nfsd: fix error values returned by nfsd4_lockt() when nfsd_open() fails
[pandora-kernel.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/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/sunrpc/svcauth_gss.h>
42 #include <linux/sunrpc/clnt.h>
43 #include "xdr4.h"
44 #include "vfs.h"
45
46 #define NFSDDBG_FACILITY                NFSDDBG_PROC
47
48 /* Globals */
49 time_t nfsd4_lease = 90;     /* default lease time */
50 time_t nfsd4_grace = 90;
51 static time_t boot_time;
52 static stateid_t zerostateid;             /* bits all 0 */
53 static stateid_t onestateid;              /* bits all 1 */
54 static u64 current_sessionid = 1;
55
56 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
57 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
58
59 /* forward declarations */
60 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
61
62 /* Locking: */
63
64 /* Currently used for almost all code touching nfsv4 state: */
65 static DEFINE_MUTEX(client_mutex);
66
67 /*
68  * Currently used for the del_recall_lru and file hash table.  In an
69  * effort to decrease the scope of the client_mutex, this spinlock may
70  * eventually cover more:
71  */
72 static DEFINE_SPINLOCK(recall_lock);
73
74 static struct kmem_cache *openowner_slab = NULL;
75 static struct kmem_cache *lockowner_slab = NULL;
76 static struct kmem_cache *file_slab = NULL;
77 static struct kmem_cache *stateid_slab = NULL;
78 static struct kmem_cache *deleg_slab = NULL;
79
80 void
81 nfs4_lock_state(void)
82 {
83         mutex_lock(&client_mutex);
84 }
85
86 void
87 nfs4_unlock_state(void)
88 {
89         mutex_unlock(&client_mutex);
90 }
91
92 static inline u32
93 opaque_hashval(const void *ptr, int nbytes)
94 {
95         unsigned char *cptr = (unsigned char *) ptr;
96
97         u32 x = 0;
98         while (nbytes--) {
99                 x *= 37;
100                 x += *cptr++;
101         }
102         return x;
103 }
104
105 static struct list_head del_recall_lru;
106
107 static void nfsd4_free_file(struct nfs4_file *f)
108 {
109         kmem_cache_free(file_slab, f);
110 }
111
112 static inline void
113 put_nfs4_file(struct nfs4_file *fi)
114 {
115         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
116                 list_del(&fi->fi_hash);
117                 spin_unlock(&recall_lock);
118                 iput(fi->fi_inode);
119                 nfsd4_free_file(fi);
120         }
121 }
122
123 static inline void
124 get_nfs4_file(struct nfs4_file *fi)
125 {
126         atomic_inc(&fi->fi_ref);
127 }
128
129 static int num_delegations;
130 unsigned int max_delegations;
131
132 /*
133  * Open owner state (share locks)
134  */
135
136 /* hash tables for open owners */
137 #define OPEN_OWNER_HASH_BITS              8
138 #define OPEN_OWNER_HASH_SIZE             (1 << OPEN_OWNER_HASH_BITS)
139 #define OPEN_OWNER_HASH_MASK             (OPEN_OWNER_HASH_SIZE - 1)
140
141 static unsigned int open_ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
142 {
143         unsigned int ret;
144
145         ret = opaque_hashval(ownername->data, ownername->len);
146         ret += clientid;
147         return ret & OPEN_OWNER_HASH_MASK;
148 }
149
150 static struct list_head open_ownerstr_hashtbl[OPEN_OWNER_HASH_SIZE];
151
152 /* hash table for nfs4_file */
153 #define FILE_HASH_BITS                   8
154 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
155
156 static unsigned int file_hashval(struct inode *ino)
157 {
158         /* XXX: why are we hashing on inode pointer, anyway? */
159         return hash_ptr(ino, FILE_HASH_BITS);
160 }
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163
164 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
165 {
166         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
167         atomic_inc(&fp->fi_access[oflag]);
168 }
169
170 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
171 {
172         if (oflag == O_RDWR) {
173                 __nfs4_file_get_access(fp, O_RDONLY);
174                 __nfs4_file_get_access(fp, O_WRONLY);
175         } else
176                 __nfs4_file_get_access(fp, oflag);
177 }
178
179 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
180 {
181         if (fp->fi_fds[oflag]) {
182                 fput(fp->fi_fds[oflag]);
183                 fp->fi_fds[oflag] = NULL;
184         }
185 }
186
187 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
188 {
189         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
190                 nfs4_file_put_fd(fp, oflag);
191                 /*
192                  * It's also safe to get rid of the RDWR open *if*
193                  * we no longer have need of the other kind of access
194                  * or if we already have the other kind of open:
195                  */
196                 if (fp->fi_fds[1-oflag]
197                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
198                         nfs4_file_put_fd(fp, O_RDWR);
199         }
200 }
201
202 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
203 {
204         if (oflag == O_RDWR) {
205                 __nfs4_file_put_access(fp, O_RDONLY);
206                 __nfs4_file_put_access(fp, O_WRONLY);
207         } else
208                 __nfs4_file_put_access(fp, oflag);
209 }
210
211 static inline int get_new_stid(struct nfs4_stid *stid)
212 {
213         static int min_stateid = 0;
214         struct idr *stateids = &stid->sc_client->cl_stateids;
215         int new_stid;
216         int error;
217
218         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
219         /*
220          * Note: the necessary preallocation was done in
221          * nfs4_alloc_stateid().  The idr code caps the number of
222          * preallocations that can exist at a time, but the state lock
223          * prevents anyone from using ours before we get here:
224          */
225         BUG_ON(error);
226         /*
227          * It shouldn't be a problem to reuse an opaque stateid value.
228          * I don't think it is for 4.1.  But with 4.0 I worry that, for
229          * example, a stray write retransmission could be accepted by
230          * the server when it should have been rejected.  Therefore,
231          * adopt a trick from the sctp code to attempt to maximize the
232          * amount of time until an id is reused, by ensuring they always
233          * "increase" (mod INT_MAX):
234          */
235
236         min_stateid = new_stid+1;
237         if (min_stateid == INT_MAX)
238                 min_stateid = 0;
239         return new_stid;
240 }
241
242 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
243 {
244         stateid_t *s = &stid->sc_stateid;
245         int new_id;
246
247         stid->sc_type = type;
248         stid->sc_client = cl;
249         s->si_opaque.so_clid = cl->cl_clientid;
250         new_id = get_new_stid(stid);
251         s->si_opaque.so_id = (u32)new_id;
252         /* Will be incremented before return to client: */
253         s->si_generation = 0;
254 }
255
256 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
257 {
258         struct idr *stateids = &cl->cl_stateids;
259
260         if (!idr_pre_get(stateids, GFP_KERNEL))
261                 return NULL;
262         /*
263          * Note: if we fail here (or any time between now and the time
264          * we actually get the new idr), we won't need to undo the idr
265          * preallocation, since the idr code caps the number of
266          * preallocated entries.
267          */
268         return kmem_cache_alloc(slab, GFP_KERNEL);
269 }
270
271 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
272 {
273         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
274 }
275
276 static struct nfs4_delegation *
277 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
278 {
279         struct nfs4_delegation *dp;
280         struct nfs4_file *fp = stp->st_file;
281
282         dprintk("NFSD alloc_init_deleg\n");
283         /*
284          * Major work on the lease subsystem (for example, to support
285          * calbacks on stat) will be required before we can support
286          * write delegations properly.
287          */
288         if (type != NFS4_OPEN_DELEGATE_READ)
289                 return NULL;
290         if (fp->fi_had_conflict)
291                 return NULL;
292         if (num_delegations > max_delegations)
293                 return NULL;
294         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
295         if (dp == NULL)
296                 return dp;
297         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
298         /*
299          * delegation seqid's are never incremented.  The 4.1 special
300          * meaning of seqid 0 isn't meaningful, really, but let's avoid
301          * 0 anyway just for consistency and use 1:
302          */
303         dp->dl_stid.sc_stateid.si_generation = 1;
304         num_delegations++;
305         INIT_LIST_HEAD(&dp->dl_perfile);
306         INIT_LIST_HEAD(&dp->dl_perclnt);
307         INIT_LIST_HEAD(&dp->dl_recall_lru);
308         get_nfs4_file(fp);
309         dp->dl_file = fp;
310         dp->dl_type = type;
311         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
312         dp->dl_time = 0;
313         atomic_set(&dp->dl_count, 1);
314         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
315         return dp;
316 }
317
318 void
319 nfs4_put_delegation(struct nfs4_delegation *dp)
320 {
321         if (atomic_dec_and_test(&dp->dl_count)) {
322                 dprintk("NFSD: freeing dp %p\n",dp);
323                 put_nfs4_file(dp->dl_file);
324                 kmem_cache_free(deleg_slab, dp);
325                 num_delegations--;
326         }
327 }
328
329 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
330 {
331         if (atomic_dec_and_test(&fp->fi_delegees)) {
332                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
333                 fp->fi_lease = NULL;
334                 fput(fp->fi_deleg_file);
335                 fp->fi_deleg_file = NULL;
336         }
337 }
338
339 static void unhash_stid(struct nfs4_stid *s)
340 {
341         struct idr *stateids = &s->sc_client->cl_stateids;
342
343         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
344 }
345
346 /* Called under the state lock. */
347 static void
348 unhash_delegation(struct nfs4_delegation *dp)
349 {
350         unhash_stid(&dp->dl_stid);
351         list_del_init(&dp->dl_perclnt);
352         spin_lock(&recall_lock);
353         list_del_init(&dp->dl_perfile);
354         list_del_init(&dp->dl_recall_lru);
355         spin_unlock(&recall_lock);
356         nfs4_put_deleg_lease(dp->dl_file);
357         nfs4_put_delegation(dp);
358 }
359
360 /* 
361  * SETCLIENTID state 
362  */
363
364 /* client_lock protects the client lru list and session hash table */
365 static DEFINE_SPINLOCK(client_lock);
366
367 /* Hash tables for nfs4_clientid state */
368 #define CLIENT_HASH_BITS                 4
369 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
370 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
371
372 static unsigned int clientid_hashval(u32 id)
373 {
374         return id & CLIENT_HASH_MASK;
375 }
376
377 static unsigned int clientstr_hashval(const char *name)
378 {
379         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
380 }
381
382 /*
383  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
384  * used in reboot/reset lease grace period processing
385  *
386  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
387  * setclientid_confirmed info. 
388  *
389  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
390  * setclientid info.
391  *
392  * client_lru holds client queue ordered by nfs4_client.cl_time
393  * for lease renewal.
394  *
395  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
396  * for last close replay.
397  */
398 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
399 static int reclaim_str_hashtbl_size = 0;
400 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
401 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
402 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
403 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
404 static struct list_head client_lru;
405 static struct list_head close_lru;
406
407 /*
408  * We store the NONE, READ, WRITE, and BOTH bits separately in the
409  * st_{access,deny}_bmap field of the stateid, in order to track not
410  * only what share bits are currently in force, but also what
411  * combinations of share bits previous opens have used.  This allows us
412  * to enforce the recommendation of rfc 3530 14.2.19 that the server
413  * return an error if the client attempt to downgrade to a combination
414  * of share bits not explicable by closing some of its previous opens.
415  *
416  * XXX: This enforcement is actually incomplete, since we don't keep
417  * track of access/deny bit combinations; so, e.g., we allow:
418  *
419  *      OPEN allow read, deny write
420  *      OPEN allow both, deny none
421  *      DOWNGRADE allow read, deny none
422  *
423  * which we should reject.
424  */
425 static void
426 set_access(unsigned int *access, unsigned long bmap) {
427         int i;
428
429         *access = 0;
430         for (i = 1; i < 4; i++) {
431                 if (test_bit(i, &bmap))
432                         *access |= i;
433         }
434 }
435
436 static void
437 set_deny(unsigned int *deny, unsigned long bmap) {
438         int i;
439
440         *deny = 0;
441         for (i = 0; i < 4; i++) {
442                 if (test_bit(i, &bmap))
443                         *deny |= i ;
444         }
445 }
446
447 static int
448 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
449         unsigned int access, deny;
450
451         set_access(&access, stp->st_access_bmap);
452         set_deny(&deny, stp->st_deny_bmap);
453         if ((access & open->op_share_deny) || (deny & open->op_share_access))
454                 return 0;
455         return 1;
456 }
457
458 static int nfs4_access_to_omode(u32 access)
459 {
460         switch (access & NFS4_SHARE_ACCESS_BOTH) {
461         case NFS4_SHARE_ACCESS_READ:
462                 return O_RDONLY;
463         case NFS4_SHARE_ACCESS_WRITE:
464                 return O_WRONLY;
465         case NFS4_SHARE_ACCESS_BOTH:
466                 return O_RDWR;
467         }
468         BUG();
469 }
470
471 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
472 {
473         list_del(&stp->st_perfile);
474         list_del(&stp->st_perstateowner);
475 }
476
477 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
478 {
479         int i;
480
481         if (stp->st_access_bmap) {
482                 for (i = 1; i < 4; i++) {
483                         if (test_bit(i, &stp->st_access_bmap))
484                                 nfs4_file_put_access(stp->st_file,
485                                                 nfs4_access_to_omode(i));
486                         __clear_bit(i, &stp->st_access_bmap);
487                 }
488         }
489         put_nfs4_file(stp->st_file);
490         stp->st_file = NULL;
491 }
492
493 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
494 {
495         kmem_cache_free(stateid_slab, stp);
496 }
497
498 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
499 {
500         struct file *file;
501
502         unhash_generic_stateid(stp);
503         unhash_stid(&stp->st_stid);
504         file = find_any_file(stp->st_file);
505         if (file)
506                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
507         close_generic_stateid(stp);
508         free_generic_stateid(stp);
509 }
510
511 static void unhash_lockowner(struct nfs4_lockowner *lo)
512 {
513         struct nfs4_ol_stateid *stp;
514
515         list_del(&lo->lo_owner.so_strhash);
516         list_del(&lo->lo_perstateid);
517         while (!list_empty(&lo->lo_owner.so_stateids)) {
518                 stp = list_first_entry(&lo->lo_owner.so_stateids,
519                                 struct nfs4_ol_stateid, st_perstateowner);
520                 release_lock_stateid(stp);
521         }
522 }
523
524 static void release_lockowner(struct nfs4_lockowner *lo)
525 {
526         unhash_lockowner(lo);
527         nfs4_free_lockowner(lo);
528 }
529
530 static void
531 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
532 {
533         struct nfs4_lockowner *lo;
534
535         while (!list_empty(&open_stp->st_lockowners)) {
536                 lo = list_entry(open_stp->st_lockowners.next,
537                                 struct nfs4_lockowner, lo_perstateid);
538                 release_lockowner(lo);
539         }
540 }
541
542 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
543 {
544         unhash_generic_stateid(stp);
545         release_stateid_lockowners(stp);
546         close_generic_stateid(stp);
547 }
548
549 static void release_open_stateid(struct nfs4_ol_stateid *stp)
550 {
551         unhash_open_stateid(stp);
552         unhash_stid(&stp->st_stid);
553         free_generic_stateid(stp);
554 }
555
556 static void unhash_openowner(struct nfs4_openowner *oo)
557 {
558         struct nfs4_ol_stateid *stp;
559
560         list_del(&oo->oo_owner.so_strhash);
561         list_del(&oo->oo_perclient);
562         while (!list_empty(&oo->oo_owner.so_stateids)) {
563                 stp = list_first_entry(&oo->oo_owner.so_stateids,
564                                 struct nfs4_ol_stateid, st_perstateowner);
565                 release_open_stateid(stp);
566         }
567 }
568
569 static void release_last_closed_stateid(struct nfs4_openowner *oo)
570 {
571         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
572
573         if (s) {
574                 unhash_stid(&s->st_stid);
575                 free_generic_stateid(s);
576                 oo->oo_last_closed_stid = NULL;
577         }
578 }
579
580 static void release_openowner(struct nfs4_openowner *oo)
581 {
582         unhash_openowner(oo);
583         list_del(&oo->oo_close_lru);
584         release_last_closed_stateid(oo);
585         nfs4_free_openowner(oo);
586 }
587
588 #define SESSION_HASH_SIZE       512
589 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
590
591 static inline int
592 hash_sessionid(struct nfs4_sessionid *sessionid)
593 {
594         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
595
596         return sid->sequence % SESSION_HASH_SIZE;
597 }
598
599 static inline void
600 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
601 {
602         u32 *ptr = (u32 *)(&sessionid->data[0]);
603         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
604 }
605
606 static void
607 gen_sessionid(struct nfsd4_session *ses)
608 {
609         struct nfs4_client *clp = ses->se_client;
610         struct nfsd4_sessionid *sid;
611
612         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
613         sid->clientid = clp->cl_clientid;
614         sid->sequence = current_sessionid++;
615         sid->reserved = 0;
616 }
617
618 /*
619  * The protocol defines ca_maxresponssize_cached to include the size of
620  * the rpc header, but all we need to cache is the data starting after
621  * the end of the initial SEQUENCE operation--the rest we regenerate
622  * each time.  Therefore we can advertise a ca_maxresponssize_cached
623  * value that is the number of bytes in our cache plus a few additional
624  * bytes.  In order to stay on the safe side, and not promise more than
625  * we can cache, those additional bytes must be the minimum possible: 24
626  * bytes of rpc header (xid through accept state, with AUTH_NULL
627  * verifier), 12 for the compound header (with zero-length tag), and 44
628  * for the SEQUENCE op response:
629  */
630 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
631
632 static void
633 free_session_slots(struct nfsd4_session *ses)
634 {
635         int i;
636
637         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
638                 kfree(ses->se_slots[i]);
639 }
640
641 /*
642  * We don't actually need to cache the rpc and session headers, so we
643  * can allocate a little less for each slot:
644  */
645 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
646 {
647         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
648 }
649
650 static int nfsd4_sanitize_slot_size(u32 size)
651 {
652         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
653         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
654
655         return size;
656 }
657
658 /*
659  * XXX: If we run out of reserved DRC memory we could (up to a point)
660  * re-negotiate active sessions and reduce their slot usage to make
661  * rooom for new connections. For now we just fail the create session.
662  */
663 static int nfsd4_get_drc_mem(int slotsize, u32 num)
664 {
665         int avail;
666
667         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
668
669         spin_lock(&nfsd_drc_lock);
670         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
671                         nfsd_drc_max_mem - nfsd_drc_mem_used);
672         num = min_t(int, num, avail / slotsize);
673         nfsd_drc_mem_used += num * slotsize;
674         spin_unlock(&nfsd_drc_lock);
675
676         return num;
677 }
678
679 static void nfsd4_put_drc_mem(int slotsize, int num)
680 {
681         spin_lock(&nfsd_drc_lock);
682         nfsd_drc_mem_used -= slotsize * num;
683         spin_unlock(&nfsd_drc_lock);
684 }
685
686 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
687 {
688         struct nfsd4_session *new;
689         int mem, i;
690
691         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
692                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
693         mem = numslots * sizeof(struct nfsd4_slot *);
694
695         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
696         if (!new)
697                 return NULL;
698         /* allocate each struct nfsd4_slot and data cache in one piece */
699         for (i = 0; i < numslots; i++) {
700                 mem = sizeof(struct nfsd4_slot) + slotsize;
701                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
702                 if (!new->se_slots[i])
703                         goto out_free;
704         }
705         return new;
706 out_free:
707         while (i--)
708                 kfree(new->se_slots[i]);
709         kfree(new);
710         return NULL;
711 }
712
713 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
714 {
715         u32 maxrpc = nfsd_serv->sv_max_mesg;
716
717         new->maxreqs = numslots;
718         new->maxresp_cached = min_t(u32, req->maxresp_cached,
719                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
720         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
721         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
722         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
723 }
724
725 static void free_conn(struct nfsd4_conn *c)
726 {
727         svc_xprt_put(c->cn_xprt);
728         kfree(c);
729 }
730
731 static void nfsd4_conn_lost(struct svc_xpt_user *u)
732 {
733         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
734         struct nfs4_client *clp = c->cn_session->se_client;
735
736         spin_lock(&clp->cl_lock);
737         if (!list_empty(&c->cn_persession)) {
738                 list_del(&c->cn_persession);
739                 free_conn(c);
740         }
741         spin_unlock(&clp->cl_lock);
742         nfsd4_probe_callback(clp);
743 }
744
745 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
746 {
747         struct nfsd4_conn *conn;
748
749         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
750         if (!conn)
751                 return NULL;
752         svc_xprt_get(rqstp->rq_xprt);
753         conn->cn_xprt = rqstp->rq_xprt;
754         conn->cn_flags = flags;
755         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
756         return conn;
757 }
758
759 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
760 {
761         conn->cn_session = ses;
762         list_add(&conn->cn_persession, &ses->se_conns);
763 }
764
765 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
766 {
767         struct nfs4_client *clp = ses->se_client;
768
769         spin_lock(&clp->cl_lock);
770         __nfsd4_hash_conn(conn, ses);
771         spin_unlock(&clp->cl_lock);
772 }
773
774 static int nfsd4_register_conn(struct nfsd4_conn *conn)
775 {
776         conn->cn_xpt_user.callback = nfsd4_conn_lost;
777         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
778 }
779
780 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
781 {
782         struct nfsd4_conn *conn;
783         int ret;
784
785         conn = alloc_conn(rqstp, dir);
786         if (!conn)
787                 return nfserr_jukebox;
788         nfsd4_hash_conn(conn, ses);
789         ret = nfsd4_register_conn(conn);
790         if (ret)
791                 /* oops; xprt is already down: */
792                 nfsd4_conn_lost(&conn->cn_xpt_user);
793         return nfs_ok;
794 }
795
796 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
797 {
798         u32 dir = NFS4_CDFC4_FORE;
799
800         if (ses->se_flags & SESSION4_BACK_CHAN)
801                 dir |= NFS4_CDFC4_BACK;
802
803         return nfsd4_new_conn(rqstp, ses, dir);
804 }
805
806 /* must be called under client_lock */
807 static void nfsd4_del_conns(struct nfsd4_session *s)
808 {
809         struct nfs4_client *clp = s->se_client;
810         struct nfsd4_conn *c;
811
812         spin_lock(&clp->cl_lock);
813         while (!list_empty(&s->se_conns)) {
814                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
815                 list_del_init(&c->cn_persession);
816                 spin_unlock(&clp->cl_lock);
817
818                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
819                 free_conn(c);
820
821                 spin_lock(&clp->cl_lock);
822         }
823         spin_unlock(&clp->cl_lock);
824 }
825
826 void free_session(struct kref *kref)
827 {
828         struct nfsd4_session *ses;
829         int mem;
830
831         ses = container_of(kref, struct nfsd4_session, se_ref);
832         nfsd4_del_conns(ses);
833         spin_lock(&nfsd_drc_lock);
834         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
835         nfsd_drc_mem_used -= mem;
836         spin_unlock(&nfsd_drc_lock);
837         free_session_slots(ses);
838         kfree(ses);
839 }
840
841 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
842 {
843         struct nfsd4_session *new;
844         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
845         int numslots, slotsize;
846         int status;
847         int idx;
848
849         /*
850          * Note decreasing slot size below client's request may
851          * make it difficult for client to function correctly, whereas
852          * decreasing the number of slots will (just?) affect
853          * performance.  When short on memory we therefore prefer to
854          * decrease number of slots instead of their size.
855          */
856         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
857         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
858         if (numslots < 1)
859                 return NULL;
860
861         new = alloc_session(slotsize, numslots);
862         if (!new) {
863                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
864                 return NULL;
865         }
866         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
867
868         new->se_client = clp;
869         gen_sessionid(new);
870
871         INIT_LIST_HEAD(&new->se_conns);
872
873         new->se_cb_seq_nr = 1;
874         new->se_flags = cses->flags;
875         new->se_cb_prog = cses->callback_prog;
876         kref_init(&new->se_ref);
877         idx = hash_sessionid(&new->se_sessionid);
878         spin_lock(&client_lock);
879         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
880         spin_lock(&clp->cl_lock);
881         list_add(&new->se_perclnt, &clp->cl_sessions);
882         spin_unlock(&clp->cl_lock);
883         spin_unlock(&client_lock);
884
885         status = nfsd4_new_conn_from_crses(rqstp, new);
886         /* whoops: benny points out, status is ignored! (err, or bogus) */
887         if (status) {
888                 free_session(&new->se_ref);
889                 return NULL;
890         }
891         if (cses->flags & SESSION4_BACK_CHAN) {
892                 struct sockaddr *sa = svc_addr(rqstp);
893                 /*
894                  * This is a little silly; with sessions there's no real
895                  * use for the callback address.  Use the peer address
896                  * as a reasonable default for now, but consider fixing
897                  * the rpc client not to require an address in the
898                  * future:
899                  */
900                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
901                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
902         }
903         nfsd4_probe_callback(clp);
904         return new;
905 }
906
907 /* caller must hold client_lock */
908 static struct nfsd4_session *
909 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
910 {
911         struct nfsd4_session *elem;
912         int idx;
913
914         dump_sessionid(__func__, sessionid);
915         idx = hash_sessionid(sessionid);
916         /* Search in the appropriate list */
917         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
918                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
919                             NFS4_MAX_SESSIONID_LEN)) {
920                         return elem;
921                 }
922         }
923
924         dprintk("%s: session not found\n", __func__);
925         return NULL;
926 }
927
928 /* caller must hold client_lock */
929 static void
930 unhash_session(struct nfsd4_session *ses)
931 {
932         list_del(&ses->se_hash);
933         spin_lock(&ses->se_client->cl_lock);
934         list_del(&ses->se_perclnt);
935         spin_unlock(&ses->se_client->cl_lock);
936 }
937
938 /* must be called under the client_lock */
939 static inline void
940 renew_client_locked(struct nfs4_client *clp)
941 {
942         if (is_client_expired(clp)) {
943                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
944                         __func__,
945                         clp->cl_clientid.cl_boot,
946                         clp->cl_clientid.cl_id);
947                 return;
948         }
949
950         dprintk("renewing client (clientid %08x/%08x)\n", 
951                         clp->cl_clientid.cl_boot, 
952                         clp->cl_clientid.cl_id);
953         list_move_tail(&clp->cl_lru, &client_lru);
954         clp->cl_time = get_seconds();
955 }
956
957 static inline void
958 renew_client(struct nfs4_client *clp)
959 {
960         spin_lock(&client_lock);
961         renew_client_locked(clp);
962         spin_unlock(&client_lock);
963 }
964
965 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
966 static int
967 STALE_CLIENTID(clientid_t *clid)
968 {
969         if (clid->cl_boot == boot_time)
970                 return 0;
971         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
972                 clid->cl_boot, clid->cl_id, boot_time);
973         return 1;
974 }
975
976 /* 
977  * XXX Should we use a slab cache ?
978  * This type of memory management is somewhat inefficient, but we use it
979  * anyway since SETCLIENTID is not a common operation.
980  */
981 static struct nfs4_client *alloc_client(struct xdr_netobj name)
982 {
983         struct nfs4_client *clp;
984
985         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
986         if (clp == NULL)
987                 return NULL;
988         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
989         if (clp->cl_name.data == NULL) {
990                 kfree(clp);
991                 return NULL;
992         }
993         memcpy(clp->cl_name.data, name.data, name.len);
994         clp->cl_name.len = name.len;
995         return clp;
996 }
997
998 static inline void
999 free_client(struct nfs4_client *clp)
1000 {
1001         while (!list_empty(&clp->cl_sessions)) {
1002                 struct nfsd4_session *ses;
1003                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1004                                 se_perclnt);
1005                 list_del(&ses->se_perclnt);
1006                 nfsd4_put_session(ses);
1007         }
1008         if (clp->cl_cred.cr_group_info)
1009                 put_group_info(clp->cl_cred.cr_group_info);
1010         kfree(clp->cl_principal);
1011         kfree(clp->cl_name.data);
1012         kfree(clp);
1013 }
1014
1015 void
1016 release_session_client(struct nfsd4_session *session)
1017 {
1018         struct nfs4_client *clp = session->se_client;
1019
1020         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1021                 return;
1022         if (is_client_expired(clp)) {
1023                 free_client(clp);
1024                 session->se_client = NULL;
1025         } else
1026                 renew_client_locked(clp);
1027         spin_unlock(&client_lock);
1028 }
1029
1030 /* must be called under the client_lock */
1031 static inline void
1032 unhash_client_locked(struct nfs4_client *clp)
1033 {
1034         struct nfsd4_session *ses;
1035
1036         mark_client_expired(clp);
1037         list_del(&clp->cl_lru);
1038         spin_lock(&clp->cl_lock);
1039         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1040                 list_del_init(&ses->se_hash);
1041         spin_unlock(&clp->cl_lock);
1042 }
1043
1044 static void
1045 expire_client(struct nfs4_client *clp)
1046 {
1047         struct nfs4_openowner *oo;
1048         struct nfs4_delegation *dp;
1049         struct list_head reaplist;
1050
1051         INIT_LIST_HEAD(&reaplist);
1052         spin_lock(&recall_lock);
1053         while (!list_empty(&clp->cl_delegations)) {
1054                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1055                 list_del_init(&dp->dl_perclnt);
1056                 list_move(&dp->dl_recall_lru, &reaplist);
1057         }
1058         spin_unlock(&recall_lock);
1059         while (!list_empty(&reaplist)) {
1060                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1061                 list_del_init(&dp->dl_recall_lru);
1062                 unhash_delegation(dp);
1063         }
1064         while (!list_empty(&clp->cl_openowners)) {
1065                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1066                 release_openowner(oo);
1067         }
1068         nfsd4_shutdown_callback(clp);
1069         if (clp->cl_cb_conn.cb_xprt)
1070                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1071         list_del(&clp->cl_idhash);
1072         list_del(&clp->cl_strhash);
1073         spin_lock(&client_lock);
1074         unhash_client_locked(clp);
1075         if (atomic_read(&clp->cl_refcount) == 0)
1076                 free_client(clp);
1077         spin_unlock(&client_lock);
1078 }
1079
1080 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1081 {
1082         memcpy(target->cl_verifier.data, source->data,
1083                         sizeof(target->cl_verifier.data));
1084 }
1085
1086 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1087 {
1088         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1089         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1090 }
1091
1092 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
1093 {
1094         target->cr_uid = source->cr_uid;
1095         target->cr_gid = source->cr_gid;
1096         target->cr_group_info = source->cr_group_info;
1097         get_group_info(target->cr_group_info);
1098 }
1099
1100 static int same_name(const char *n1, const char *n2)
1101 {
1102         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1103 }
1104
1105 static int
1106 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1107 {
1108         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1109 }
1110
1111 static int
1112 same_clid(clientid_t *cl1, clientid_t *cl2)
1113 {
1114         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1115 }
1116
1117 /* XXX what about NGROUP */
1118 static int
1119 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1120 {
1121         return cr1->cr_uid == cr2->cr_uid;
1122 }
1123
1124 static void gen_clid(struct nfs4_client *clp)
1125 {
1126         static u32 current_clientid = 1;
1127
1128         clp->cl_clientid.cl_boot = boot_time;
1129         clp->cl_clientid.cl_id = current_clientid++; 
1130 }
1131
1132 static void gen_confirm(struct nfs4_client *clp)
1133 {
1134         static u32 i;
1135         u32 *p;
1136
1137         p = (u32 *)clp->cl_confirm.data;
1138         *p++ = get_seconds();
1139         *p++ = i++;
1140 }
1141
1142 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1143 {
1144         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1145 }
1146
1147 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1148 {
1149         struct nfs4_stid *s;
1150
1151         s = find_stateid(cl, t);
1152         if (!s)
1153                 return NULL;
1154         if (typemask & s->sc_type)
1155                 return s;
1156         return NULL;
1157 }
1158
1159 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1160                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1161 {
1162         struct nfs4_client *clp;
1163         struct sockaddr *sa = svc_addr(rqstp);
1164         char *princ;
1165
1166         clp = alloc_client(name);
1167         if (clp == NULL)
1168                 return NULL;
1169
1170         INIT_LIST_HEAD(&clp->cl_sessions);
1171
1172         princ = svc_gss_principal(rqstp);
1173         if (princ) {
1174                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1175                 if (clp->cl_principal == NULL) {
1176                         free_client(clp);
1177                         return NULL;
1178                 }
1179         }
1180
1181         idr_init(&clp->cl_stateids);
1182         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1183         atomic_set(&clp->cl_refcount, 0);
1184         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1185         INIT_LIST_HEAD(&clp->cl_idhash);
1186         INIT_LIST_HEAD(&clp->cl_strhash);
1187         INIT_LIST_HEAD(&clp->cl_openowners);
1188         INIT_LIST_HEAD(&clp->cl_delegations);
1189         INIT_LIST_HEAD(&clp->cl_lru);
1190         INIT_LIST_HEAD(&clp->cl_callbacks);
1191         spin_lock_init(&clp->cl_lock);
1192         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1193         clp->cl_time = get_seconds();
1194         clear_bit(0, &clp->cl_cb_slot_busy);
1195         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1196         copy_verf(clp, verf);
1197         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1198         clp->cl_flavor = rqstp->rq_flavor;
1199         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1200         gen_confirm(clp);
1201         clp->cl_cb_session = NULL;
1202         return clp;
1203 }
1204
1205 static void
1206 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1207 {
1208         unsigned int idhashval;
1209
1210         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1211         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1212         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1213         renew_client(clp);
1214 }
1215
1216 static void
1217 move_to_confirmed(struct nfs4_client *clp)
1218 {
1219         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1220         unsigned int strhashval;
1221
1222         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1223         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1224         strhashval = clientstr_hashval(clp->cl_recdir);
1225         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1226         renew_client(clp);
1227 }
1228
1229 static struct nfs4_client *
1230 find_confirmed_client(clientid_t *clid)
1231 {
1232         struct nfs4_client *clp;
1233         unsigned int idhashval = clientid_hashval(clid->cl_id);
1234
1235         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1236                 if (same_clid(&clp->cl_clientid, clid)) {
1237                         renew_client(clp);
1238                         return clp;
1239                 }
1240         }
1241         return NULL;
1242 }
1243
1244 static struct nfs4_client *
1245 find_unconfirmed_client(clientid_t *clid)
1246 {
1247         struct nfs4_client *clp;
1248         unsigned int idhashval = clientid_hashval(clid->cl_id);
1249
1250         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1251                 if (same_clid(&clp->cl_clientid, clid))
1252                         return clp;
1253         }
1254         return NULL;
1255 }
1256
1257 static bool clp_used_exchangeid(struct nfs4_client *clp)
1258 {
1259         return clp->cl_exchange_flags != 0;
1260
1261
1262 static struct nfs4_client *
1263 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1264 {
1265         struct nfs4_client *clp;
1266
1267         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1268                 if (same_name(clp->cl_recdir, dname))
1269                         return clp;
1270         }
1271         return NULL;
1272 }
1273
1274 static struct nfs4_client *
1275 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1276 {
1277         struct nfs4_client *clp;
1278
1279         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1280                 if (same_name(clp->cl_recdir, dname))
1281                         return clp;
1282         }
1283         return NULL;
1284 }
1285
1286 static void
1287 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1288 {
1289         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1290         struct sockaddr *sa = svc_addr(rqstp);
1291         u32 scopeid = rpc_get_scope_id(sa);
1292         unsigned short expected_family;
1293
1294         /* Currently, we only support tcp and tcp6 for the callback channel */
1295         if (se->se_callback_netid_len == 3 &&
1296             !memcmp(se->se_callback_netid_val, "tcp", 3))
1297                 expected_family = AF_INET;
1298         else if (se->se_callback_netid_len == 4 &&
1299                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1300                 expected_family = AF_INET6;
1301         else
1302                 goto out_err;
1303
1304         conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1305                                             se->se_callback_addr_len,
1306                                             (struct sockaddr *)&conn->cb_addr,
1307                                             sizeof(conn->cb_addr));
1308
1309         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1310                 goto out_err;
1311
1312         if (conn->cb_addr.ss_family == AF_INET6)
1313                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1314
1315         conn->cb_prog = se->se_callback_prog;
1316         conn->cb_ident = se->se_callback_ident;
1317         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1318         return;
1319 out_err:
1320         conn->cb_addr.ss_family = AF_UNSPEC;
1321         conn->cb_addrlen = 0;
1322         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1323                 "will not receive delegations\n",
1324                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1325
1326         return;
1327 }
1328
1329 /*
1330  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1331  */
1332 void
1333 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1334 {
1335         struct nfsd4_slot *slot = resp->cstate.slot;
1336         unsigned int base;
1337
1338         dprintk("--> %s slot %p\n", __func__, slot);
1339
1340         slot->sl_opcnt = resp->opcnt;
1341         slot->sl_status = resp->cstate.status;
1342
1343         if (nfsd4_not_cached(resp)) {
1344                 slot->sl_datalen = 0;
1345                 return;
1346         }
1347         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1348         base = (char *)resp->cstate.datap -
1349                                         (char *)resp->xbuf->head[0].iov_base;
1350         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1351                                     slot->sl_datalen))
1352                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1353         return;
1354 }
1355
1356 /*
1357  * Encode the replay sequence operation from the slot values.
1358  * If cachethis is FALSE encode the uncached rep error on the next
1359  * operation which sets resp->p and increments resp->opcnt for
1360  * nfs4svc_encode_compoundres.
1361  *
1362  */
1363 static __be32
1364 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1365                           struct nfsd4_compoundres *resp)
1366 {
1367         struct nfsd4_op *op;
1368         struct nfsd4_slot *slot = resp->cstate.slot;
1369
1370         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1371                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1372
1373         /* Encode the replayed sequence operation */
1374         op = &args->ops[resp->opcnt - 1];
1375         nfsd4_encode_operation(resp, op);
1376
1377         /* Return nfserr_retry_uncached_rep in next operation. */
1378         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1379                 op = &args->ops[resp->opcnt++];
1380                 op->status = nfserr_retry_uncached_rep;
1381                 nfsd4_encode_operation(resp, op);
1382         }
1383         return op->status;
1384 }
1385
1386 /*
1387  * The sequence operation is not cached because we can use the slot and
1388  * session values.
1389  */
1390 __be32
1391 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1392                          struct nfsd4_sequence *seq)
1393 {
1394         struct nfsd4_slot *slot = resp->cstate.slot;
1395         __be32 status;
1396
1397         dprintk("--> %s slot %p\n", __func__, slot);
1398
1399         /* Either returns 0 or nfserr_retry_uncached */
1400         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1401         if (status == nfserr_retry_uncached_rep)
1402                 return status;
1403
1404         /* The sequence operation has been encoded, cstate->datap set. */
1405         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1406
1407         resp->opcnt = slot->sl_opcnt;
1408         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1409         status = slot->sl_status;
1410
1411         return status;
1412 }
1413
1414 /*
1415  * Set the exchange_id flags returned by the server.
1416  */
1417 static void
1418 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1419 {
1420         /* pNFS is not supported */
1421         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1422
1423         /* Referrals are supported, Migration is not. */
1424         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1425
1426         /* set the wire flags to return to client. */
1427         clid->flags = new->cl_exchange_flags;
1428 }
1429
1430 __be32
1431 nfsd4_exchange_id(struct svc_rqst *rqstp,
1432                   struct nfsd4_compound_state *cstate,
1433                   struct nfsd4_exchange_id *exid)
1434 {
1435         struct nfs4_client *unconf, *conf, *new;
1436         int status;
1437         unsigned int            strhashval;
1438         char                    dname[HEXDIR_LEN];
1439         char                    addr_str[INET6_ADDRSTRLEN];
1440         nfs4_verifier           verf = exid->verifier;
1441         struct sockaddr         *sa = svc_addr(rqstp);
1442
1443         rpc_ntop(sa, addr_str, sizeof(addr_str));
1444         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1445                 "ip_addr=%s flags %x, spa_how %d\n",
1446                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1447                 addr_str, exid->flags, exid->spa_how);
1448
1449         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1450                 return nfserr_inval;
1451
1452         /* Currently only support SP4_NONE */
1453         switch (exid->spa_how) {
1454         case SP4_NONE:
1455                 break;
1456         case SP4_SSV:
1457                 return nfserr_serverfault;
1458         default:
1459                 BUG();                          /* checked by xdr code */
1460         case SP4_MACH_CRED:
1461                 return nfserr_serverfault;      /* no excuse :-/ */
1462         }
1463
1464         status = nfs4_make_rec_clidname(dname, &exid->clname);
1465
1466         if (status)
1467                 goto error;
1468
1469         strhashval = clientstr_hashval(dname);
1470
1471         nfs4_lock_state();
1472         status = nfs_ok;
1473
1474         conf = find_confirmed_client_by_str(dname, strhashval);
1475         if (conf) {
1476                 if (!clp_used_exchangeid(conf)) {
1477                         status = nfserr_clid_inuse; /* XXX: ? */
1478                         goto out;
1479                 }
1480                 if (!same_verf(&verf, &conf->cl_verifier)) {
1481                         /* 18.35.4 case 8 */
1482                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1483                                 status = nfserr_not_same;
1484                                 goto out;
1485                         }
1486                         /* Client reboot: destroy old state */
1487                         expire_client(conf);
1488                         goto out_new;
1489                 }
1490                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1491                         /* 18.35.4 case 9 */
1492                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1493                                 status = nfserr_perm;
1494                                 goto out;
1495                         }
1496                         expire_client(conf);
1497                         goto out_new;
1498                 }
1499                 /*
1500                  * Set bit when the owner id and verifier map to an already
1501                  * confirmed client id (18.35.3).
1502                  */
1503                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1504
1505                 /*
1506                  * Falling into 18.35.4 case 2, possible router replay.
1507                  * Leave confirmed record intact and return same result.
1508                  */
1509                 copy_verf(conf, &verf);
1510                 new = conf;
1511                 goto out_copy;
1512         }
1513
1514         /* 18.35.4 case 7 */
1515         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1516                 status = nfserr_noent;
1517                 goto out;
1518         }
1519
1520         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1521         if (unconf) {
1522                 /*
1523                  * Possible retry or client restart.  Per 18.35.4 case 4,
1524                  * a new unconfirmed record should be generated regardless
1525                  * of whether any properties have changed.
1526                  */
1527                 expire_client(unconf);
1528         }
1529
1530 out_new:
1531         /* Normal case */
1532         new = create_client(exid->clname, dname, rqstp, &verf);
1533         if (new == NULL) {
1534                 status = nfserr_jukebox;
1535                 goto out;
1536         }
1537
1538         gen_clid(new);
1539         add_to_unconfirmed(new, strhashval);
1540 out_copy:
1541         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1542         exid->clientid.cl_id = new->cl_clientid.cl_id;
1543
1544         exid->seqid = 1;
1545         nfsd4_set_ex_flags(new, exid);
1546
1547         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1548                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1549         status = nfs_ok;
1550
1551 out:
1552         nfs4_unlock_state();
1553 error:
1554         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1555         return status;
1556 }
1557
1558 static int
1559 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1560 {
1561         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1562                 slot_seqid);
1563
1564         /* The slot is in use, and no response has been sent. */
1565         if (slot_inuse) {
1566                 if (seqid == slot_seqid)
1567                         return nfserr_jukebox;
1568                 else
1569                         return nfserr_seq_misordered;
1570         }
1571         /* Normal */
1572         if (likely(seqid == slot_seqid + 1))
1573                 return nfs_ok;
1574         /* Replay */
1575         if (seqid == slot_seqid)
1576                 return nfserr_replay_cache;
1577         /* Wraparound */
1578         if (seqid == 1 && (slot_seqid + 1) == 0)
1579                 return nfs_ok;
1580         /* Misordered replay or misordered new request */
1581         return nfserr_seq_misordered;
1582 }
1583
1584 /*
1585  * Cache the create session result into the create session single DRC
1586  * slot cache by saving the xdr structure. sl_seqid has been set.
1587  * Do this for solo or embedded create session operations.
1588  */
1589 static void
1590 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1591                            struct nfsd4_clid_slot *slot, int nfserr)
1592 {
1593         slot->sl_status = nfserr;
1594         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1595 }
1596
1597 static __be32
1598 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1599                             struct nfsd4_clid_slot *slot)
1600 {
1601         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1602         return slot->sl_status;
1603 }
1604
1605 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1606                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1607                         1 +     /* MIN tag is length with zero, only length */ \
1608                         3 +     /* version, opcount, opcode */ \
1609                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1610                                 /* seqid, slotID, slotID, cache */ \
1611                         4 ) * sizeof(__be32))
1612
1613 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1614                         2 +     /* verifier: AUTH_NULL, length 0 */\
1615                         1 +     /* status */ \
1616                         1 +     /* MIN tag is length with zero, only length */ \
1617                         3 +     /* opcount, opcode, opstatus*/ \
1618                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1619                                 /* seqid, slotID, slotID, slotID, status */ \
1620                         5 ) * sizeof(__be32))
1621
1622 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1623 {
1624         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1625                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1626 }
1627
1628 __be32
1629 nfsd4_create_session(struct svc_rqst *rqstp,
1630                      struct nfsd4_compound_state *cstate,
1631                      struct nfsd4_create_session *cr_ses)
1632 {
1633         struct sockaddr *sa = svc_addr(rqstp);
1634         struct nfs4_client *conf, *unconf;
1635         struct nfsd4_session *new;
1636         struct nfsd4_clid_slot *cs_slot = NULL;
1637         bool confirm_me = false;
1638         int status = 0;
1639
1640         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1641                 return nfserr_inval;
1642
1643         nfs4_lock_state();
1644         unconf = find_unconfirmed_client(&cr_ses->clientid);
1645         conf = find_confirmed_client(&cr_ses->clientid);
1646
1647         if (conf) {
1648                 cs_slot = &conf->cl_cs_slot;
1649                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1650                 if (status == nfserr_replay_cache) {
1651                         dprintk("Got a create_session replay! seqid= %d\n",
1652                                 cs_slot->sl_seqid);
1653                         /* Return the cached reply status */
1654                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1655                         goto out;
1656                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1657                         status = nfserr_seq_misordered;
1658                         dprintk("Sequence misordered!\n");
1659                         dprintk("Expected seqid= %d but got seqid= %d\n",
1660                                 cs_slot->sl_seqid, cr_ses->seqid);
1661                         goto out;
1662                 }
1663         } else if (unconf) {
1664                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1665                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1666                         status = nfserr_clid_inuse;
1667                         goto out;
1668                 }
1669
1670                 cs_slot = &unconf->cl_cs_slot;
1671                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1672                 if (status) {
1673                         /* an unconfirmed replay returns misordered */
1674                         status = nfserr_seq_misordered;
1675                         goto out;
1676                 }
1677
1678                 confirm_me = true;
1679                 conf = unconf;
1680         } else {
1681                 status = nfserr_stale_clientid;
1682                 goto out;
1683         }
1684
1685         /*
1686          * XXX: we should probably set this at creation time, and check
1687          * for consistent minorversion use throughout:
1688          */
1689         conf->cl_minorversion = 1;
1690         /*
1691          * We do not support RDMA or persistent sessions
1692          */
1693         cr_ses->flags &= ~SESSION4_PERSIST;
1694         cr_ses->flags &= ~SESSION4_RDMA;
1695
1696         status = nfserr_toosmall;
1697         if (check_forechannel_attrs(cr_ses->fore_channel))
1698                 goto out;
1699
1700         status = nfserr_jukebox;
1701         new = alloc_init_session(rqstp, conf, cr_ses);
1702         if (!new)
1703                 goto out;
1704         status = nfs_ok;
1705         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1706                NFS4_MAX_SESSIONID_LEN);
1707         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1708                 sizeof(struct nfsd4_channel_attrs));
1709         cs_slot->sl_seqid++;
1710         cr_ses->seqid = cs_slot->sl_seqid;
1711
1712         /* cache solo and embedded create sessions under the state lock */
1713         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1714         if (confirm_me)
1715                 move_to_confirmed(conf);
1716 out:
1717         nfs4_unlock_state();
1718         dprintk("%s returns %d\n", __func__, ntohl(status));
1719         return status;
1720 }
1721
1722 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1723 {
1724         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1725         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1726
1727         return argp->opcnt == resp->opcnt;
1728 }
1729
1730 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1731 {
1732         switch (*dir) {
1733         case NFS4_CDFC4_FORE:
1734         case NFS4_CDFC4_BACK:
1735                 return nfs_ok;
1736         case NFS4_CDFC4_FORE_OR_BOTH:
1737         case NFS4_CDFC4_BACK_OR_BOTH:
1738                 *dir = NFS4_CDFC4_BOTH;
1739                 return nfs_ok;
1740         };
1741         return nfserr_inval;
1742 }
1743
1744 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1745                      struct nfsd4_compound_state *cstate,
1746                      struct nfsd4_bind_conn_to_session *bcts)
1747 {
1748         __be32 status;
1749
1750         if (!nfsd4_last_compound_op(rqstp))
1751                 return nfserr_not_only_op;
1752         spin_lock(&client_lock);
1753         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1754         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1755          * client_lock iself: */
1756         if (cstate->session) {
1757                 nfsd4_get_session(cstate->session);
1758                 atomic_inc(&cstate->session->se_client->cl_refcount);
1759         }
1760         spin_unlock(&client_lock);
1761         if (!cstate->session)
1762                 return nfserr_badsession;
1763
1764         status = nfsd4_map_bcts_dir(&bcts->dir);
1765         if (!status)
1766                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1767         return status;
1768 }
1769
1770 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1771 {
1772         if (!session)
1773                 return 0;
1774         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1775 }
1776
1777 __be32
1778 nfsd4_destroy_session(struct svc_rqst *r,
1779                       struct nfsd4_compound_state *cstate,
1780                       struct nfsd4_destroy_session *sessionid)
1781 {
1782         struct nfsd4_session *ses;
1783         u32 status = nfserr_badsession;
1784
1785         /* Notes:
1786          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1787          * - Should we return nfserr_back_chan_busy if waiting for
1788          *   callbacks on to-be-destroyed session?
1789          * - Do we need to clear any callback info from previous session?
1790          */
1791
1792         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1793                 if (!nfsd4_last_compound_op(r))
1794                         return nfserr_not_only_op;
1795         }
1796         dump_sessionid(__func__, &sessionid->sessionid);
1797         spin_lock(&client_lock);
1798         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1799         if (!ses) {
1800                 spin_unlock(&client_lock);
1801                 goto out;
1802         }
1803
1804         unhash_session(ses);
1805         spin_unlock(&client_lock);
1806
1807         nfs4_lock_state();
1808         nfsd4_probe_callback_sync(ses->se_client);
1809         nfs4_unlock_state();
1810
1811         nfsd4_del_conns(ses);
1812
1813         nfsd4_put_session(ses);
1814         status = nfs_ok;
1815 out:
1816         dprintk("%s returns %d\n", __func__, ntohl(status));
1817         return status;
1818 }
1819
1820 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1821 {
1822         struct nfsd4_conn *c;
1823
1824         list_for_each_entry(c, &s->se_conns, cn_persession) {
1825                 if (c->cn_xprt == xpt) {
1826                         return c;
1827                 }
1828         }
1829         return NULL;
1830 }
1831
1832 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1833 {
1834         struct nfs4_client *clp = ses->se_client;
1835         struct nfsd4_conn *c;
1836         int ret;
1837
1838         spin_lock(&clp->cl_lock);
1839         c = __nfsd4_find_conn(new->cn_xprt, ses);
1840         if (c) {
1841                 spin_unlock(&clp->cl_lock);
1842                 free_conn(new);
1843                 return;
1844         }
1845         __nfsd4_hash_conn(new, ses);
1846         spin_unlock(&clp->cl_lock);
1847         ret = nfsd4_register_conn(new);
1848         if (ret)
1849                 /* oops; xprt is already down: */
1850                 nfsd4_conn_lost(&new->cn_xpt_user);
1851         return;
1852 }
1853
1854 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1855 {
1856         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1857
1858         return args->opcnt > session->se_fchannel.maxops;
1859 }
1860
1861 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
1862                                   struct nfsd4_session *session)
1863 {
1864         struct xdr_buf *xb = &rqstp->rq_arg;
1865
1866         return xb->len > session->se_fchannel.maxreq_sz;
1867 }
1868
1869 __be32
1870 nfsd4_sequence(struct svc_rqst *rqstp,
1871                struct nfsd4_compound_state *cstate,
1872                struct nfsd4_sequence *seq)
1873 {
1874         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1875         struct nfsd4_session *session;
1876         struct nfsd4_slot *slot;
1877         struct nfsd4_conn *conn;
1878         int status;
1879
1880         if (resp->opcnt != 1)
1881                 return nfserr_sequence_pos;
1882
1883         /*
1884          * Will be either used or freed by nfsd4_sequence_check_conn
1885          * below.
1886          */
1887         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1888         if (!conn)
1889                 return nfserr_jukebox;
1890
1891         spin_lock(&client_lock);
1892         status = nfserr_badsession;
1893         session = find_in_sessionid_hashtbl(&seq->sessionid);
1894         if (!session)
1895                 goto out;
1896
1897         status = nfserr_too_many_ops;
1898         if (nfsd4_session_too_many_ops(rqstp, session))
1899                 goto out;
1900
1901         status = nfserr_req_too_big;
1902         if (nfsd4_request_too_big(rqstp, session))
1903                 goto out;
1904
1905         status = nfserr_badslot;
1906         if (seq->slotid >= session->se_fchannel.maxreqs)
1907                 goto out;
1908
1909         slot = session->se_slots[seq->slotid];
1910         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1911
1912         /* We do not negotiate the number of slots yet, so set the
1913          * maxslots to the session maxreqs which is used to encode
1914          * sr_highest_slotid and the sr_target_slot id to maxslots */
1915         seq->maxslots = session->se_fchannel.maxreqs;
1916
1917         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1918         if (status == nfserr_replay_cache) {
1919                 cstate->slot = slot;
1920                 cstate->session = session;
1921                 /* Return the cached reply status and set cstate->status
1922                  * for nfsd4_proc_compound processing */
1923                 status = nfsd4_replay_cache_entry(resp, seq);
1924                 cstate->status = nfserr_replay_cache;
1925                 goto out;
1926         }
1927         if (status)
1928                 goto out;
1929
1930         nfsd4_sequence_check_conn(conn, session);
1931         conn = NULL;
1932
1933         /* Success! bump slot seqid */
1934         slot->sl_inuse = true;
1935         slot->sl_seqid = seq->seqid;
1936         slot->sl_cachethis = seq->cachethis;
1937
1938         cstate->slot = slot;
1939         cstate->session = session;
1940
1941 out:
1942         /* Hold a session reference until done processing the compound. */
1943         if (cstate->session) {
1944                 struct nfs4_client *clp = session->se_client;
1945
1946                 nfsd4_get_session(cstate->session);
1947                 atomic_inc(&clp->cl_refcount);
1948                 switch (clp->cl_cb_state) {
1949                 case NFSD4_CB_DOWN:
1950                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
1951                         break;
1952                 case NFSD4_CB_FAULT:
1953                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
1954                         break;
1955                 default:
1956                         seq->status_flags = 0;
1957                 }
1958         }
1959         kfree(conn);
1960         spin_unlock(&client_lock);
1961         dprintk("%s: return %d\n", __func__, ntohl(status));
1962         return status;
1963 }
1964
1965 static inline bool has_resources(struct nfs4_client *clp)
1966 {
1967         return !list_empty(&clp->cl_openowners)
1968                 || !list_empty(&clp->cl_delegations)
1969                 || !list_empty(&clp->cl_sessions);
1970 }
1971
1972 __be32
1973 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
1974 {
1975         struct nfs4_client *conf, *unconf, *clp;
1976         int status = 0;
1977
1978         nfs4_lock_state();
1979         unconf = find_unconfirmed_client(&dc->clientid);
1980         conf = find_confirmed_client(&dc->clientid);
1981
1982         if (conf) {
1983                 clp = conf;
1984
1985                 if (!is_client_expired(conf) && has_resources(conf)) {
1986                         status = nfserr_clientid_busy;
1987                         goto out;
1988                 }
1989
1990                 /* rfc5661 18.50.3 */
1991                 if (cstate->session && conf == cstate->session->se_client) {
1992                         status = nfserr_clientid_busy;
1993                         goto out;
1994                 }
1995         } else if (unconf)
1996                 clp = unconf;
1997         else {
1998                 status = nfserr_stale_clientid;
1999                 goto out;
2000         }
2001
2002         expire_client(clp);
2003 out:
2004         nfs4_unlock_state();
2005         dprintk("%s return %d\n", __func__, ntohl(status));
2006         return status;
2007 }
2008
2009 __be32
2010 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2011 {
2012         int status = 0;
2013
2014         if (rc->rca_one_fs) {
2015                 if (!cstate->current_fh.fh_dentry)
2016                         return nfserr_nofilehandle;
2017                 /*
2018                  * We don't take advantage of the rca_one_fs case.
2019                  * That's OK, it's optional, we can safely ignore it.
2020                  */
2021                  return nfs_ok;
2022         }
2023
2024         nfs4_lock_state();
2025         status = nfserr_complete_already;
2026         if (cstate->session->se_client->cl_firststate)
2027                 goto out;
2028
2029         status = nfserr_stale_clientid;
2030         if (is_client_expired(cstate->session->se_client))
2031                 /*
2032                  * The following error isn't really legal.
2033                  * But we only get here if the client just explicitly
2034                  * destroyed the client.  Surely it no longer cares what
2035                  * error it gets back on an operation for the dead
2036                  * client.
2037                  */
2038                 goto out;
2039
2040         status = nfs_ok;
2041         nfsd4_create_clid_dir(cstate->session->se_client);
2042 out:
2043         nfs4_unlock_state();
2044         return status;
2045 }
2046
2047 __be32
2048 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2049                   struct nfsd4_setclientid *setclid)
2050 {
2051         struct xdr_netobj       clname = setclid->se_name;
2052         nfs4_verifier           clverifier = setclid->se_verf;
2053         unsigned int            strhashval;
2054         struct nfs4_client      *conf, *unconf, *new;
2055         __be32                  status;
2056         char                    dname[HEXDIR_LEN];
2057         
2058         status = nfs4_make_rec_clidname(dname, &clname);
2059         if (status)
2060                 return status;
2061
2062         /* 
2063          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2064          * We get here on a DRC miss.
2065          */
2066
2067         strhashval = clientstr_hashval(dname);
2068
2069         nfs4_lock_state();
2070         conf = find_confirmed_client_by_str(dname, strhashval);
2071         if (conf) {
2072                 /* RFC 3530 14.2.33 CASE 0: */
2073                 status = nfserr_clid_inuse;
2074                 if (clp_used_exchangeid(conf))
2075                         goto out;
2076                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2077                         char addr_str[INET6_ADDRSTRLEN];
2078                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2079                                  sizeof(addr_str));
2080                         dprintk("NFSD: setclientid: string in use by client "
2081                                 "at %s\n", addr_str);
2082                         goto out;
2083                 }
2084         }
2085         /*
2086          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
2087          * has a description of SETCLIENTID request processing consisting
2088          * of 5 bullet points, labeled as CASE0 - CASE4 below.
2089          */
2090         unconf = find_unconfirmed_client_by_str(dname, strhashval);
2091         status = nfserr_jukebox;
2092         if (!conf) {
2093                 /*
2094                  * RFC 3530 14.2.33 CASE 4:
2095                  * placed first, because it is the normal case
2096                  */
2097                 if (unconf)
2098                         expire_client(unconf);
2099                 new = create_client(clname, dname, rqstp, &clverifier);
2100                 if (new == NULL)
2101                         goto out;
2102                 gen_clid(new);
2103         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
2104                 /*
2105                  * RFC 3530 14.2.33 CASE 1:
2106                  * probable callback update
2107                  */
2108                 if (unconf) {
2109                         /* Note this is removing unconfirmed {*x***},
2110                          * which is stronger than RFC recommended {vxc**}.
2111                          * This has the advantage that there is at most
2112                          * one {*x***} in either list at any time.
2113                          */
2114                         expire_client(unconf);
2115                 }
2116                 new = create_client(clname, dname, rqstp, &clverifier);
2117                 if (new == NULL)
2118                         goto out;
2119                 copy_clid(new, conf);
2120         } else if (!unconf) {
2121                 /*
2122                  * RFC 3530 14.2.33 CASE 2:
2123                  * probable client reboot; state will be removed if
2124                  * confirmed.
2125                  */
2126                 new = create_client(clname, dname, rqstp, &clverifier);
2127                 if (new == NULL)
2128                         goto out;
2129                 gen_clid(new);
2130         } else {
2131                 /*
2132                  * RFC 3530 14.2.33 CASE 3:
2133                  * probable client reboot; state will be removed if
2134                  * confirmed.
2135                  */
2136                 expire_client(unconf);
2137                 new = create_client(clname, dname, rqstp, &clverifier);
2138                 if (new == NULL)
2139                         goto out;
2140                 gen_clid(new);
2141         }
2142         /*
2143          * XXX: we should probably set this at creation time, and check
2144          * for consistent minorversion use throughout:
2145          */
2146         new->cl_minorversion = 0;
2147         gen_callback(new, setclid, rqstp);
2148         add_to_unconfirmed(new, strhashval);
2149         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2150         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2151         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2152         status = nfs_ok;
2153 out:
2154         nfs4_unlock_state();
2155         return status;
2156 }
2157
2158
2159 /*
2160  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
2161  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
2162  * bullets, labeled as CASE1 - CASE4 below.
2163  */
2164 __be32
2165 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2166                          struct nfsd4_compound_state *cstate,
2167                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2168 {
2169         struct sockaddr *sa = svc_addr(rqstp);
2170         struct nfs4_client *conf, *unconf;
2171         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
2172         clientid_t * clid = &setclientid_confirm->sc_clientid;
2173         __be32 status;
2174
2175         if (STALE_CLIENTID(clid))
2176                 return nfserr_stale_clientid;
2177         /* 
2178          * XXX The Duplicate Request Cache (DRC) has been checked (??)
2179          * We get here on a DRC miss.
2180          */
2181
2182         nfs4_lock_state();
2183
2184         conf = find_confirmed_client(clid);
2185         unconf = find_unconfirmed_client(clid);
2186
2187         status = nfserr_clid_inuse;
2188         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2189                 goto out;
2190         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2191                 goto out;
2192
2193         /*
2194          * section 14.2.34 of RFC 3530 has a description of
2195          * SETCLIENTID_CONFIRM request processing consisting
2196          * of 4 bullet points, labeled as CASE1 - CASE4 below.
2197          */
2198         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2199                 /*
2200                  * RFC 3530 14.2.34 CASE 1:
2201                  * callback update
2202                  */
2203                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2204                         status = nfserr_clid_inuse;
2205                 else {
2206                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2207                         nfsd4_probe_callback(conf);
2208                         expire_client(unconf);
2209                         status = nfs_ok;
2210
2211                 }
2212         } else if (conf && !unconf) {
2213                 /*
2214                  * RFC 3530 14.2.34 CASE 2:
2215                  * probable retransmitted request; play it safe and
2216                  * do nothing.
2217                  */
2218                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2219                         status = nfserr_clid_inuse;
2220                 else
2221                         status = nfs_ok;
2222         } else if (!conf && unconf
2223                         && same_verf(&unconf->cl_confirm, &confirm)) {
2224                 /*
2225                  * RFC 3530 14.2.34 CASE 3:
2226                  * Normal case; new or rebooted client:
2227                  */
2228                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2229                         status = nfserr_clid_inuse;
2230                 } else {
2231                         unsigned int hash =
2232                                 clientstr_hashval(unconf->cl_recdir);
2233                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2234                                                             hash);
2235                         if (conf) {
2236                                 nfsd4_remove_clid_dir(conf);
2237                                 expire_client(conf);
2238                         }
2239                         move_to_confirmed(unconf);
2240                         conf = unconf;
2241                         nfsd4_probe_callback(conf);
2242                         status = nfs_ok;
2243                 }
2244         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2245             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2246                                                                 &confirm)))) {
2247                 /*
2248                  * RFC 3530 14.2.34 CASE 4:
2249                  * Client probably hasn't noticed that we rebooted yet.
2250                  */
2251                 status = nfserr_stale_clientid;
2252         } else {
2253                 /* check that we have hit one of the cases...*/
2254                 status = nfserr_clid_inuse;
2255         }
2256 out:
2257         nfs4_unlock_state();
2258         return status;
2259 }
2260
2261 static struct nfs4_file *nfsd4_alloc_file(void)
2262 {
2263         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2264 }
2265
2266 /* OPEN Share state helper functions */
2267 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2268 {
2269         unsigned int hashval = file_hashval(ino);
2270
2271         atomic_set(&fp->fi_ref, 1);
2272         INIT_LIST_HEAD(&fp->fi_hash);
2273         INIT_LIST_HEAD(&fp->fi_stateids);
2274         INIT_LIST_HEAD(&fp->fi_delegations);
2275         fp->fi_inode = igrab(ino);
2276         fp->fi_had_conflict = false;
2277         fp->fi_lease = NULL;
2278         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2279         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2280         spin_lock(&recall_lock);
2281         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2282         spin_unlock(&recall_lock);
2283 }
2284
2285 static void
2286 nfsd4_free_slab(struct kmem_cache **slab)
2287 {
2288         if (*slab == NULL)
2289                 return;
2290         kmem_cache_destroy(*slab);
2291         *slab = NULL;
2292 }
2293
2294 void
2295 nfsd4_free_slabs(void)
2296 {
2297         nfsd4_free_slab(&openowner_slab);
2298         nfsd4_free_slab(&lockowner_slab);
2299         nfsd4_free_slab(&file_slab);
2300         nfsd4_free_slab(&stateid_slab);
2301         nfsd4_free_slab(&deleg_slab);
2302 }
2303
2304 static int
2305 nfsd4_init_slabs(void)
2306 {
2307         openowner_slab = kmem_cache_create("nfsd4_openowners",
2308                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2309         if (openowner_slab == NULL)
2310                 goto out_nomem;
2311         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2312                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2313         if (lockowner_slab == NULL)
2314                 goto out_nomem;
2315         file_slab = kmem_cache_create("nfsd4_files",
2316                         sizeof(struct nfs4_file), 0, 0, NULL);
2317         if (file_slab == NULL)
2318                 goto out_nomem;
2319         stateid_slab = kmem_cache_create("nfsd4_stateids",
2320                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2321         if (stateid_slab == NULL)
2322                 goto out_nomem;
2323         deleg_slab = kmem_cache_create("nfsd4_delegations",
2324                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2325         if (deleg_slab == NULL)
2326                 goto out_nomem;
2327         return 0;
2328 out_nomem:
2329         nfsd4_free_slabs();
2330         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2331         return -ENOMEM;
2332 }
2333
2334 void nfs4_free_openowner(struct nfs4_openowner *oo)
2335 {
2336         kfree(oo->oo_owner.so_owner.data);
2337         kmem_cache_free(openowner_slab, oo);
2338 }
2339
2340 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2341 {
2342         kfree(lo->lo_owner.so_owner.data);
2343         kmem_cache_free(lockowner_slab, lo);
2344 }
2345
2346 static void init_nfs4_replay(struct nfs4_replay *rp)
2347 {
2348         rp->rp_status = nfserr_serverfault;
2349         rp->rp_buflen = 0;
2350         rp->rp_buf = rp->rp_ibuf;
2351 }
2352
2353 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2354 {
2355         struct nfs4_stateowner *sop;
2356
2357         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2358         if (!sop)
2359                 return NULL;
2360
2361         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2362         if (!sop->so_owner.data) {
2363                 kmem_cache_free(slab, sop);
2364                 return NULL;
2365         }
2366         sop->so_owner.len = owner->len;
2367
2368         INIT_LIST_HEAD(&sop->so_stateids);
2369         sop->so_client = clp;
2370         init_nfs4_replay(&sop->so_replay);
2371         return sop;
2372 }
2373
2374 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2375 {
2376         list_add(&oo->oo_owner.so_strhash, &open_ownerstr_hashtbl[strhashval]);
2377         list_add(&oo->oo_perclient, &clp->cl_openowners);
2378 }
2379
2380 static struct nfs4_openowner *
2381 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2382         struct nfs4_openowner *oo;
2383
2384         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2385         if (!oo)
2386                 return NULL;
2387         oo->oo_owner.so_is_open_owner = 1;
2388         oo->oo_owner.so_seqid = open->op_seqid;
2389         oo->oo_flags = NFS4_OO_NEW;
2390         oo->oo_time = 0;
2391         oo->oo_last_closed_stid = NULL;
2392         INIT_LIST_HEAD(&oo->oo_close_lru);
2393         hash_openowner(oo, clp, strhashval);
2394         return oo;
2395 }
2396
2397 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2398         struct nfs4_openowner *oo = open->op_openowner;
2399         struct nfs4_client *clp = oo->oo_owner.so_client;
2400
2401         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2402         INIT_LIST_HEAD(&stp->st_lockowners);
2403         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2404         list_add(&stp->st_perfile, &fp->fi_stateids);
2405         stp->st_stateowner = &oo->oo_owner;
2406         get_nfs4_file(fp);
2407         stp->st_file = fp;
2408         stp->st_access_bmap = 0;
2409         stp->st_deny_bmap = 0;
2410         __set_bit(open->op_share_access, &stp->st_access_bmap);
2411         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2412         stp->st_openstp = NULL;
2413 }
2414
2415 static void
2416 move_to_close_lru(struct nfs4_openowner *oo)
2417 {
2418         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2419
2420         list_move_tail(&oo->oo_close_lru, &close_lru);
2421         oo->oo_time = get_seconds();
2422 }
2423
2424 static int
2425 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2426                                                         clientid_t *clid)
2427 {
2428         return (sop->so_owner.len == owner->len) &&
2429                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2430                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2431 }
2432
2433 static struct nfs4_openowner *
2434 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2435 {
2436         struct nfs4_stateowner *so;
2437         struct nfs4_openowner *oo;
2438
2439         list_for_each_entry(so, &open_ownerstr_hashtbl[hashval], so_strhash) {
2440                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2441                         oo = openowner(so);
2442                         renew_client(oo->oo_owner.so_client);
2443                         return oo;
2444                 }
2445         }
2446         return NULL;
2447 }
2448
2449 /* search file_hashtbl[] for file */
2450 static struct nfs4_file *
2451 find_file(struct inode *ino)
2452 {
2453         unsigned int hashval = file_hashval(ino);
2454         struct nfs4_file *fp;
2455
2456         spin_lock(&recall_lock);
2457         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2458                 if (fp->fi_inode == ino) {
2459                         get_nfs4_file(fp);
2460                         spin_unlock(&recall_lock);
2461                         return fp;
2462                 }
2463         }
2464         spin_unlock(&recall_lock);
2465         return NULL;
2466 }
2467
2468 /*
2469  * Called to check deny when READ with all zero stateid or
2470  * WRITE with all zero or all one stateid
2471  */
2472 static __be32
2473 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2474 {
2475         struct inode *ino = current_fh->fh_dentry->d_inode;
2476         struct nfs4_file *fp;
2477         struct nfs4_ol_stateid *stp;
2478         __be32 ret;
2479
2480         dprintk("NFSD: nfs4_share_conflict\n");
2481
2482         fp = find_file(ino);
2483         if (!fp)
2484                 return nfs_ok;
2485         ret = nfserr_locked;
2486         /* Search for conflicting share reservations */
2487         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2488                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2489                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2490                         goto out;
2491         }
2492         ret = nfs_ok;
2493 out:
2494         put_nfs4_file(fp);
2495         return ret;
2496 }
2497
2498 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2499 {
2500         /* We're assuming the state code never drops its reference
2501          * without first removing the lease.  Since we're in this lease
2502          * callback (and since the lease code is serialized by the kernel
2503          * lock) we know the server hasn't removed the lease yet, we know
2504          * it's safe to take a reference: */
2505         atomic_inc(&dp->dl_count);
2506
2507         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2508
2509         /* only place dl_time is set. protected by lock_flocks*/
2510         dp->dl_time = get_seconds();
2511
2512         nfsd4_cb_recall(dp);
2513 }
2514
2515 /* Called from break_lease() with lock_flocks() held. */
2516 static void nfsd_break_deleg_cb(struct file_lock *fl)
2517 {
2518         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2519         struct nfs4_delegation *dp;
2520
2521         BUG_ON(!fp);
2522         /* We assume break_lease is only called once per lease: */
2523         BUG_ON(fp->fi_had_conflict);
2524         /*
2525          * We don't want the locks code to timeout the lease for us;
2526          * we'll remove it ourself if a delegation isn't returned
2527          * in time:
2528          */
2529         fl->fl_break_time = 0;
2530
2531         spin_lock(&recall_lock);
2532         fp->fi_had_conflict = true;
2533         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2534                 nfsd_break_one_deleg(dp);
2535         spin_unlock(&recall_lock);
2536 }
2537
2538 static
2539 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2540 {
2541         if (arg & F_UNLCK)
2542                 return lease_modify(onlist, arg);
2543         else
2544                 return -EAGAIN;
2545 }
2546
2547 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2548         .lm_break = nfsd_break_deleg_cb,
2549         .lm_change = nfsd_change_deleg_cb,
2550 };
2551
2552 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2553 {
2554         if (nfsd4_has_session(cstate))
2555                 return nfs_ok;
2556         if (seqid == so->so_seqid - 1)
2557                 return nfserr_replay_me;
2558         if (seqid == so->so_seqid)
2559                 return nfs_ok;
2560         return nfserr_bad_seqid;
2561 }
2562
2563 __be32
2564 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2565                     struct nfsd4_open *open)
2566 {
2567         clientid_t *clientid = &open->op_clientid;
2568         struct nfs4_client *clp = NULL;
2569         unsigned int strhashval;
2570         struct nfs4_openowner *oo = NULL;
2571         __be32 status;
2572
2573         if (STALE_CLIENTID(&open->op_clientid))
2574                 return nfserr_stale_clientid;
2575         /*
2576          * In case we need it later, after we've already created the
2577          * file and don't want to risk a further failure:
2578          */
2579         open->op_file = nfsd4_alloc_file();
2580         if (open->op_file == NULL)
2581                 return nfserr_jukebox;
2582
2583         strhashval = open_ownerstr_hashval(clientid->cl_id, &open->op_owner);
2584         oo = find_openstateowner_str(strhashval, open);
2585         open->op_openowner = oo;
2586         if (!oo) {
2587                 clp = find_confirmed_client(clientid);
2588                 if (clp == NULL)
2589                         return nfserr_expired;
2590                 goto new_owner;
2591         }
2592         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2593                 /* Replace unconfirmed owners without checking for replay. */
2594                 clp = oo->oo_owner.so_client;
2595                 release_openowner(oo);
2596                 open->op_openowner = NULL;
2597                 goto new_owner;
2598         }
2599         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2600         if (status)
2601                 return status;
2602         clp = oo->oo_owner.so_client;
2603         goto alloc_stateid;
2604 new_owner:
2605         oo = alloc_init_open_stateowner(strhashval, clp, open);
2606         if (oo == NULL)
2607                 return nfserr_jukebox;
2608         open->op_openowner = oo;
2609 alloc_stateid:
2610         open->op_stp = nfs4_alloc_stateid(clp);
2611         if (!open->op_stp)
2612                 return nfserr_jukebox;
2613         return nfs_ok;
2614 }
2615
2616 static inline __be32
2617 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2618 {
2619         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2620                 return nfserr_openmode;
2621         else
2622                 return nfs_ok;
2623 }
2624
2625 static int share_access_to_flags(u32 share_access)
2626 {
2627         share_access &= ~NFS4_SHARE_WANT_MASK;
2628
2629         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2630 }
2631
2632 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2633 {
2634         struct nfs4_stid *ret;
2635
2636         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2637         if (!ret)
2638                 return NULL;
2639         return delegstateid(ret);
2640 }
2641
2642 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2643 {
2644         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2645                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2646 }
2647
2648 static __be32
2649 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2650                 struct nfs4_delegation **dp)
2651 {
2652         int flags;
2653         __be32 status = nfserr_bad_stateid;
2654
2655         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2656         if (*dp == NULL)
2657                 goto out;
2658         flags = share_access_to_flags(open->op_share_access);
2659         status = nfs4_check_delegmode(*dp, flags);
2660         if (status)
2661                 *dp = NULL;
2662 out:
2663         if (!nfsd4_is_deleg_cur(open))
2664                 return nfs_ok;
2665         if (status)
2666                 return status;
2667         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2668         return nfs_ok;
2669 }
2670
2671 static __be32
2672 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2673 {
2674         struct nfs4_ol_stateid *local;
2675         struct nfs4_openowner *oo = open->op_openowner;
2676
2677         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2678                 /* ignore lock owners */
2679                 if (local->st_stateowner->so_is_open_owner == 0)
2680                         continue;
2681                 /* remember if we have seen this open owner */
2682                 if (local->st_stateowner == &oo->oo_owner)
2683                         *stpp = local;
2684                 /* check for conflicting share reservations */
2685                 if (!test_share(local, open))
2686                         return nfserr_share_denied;
2687         }
2688         return nfs_ok;
2689 }
2690
2691 static void nfs4_free_stateid(struct nfs4_ol_stateid *s)
2692 {
2693         kmem_cache_free(stateid_slab, s);
2694 }
2695
2696 static inline int nfs4_access_to_access(u32 nfs4_access)
2697 {
2698         int flags = 0;
2699
2700         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2701                 flags |= NFSD_MAY_READ;
2702         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2703                 flags |= NFSD_MAY_WRITE;
2704         return flags;
2705 }
2706
2707 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2708                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2709 {
2710         __be32 status;
2711         int oflag = nfs4_access_to_omode(open->op_share_access);
2712         int access = nfs4_access_to_access(open->op_share_access);
2713
2714         if (!fp->fi_fds[oflag]) {
2715                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2716                         &fp->fi_fds[oflag]);
2717                 if (status)
2718                         return status;
2719         }
2720         nfs4_file_get_access(fp, oflag);
2721
2722         return nfs_ok;
2723 }
2724
2725 static inline __be32
2726 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2727                 struct nfsd4_open *open)
2728 {
2729         struct iattr iattr = {
2730                 .ia_valid = ATTR_SIZE,
2731                 .ia_size = 0,
2732         };
2733         if (!open->op_truncate)
2734                 return 0;
2735         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2736                 return nfserr_inval;
2737         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2738 }
2739
2740 static __be32
2741 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
2742 {
2743         u32 op_share_access = open->op_share_access;
2744         bool new_access;
2745         __be32 status;
2746
2747         new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2748         if (new_access) {
2749                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2750                 if (status)
2751                         return status;
2752         }
2753         status = nfsd4_truncate(rqstp, cur_fh, open);
2754         if (status) {
2755                 if (new_access) {
2756                         int oflag = nfs4_access_to_omode(op_share_access);
2757                         nfs4_file_put_access(fp, oflag);
2758                 }
2759                 return status;
2760         }
2761         /* remember the open */
2762         __set_bit(op_share_access, &stp->st_access_bmap);
2763         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2764
2765         return nfs_ok;
2766 }
2767
2768
2769 static void
2770 nfs4_set_claim_prev(struct nfsd4_open *open)
2771 {
2772         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2773         open->op_openowner->oo_owner.so_client->cl_firststate = 1;
2774 }
2775
2776 /* Should we give out recallable state?: */
2777 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2778 {
2779         if (clp->cl_cb_state == NFSD4_CB_UP)
2780                 return true;
2781         /*
2782          * In the sessions case, since we don't have to establish a
2783          * separate connection for callbacks, we assume it's OK
2784          * until we hear otherwise:
2785          */
2786         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2787 }
2788
2789 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2790 {
2791         struct file_lock *fl;
2792
2793         fl = locks_alloc_lock();
2794         if (!fl)
2795                 return NULL;
2796         locks_init_lock(fl);
2797         fl->fl_lmops = &nfsd_lease_mng_ops;
2798         fl->fl_flags = FL_LEASE;
2799         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2800         fl->fl_end = OFFSET_MAX;
2801         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2802         fl->fl_pid = current->tgid;
2803         return fl;
2804 }
2805
2806 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2807 {
2808         struct nfs4_file *fp = dp->dl_file;
2809         struct file_lock *fl;
2810         int status;
2811
2812         fl = nfs4_alloc_init_lease(dp, flag);
2813         if (!fl)
2814                 return -ENOMEM;
2815         fl->fl_file = find_readable_file(fp);
2816         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2817         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2818         if (status) {
2819                 list_del_init(&dp->dl_perclnt);
2820                 locks_free_lock(fl);
2821                 return -ENOMEM;
2822         }
2823         fp->fi_lease = fl;
2824         fp->fi_deleg_file = fl->fl_file;
2825         get_file(fp->fi_deleg_file);
2826         atomic_set(&fp->fi_delegees, 1);
2827         list_add(&dp->dl_perfile, &fp->fi_delegations);
2828         return 0;
2829 }
2830
2831 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2832 {
2833         struct nfs4_file *fp = dp->dl_file;
2834
2835         if (!fp->fi_lease)
2836                 return nfs4_setlease(dp, flag);
2837         spin_lock(&recall_lock);
2838         if (fp->fi_had_conflict) {
2839                 spin_unlock(&recall_lock);
2840                 return -EAGAIN;
2841         }
2842         atomic_inc(&fp->fi_delegees);
2843         list_add(&dp->dl_perfile, &fp->fi_delegations);
2844         spin_unlock(&recall_lock);
2845         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2846         return 0;
2847 }
2848
2849 /*
2850  * Attempt to hand out a delegation.
2851  */
2852 static void
2853 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2854 {
2855         struct nfs4_delegation *dp;
2856         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2857         int cb_up;
2858         int status, flag = 0;
2859
2860         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2861         flag = NFS4_OPEN_DELEGATE_NONE;
2862         open->op_recall = 0;
2863         switch (open->op_claim_type) {
2864                 case NFS4_OPEN_CLAIM_PREVIOUS:
2865                         if (!cb_up)
2866                                 open->op_recall = 1;
2867                         flag = open->op_delegate_type;
2868                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2869                                 goto out;
2870                         break;
2871                 case NFS4_OPEN_CLAIM_NULL:
2872                         /* Let's not give out any delegations till everyone's
2873                          * had the chance to reclaim theirs.... */
2874                         if (locks_in_grace())
2875                                 goto out;
2876                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2877                                 goto out;
2878                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2879                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2880                         else
2881                                 flag = NFS4_OPEN_DELEGATE_READ;
2882                         break;
2883                 default:
2884                         goto out;
2885         }
2886
2887         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2888         if (dp == NULL)
2889                 goto out_no_deleg;
2890         status = nfs4_set_delegation(dp, flag);
2891         if (status)
2892                 goto out_free;
2893
2894         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2895
2896         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2897                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2898 out:
2899         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2900                         && flag == NFS4_OPEN_DELEGATE_NONE
2901                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2902                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2903         open->op_delegate_type = flag;
2904         return;
2905 out_free:
2906         nfs4_put_delegation(dp);
2907 out_no_deleg:
2908         flag = NFS4_OPEN_DELEGATE_NONE;
2909         goto out;
2910 }
2911
2912 /*
2913  * called with nfs4_lock_state() held.
2914  */
2915 __be32
2916 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2917 {
2918         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2919         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
2920         struct nfs4_file *fp = NULL;
2921         struct inode *ino = current_fh->fh_dentry->d_inode;
2922         struct nfs4_ol_stateid *stp = NULL;
2923         struct nfs4_delegation *dp = NULL;
2924         __be32 status;
2925
2926         /*
2927          * Lookup file; if found, lookup stateid and check open request,
2928          * and check for delegations in the process of being recalled.
2929          * If not found, create the nfs4_file struct
2930          */
2931         fp = find_file(ino);
2932         if (fp) {
2933                 if ((status = nfs4_check_open(fp, open, &stp)))
2934                         goto out;
2935                 status = nfs4_check_deleg(cl, fp, open, &dp);
2936                 if (status)
2937                         goto out;
2938         } else {
2939                 status = nfserr_bad_stateid;
2940                 if (nfsd4_is_deleg_cur(open))
2941                         goto out;
2942                 status = nfserr_jukebox;
2943                 fp = open->op_file;
2944                 open->op_file = NULL;
2945                 nfsd4_init_file(fp, ino);
2946         }
2947
2948         /*
2949          * OPEN the file, or upgrade an existing OPEN.
2950          * If truncate fails, the OPEN fails.
2951          */
2952         if (stp) {
2953                 /* Stateid was found, this is an OPEN upgrade */
2954                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2955                 if (status)
2956                         goto out;
2957         } else {
2958                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
2959                 if (status)
2960                         goto out;
2961                 stp = open->op_stp;
2962                 open->op_stp = NULL;
2963                 init_open_stateid(stp, fp, open);
2964                 status = nfsd4_truncate(rqstp, current_fh, open);
2965                 if (status) {
2966                         release_open_stateid(stp);
2967                         goto out;
2968                 }
2969         }
2970         update_stateid(&stp->st_stid.sc_stateid);
2971         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
2972
2973         if (nfsd4_has_session(&resp->cstate))
2974                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2975
2976         /*
2977         * Attempt to hand out a delegation. No error return, because the
2978         * OPEN succeeds even if we fail.
2979         */
2980         nfs4_open_delegation(current_fh, open, stp);
2981
2982         status = nfs_ok;
2983
2984         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2985                 STATEID_VAL(&stp->st_stid.sc_stateid));
2986 out:
2987         if (fp)
2988                 put_nfs4_file(fp);
2989         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2990                 nfs4_set_claim_prev(open);
2991         /*
2992         * To finish the open response, we just need to set the rflags.
2993         */
2994         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2995         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
2996             !nfsd4_has_session(&resp->cstate))
2997                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2998
2999         return status;
3000 }
3001
3002 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3003 {
3004         if (open->op_openowner) {
3005                 struct nfs4_openowner *oo = open->op_openowner;
3006
3007                 if (!list_empty(&oo->oo_owner.so_stateids))
3008                         list_del_init(&oo->oo_close_lru);
3009                 if (oo->oo_flags & NFS4_OO_NEW) {
3010                         if (status) {
3011                                 release_openowner(oo);
3012                                 open->op_openowner = NULL;
3013                         } else
3014                                 oo->oo_flags &= ~NFS4_OO_NEW;
3015                 }
3016         }
3017         if (open->op_file)
3018                 nfsd4_free_file(open->op_file);
3019         if (open->op_stp)
3020                 nfs4_free_stateid(open->op_stp);
3021 }
3022
3023 __be32
3024 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3025             clientid_t *clid)
3026 {
3027         struct nfs4_client *clp;
3028         __be32 status;
3029
3030         nfs4_lock_state();
3031         dprintk("process_renew(%08x/%08x): starting\n", 
3032                         clid->cl_boot, clid->cl_id);
3033         status = nfserr_stale_clientid;
3034         if (STALE_CLIENTID(clid))
3035                 goto out;
3036         clp = find_confirmed_client(clid);
3037         status = nfserr_expired;
3038         if (clp == NULL) {
3039                 /* We assume the client took too long to RENEW. */
3040                 dprintk("nfsd4_renew: clientid not found!\n");
3041                 goto out;
3042         }
3043         status = nfserr_cb_path_down;
3044         if (!list_empty(&clp->cl_delegations)
3045                         && clp->cl_cb_state != NFSD4_CB_UP)
3046                 goto out;
3047         status = nfs_ok;
3048 out:
3049         nfs4_unlock_state();
3050         return status;
3051 }
3052
3053 static struct lock_manager nfsd4_manager = {
3054 };
3055
3056 static void
3057 nfsd4_end_grace(void)
3058 {
3059         dprintk("NFSD: end of grace period\n");
3060         nfsd4_recdir_purge_old();
3061         locks_end_grace(&nfsd4_manager);
3062         /*
3063          * Now that every NFSv4 client has had the chance to recover and
3064          * to see the (possibly new, possibly shorter) lease time, we
3065          * can safely set the next grace time to the current lease time:
3066          */
3067         nfsd4_grace = nfsd4_lease;
3068 }
3069
3070 static time_t
3071 nfs4_laundromat(void)
3072 {
3073         struct nfs4_client *clp;
3074         struct nfs4_openowner *oo;
3075         struct nfs4_delegation *dp;
3076         struct list_head *pos, *next, reaplist;
3077         time_t cutoff = get_seconds() - nfsd4_lease;
3078         time_t t, clientid_val = nfsd4_lease;
3079         time_t u, test_val = nfsd4_lease;
3080
3081         nfs4_lock_state();
3082
3083         dprintk("NFSD: laundromat service - starting\n");
3084         if (locks_in_grace())
3085                 nfsd4_end_grace();
3086         INIT_LIST_HEAD(&reaplist);
3087         spin_lock(&client_lock);
3088         list_for_each_safe(pos, next, &client_lru) {
3089                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3090                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3091                         t = clp->cl_time - cutoff;
3092                         if (clientid_val > t)
3093                                 clientid_val = t;
3094                         break;
3095                 }
3096                 if (atomic_read(&clp->cl_refcount)) {
3097                         dprintk("NFSD: client in use (clientid %08x)\n",
3098                                 clp->cl_clientid.cl_id);
3099                         continue;
3100                 }
3101                 unhash_client_locked(clp);
3102                 list_add(&clp->cl_lru, &reaplist);
3103         }
3104         spin_unlock(&client_lock);
3105         list_for_each_safe(pos, next, &reaplist) {
3106                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3107                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3108                         clp->cl_clientid.cl_id);
3109                 nfsd4_remove_clid_dir(clp);
3110                 expire_client(clp);
3111         }
3112         spin_lock(&recall_lock);
3113         list_for_each_safe(pos, next, &del_recall_lru) {
3114                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3115                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3116                         u = dp->dl_time - cutoff;
3117                         if (test_val > u)
3118                                 test_val = u;
3119                         break;
3120                 }
3121                 list_move(&dp->dl_recall_lru, &reaplist);
3122         }
3123         spin_unlock(&recall_lock);
3124         list_for_each_safe(pos, next, &reaplist) {
3125                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3126                 list_del_init(&dp->dl_recall_lru);
3127                 unhash_delegation(dp);
3128         }
3129         test_val = nfsd4_lease;
3130         list_for_each_safe(pos, next, &close_lru) {
3131                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3132                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3133                         u = oo->oo_time - cutoff;
3134                         if (test_val > u)
3135                                 test_val = u;
3136                         break;
3137                 }
3138                 release_openowner(oo);
3139         }
3140         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3141                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3142         nfs4_unlock_state();
3143         return clientid_val;
3144 }
3145
3146 static struct workqueue_struct *laundry_wq;
3147 static void laundromat_main(struct work_struct *);
3148 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3149
3150 static void
3151 laundromat_main(struct work_struct *not_used)
3152 {
3153         time_t t;
3154
3155         t = nfs4_laundromat();
3156         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3157         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3158 }
3159
3160 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3161 {
3162         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3163                 return nfserr_bad_stateid;
3164         return nfs_ok;
3165 }
3166
3167 static int
3168 STALE_STATEID(stateid_t *stateid)
3169 {
3170         if (stateid->si_opaque.so_clid.cl_boot == boot_time)
3171                 return 0;
3172         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3173                 STATEID_VAL(stateid));
3174         return 1;
3175 }
3176
3177 static inline int
3178 access_permit_read(unsigned long access_bmap)
3179 {
3180         return test_bit(NFS4_SHARE_ACCESS_READ, &access_bmap) ||
3181                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap) ||
3182                 test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap);
3183 }
3184
3185 static inline int
3186 access_permit_write(unsigned long access_bmap)
3187 {
3188         return test_bit(NFS4_SHARE_ACCESS_WRITE, &access_bmap) ||
3189                 test_bit(NFS4_SHARE_ACCESS_BOTH, &access_bmap);
3190 }
3191
3192 static
3193 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3194 {
3195         __be32 status = nfserr_openmode;
3196
3197         /* For lock stateid's, we test the parent open, not the lock: */
3198         if (stp->st_openstp)
3199                 stp = stp->st_openstp;
3200         if ((flags & WR_STATE) && (!access_permit_write(stp->st_access_bmap)))
3201                 goto out;
3202         if ((flags & RD_STATE) && (!access_permit_read(stp->st_access_bmap)))
3203                 goto out;
3204         status = nfs_ok;
3205 out:
3206         return status;
3207 }
3208
3209 static inline __be32
3210 check_special_stateids(svc_fh *current_fh, stateid_t *stateid, int flags)
3211 {
3212         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3213                 return nfs_ok;
3214         else if (locks_in_grace()) {
3215                 /* Answer in remaining cases depends on existence of
3216                  * conflicting state; so we must wait out the grace period. */
3217                 return nfserr_grace;
3218         } else if (flags & WR_STATE)
3219                 return nfs4_share_conflict(current_fh,
3220                                 NFS4_SHARE_DENY_WRITE);
3221         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3222                 return nfs4_share_conflict(current_fh,
3223                                 NFS4_SHARE_DENY_READ);
3224 }
3225
3226 /*
3227  * Allow READ/WRITE during grace period on recovered state only for files
3228  * that are not able to provide mandatory locking.
3229  */
3230 static inline int
3231 grace_disallows_io(struct inode *inode)
3232 {
3233         return locks_in_grace() && mandatory_lock(inode);
3234 }
3235
3236 /* Returns true iff a is later than b: */
3237 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3238 {
3239         return (s32)a->si_generation - (s32)b->si_generation > 0;
3240 }
3241
3242 static int check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3243 {
3244         /*
3245          * When sessions are used the stateid generation number is ignored
3246          * when it is zero.
3247          */
3248         if (has_session && in->si_generation == 0)
3249                 return nfs_ok;
3250
3251         if (in->si_generation == ref->si_generation)
3252                 return nfs_ok;
3253
3254         /* If the client sends us a stateid from the future, it's buggy: */
3255         if (stateid_generation_after(in, ref))
3256                 return nfserr_bad_stateid;
3257         /*
3258          * However, we could see a stateid from the past, even from a
3259          * non-buggy client.  For example, if the client sends a lock
3260          * while some IO is outstanding, the lock may bump si_generation
3261          * while the IO is still in flight.  The client could avoid that
3262          * situation by waiting for responses on all the IO requests,
3263          * but better performance may result in retrying IO that
3264          * receives an old_stateid error if requests are rarely
3265          * reordered in flight:
3266          */
3267         return nfserr_old_stateid;
3268 }
3269
3270 __be32 nfs4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3271 {
3272         struct nfs4_stid *s;
3273         struct nfs4_ol_stateid *ols;
3274         __be32 status;
3275
3276         if (STALE_STATEID(stateid))
3277                 return nfserr_stale_stateid;
3278
3279         s = find_stateid(cl, stateid);
3280         if (!s)
3281                  return nfserr_stale_stateid;
3282         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3283         if (status)
3284                 return status;
3285         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3286                 return nfs_ok;
3287         ols = openlockstateid(s);
3288         if (ols->st_stateowner->so_is_open_owner
3289             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3290                 return nfserr_bad_stateid;
3291         return nfs_ok;
3292 }
3293
3294 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s)
3295 {
3296         struct nfs4_client *cl;
3297
3298         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3299                 return nfserr_bad_stateid;
3300         if (STALE_STATEID(stateid))
3301                 return nfserr_stale_stateid;
3302         cl = find_confirmed_client(&stateid->si_opaque.so_clid);
3303         if (!cl)
3304                 return nfserr_expired;
3305         *s = find_stateid_by_type(cl, stateid, typemask);
3306         if (!*s)
3307                 return nfserr_bad_stateid;
3308         return nfs_ok;
3309
3310 }
3311
3312 /*
3313 * Checks for stateid operations
3314 */
3315 __be32
3316 nfs4_preprocess_stateid_op(struct nfsd4_compound_state *cstate,
3317                            stateid_t *stateid, int flags, struct file **filpp)
3318 {
3319         struct nfs4_stid *s;
3320         struct nfs4_ol_stateid *stp = NULL;
3321         struct nfs4_delegation *dp = NULL;
3322         struct svc_fh *current_fh = &cstate->current_fh;
3323         struct inode *ino = current_fh->fh_dentry->d_inode;
3324         __be32 status;
3325
3326         if (filpp)
3327                 *filpp = NULL;
3328
3329         if (grace_disallows_io(ino))
3330                 return nfserr_grace;
3331