nfsd: get rid of cl_recdir field
[cascardo/linux.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/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/clnt.h>
44 #include "xdr4.h"
45 #include "vfs.h"
46 #include "current_stateid.h"
47 #include "fault_inject.h"
48
49 #include "netns.h"
50
51 #define NFSDDBG_FACILITY                NFSDDBG_PROC
52
53 /* Globals */
54 time_t nfsd4_lease = 90;     /* default lease time */
55 time_t nfsd4_grace = 90;
56
57 #define all_ones {{~0,~0},~0}
58 static const stateid_t one_stateid = {
59         .si_generation = ~0,
60         .si_opaque = all_ones,
61 };
62 static const stateid_t zero_stateid = {
63         /* all fields zero */
64 };
65 static const stateid_t currentstateid = {
66         .si_generation = 1,
67 };
68
69 static u64 current_sessionid = 1;
70
71 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
72 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
73 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
74
75 /* forward declarations */
76 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
77
78 /* Locking: */
79
80 /* Currently used for almost all code touching nfsv4 state: */
81 static DEFINE_MUTEX(client_mutex);
82
83 /*
84  * Currently used for the del_recall_lru and file hash table.  In an
85  * effort to decrease the scope of the client_mutex, this spinlock may
86  * eventually cover more:
87  */
88 static DEFINE_SPINLOCK(recall_lock);
89
90 static struct kmem_cache *openowner_slab = NULL;
91 static struct kmem_cache *lockowner_slab = NULL;
92 static struct kmem_cache *file_slab = NULL;
93 static struct kmem_cache *stateid_slab = NULL;
94 static struct kmem_cache *deleg_slab = NULL;
95
96 void
97 nfs4_lock_state(void)
98 {
99         mutex_lock(&client_mutex);
100 }
101
102 static void free_session(struct kref *);
103
104 /* Must be called under the client_lock */
105 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
106 {
107         kref_put(&ses->se_ref, free_session);
108 }
109
110 static void nfsd4_get_session(struct nfsd4_session *ses)
111 {
112         kref_get(&ses->se_ref);
113 }
114
115 void
116 nfs4_unlock_state(void)
117 {
118         mutex_unlock(&client_mutex);
119 }
120
121 static inline u32
122 opaque_hashval(const void *ptr, int nbytes)
123 {
124         unsigned char *cptr = (unsigned char *) ptr;
125
126         u32 x = 0;
127         while (nbytes--) {
128                 x *= 37;
129                 x += *cptr++;
130         }
131         return x;
132 }
133
134 static struct list_head del_recall_lru;
135
136 static void nfsd4_free_file(struct nfs4_file *f)
137 {
138         kmem_cache_free(file_slab, f);
139 }
140
141 static inline void
142 put_nfs4_file(struct nfs4_file *fi)
143 {
144         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
145                 list_del(&fi->fi_hash);
146                 spin_unlock(&recall_lock);
147                 iput(fi->fi_inode);
148                 nfsd4_free_file(fi);
149         }
150 }
151
152 static inline void
153 get_nfs4_file(struct nfs4_file *fi)
154 {
155         atomic_inc(&fi->fi_ref);
156 }
157
158 static int num_delegations;
159 unsigned int max_delegations;
160
161 /*
162  * Open owner state (share locks)
163  */
164
165 /* hash tables for lock and open owners */
166 #define OWNER_HASH_BITS              8
167 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
168 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
169
170 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
171 {
172         unsigned int ret;
173
174         ret = opaque_hashval(ownername->data, ownername->len);
175         ret += clientid;
176         return ret & OWNER_HASH_MASK;
177 }
178
179 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
180
181 /* hash table for nfs4_file */
182 #define FILE_HASH_BITS                   8
183 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
184
185 static unsigned int file_hashval(struct inode *ino)
186 {
187         /* XXX: why are we hashing on inode pointer, anyway? */
188         return hash_ptr(ino, FILE_HASH_BITS);
189 }
190
191 static struct list_head file_hashtbl[FILE_HASH_SIZE];
192
193 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
194 {
195         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
196         atomic_inc(&fp->fi_access[oflag]);
197 }
198
199 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
200 {
201         if (oflag == O_RDWR) {
202                 __nfs4_file_get_access(fp, O_RDONLY);
203                 __nfs4_file_get_access(fp, O_WRONLY);
204         } else
205                 __nfs4_file_get_access(fp, oflag);
206 }
207
208 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
209 {
210         if (fp->fi_fds[oflag]) {
211                 fput(fp->fi_fds[oflag]);
212                 fp->fi_fds[oflag] = NULL;
213         }
214 }
215
216 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
217 {
218         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
219                 nfs4_file_put_fd(fp, oflag);
220                 /*
221                  * It's also safe to get rid of the RDWR open *if*
222                  * we no longer have need of the other kind of access
223                  * or if we already have the other kind of open:
224                  */
225                 if (fp->fi_fds[1-oflag]
226                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
227                         nfs4_file_put_fd(fp, O_RDWR);
228         }
229 }
230
231 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
232 {
233         if (oflag == O_RDWR) {
234                 __nfs4_file_put_access(fp, O_RDONLY);
235                 __nfs4_file_put_access(fp, O_WRONLY);
236         } else
237                 __nfs4_file_put_access(fp, oflag);
238 }
239
240 static inline int get_new_stid(struct nfs4_stid *stid)
241 {
242         static int min_stateid = 0;
243         struct idr *stateids = &stid->sc_client->cl_stateids;
244         int new_stid;
245         int error;
246
247         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
248         /*
249          * Note: the necessary preallocation was done in
250          * nfs4_alloc_stateid().  The idr code caps the number of
251          * preallocations that can exist at a time, but the state lock
252          * prevents anyone from using ours before we get here:
253          */
254         BUG_ON(error);
255         /*
256          * It shouldn't be a problem to reuse an opaque stateid value.
257          * I don't think it is for 4.1.  But with 4.0 I worry that, for
258          * example, a stray write retransmission could be accepted by
259          * the server when it should have been rejected.  Therefore,
260          * adopt a trick from the sctp code to attempt to maximize the
261          * amount of time until an id is reused, by ensuring they always
262          * "increase" (mod INT_MAX):
263          */
264
265         min_stateid = new_stid+1;
266         if (min_stateid == INT_MAX)
267                 min_stateid = 0;
268         return new_stid;
269 }
270
271 static void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
272 {
273         stateid_t *s = &stid->sc_stateid;
274         int new_id;
275
276         stid->sc_type = type;
277         stid->sc_client = cl;
278         s->si_opaque.so_clid = cl->cl_clientid;
279         new_id = get_new_stid(stid);
280         s->si_opaque.so_id = (u32)new_id;
281         /* Will be incremented before return to client: */
282         s->si_generation = 0;
283 }
284
285 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
286 {
287         struct idr *stateids = &cl->cl_stateids;
288
289         if (!idr_pre_get(stateids, GFP_KERNEL))
290                 return NULL;
291         /*
292          * Note: if we fail here (or any time between now and the time
293          * we actually get the new idr), we won't need to undo the idr
294          * preallocation, since the idr code caps the number of
295          * preallocated entries.
296          */
297         return kmem_cache_alloc(slab, GFP_KERNEL);
298 }
299
300 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
301 {
302         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
303 }
304
305 static struct nfs4_delegation *
306 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
307 {
308         struct nfs4_delegation *dp;
309         struct nfs4_file *fp = stp->st_file;
310
311         dprintk("NFSD alloc_init_deleg\n");
312         /*
313          * Major work on the lease subsystem (for example, to support
314          * calbacks on stat) will be required before we can support
315          * write delegations properly.
316          */
317         if (type != NFS4_OPEN_DELEGATE_READ)
318                 return NULL;
319         if (fp->fi_had_conflict)
320                 return NULL;
321         if (num_delegations > max_delegations)
322                 return NULL;
323         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
324         if (dp == NULL)
325                 return dp;
326         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
327         /*
328          * delegation seqid's are never incremented.  The 4.1 special
329          * meaning of seqid 0 isn't meaningful, really, but let's avoid
330          * 0 anyway just for consistency and use 1:
331          */
332         dp->dl_stid.sc_stateid.si_generation = 1;
333         num_delegations++;
334         INIT_LIST_HEAD(&dp->dl_perfile);
335         INIT_LIST_HEAD(&dp->dl_perclnt);
336         INIT_LIST_HEAD(&dp->dl_recall_lru);
337         get_nfs4_file(fp);
338         dp->dl_file = fp;
339         dp->dl_type = type;
340         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
341         dp->dl_time = 0;
342         atomic_set(&dp->dl_count, 1);
343         nfsd4_init_callback(&dp->dl_recall);
344         return dp;
345 }
346
347 void
348 nfs4_put_delegation(struct nfs4_delegation *dp)
349 {
350         if (atomic_dec_and_test(&dp->dl_count)) {
351                 dprintk("NFSD: freeing dp %p\n",dp);
352                 put_nfs4_file(dp->dl_file);
353                 kmem_cache_free(deleg_slab, dp);
354                 num_delegations--;
355         }
356 }
357
358 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
359 {
360         if (atomic_dec_and_test(&fp->fi_delegees)) {
361                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
362                 fp->fi_lease = NULL;
363                 fput(fp->fi_deleg_file);
364                 fp->fi_deleg_file = NULL;
365         }
366 }
367
368 static void unhash_stid(struct nfs4_stid *s)
369 {
370         struct idr *stateids = &s->sc_client->cl_stateids;
371
372         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
373 }
374
375 /* Called under the state lock. */
376 static void
377 unhash_delegation(struct nfs4_delegation *dp)
378 {
379         unhash_stid(&dp->dl_stid);
380         list_del_init(&dp->dl_perclnt);
381         spin_lock(&recall_lock);
382         list_del_init(&dp->dl_perfile);
383         list_del_init(&dp->dl_recall_lru);
384         spin_unlock(&recall_lock);
385         nfs4_put_deleg_lease(dp->dl_file);
386         nfs4_put_delegation(dp);
387 }
388
389 /* 
390  * SETCLIENTID state 
391  */
392
393 /* client_lock protects the client lru list and session hash table */
394 static DEFINE_SPINLOCK(client_lock);
395
396 /* Hash tables for nfs4_clientid state */
397 #define CLIENT_HASH_BITS                 4
398 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
399 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
400
401 static unsigned int clientid_hashval(u32 id)
402 {
403         return id & CLIENT_HASH_MASK;
404 }
405
406 static unsigned int clientstr_hashval(const char *name)
407 {
408         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
409 }
410
411 /*
412  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
413  * used in reboot/reset lease grace period processing
414  *
415  * conf_id_hashtbl[], and conf_name_tree hold confirmed
416  * setclientid_confirmed info. 
417  *
418  * unconf_id_hashtbl[] and unconf_name_tree hold unconfirmed
419  * setclientid info.
420  *
421  * client_lru holds client queue ordered by nfs4_client.cl_time
422  * for lease renewal.
423  *
424  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
425  * for last close replay.
426  *
427  * All of the above fields are protected by the client_mutex.
428  */
429 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
430 static int reclaim_str_hashtbl_size = 0;
431 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
432 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
433 static struct rb_root conf_name_tree;
434 static struct rb_root unconf_name_tree;
435 static struct list_head client_lru;
436 static struct list_head close_lru;
437
438 /*
439  * We store the NONE, READ, WRITE, and BOTH bits separately in the
440  * st_{access,deny}_bmap field of the stateid, in order to track not
441  * only what share bits are currently in force, but also what
442  * combinations of share bits previous opens have used.  This allows us
443  * to enforce the recommendation of rfc 3530 14.2.19 that the server
444  * return an error if the client attempt to downgrade to a combination
445  * of share bits not explicable by closing some of its previous opens.
446  *
447  * XXX: This enforcement is actually incomplete, since we don't keep
448  * track of access/deny bit combinations; so, e.g., we allow:
449  *
450  *      OPEN allow read, deny write
451  *      OPEN allow both, deny none
452  *      DOWNGRADE allow read, deny none
453  *
454  * which we should reject.
455  */
456 static unsigned int
457 bmap_to_share_mode(unsigned long bmap) {
458         int i;
459         unsigned int access = 0;
460
461         for (i = 1; i < 4; i++) {
462                 if (test_bit(i, &bmap))
463                         access |= i;
464         }
465         return access;
466 }
467
468 static bool
469 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
470         unsigned int access, deny;
471
472         access = bmap_to_share_mode(stp->st_access_bmap);
473         deny = bmap_to_share_mode(stp->st_deny_bmap);
474         if ((access & open->op_share_deny) || (deny & open->op_share_access))
475                 return false;
476         return true;
477 }
478
479 /* set share access for a given stateid */
480 static inline void
481 set_access(u32 access, struct nfs4_ol_stateid *stp)
482 {
483         __set_bit(access, &stp->st_access_bmap);
484 }
485
486 /* clear share access for a given stateid */
487 static inline void
488 clear_access(u32 access, struct nfs4_ol_stateid *stp)
489 {
490         __clear_bit(access, &stp->st_access_bmap);
491 }
492
493 /* test whether a given stateid has access */
494 static inline bool
495 test_access(u32 access, struct nfs4_ol_stateid *stp)
496 {
497         return test_bit(access, &stp->st_access_bmap);
498 }
499
500 /* set share deny for a given stateid */
501 static inline void
502 set_deny(u32 access, struct nfs4_ol_stateid *stp)
503 {
504         __set_bit(access, &stp->st_deny_bmap);
505 }
506
507 /* clear share deny for a given stateid */
508 static inline void
509 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
510 {
511         __clear_bit(access, &stp->st_deny_bmap);
512 }
513
514 /* test whether a given stateid is denying specific access */
515 static inline bool
516 test_deny(u32 access, struct nfs4_ol_stateid *stp)
517 {
518         return test_bit(access, &stp->st_deny_bmap);
519 }
520
521 static int nfs4_access_to_omode(u32 access)
522 {
523         switch (access & NFS4_SHARE_ACCESS_BOTH) {
524         case NFS4_SHARE_ACCESS_READ:
525                 return O_RDONLY;
526         case NFS4_SHARE_ACCESS_WRITE:
527                 return O_WRONLY;
528         case NFS4_SHARE_ACCESS_BOTH:
529                 return O_RDWR;
530         }
531         BUG();
532 }
533
534 /* release all access and file references for a given stateid */
535 static void
536 release_all_access(struct nfs4_ol_stateid *stp)
537 {
538         int i;
539
540         for (i = 1; i < 4; i++) {
541                 if (test_access(i, stp))
542                         nfs4_file_put_access(stp->st_file,
543                                              nfs4_access_to_omode(i));
544                 clear_access(i, stp);
545         }
546 }
547
548 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
549 {
550         list_del(&stp->st_perfile);
551         list_del(&stp->st_perstateowner);
552 }
553
554 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
555 {
556         release_all_access(stp);
557         put_nfs4_file(stp->st_file);
558         stp->st_file = NULL;
559 }
560
561 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
562 {
563         kmem_cache_free(stateid_slab, stp);
564 }
565
566 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
567 {
568         struct file *file;
569
570         unhash_generic_stateid(stp);
571         unhash_stid(&stp->st_stid);
572         file = find_any_file(stp->st_file);
573         if (file)
574                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
575         close_generic_stateid(stp);
576         free_generic_stateid(stp);
577 }
578
579 static void unhash_lockowner(struct nfs4_lockowner *lo)
580 {
581         struct nfs4_ol_stateid *stp;
582
583         list_del(&lo->lo_owner.so_strhash);
584         list_del(&lo->lo_perstateid);
585         list_del(&lo->lo_owner_ino_hash);
586         while (!list_empty(&lo->lo_owner.so_stateids)) {
587                 stp = list_first_entry(&lo->lo_owner.so_stateids,
588                                 struct nfs4_ol_stateid, st_perstateowner);
589                 release_lock_stateid(stp);
590         }
591 }
592
593 static void release_lockowner(struct nfs4_lockowner *lo)
594 {
595         unhash_lockowner(lo);
596         nfs4_free_lockowner(lo);
597 }
598
599 static void
600 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
601 {
602         struct nfs4_lockowner *lo;
603
604         while (!list_empty(&open_stp->st_lockowners)) {
605                 lo = list_entry(open_stp->st_lockowners.next,
606                                 struct nfs4_lockowner, lo_perstateid);
607                 release_lockowner(lo);
608         }
609 }
610
611 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
612 {
613         unhash_generic_stateid(stp);
614         release_stateid_lockowners(stp);
615         close_generic_stateid(stp);
616 }
617
618 static void release_open_stateid(struct nfs4_ol_stateid *stp)
619 {
620         unhash_open_stateid(stp);
621         unhash_stid(&stp->st_stid);
622         free_generic_stateid(stp);
623 }
624
625 static void unhash_openowner(struct nfs4_openowner *oo)
626 {
627         struct nfs4_ol_stateid *stp;
628
629         list_del(&oo->oo_owner.so_strhash);
630         list_del(&oo->oo_perclient);
631         while (!list_empty(&oo->oo_owner.so_stateids)) {
632                 stp = list_first_entry(&oo->oo_owner.so_stateids,
633                                 struct nfs4_ol_stateid, st_perstateowner);
634                 release_open_stateid(stp);
635         }
636 }
637
638 static void release_last_closed_stateid(struct nfs4_openowner *oo)
639 {
640         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
641
642         if (s) {
643                 unhash_stid(&s->st_stid);
644                 free_generic_stateid(s);
645                 oo->oo_last_closed_stid = NULL;
646         }
647 }
648
649 static void release_openowner(struct nfs4_openowner *oo)
650 {
651         unhash_openowner(oo);
652         list_del(&oo->oo_close_lru);
653         release_last_closed_stateid(oo);
654         nfs4_free_openowner(oo);
655 }
656
657 #define SESSION_HASH_SIZE       512
658 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
659
660 static inline int
661 hash_sessionid(struct nfs4_sessionid *sessionid)
662 {
663         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
664
665         return sid->sequence % SESSION_HASH_SIZE;
666 }
667
668 #ifdef NFSD_DEBUG
669 static inline void
670 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
671 {
672         u32 *ptr = (u32 *)(&sessionid->data[0]);
673         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
674 }
675 #else
676 static inline void
677 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
678 {
679 }
680 #endif
681
682
683 static void
684 gen_sessionid(struct nfsd4_session *ses)
685 {
686         struct nfs4_client *clp = ses->se_client;
687         struct nfsd4_sessionid *sid;
688
689         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
690         sid->clientid = clp->cl_clientid;
691         sid->sequence = current_sessionid++;
692         sid->reserved = 0;
693 }
694
695 /*
696  * The protocol defines ca_maxresponssize_cached to include the size of
697  * the rpc header, but all we need to cache is the data starting after
698  * the end of the initial SEQUENCE operation--the rest we regenerate
699  * each time.  Therefore we can advertise a ca_maxresponssize_cached
700  * value that is the number of bytes in our cache plus a few additional
701  * bytes.  In order to stay on the safe side, and not promise more than
702  * we can cache, those additional bytes must be the minimum possible: 24
703  * bytes of rpc header (xid through accept state, with AUTH_NULL
704  * verifier), 12 for the compound header (with zero-length tag), and 44
705  * for the SEQUENCE op response:
706  */
707 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
708
709 static void
710 free_session_slots(struct nfsd4_session *ses)
711 {
712         int i;
713
714         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
715                 kfree(ses->se_slots[i]);
716 }
717
718 /*
719  * We don't actually need to cache the rpc and session headers, so we
720  * can allocate a little less for each slot:
721  */
722 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
723 {
724         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
725 }
726
727 static int nfsd4_sanitize_slot_size(u32 size)
728 {
729         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
730         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
731
732         return size;
733 }
734
735 /*
736  * XXX: If we run out of reserved DRC memory we could (up to a point)
737  * re-negotiate active sessions and reduce their slot usage to make
738  * room for new connections. For now we just fail the create session.
739  */
740 static int nfsd4_get_drc_mem(int slotsize, u32 num)
741 {
742         int avail;
743
744         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
745
746         spin_lock(&nfsd_drc_lock);
747         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
748                         nfsd_drc_max_mem - nfsd_drc_mem_used);
749         num = min_t(int, num, avail / slotsize);
750         nfsd_drc_mem_used += num * slotsize;
751         spin_unlock(&nfsd_drc_lock);
752
753         return num;
754 }
755
756 static void nfsd4_put_drc_mem(int slotsize, int num)
757 {
758         spin_lock(&nfsd_drc_lock);
759         nfsd_drc_mem_used -= slotsize * num;
760         spin_unlock(&nfsd_drc_lock);
761 }
762
763 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
764 {
765         struct nfsd4_session *new;
766         int mem, i;
767
768         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
769                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
770         mem = numslots * sizeof(struct nfsd4_slot *);
771
772         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
773         if (!new)
774                 return NULL;
775         /* allocate each struct nfsd4_slot and data cache in one piece */
776         for (i = 0; i < numslots; i++) {
777                 mem = sizeof(struct nfsd4_slot) + slotsize;
778                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
779                 if (!new->se_slots[i])
780                         goto out_free;
781         }
782         return new;
783 out_free:
784         while (i--)
785                 kfree(new->se_slots[i]);
786         kfree(new);
787         return NULL;
788 }
789
790 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
791 {
792         u32 maxrpc = nfsd_serv->sv_max_mesg;
793
794         new->maxreqs = numslots;
795         new->maxresp_cached = min_t(u32, req->maxresp_cached,
796                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
797         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
798         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
799         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
800 }
801
802 static void free_conn(struct nfsd4_conn *c)
803 {
804         svc_xprt_put(c->cn_xprt);
805         kfree(c);
806 }
807
808 static void nfsd4_conn_lost(struct svc_xpt_user *u)
809 {
810         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
811         struct nfs4_client *clp = c->cn_session->se_client;
812
813         spin_lock(&clp->cl_lock);
814         if (!list_empty(&c->cn_persession)) {
815                 list_del(&c->cn_persession);
816                 free_conn(c);
817         }
818         spin_unlock(&clp->cl_lock);
819         nfsd4_probe_callback(clp);
820 }
821
822 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
823 {
824         struct nfsd4_conn *conn;
825
826         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
827         if (!conn)
828                 return NULL;
829         svc_xprt_get(rqstp->rq_xprt);
830         conn->cn_xprt = rqstp->rq_xprt;
831         conn->cn_flags = flags;
832         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
833         return conn;
834 }
835
836 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
837 {
838         conn->cn_session = ses;
839         list_add(&conn->cn_persession, &ses->se_conns);
840 }
841
842 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
843 {
844         struct nfs4_client *clp = ses->se_client;
845
846         spin_lock(&clp->cl_lock);
847         __nfsd4_hash_conn(conn, ses);
848         spin_unlock(&clp->cl_lock);
849 }
850
851 static int nfsd4_register_conn(struct nfsd4_conn *conn)
852 {
853         conn->cn_xpt_user.callback = nfsd4_conn_lost;
854         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
855 }
856
857 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
858 {
859         int ret;
860
861         nfsd4_hash_conn(conn, ses);
862         ret = nfsd4_register_conn(conn);
863         if (ret)
864                 /* oops; xprt is already down: */
865                 nfsd4_conn_lost(&conn->cn_xpt_user);
866         if (conn->cn_flags & NFS4_CDFC4_BACK) {
867                 /* callback channel may be back up */
868                 nfsd4_probe_callback(ses->se_client);
869         }
870 }
871
872 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
873 {
874         u32 dir = NFS4_CDFC4_FORE;
875
876         if (cses->flags & SESSION4_BACK_CHAN)
877                 dir |= NFS4_CDFC4_BACK;
878         return alloc_conn(rqstp, dir);
879 }
880
881 /* must be called under client_lock */
882 static void nfsd4_del_conns(struct nfsd4_session *s)
883 {
884         struct nfs4_client *clp = s->se_client;
885         struct nfsd4_conn *c;
886
887         spin_lock(&clp->cl_lock);
888         while (!list_empty(&s->se_conns)) {
889                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
890                 list_del_init(&c->cn_persession);
891                 spin_unlock(&clp->cl_lock);
892
893                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
894                 free_conn(c);
895
896                 spin_lock(&clp->cl_lock);
897         }
898         spin_unlock(&clp->cl_lock);
899 }
900
901 static void __free_session(struct nfsd4_session *ses)
902 {
903         nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
904         free_session_slots(ses);
905         kfree(ses);
906 }
907
908 static void free_session(struct kref *kref)
909 {
910         struct nfsd4_session *ses;
911
912         lockdep_assert_held(&client_lock);
913         ses = container_of(kref, struct nfsd4_session, se_ref);
914         nfsd4_del_conns(ses);
915         __free_session(ses);
916 }
917
918 void nfsd4_put_session(struct nfsd4_session *ses)
919 {
920         spin_lock(&client_lock);
921         nfsd4_put_session_locked(ses);
922         spin_unlock(&client_lock);
923 }
924
925 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan)
926 {
927         struct nfsd4_session *new;
928         int numslots, slotsize;
929         /*
930          * Note decreasing slot size below client's request may
931          * make it difficult for client to function correctly, whereas
932          * decreasing the number of slots will (just?) affect
933          * performance.  When short on memory we therefore prefer to
934          * decrease number of slots instead of their size.
935          */
936         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
937         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
938         if (numslots < 1)
939                 return NULL;
940
941         new = __alloc_session(slotsize, numslots);
942         if (!new) {
943                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
944                 return NULL;
945         }
946         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
947         return new;
948 }
949
950 void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
951 {
952         int idx;
953
954         new->se_client = clp;
955         gen_sessionid(new);
956
957         INIT_LIST_HEAD(&new->se_conns);
958
959         new->se_cb_seq_nr = 1;
960         new->se_flags = cses->flags;
961         new->se_cb_prog = cses->callback_prog;
962         new->se_cb_sec = cses->cb_sec;
963         kref_init(&new->se_ref);
964         idx = hash_sessionid(&new->se_sessionid);
965         spin_lock(&client_lock);
966         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
967         spin_lock(&clp->cl_lock);
968         list_add(&new->se_perclnt, &clp->cl_sessions);
969         spin_unlock(&clp->cl_lock);
970         spin_unlock(&client_lock);
971
972         if (cses->flags & SESSION4_BACK_CHAN) {
973                 struct sockaddr *sa = svc_addr(rqstp);
974                 /*
975                  * This is a little silly; with sessions there's no real
976                  * use for the callback address.  Use the peer address
977                  * as a reasonable default for now, but consider fixing
978                  * the rpc client not to require an address in the
979                  * future:
980                  */
981                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
982                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
983         }
984 }
985
986 /* caller must hold client_lock */
987 static struct nfsd4_session *
988 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
989 {
990         struct nfsd4_session *elem;
991         int idx;
992
993         dump_sessionid(__func__, sessionid);
994         idx = hash_sessionid(sessionid);
995         /* Search in the appropriate list */
996         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
997                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
998                             NFS4_MAX_SESSIONID_LEN)) {
999                         return elem;
1000                 }
1001         }
1002
1003         dprintk("%s: session not found\n", __func__);
1004         return NULL;
1005 }
1006
1007 /* caller must hold client_lock */
1008 static void
1009 unhash_session(struct nfsd4_session *ses)
1010 {
1011         list_del(&ses->se_hash);
1012         spin_lock(&ses->se_client->cl_lock);
1013         list_del(&ses->se_perclnt);
1014         spin_unlock(&ses->se_client->cl_lock);
1015 }
1016
1017 /* must be called under the client_lock */
1018 static inline void
1019 renew_client_locked(struct nfs4_client *clp)
1020 {
1021         if (is_client_expired(clp)) {
1022                 WARN_ON(1);
1023                 printk("%s: client (clientid %08x/%08x) already expired\n",
1024                         __func__,
1025                         clp->cl_clientid.cl_boot,
1026                         clp->cl_clientid.cl_id);
1027                 return;
1028         }
1029
1030         dprintk("renewing client (clientid %08x/%08x)\n", 
1031                         clp->cl_clientid.cl_boot, 
1032                         clp->cl_clientid.cl_id);
1033         list_move_tail(&clp->cl_lru, &client_lru);
1034         clp->cl_time = get_seconds();
1035 }
1036
1037 static inline void
1038 renew_client(struct nfs4_client *clp)
1039 {
1040         spin_lock(&client_lock);
1041         renew_client_locked(clp);
1042         spin_unlock(&client_lock);
1043 }
1044
1045 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1046 static int
1047 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1048 {
1049         if (clid->cl_boot == nn->boot_time)
1050                 return 0;
1051         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1052                 clid->cl_boot, clid->cl_id, nn->boot_time);
1053         return 1;
1054 }
1055
1056 /* 
1057  * XXX Should we use a slab cache ?
1058  * This type of memory management is somewhat inefficient, but we use it
1059  * anyway since SETCLIENTID is not a common operation.
1060  */
1061 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1062 {
1063         struct nfs4_client *clp;
1064
1065         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1066         if (clp == NULL)
1067                 return NULL;
1068         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1069         if (clp->cl_name.data == NULL) {
1070                 kfree(clp);
1071                 return NULL;
1072         }
1073         clp->cl_name.len = name.len;
1074         return clp;
1075 }
1076
1077 static inline void
1078 free_client(struct nfs4_client *clp)
1079 {
1080         lockdep_assert_held(&client_lock);
1081         while (!list_empty(&clp->cl_sessions)) {
1082                 struct nfsd4_session *ses;
1083                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1084                                 se_perclnt);
1085                 list_del(&ses->se_perclnt);
1086                 nfsd4_put_session_locked(ses);
1087         }
1088         free_svc_cred(&clp->cl_cred);
1089         kfree(clp->cl_name.data);
1090         kfree(clp);
1091 }
1092
1093 void
1094 release_session_client(struct nfsd4_session *session)
1095 {
1096         struct nfs4_client *clp = session->se_client;
1097
1098         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
1099                 return;
1100         if (is_client_expired(clp)) {
1101                 free_client(clp);
1102                 session->se_client = NULL;
1103         } else
1104                 renew_client_locked(clp);
1105         spin_unlock(&client_lock);
1106 }
1107
1108 /* must be called under the client_lock */
1109 static inline void
1110 unhash_client_locked(struct nfs4_client *clp)
1111 {
1112         struct nfsd4_session *ses;
1113
1114         mark_client_expired(clp);
1115         list_del(&clp->cl_lru);
1116         spin_lock(&clp->cl_lock);
1117         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1118                 list_del_init(&ses->se_hash);
1119         spin_unlock(&clp->cl_lock);
1120 }
1121
1122 static void
1123 destroy_client(struct nfs4_client *clp)
1124 {
1125         struct nfs4_openowner *oo;
1126         struct nfs4_delegation *dp;
1127         struct list_head reaplist;
1128
1129         INIT_LIST_HEAD(&reaplist);
1130         spin_lock(&recall_lock);
1131         while (!list_empty(&clp->cl_delegations)) {
1132                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1133                 list_del_init(&dp->dl_perclnt);
1134                 list_move(&dp->dl_recall_lru, &reaplist);
1135         }
1136         spin_unlock(&recall_lock);
1137         while (!list_empty(&reaplist)) {
1138                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1139                 unhash_delegation(dp);
1140         }
1141         while (!list_empty(&clp->cl_openowners)) {
1142                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1143                 release_openowner(oo);
1144         }
1145         nfsd4_shutdown_callback(clp);
1146         if (clp->cl_cb_conn.cb_xprt)
1147                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1148         list_del(&clp->cl_idhash);
1149         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1150                 rb_erase(&clp->cl_namenode, &conf_name_tree);
1151         else
1152                 rb_erase(&clp->cl_namenode, &unconf_name_tree);
1153         spin_lock(&client_lock);
1154         unhash_client_locked(clp);
1155         if (atomic_read(&clp->cl_refcount) == 0)
1156                 free_client(clp);
1157         spin_unlock(&client_lock);
1158 }
1159
1160 static void expire_client(struct nfs4_client *clp)
1161 {
1162         nfsd4_client_record_remove(clp);
1163         destroy_client(clp);
1164 }
1165
1166 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1167 {
1168         memcpy(target->cl_verifier.data, source->data,
1169                         sizeof(target->cl_verifier.data));
1170 }
1171
1172 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1173 {
1174         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1175         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1176 }
1177
1178 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1179 {
1180         if (source->cr_principal) {
1181                 target->cr_principal =
1182                                 kstrdup(source->cr_principal, GFP_KERNEL);
1183                 if (target->cr_principal == NULL)
1184                         return -ENOMEM;
1185         } else
1186                 target->cr_principal = NULL;
1187         target->cr_flavor = source->cr_flavor;
1188         target->cr_uid = source->cr_uid;
1189         target->cr_gid = source->cr_gid;
1190         target->cr_group_info = source->cr_group_info;
1191         get_group_info(target->cr_group_info);
1192         return 0;
1193 }
1194
1195 static long long
1196 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1197 {
1198         long long res;
1199
1200         res = o1->len - o2->len;
1201         if (res)
1202                 return res;
1203         return (long long)memcmp(o1->data, o2->data, o1->len);
1204 }
1205
1206 static int same_name(const char *n1, const char *n2)
1207 {
1208         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1209 }
1210
1211 static int
1212 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1213 {
1214         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1215 }
1216
1217 static int
1218 same_clid(clientid_t *cl1, clientid_t *cl2)
1219 {
1220         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1221 }
1222
1223 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1224 {
1225         int i;
1226
1227         if (g1->ngroups != g2->ngroups)
1228                 return false;
1229         for (i=0; i<g1->ngroups; i++)
1230                 if (GROUP_AT(g1, i) != GROUP_AT(g2, i))
1231                         return false;
1232         return true;
1233 }
1234
1235 /*
1236  * RFC 3530 language requires clid_inuse be returned when the
1237  * "principal" associated with a requests differs from that previously
1238  * used.  We use uid, gid's, and gss principal string as our best
1239  * approximation.  We also don't want to allow non-gss use of a client
1240  * established using gss: in theory cr_principal should catch that
1241  * change, but in practice cr_principal can be null even in the gss case
1242  * since gssd doesn't always pass down a principal string.
1243  */
1244 static bool is_gss_cred(struct svc_cred *cr)
1245 {
1246         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1247         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1248 }
1249
1250
1251 static bool
1252 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1253 {
1254         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1255                 || (cr1->cr_uid != cr2->cr_uid)
1256                 || (cr1->cr_gid != cr2->cr_gid)
1257                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1258                 return false;
1259         if (cr1->cr_principal == cr2->cr_principal)
1260                 return true;
1261         if (!cr1->cr_principal || !cr2->cr_principal)
1262                 return false;
1263         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1264 }
1265
1266 static void gen_clid(struct nfs4_client *clp)
1267 {
1268         static u32 current_clientid = 1;
1269         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
1270
1271         clp->cl_clientid.cl_boot = nn->boot_time;
1272         clp->cl_clientid.cl_id = current_clientid++; 
1273 }
1274
1275 static void gen_confirm(struct nfs4_client *clp)
1276 {
1277         __be32 verf[2];
1278         static u32 i;
1279
1280         verf[0] = (__be32)get_seconds();
1281         verf[1] = (__be32)i++;
1282         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1283 }
1284
1285 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1286 {
1287         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1288 }
1289
1290 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1291 {
1292         struct nfs4_stid *s;
1293
1294         s = find_stateid(cl, t);
1295         if (!s)
1296                 return NULL;
1297         if (typemask & s->sc_type)
1298                 return s;
1299         return NULL;
1300 }
1301
1302 static struct nfs4_client *create_client(struct xdr_netobj name,
1303                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1304 {
1305         struct nfs4_client *clp;
1306         struct sockaddr *sa = svc_addr(rqstp);
1307         int ret;
1308
1309         clp = alloc_client(name);
1310         if (clp == NULL)
1311                 return NULL;
1312
1313         INIT_LIST_HEAD(&clp->cl_sessions);
1314         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1315         if (ret) {
1316                 spin_lock(&client_lock);
1317                 free_client(clp);
1318                 spin_unlock(&client_lock);
1319                 return NULL;
1320         }
1321         idr_init(&clp->cl_stateids);
1322         atomic_set(&clp->cl_refcount, 0);
1323         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1324         INIT_LIST_HEAD(&clp->cl_idhash);
1325         INIT_LIST_HEAD(&clp->cl_openowners);
1326         INIT_LIST_HEAD(&clp->cl_delegations);
1327         INIT_LIST_HEAD(&clp->cl_lru);
1328         INIT_LIST_HEAD(&clp->cl_callbacks);
1329         spin_lock_init(&clp->cl_lock);
1330         nfsd4_init_callback(&clp->cl_cb_null);
1331         clp->cl_time = get_seconds();
1332         clear_bit(0, &clp->cl_cb_slot_busy);
1333         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1334         copy_verf(clp, verf);
1335         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1336         gen_confirm(clp);
1337         clp->cl_cb_session = NULL;
1338         return clp;
1339 }
1340
1341 static void
1342 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1343 {
1344         struct rb_node **new = &(root->rb_node), *parent = NULL;
1345         struct nfs4_client *clp;
1346
1347         while (*new) {
1348                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1349                 parent = *new;
1350
1351                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1352                         new = &((*new)->rb_left);
1353                 else
1354                         new = &((*new)->rb_right);
1355         }
1356
1357         rb_link_node(&new_clp->cl_namenode, parent, new);
1358         rb_insert_color(&new_clp->cl_namenode, root);
1359 }
1360
1361 static struct nfs4_client *
1362 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1363 {
1364         long long cmp;
1365         struct rb_node *node = root->rb_node;
1366         struct nfs4_client *clp;
1367
1368         while (node) {
1369                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1370                 cmp = compare_blob(&clp->cl_name, name);
1371                 if (cmp > 0)
1372                         node = node->rb_left;
1373                 else if (cmp < 0)
1374                         node = node->rb_right;
1375                 else
1376                         return clp;
1377         }
1378         return NULL;
1379 }
1380
1381 static void
1382 add_to_unconfirmed(struct nfs4_client *clp)
1383 {
1384         unsigned int idhashval;
1385
1386         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1387         add_clp_to_name_tree(clp, &unconf_name_tree);
1388         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1389         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1390         renew_client(clp);
1391 }
1392
1393 static void
1394 move_to_confirmed(struct nfs4_client *clp)
1395 {
1396         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1397
1398         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1399         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1400         rb_erase(&clp->cl_namenode, &unconf_name_tree);
1401         add_clp_to_name_tree(clp, &conf_name_tree);
1402         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1403         renew_client(clp);
1404 }
1405
1406 static struct nfs4_client *
1407 find_confirmed_client(clientid_t *clid, bool sessions)
1408 {
1409         struct nfs4_client *clp;
1410         unsigned int idhashval = clientid_hashval(clid->cl_id);
1411
1412         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1413                 if (same_clid(&clp->cl_clientid, clid)) {
1414                         if ((bool)clp->cl_minorversion != sessions)
1415                                 return NULL;
1416                         renew_client(clp);
1417                         return clp;
1418                 }
1419         }
1420         return NULL;
1421 }
1422
1423 static struct nfs4_client *
1424 find_unconfirmed_client(clientid_t *clid, bool sessions)
1425 {
1426         struct nfs4_client *clp;
1427         unsigned int idhashval = clientid_hashval(clid->cl_id);
1428
1429         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1430                 if (same_clid(&clp->cl_clientid, clid)) {
1431                         if ((bool)clp->cl_minorversion != sessions)
1432                                 return NULL;
1433                         return clp;
1434                 }
1435         }
1436         return NULL;
1437 }
1438
1439 static bool clp_used_exchangeid(struct nfs4_client *clp)
1440 {
1441         return clp->cl_exchange_flags != 0;
1442
1443
1444 static struct nfs4_client *
1445 find_confirmed_client_by_name(struct xdr_netobj *name)
1446 {
1447         return find_clp_in_name_tree(name, &conf_name_tree);
1448 }
1449
1450 static struct nfs4_client *
1451 find_unconfirmed_client_by_name(struct xdr_netobj *name)
1452 {
1453         return find_clp_in_name_tree(name, &unconf_name_tree);
1454 }
1455
1456 static void
1457 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1458 {
1459         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1460         struct sockaddr *sa = svc_addr(rqstp);
1461         u32 scopeid = rpc_get_scope_id(sa);
1462         unsigned short expected_family;
1463
1464         /* Currently, we only support tcp and tcp6 for the callback channel */
1465         if (se->se_callback_netid_len == 3 &&
1466             !memcmp(se->se_callback_netid_val, "tcp", 3))
1467                 expected_family = AF_INET;
1468         else if (se->se_callback_netid_len == 4 &&
1469                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1470                 expected_family = AF_INET6;
1471         else
1472                 goto out_err;
1473
1474         conn->cb_addrlen = rpc_uaddr2sockaddr(&init_net, se->se_callback_addr_val,
1475                                             se->se_callback_addr_len,
1476                                             (struct sockaddr *)&conn->cb_addr,
1477                                             sizeof(conn->cb_addr));
1478
1479         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1480                 goto out_err;
1481
1482         if (conn->cb_addr.ss_family == AF_INET6)
1483                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1484
1485         conn->cb_prog = se->se_callback_prog;
1486         conn->cb_ident = se->se_callback_ident;
1487         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1488         return;
1489 out_err:
1490         conn->cb_addr.ss_family = AF_UNSPEC;
1491         conn->cb_addrlen = 0;
1492         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1493                 "will not receive delegations\n",
1494                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1495
1496         return;
1497 }
1498
1499 /*
1500  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1501  */
1502 void
1503 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1504 {
1505         struct nfsd4_slot *slot = resp->cstate.slot;
1506         unsigned int base;
1507
1508         dprintk("--> %s slot %p\n", __func__, slot);
1509
1510         slot->sl_opcnt = resp->opcnt;
1511         slot->sl_status = resp->cstate.status;
1512
1513         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1514         if (nfsd4_not_cached(resp)) {
1515                 slot->sl_datalen = 0;
1516                 return;
1517         }
1518         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1519         base = (char *)resp->cstate.datap -
1520                                         (char *)resp->xbuf->head[0].iov_base;
1521         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1522                                     slot->sl_datalen))
1523                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1524         return;
1525 }
1526
1527 /*
1528  * Encode the replay sequence operation from the slot values.
1529  * If cachethis is FALSE encode the uncached rep error on the next
1530  * operation which sets resp->p and increments resp->opcnt for
1531  * nfs4svc_encode_compoundres.
1532  *
1533  */
1534 static __be32
1535 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1536                           struct nfsd4_compoundres *resp)
1537 {
1538         struct nfsd4_op *op;
1539         struct nfsd4_slot *slot = resp->cstate.slot;
1540
1541         /* Encode the replayed sequence operation */
1542         op = &args->ops[resp->opcnt - 1];
1543         nfsd4_encode_operation(resp, op);
1544
1545         /* Return nfserr_retry_uncached_rep in next operation. */
1546         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1547                 op = &args->ops[resp->opcnt++];
1548                 op->status = nfserr_retry_uncached_rep;
1549                 nfsd4_encode_operation(resp, op);
1550         }
1551         return op->status;
1552 }
1553
1554 /*
1555  * The sequence operation is not cached because we can use the slot and
1556  * session values.
1557  */
1558 __be32
1559 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1560                          struct nfsd4_sequence *seq)
1561 {
1562         struct nfsd4_slot *slot = resp->cstate.slot;
1563         __be32 status;
1564
1565         dprintk("--> %s slot %p\n", __func__, slot);
1566
1567         /* Either returns 0 or nfserr_retry_uncached */
1568         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1569         if (status == nfserr_retry_uncached_rep)
1570                 return status;
1571
1572         /* The sequence operation has been encoded, cstate->datap set. */
1573         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1574
1575         resp->opcnt = slot->sl_opcnt;
1576         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1577         status = slot->sl_status;
1578
1579         return status;
1580 }
1581
1582 /*
1583  * Set the exchange_id flags returned by the server.
1584  */
1585 static void
1586 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1587 {
1588         /* pNFS is not supported */
1589         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1590
1591         /* Referrals are supported, Migration is not. */
1592         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1593
1594         /* set the wire flags to return to client. */
1595         clid->flags = new->cl_exchange_flags;
1596 }
1597
1598 static bool client_has_state(struct nfs4_client *clp)
1599 {
1600         /*
1601          * Note clp->cl_openowners check isn't quite right: there's no
1602          * need to count owners without stateid's.
1603          *
1604          * Also note we should probably be using this in 4.0 case too.
1605          */
1606         return !list_empty(&clp->cl_openowners)
1607                 || !list_empty(&clp->cl_delegations)
1608                 || !list_empty(&clp->cl_sessions);
1609 }
1610
1611 __be32
1612 nfsd4_exchange_id(struct svc_rqst *rqstp,
1613                   struct nfsd4_compound_state *cstate,
1614                   struct nfsd4_exchange_id *exid)
1615 {
1616         struct nfs4_client *unconf, *conf, *new;
1617         __be32 status;
1618         char                    addr_str[INET6_ADDRSTRLEN];
1619         nfs4_verifier           verf = exid->verifier;
1620         struct sockaddr         *sa = svc_addr(rqstp);
1621         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1622
1623         rpc_ntop(sa, addr_str, sizeof(addr_str));
1624         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1625                 "ip_addr=%s flags %x, spa_how %d\n",
1626                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1627                 addr_str, exid->flags, exid->spa_how);
1628
1629         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1630                 return nfserr_inval;
1631
1632         /* Currently only support SP4_NONE */
1633         switch (exid->spa_how) {
1634         case SP4_NONE:
1635                 break;
1636         case SP4_SSV:
1637                 return nfserr_serverfault;
1638         default:
1639                 BUG();                          /* checked by xdr code */
1640         case SP4_MACH_CRED:
1641                 return nfserr_serverfault;      /* no excuse :-/ */
1642         }
1643
1644         /* Cases below refer to rfc 5661 section 18.35.4: */
1645         nfs4_lock_state();
1646         conf = find_confirmed_client_by_name(&exid->clname);
1647         if (conf) {
1648                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1649                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1650
1651                 if (update) {
1652                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1653                                 status = nfserr_inval;
1654                                 goto out;
1655                         }
1656                         if (!creds_match) { /* case 9 */
1657                                 status = nfserr_perm;
1658                                 goto out;
1659                         }
1660                         if (!verfs_match) { /* case 8 */
1661                                 status = nfserr_not_same;
1662                                 goto out;
1663                         }
1664                         /* case 6 */
1665                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1666                         new = conf;
1667                         goto out_copy;
1668                 }
1669                 if (!creds_match) { /* case 3 */
1670                         if (client_has_state(conf)) {
1671                                 status = nfserr_clid_inuse;
1672                                 goto out;
1673                         }
1674                         expire_client(conf);
1675                         goto out_new;
1676                 }
1677                 if (verfs_match) { /* case 2 */
1678                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1679                         new = conf;
1680                         goto out_copy;
1681                 }
1682                 /* case 5, client reboot */
1683                 goto out_new;
1684         }
1685
1686         if (update) { /* case 7 */
1687                 status = nfserr_noent;
1688                 goto out;
1689         }
1690
1691         unconf  = find_unconfirmed_client_by_name(&exid->clname);
1692         if (unconf) /* case 4, possible retry or client restart */
1693                 expire_client(unconf);
1694
1695         /* case 1 (normal case) */
1696 out_new:
1697         new = create_client(exid->clname, rqstp, &verf);
1698         if (new == NULL) {
1699                 status = nfserr_jukebox;
1700                 goto out;
1701         }
1702         new->cl_minorversion = 1;
1703
1704         gen_clid(new);
1705         add_to_unconfirmed(new);
1706 out_copy:
1707         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1708         exid->clientid.cl_id = new->cl_clientid.cl_id;
1709
1710         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1711         nfsd4_set_ex_flags(new, exid);
1712
1713         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1714                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1715         status = nfs_ok;
1716
1717 out:
1718         nfs4_unlock_state();
1719         return status;
1720 }
1721
1722 static __be32
1723 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1724 {
1725         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1726                 slot_seqid);
1727
1728         /* The slot is in use, and no response has been sent. */
1729         if (slot_inuse) {
1730                 if (seqid == slot_seqid)
1731                         return nfserr_jukebox;
1732                 else
1733                         return nfserr_seq_misordered;
1734         }
1735         /* Note unsigned 32-bit arithmetic handles wraparound: */
1736         if (likely(seqid == slot_seqid + 1))
1737                 return nfs_ok;
1738         if (seqid == slot_seqid)
1739                 return nfserr_replay_cache;
1740         return nfserr_seq_misordered;
1741 }
1742
1743 /*
1744  * Cache the create session result into the create session single DRC
1745  * slot cache by saving the xdr structure. sl_seqid has been set.
1746  * Do this for solo or embedded create session operations.
1747  */
1748 static void
1749 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1750                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1751 {
1752         slot->sl_status = nfserr;
1753         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1754 }
1755
1756 static __be32
1757 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1758                             struct nfsd4_clid_slot *slot)
1759 {
1760         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1761         return slot->sl_status;
1762 }
1763
1764 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1765                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1766                         1 +     /* MIN tag is length with zero, only length */ \
1767                         3 +     /* version, opcount, opcode */ \
1768                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1769                                 /* seqid, slotID, slotID, cache */ \
1770                         4 ) * sizeof(__be32))
1771
1772 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1773                         2 +     /* verifier: AUTH_NULL, length 0 */\
1774                         1 +     /* status */ \
1775                         1 +     /* MIN tag is length with zero, only length */ \
1776                         3 +     /* opcount, opcode, opstatus*/ \
1777                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1778                                 /* seqid, slotID, slotID, slotID, status */ \
1779                         5 ) * sizeof(__be32))
1780
1781 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1782 {
1783         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1784                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1785 }
1786
1787 __be32
1788 nfsd4_create_session(struct svc_rqst *rqstp,
1789                      struct nfsd4_compound_state *cstate,
1790                      struct nfsd4_create_session *cr_ses)
1791 {
1792         struct sockaddr *sa = svc_addr(rqstp);
1793         struct nfs4_client *conf, *unconf;
1794         struct nfsd4_session *new;
1795         struct nfsd4_conn *conn;
1796         struct nfsd4_clid_slot *cs_slot = NULL;
1797         __be32 status = 0;
1798
1799         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1800                 return nfserr_inval;
1801         if (check_forechannel_attrs(cr_ses->fore_channel))
1802                 return nfserr_toosmall;
1803         new = alloc_session(&cr_ses->fore_channel);
1804         if (!new)
1805                 return nfserr_jukebox;
1806         status = nfserr_jukebox;
1807         conn = alloc_conn_from_crses(rqstp, cr_ses);
1808         if (!conn)
1809                 goto out_free_session;
1810
1811         nfs4_lock_state();
1812         unconf = find_unconfirmed_client(&cr_ses->clientid, true);
1813         conf = find_confirmed_client(&cr_ses->clientid, true);
1814
1815         if (conf) {
1816                 cs_slot = &conf->cl_cs_slot;
1817                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1818                 if (status == nfserr_replay_cache) {
1819                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1820                         goto out_free_conn;
1821                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1822                         status = nfserr_seq_misordered;
1823                         goto out_free_conn;
1824                 }
1825         } else if (unconf) {
1826                 struct nfs4_client *old;
1827                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1828                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1829                         status = nfserr_clid_inuse;
1830                         goto out_free_conn;
1831                 }
1832                 cs_slot = &unconf->cl_cs_slot;
1833                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1834                 if (status) {
1835                         /* an unconfirmed replay returns misordered */
1836                         status = nfserr_seq_misordered;
1837                         goto out_free_conn;
1838                 }
1839                 old = find_confirmed_client_by_name(&unconf->cl_name);
1840                 if (old)
1841                         expire_client(old);
1842                 move_to_confirmed(unconf);
1843                 conf = unconf;
1844         } else {
1845                 status = nfserr_stale_clientid;
1846                 goto out_free_conn;
1847         }
1848         status = nfs_ok;
1849         /*
1850          * We do not support RDMA or persistent sessions
1851          */
1852         cr_ses->flags &= ~SESSION4_PERSIST;
1853         cr_ses->flags &= ~SESSION4_RDMA;
1854
1855         init_session(rqstp, new, conf, cr_ses);
1856         nfsd4_init_conn(rqstp, conn, new);
1857
1858         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1859                NFS4_MAX_SESSIONID_LEN);
1860         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1861                 sizeof(struct nfsd4_channel_attrs));
1862         cs_slot->sl_seqid++;
1863         cr_ses->seqid = cs_slot->sl_seqid;
1864
1865         /* cache solo and embedded create sessions under the state lock */
1866         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1867 out:
1868         nfs4_unlock_state();
1869         dprintk("%s returns %d\n", __func__, ntohl(status));
1870         return status;
1871 out_free_conn:
1872         free_conn(conn);
1873 out_free_session:
1874         __free_session(new);
1875         goto out;
1876 }
1877
1878 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1879 {
1880         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1881         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1882
1883         return argp->opcnt == resp->opcnt;
1884 }
1885
1886 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1887 {
1888         switch (*dir) {
1889         case NFS4_CDFC4_FORE:
1890         case NFS4_CDFC4_BACK:
1891                 return nfs_ok;
1892         case NFS4_CDFC4_FORE_OR_BOTH:
1893         case NFS4_CDFC4_BACK_OR_BOTH:
1894                 *dir = NFS4_CDFC4_BOTH;
1895                 return nfs_ok;
1896         };
1897         return nfserr_inval;
1898 }
1899
1900 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1901 {
1902         struct nfsd4_session *session = cstate->session;
1903
1904         spin_lock(&client_lock);
1905         session->se_cb_prog = bc->bc_cb_program;
1906         session->se_cb_sec = bc->bc_cb_sec;
1907         spin_unlock(&client_lock);
1908
1909         nfsd4_probe_callback(session->se_client);
1910
1911         return nfs_ok;
1912 }
1913
1914 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1915                      struct nfsd4_compound_state *cstate,
1916                      struct nfsd4_bind_conn_to_session *bcts)
1917 {
1918         __be32 status;
1919         struct nfsd4_conn *conn;
1920
1921         if (!nfsd4_last_compound_op(rqstp))
1922                 return nfserr_not_only_op;
1923         spin_lock(&client_lock);
1924         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1925         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1926          * client_lock iself: */
1927         if (cstate->session) {
1928                 nfsd4_get_session(cstate->session);
1929                 atomic_inc(&cstate->session->se_client->cl_refcount);
1930         }
1931         spin_unlock(&client_lock);
1932         if (!cstate->session)
1933                 return nfserr_badsession;
1934
1935         status = nfsd4_map_bcts_dir(&bcts->dir);
1936         if (status)
1937                 return status;
1938         conn = alloc_conn(rqstp, bcts->dir);
1939         if (!conn)
1940                 return nfserr_jukebox;
1941         nfsd4_init_conn(rqstp, conn, cstate->session);
1942         return nfs_ok;
1943 }
1944
1945 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1946 {
1947         if (!session)
1948                 return 0;
1949         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1950 }
1951
1952 __be32
1953 nfsd4_destroy_session(struct svc_rqst *r,
1954                       struct nfsd4_compound_state *cstate,
1955                       struct nfsd4_destroy_session *sessionid)
1956 {
1957         struct nfsd4_session *ses;
1958         __be32 status = nfserr_badsession;
1959
1960         /* Notes:
1961          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1962          * - Should we return nfserr_back_chan_busy if waiting for
1963          *   callbacks on to-be-destroyed session?
1964          * - Do we need to clear any callback info from previous session?
1965          */
1966
1967         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1968                 if (!nfsd4_last_compound_op(r))
1969                         return nfserr_not_only_op;
1970         }
1971         dump_sessionid(__func__, &sessionid->sessionid);
1972         spin_lock(&client_lock);
1973         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1974         if (!ses) {
1975                 spin_unlock(&client_lock);
1976                 goto out;
1977         }
1978
1979         unhash_session(ses);
1980         spin_unlock(&client_lock);
1981
1982         nfs4_lock_state();
1983         nfsd4_probe_callback_sync(ses->se_client);
1984         nfs4_unlock_state();
1985
1986         spin_lock(&client_lock);
1987         nfsd4_del_conns(ses);
1988         nfsd4_put_session_locked(ses);
1989         spin_unlock(&client_lock);
1990         status = nfs_ok;
1991 out:
1992         dprintk("%s returns %d\n", __func__, ntohl(status));
1993         return status;
1994 }
1995
1996 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1997 {
1998         struct nfsd4_conn *c;
1999
2000         list_for_each_entry(c, &s->se_conns, cn_persession) {
2001                 if (c->cn_xprt == xpt) {
2002                         return c;
2003                 }
2004         }
2005         return NULL;
2006 }
2007
2008 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2009 {
2010         struct nfs4_client *clp = ses->se_client;
2011         struct nfsd4_conn *c;
2012         int ret;
2013
2014         spin_lock(&clp->cl_lock);
2015         c = __nfsd4_find_conn(new->cn_xprt, ses);
2016         if (c) {
2017                 spin_unlock(&clp->cl_lock);
2018                 free_conn(new);
2019                 return;
2020         }
2021         __nfsd4_hash_conn(new, ses);
2022         spin_unlock(&clp->cl_lock);
2023         ret = nfsd4_register_conn(new);
2024         if (ret)
2025                 /* oops; xprt is already down: */
2026                 nfsd4_conn_lost(&new->cn_xpt_user);
2027         return;
2028 }
2029
2030 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2031 {
2032         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2033
2034         return args->opcnt > session->se_fchannel.maxops;
2035 }
2036
2037 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2038                                   struct nfsd4_session *session)
2039 {
2040         struct xdr_buf *xb = &rqstp->rq_arg;
2041
2042         return xb->len > session->se_fchannel.maxreq_sz;
2043 }
2044
2045 __be32
2046 nfsd4_sequence(struct svc_rqst *rqstp,
2047                struct nfsd4_compound_state *cstate,
2048                struct nfsd4_sequence *seq)
2049 {
2050         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2051         struct nfsd4_session *session;
2052         struct nfsd4_slot *slot;
2053         struct nfsd4_conn *conn;
2054         __be32 status;
2055
2056         if (resp->opcnt != 1)
2057                 return nfserr_sequence_pos;
2058
2059         /*
2060          * Will be either used or freed by nfsd4_sequence_check_conn
2061          * below.
2062          */
2063         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2064         if (!conn)
2065                 return nfserr_jukebox;
2066
2067         spin_lock(&client_lock);
2068         status = nfserr_badsession;
2069         session = find_in_sessionid_hashtbl(&seq->sessionid);
2070         if (!session)
2071                 goto out;
2072
2073         status = nfserr_too_many_ops;
2074         if (nfsd4_session_too_many_ops(rqstp, session))
2075                 goto out;
2076
2077         status = nfserr_req_too_big;
2078         if (nfsd4_request_too_big(rqstp, session))
2079                 goto out;
2080
2081         status = nfserr_badslot;
2082         if (seq->slotid >= session->se_fchannel.maxreqs)
2083                 goto out;
2084
2085         slot = session->se_slots[seq->slotid];
2086         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2087
2088         /* We do not negotiate the number of slots yet, so set the
2089          * maxslots to the session maxreqs which is used to encode
2090          * sr_highest_slotid and the sr_target_slot id to maxslots */
2091         seq->maxslots = session->se_fchannel.maxreqs;
2092
2093         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2094                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2095         if (status == nfserr_replay_cache) {
2096                 status = nfserr_seq_misordered;
2097                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2098                         goto out;
2099                 cstate->slot = slot;
2100                 cstate->session = session;
2101                 /* Return the cached reply status and set cstate->status
2102                  * for nfsd4_proc_compound processing */
2103                 status = nfsd4_replay_cache_entry(resp, seq);
2104                 cstate->status = nfserr_replay_cache;
2105                 goto out;
2106         }
2107         if (status)
2108                 goto out;
2109
2110         nfsd4_sequence_check_conn(conn, session);
2111         conn = NULL;
2112
2113         /* Success! bump slot seqid */
2114         slot->sl_seqid = seq->seqid;
2115         slot->sl_flags |= NFSD4_SLOT_INUSE;
2116         if (seq->cachethis)
2117                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2118         else
2119                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2120
2121         cstate->slot = slot;
2122         cstate->session = session;
2123
2124 out:
2125         /* Hold a session reference until done processing the compound. */
2126         if (cstate->session) {
2127                 struct nfs4_client *clp = session->se_client;
2128
2129                 nfsd4_get_session(cstate->session);
2130                 atomic_inc(&clp->cl_refcount);
2131                 switch (clp->cl_cb_state) {
2132                 case NFSD4_CB_DOWN:
2133                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2134                         break;
2135                 case NFSD4_CB_FAULT:
2136                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2137                         break;
2138                 default:
2139                         seq->status_flags = 0;
2140                 }
2141         }
2142         kfree(conn);
2143         spin_unlock(&client_lock);
2144         dprintk("%s: return %d\n", __func__, ntohl(status));
2145         return status;
2146 }
2147
2148 __be32
2149 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2150 {
2151         struct nfs4_client *conf, *unconf, *clp;
2152         __be32 status = 0;
2153
2154         nfs4_lock_state();
2155         unconf = find_unconfirmed_client(&dc->clientid, true);
2156         conf = find_confirmed_client(&dc->clientid, true);
2157
2158         if (conf) {
2159                 clp = conf;
2160
2161                 if (!is_client_expired(conf) && client_has_state(conf)) {
2162                         status = nfserr_clientid_busy;
2163                         goto out;
2164                 }
2165
2166                 /* rfc5661 18.50.3 */
2167                 if (cstate->session && conf == cstate->session->se_client) {
2168                         status = nfserr_clientid_busy;
2169                         goto out;
2170                 }
2171         } else if (unconf)
2172                 clp = unconf;
2173         else {
2174                 status = nfserr_stale_clientid;
2175                 goto out;
2176         }
2177
2178         expire_client(clp);
2179 out:
2180         nfs4_unlock_state();
2181         dprintk("%s return %d\n", __func__, ntohl(status));
2182         return status;
2183 }
2184
2185 __be32
2186 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2187 {
2188         __be32 status = 0;
2189
2190         if (rc->rca_one_fs) {
2191                 if (!cstate->current_fh.fh_dentry)
2192                         return nfserr_nofilehandle;
2193                 /*
2194                  * We don't take advantage of the rca_one_fs case.
2195                  * That's OK, it's optional, we can safely ignore it.
2196                  */
2197                  return nfs_ok;
2198         }
2199
2200         nfs4_lock_state();
2201         status = nfserr_complete_already;
2202         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2203                              &cstate->session->se_client->cl_flags))
2204                 goto out;
2205
2206         status = nfserr_stale_clientid;
2207         if (is_client_expired(cstate->session->se_client))
2208                 /*
2209                  * The following error isn't really legal.
2210                  * But we only get here if the client just explicitly
2211                  * destroyed the client.  Surely it no longer cares what
2212                  * error it gets back on an operation for the dead
2213                  * client.
2214                  */
2215                 goto out;
2216
2217         status = nfs_ok;
2218         nfsd4_client_record_create(cstate->session->se_client);
2219 out:
2220         nfs4_unlock_state();
2221         return status;
2222 }
2223
2224 __be32
2225 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2226                   struct nfsd4_setclientid *setclid)
2227 {
2228         struct xdr_netobj       clname = setclid->se_name;
2229         nfs4_verifier           clverifier = setclid->se_verf;
2230         struct nfs4_client      *conf, *unconf, *new;
2231         __be32                  status;
2232         
2233         /* Cases below refer to rfc 3530 section 14.2.33: */
2234         nfs4_lock_state();
2235         conf = find_confirmed_client_by_name(&clname);
2236         if (conf) {
2237                 /* case 0: */
2238                 status = nfserr_clid_inuse;
2239                 if (clp_used_exchangeid(conf))
2240                         goto out;
2241                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2242                         char addr_str[INET6_ADDRSTRLEN];
2243                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2244                                  sizeof(addr_str));
2245                         dprintk("NFSD: setclientid: string in use by client "
2246                                 "at %s\n", addr_str);
2247                         goto out;
2248                 }
2249         }
2250         unconf = find_unconfirmed_client_by_name(&clname);
2251         if (unconf)
2252                 expire_client(unconf);
2253         status = nfserr_jukebox;
2254         new = create_client(clname, rqstp, &clverifier);
2255         if (new == NULL)
2256                 goto out;
2257         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2258                 /* case 1: probable callback update */
2259                 copy_clid(new, conf);
2260         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2261                 gen_clid(new);
2262         new->cl_minorversion = 0;
2263         gen_callback(new, setclid, rqstp);
2264         add_to_unconfirmed(new);
2265         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2266         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2267         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2268         status = nfs_ok;
2269 out:
2270         nfs4_unlock_state();
2271         return status;
2272 }
2273
2274
2275 __be32
2276 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2277                          struct nfsd4_compound_state *cstate,
2278                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2279 {
2280         struct nfs4_client *conf, *unconf;
2281         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
2282         clientid_t * clid = &setclientid_confirm->sc_clientid;
2283         __be32 status;
2284         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
2285
2286         if (STALE_CLIENTID(clid, nn))
2287                 return nfserr_stale_clientid;
2288         nfs4_lock_state();
2289
2290         conf = find_confirmed_client(clid, false);
2291         unconf = find_unconfirmed_client(clid, false);
2292         /*
2293          * We try hard to give out unique clientid's, so if we get an
2294          * attempt to confirm the same clientid with a different cred,
2295          * there's a bug somewhere.  Let's charitably assume it's our
2296          * bug.
2297          */
2298         status = nfserr_serverfault;
2299         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2300                 goto out;
2301         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2302                 goto out;
2303         /* cases below refer to rfc 3530 section 14.2.34: */
2304         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2305                 if (conf && !unconf) /* case 2: probable retransmit */
2306                         status = nfs_ok;
2307                 else /* case 4: client hasn't noticed we rebooted yet? */
2308                         status = nfserr_stale_clientid;
2309                 goto out;
2310         }
2311         status = nfs_ok;
2312         if (conf) { /* case 1: callback update */
2313                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2314                 nfsd4_probe_callback(conf);
2315                 expire_client(unconf);
2316         } else { /* case 3: normal case; new or rebooted client */
2317                 conf = find_confirmed_client_by_name(&unconf->cl_name);
2318                 if (conf)
2319                         expire_client(conf);
2320                 move_to_confirmed(unconf);
2321                 nfsd4_probe_callback(unconf);
2322         }
2323 out:
2324         nfs4_unlock_state();
2325         return status;
2326 }
2327
2328 static struct nfs4_file *nfsd4_alloc_file(void)
2329 {
2330         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2331 }
2332
2333 /* OPEN Share state helper functions */
2334 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2335 {
2336         unsigned int hashval = file_hashval(ino);
2337
2338         atomic_set(&fp->fi_ref, 1);
2339         INIT_LIST_HEAD(&fp->fi_hash);
2340         INIT_LIST_HEAD(&fp->fi_stateids);
2341         INIT_LIST_HEAD(&fp->fi_delegations);
2342         fp->fi_inode = igrab(ino);
2343         fp->fi_had_conflict = false;
2344         fp->fi_lease = NULL;
2345         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2346         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2347         spin_lock(&recall_lock);
2348         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2349         spin_unlock(&recall_lock);
2350 }
2351
2352 static void
2353 nfsd4_free_slab(struct kmem_cache **slab)
2354 {
2355         if (*slab == NULL)
2356                 return;
2357         kmem_cache_destroy(*slab);
2358         *slab = NULL;
2359 }
2360
2361 void
2362 nfsd4_free_slabs(void)
2363 {
2364         nfsd4_free_slab(&openowner_slab);
2365         nfsd4_free_slab(&lockowner_slab);
2366         nfsd4_free_slab(&file_slab);
2367         nfsd4_free_slab(&stateid_slab);
2368         nfsd4_free_slab(&deleg_slab);
2369 }
2370
2371 int
2372 nfsd4_init_slabs(void)
2373 {
2374         openowner_slab = kmem_cache_create("nfsd4_openowners",
2375                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2376         if (openowner_slab == NULL)
2377                 goto out_nomem;
2378         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2379                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2380         if (lockowner_slab == NULL)
2381                 goto out_nomem;
2382         file_slab = kmem_cache_create("nfsd4_files",
2383                         sizeof(struct nfs4_file), 0, 0, NULL);
2384         if (file_slab == NULL)
2385                 goto out_nomem;
2386         stateid_slab = kmem_cache_create("nfsd4_stateids",
2387                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2388         if (stateid_slab == NULL)
2389                 goto out_nomem;
2390         deleg_slab = kmem_cache_create("nfsd4_delegations",
2391                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2392         if (deleg_slab == NULL)
2393                 goto out_nomem;
2394         return 0;
2395 out_nomem:
2396         nfsd4_free_slabs();
2397         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2398         return -ENOMEM;
2399 }
2400
2401 void nfs4_free_openowner(struct nfs4_openowner *oo)
2402 {
2403         kfree(oo->oo_owner.so_owner.data);
2404         kmem_cache_free(openowner_slab, oo);
2405 }
2406
2407 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2408 {
2409         kfree(lo->lo_owner.so_owner.data);
2410         kmem_cache_free(lockowner_slab, lo);
2411 }
2412
2413 static void init_nfs4_replay(struct nfs4_replay *rp)
2414 {
2415         rp->rp_status = nfserr_serverfault;
2416         rp->rp_buflen = 0;
2417         rp->rp_buf = rp->rp_ibuf;
2418 }
2419
2420 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2421 {
2422         struct nfs4_stateowner *sop;
2423
2424         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2425         if (!sop)
2426                 return NULL;
2427
2428         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2429         if (!sop->so_owner.data) {
2430                 kmem_cache_free(slab, sop);
2431                 return NULL;
2432         }
2433         sop->so_owner.len = owner->len;
2434
2435         INIT_LIST_HEAD(&sop->so_stateids);
2436         sop->so_client = clp;
2437         init_nfs4_replay(&sop->so_replay);
2438         return sop;
2439 }
2440
2441 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2442 {
2443         list_add(&oo->oo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
2444         list_add(&oo->oo_perclient, &clp->cl_openowners);
2445 }
2446
2447 static struct nfs4_openowner *
2448 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2449         struct nfs4_openowner *oo;
2450
2451         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2452         if (!oo)
2453                 return NULL;
2454         oo->oo_owner.so_is_open_owner = 1;
2455         oo->oo_owner.so_seqid = open->op_seqid;
2456         oo->oo_flags = NFS4_OO_NEW;
2457         oo->oo_time = 0;
2458         oo->oo_last_closed_stid = NULL;
2459         INIT_LIST_HEAD(&oo->oo_close_lru);
2460         hash_openowner(oo, clp, strhashval);
2461         return oo;
2462 }
2463
2464 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2465         struct nfs4_openowner *oo = open->op_openowner;
2466         struct nfs4_client *clp = oo->oo_owner.so_client;
2467
2468         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2469         INIT_LIST_HEAD(&stp->st_lockowners);
2470         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2471         list_add(&stp->st_perfile, &fp->fi_stateids);
2472         stp->st_stateowner = &oo->oo_owner;
2473         get_nfs4_file(fp);
2474         stp->st_file = fp;
2475         stp->st_access_bmap = 0;
2476         stp->st_deny_bmap = 0;
2477         set_access(open->op_share_access, stp);
2478         set_deny(open->op_share_deny, stp);
2479         stp->st_openstp = NULL;
2480 }
2481
2482 static void
2483 move_to_close_lru(struct nfs4_openowner *oo)
2484 {
2485         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2486
2487         list_move_tail(&oo->oo_close_lru, &close_lru);
2488         oo->oo_time = get_seconds();
2489 }
2490
2491 static int
2492 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2493                                                         clientid_t *clid)
2494 {
2495         return (sop->so_owner.len == owner->len) &&
2496                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2497                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2498 }
2499
2500 static struct nfs4_openowner *
2501 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open, bool sessions)
2502 {
2503         struct nfs4_stateowner *so;
2504         struct nfs4_openowner *oo;
2505         struct nfs4_client *clp;
2506
2507         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2508                 if (!so->so_is_open_owner)
2509                         continue;
2510                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2511                         oo = openowner(so);
2512                         clp = oo->oo_owner.so_client;
2513                         if ((bool)clp->cl_minorversion != sessions)
2514                                 return NULL;
2515                         renew_client(oo->oo_owner.so_client);
2516                         return oo;
2517                 }
2518         }
2519         return NULL;
2520 }
2521
2522 /* search file_hashtbl[] for file */
2523 static struct nfs4_file *
2524 find_file(struct inode *ino)
2525 {
2526         unsigned int hashval = file_hashval(ino);
2527         struct nfs4_file *fp;
2528
2529         spin_lock(&recall_lock);
2530         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2531                 if (fp->fi_inode == ino) {
2532                         get_nfs4_file(fp);
2533                         spin_unlock(&recall_lock);
2534                         return fp;
2535                 }
2536         }
2537         spin_unlock(&recall_lock);
2538         return NULL;
2539 }
2540
2541 /*
2542  * Called to check deny when READ with all zero stateid or
2543  * WRITE with all zero or all one stateid
2544  */
2545 static __be32
2546 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2547 {
2548         struct inode *ino = current_fh->fh_dentry->d_inode;
2549         struct nfs4_file *fp;
2550         struct nfs4_ol_stateid *stp;
2551         __be32 ret;
2552
2553         dprintk("NFSD: nfs4_share_conflict\n");
2554
2555         fp = find_file(ino);
2556         if (!fp)
2557                 return nfs_ok;
2558         ret = nfserr_locked;
2559         /* Search for conflicting share reservations */
2560         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2561                 if (test_deny(deny_type, stp) ||
2562                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2563                         goto out;
2564         }
2565         ret = nfs_ok;
2566 out:
2567         put_nfs4_file(fp);
2568         return ret;
2569 }
2570
2571 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2572 {
2573         /* We're assuming the state code never drops its reference
2574          * without first removing the lease.  Since we're in this lease
2575          * callback (and since the lease code is serialized by the kernel
2576          * lock) we know the server hasn't removed the lease yet, we know
2577          * it's safe to take a reference: */
2578         atomic_inc(&dp->dl_count);
2579
2580         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2581
2582         /* only place dl_time is set. protected by lock_flocks*/
2583         dp->dl_time = get_seconds();
2584
2585         nfsd4_cb_recall(dp);
2586 }
2587
2588 /* Called from break_lease() with lock_flocks() held. */
2589 static void nfsd_break_deleg_cb(struct file_lock *fl)
2590 {
2591         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2592         struct nfs4_delegation *dp;
2593
2594         if (!fp) {
2595                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2596                 return;
2597         }
2598         if (fp->fi_had_conflict) {
2599                 WARN(1, "duplicate break on %p\n", fp);
2600                 return;
2601         }
2602         /*
2603          * We don't want the locks code to timeout the lease for us;
2604          * we'll remove it ourself if a delegation isn't returned
2605          * in time:
2606          */
2607         fl->fl_break_time = 0;
2608
2609         spin_lock(&recall_lock);
2610         fp->fi_had_conflict = true;
2611         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2612                 nfsd_break_one_deleg(dp);
2613         spin_unlock(&recall_lock);
2614 }
2615
2616 static
2617 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2618 {
2619         if (arg & F_UNLCK)
2620                 return lease_modify(onlist, arg);
2621         else
2622                 return -EAGAIN;
2623 }
2624
2625 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2626         .lm_break = nfsd_break_deleg_cb,
2627         .lm_change = nfsd_change_deleg_cb,
2628 };
2629
2630 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2631 {
2632         if (nfsd4_has_session(cstate))
2633                 return nfs_ok;
2634         if (seqid == so->so_seqid - 1)
2635                 return nfserr_replay_me;
2636         if (seqid == so->so_seqid)
2637                 return nfs_ok;
2638         return nfserr_bad_seqid;
2639 }
2640
2641 __be32
2642 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2643                     struct nfsd4_open *open)
2644 {
2645         clientid_t *clientid = &open->op_clientid;
2646         struct nfs4_client *clp = NULL;
2647         unsigned int strhashval;
2648         struct nfs4_openowner *oo = NULL;
2649         __be32 status;
2650         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
2651
2652         if (STALE_CLIENTID(&open->op_clientid, nn))
2653                 return nfserr_stale_clientid;
2654         /*
2655          * In case we need it later, after we've already created the
2656          * file and don't want to risk a further failure:
2657          */
2658         open->op_file = nfsd4_alloc_file();
2659         if (open->op_file == NULL)
2660                 return nfserr_jukebox;
2661
2662         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2663         oo = find_openstateowner_str(strhashval, open, cstate->minorversion);
2664         open->op_openowner = oo;
2665         if (!oo) {
2666                 clp = find_confirmed_client(clientid, cstate->minorversion);
2667                 if (clp == NULL)
2668                         return nfserr_expired;
2669                 goto new_owner;
2670         }
2671         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2672                 /* Replace unconfirmed owners without checking for replay. */
2673                 clp = oo->oo_owner.so_client;
2674                 release_openowner(oo);
2675                 open->op_openowner = NULL;
2676                 goto new_owner;
2677         }
2678         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2679         if (status)
2680                 return status;
2681         clp = oo->oo_owner.so_client;
2682         goto alloc_stateid;
2683 new_owner:
2684         oo = alloc_init_open_stateowner(strhashval, clp, open);
2685         if (oo == NULL)
2686                 return nfserr_jukebox;
2687         open->op_openowner = oo;
2688 alloc_stateid:
2689         open->op_stp = nfs4_alloc_stateid(clp);
2690         if (!open->op_stp)
2691                 return nfserr_jukebox;
2692         return nfs_ok;
2693 }
2694
2695 static inline __be32
2696 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2697 {
2698         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2699                 return nfserr_openmode;
2700         else
2701                 return nfs_ok;
2702 }
2703
2704 static int share_access_to_flags(u32 share_access)
2705 {
2706         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2707 }
2708
2709 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2710 {
2711         struct nfs4_stid *ret;
2712
2713         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2714         if (!ret)
2715                 return NULL;
2716         return delegstateid(ret);
2717 }
2718
2719 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2720 {
2721         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2722                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2723 }
2724
2725 static __be32
2726 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2727                 struct nfs4_delegation **dp)
2728 {
2729         int flags;
2730         __be32 status = nfserr_bad_stateid;
2731
2732         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2733         if (*dp == NULL)
2734                 goto out;
2735         flags = share_access_to_flags(open->op_share_access);
2736         status = nfs4_check_delegmode(*dp, flags);
2737         if (status)
2738                 *dp = NULL;
2739 out:
2740         if (!nfsd4_is_deleg_cur(open))
2741                 return nfs_ok;
2742         if (status)
2743                 return status;
2744         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2745         return nfs_ok;
2746 }
2747
2748 static __be32
2749 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2750 {
2751         struct nfs4_ol_stateid *local;
2752         struct nfs4_openowner *oo = open->op_openowner;
2753
2754         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2755                 /* ignore lock owners */
2756                 if (local->st_stateowner->so_is_open_owner == 0)
2757                         continue;
2758                 /* remember if we have seen this open owner */
2759                 if (local->st_stateowner == &oo->oo_owner)
2760                         *stpp = local;
2761                 /* check for conflicting share reservations */
2762                 if (!test_share(local, open))
2763                         return nfserr_share_denied;
2764         }
2765         return nfs_ok;
2766 }
2767
2768 static inline int nfs4_access_to_access(u32 nfs4_access)
2769 {
2770         int flags = 0;
2771
2772         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2773                 flags |= NFSD_MAY_READ;
2774         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2775                 flags |= NFSD_MAY_WRITE;
2776         return flags;
2777 }
2778
2779 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2780                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2781 {
2782         __be32 status;
2783         int oflag = nfs4_access_to_omode(open->op_share_access);
2784         int access = nfs4_access_to_access(open->op_share_access);
2785
2786         if (!fp->fi_fds[oflag]) {
2787                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2788                         &fp->fi_fds[oflag]);
2789                 if (status)
2790                         return status;
2791         }
2792         nfs4_file_get_access(fp, oflag);
2793
2794         return nfs_ok;
2795 }
2796
2797 static inline __be32
2798 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2799                 struct nfsd4_open *open)
2800 {
2801         struct iattr iattr = {
2802                 .ia_valid = ATTR_SIZE,
2803                 .ia_size = 0,
2804         };
2805         if (!open->op_truncate)
2806                 return 0;
2807         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2808                 return nfserr_inval;
2809         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2810 }
2811
2812 static __be32
2813 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)
2814 {
2815         u32 op_share_access = open->op_share_access;
2816         bool new_access;
2817         __be32 status;
2818
2819         new_access = !test_access(op_share_access, stp);
2820         if (new_access) {
2821                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2822                 if (status)
2823                         return status;
2824         }
2825         status = nfsd4_truncate(rqstp, cur_fh, open);
2826         if (status) {
2827                 if (new_access) {
2828                         int oflag = nfs4_access_to_omode(op_share_access);
2829                         nfs4_file_put_access(fp, oflag);
2830                 }
2831                 return status;
2832         }
2833         /* remember the open */
2834         set_access(op_share_access, stp);
2835         set_deny(open->op_share_deny, stp);
2836
2837         return nfs_ok;
2838 }
2839
2840
2841 static void
2842 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2843 {
2844         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2845 }
2846
2847 /* Should we give out recallable state?: */
2848 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2849 {
2850         if (clp->cl_cb_state == NFSD4_CB_UP)
2851                 return true;
2852         /*
2853          * In the sessions case, since we don't have to establish a
2854          * separate connection for callbacks, we assume it's OK
2855          * until we hear otherwise:
2856          */
2857         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2858 }
2859
2860 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2861 {
2862         struct file_lock *fl;
2863
2864         fl = locks_alloc_lock();
2865         if (!fl)
2866                 return NULL;
2867         locks_init_lock(fl);
2868         fl->fl_lmops = &nfsd_lease_mng_ops;
2869         fl->fl_flags = FL_LEASE;
2870         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2871         fl->fl_end = OFFSET_MAX;
2872         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2873         fl->fl_pid = current->tgid;
2874         return fl;
2875 }
2876
2877 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2878 {
2879         struct nfs4_file *fp = dp->dl_file;
2880         struct file_lock *fl;
2881         int status;
2882
2883         fl = nfs4_alloc_init_lease(dp, flag);
2884         if (!fl)
2885                 return -ENOMEM;
2886         fl->fl_file = find_readable_file(fp);
2887         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2888         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2889         if (status) {
2890                 list_del_init(&dp->dl_perclnt);
2891                 locks_free_lock(fl);
2892                 return -ENOMEM;
2893         }
2894         fp->fi_lease = fl;
2895         fp->fi_deleg_file = get_file(fl->fl_file);
2896         atomic_set(&fp->fi_delegees, 1);
2897         list_add(&dp->dl_perfile, &fp->fi_delegations);
2898         return 0;
2899 }
2900
2901 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2902 {
2903         struct nfs4_file *fp = dp->dl_file;
2904
2905         if (!fp->fi_lease)
2906                 return nfs4_setlease(dp, flag);
2907         spin_lock(&recall_lock);
2908         if (fp->fi_had_conflict) {
2909                 spin_unlock(&recall_lock);
2910                 return -EAGAIN;
2911         }
2912         atomic_inc(&fp->fi_delegees);
2913         list_add(&dp->dl_perfile, &fp->fi_delegations);
2914         spin_unlock(&recall_lock);
2915         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2916         return 0;
2917 }
2918
2919 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2920 {
2921         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2922         if (status == -EAGAIN)
2923                 open->op_why_no_deleg = WND4_CONTENTION;
2924         else {
2925                 open->op_why_no_deleg = WND4_RESOURCE;
2926                 switch (open->op_deleg_want) {
2927                 case NFS4_SHARE_WANT_READ_DELEG:
2928                 case NFS4_SHARE_WANT_WRITE_DELEG:
2929                 case NFS4_SHARE_WANT_ANY_DELEG:
2930                         break;
2931                 case NFS4_SHARE_WANT_CANCEL:
2932                         open->op_why_no_deleg = WND4_CANCELLED;
2933                         break;
2934                 case NFS4_SHARE_WANT_NO_DELEG:
2935                         BUG();  /* not supposed to get here */
2936                 }
2937         }
2938 }
2939
2940 /*
2941  * Attempt to hand out a delegation.
2942  */
2943 static void
2944 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2945                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2946 {
2947         struct nfs4_delegation *dp;
2948         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2949         int cb_up;
2950         int status = 0, flag = 0;
2951
2952         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2953         flag = NFS4_OPEN_DELEGATE_NONE;
2954         open->op_recall = 0;
2955         switch (open->op_claim_type) {
2956                 case NFS4_OPEN_CLAIM_PREVIOUS:
2957                         if (!cb_up)
2958                                 open->op_recall = 1;
2959                         flag = open->op_delegate_type;
2960                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2961                                 goto out;
2962                         break;
2963                 case NFS4_OPEN_CLAIM_NULL:
2964                         /* Let's not give out any delegations till everyone's
2965                          * had the chance to reclaim theirs.... */
2966                         if (locks_in_grace(net))
2967                                 goto out;
2968                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2969                                 goto out;
2970                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2971                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2972                         else
2973                                 flag = NFS4_OPEN_DELEGATE_READ;
2974                         break;
2975                 default:
2976                         goto out;
2977         }
2978
2979         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2980         if (dp == NULL)
2981                 goto out_no_deleg;
2982         status = nfs4_set_delegation(dp, flag);
2983         if (status)
2984                 goto out_free;
2985
2986         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2987
2988         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2989                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2990 out:
2991         open->op_delegate_type = flag;
2992         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2993                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2994                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2995                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2996
2997                 /* 4.1 client asking for a delegation? */
2998                 if (open->op_deleg_want)
2999                         nfsd4_open_deleg_none_ext(open, status);
3000         }
3001         return;
3002 out_free:
3003         nfs4_put_delegation(dp);
3004 out_no_deleg:
3005         flag = NFS4_OPEN_DELEGATE_NONE;
3006         goto out;
3007 }
3008
3009 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3010                                         struct nfs4_delegation *dp)
3011 {
3012         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3013             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3014                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3015                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3016         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3017                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3018                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3019                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3020         }
3021         /* Otherwise the client must be confused wanting a delegation
3022          * it already has, therefore we don't return
3023          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3024          */
3025 }
3026
3027 /*
3028  * called with nfs4_lock_state() held.
3029  */
3030 __be32
3031 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3032 {
3033         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3034         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3035         struct nfs4_file *fp = NULL;
3036         struct inode *ino = current_fh->fh_dentry->d_inode;
3037         struct nfs4_ol_stateid *stp = NULL;
3038         struct nfs4_delegation *dp = NULL;
3039         __be32 status;
3040
3041         /*
3042          * Lookup file; if found, lookup stateid and check open request,
3043          * and check for delegations in the process of being recalled.
3044          * If not found, create the nfs4_file struct
3045          */
3046         fp = find_file(ino);
3047         if (fp) {
3048                 if ((status = nfs4_check_open(fp, open, &stp)))
3049                         goto out;
3050                 status = nfs4_check_deleg(cl, fp, open, &dp);
3051                 if (status)
3052                         goto out;
3053         } else {
3054                 status = nfserr_bad_stateid;
3055                 if (nfsd4_is_deleg_cur(open))
3056                         goto out;
3057                 status = nfserr_jukebox;
3058                 fp = open->op_file;
3059                 open->op_file = NULL;
3060                 nfsd4_init_file(fp, ino);
3061         }
3062
3063         /*
3064          * OPEN the file, or upgrade an existing OPEN.
3065          * If truncate fails, the OPEN fails.
3066          */
3067         if (stp) {
3068                 /* Stateid was found, this is an OPEN upgrade */
3069                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3070                 if (status)
3071                         goto out;
3072         } else {
3073                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3074                 if (status)
3075                         goto out;
3076                 status = nfsd4_truncate(rqstp, current_fh, open);
3077                 if (status)
3078                         goto out;
3079                 stp = open->op_stp;
3080                 open->op_stp = NULL;
3081                 init_open_stateid(stp, fp, open);
3082         }
3083         update_stateid(&stp->st_stid.sc_stateid);
3084         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3085
3086         if (nfsd4_has_session(&resp->cstate)) {
3087                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3088
3089                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3090                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3091                         open->op_why_no_deleg = WND4_NOT_WANTED;
3092                         goto nodeleg;
3093                 }
3094         }
3095
3096         /*
3097         * Attempt to hand out a delegation. No error return, because the
3098         * OPEN succeeds even if we fail.
3099         */
3100         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3101 nodeleg:
3102         status = nfs_ok;
3103
3104         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3105                 STATEID_VAL(&stp->st_stid.sc_stateid));
3106 out:
3107         /* 4.1 client trying to upgrade/downgrade delegation? */
3108         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3109             open->op_deleg_want)
3110                 nfsd4_deleg_xgrade_none_ext(open, dp);
3111
3112         if (fp)
3113                 put_nfs4_file(fp);
3114         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3115                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3116         /*
3117         * To finish the open response, we just need to set the rflags.
3118         */
3119         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3120         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3121             !nfsd4_has_session(&resp->cstate))
3122                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3123
3124         return status;
3125 }
3126
3127 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3128 {
3129         if (open->op_openowner) {
3130                 struct nfs4_openowner *oo = open->op_openowner;
3131
3132                 if (!list_empty(&oo->oo_owner.so_stateids))
3133                         list_del_init(&oo->oo_close_lru);
3134                 if (oo->oo_flags & NFS4_OO_NEW) {
3135                         if (status) {
3136                                 release_openowner(oo);
3137                                 open->op_openowner = NULL;
3138                         } else
3139                                 oo->oo_flags &= ~NFS4_OO_NEW;
3140                 }
3141         }
3142         if (open->op_file)
3143                 nfsd4_free_file(open->op_file);
3144         if (open->op_stp)
3145                 free_generic_stateid(open->op_stp);
3146 }
3147
3148 __be32
3149 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3150             clientid_t *clid)
3151 {
3152         struct nfs4_client *clp;
3153         __be32 status;
3154         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
3155
3156         nfs4_lock_state();
3157         dprintk("process_renew(%08x/%08x): starting\n", 
3158                         clid->cl_boot, clid->cl_id);
3159         status = nfserr_stale_clientid;
3160         if (STALE_CLIENTID(clid, nn))
3161                 goto out;
3162         clp = find_confirmed_client(clid, cstate->minorversion);
3163         status = nfserr_expired;
3164         if (clp == NULL) {
3165                 /* We assume the client took too long to RENEW. */
3166                 dprintk("nfsd4_renew: clientid not found!\n");
3167                 goto out;
3168         }
3169         status = nfserr_cb_path_down;
3170         if (!list_empty(&clp->cl_delegations)
3171                         && clp->cl_cb_state != NFSD4_CB_UP)
3172                 goto out;
3173         status = nfs_ok;
3174 out:
3175         nfs4_unlock_state();
3176         return status;
3177 }
3178
3179 static void
3180 nfsd4_end_grace(struct net *net)
3181 {
3182         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3183
3184         /* do nothing if grace period already ended */
3185         if (nn->grace_ended)
3186                 return;
3187
3188         dprintk("NFSD: end of grace period\n");
3189         nn->grace_ended = true;
3190         nfsd4_record_grace_done(net, nn->boot_time);
3191         locks_end_grace(&nn->nfsd4_manager);
3192         /*
3193          * Now that every NFSv4 client has had the chance to recover and
3194          * to see the (possibly new, possibly shorter) lease time, we
3195          * can safely set the next grace time to the current lease time:
3196          */
3197         nfsd4_grace = nfsd4_lease;
3198 }
3199
3200 static time_t
3201 nfs4_laundromat(void)
3202 {
3203         struct nfs4_client *clp;
3204         struct nfs4_openowner *oo;
3205         struct nfs4_delegation *dp;
3206         struct list_head *pos, *next, reaplist;
3207         time_t cutoff = get_seconds() - nfsd4_lease;
3208         time_t t, clientid_val = nfsd4_lease;
3209         time_t u, test_val = nfsd4_lease;
3210
3211         nfs4_lock_state();
3212
3213         dprintk("NFSD: laundromat service - starting\n");
3214         nfsd4_end_grace(&init_net);
3215         INIT_LIST_HEAD(&reaplist);
3216         spin_lock(&client_lock);
3217         list_for_each_safe(pos, next, &client_lru) {
3218                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3219                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3220                         t = clp->cl_time - cutoff;
3221                         if (clientid_val > t)
3222                                 clientid_val = t;
3223                         break;
3224                 }
3225                 if (atomic_read(&clp->cl_refcount)) {
3226                         dprintk("NFSD: client in use (clientid %08x)\n",
3227                                 clp->cl_clientid.cl_id);
3228                         continue;
3229                 }
3230                 unhash_client_locked(clp);
3231                 list_add(&clp->cl_lru, &reaplist);
3232         }
3233         spin_unlock(&client_lock);
3234         list_for_each_safe(pos, next, &reaplist) {
3235                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3236                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3237                         clp->cl_clientid.cl_id);
3238                 expire_client(clp);
3239         }
3240         spin_lock(&recall_lock);
3241         list_for_each_safe(pos, next, &del_recall_lru) {
3242                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3243                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3244                         u = dp->dl_time - cutoff;
3245                         if (test_val > u)
3246                                 test_val = u;
3247                         break;
3248                 }
3249                 list_move(&dp->dl_recall_lru, &reaplist);
3250         }
3251         spin_unlock(&recall_lock);
3252         list_for_each_safe(pos, next, &reaplist) {
3253                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3254                 unhash_delegation(dp);
3255         }
3256         test_val = nfsd4_lease;
3257         list_for_each_safe(pos, next, &close_lru) {
3258                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3259                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3260                         u = oo->oo_time - cutoff;
3261                         if (test_val > u)
3262                                 test_val = u;
3263                         break;
3264                 }
3265                 release_openowner(oo);
3266         }
3267         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3268                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3269         nfs4_unlock_state();
3270         return clientid_val;
3271 }
3272
3273 static struct workqueue_struct *laundry_wq;
3274 static void laundromat_main(struct work_struct *);
3275 static DECLARE_DELAYED_WORK(laundromat_work, laundromat_main);
3276
3277 static void
3278 laundromat_main(struct work_struct *not_used)
3279 {
3280         time_t t;
3281
3282         t = nfs4_laundromat();
3283         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3284         queue_delayed_work(laundry_wq, &laundromat_work, t*HZ);
3285 }
3286
3287 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3288 {
3289         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3290                 return nfserr_bad_stateid;
3291         return nfs_ok;
3292 }
3293
3294 static int
3295 STALE_STATEID(stateid_t *stateid, struct nfsd_net *nn)
3296 {
3297         if (stateid->si_opaque.so_clid.cl_boot == nn->boot_time)
3298                 return 0;
3299         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3300                 STATEID_VAL(stateid));
3301         return 1;
3302 }
3303
3304 static inline int
3305 access_permit_read(struct nfs4_ol_stateid *stp)
3306 {
3307         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3308                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3309                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3310 }
3311
3312 static inline int
3313 access_permit_write(struct nfs4_ol_stateid *stp)
3314 {
3315         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3316                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3317 }
3318
3319 static
3320 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3321 {
3322         __be32 status = nfserr_openmode;
3323
3324         /* For lock stateid's, we test the parent open, not the lock: */
3325         if (stp->st_openstp)
3326                 stp = stp->st_openstp;
3327         if ((flags & WR_STATE) && !access_permit_write(stp))
3328                 goto out;
3329         if ((flags & RD_STATE) && !access_permit_read(stp))
3330                 goto out;
3331         status = nfs_ok;
3332 out:
3333         return status;
3334 }
3335
3336 static inline __be32
3337 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3338 {
3339         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3340                 return nfs_ok;
3341         else if (locks_in_grace(net)) {
3342                 /* Answer in remaining cases depends on existence of
3343                  * conflicting state; so we must wait out the grace period. */
3344                 return nfserr_grace;
3345         } else if (flags & WR_STATE)
3346                 return nfs4_share_conflict(current_fh,
3347                                 NFS4_SHARE_DENY_WRITE);
3348         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3349                 return nfs4_share_conflict(current_fh,
3350                                 NFS4_SHARE_DENY_READ);
3351 }
3352
3353 /*
3354  * Allow READ/WRITE during grace period on recovered state only for files
3355  * that are not able to provide mandatory locking.
3356  */
3357 static inline int
3358 grace_disallows_io(struct net *net, struct inode *inode)
3359 {
3360         return locks_in_grace(net) && mandatory_lock(inode);
3361 }
3362
3363 /* Returns true iff a is later than b: */
3364 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3365 {
3366         return (s32)a->si_generation - (s32)b->si_generation > 0;
3367 }
3368
3369 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3370 {
3371         /*
3372          * When sessions are used the stateid generation number is ignored
3373          * when it is zero.
3374          */
3375         if (has_session && in->si_generation == 0)
3376                 return nfs_ok;
3377
3378         if (in->si_generation == ref->si_generation)
3379                 return nfs_ok;
3380
3381         /* If the client sends us a stateid from the future, it's buggy: */
3382         if (stateid_generation_after(in, ref))
3383                 return nfserr_bad_stateid;
3384         /*
3385          * However, we could see a stateid from the past, even from a
3386          * non-buggy client.  For example, if the client sends a lock
3387          * while some IO is outstanding, the lock may bump si_generation
3388          * while the IO is still in flight.  The client could avoid that
3389          * situation by waiting for responses on all the IO requests,
3390          * but better performance may result in retrying IO that
3391          * receives an old_stateid error if requests are rarely
3392          * reordered in flight:
3393          */
3394         return nfserr_old_stateid;
3395 }
3396
3397 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3398 {
3399         struct nfs4_stid *s;
3400         struct nfs4_ol_stateid *ols;
3401         __be32 status;
3402
3403         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3404                 return nfserr_bad_stateid;
3405         /* Client debugging aid. */
3406         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3407                 char addr_str[INET6_ADDRSTRLEN];
3408                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3409                                  sizeof(addr_str));
3410                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3411                                         "with incorrect client ID\n", addr_str);
3412                 return nfserr_bad_stateid;
3413         }
3414         s = find_stateid(cl, stateid);
3415         if (!s)
3416                 return nfserr_bad_stateid;
3417         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3418         if (status)
3419                 return status;
3420         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3421                 return nfs_ok;
3422         ols = openlockstateid(s);
3423         if (ols->st_stateowner->so_is_open_owner
3424             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3425                 return nfserr_bad_stateid;
3426         return nfs_ok;
3427 }
3428
3429 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask, struct nfs4_stid **s, bool sessions)
3430 {
3431         struct nfs4_client *cl;
3432         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
3433
3434         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3435                 return nfserr_bad_stateid;
3436         if (STALE_STATEID(stateid, nn))
3437                 return nfserr_stale_stateid;
3438         cl = find_confirmed_client(&stateid->si_opaque.so_clid, sessions);
3439         if (!cl)
3440                 return nfserr_expired;
3441         *s = find_stateid_by_type(cl, stateid, typemask);
3442         if (!*s)
3443                 return nfserr_bad_stateid;
3444         return nfs_ok;
3445
3446 }
3447
3448 /*
3449 * Checks for stateid operations
3450 */
3451 __be32
3452 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3453                            stateid_t *stateid, int flags, struct file **filpp)
3454 {
3455         struct nfs4_stid *s;
3456         struct nfs4_ol_stateid *stp = NULL;
3457         struct nfs4_delegation *dp = NULL;
3458         struct svc_fh *current_fh = &cstate->current_fh;
3459         struct inode *ino = current_fh->fh_dentry->d_inode;
3460         __be32 status;
3461
3462         if (filpp)
3463                 *filpp = NULL;
3464
3465         if (grace_disallows_io(net, ino))
3466                 return nfserr_grace;
3467
3468         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3469                 return check_special_stateids(net, current_fh, stateid, flags);
3470
3471         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID, &s, cstate->minorversion);
3472         if (status)
3473                 return status;
3474         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3475         if (status)
3476                 goto out;
3477         switch (s->sc_type) {
3478         case NFS4_DELEG_STID:
3479                 dp = delegstateid(s);
3480                 status = nfs4_check_delegmode(dp, flags);
3481                 if (status)
3482                         goto out;
3483                 if (filpp) {
3484                         *filpp = dp->dl_file->fi_deleg_file;
3485                         BUG_ON(!*filpp);
3486                 }
3487                 break;
3488         case NFS4_OPEN_STID:
3489         case NFS4_LOCK_STID:
3490                 stp = openlockstateid(s);
3491                 status = nfs4_check_fh(current_fh, stp);
3492                 if (status)
3493                         goto out;
3494                 if (stp->st_stateowner->so_is_open_owner
3495                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3496                         goto out;
3497                 status = nfs4_check_openmode(stp, flags);
3498                 if (status)
3499                         goto out;
3500                 if (filpp) {
3501                         if (flags & RD_STATE)
3502                                 *filpp = find_readable_file(stp->st_file);
3503                         else
3504                                 *filpp = find_writeable_file(stp->st_file);
3505                 }
3506                 break;
3507         default:
3508                 return nfserr_bad_stateid;
3509         }
3510         status = nfs_ok;
3511 out:
3512         return status;
3513 }
3514
3515 static __be32
3516 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3517 {
3518         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3519                 return nfserr_locks_held;
3520         release_lock_stateid(stp);
3521         return nfs_ok;
3522 }
3523
3524 /*
3525  * Test if the stateid is valid
3526  */
3527 __be32
3528 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3529                    struct nfsd4_test_stateid *test_stateid)
3530 {
3531         struct nfsd4_test_stateid_id *stateid;
3532         struct nfs4_client *cl = cstate->session->se_client;
3533
3534         nfs4_lock_state();
3535         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3536                 stateid->ts_id_status =
3537                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3538         nfs4_unlock_state();
3539
3540         return nfs_ok;
3541 }
3542
3543 __be32
3544 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3545                    struct nfsd4_free_stateid *free_stateid)
3546 {
3547         stateid_t *stateid = &free_stateid->fr_stateid;
3548         struct nfs4_stid *s;
3549         struct nfs4_client *cl = cstate->session->se_client;
3550         __be32 ret = nfserr_bad_stateid;
3551
3552         nfs4_lock_state();
3553         s = find_stateid(cl, stateid);
3554         if (!s)
3555                 goto out;
3556         switch (s->sc_type) {
3557         case NFS4_DELEG_STID:
3558                 ret = nfserr_locks_held;
3559                 goto out;
3560         case NFS4_OPEN_STID:
3561         case NFS4_LOCK_STID:
3562                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3563                 if (ret)
3564                         goto out;
3565                 if (s->sc_type == NFS4_LOCK_STID)
3566                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3567                 else
3568                         ret = nfserr_locks_held;
3569                 break;
3570         default:
3571                 ret = nfserr_bad_stateid;
3572         }
3573 out:
3574         nfs4_unlock_state();
3575         return ret;
3576 }
3577
3578 static inline int
3579 setlkflg (int type)
3580 {
3581         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3582                 RD_STATE : WR_STATE;
3583 }
3584
3585 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3586 {
3587         struct svc_fh *current_fh = &cstate->current_fh;
3588         struct nfs4_stateowner *sop = stp->st_stateowner;
3589         __be32 status;
3590
3591         status = nfsd4_check_seqid(cstate, sop, seqid);
3592         if (status)
3593                 return status;
3594         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3595                 /*
3596                  * "Closed" stateid's exist *only* to return
3597                  * nfserr_replay_me from the previous step.
3598                  */
3599                 return nfserr_bad_stateid;
3600         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3601         if (status)
3602                 return status;
3603         return nfs4_check_fh(current_fh, stp);
3604 }
3605
3606 /* 
3607  * Checks for sequence id mutating operations. 
3608  */
3609 static __be32
3610 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3611                          stateid_t *stateid, char typemask,
3612                          struct nfs4_ol_stateid **stpp)
3613 {
3614         __be32 status;
3615         struct nfs4_stid *s;
3616
3617         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3618                 seqid, STATEID_VAL(stateid));
3619
3620         *stpp = NULL;
3621         status = nfsd4_lookup_stateid(stateid, typemask, &s, cstate->minorversion);
3622         if (status)
3623                 return status;
3624         *stpp = openlockstateid(s);
3625         cstate->replay_owner = (*stpp)->st_stateowner;
3626
3627         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3628 }
3629
3630 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid, stateid_t *stateid, struct nfs4_ol_stateid **stpp)
3631 {
3632         __be32 status;
3633         struct nfs4_openowner *oo;
3634
3635         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3636                                                 NFS4_OPEN_STID, stpp);
3637         if (status)
3638                 return status;
3639         oo = openowner((*stpp)->st_stateowner);
3640         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3641                 return nfserr_bad_stateid;
3642         return nfs_ok;
3643 }
3644
3645 __be32
3646 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3647                    struct nfsd4_open_confirm *oc)
3648 {
3649         __be32 status;
3650         struct nfs4_openowner *oo;
3651         struct nfs4_ol_stateid *stp;
3652
3653         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3654                         (int)cstate->current_fh.fh_dentry->d_name.len,
3655                         cstate->current_fh.fh_dentry->d_name.name);
3656
3657         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3658         if (status)
3659                 return status;
3660
3661         nfs4_lock_state();
3662
3663         status = nfs4_preprocess_seqid_op(cstate,
3664                                         oc->oc_seqid, &oc->oc_req_stateid,
3665                                         NFS4_OPEN_STID, &stp);
3666         if (status)
3667                 goto out;
3668         oo = openowner(stp->st_stateowner);
3669         status = nfserr_bad_stateid;
3670         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3671                 goto out;
3672         oo->oo_flags |= NFS4_OO_CONFIRMED;
3673         update_stateid(&stp->st_stid.sc_stateid);
3674         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3675         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3676                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3677
3678         nfsd4_client_record_create(oo->oo_owner.so_client);
3679         status = nfs_ok;
3680 out:
3681         if (!cstate->replay_owner)
3682                 nfs4_unlock_state();
3683         return status;
3684 }
3685
3686 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3687 {
3688         if (!test_access(access, stp))
3689                 return;
3690         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3691         clear_access(access, stp);
3692 }
3693
3694 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3695 {
3696         switch (to_access) {
3697         case NFS4_SHARE_ACCESS_READ:
3698                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3699                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3700                 break;
3701         case NFS4_SHARE_ACCESS_WRITE:
3702                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3703                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3704                 break;
3705         case NFS4_SHARE_ACCESS_BOTH:
3706                 break;
3707         default:
3708                 BUG();
3709         }
3710 }
3711
3712 static void
3713 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3714 {
3715         int i;
3716         for (i = 0; i < 4; i++) {
3717                 if ((i & deny) != i)
3718                         clear_deny(i, stp);
3719         }
3720 }
3721
3722 __be32
3723 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3724                      struct nfsd4_compound_state *cstate,
3725                      struct nfsd4_open_downgrade *od)
3726 {
3727         __be32 status;
3728         struct nfs4_ol_stateid *stp;
3729
3730         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3731                         (int)cstate->current_fh.fh_dentry->d_name.len,
3732                         cstate->current_fh.fh_dentry->d_name.name);
3733
3734         /* We don't yet support WANT bits: */
3735         if (od->od_deleg_want)
3736                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3737                         od->od_deleg_want);
3738
3739         nfs4_lock_state();
3740         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3741                                         &od->od_stateid, &stp);
3742         if (status)
3743                 goto out; 
3744         status = nfserr_inval;
3745         if (!test_access(od->od_share_access, stp)) {
3746                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3747                         stp->st_access_bmap, od->od_share_access);
3748                 goto out;
3749         }
3750         if (!test_deny(od->od_share_deny, stp)) {
3751                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3752                         stp->st_deny_bmap, od->od_share_deny);
3753                 goto out;
3754         }
3755         nfs4_stateid_downgrade(stp, od->od_share_access);
3756
3757         reset_union_bmap_deny(od->od_share_deny, stp);
3758
3759         update_stateid(&stp->st_stid.sc_stateid);
3760         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3761         status = nfs_ok;
3762 out:
3763         if (!cstate->replay_owner)
3764                 nfs4_unlock_state();
3765         return status;
3766 }
3767
3768 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3769 {
3770         struct nfs4_openowner *oo;
3771         struct nfs4_ol_stateid *s;
3772
3773         if (!so->so_is_open_owner)
3774                 return;
3775         oo = openowner(so);
3776         s = oo->oo_last_closed_stid;
3777         if (!s)
3778                 return;
3779         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3780                 /* Release the last_closed_stid on the next seqid bump: */
3781                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3782                 return;
3783         }
3784         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3785         release_last_closed_stateid(oo);
3786 }
3787
3788 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3789 {
3790         unhash_open_stateid(s);
3791         s->st_stid.sc_type = NFS4_CLOSED_STID;
3792 }
3793
3794 /*
3795  * nfs4_unlock_state() called after encode
3796  */
3797 __be32
3798 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3799             struct nfsd4_close *close)
3800 {
3801         __be32 status;
3802         struct nfs4_openowner *oo;
3803         struct nfs4_ol_stateid *stp;
3804
3805         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3806                         (int)cstate->current_fh.fh_dentry->d_name.len,
3807                         cstate->current_fh.fh_dentry->d_name.name);
3808
3809         nfs4_lock_state();
3810         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3811                                         &close->cl_stateid,
3812                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3813                                         &stp);
3814         if (status)
3815                 goto out; 
3816         oo = openowner(stp->st_stateowner);
3817         status = nfs_ok;
3818         update_stateid(&stp->st_stid.sc_stateid);
3819         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3820
3821         nfsd4_close_open_stateid(stp);
3822         release_last_closed_stateid(oo);
3823         oo->oo_last_closed_stid = stp;
3824
3825         if (list_empty(&oo->oo_owner.so_stateids)) {
3826                 if (cstate->minorversion) {
3827                         release_openowner(oo);
3828                         cstate->replay_owner = NULL;
3829                 } else {
3830                         /*
3831                          * In the 4.0 case we need to keep the owners around a
3832                          * little while to handle CLOSE replay.
3833                          */
3834                         if (list_empty(&oo->oo_owner.so_stateids))
3835                                 move_to_close_lru(oo);
3836                 }
3837         }
3838 out:
3839         if (!cstate->replay_owner)
3840                 nfs4_unlock_state();
3841         return status;
3842 }
3843
3844 __be32
3845 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3846                   struct nfsd4_delegreturn *dr)
3847 {
3848         struct nfs4_delegation *dp;
3849         stateid_t *stateid = &dr->dr_stateid;
3850         struct nfs4_stid *s;
3851         __be32 status;
3852
3853         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3854                 return status;
3855
3856         nfs4_lock_state();
3857         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s, cstate->minorversion);
3858         if (status)
3859                 goto out;
3860         dp = delegstateid(s);
3861         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3862         if (status)
3863                 goto out;
3864
3865         unhash_delegation(dp);
3866 out:
3867         nfs4_unlock_state();
3868
3869         return status;
3870 }
3871
3872
3873 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3874
3875 #define LOCKOWNER_INO_HASH_BITS 8
3876 #define LOCKOWNER_INO_HASH_SIZE (1 << LOCKOWNER_INO_HASH_BITS)
3877 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3878
3879 static inline u64
3880 end_offset(u64 start, u64 len)
3881 {
3882         u64 end;
3883
3884         end = start + len;
3885         return end >= start ? end: NFS4_MAX_UINT64;
3886 }
3887
3888 /* last octet in a range */
3889 static inline u64
3890 last_byte_offset(u64 start, u64 len)
3891 {
3892         u64 end;
3893
3894         BUG_ON(!len);
3895         end = start + len;
3896         return end > start ? end - 1: NFS4_MAX_UINT64;
3897 }
3898
3899 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3900 {
3901         return (file_hashval(inode) + cl_id
3902                         + opaque_hashval(ownername->data, ownername->len))
3903                 & LOCKOWNER_INO_HASH_MASK;
3904 }
3905
3906 static struct list_head lockowner_ino_hashtbl[LOCKOWNER_INO_HASH_SIZE];
3907
3908 /*
3909  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3910  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3911  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3912  * locking, this prevents us from being completely protocol-compliant.  The
3913  * real solution to this problem is to start using unsigned file offsets in
3914  * the VFS, but this is a very deep change!
3915  */
3916 static inline void
3917 nfs4_transform_lock_offset(struct file_lock *lock)
3918 {
3919         if (lock->fl_start < 0)
3920                 lock->fl_start = OFFSET_MAX;
3921         if (lock->fl_end < 0)
3922                 lock->fl_end = OFFSET_MAX;
3923 }
3924
3925 /* Hack!: For now, we're defining this just so we can use a pointer to it
3926  * as a unique cookie to identify our (NFSv4's) posix locks. */
3927 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3928 };
3929
3930 static inline void
3931 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3932 {
3933         struct nfs4_lockowner *lo;
3934
3935         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3936                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3937                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3938                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3939                 if (!deny->ld_owner.data)
3940                         /* We just don't care that much */
3941                         goto nevermind;
3942                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3943                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3944         } else {
3945 nevermind:
3946                 deny->ld_owner.len = 0;
3947                 deny->ld_owner.data = NULL;
3948                 deny->ld_clientid.cl_boot = 0;
3949                 deny->ld_clientid.cl_id = 0;
3950         }
3951         deny->ld_start = fl->fl_start;
3952         deny->ld_length = NFS4_MAX_UINT64;
3953         if (fl->fl_end != NFS4_MAX_UINT64)
3954                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3955         deny->ld_type = NFS4_READ_LT;
3956         if (fl->fl_type != F_RDLCK)
3957                 deny->ld_type = NFS4_WRITE_LT;
3958 }
3959
3960 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3961 {
3962         struct nfs4_ol_stateid *lst;
3963
3964         if (!same_owner_str(&lo->lo_owner, owner, clid))
3965                 return false;
3966         lst = list_first_entry(&lo->lo_owner.so_stateids,
3967                                struct nfs4_ol_stateid, st_perstateowner);
3968         return lst->st_file->fi_inode == inode;
3969 }
3970
3971 static struct nfs4_lockowner *
3972 find_lockowner_str(struct inode *inode, clientid_t *clid,
3973                 struct xdr_netobj *owner)
3974 {
3975         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
3976         struct nfs4_lockowner *lo;
3977
3978         list_for_each_entry(lo, &lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
3979                 if (same_lockowner_ino(lo, inode, clid, owner))
3980                         return lo;
3981         }
3982         return NULL;
3983 }
3984
3985 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
3986 {
3987         struct inode *inode = open_stp->st_file->fi_inode;
3988         unsigned int inohash = lockowner_ino_hashval(inode,
3989                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
3990
3991         list_add(&lo->lo_owner.so_strhash, &ownerstr_hashtbl[strhashval]);
3992         list_add(&lo->lo_owner_ino_hash, &lockowner_ino_hashtbl[inohash]);
3993         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
3994 }
3995
3996 /*
3997  * Alloc a lock owner structure.
3998  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
3999  * occurred. 
4000  *
4001  * strhashval = ownerstr_hashval
4002  */
4003
4004 static struct nfs4_lockowner *
4005 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4006         struct nfs4_lockowner *lo;
4007
4008         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4009         if (!lo)
4010                 return NULL;
4011         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4012         lo->lo_owner.so_is_open_owner = 0;
4013         /* It is the openowner seqid that will be incremented in encode in the
4014          * case of new lockowners; so increment the lock seqid manually: */
4015         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4016         hash_lockowner(lo, strhashval, clp, open_stp);
4017         return lo;
4018 }
4019
4020 static struct nfs4_ol_stateid *
4021 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4022 {
4023         struct nfs4_ol_stateid *stp;
4024         struct nfs4_client *clp = lo->lo_owner.so_client;
4025
4026         stp = nfs4_alloc_stateid(clp);
4027         if (stp == NULL)
4028                 return NULL;
4029         init_stid(&stp->st_stid, clp, NFS4_LOCK_STID);
4030         list_add(&stp->st_perfile, &fp->fi_stateids);
4031         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4032         stp->st_stateowner = &lo->lo_owner;
4033         get_nfs4_file(fp);
4034         stp->st_file = fp;
4035         stp->st_access_bmap = 0;
4036         stp->st_deny_bmap = open_stp->st_deny_bmap;
4037         stp->st_openstp = open_stp;
4038         return stp;
4039 }
4040
4041 static int
4042 check_lock_length(u64 offset, u64 length)
4043 {
4044         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4045              LOFF_OVERFLOW(offset, length)));
4046 }
4047
4048 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4049 {
4050         struct nfs4_file *fp = lock_stp->st_file;
4051         int oflag = nfs4_access_to_omode(access);
4052
4053         if (test_access(access, lock_stp))
4054                 return;
4055         nfs4_file_get_access(fp, oflag);
4056         set_access(access, lock_stp);
4057 }
4058
4059 static __be32 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate, struct nfs4_ol_stateid *ost, struct nfsd4_lock *lock, struct nfs4_ol_stateid **lst, bool *new)
4060 {
4061         struct nfs4_file *fi = ost->st_file;
4062         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4063         struct nfs4_client *cl = oo->oo_owner.so_client;
4064         struct nfs4_lockowner *lo;
4065         unsigned int strhashval;
4066
4067         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid, &lock->v.new.owner);
4068         if (lo) {
4069                 if (!cstate->minorversion)
4070                         return nfserr_bad_seqid;
4071                 /* XXX: a lockowner always has exactly one stateid: */
4072                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4073                                 struct nfs4_ol_stateid, st_perstateowner);
4074                 return nfs_ok;
4075         }
4076         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4077                         &lock->v.new.owner);
4078         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4079         if (lo == NULL)
4080                 return nfserr_jukebox;
4081         *lst = alloc_init_lock_stateid(lo, fi, ost);
4082         if (*lst == NULL) {
4083                 release_lockowner(lo);
4084                 return nfserr_jukebox;
4085         }
4086         *new = true;
4087         return nfs_ok;
4088 }
4089
4090 /*
4091  *  LOCK operation 
4092  */
4093 __be32
4094 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4095            struct nfsd4_lock *lock)
4096 {
4097         struct nfs4_openowner *open_sop = NULL;
4098         struct nfs4_lockowner *lock_sop = NULL;
4099         struct nfs4_ol_stateid *lock_stp;
4100         struct file *filp = NULL;
4101         struct file_lock *file_lock = NULL;
4102         struct file_lock *conflock = NULL;
4103         __be32 status = 0;
4104         bool new_state = false;
4105         int lkflg;
4106         int err;
4107         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
4108
4109         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4110                 (long long) lock->lk_offset,
4111                 (long long) lock->lk_length);
4112
4113         if (check_lock_length(lock->lk_offset, lock->lk_length))
4114                  return nfserr_inval;
4115
4116         if ((status = fh_verify(rqstp, &cstate->current_fh,
4117                                 S_IFREG, NFSD_MAY_LOCK))) {
4118                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4119                 return status;
4120         }
4121
4122         nfs4_lock_state();
4123
4124         if (lock->lk_is_new) {
4125                 struct nfs4_ol_stateid *open_stp = NULL;
4126
4127                 if (nfsd4_has_session(cstate))
4128                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4129                         memcpy(&lock->v.new.clientid,
4130                                 &cstate->session->se_client->cl_clientid,
4131                                 sizeof(clientid_t));
4132
4133                 status = nfserr_stale_clientid;
4134                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4135                         goto out;
4136
4137                 /* validate and update open stateid and open seqid */
4138                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4139                                         lock->lk_new_open_seqid,
4140                                         &lock->lk_new_open_stateid,
4141                                         &open_stp);
4142                 if (status)
4143                         goto out;
4144                 open_sop = openowner(open_stp->st_stateowner);
4145                 status = nfserr_bad_stateid;
4146                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4147                                                 &lock->v.new.clientid))
4148                         goto out;
4149                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4150                                                         &lock_stp, &new_state);
4151         } else
4152                 status = nfs4_preprocess_seqid_op(cstate,
4153                                        lock->lk_old_lock_seqid,
4154                                        &lock->lk_old_lock_stateid,
4155                                        NFS4_LOCK_STID, &lock_stp);
4156         if (status)
4157                 goto out;
4158         lock_sop = lockowner(lock_stp->st_stateowner);
4159
4160         lkflg = setlkflg(lock->lk_type);
4161         status = nfs4_check_openmode(lock_stp, lkflg);
4162         if (status)
4163                 goto out;
4164
4165         status = nfserr_grace;
4166         if (locks_in_grace(SVC_NET(rqstp)) && !lock->lk_reclaim)
4167                 goto out;
4168         status = nfserr_no_grace;
4169         if (!locks_in_grace(SVC_NET(rqstp)) && lock->lk_reclaim)
4170                 goto out;
4171
4172         file_lock = locks_alloc_lock();
4173         if (!file_lock) {
4174                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4175                 status = nfserr_jukebox;
4176                 goto out;
4177         }
4178
4179         locks_init_lock(file_lock);
4180         switch (lock->lk_type) {
4181                 case NFS4_READ_LT:
4182                 case NFS4_READW_LT:
4183                         filp = find_readable_file(lock_stp->st_file);
4184                         if (filp)
4185                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4186                         file_lock->fl_type = F_RDLCK;
4187                         break;
4188                 case NFS4_WRITE_LT:
4189                 case NFS4_WRITEW_LT:
4190                         filp = find_writeable_file(lock_stp->st_file);
4191                         if (filp)
4192                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4193                         file_lock->fl_type = F_WRLCK;
4194                         break;
4195                 default:
4196                         status = nfserr_inval;
4197                 goto out;
4198         }
4199         if (!filp) {
4200                 status = nfserr_openmode;
4201                 goto out;
4202         }
4203         file_lock->fl_owner = (fl_owner_t)lock_sop;
4204         file_lock->fl_pid = current->tgid;
4205         file_lock->fl_file = filp;
4206         file_lock->fl_flags = FL_POSIX;
4207         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4208         file_lock->fl_start = lock->lk_offset;
4209         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4210         nfs4_transform_lock_offset(file_lock);
4211
4212         conflock = locks_alloc_lock();
4213         if (!conflock) {
4214                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4215                 status = nfserr_jukebox;
4216                 goto out;
4217         }
4218
4219         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4220         switch (-err) {
4221         case 0: /* success! */
4222                 update_stateid(&lock_stp->st_stid.sc_stateid);
4223                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid, 
4224                                 sizeof(stateid_t));
4225                 status = 0;
4226                 break;
4227         case (EAGAIN):          /* conflock holds conflicting lock */
4228                 status = nfserr_denied;
4229                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4230                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4231                 break;
4232         case (EDEADLK):
4233                 status = nfserr_deadlock;
4234                 break;
4235         default:
4236                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4237                 status = nfserrno(err);
4238                 break;
4239         }
4240 out:
4241         if (status && new_state)
4242                 release_lockowner(lock_sop);
4243         if (!cstate->replay_owner)
4244                 nfs4_unlock_state();
4245         if (file_lock)
4246                 locks_free_lock(file_lock);
4247         if (conflock)
4248                 locks_free_lock(conflock);
4249         return status;
4250 }
4251
4252 /*
4253  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4254  * so we do a temporary open here just to get an open file to pass to
4255  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4256  * inode operation.)
4257  */
4258 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4259 {
4260         struct file *file;
4261         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4262         if (!err) {
4263                 err = nfserrno(vfs_test_lock(file, lock));
4264                 nfsd_close(file);
4265         }
4266         return err;
4267 }
4268
4269 /*
4270  * LOCKT operation
4271  */
4272 __be32
4273 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4274             struct nfsd4_lockt *lockt)
4275 {
4276         struct inode *inode;
4277         struct file_lock *file_lock = NULL;
4278         struct nfs4_lockowner *lo;
4279         __be32 status;
4280         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
4281
4282         if (locks_in_grace(SVC_NET(rqstp)))
4283                 return nfserr_grace;
4284
4285         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4286                  return nfserr_inval;
4287
4288         nfs4_lock_state();
4289
4290         status = nfserr_stale_clientid;
4291         if (!nfsd4_has_session(cstate) && STALE_CLIENTID(&lockt->lt_clientid, nn))
4292                 goto out;
4293
4294         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4295                 goto out;
4296
4297         inode = cstate->current_fh.fh_dentry->d_inode;
4298         file_lock = locks_alloc_lock();
4299         if (!file_lock) {
4300                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4301                 status = nfserr_jukebox;
4302                 goto out;
4303         }
4304         locks_init_lock(file_lock);
4305         switch (lockt->lt_type) {
4306                 case NFS4_READ_LT:
4307                 case NFS4_READW_LT:
4308                         file_lock->fl_type = F_RDLCK;
4309                 break;
4310                 case NFS4_WRITE_LT:
4311                 case NFS4_WRITEW_LT:
4312                         file_lock->fl_type = F_WRLCK;
4313                 break;
4314                 default:
4315                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4316                         status = nfserr_inval;
4317                 goto out;
4318         }
4319
4320         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner);
4321         if (lo)
4322                 file_lock->fl_owner = (fl_owner_t)lo;
4323         file_lock->fl_pid = current->tgid;
4324         file_lock->fl_flags = FL_POSIX;
4325
4326         file_lock->fl_start = lockt->lt_offset;
4327         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4328
4329         nfs4_transform_lock_offset(file_lock);
4330
4331         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4332         if (status)
4333                 goto out;
4334
4335         if (file_lock->fl_type != F_UNLCK) {
4336                 status = nfserr_denied;
4337                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4338         }
4339 out:
4340         nfs4_unlock_state();
4341         if (file_lock)
4342                 locks_free_lock(file_lock);
4343         return status;
4344 }
4345
4346 __be32
4347 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4348             struct nfsd4_locku *locku)
4349 {
4350         struct nfs4_ol_stateid *stp;
4351         struct file *filp = NULL;
4352         struct file_lock *file_lock = NULL;
4353         __be32 status;
4354         int err;
4355                                                         
4356         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4357                 (long long) locku->lu_offset,
4358                 (long long) locku->lu_length);
4359
4360         if (check_lock_length(locku->lu_offset, locku->lu_length))
4361                  return nfserr_inval;
4362
4363         nfs4_lock_state();
4364                                                                                 
4365         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4366                                         &locku->lu_stateid, NFS4_LOCK_STID, &stp);
4367         if (status)
4368                 goto out;
4369         filp = find_any_file(stp->st_file);
4370         if (!filp) {
4371                 status = nfserr_lock_range;
4372                 goto out;
4373         }
4374         file_lock = locks_alloc_lock();
4375         if (!file_lock) {
4376                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4377                 status = nfserr_jukebox;
4378                 goto out;
4379         }
4380         locks_init_lock(file_lock);
4381         file_lock->fl_type = F_UNLCK;
4382         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4383         file_lock->fl_pid = current->tgid;
4384         file_lock->fl_file = filp;
4385         file_lock->fl_flags = FL_POSIX;
4386         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4387         file_lock->fl_start = locku->lu_offset;
4388
4389         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4390                                                 locku->lu_length);
4391         nfs4_transform_lock_offset(file_lock);
4392
4393         /*
4394         *  Try to unlock the file in the VFS.
4395         */
4396         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4397         if (err) {
4398                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4399                 goto out_nfserr;
4400         }
4401         /*
4402         * OK, unlock succeeded; the only thing left to do is update the stateid.
4403         */
4404         update_stateid(&stp->st_stid.sc_stateid);
4405         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4406
4407 out:
4408         if (!cstate->replay_owner)
4409                 nfs4_unlock_state();
4410         if (file_lock)
4411                 locks_free_lock(file_lock);
4412         return status;
4413
4414 out_nfserr:
4415         status = nfserrno(err);
4416         goto out;
4417 }
4418
4419 /*
4420  * returns
4421  *      1: locks held by lockowner
4422  *      0: no locks held by lockowner
4423  */
4424 static int
4425 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4426 {
4427         struct file_lock **flpp;
4428         struct inode *inode = filp->fi_inode;
4429         int status = 0;
4430
4431         lock_flocks();
4432         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4433                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4434                         status = 1;
4435                         goto out;
4436                 }
4437         }
4438 out:
4439         unlock_flocks();
4440         return status;
4441 }
4442
4443 __be32
4444 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4445                         struct nfsd4_compound_state *cstate,
4446                         struct nfsd4_release_lockowner *rlockowner)
4447 {
4448         clientid_t *clid = &rlockowner->rl_clientid;
4449         struct nfs4_stateowner *sop;
4450         struct nfs4_lockowner *lo;
4451         struct nfs4_ol_stateid *stp;
4452         struct xdr_netobj *owner = &rlockowner->rl_owner;
4453         struct list_head matches;
4454         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4455         __be32 status;
4456         struct nfsd_net *nn = net_generic(&init_net, nfsd_net_id);
4457
4458         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4459                 clid->cl_boot, clid->cl_id);
4460
4461         /* XXX check for lease expiration */
4462
4463         status = nfserr_stale_clientid;
4464         if (STALE_CLIENTID(clid, nn))
4465                 return status;
4466
4467         nfs4_lock_state();
4468
4469         status = nfserr_locks_held;
4470         INIT_LIST_HEAD(&matches);
4471
4472         list_for_each_entry(sop, &ownerstr_hashtbl[hashval], so_strhash) {
4473                 if (sop->so_is_open_owner)
4474                         continue;
4475                 if (!same_owner_str(sop, owner, clid))
4476                         continue;
4477                 list_for_each_entry(stp, &sop->so_stateids,
4478                                 st_perstateowner) {
4479                         lo = lockowner(sop);
4480                         if (check_for_locks(stp->st_file, lo))
4481                                 goto out;
4482                         list_add(&lo->lo_list, &matches);
4483                 }
4484         }
4485         /* Clients probably won't expect us to return with some (but not all)
4486          * of the lockowner state released; so don't release any until all
4487          * have been checked. */
4488         status = nfs_ok;
4489         while (!list_empty(&matches)) {
4490                 lo = list_entry(matches.next, struct nfs4_lockowner,
4491                                                                 lo_list);
4492                 /* unhash_stateowner deletes so_perclient only
4493                  * for openowners. */
4494                 list_del(&lo->lo_list);
4495                 release_lockowner(lo);
4496         }
4497 out:
4498         nfs4_unlock_state();
4499         return status;
4500 }
4501
4502 static inline struct nfs4_client_reclaim *
4503 alloc_reclaim(void)
4504 {
4505         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4506 }
4507
4508 bool
4509 nfs4_has_reclaimed_state(const char *name)
4510 {
4511         struct nfs4_client_reclaim *crp;
4512
4513         crp = nfsd4_find_reclaim_client(name);
4514         return (crp && crp->cr_clp);
4515 }
4516
4517 /*
4518  * failure => all reset bets are off, nfserr_no_grace...
4519  */
4520 struct nfs4_client_reclaim *
4521 nfs4_client_to_reclaim(const char *name)
4522 {
4523         unsigned int strhashval;
4524         struct nfs4_client_reclaim *crp;
4525
4526         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4527         crp = alloc_reclaim();
4528         if (crp) {
4529                 strhashval = clientstr_hashval(name);
4530                 INIT_LIST_HEAD(&crp->cr_strhash);
4531                 list_add(&crp->cr_strhash, &reclaim_str_hashtbl[strhashval]);
4532                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4533                 crp->cr_clp = NULL;
4534                 reclaim_str_hashtbl_size++;
4535         }
4536         return crp;
4537 }
4538
4539 void
4540 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp)
4541 {
4542         list_del(&crp->cr_strhash);
4543         kfree(crp);
4544         reclaim_str_hashtbl_size--;
4545 }
4546
4547 void
4548 nfs4_release_reclaim(void)
4549 {
4550         struct nfs4_client_reclaim *crp = NULL;
4551         int i;
4552
4553         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4554                 while (!list_empty(&reclaim_str_hashtbl[i])) {
4555                         crp = list_entry(reclaim_str_hashtbl[i].next,
4556                                         struct nfs4_client_reclaim, cr_strhash);
4557                         nfs4_remove_reclaim_record(crp);
4558                 }
4559         }
4560         BUG_ON(reclaim_str_hashtbl_size);
4561 }
4562
4563 /*
4564  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4565 struct nfs4_client_reclaim *
4566 nfsd4_find_reclaim_client(const char *recdir)
4567 {
4568         unsigned int strhashval;
4569         struct nfs4_client_reclaim *crp = NULL;
4570
4571         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4572
4573         strhashval = clientstr_hashval(recdir);
4574         list_for_each_entry(crp, &reclaim_str_hashtbl[strhashval], cr_strhash) {
4575                 if (same_name(crp->cr_recdir, recdir)) {
4576                         return crp;
4577                 }
4578         }
4579         return NULL;
4580 }
4581
4582 /*
4583 * Called from OPEN. Look for clientid in reclaim list.
4584 */
4585 __be32
4586 nfs4_check_open_reclaim(clientid_t *clid, bool sessions)
4587 {
4588         struct nfs4_client *clp;
4589
4590         /* find clientid in conf_id_hashtbl */
4591         clp = find_confirmed_client(clid, sessions);
4592         if (clp == NULL)
4593                 return nfserr_reclaim_bad;
4594
4595         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4596 }
4597
4598 #ifdef CONFIG_NFSD_FAULT_INJECTION
4599
4600 void nfsd_forget_clients(u64 num)
4601 {
4602         struct nfs4_client *clp, *next;
4603         int count = 0;
4604
4605         nfs4_lock_state();
4606         list_for_each_entry_safe(clp, next, &client_lru, cl_lru) {
4607                 expire_client(clp);
4608                 if (++count == num)
4609                         break;
4610         }
4611         nfs4_unlock_state();
4612
4613         printk(KERN_INFO "NFSD: Forgot %d clients", count);
4614 }
4615
4616 static void release_lockowner_sop(struct nfs4_stateowner *sop)
4617 {
4618         release_lockowner(lockowner(sop));
4619 }
4620
4621 static void release_openowner_sop(struct nfs4_stateowner *sop)
4622 {
4623         release_openowner(openowner(sop));
4624 }
4625
4626 static int nfsd_release_n_owners(u64 num, bool is_open_owner,
4627                                 void (*release_sop)(struct nfs4_stateowner *))
4628 {
4629         int i, count = 0;
4630         struct nfs4_stateowner *sop, *next;
4631
4632         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4633                 list_for_each_entry_safe(sop, next, &ownerstr_hashtbl[i], so_strhash) {
4634                         if (sop->so_is_open_owner != is_open_owner)
4635                                 continue;
4636                         release_sop(sop);
4637                         if (++count == num)
4638                                 return count;
4639                 }
4640         }
4641         return count;
4642 }
4643
4644 void nfsd_forget_locks(u64 num)
4645 {
4646         int count;
4647
4648         nfs4_lock_state();
4649         count = nfsd_release_n_owners(num, false, release_lockowner_sop);
4650         nfs4_unlock_state();
4651
4652         printk(KERN_INFO "NFSD: Forgot %d locks", count);
4653 }
4654
4655 void nfsd_forget_openowners(u64 num)
4656 {
4657         int count;
4658
4659         nfs4_lock_state();
4660         count = nfsd_release_n_owners(num, true, release_openowner_sop);
4661         nfs4_unlock_state();
4662
4663         printk(KERN_INFO "NFSD: Forgot %d open owners", count);
4664 }
4665
4666 static int nfsd_process_n_delegations(u64 num, struct list_head *list)
4667 {
4668         int i, count = 0;
4669         struct nfs4_file *fp, *fnext;
4670         struct nfs4_delegation *dp, *dnext;
4671
4672         for (i = 0; i < FILE_HASH_SIZE; i++) {
4673                 list_for_each_entry_safe(fp, fnext, &file_hashtbl[i], fi_hash) {
4674                         list_for_each_entry_safe(dp, dnext, &fp->fi_delegations, dl_perfile) {
4675                                 list_move(&dp->dl_recall_lru, list);
4676                                 if (++count == num)
4677                                         return count;
4678                         }
4679                 }
4680         }
4681
4682         return count;
4683 }
4684
4685 void nfsd_forget_delegations(u64 num)
4686 {
4687         unsigned int count;
4688         LIST_HEAD(victims);
4689         struct nfs4_delegation *dp, *dnext;
4690
4691         spin_lock(&recall_lock);
4692         count = nfsd_process_n_delegations(num, &victims);
4693         spin_unlock(&recall_lock);
4694
4695         nfs4_lock_state();
4696         list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru)
4697                 unhash_delegation(dp);
4698         nfs4_unlock_state();
4699
4700         printk(KERN_INFO "NFSD: Forgot %d delegations", count);
4701 }
4702
4703 void nfsd_recall_delegations(u64 num)
4704 {
4705         unsigned int count;
4706         LIST_HEAD(victims);
4707         struct nfs4_delegation *dp, *dnext;
4708
4709         spin_lock(&recall_lock);
4710         count = nfsd_process_n_delegations(num, &victims);
4711         list_for_each_entry_safe(dp, dnext, &victims, dl_recall_lru) {
4712                 list_del(&dp->dl_recall_lru);
4713                 nfsd_break_one_deleg(dp);
4714         }
4715         spin_unlock(&recall_lock);
4716
4717         printk(KERN_INFO "NFSD: Recalled %d delegations", count);
4718 }
4719
4720 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4721
4722 /* initialization to perform at module load time: */
4723
4724 void
4725 nfs4_state_init(void)
4726 {
4727         int i;
4728
4729         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4730                 INIT_LIST_HEAD(&conf_id_hashtbl[i]);
4731                 INIT_LIST_HEAD(&unconf_id_hashtbl[i]);
4732                 INIT_LIST_HEAD(&reclaim_str_hashtbl[i]);
4733         }
4734         conf_name_tree = RB_ROOT;
4735         unconf_name_tree = RB_ROOT;
4736         for (i = 0; i < SESSION_HASH_SIZE; i++)
4737                 INIT_LIST_HEAD(&sessionid_hashtbl[i]);
4738         for (i = 0; i < FILE_HASH_SIZE; i++) {
4739                 INIT_LIST_HEAD(&file_hashtbl[i]);
4740         }
4741         for (i = 0; i < OWNER_HASH_SIZE; i++) {
4742                 INIT_LIST_HEAD(&ownerstr_hashtbl[i]);
4743         }
4744         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4745                 INIT_LIST_HEAD(&lockowner_ino_hashtbl[i]);
4746         INIT_LIST_HEAD(&close_lru);
4747         INIT_LIST_HEAD(&client_lru);
4748         INIT_LIST_HEAD(&del_recall_lru);
4749         reclaim_str_hashtbl_size = 0;
4750 }
4751
4752 /*
4753  * Since the lifetime of a delegation isn't limited to that of an open, a
4754  * client may quite reasonably hang on to a delegation as long as it has
4755  * the inode cached.  This becomes an obvious problem the first time a
4756  * client's inode cache approaches the size of the server's total memory.
4757  *
4758  * For now we avoid this problem by imposing a hard limit on the number
4759  * of delegations, which varies according to the server's memory size.
4760  */
4761 static void
4762 set_max_delegations(void)
4763 {
4764         /*
4765          * Allow at most 4 delegations per megabyte of RAM.  Quick
4766          * estimates suggest that in the worst case (where every delegation
4767          * is for a different inode), a delegation could take about 1.5K,
4768          * giving a worst case usage of about 6% of memory.
4769          */
4770         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4771 }
4772
4773 /* initialization to perform when the nfsd service is started: */
4774
4775 int
4776 nfs4_state_start(void)
4777 {
4778         struct net *net = &init_net;
4779         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4780         int ret;
4781
4782         /*
4783          * FIXME: For now, we hang most of the pernet global stuff off of
4784          * init_net until nfsd is fully containerized. Eventually, we'll
4785          * need to pass a net pointer into this function, take a reference
4786          * to that instead and then do most of the rest of this on a per-net
4787          * basis.
4788          */
4789         get_net(net);
4790         nfsd4_client_tracking_init(net);
4791         nn->boot_time = get_seconds();
4792         locks_start_grace(net, &nn->nfsd4_manager);
4793         nn->grace_ended = false;
4794         printk(KERN_INFO "NFSD: starting %ld-second grace period\n",
4795                nfsd4_grace);
4796         ret = set_callback_cred();
4797         if (ret) {
4798                 ret = -ENOMEM;
4799                 goto out_recovery;
4800         }
4801         laundry_wq = create_singlethread_workqueue("nfsd4");
4802         if (laundry_wq == NULL) {
4803                 ret = -ENOMEM;
4804                 goto out_recovery;
4805         }
4806         ret = nfsd4_create_callback_queue();
4807         if (ret)
4808                 goto out_free_laundry;
4809         queue_delayed_work(laundry_wq, &laundromat_work, nfsd4_grace * HZ);
4810         set_max_delegations();
4811         return 0;
4812 out_free_laundry:
4813         destroy_workqueue(laundry_wq);
4814 out_recovery:
4815         nfsd4_client_tracking_exit(net);
4816         put_net(net);
4817         return ret;
4818 }
4819
4820 /* should be called with the state lock held */
4821 static void
4822 __nfs4_state_shutdown(void)
4823 {
4824         int i;
4825         struct nfs4_client *clp = NULL;
4826         struct nfs4_delegation *dp = NULL;
4827         struct list_head *pos, *next, reaplist;
4828         struct rb_node *node, *tmp;
4829
4830         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4831                 while (!list_empty(&conf_id_hashtbl[i])) {
4832                         clp = list_entry(conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4833                         destroy_client(clp);
4834                 }
4835         }
4836
4837         node = rb_first(&unconf_name_tree);
4838         while (node != NULL) {
4839                 tmp = node;
4840                 node = rb_next(tmp);
4841                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4842                 rb_erase(tmp, &unconf_name_tree);
4843                 destroy_client(clp);
4844         }
4845
4846         INIT_LIST_HEAD(&reaplist);
4847         spin_lock(&recall_lock);
4848         list_for_each_safe(pos, next, &del_recall_lru) {
4849                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4850                 list_move(&dp->dl_recall_lru, &reaplist);
4851         }
4852         spin_unlock(&recall_lock);
4853         list_for_each_safe(pos, next, &reaplist) {
4854                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4855                 unhash_delegation(dp);
4856         }
4857
4858         nfsd4_client_tracking_exit(&init_net);
4859         put_net(&init_net);
4860 }
4861
4862 void
4863 nfs4_state_shutdown(void)
4864 {
4865         struct net *net = &init_net;
4866         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4867
4868         cancel_delayed_work_sync(&laundromat_work);
4869         destroy_workqueue(laundry_wq);
4870         locks_end_grace(&nn->nfsd4_manager);
4871         nfs4_lock_state();
4872         __nfs4_state_shutdown();
4873         nfs4_unlock_state();
4874         nfsd4_destroy_callback_queue();
4875 }
4876
4877 static void
4878 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4879 {
4880         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
4881                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
4882 }
4883
4884 static void
4885 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
4886 {
4887         if (cstate->minorversion) {
4888                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
4889                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4890         }
4891 }
4892
4893 void
4894 clear_current_stateid(struct nfsd4_compound_state *cstate)
4895 {
4896         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
4897 }
4898
4899 /*
4900  * functions to set current state id
4901  */
4902 void
4903 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4904 {
4905         put_stateid(cstate, &odp->od_stateid);
4906 }
4907
4908 void
4909 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
4910 {
4911         put_stateid(cstate, &open->op_stateid);
4912 }
4913
4914 void
4915 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4916 {
4917         put_stateid(cstate, &close->cl_stateid);
4918 }
4919
4920 void
4921 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
4922 {
4923         put_stateid(cstate, &lock->lk_resp_stateid);
4924 }
4925
4926 /*
4927  * functions to consume current state id
4928  */
4929
4930 void
4931 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
4932 {
4933         get_stateid(cstate, &odp->od_stateid);
4934 }
4935
4936 void
4937 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
4938 {
4939         get_stateid(cstate, &drp->dr_stateid);
4940 }
4941
4942 void
4943 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
4944 {
4945         get_stateid(cstate, &fsp->fr_stateid);
4946 }
4947
4948 void
4949 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
4950 {
4951         get_stateid(cstate, &setattr->sa_stateid);
4952 }
4953
4954 void
4955 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
4956 {
4957         get_stateid(cstate, &close->cl_stateid);
4958 }
4959
4960 void
4961 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
4962 {
4963         get_stateid(cstate, &locku->lu_stateid);
4964 }
4965
4966 void
4967 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
4968 {
4969         get_stateid(cstate, &read->rd_stateid);
4970 }
4971
4972 void
4973 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
4974 {
4975         get_stateid(cstate, &write->wr_stateid);
4976 }