return nfserr_bad_stateid;
3046                 *sopp = sop;
3047                 goto check_replay;
3048         }
3049
3050         *stpp = stp;
3051         *sopp = sop = stp->st_stateowner;
3052
3053         if (lock) {
3054                 clientid_t *lockclid = &lock->v.new.clientid;
3055                 struct nfs4_client *clp = sop->so_client;
3056                 int lkflg = 0;
3057                 __be32 status;
3058
3059                 lkflg = setlkflg(lock->lk_type);
3060
3061                 if (lock->lk_is_new) {
3062                         if (!sop->so_is_open_owner)
3063                                 return nfserr_bad_stateid;
3064                         if (!(flags & HAS_SESSION) &&
3065                             !same_clid(&clp->cl_clientid, lockclid))
3066                                 return nfserr_bad_stateid;
3067                         /* stp is the open stateid */
3068                         status = nfs4_check_openmode(stp, lkflg);
3069                         if (status)
3070                                 return status;
3071                 } else {
3072                         /* stp is the lock stateid */
3073                         status = nfs4_check_openmode(stp->st_openstp, lkflg);
3074                         if (status)
3075                                 return status;
3076                }
3077         }
3078
3079         if (nfs4_check_fh(current_fh, stp)) {
3080                 dprintk("NFSD: preprocess_seqid_op: fh-stateid mismatch!\n");
3081                 return nfserr_bad_stateid;
3082         }
3083
3084         /*
3085         *  We now validate the seqid and stateid generation numbers.
3086         *  For the moment, we ignore the possibility of 
3087         *  generation number wraparound.
3088         */
3089         if (!(flags & HAS_SESSION) && seqid != sop->so_seqid)
3090                 goto check_replay;
3091
3092         if (sop->so_confirmed && flags & CONFIRM) {
3093                 dprintk("NFSD: preprocess_seqid_op: expected"
3094                                 " unconfirmed stateowner!\n");
3095                 return nfserr_bad_stateid;
3096         }
3097         if (!sop->so_confirmed && !(flags & CONFIRM)) {
3098                 dprintk("NFSD: preprocess_seqid_op: stateowner not"
3099                                 " confirmed yet!\n");
3100                 return nfserr_bad_stateid;
3101         }
3102         status = check_stateid_generation(stateid, &stp->st_stateid, flags);
3103         if (status)
3104                 return status;
3105         renew_client(sop->so_client);
3106         return nfs_ok;
3107
3108 check_replay:
3109         if (seqid == sop->so_seqid - 1) {
3110                 dprintk("NFSD: preprocess_seqid_op: retransmission?\n");
3111                 /* indicate replay to calling function */
3112                 return nfserr_replay_me;
3113         }
3114         dprintk("NFSD: preprocess_seqid_op: bad seqid (expected %d, got %d)\n",
3115                         sop->so_seqid, seqid);
3116         *sopp = NULL;
3117         return nfserr_bad_seqid;
3118 }
3119
3120 __be32
3121 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3122                    struct nfsd4_open_confirm *oc)
3123 {
3124         __be32 status;
3125         struct nfs4_stateowner *sop;
3126         struct nfs4_stateid *stp;
3127
3128         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3129                         (int)cstate->current_fh.fh_dentry->d_name.len,
3130                         cstate->current_fh.fh_dentry->d_name.name);
3131
3132         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3133         if (status)
3134                 return status;
3135
3136         nfs4_lock_state();
3137
3138         if ((status = nfs4_preprocess_seqid_op(cstate,
3139                                         oc->oc_seqid, &oc->oc_req_stateid,
3140                                         CONFIRM | OPEN_STATE,
3141                                         &oc->oc_stateowner, &stp, NULL)))
3142                 goto out; 
3143
3144         sop = oc->oc_stateowner;
3145         sop->so_confirmed = 1;
3146         update_stateid(&stp->st_stateid);
3147         memcpy(&oc->oc_resp_stateid, &stp->st_stateid, sizeof(stateid_t));
3148         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3149                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stateid));
3150
3151         nfsd4_create_clid_dir(sop->so_client);
3152 out:
3153         if (oc->oc_stateowner) {
3154                 nfs4_get_stateowner(oc->oc_stateowner);
3155                 cstate->replay_owner = oc->oc_stateowner;
3156         }
3157         nfs4_unlock_state();
3158         return status;
3159 }
3160
3161
3162 /*
3163  * unset all bits in union bitmap (bmap) that
3164  * do not exist in share (from successful OPEN_DOWNGRADE)
3165  */
3166 static void
3167 reset_union_bmap_access(unsigned long access, unsigned long *bmap)
3168 {
3169         int i;
3170         for (i = 1; i < 4; i++) {
3171                 if ((i & access) != i)
3172                         __clear_bit(i, bmap);
3173         }
3174 }
3175
3176 static void
3177 reset_union_bmap_deny(unsigned long deny, unsigned long *bmap)
3178 {
3179         int i;
3180         for (i = 0; i < 4; i++) {
3181                 if ((i & deny) != i)
3182                         __clear_bit(i, bmap);
3183         }
3184 }
3185
3186 __be32
3187 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3188                      struct nfsd4_compound_state *cstate,
3189                      struct nfsd4_open_downgrade *od)
3190 {
3191         __be32 status;
3192         struct nfs4_stateid *stp;
3193         unsigned int share_access;
3194
3195         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3196                         (int)cstate->current_fh.fh_dentry->d_name.len,
3197                         cstate->current_fh.fh_dentry->d_name.name);
3198
3199         if (!access_valid(od->od_share_access, cstate->minorversion)
3200                         || !deny_valid(od->od_share_deny))
3201                 return nfserr_inval;
3202
3203         nfs4_lock_state();
3204         if ((status = nfs4_preprocess_seqid_op(cstate,
3205                                         od->od_seqid,
3206                                         &od->od_stateid, 
3207                                         OPEN_STATE,
3208                                         &od->od_stateowner, &stp, NULL)))
3209                 goto out; 
3210
3211         status = nfserr_inval;
3212         if (!test_bit(od->od_share_access, &stp->st_access_bmap)) {
3213                 dprintk("NFSD:access not a subset current bitmap: 0x%lx, input access=%08x\n",
3214                         stp->st_access_bmap, od->od_share_access);
3215                 goto out;
3216         }
3217         if (!test_bit(od->od_share_deny, &stp->st_deny_bmap)) {
3218                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3219                         stp->st_deny_bmap, od->od_share_deny);
3220                 goto out;
3221         }
3222         set_access(&share_access, stp->st_access_bmap);
3223         nfs4_file_downgrade(stp->st_file, share_access & ~od->od_share_access);
3224
3225         reset_union_bmap_access(od->od_share_access, &stp->st_access_bmap);
3226         reset_union_bmap_deny(od->od_share_deny, &stp->st_deny_bmap);
3227
3228         update_stateid(&stp->st_stateid);
3229         memcpy(&od->od_stateid, &stp->st_stateid, sizeof(stateid_t));
3230         status = nfs_ok;
3231 out:
3232         if (od->od_stateowner) {
3233                 nfs4_get_stateowner(od->od_stateowner);
3234                 cstate->replay_owner = od->od_stateowner;
3235         }
3236         nfs4_unlock_state();
3237         return status;
3238 }
3239
3240 /*
3241  * nfs4_unlock_state() called after encode
3242  */
3243 __be32
3244 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3245             struct nfsd4_close *close)
3246 {
3247         __be32 status;
3248         struct nfs4_stateid *stp;
3249
3250         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3251                         (int)cstate->current_fh.fh_dentry->d_name.len,
3252                         cstate->current_fh.fh_dentry->d_name.name);
3253
3254         nfs4_lock_state();
3255         /* check close_lru for replay */
3256         if ((status = nfs4_preprocess_seqid_op(cstate,
3257                                         close->cl_seqid,
3258                                         &close->cl_stateid, 
3259                                         OPEN_STATE | CLOSE_STATE,
3260                                         &close->cl_stateowner, &stp, NULL)))
3261                 goto out; 
3262         status = nfs_ok;
3263         update_stateid(&stp->st_stateid);
3264         memcpy(&close->cl_stateid, &stp->st_stateid, sizeof(stateid_t));
3265
3266         /* release_stateid() calls nfsd_close() if needed */
3267         release_open_stateid(stp);
3268
3269         /* place unused nfs4_stateowners on so_close_lru list to be
3270          * released by the laundromat service after the lease period
3271          * to enable us to handle CLOSE replay
3272          */
3273         if (list_empty(&close->cl_stateowner->so_stateids))
3274                 move_to_close_lru(close->cl_stateowner);
3275 out:
3276         if (close->cl_stateowner) {
3277                 nfs4_get_stateowner(close->cl_stateowner);
3278                 cstate->replay_owner = close->cl_stateowner;
3279         }
3280         nfs4_unlock_state();
3281         return status;
3282 }
3283
3284 __be32
3285 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3286                   struct nfsd4_delegreturn *dr)
3287 {
3288         struct nfs4_delegation *dp;
3289         stateid_t *stateid = &dr->dr_stateid;
3290         struct inode *inode;
3291         __be32 status;
3292         int flags = 0;
3293
3294         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3295                 return status;
3296         inode = cstate->current_fh.fh_dentry->d_inode;
3297
3298         if (nfsd4_has_session(cstate))
3299                 flags |= HAS_SESSION;
3300         nfs4_lock_state();
3301         status = nfserr_bad_stateid;
3302         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3303                 goto out;
3304         status = nfserr_stale_stateid;
3305         if (STALE_STATEID(stateid))
3306                 goto out;
3307         status = nfserr_bad_stateid;
3308         if (!is_delegation_stateid(stateid))
3309                 goto out;
3310         dp = find_delegation_stateid(inode, stateid);
3311         if (!dp)
3312                 goto out;
3313         status = check_stateid_generation(stateid, &dp->dl_stateid, flags);
3314         if (status)
3315                 goto out;
3316         renew_client(dp->dl_client);
3317
3318         unhash_delegation(dp);
3319 out:
3320         nfs4_unlock_state();
3321
3322         return status;
3323 }
3324
3325
3326 /* 
3327  * Lock owner state (byte-range locks)
3328  */
3329 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3330 #define LOCK_HASH_BITS              8
3331 #define LOCK_HASH_SIZE             (1 << LOCK_HASH_BITS)
3332 #define LOCK_HASH_MASK             (LOCK_HASH_SIZE - 1)
3333
3334 static inline u64
3335 end_offset(u64 start, u64 len)
3336 {
3337         u64 end;
3338
3339         end = start + len;
3340         return end >= start ? end: NFS4_MAX_UINT64;
3341 }
3342
3343 /* last octet in a range */
3344 static inline u64
3345 last_byte_offset(u64 start, u64 len)
3346 {
3347         u64 end;
3348
3349         BUG_ON(!len);
3350         end = start + len;
3351         return end > start ? end - 1: NFS4_MAX_UINT64;
3352 }
3353
3354 #define lockownerid_hashval(id) \
3355         ((id) & LOCK_HASH_MASK)
3356
3357 static inline unsigned int
3358 lock_ownerstr_hashval(struct inode *inode, u32 cl_id,
3359                 struct xdr_netobj *ownername)
3360 {
3361         return (file_hashval(inode) + cl_id
3362                         + opaque_hashval(ownername->data, ownername->len))
3363                 & LOCK_HASH_MASK;
3364 }
3365
3366 static struct list_head lock_ownerid_hashtbl[LOCK_HASH_SIZE];
3367 static struct list_head lock_ownerstr_hashtbl[LOCK_HASH_SIZE];
3368 static struct list_head lockstateid_hashtbl[STATEID_HASH_SIZE];
3369
3370 static struct nfs4_stateid *
3371 find_stateid(stateid_t *stid, int flags)
3372 {
3373         struct nfs4_stateid *local;
3374         u32 st_id = stid->si_stateownerid;
3375         u32 f_id = stid->si_fileid;
3376         unsigned int hashval;
3377
3378         dprintk("NFSD: find_stateid flags 0x%x\n",flags);
3379         if (flags & (LOCK_STATE | RD_STATE | WR_STATE)) {
3380                 hashval = stateid_hashval(st_id, f_id);
3381                 list_for_each_entry(local, &lockstateid_hashtbl[hashval], st_hash) {
3382                         if ((local->st_stateid.si_stateownerid == st_id) &&
3383                             (local->st_stateid.si_fileid == f_id))
3384                                 return local;
3385                 }
3386         } 
3387
3388         if (flags & (OPEN_STATE | RD_STATE | WR_STATE)) {
3389                 hashval = stateid_hashval(st_id, f_id);
3390                 list_for_each_entry(local, &stateid_hashtbl[hashval], st_hash) {
3391                         if ((local->st_stateid.si_stateownerid == st_id) &&
3392                             (local->st_stateid.si_fileid == f_id))
3393                                 return local;
3394                 }
3395         }
3396         return NULL;
3397 }
3398
3399 static struct nfs4_delegation *
3400 find_delegation_stateid(struct inode *ino, stateid_t *stid)
3401 {
3402         struct nfs4_file *fp;
3403         struct nfs4_delegation *dl;
3404
3405         dprintk("NFSD: %s: stateid=" STATEID_FMT "\n", __func__,
3406                 STATEID_VAL(stid));
3407
3408         fp = find_file(ino);
3409         if (!fp)
3410                 return NULL;
3411         dl = find_delegation_file(fp, stid);
3412         put_nfs4_file(fp);
3413         return dl;
3414 }
3415
3416 /*
3417  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3418  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3419  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3420  * locking, this prevents us from being completely protocol-compliant.  The
3421  * real solution to this problem is to start using unsigned file offsets in
3422  * the VFS, but this is a very deep change!
3423  */
3424 static inline void
3425 nfs4_transform_lock_offset(struct file_lock *lock)
3426 {
3427         if (lock->fl_start < 0)
3428                 lock->fl_start = OFFSET_MAX;
3429         if (lock->fl_end < 0)
3430                 lock->fl_end = OFFSET_MAX;
3431 }
3432
3433 /* Hack!: For now, we're defining this just so we can use a pointer to it
3434  * as a unique cookie to identify our (NFSv4's) posix locks. */
3435 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3436 };
3437
3438 static inline void
3439 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3440 {
3441         struct nfs4_stateowner *sop;
3442
3443         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3444                 sop = (struct nfs4_stateowner *) fl->fl_owner;
3445                 kref_get(&sop->so_ref);
3446                 deny->ld_sop = sop;
3447                 deny->ld_clientid = sop->so_client->cl_clientid;
3448         } else {
3449                 deny->ld_sop = NULL;
3450                 deny->ld_clientid.cl_boot = 0;
3451                 deny->ld_clientid.cl_id = 0;
3452         }
3453         deny->ld_start = fl->fl_start;
3454         deny->ld_length = NFS4_MAX_UINT64;
3455         if (fl->fl_end != NFS4_MAX_UINT64)
3456                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3457         deny->ld_type = NFS4_READ_LT;
3458         if (fl->fl_type != F_RDLCK)
3459                 deny->ld_type = NFS4_WRITE_LT;
3460 }
3461
3462 static struct nfs4_stateowner *
3463 find_lockstateowner_str(struct inode *inode, clientid_t *clid,
3464                 struct xdr_netobj *owner)
3465 {
3466         unsigned int hashval = lock_ownerstr_hashval(inode, clid->cl_id, owner);
3467         struct nfs4_stateowner *op;
3468
3469         list_for_each_entry(op, &lock_ownerstr_hashtbl[hashval], so_strhash) {
3470                 if (same_owner_str(op, owner, clid))
3471                         return op;
3472         }
3473         return NULL;
3474 }
3475
3476 /*
3477  * Alloc a lock owner structure.
3478  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3479  * occured. 
3480  *
3481  * strhashval = lock_ownerstr_hashval 
3482  */
3483
3484 static struct nfs4_stateowner *
3485 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_stateid *open_stp, struct nfsd4_lock *lock) {
3486         struct nfs4_stateowner *sop;
3487         struct nfs4_replay *rp;
3488         unsigned int idhashval;
3489
3490         if (!(sop = alloc_stateowner(&lock->lk_new_owner)))
3491                 return NULL;
3492         idhashval = lockownerid_hashval(current_ownerid);
3493         INIT_LIST_HEAD(&sop->so_idhash);
3494         INIT_LIST_HEAD(&sop->so_strhash);
3495         INIT_LIST_HEAD(&sop->so_perclient);
3496         INIT_LIST_HEAD(&sop->so_stateids);
3497         INIT_LIST_HEAD(&sop->so_perstateid);
3498         INIT_LIST_HEAD(&sop->so_close_lru); /* not used */
3499         sop->so_time = 0;
3500         list_add(&sop->so_idhash, &lock_ownerid_hashtbl[idhashval]);
3501         list_add(&sop->so_strhash, &lock_ownerstr_hashtbl[strhashval]);
3502         list_add(&sop->so_perstateid, &open_stp->st_lockowners);
3503         sop->so_is_open_owner = 0;
3504         sop->so_id = current_ownerid++;
3505         sop->so_client = clp;
3506         /* It is the openowner seqid that will be incremented in encode in the
3507          * case of new lockowners; so increment the lock seqid manually: */
3508         sop->so_seqid = lock->lk_new_lock_seqid + 1;
3509         sop->so_confirmed = 1;
3510         rp = &sop->so_replay;
3511         rp->rp_status = nfserr_serverfault;
3512         rp->rp_buflen = 0;
3513         rp->rp_buf = rp->rp_ibuf;
3514         return sop;
3515 }
3516
3517 static struct nfs4_stateid *
3518 alloc_init_lock_stateid(struct nfs4_stateowner *sop, struct nfs4_file *fp, struct nfs4_stateid *open_stp)
3519 {
3520         struct nfs4_stateid *stp;
3521         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
3522
3523         stp = nfs4_alloc_stateid();
3524         if (stp == NULL)
3525                 goto out;
3526         INIT_LIST_HEAD(&stp->st_hash);
3527         INIT_LIST_HEAD(&stp->st_perfile);
3528         INIT_LIST_HEAD(&stp->st_perstateowner);
3529         INIT_LIST_HEAD(&stp->st_lockowners); /* not used */
3530         list_add(&stp->st_hash, &lockstateid_hashtbl[hashval]);
3531         list_add(&stp->st_perfile, &fp->fi_stateids);
3532         list_add(&stp->st_perstateowner, &sop->so_stateids);
3533         stp->st_stateowner = sop;
3534         get_nfs4_file(fp);
3535         stp->st_file = fp;
3536         stp->st_stateid.si_boot = boot_time;
3537         stp->st_stateid.si_stateownerid = sop->so_id;
3538         stp->st_stateid.si_fileid = fp->fi_id;
3539         stp->st_stateid.si_generation = 0;
3540         stp->st_deny_bmap = open_stp->st_deny_bmap;
3541         stp->st_openstp = open_stp;
3542
3543 out:
3544         return stp;
3545 }
3546
3547 static int
3548 check_lock_length(u64 offset, u64 length)
3549 {
3550         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
3551              LOFF_OVERFLOW(offset, length)));
3552 }
3553
3554 /*
3555  *  LOCK operation 
3556  */
3557 __be32
3558 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3559            struct nfsd4_lock *lock)
3560 {
3561         struct nfs4_stateowner *open_sop = NULL;
3562         struct nfs4_stateowner *lock_sop = NULL;
3563         struct nfs4_stateid *lock_stp;
3564         struct file *filp;
3565         struct file_lock file_lock;
3566         struct file_lock conflock;
3567         __be32 status = 0;
3568         unsigned int strhashval;
3569         unsigned int cmd;
3570         int err;
3571
3572         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
3573                 (long long) lock->lk_offset,
3574                 (long long) lock->lk_length);
3575
3576         if (check_lock_length(lock->lk_offset, lock->lk_length))
3577                  return nfserr_inval;
3578
3579         if ((status = fh_verify(rqstp, &cstate->current_fh,
3580                                 S_IFREG, NFSD_MAY_LOCK))) {
3581                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
3582                 return status;
3583         }
3584
3585         nfs4_lock_state();
3586
3587         if (lock->lk_is_new) {
3588                 /*
3589                  * Client indicates that this is a new lockowner.
3590                  * Use open owner and open stateid to create lock owner and
3591                  * lock stateid.
3592                  */
3593                 struct nfs4_stateid *open_stp = NULL;
3594                 struct nfs4_file *fp;
3595                 
3596                 status = nfserr_stale_clientid;
3597                 if (!nfsd4_has_session(cstate) &&
3598                     STALE_CLIENTID(&lock->lk_new_clientid))
3599                         goto out;
3600
3601                 /* validate and update open stateid and open seqid */
3602                 status = nfs4_preprocess_seqid_op(cstate,
3603                                         lock->lk_new_open_seqid,
3604                                         &lock->lk_new_open_stateid,
3605                                         OPEN_STATE,
3606                                         &lock->lk_replay_owner, &open_stp,
3607                                         lock);
3608                 if (status)
3609                         goto out;
3610                 open_sop = lock->lk_replay_owner;
3611                 /* create lockowner and lock stateid */
3612                 fp = open_stp->st_file;
3613                 strhashval = lock_ownerstr_hashval(fp->fi_inode, 
3614                                 open_sop->so_client->cl_clientid.cl_id, 
3615                                 &lock->v.new.owner);
3616                 /* XXX: Do we need to check for duplicate stateowners on
3617                  * the same file, or should they just be allowed (and
3618                  * create new stateids)? */
3619                 status = nfserr_resource;
3620                 lock_sop = alloc_init_lock_stateowner(strhashval,
3621                                 open_sop->so_client, open_stp, lock);
3622                 if (lock_sop == NULL)
3623                         goto out;
3624                 lock_stp = alloc_init_lock_stateid(lock_sop, fp, open_stp);
3625                 if (lock_stp == NULL)
3626                         goto out;
3627         } else {
3628                 /* lock (lock owner + lock stateid) already exists */
3629                 status = nfs4_preprocess_seqid_op(cstate,
3630                                        lock->lk_old_lock_seqid, 
3631                                        &lock->lk_old_lock_stateid, 
3632                                        LOCK_STATE,
3633                                        &lock->lk_replay_owner, &lock_stp, lock);
3634                 if (status)
3635                         goto out;
3636                 lock_sop = lock->lk_replay_owner;
3637         }
3638         /* lock->lk_replay_owner and lock_stp have been created or found */
3639
3640         status = nfserr_grace;
3641         if (locks_in_grace() && !lock->lk_reclaim)
3642                 goto out;
3643         status = nfserr_no_grace;
3644         if (!locks_in_grace() && lock->lk_reclaim)
3645                 goto out;
3646
3647         locks_init_lock(&file_lock);
3648         switch (lock->lk_type) {
3649                 case NFS4_READ_LT:
3650                 case NFS4_READW_LT:
3651                         filp = find_readable_file(lock_stp->st_file);
3652                         file_lock.fl_type = F_RDLCK;
3653                         cmd = F_SETLK;
3654                 break;
3655                 case NFS4_WRITE_LT:
3656                 case NFS4_WRITEW_LT:
3657                         filp = find_writeable_file(lock_stp->st_file);
3658                         file_lock.fl_type = F_WRLCK;
3659                         cmd = F_SETLK;
3660                 break;
3661                 default:
3662                         status = nfserr_inval;
3663                 goto out;
3664         }
3665         if (!filp) {
3666                 status = nfserr_openmode;
3667                 goto out;
3668         }
3669         file_lock.fl_owner = (fl_owner_t)lock_sop;
3670         file_lock.fl_pid = current->tgid;
3671         file_lock.fl_file = filp;
3672         file_lock.fl_flags = FL_POSIX;
3673         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3674
3675         file_lock.fl_start = lock->lk_offset;
3676         file_lock.fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
3677         nfs4_transform_lock_offset(&file_lock);
3678
3679         /*
3680         * Try to lock the file in the VFS.
3681         * Note: locks.c uses the BKL to protect the inode's lock list.
3682         */
3683
3684         err = vfs_lock_file(filp, cmd, &file_lock, &conflock);
3685         switch (-err) {
3686         case 0: /* success! */
3687                 update_stateid(&lock_stp->st_stateid);
3688                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stateid, 
3689                                 sizeof(stateid_t));
3690                 status = 0;
3691                 break;
3692         case (EAGAIN):          /* conflock holds conflicting lock */
3693                 status = nfserr_denied;
3694                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
3695                 nfs4_set_lock_denied(&conflock, &lock->lk_denied);
3696                 break;
3697         case (EDEADLK):
3698                 status = nfserr_deadlock;
3699                 break;
3700         default:        
3701                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
3702                 status = nfserr_resource;
3703                 break;
3704         }
3705 out:
3706         if (status && lock->lk_is_new && lock_sop)
3707                 release_lockowner(lock_sop);
3708         if (lock->lk_replay_owner) {
3709                 nfs4_get_stateowner(lock->lk_replay_owner);
3710                 cstate->replay_owner = lock->lk_replay_owner;
3711         }
3712         nfs4_unlock_state();
3713         return status;
3714 }
3715
3716 /*
3717  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
3718  * so we do a temporary open here just to get an open file to pass to
3719  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
3720  * inode operation.)
3721  */
3722 static int nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
3723 {
3724         struct file *file;
3725         int err;
3726
3727         err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
3728         if (err)
3729                 return err;
3730         err = vfs_test_lock(file, lock);
3731         nfsd_close(file);
3732         return err;
3733 }
3734
3735 /*
3736  * LOCKT operation
3737  */
3738 __be32
3739 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3740             struct nfsd4_lockt *lockt)
3741 {
3742         struct inode *inode;
3743         struct file_lock file_lock;
3744         int error;
3745         __be32 status;
3746
3747         if (locks_in_grace())
3748                 return nfserr_grace;
3749
3750         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
3751                  return nfserr_inval;
3752
3753         lockt->lt_stateowner = NULL;
3754         nfs4_lock_state();
3755
3756         status = nfserr_stale_clientid;
3757         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid))
3758                 goto out;
3759
3760         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0))) {
3761                 dprintk("NFSD: nfsd4_lockt: fh_verify() failed!\n");
3762                 if (status == nfserr_symlink)
3763                         status = nfserr_inval;
3764                 goto out;
3765         }
3766
3767         inode = cstate->current_fh.fh_dentry->d_inode;
3768         locks_init_lock(&file_lock);
3769         switch (lockt->lt_type) {
3770                 case NFS4_READ_LT:
3771                 case NFS4_READW_LT:
3772                         file_lock.fl_type = F_RDLCK;
3773                 break;
3774                 case NFS4_WRITE_LT:
3775                 case NFS4_WRITEW_LT:
3776                         file_lock.fl_type = F_WRLCK;
3777                 break;
3778                 default:
3779                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
3780                         status = nfserr_inval;
3781                 goto out;
3782         }
3783
3784         lockt->lt_stateowner = find_lockstateowner_str(inode,
3785                         &lockt->lt_clientid, &lockt->lt_owner);
3786         if (lockt->lt_stateowner)
3787                 file_lock.fl_owner = (fl_owner_t)lockt->lt_stateowner;
3788         file_lock.fl_pid = current->tgid;
3789         file_lock.fl_flags = FL_POSIX;
3790
3791         file_lock.fl_start = lockt->lt_offset;
3792         file_lock.fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
3793
3794         nfs4_transform_lock_offset(&file_lock);
3795
3796         status = nfs_ok;
3797         error = nfsd_test_lock(rqstp, &cstate->current_fh, &file_lock);
3798         if (error) {
3799                 status = nfserrno(error);
3800                 goto out;
3801         }
3802         if (file_lock.fl_type != F_UNLCK) {
3803                 status = nfserr_denied;
3804                 nfs4_set_lock_denied(&file_lock, &lockt->lt_denied);
3805         }
3806 out:
3807         nfs4_unlock_state();
3808         return status;
3809 }
3810
3811 __be32
3812 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3813             struct nfsd4_locku *locku)
3814 {
3815         struct nfs4_stateid *stp;
3816         struct file *filp = NULL;
3817         struct file_lock file_lock;
3818         __be32 status;
3819         int err;
3820                                                         
3821         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
3822                 (long long) locku->lu_offset,
3823                 (long long) locku->lu_length);
3824
3825         if (check_lock_length(locku->lu_offset, locku->lu_length))
3826                  return nfserr_inval;
3827
3828         nfs4_lock_state();
3829                                                                                 
3830         if ((status = nfs4_preprocess_seqid_op(cstate,
3831                                         locku->lu_seqid, 
3832                                         &locku->lu_stateid, 
3833                                         LOCK_STATE,
3834                                         &locku->lu_stateowner, &stp, NULL)))
3835                 goto out;
3836
3837         filp = find_any_file(stp->st_file);
3838         if (!filp) {
3839                 status = nfserr_lock_range;
3840                 goto out;
3841         }
3842         BUG_ON(!filp);
3843         locks_init_lock(&file_lock);
3844         file_lock.fl_type = F_UNLCK;
3845         file_lock.fl_owner = (fl_owner_t) locku->lu_stateowner;
3846         file_lock.fl_pid = current->tgid;
3847         file_lock.fl_file = filp;
3848         file_lock.fl_flags = FL_POSIX; 
3849         file_lock.fl_lmops = &nfsd_posix_mng_ops;
3850         file_lock.fl_start = locku->lu_offset;
3851
3852         file_lock.fl_end = last_byte_offset(locku->lu_offset, locku->lu_length);
3853         nfs4_transform_lock_offset(&file_lock);
3854
3855         /*
3856         *  Try to unlock the file in the VFS.
3857         */
3858         err = vfs_lock_file(filp, F_SETLK, &file_lock, NULL);
3859         if (err) {
3860                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
3861                 goto out_nfserr;
3862         }
3863         /*
3864         * OK, unlock succeeded; the only thing left to do is update the stateid.
3865         */
3866         update_stateid(&stp->st_stateid);
3867         memcpy(&locku->lu_stateid, &stp->st_stateid, sizeof(stateid_t));
3868
3869 out:
3870         if (locku->lu_stateowner) {
3871                 nfs4_get_stateowner(locku->lu_stateowner);
3872                 cstate->replay_owner = locku->lu_stateowner;
3873         }
3874         nfs4_unlock_state();
3875         return status;
3876
3877 out_nfserr:
3878         status = nfserrno(err);
3879         goto out;
3880 }
3881
3882 /*
3883  * returns
3884  *      1: locks held by lockowner
3885  *      0: no locks held by lockowner
3886  */
3887 static int
3888 check_for_locks(struct nfs4_file *filp, struct nfs4_stateowner *lowner)
3889 {
3890         struct file_lock **flpp;
3891         struct inode *inode = filp->fi_inode;
3892         int status = 0;
3893
3894         lock_kernel();
3895         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
3896                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
3897                         status = 1;
3898                         goto out;
3899                 }
3900         }
3901 out:
3902         unlock_kernel();
3903         return status;
3904 }
3905
3906 __be32
3907 nfsd4_release_lockowner(struct svc_rqst *rqstp,
3908                         struct nfsd4_compound_state *cstate,
3909                         struct nfsd4_release_lockowner *rlockowner)
3910 {
3911         clientid_t *clid = &rlockowner->rl_clientid;
3912         struct nfs4_stateowner *sop;
3913         struct nfs4_stateid *stp;
3914         struct xdr_netobj *owner = &rlockowner->rl_owner;
3915         struct list_head matches;
3916         int i;
3917         __be32 status;
3918
3919         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
3920                 clid->cl_boot, clid->cl_id);
3921
3922         /* XXX check for lease expiration */
3923
3924         status = nfserr_stale_clientid;
3925         if (STALE_CLIENTID(clid))
3926                 return status;
3927
3928         nfs4_lock_state();
3929
3930         status = nfserr_locks_held;
3931         /* XXX: we're doing a linear search through all the lockowners.
3932          * Yipes!  For now we'll just hope clients aren't really using
3933          * release_lockowner much, but eventually we have to fix these
3934          * data structures. */
3935         INIT_LIST_HEAD(&matches);
3936         for (i = 0; i < LOCK_HASH_SIZE; i++) {
3937                 list_for_each_entry(sop, &lock_ownerid_hashtbl[i], so_idhash) {
3938                         if (!same_owner_str(sop, owner, clid))
3939                                 continue;
3940                         list_for_each_entry(stp, &sop->so_stateids,
3941                                         st_perstateowner) {
3942                                 if (check_for_locks(stp->st_file, sop))
3943                                         goto out;
3944                                 /* Note: so_perclient unused for lockowners,
3945                                  * so it's OK to fool with here. */
3946                                 list_add(&sop->so_perclient, &matches);
3947                         }
3948                 }
3949         }
3950         /* Clients probably won't expect us to return with some (but not all)
3951          * of the lockowner state released; so don't release any until all
3952          * have been checked. */
3953         status = nfs_ok;
3954         while (!list_empty(&matches)) {
3955                 sop = list_entry(matches.next, struct nfs4_stateowner,
3956                                                                 so_perclient);
3957                 /* unhash_stateowner deletes so_perclient only
3958                  * for openowners. */
3959                 list_del(&sop->so_perclient);
3960                 release_lockowner(sop);
3961         }
3962 out:
3963         nfs4_unlock_state();
3964         return status;
3965 }
3966
3967 static inline struct nfs4_client_reclaim *
3968 alloc_reclaim(void)
3969 {
3970         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
3971 }
3972
3973 int
3974 nfs4_has_reclaimed_state(const char *name, bool use_exchange_id)
3975 {
3976         unsigned int strhashval = clientstr_hashval(name);
3977         struct nfs4_client *clp;
3978
3979         clp = find_confirmed_client_by_str(name, strhashval, use_exchange_id);
3980         return clp ? 1 : 0;
3981 }
3982
3983 /*
3984  * failure => all reset bets are off, nfserr_no_grace...
3985  */
3986 int
3987 nfs4_client_to_reclaim(const char *name)
3988 {
3989         unsigned int strhashval;
3990         struct nfs4_client_reclaim *crp = NULL;
3991
3992         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
3993         crp = alloc_reclaim();
3994         if (!crp)
3995                 return 0;
3996         strhashval = clientstr_hashval(name);
3997         INIT_LIST_HEAD(&crp->cr_strhash);
3998         list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
3999         memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4000         reclaim_str_hashtbl_size++;
4001         return 1;
4002 }
4003
4004 static void
4005 nfs4_release_reclaim(void)
4006 {
4007         struct nfs4_client_reclaim *crp = NULL;
4008         int i;
4009
4010         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4011                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4012                         crp = list_entry(reclaim_str_hashtbl[i].next,
4013                                         struct nfs4_client_reclaim, cr_strhash);
4014                         list_del(&crp->cr_strhash);
4015                         kfree(crp);
4016                         reclaim_str_hashtbl_size--;
4017                 }
4018         }
4019         BUG_ON(reclaim_str_hashtbl_size);
4020 }
4021
4022 /*
4023  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4024 static struct nfs4_client_reclaim *
4025 nfs4_find_reclaim_client(clientid_t *clid)
4026 {
4027         unsigned int strhashval;
4028         struct nfs4_client *clp;
4029         struct nfs4_client_reclaim *crp = NULL;
4030
4031
4032         /* find clientid in conf_id_hashtbl */
4033         clp = find_confirmed_client(clid);
4034         if (clp == NULL)
4035                 return NULL;
4036
4037         dprintk("NFSD: nfs4_find_reclaim_client for %.*s with recdir %s\n",
4038                             clp->cl_name.len, clp->cl_name.data,
4039                             clp->cl_recdir);
4040
4041         /* find clp->cl_name in reclaim_str_hashtbl */
4042         strhashval = clientstr_hashval(clp->cl_recdir);
4043         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4044                 if (same_name(crp->cr_recdir, clp->cl_recdir)) {
4045                         return crp;
4046                 }
4047         }
4048         return NULL;
4049 }
4050
4051 /*
4052 * Called from OPEN. Look for clientid in reclaim list.
4053 */
4054 __be32
4055 nfs4_check_open_reclaim(clientid_t *clid)
4056 {
4057         return nfs4_find_reclaim_client(clid) ? nfs_ok : nfserr_reclaim_bad;
4058 }
4059
4060 /* initialization to perform at module load time: */
4061
4062 int
4063 nfs4_state_init(void)
4064 {
4065         int i, status;
4066
4067         status = nfsd4_init_slabs();
4068         if (status)
4069                 return status;
4070         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4071                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4072                 INIT_LIST_HEAD(&conf_str_hashtbl[i]);
4073                 INIT_LIST_HEAD(&unconf_str_hashtbl[i]);
4074                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4075                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4076         }
4077         for (i = 0; i < SESSION_HASH_SIZE; i++)
4078                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4079         for (i = 0; i < FILE_HASH_SIZE; i++) {
4080                 INIT_LIST_HEAD(&file_hashtbl[i]);
4081         }
4082         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4083                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4084                 INIT_LIST_HEAD(&ownerid_hashtbl[i]);
4085         }
4086         for (i = 0; i < STATEID_HASH_SIZE; i++) {
4087                 INIT_LIST_HEAD(&stateid_hashtbl[i]);
4088                 INIT_LIST_HEAD(&lockstateid_hashtbl[i]);
4089         }
4090         for (i = 0; i < LOCK_HASH_SIZE; i++) {
4091                 INIT_LIST_HEAD(&lock_ownerid_hashtbl[i]);
4092                 INIT_LIST_HEAD(&lock_ownerstr_hashtbl[i]);
4093         }
4094         memset(&onestateid, ~0, sizeof(stateid_t));
4095         INIT_LIST_HEAD(&close_lru);
4096         INIT_LIST_HEAD(&client_lru);
4097         INIT_LIST_HEAD(&del_recall_lru);
4098         reclaim_str_hashtbl_size = 0;
4099         return 0;
4100 }
4101
4102 static void
4103 nfsd4_load_reboot_recovery_data(void)
4104 {
4105         int status;
4106
4107         nfs4_lock_state();
4108         nfsd4_init_recdir(user_recovery_dirname);
4109         status = nfsd4_recdir_load();
4110         nfs4_unlock_state();
4111         if (status)
4112                 printk("NFSD: Failure reading reboot recovery data\n");
4113 }
4114
4115 /*
4116  * Since the lifetime of a delegation isn't limited to that of an open, a
4117  * client may quite reasonably hang on to a delegation as long as it has
4118  * the inode cached.  This becomes an obvious problem the first time a
4119  * client's inode cache approaches the size of the server's total memory.
4120  *
4121  * For now we avoid this problem by imposing a hard limit on the number
4122  * of delegations, which varies according to the server's memory size.
4123  */
4124 static void
4125 set_max_delegations(void)
4126 {
4127         /*
4128          * Allow at most 4 delegations per megabyte of RAM.  Quick
4129          * estimates suggest that in the worst case (where every delegation
4130          * is for a different inode), a delegation could take about 1.5K,
4131          * giving a worst case usage of about 6% of memory.
4132          */
4133         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4134 }
4135
4136 /* initialization to perform when the nfsd service is started: */
4137
4138 static int
4139 __nfs4_state_start(void)
4140 {
4141         int ret;
4142
4143         boot_time = get_seconds();
4144         locks_start_grace(&nfsd4_manager);
4145         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4146                nfsd4_grace);
4147         ret = set_callback_cred();
4148         if (ret)
4149                 return -ENOMEM;
4150         laundry_wq = create_singlethread_workqueue("nfsd4");
4151         if (laundry_wq == NULL)
4152                 return -ENOMEM;
4153         ret = nfsd4_create_callback_queue();
4154         if (ret)
4155                 goto out_free_laundry;
4156         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4157         set_max_delegations();
4158         return 0;
4159 out_free_laundry:
4160         destroy_workqueue(laundry_wq);
4161         return ret;
4162 }
4163
4164 int
4165 nfs4_state_start(void)
4166 {
4167         nfsd4_load_reboot_recovery_data();
4168         return __nfs4_state_start();
4169 }
4170
4171 static void
4172 __nfs4_state_shutdown(void)
4173 {
4174         int i;
4175         struct nfs4_client *clp = NULL;
4176         struct nfs4_delegation *dp = NULL;
4177         struct list_head *pos, *next, reaplist;
4178
4179         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4180                 while (!list_empty(&conf_id_hashtbl[i])) {
4181                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4182                         expire_client(clp);
4183                 }
4184                 while (!list_empty(&unconf_str_hashtbl[i])) {
4185                         clp = list_entry(unconf_str_hashtbl[i].next, struct nfs4_client, cl_strhash);
4186                         expire_client(clp);
4187                 }
4188         }
4189         INIT_LIST_HEAD(&reaplist);
4190         spin_lock(&recall_lock);
4191         list_for_each_safe(pos, next, &del_recall_lru) {
4192                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4193                 list_move(&dp->dl_recall_lru, &reaplist);
4194         }
4195         spin_unlock(&recall_lock);
4196         list_for_each_safe(pos, next, &reaplist) {
4197                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4198                 list_del_init(&dp->dl_recall_lru);
4199                 unhash_delegation(dp);
4200         }
4201
4202         nfsd4_shutdown_recdir();
4203 }
4204
4205 void
4206 nfs4_state_shutdown(void)
4207 {
4208         cancel_rearming_delayed_workqueue(laundry_wq, &laundromat_work);
4209         destroy_workqueue(laundry_wq);
4210         locks_end_grace(&nfsd4_manager);
4211         nfs4_lock_state();
4212         nfs4_release_reclaim();
4213         __nfs4_state_shutdown();
4214         nfsd4_destroy_callback_queue();
4215         nfs4_unlock_state();
4216 }
4217
4218 /*
4219  * user_recovery_dirname is protected by the nfsd_mutex since it's only
4220  * accessed when nfsd is starting.
4221  */
4222 static void
4223 nfs4_set_recdir(char *recdir)
4224 {
4225         strcpy(user_recovery_dirname, recdir);
4226 }
4227
4228 /*
4229  * Change the NFSv4 recovery directory to recdir.
4230  */
4231 int
4232 nfs4_reset_recoverydir(char *recdir)
4233 {
4234         int status;
4235         struct path path;
4236
4237         status = kern_path(recdir, LOOKUP_FOLLOW, &path);
4238         if (status)
4239                 return status;
4240         status = -ENOTDIR;
4241         if (S_ISDIR(path.dentry->d_inode->i_mode)) {
4242                 nfs4_set_recdir(recdir);
4243                 status = 0;
4244         }
4245         path_put(&path);
4246         return status;
4247 }
4248
4249 char *
4250 nfs4_recoverydir(void)
4251 {
4252         return user_recovery_dirname;
4253 }