16d39c6c4fbb56b11c25ac8927b8913b1a928a15
[linux-3.10.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/addr.h>
44 #include "xdr4.h"
45 #include "vfs.h"
46 #include "current_stateid.h"
47
48 #include "netns.h"
49
50 #define NFSDDBG_FACILITY                NFSDDBG_PROC
51
52 #define all_ones {{~0,~0},~0}
53 static const stateid_t one_stateid = {
54         .si_generation = ~0,
55         .si_opaque = all_ones,
56 };
57 static const stateid_t zero_stateid = {
58         /* all fields zero */
59 };
60 static const stateid_t currentstateid = {
61         .si_generation = 1,
62 };
63
64 static u64 current_sessionid = 1;
65
66 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
67 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
68 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
69
70 /* forward declarations */
71 static int check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner);
72
73 /* Locking: */
74
75 /* Currently used for almost all code touching nfsv4 state: */
76 static DEFINE_MUTEX(client_mutex);
77
78 /*
79  * Currently used for the del_recall_lru and file hash table.  In an
80  * effort to decrease the scope of the client_mutex, this spinlock may
81  * eventually cover more:
82  */
83 static DEFINE_SPINLOCK(recall_lock);
84
85 static struct kmem_cache *openowner_slab = NULL;
86 static struct kmem_cache *lockowner_slab = NULL;
87 static struct kmem_cache *file_slab = NULL;
88 static struct kmem_cache *stateid_slab = NULL;
89 static struct kmem_cache *deleg_slab = NULL;
90
91 void
92 nfs4_lock_state(void)
93 {
94         mutex_lock(&client_mutex);
95 }
96
97 static void free_session(struct kref *);
98
99 /* Must be called under the client_lock */
100 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
101 {
102         kref_put(&ses->se_ref, free_session);
103 }
104
105 static void nfsd4_get_session(struct nfsd4_session *ses)
106 {
107         kref_get(&ses->se_ref);
108 }
109
110 void
111 nfs4_unlock_state(void)
112 {
113         mutex_unlock(&client_mutex);
114 }
115
116 static inline u32
117 opaque_hashval(const void *ptr, int nbytes)
118 {
119         unsigned char *cptr = (unsigned char *) ptr;
120
121         u32 x = 0;
122         while (nbytes--) {
123                 x *= 37;
124                 x += *cptr++;
125         }
126         return x;
127 }
128
129 static struct list_head del_recall_lru;
130
131 static void nfsd4_free_file(struct nfs4_file *f)
132 {
133         kmem_cache_free(file_slab, f);
134 }
135
136 static inline void
137 put_nfs4_file(struct nfs4_file *fi)
138 {
139         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
140                 list_del(&fi->fi_hash);
141                 spin_unlock(&recall_lock);
142                 iput(fi->fi_inode);
143                 nfsd4_free_file(fi);
144         }
145 }
146
147 static inline void
148 get_nfs4_file(struct nfs4_file *fi)
149 {
150         atomic_inc(&fi->fi_ref);
151 }
152
153 static int num_delegations;
154 unsigned long max_delegations;
155
156 /*
157  * Open owner state (share locks)
158  */
159
160 /* hash tables for lock and open owners */
161 #define OWNER_HASH_BITS              8
162 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
163 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
164
165 static unsigned int ownerstr_hashval(u32 clientid, struct xdr_netobj *ownername)
166 {
167         unsigned int ret;
168
169         ret = opaque_hashval(ownername->data, ownername->len);
170         ret += clientid;
171         return ret & OWNER_HASH_MASK;
172 }
173
174 /* hash table for nfs4_file */
175 #define FILE_HASH_BITS                   8
176 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
177
178 static unsigned int file_hashval(struct inode *ino)
179 {
180         /* XXX: why are we hashing on inode pointer, anyway? */
181         return hash_ptr(ino, FILE_HASH_BITS);
182 }
183
184 static struct list_head file_hashtbl[FILE_HASH_SIZE];
185
186 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
187 {
188         WARN_ON_ONCE(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
189         atomic_inc(&fp->fi_access[oflag]);
190 }
191
192 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
193 {
194         if (oflag == O_RDWR) {
195                 __nfs4_file_get_access(fp, O_RDONLY);
196                 __nfs4_file_get_access(fp, O_WRONLY);
197         } else
198                 __nfs4_file_get_access(fp, oflag);
199 }
200
201 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
202 {
203         if (fp->fi_fds[oflag]) {
204                 fput(fp->fi_fds[oflag]);
205                 fp->fi_fds[oflag] = NULL;
206         }
207 }
208
209 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
210 {
211         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
212                 nfs4_file_put_fd(fp, oflag);
213                 /*
214                  * It's also safe to get rid of the RDWR open *if*
215                  * we no longer have need of the other kind of access
216                  * or if we already have the other kind of open:
217                  */
218                 if (fp->fi_fds[1-oflag]
219                         || atomic_read(&fp->fi_access[1 - oflag]) == 0)
220                         nfs4_file_put_fd(fp, O_RDWR);
221         }
222 }
223
224 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
225 {
226         if (oflag == O_RDWR) {
227                 __nfs4_file_put_access(fp, O_RDONLY);
228                 __nfs4_file_put_access(fp, O_WRONLY);
229         } else
230                 __nfs4_file_put_access(fp, oflag);
231 }
232
233 static inline int get_new_stid(struct nfs4_stid *stid)
234 {
235         static int min_stateid = 0;
236         struct idr *stateids = &stid->sc_client->cl_stateids;
237         int new_stid;
238         int error;
239
240         error = idr_get_new_above(stateids, stid, min_stateid, &new_stid);
241         /*
242          * Note: the necessary preallocation was done in
243          * nfs4_alloc_stateid().  The idr code caps the number of
244          * preallocations that can exist at a time, but the state lock
245          * prevents anyone from using ours before we get here:
246          */
247         WARN_ON_ONCE(error);
248         /*
249          * It shouldn't be a problem to reuse an opaque stateid value.
250          * I don't think it is for 4.1.  But with 4.0 I worry that, for
251          * example, a stray write retransmission could be accepted by
252          * the server when it should have been rejected.  Therefore,
253          * adopt a trick from the sctp code to attempt to maximize the
254          * amount of time until an id is reused, by ensuring they always
255          * "increase" (mod INT_MAX):
256          */
257
258         min_stateid = new_stid+1;
259         if (min_stateid == INT_MAX)
260                 min_stateid = 0;
261         return new_stid;
262 }
263
264 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct
265 kmem_cache *slab)
266 {
267         struct idr *stateids = &cl->cl_stateids;
268         static int min_stateid = 0;
269         struct nfs4_stid *stid;
270         int new_id;
271
272         stid = kmem_cache_alloc(slab, GFP_KERNEL);
273         if (!stid)
274                 return NULL;
275
276         if (!idr_pre_get(stateids, GFP_KERNEL))
277                 goto out_free;
278         if (idr_get_new_above(stateids, stid, min_stateid, &new_id))
279                 goto out_free;
280         stid->sc_client = cl;
281         stid->sc_type = 0;
282         stid->sc_stateid.si_opaque.so_id = new_id;
283         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
284         /* Will be incremented before return to client: */
285         stid->sc_stateid.si_generation = 0;
286
287         /*
288          * It shouldn't be a problem to reuse an opaque stateid value.
289          * I don't think it is for 4.1.  But with 4.0 I worry that, for
290          * example, a stray write retransmission could be accepted by
291          * the server when it should have been rejected.  Therefore,
292          * adopt a trick from the sctp code to attempt to maximize the
293          * amount of time until an id is reused, by ensuring they always
294          * "increase" (mod INT_MAX):
295          */
296
297         min_stateid = new_id+1;
298         if (min_stateid == INT_MAX)
299                 min_stateid = 0;
300         return stid;
301 out_free:
302         kfree(stid);
303         return NULL;
304 }
305
306 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
307 {
308         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
309 }
310
311 static struct nfs4_delegation *
312 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
313 {
314         struct nfs4_delegation *dp;
315         struct nfs4_file *fp = stp->st_file;
316
317         dprintk("NFSD alloc_init_deleg\n");
318         /*
319          * Major work on the lease subsystem (for example, to support
320          * calbacks on stat) will be required before we can support
321          * write delegations properly.
322          */
323         if (type != NFS4_OPEN_DELEGATE_READ)
324                 return NULL;
325         if (fp->fi_had_conflict)
326                 return NULL;
327         if (num_delegations > max_delegations)
328                 return NULL;
329         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
330         if (dp == NULL)
331                 return dp;
332         dp->dl_stid.sc_type = NFS4_DELEG_STID;
333         /*
334          * delegation seqid's are never incremented.  The 4.1 special
335          * meaning of seqid 0 isn't meaningful, really, but let's avoid
336          * 0 anyway just for consistency and use 1:
337          */
338         dp->dl_stid.sc_stateid.si_generation = 1;
339         num_delegations++;
340         INIT_LIST_HEAD(&dp->dl_perfile);
341         INIT_LIST_HEAD(&dp->dl_perclnt);
342         INIT_LIST_HEAD(&dp->dl_recall_lru);
343         get_nfs4_file(fp);
344         dp->dl_file = fp;
345         dp->dl_type = type;
346         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
347         dp->dl_time = 0;
348         atomic_set(&dp->dl_count, 1);
349         nfsd4_init_callback(&dp->dl_recall);
350         return dp;
351 }
352
353 static void free_stid(struct nfs4_stid *s, struct kmem_cache *slab)
354 {
355         struct idr *stateids = &s->sc_client->cl_stateids;
356
357         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
358         kmem_cache_free(slab, s);
359 }
360
361 void
362 nfs4_put_delegation(struct nfs4_delegation *dp)
363 {
364         if (atomic_dec_and_test(&dp->dl_count)) {
365                 dprintk("NFSD: freeing dp %p\n",dp);
366                 put_nfs4_file(dp->dl_file);
367                 free_stid(&dp->dl_stid, deleg_slab);
368                 num_delegations--;
369         }
370 }
371
372 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
373 {
374         if (atomic_dec_and_test(&fp->fi_delegees)) {
375                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
376                 fp->fi_lease = NULL;
377                 fput(fp->fi_deleg_file);
378                 fp->fi_deleg_file = NULL;
379         }
380 }
381
382 static void unhash_stid(struct nfs4_stid *s)
383 {
384         s->sc_type = 0;
385 }
386
387 /* Called under the state lock. */
388 static void
389 unhash_delegation(struct nfs4_delegation *dp)
390 {
391         unhash_stid(&dp->dl_stid);
392         list_del_init(&dp->dl_perclnt);
393         spin_lock(&recall_lock);
394         list_del_init(&dp->dl_perfile);
395         list_del_init(&dp->dl_recall_lru);
396         spin_unlock(&recall_lock);
397         nfs4_put_deleg_lease(dp->dl_file);
398         nfs4_put_delegation(dp);
399 }
400
401 /* 
402  * SETCLIENTID state 
403  */
404
405 static unsigned int clientid_hashval(u32 id)
406 {
407         return id & CLIENT_HASH_MASK;
408 }
409
410 static unsigned int clientstr_hashval(const char *name)
411 {
412         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
413 }
414
415 /*
416  * We store the NONE, READ, WRITE, and BOTH bits separately in the
417  * st_{access,deny}_bmap field of the stateid, in order to track not
418  * only what share bits are currently in force, but also what
419  * combinations of share bits previous opens have used.  This allows us
420  * to enforce the recommendation of rfc 3530 14.2.19 that the server
421  * return an error if the client attempt to downgrade to a combination
422  * of share bits not explicable by closing some of its previous opens.
423  *
424  * XXX: This enforcement is actually incomplete, since we don't keep
425  * track of access/deny bit combinations; so, e.g., we allow:
426  *
427  *      OPEN allow read, deny write
428  *      OPEN allow both, deny none
429  *      DOWNGRADE allow read, deny none
430  *
431  * which we should reject.
432  */
433 static unsigned int
434 bmap_to_share_mode(unsigned long bmap) {
435         int i;
436         unsigned int access = 0;
437
438         for (i = 1; i < 4; i++) {
439                 if (test_bit(i, &bmap))
440                         access |= i;
441         }
442         return access;
443 }
444
445 static bool
446 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
447         unsigned int access, deny;
448
449         access = bmap_to_share_mode(stp->st_access_bmap);
450         deny = bmap_to_share_mode(stp->st_deny_bmap);
451         if ((access & open->op_share_deny) || (deny & open->op_share_access))
452                 return false;
453         return true;
454 }
455
456 /* set share access for a given stateid */
457 static inline void
458 set_access(u32 access, struct nfs4_ol_stateid *stp)
459 {
460         __set_bit(access, &stp->st_access_bmap);
461 }
462
463 /* clear share access for a given stateid */
464 static inline void
465 clear_access(u32 access, struct nfs4_ol_stateid *stp)
466 {
467         __clear_bit(access, &stp->st_access_bmap);
468 }
469
470 /* test whether a given stateid has access */
471 static inline bool
472 test_access(u32 access, struct nfs4_ol_stateid *stp)
473 {
474         return test_bit(access, &stp->st_access_bmap);
475 }
476
477 /* set share deny for a given stateid */
478 static inline void
479 set_deny(u32 access, struct nfs4_ol_stateid *stp)
480 {
481         __set_bit(access, &stp->st_deny_bmap);
482 }
483
484 /* clear share deny for a given stateid */
485 static inline void
486 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
487 {
488         __clear_bit(access, &stp->st_deny_bmap);
489 }
490
491 /* test whether a given stateid is denying specific access */
492 static inline bool
493 test_deny(u32 access, struct nfs4_ol_stateid *stp)
494 {
495         return test_bit(access, &stp->st_deny_bmap);
496 }
497
498 static int nfs4_access_to_omode(u32 access)
499 {
500         switch (access & NFS4_SHARE_ACCESS_BOTH) {
501         case NFS4_SHARE_ACCESS_READ:
502                 return O_RDONLY;
503         case NFS4_SHARE_ACCESS_WRITE:
504                 return O_WRONLY;
505         case NFS4_SHARE_ACCESS_BOTH:
506                 return O_RDWR;
507         }
508         WARN_ON_ONCE(1);
509         return O_RDONLY;
510 }
511
512 /* release all access and file references for a given stateid */
513 static void
514 release_all_access(struct nfs4_ol_stateid *stp)
515 {
516         int i;
517
518         for (i = 1; i < 4; i++) {
519                 if (test_access(i, stp))
520                         nfs4_file_put_access(stp->st_file,
521                                              nfs4_access_to_omode(i));
522                 clear_access(i, stp);
523         }
524 }
525
526 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
527 {
528         list_del(&stp->st_perfile);
529         list_del(&stp->st_perstateowner);
530 }
531
532 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
533 {
534         release_all_access(stp);
535         put_nfs4_file(stp->st_file);
536         stp->st_file = NULL;
537 }
538
539 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
540 {
541         free_stid(&stp->st_stid, stateid_slab);
542 }
543
544 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
545 {
546         struct file *file;
547
548         unhash_generic_stateid(stp);
549         unhash_stid(&stp->st_stid);
550         file = find_any_file(stp->st_file);
551         if (file)
552                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
553         close_generic_stateid(stp);
554         free_generic_stateid(stp);
555 }
556
557 static void unhash_lockowner(struct nfs4_lockowner *lo)
558 {
559         struct nfs4_ol_stateid *stp;
560
561         list_del(&lo->lo_owner.so_strhash);
562         list_del(&lo->lo_perstateid);
563         list_del(&lo->lo_owner_ino_hash);
564         while (!list_empty(&lo->lo_owner.so_stateids)) {
565                 stp = list_first_entry(&lo->lo_owner.so_stateids,
566                                 struct nfs4_ol_stateid, st_perstateowner);
567                 release_lock_stateid(stp);
568         }
569 }
570
571 static void release_lockowner(struct nfs4_lockowner *lo)
572 {
573         unhash_lockowner(lo);
574         nfs4_free_lockowner(lo);
575 }
576
577 static void
578 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
579 {
580         struct nfs4_lockowner *lo;
581
582         while (!list_empty(&open_stp->st_lockowners)) {
583                 lo = list_entry(open_stp->st_lockowners.next,
584                                 struct nfs4_lockowner, lo_perstateid);
585                 release_lockowner(lo);
586         }
587 }
588
589 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
590 {
591         unhash_generic_stateid(stp);
592         release_stateid_lockowners(stp);
593         close_generic_stateid(stp);
594 }
595
596 static void release_open_stateid(struct nfs4_ol_stateid *stp)
597 {
598         unhash_open_stateid(stp);
599         unhash_stid(&stp->st_stid);
600         free_generic_stateid(stp);
601 }
602
603 static void unhash_openowner(struct nfs4_openowner *oo)
604 {
605         struct nfs4_ol_stateid *stp;
606
607         list_del(&oo->oo_owner.so_strhash);
608         list_del(&oo->oo_perclient);
609         while (!list_empty(&oo->oo_owner.so_stateids)) {
610                 stp = list_first_entry(&oo->oo_owner.so_stateids,
611                                 struct nfs4_ol_stateid, st_perstateowner);
612                 release_open_stateid(stp);
613         }
614 }
615
616 static void release_last_closed_stateid(struct nfs4_openowner *oo)
617 {
618         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
619
620         if (s) {
621                 unhash_stid(&s->st_stid);
622                 free_generic_stateid(s);
623                 oo->oo_last_closed_stid = NULL;
624         }
625 }
626
627 static void release_openowner(struct nfs4_openowner *oo)
628 {
629         unhash_openowner(oo);
630         list_del(&oo->oo_close_lru);
631         release_last_closed_stateid(oo);
632         nfs4_free_openowner(oo);
633 }
634
635 static inline int
636 hash_sessionid(struct nfs4_sessionid *sessionid)
637 {
638         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
639
640         return sid->sequence % SESSION_HASH_SIZE;
641 }
642
643 #ifdef NFSD_DEBUG
644 static inline void
645 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
646 {
647         u32 *ptr = (u32 *)(&sessionid->data[0]);
648         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
649 }
650 #else
651 static inline void
652 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
653 {
654 }
655 #endif
656
657
658 static void
659 gen_sessionid(struct nfsd4_session *ses)
660 {
661         struct nfs4_client *clp = ses->se_client;
662         struct nfsd4_sessionid *sid;
663
664         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
665         sid->clientid = clp->cl_clientid;
666         sid->sequence = current_sessionid++;
667         sid->reserved = 0;
668 }
669
670 /*
671  * The protocol defines ca_maxresponssize_cached to include the size of
672  * the rpc header, but all we need to cache is the data starting after
673  * the end of the initial SEQUENCE operation--the rest we regenerate
674  * each time.  Therefore we can advertise a ca_maxresponssize_cached
675  * value that is the number of bytes in our cache plus a few additional
676  * bytes.  In order to stay on the safe side, and not promise more than
677  * we can cache, those additional bytes must be the minimum possible: 24
678  * bytes of rpc header (xid through accept state, with AUTH_NULL
679  * verifier), 12 for the compound header (with zero-length tag), and 44
680  * for the SEQUENCE op response:
681  */
682 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
683
684 static void
685 free_session_slots(struct nfsd4_session *ses)
686 {
687         int i;
688
689         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
690                 kfree(ses->se_slots[i]);
691 }
692
693 /*
694  * We don't actually need to cache the rpc and session headers, so we
695  * can allocate a little less for each slot:
696  */
697 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
698 {
699         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
700 }
701
702 static int nfsd4_sanitize_slot_size(u32 size)
703 {
704         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
705         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
706
707         return size;
708 }
709
710 /*
711  * XXX: If we run out of reserved DRC memory we could (up to a point)
712  * re-negotiate active sessions and reduce their slot usage to make
713  * room for new connections. For now we just fail the create session.
714  */
715 static int nfsd4_get_drc_mem(int slotsize, u32 num)
716 {
717         int avail;
718
719         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
720
721         spin_lock(&nfsd_drc_lock);
722         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
723                     nfsd_drc_max_mem - nfsd_drc_mem_used);
724         num = min_t(int, num, avail / slotsize);
725         nfsd_drc_mem_used += num * slotsize;
726         spin_unlock(&nfsd_drc_lock);
727
728         return num;
729 }
730
731 static void nfsd4_put_drc_mem(int slotsize, int num)
732 {
733         spin_lock(&nfsd_drc_lock);
734         nfsd_drc_mem_used -= slotsize * num;
735         spin_unlock(&nfsd_drc_lock);
736 }
737
738 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
739 {
740         struct nfsd4_session *new;
741         int mem, i;
742
743         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
744                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
745         mem = numslots * sizeof(struct nfsd4_slot *);
746
747         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
748         if (!new)
749                 return NULL;
750         /* allocate each struct nfsd4_slot and data cache in one piece */
751         for (i = 0; i < numslots; i++) {
752                 mem = sizeof(struct nfsd4_slot) + slotsize;
753                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
754                 if (!new->se_slots[i])
755                         goto out_free;
756         }
757         return new;
758 out_free:
759         while (i--)
760                 kfree(new->se_slots[i]);
761         kfree(new);
762         return NULL;
763 }
764
765 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new,
766                                    struct nfsd4_channel_attrs *req,
767                                    int numslots, int slotsize,
768                                    struct nfsd_net *nn)
769 {
770         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
771
772         new->maxreqs = numslots;
773         new->maxresp_cached = min_t(u32, req->maxresp_cached,
774                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
775         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
776         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
777         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
778 }
779
780 static void free_conn(struct nfsd4_conn *c)
781 {
782         svc_xprt_put(c->cn_xprt);
783         kfree(c);
784 }
785
786 static void nfsd4_conn_lost(struct svc_xpt_user *u)
787 {
788         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
789         struct nfs4_client *clp = c->cn_session->se_client;
790
791         spin_lock(&clp->cl_lock);
792         if (!list_empty(&c->cn_persession)) {
793                 list_del(&c->cn_persession);
794                 free_conn(c);
795         }
796         spin_unlock(&clp->cl_lock);
797         nfsd4_probe_callback(clp);
798 }
799
800 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
801 {
802         struct nfsd4_conn *conn;
803
804         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
805         if (!conn)
806                 return NULL;
807         svc_xprt_get(rqstp->rq_xprt);
808         conn->cn_xprt = rqstp->rq_xprt;
809         conn->cn_flags = flags;
810         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
811         return conn;
812 }
813
814 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
815 {
816         conn->cn_session = ses;
817         list_add(&conn->cn_persession, &ses->se_conns);
818 }
819
820 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
821 {
822         struct nfs4_client *clp = ses->se_client;
823
824         spin_lock(&clp->cl_lock);
825         __nfsd4_hash_conn(conn, ses);
826         spin_unlock(&clp->cl_lock);
827 }
828
829 static int nfsd4_register_conn(struct nfsd4_conn *conn)
830 {
831         conn->cn_xpt_user.callback = nfsd4_conn_lost;
832         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
833 }
834
835 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
836 {
837         int ret;
838
839         nfsd4_hash_conn(conn, ses);
840         ret = nfsd4_register_conn(conn);
841         if (ret)
842                 /* oops; xprt is already down: */
843                 nfsd4_conn_lost(&conn->cn_xpt_user);
844         if (conn->cn_flags & NFS4_CDFC4_BACK) {
845                 /* callback channel may be back up */
846                 nfsd4_probe_callback(ses->se_client);
847         }
848 }
849
850 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
851 {
852         u32 dir = NFS4_CDFC4_FORE;
853
854         if (cses->flags & SESSION4_BACK_CHAN)
855                 dir |= NFS4_CDFC4_BACK;
856         return alloc_conn(rqstp, dir);
857 }
858
859 /* must be called under client_lock */
860 static void nfsd4_del_conns(struct nfsd4_session *s)
861 {
862         struct nfs4_client *clp = s->se_client;
863         struct nfsd4_conn *c;
864
865         spin_lock(&clp->cl_lock);
866         while (!list_empty(&s->se_conns)) {
867                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
868                 list_del_init(&c->cn_persession);
869                 spin_unlock(&clp->cl_lock);
870
871                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
872                 free_conn(c);
873
874                 spin_lock(&clp->cl_lock);
875         }
876         spin_unlock(&clp->cl_lock);
877 }
878
879 static void __free_session(struct nfsd4_session *ses)
880 {
881         nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
882         free_session_slots(ses);
883         kfree(ses);
884 }
885
886 static void free_session(struct kref *kref)
887 {
888         struct nfsd4_session *ses;
889         struct nfsd_net *nn;
890
891         ses = container_of(kref, struct nfsd4_session, se_ref);
892         nn = net_generic(ses->se_client->net, nfsd_net_id);
893
894         lockdep_assert_held(&nn->client_lock);
895         nfsd4_del_conns(ses);
896         __free_session(ses);
897 }
898
899 void nfsd4_put_session(struct nfsd4_session *ses)
900 {
901         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
902
903         spin_lock(&nn->client_lock);
904         nfsd4_put_session_locked(ses);
905         spin_unlock(&nn->client_lock);
906 }
907
908 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan,
909                                            struct nfsd_net *nn)
910 {
911         struct nfsd4_session *new;
912         int numslots, slotsize;
913         /*
914          * Note decreasing slot size below client's request may
915          * make it difficult for client to function correctly, whereas
916          * decreasing the number of slots will (just?) affect
917          * performance.  When short on memory we therefore prefer to
918          * decrease number of slots instead of their size.
919          */
920         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
921         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
922         if (numslots < 1)
923                 return NULL;
924
925         new = __alloc_session(slotsize, numslots);
926         if (!new) {
927                 nfsd4_put_drc_mem(slotsize, numslots);
928                 return NULL;
929         }
930         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize, nn);
931         return new;
932 }
933
934 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
935 {
936         int idx;
937         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
938
939         new->se_client = clp;
940         gen_sessionid(new);
941
942         INIT_LIST_HEAD(&new->se_conns);
943
944         new->se_cb_seq_nr = 1;
945         new->se_flags = cses->flags;
946         new->se_cb_prog = cses->callback_prog;
947         new->se_cb_sec = cses->cb_sec;
948         kref_init(&new->se_ref);
949         idx = hash_sessionid(&new->se_sessionid);
950         spin_lock(&nn->client_lock);
951         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
952         spin_lock(&clp->cl_lock);
953         list_add(&new->se_perclnt, &clp->cl_sessions);
954         spin_unlock(&clp->cl_lock);
955         spin_unlock(&nn->client_lock);
956
957         if (cses->flags & SESSION4_BACK_CHAN) {
958                 struct sockaddr *sa = svc_addr(rqstp);
959                 /*
960                  * This is a little silly; with sessions there's no real
961                  * use for the callback address.  Use the peer address
962                  * as a reasonable default for now, but consider fixing
963                  * the rpc client not to require an address in the
964                  * future:
965                  */
966                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
967                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
968         }
969 }
970
971 /* caller must hold client_lock */
972 static struct nfsd4_session *
973 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
974 {
975         struct nfsd4_session *elem;
976         int idx;
977         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
978
979         dump_sessionid(__func__, sessionid);
980         idx = hash_sessionid(sessionid);
981         /* Search in the appropriate list */
982         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
983                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
984                             NFS4_MAX_SESSIONID_LEN)) {
985                         return elem;
986                 }
987         }
988
989         dprintk("%s: session not found\n", __func__);
990         return NULL;
991 }
992
993 /* caller must hold client_lock */
994 static void
995 unhash_session(struct nfsd4_session *ses)
996 {
997         list_del(&ses->se_hash);
998         spin_lock(&ses->se_client->cl_lock);
999         list_del(&ses->se_perclnt);
1000         spin_unlock(&ses->se_client->cl_lock);
1001 }
1002
1003 /* must be called under the client_lock */
1004 static inline void
1005 renew_client_locked(struct nfs4_client *clp)
1006 {
1007         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1008
1009         if (is_client_expired(clp)) {
1010                 WARN_ON(1);
1011                 printk("%s: client (clientid %08x/%08x) already expired\n",
1012                         __func__,
1013                         clp->cl_clientid.cl_boot,
1014                         clp->cl_clientid.cl_id);
1015                 return;
1016         }
1017
1018         dprintk("renewing client (clientid %08x/%08x)\n", 
1019                         clp->cl_clientid.cl_boot, 
1020                         clp->cl_clientid.cl_id);
1021         list_move_tail(&clp->cl_lru, &nn->client_lru);
1022         clp->cl_time = get_seconds();
1023 }
1024
1025 static inline void
1026 renew_client(struct nfs4_client *clp)
1027 {
1028         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1029
1030         spin_lock(&nn->client_lock);
1031         renew_client_locked(clp);
1032         spin_unlock(&nn->client_lock);
1033 }
1034
1035 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1036 static int
1037 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1038 {
1039         if (clid->cl_boot == nn->boot_time)
1040                 return 0;
1041         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1042                 clid->cl_boot, clid->cl_id, nn->boot_time);
1043         return 1;
1044 }
1045
1046 /* 
1047  * XXX Should we use a slab cache ?
1048  * This type of memory management is somewhat inefficient, but we use it
1049  * anyway since SETCLIENTID is not a common operation.
1050  */
1051 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1052 {
1053         struct nfs4_client *clp;
1054
1055         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1056         if (clp == NULL)
1057                 return NULL;
1058         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1059         if (clp->cl_name.data == NULL) {
1060                 kfree(clp);
1061                 return NULL;
1062         }
1063         clp->cl_name.len = name.len;
1064         return clp;
1065 }
1066
1067 static inline void
1068 free_client(struct nfs4_client *clp)
1069 {
1070         struct nfsd_net __maybe_unused *nn = net_generic(clp->net, nfsd_net_id);
1071
1072         lockdep_assert_held(&nn->client_lock);
1073         while (!list_empty(&clp->cl_sessions)) {
1074                 struct nfsd4_session *ses;
1075                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1076                                 se_perclnt);
1077                 list_del(&ses->se_perclnt);
1078                 nfsd4_put_session_locked(ses);
1079         }
1080         free_svc_cred(&clp->cl_cred);
1081         kfree(clp->cl_name.data);
1082         idr_destroy(&clp->cl_stateids);
1083         kfree(clp);
1084 }
1085
1086 void
1087 release_session_client(struct nfsd4_session *session)
1088 {
1089         struct nfs4_client *clp = session->se_client;
1090         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1091
1092         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
1093                 return;
1094         if (is_client_expired(clp)) {
1095                 free_client(clp);
1096                 session->se_client = NULL;
1097         } else
1098                 renew_client_locked(clp);
1099         spin_unlock(&nn->client_lock);
1100 }
1101
1102 /* must be called under the client_lock */
1103 static inline void
1104 unhash_client_locked(struct nfs4_client *clp)
1105 {
1106         struct nfsd4_session *ses;
1107
1108         mark_client_expired(clp);
1109         list_del(&clp->cl_lru);
1110         spin_lock(&clp->cl_lock);
1111         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1112                 list_del_init(&ses->se_hash);
1113         spin_unlock(&clp->cl_lock);
1114 }
1115
1116 static void
1117 destroy_client(struct nfs4_client *clp)
1118 {
1119         struct nfs4_openowner *oo;
1120         struct nfs4_delegation *dp;
1121         struct list_head reaplist;
1122         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1123
1124         INIT_LIST_HEAD(&reaplist);
1125         spin_lock(&recall_lock);
1126         while (!list_empty(&clp->cl_delegations)) {
1127                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1128                 list_del_init(&dp->dl_perclnt);
1129                 list_move(&dp->dl_recall_lru, &reaplist);
1130         }
1131         spin_unlock(&recall_lock);
1132         while (!list_empty(&reaplist)) {
1133                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1134                 unhash_delegation(dp);
1135         }
1136         while (!list_empty(&clp->cl_openowners)) {
1137                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1138                 release_openowner(oo);
1139         }
1140         nfsd4_shutdown_callback(clp);
1141         if (clp->cl_cb_conn.cb_xprt)
1142                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1143         list_del(&clp->cl_idhash);
1144         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1145                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1146         else
1147                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1148         spin_lock(&nn->client_lock);
1149         unhash_client_locked(clp);
1150         if (atomic_read(&clp->cl_refcount) == 0)
1151                 free_client(clp);
1152         spin_unlock(&nn->client_lock);
1153 }
1154
1155 static void expire_client(struct nfs4_client *clp)
1156 {
1157         nfsd4_client_record_remove(clp);
1158         destroy_client(clp);
1159 }
1160
1161 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1162 {
1163         memcpy(target->cl_verifier.data, source->data,
1164                         sizeof(target->cl_verifier.data));
1165 }
1166
1167 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1168 {
1169         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1170         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1171 }
1172
1173 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1174 {
1175         if (source->cr_principal) {
1176                 target->cr_principal =
1177                                 kstrdup(source->cr_principal, GFP_KERNEL);
1178                 if (target->cr_principal == NULL)
1179                         return -ENOMEM;
1180         } else
1181                 target->cr_principal = NULL;
1182         target->cr_flavor = source->cr_flavor;
1183         target->cr_uid = source->cr_uid;
1184         target->cr_gid = source->cr_gid;
1185         target->cr_group_info = source->cr_group_info;
1186         get_group_info(target->cr_group_info);
1187         return 0;
1188 }
1189
1190 static long long
1191 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1192 {
1193         long long res;
1194
1195         res = o1->len - o2->len;
1196         if (res)
1197                 return res;
1198         return (long long)memcmp(o1->data, o2->data, o1->len);
1199 }
1200
1201 static int same_name(const char *n1, const char *n2)
1202 {
1203         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1204 }
1205
1206 static int
1207 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1208 {
1209         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1210 }
1211
1212 static int
1213 same_clid(clientid_t *cl1, clientid_t *cl2)
1214 {
1215         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1216 }
1217
1218 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1219 {
1220         int i;
1221
1222         if (g1->ngroups != g2->ngroups)
1223                 return false;
1224         for (i=0; i<g1->ngroups; i++)
1225                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1226                         return false;
1227         return true;
1228 }
1229
1230 /*
1231  * RFC 3530 language requires clid_inuse be returned when the
1232  * "principal" associated with a requests differs from that previously
1233  * used.  We use uid, gid's, and gss principal string as our best
1234  * approximation.  We also don't want to allow non-gss use of a client
1235  * established using gss: in theory cr_principal should catch that
1236  * change, but in practice cr_principal can be null even in the gss case
1237  * since gssd doesn't always pass down a principal string.
1238  */
1239 static bool is_gss_cred(struct svc_cred *cr)
1240 {
1241         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1242         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1243 }
1244
1245
1246 static bool
1247 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1248 {
1249         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1250                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1251                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1252                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1253                 return false;
1254         if (cr1->cr_principal == cr2->cr_principal)
1255                 return true;
1256         if (!cr1->cr_principal || !cr2->cr_principal)
1257                 return false;
1258         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1259 }
1260
1261 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1262 {
1263         static u32 current_clientid = 1;
1264
1265         clp->cl_clientid.cl_boot = nn->boot_time;
1266         clp->cl_clientid.cl_id = current_clientid++; 
1267 }
1268
1269 static void gen_confirm(struct nfs4_client *clp)
1270 {
1271         __be32 verf[2];
1272         static u32 i;
1273
1274         verf[0] = (__be32)get_seconds();
1275         verf[1] = (__be32)i++;
1276         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1277 }
1278
1279 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1280 {
1281         struct nfs4_stid *ret;
1282
1283         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1284         if (!ret || !ret->sc_type)
1285                 return NULL;
1286         return ret;
1287 }
1288
1289 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1290 {
1291         struct nfs4_stid *s;
1292
1293         s = find_stateid(cl, t);
1294         if (!s)
1295                 return NULL;
1296         if (typemask & s->sc_type)
1297                 return s;
1298         return NULL;
1299 }
1300
1301 static struct nfs4_client *create_client(struct xdr_netobj name,
1302                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1303 {
1304         struct nfs4_client *clp;
1305         struct sockaddr *sa = svc_addr(rqstp);
1306         int ret;
1307         struct net *net = SVC_NET(rqstp);
1308         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1309
1310         clp = alloc_client(name);
1311         if (clp == NULL)
1312                 return NULL;
1313
1314         INIT_LIST_HEAD(&clp->cl_sessions);
1315         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1316         if (ret) {
1317                 spin_lock(&nn->client_lock);
1318                 free_client(clp);
1319                 spin_unlock(&nn->client_lock);
1320                 return NULL;
1321         }
1322         idr_init(&clp->cl_stateids);
1323         atomic_set(&clp->cl_refcount, 0);
1324         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1325         INIT_LIST_HEAD(&clp->cl_idhash);
1326         INIT_LIST_HEAD(&clp->cl_openowners);
1327         INIT_LIST_HEAD(&clp->cl_delegations);
1328         INIT_LIST_HEAD(&clp->cl_lru);
1329         INIT_LIST_HEAD(&clp->cl_callbacks);
1330         spin_lock_init(&clp->cl_lock);
1331         nfsd4_init_callback(&clp->cl_cb_null);
1332         clp->cl_time = get_seconds();
1333         clear_bit(0, &clp->cl_cb_slot_busy);
1334         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1335         copy_verf(clp, verf);
1336         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1337         gen_confirm(clp);
1338         clp->cl_cb_session = NULL;
1339         clp->net = net;
1340         return clp;
1341 }
1342
1343 static void
1344 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1345 {
1346         struct rb_node **new = &(root->rb_node), *parent = NULL;
1347         struct nfs4_client *clp;
1348
1349         while (*new) {
1350                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1351                 parent = *new;
1352
1353                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1354                         new = &((*new)->rb_left);
1355                 else
1356                         new = &((*new)->rb_right);
1357         }
1358
1359         rb_link_node(&new_clp->cl_namenode, parent, new);
1360         rb_insert_color(&new_clp->cl_namenode, root);
1361 }
1362
1363 static struct nfs4_client *
1364 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1365 {
1366         long long cmp;
1367         struct rb_node *node = root->rb_node;
1368         struct nfs4_client *clp;
1369
1370         while (node) {
1371                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1372                 cmp = compare_blob(&clp->cl_name, name);
1373                 if (cmp > 0)
1374                         node = node->rb_left;
1375                 else if (cmp < 0)
1376                         node = node->rb_right;
1377                 else
1378                         return clp;
1379         }
1380         return NULL;
1381 }
1382
1383 static void
1384 add_to_unconfirmed(struct nfs4_client *clp)
1385 {
1386         unsigned int idhashval;
1387         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1388
1389         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1390         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1391         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1392         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1393         renew_client(clp);
1394 }
1395
1396 static void
1397 move_to_confirmed(struct nfs4_client *clp)
1398 {
1399         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1400         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1401
1402         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1403         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1404         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1405         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1406         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1407         renew_client(clp);
1408 }
1409
1410 static struct nfs4_client *
1411 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1412 {
1413         struct nfs4_client *clp;
1414         unsigned int idhashval = clientid_hashval(clid->cl_id);
1415
1416         list_for_each_entry(clp, &nn->conf_id_hashtbl[idhashval], cl_idhash) {
1417                 if (same_clid(&clp->cl_clientid, clid)) {
1418                         if ((bool)clp->cl_minorversion != sessions)
1419                                 return NULL;
1420                         renew_client(clp);
1421                         return clp;
1422                 }
1423         }
1424         return NULL;
1425 }
1426
1427 static struct nfs4_client *
1428 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1429 {
1430         struct nfs4_client *clp;
1431         unsigned int idhashval = clientid_hashval(clid->cl_id);
1432
1433         list_for_each_entry(clp, &nn->unconf_id_hashtbl[idhashval], cl_idhash) {
1434                 if (same_clid(&clp->cl_clientid, clid)) {
1435                         if ((bool)clp->cl_minorversion != sessions)
1436                                 return NULL;
1437                         return clp;
1438                 }
1439         }
1440         return NULL;
1441 }
1442
1443 static bool clp_used_exchangeid(struct nfs4_client *clp)
1444 {
1445         return clp->cl_exchange_flags != 0;
1446
1447
1448 static struct nfs4_client *
1449 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1450 {
1451         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1452 }
1453
1454 static struct nfs4_client *
1455 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1456 {
1457         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1458 }
1459
1460 static void
1461 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1462 {
1463         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1464         struct sockaddr *sa = svc_addr(rqstp);
1465         u32 scopeid = rpc_get_scope_id(sa);
1466         unsigned short expected_family;
1467
1468         /* Currently, we only support tcp and tcp6 for the callback channel */
1469         if (se->se_callback_netid_len == 3 &&
1470             !memcmp(se->se_callback_netid_val, "tcp", 3))
1471                 expected_family = AF_INET;
1472         else if (se->se_callback_netid_len == 4 &&
1473                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1474                 expected_family = AF_INET6;
1475         else
1476                 goto out_err;
1477
1478         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1479                                             se->se_callback_addr_len,
1480                                             (struct sockaddr *)&conn->cb_addr,
1481                                             sizeof(conn->cb_addr));
1482
1483         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1484                 goto out_err;
1485
1486         if (conn->cb_addr.ss_family == AF_INET6)
1487                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1488
1489         conn->cb_prog = se->se_callback_prog;
1490         conn->cb_ident = se->se_callback_ident;
1491         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1492         return;
1493 out_err:
1494         conn->cb_addr.ss_family = AF_UNSPEC;
1495         conn->cb_addrlen = 0;
1496         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1497                 "will not receive delegations\n",
1498                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1499
1500         return;
1501 }
1502
1503 /*
1504  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1505  */
1506 void
1507 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1508 {
1509         struct nfsd4_slot *slot = resp->cstate.slot;
1510         unsigned int base;
1511
1512         dprintk("--> %s slot %p\n", __func__, slot);
1513
1514         slot->sl_opcnt = resp->opcnt;
1515         slot->sl_status = resp->cstate.status;
1516
1517         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1518         if (nfsd4_not_cached(resp)) {
1519                 slot->sl_datalen = 0;
1520                 return;
1521         }
1522         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1523         base = (char *)resp->cstate.datap -
1524                                         (char *)resp->xbuf->head[0].iov_base;
1525         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1526                                     slot->sl_datalen))
1527                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1528         return;
1529 }
1530
1531 /*
1532  * Encode the replay sequence operation from the slot values.
1533  * If cachethis is FALSE encode the uncached rep error on the next
1534  * operation which sets resp->p and increments resp->opcnt for
1535  * nfs4svc_encode_compoundres.
1536  *
1537  */
1538 static __be32
1539 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1540                           struct nfsd4_compoundres *resp)
1541 {
1542         struct nfsd4_op *op;
1543         struct nfsd4_slot *slot = resp->cstate.slot;
1544
1545         /* Encode the replayed sequence operation */
1546         op = &args->ops[resp->opcnt - 1];
1547         nfsd4_encode_operation(resp, op);
1548
1549         /* Return nfserr_retry_uncached_rep in next operation. */
1550         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1551                 op = &args->ops[resp->opcnt++];
1552                 op->status = nfserr_retry_uncached_rep;
1553                 nfsd4_encode_operation(resp, op);
1554         }
1555         return op->status;
1556 }
1557
1558 /*
1559  * The sequence operation is not cached because we can use the slot and
1560  * session values.
1561  */
1562 __be32
1563 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1564                          struct nfsd4_sequence *seq)
1565 {
1566         struct nfsd4_slot *slot = resp->cstate.slot;
1567         __be32 status;
1568
1569         dprintk("--> %s slot %p\n", __func__, slot);
1570
1571         /* Either returns 0 or nfserr_retry_uncached */
1572         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1573         if (status == nfserr_retry_uncached_rep)
1574                 return status;
1575
1576         /* The sequence operation has been encoded, cstate->datap set. */
1577         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1578
1579         resp->opcnt = slot->sl_opcnt;
1580         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1581         status = slot->sl_status;
1582
1583         return status;
1584 }
1585
1586 /*
1587  * Set the exchange_id flags returned by the server.
1588  */
1589 static void
1590 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1591 {
1592         /* pNFS is not supported */
1593         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1594
1595         /* Referrals are supported, Migration is not. */
1596         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1597
1598         /* set the wire flags to return to client. */
1599         clid->flags = new->cl_exchange_flags;
1600 }
1601
1602 static bool client_has_state(struct nfs4_client *clp)
1603 {
1604         /*
1605          * Note clp->cl_openowners check isn't quite right: there's no
1606          * need to count owners without stateid's.
1607          *
1608          * Also note we should probably be using this in 4.0 case too.
1609          */
1610         return !list_empty(&clp->cl_openowners)
1611                 || !list_empty(&clp->cl_delegations)
1612                 || !list_empty(&clp->cl_sessions);
1613 }
1614
1615 __be32
1616 nfsd4_exchange_id(struct svc_rqst *rqstp,
1617                   struct nfsd4_compound_state *cstate,
1618                   struct nfsd4_exchange_id *exid)
1619 {
1620         struct nfs4_client *unconf, *conf, *new;
1621         __be32 status;
1622         char                    addr_str[INET6_ADDRSTRLEN];
1623         nfs4_verifier           verf = exid->verifier;
1624         struct sockaddr         *sa = svc_addr(rqstp);
1625         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1626         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1627
1628         rpc_ntop(sa, addr_str, sizeof(addr_str));
1629         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1630                 "ip_addr=%s flags %x, spa_how %d\n",
1631                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1632                 addr_str, exid->flags, exid->spa_how);
1633
1634         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1635                 return nfserr_inval;
1636
1637         /* Currently only support SP4_NONE */
1638         switch (exid->spa_how) {
1639         case SP4_NONE:
1640                 break;
1641         default:                                /* checked by xdr code */
1642                 WARN_ON_ONCE(1);
1643         case SP4_SSV:
1644         case SP4_MACH_CRED:
1645                 return nfserr_serverfault;      /* no excuse :-/ */
1646         }
1647
1648         /* Cases below refer to rfc 5661 section 18.35.4: */
1649         nfs4_lock_state();
1650         conf = find_confirmed_client_by_name(&exid->clname, nn);
1651         if (conf) {
1652                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1653                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1654
1655                 if (update) {
1656                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1657                                 status = nfserr_inval;
1658                                 goto out;
1659                         }
1660                         if (!creds_match) { /* case 9 */
1661                                 status = nfserr_perm;
1662                                 goto out;
1663                         }
1664                         if (!verfs_match) { /* case 8 */
1665                                 status = nfserr_not_same;
1666                                 goto out;
1667                         }
1668                         /* case 6 */
1669                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1670                         new = conf;
1671                         goto out_copy;
1672                 }
1673                 if (!creds_match) { /* case 3 */
1674                         if (client_has_state(conf)) {
1675                                 status = nfserr_clid_inuse;
1676                                 goto out;
1677                         }
1678                         expire_client(conf);
1679                         goto out_new;
1680                 }
1681                 if (verfs_match) { /* case 2 */
1682                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1683                         new = conf;
1684                         goto out_copy;
1685                 }
1686                 /* case 5, client reboot */
1687                 goto out_new;
1688         }
1689
1690         if (update) { /* case 7 */
1691                 status = nfserr_noent;
1692                 goto out;
1693         }
1694
1695         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1696         if (unconf) /* case 4, possible retry or client restart */
1697                 expire_client(unconf);
1698
1699         /* case 1 (normal case) */
1700 out_new:
1701         new = create_client(exid->clname, rqstp, &verf);
1702         if (new == NULL) {
1703                 status = nfserr_jukebox;
1704                 goto out;
1705         }
1706         new->cl_minorversion = 1;
1707
1708         gen_clid(new, nn);
1709         add_to_unconfirmed(new);
1710 out_copy:
1711         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1712         exid->clientid.cl_id = new->cl_clientid.cl_id;
1713
1714         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1715         nfsd4_set_ex_flags(new, exid);
1716
1717         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1718                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1719         status = nfs_ok;
1720
1721 out:
1722         nfs4_unlock_state();
1723         return status;
1724 }
1725
1726 static __be32
1727 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1728 {
1729         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1730                 slot_seqid);
1731
1732         /* The slot is in use, and no response has been sent. */
1733         if (slot_inuse) {
1734                 if (seqid == slot_seqid)
1735                         return nfserr_jukebox;
1736                 else
1737                         return nfserr_seq_misordered;
1738         }
1739         /* Note unsigned 32-bit arithmetic handles wraparound: */
1740         if (likely(seqid == slot_seqid + 1))
1741                 return nfs_ok;
1742         if (seqid == slot_seqid)
1743                 return nfserr_replay_cache;
1744         return nfserr_seq_misordered;
1745 }
1746
1747 /*
1748  * Cache the create session result into the create session single DRC
1749  * slot cache by saving the xdr structure. sl_seqid has been set.
1750  * Do this for solo or embedded create session operations.
1751  */
1752 static void
1753 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1754                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1755 {
1756         slot->sl_status = nfserr;
1757         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1758 }
1759
1760 static __be32
1761 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1762                             struct nfsd4_clid_slot *slot)
1763 {
1764         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1765         return slot->sl_status;
1766 }
1767
1768 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1769                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1770                         1 +     /* MIN tag is length with zero, only length */ \
1771                         3 +     /* version, opcount, opcode */ \
1772                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1773                                 /* seqid, slotID, slotID, cache */ \
1774                         4 ) * sizeof(__be32))
1775
1776 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1777                         2 +     /* verifier: AUTH_NULL, length 0 */\
1778                         1 +     /* status */ \
1779                         1 +     /* MIN tag is length with zero, only length */ \
1780                         3 +     /* opcount, opcode, opstatus*/ \
1781                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1782                                 /* seqid, slotID, slotID, slotID, status */ \
1783                         5 ) * sizeof(__be32))
1784
1785 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1786 {
1787         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1788                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1789 }
1790
1791 __be32
1792 nfsd4_create_session(struct svc_rqst *rqstp,
1793                      struct nfsd4_compound_state *cstate,
1794                      struct nfsd4_create_session *cr_ses)
1795 {
1796         struct sockaddr *sa = svc_addr(rqstp);
1797         struct nfs4_client *conf, *unconf;
1798         struct nfsd4_session *new;
1799         struct nfsd4_conn *conn;
1800         struct nfsd4_clid_slot *cs_slot = NULL;
1801         __be32 status = 0;
1802         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1803
1804         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1805                 return nfserr_inval;
1806         if (check_forechannel_attrs(cr_ses->fore_channel))
1807                 return nfserr_toosmall;
1808         new = alloc_session(&cr_ses->fore_channel, nn);
1809         if (!new)
1810                 return nfserr_jukebox;
1811         status = nfserr_jukebox;
1812         conn = alloc_conn_from_crses(rqstp, cr_ses);
1813         if (!conn)
1814                 goto out_free_session;
1815
1816         nfs4_lock_state();
1817         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1818         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1819
1820         if (conf) {
1821                 cs_slot = &conf->cl_cs_slot;
1822                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1823                 if (status == nfserr_replay_cache) {
1824                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1825                         goto out_free_conn;
1826                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1827                         status = nfserr_seq_misordered;
1828                         goto out_free_conn;
1829                 }
1830         } else if (unconf) {
1831                 struct nfs4_client *old;
1832                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1833                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1834                         status = nfserr_clid_inuse;
1835                         goto out_free_conn;
1836                 }
1837                 cs_slot = &unconf->cl_cs_slot;
1838                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1839                 if (status) {
1840                         /* an unconfirmed replay returns misordered */
1841                         status = nfserr_seq_misordered;
1842                         goto out_free_conn;
1843                 }
1844                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1845                 if (old)
1846                         expire_client(old);
1847                 move_to_confirmed(unconf);
1848                 conf = unconf;
1849         } else {
1850                 status = nfserr_stale_clientid;
1851                 goto out_free_conn;
1852         }
1853         status = nfs_ok;
1854         /*
1855          * We do not support RDMA or persistent sessions
1856          */
1857         cr_ses->flags &= ~SESSION4_PERSIST;
1858         cr_ses->flags &= ~SESSION4_RDMA;
1859
1860         init_session(rqstp, new, conf, cr_ses);
1861         nfsd4_init_conn(rqstp, conn, new);
1862
1863         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1864                NFS4_MAX_SESSIONID_LEN);
1865         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1866                 sizeof(struct nfsd4_channel_attrs));
1867         cs_slot->sl_seqid++;
1868         cr_ses->seqid = cs_slot->sl_seqid;
1869
1870         /* cache solo and embedded create sessions under the state lock */
1871         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1872         nfs4_unlock_state();
1873 out:
1874         dprintk("%s returns %d\n", __func__, ntohl(status));
1875         return status;
1876 out_free_conn:
1877         nfs4_unlock_state();
1878         free_conn(conn);
1879 out_free_session:
1880         __free_session(new);
1881         goto out;
1882 }
1883
1884 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1885 {
1886         switch (*dir) {
1887         case NFS4_CDFC4_FORE:
1888         case NFS4_CDFC4_BACK:
1889                 return nfs_ok;
1890         case NFS4_CDFC4_FORE_OR_BOTH:
1891         case NFS4_CDFC4_BACK_OR_BOTH:
1892                 *dir = NFS4_CDFC4_BOTH;
1893                 return nfs_ok;
1894         };
1895         return nfserr_inval;
1896 }
1897
1898 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1899 {
1900         struct nfsd4_session *session = cstate->session;
1901         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1902
1903         spin_lock(&nn->client_lock);
1904         session->se_cb_prog = bc->bc_cb_program;
1905         session->se_cb_sec = bc->bc_cb_sec;
1906         spin_unlock(&nn->client_lock);
1907
1908         nfsd4_probe_callback(session->se_client);
1909
1910         return nfs_ok;
1911 }
1912
1913 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1914                      struct nfsd4_compound_state *cstate,
1915                      struct nfsd4_bind_conn_to_session *bcts)
1916 {
1917         __be32 status;
1918         struct nfsd4_conn *conn;
1919         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1920
1921         if (!nfsd4_last_compound_op(rqstp))
1922                 return nfserr_not_only_op;
1923         spin_lock(&nn->client_lock);
1924         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
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(&nn->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         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
1960
1961         /* Notes:
1962          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1963          * - Should we return nfserr_back_chan_busy if waiting for
1964          *   callbacks on to-be-destroyed session?
1965          * - Do we need to clear any callback info from previous session?
1966          */
1967
1968         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1969                 if (!nfsd4_last_compound_op(r))
1970                         return nfserr_not_only_op;
1971         }
1972         dump_sessionid(__func__, &sessionid->sessionid);
1973         spin_lock(&nn->client_lock);
1974         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
1975         if (!ses) {
1976                 spin_unlock(&nn->client_lock);
1977                 goto out;
1978         }
1979
1980         unhash_session(ses);
1981         spin_unlock(&nn->client_lock);
1982
1983         nfs4_lock_state();
1984         nfsd4_probe_callback_sync(ses->se_client);
1985         nfs4_unlock_state();
1986
1987         spin_lock(&nn->client_lock);
1988         nfsd4_del_conns(ses);
1989         nfsd4_put_session_locked(ses);
1990         spin_unlock(&nn->client_lock);
1991         status = nfs_ok;
1992 out:
1993         dprintk("%s returns %d\n", __func__, ntohl(status));
1994         return status;
1995 }
1996
1997 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1998 {
1999         struct nfsd4_conn *c;
2000
2001         list_for_each_entry(c, &s->se_conns, cn_persession) {
2002                 if (c->cn_xprt == xpt) {
2003                         return c;
2004                 }
2005         }
2006         return NULL;
2007 }
2008
2009 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2010 {
2011         struct nfs4_client *clp = ses->se_client;
2012         struct nfsd4_conn *c;
2013         int ret;
2014
2015         spin_lock(&clp->cl_lock);
2016         c = __nfsd4_find_conn(new->cn_xprt, ses);
2017         if (c) {
2018                 spin_unlock(&clp->cl_lock);
2019                 free_conn(new);
2020                 return;
2021         }
2022         __nfsd4_hash_conn(new, ses);
2023         spin_unlock(&clp->cl_lock);
2024         ret = nfsd4_register_conn(new);
2025         if (ret)
2026                 /* oops; xprt is already down: */
2027                 nfsd4_conn_lost(&new->cn_xpt_user);
2028         return;
2029 }
2030
2031 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2032 {
2033         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2034
2035         return args->opcnt > session->se_fchannel.maxops;
2036 }
2037
2038 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2039                                   struct nfsd4_session *session)
2040 {
2041         struct xdr_buf *xb = &rqstp->rq_arg;
2042
2043         return xb->len > session->se_fchannel.maxreq_sz;
2044 }
2045
2046 __be32
2047 nfsd4_sequence(struct svc_rqst *rqstp,
2048                struct nfsd4_compound_state *cstate,
2049                struct nfsd4_sequence *seq)
2050 {
2051         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2052         struct nfsd4_session *session;
2053         struct nfsd4_slot *slot;
2054         struct nfsd4_conn *conn;
2055         __be32 status;
2056         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2057
2058         if (resp->opcnt != 1)
2059                 return nfserr_sequence_pos;
2060
2061         /*
2062          * Will be either used or freed by nfsd4_sequence_check_conn
2063          * below.
2064          */
2065         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2066         if (!conn)
2067                 return nfserr_jukebox;
2068
2069         spin_lock(&nn->client_lock);
2070         status = nfserr_badsession;
2071         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2072         if (!session)
2073                 goto out;
2074
2075         status = nfserr_too_many_ops;
2076         if (nfsd4_session_too_many_ops(rqstp, session))
2077                 goto out;
2078
2079         status = nfserr_req_too_big;
2080         if (nfsd4_request_too_big(rqstp, session))
2081                 goto out;
2082
2083         status = nfserr_badslot;
2084         if (seq->slotid >= session->se_fchannel.maxreqs)
2085                 goto out;
2086
2087         slot = session->se_slots[seq->slotid];
2088         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2089
2090         /* We do not negotiate the number of slots yet, so set the
2091          * maxslots to the session maxreqs which is used to encode
2092          * sr_highest_slotid and the sr_target_slot id to maxslots */
2093         seq->maxslots = session->se_fchannel.maxreqs;
2094
2095         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2096                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2097         if (status == nfserr_replay_cache) {
2098                 status = nfserr_seq_misordered;
2099                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2100                         goto out;
2101                 cstate->slot = slot;
2102                 cstate->session = session;
2103                 /* Return the cached reply status and set cstate->status
2104                  * for nfsd4_proc_compound processing */
2105                 status = nfsd4_replay_cache_entry(resp, seq);
2106                 cstate->status = nfserr_replay_cache;
2107                 goto out;
2108         }
2109         if (status)
2110                 goto out;
2111
2112         nfsd4_sequence_check_conn(conn, session);
2113         conn = NULL;
2114
2115         /* Success! bump slot seqid */
2116         slot->sl_seqid = seq->seqid;
2117         slot->sl_flags |= NFSD4_SLOT_INUSE;
2118         if (seq->cachethis)
2119                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2120         else
2121                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2122
2123         cstate->slot = slot;
2124         cstate->session = session;
2125
2126 out:
2127         /* Hold a session reference until done processing the compound. */
2128         if (cstate->session) {
2129                 struct nfs4_client *clp = session->se_client;
2130
2131                 nfsd4_get_session(cstate->session);
2132                 atomic_inc(&clp->cl_refcount);
2133                 switch (clp->cl_cb_state) {
2134                 case NFSD4_CB_DOWN:
2135                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2136                         break;
2137                 case NFSD4_CB_FAULT:
2138                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2139                         break;
2140                 default:
2141                         seq->status_flags = 0;
2142                 }
2143         }
2144         kfree(conn);
2145         spin_unlock(&nn->client_lock);
2146         dprintk("%s: return %d\n", __func__, ntohl(status));
2147         return status;
2148 }
2149
2150 __be32
2151 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2152 {
2153         struct nfs4_client *conf, *unconf, *clp;
2154         __be32 status = 0;
2155         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2156
2157         nfs4_lock_state();
2158         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2159         conf = find_confirmed_client(&dc->clientid, true, nn);
2160
2161         if (conf) {
2162                 clp = conf;
2163
2164                 if (!is_client_expired(conf) && client_has_state(conf)) {
2165                         status = nfserr_clientid_busy;
2166                         goto out;
2167                 }
2168
2169                 /* rfc5661 18.50.3 */
2170                 if (cstate->session && conf == cstate->session->se_client) {
2171                         status = nfserr_clientid_busy;
2172                         goto out;
2173                 }
2174         } else if (unconf)
2175                 clp = unconf;
2176         else {
2177                 status = nfserr_stale_clientid;
2178                 goto out;
2179         }
2180
2181         expire_client(clp);
2182 out:
2183         nfs4_unlock_state();
2184         dprintk("%s return %d\n", __func__, ntohl(status));
2185         return status;
2186 }
2187
2188 __be32
2189 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2190 {
2191         __be32 status = 0;
2192
2193         if (rc->rca_one_fs) {
2194                 if (!cstate->current_fh.fh_dentry)
2195                         return nfserr_nofilehandle;
2196                 /*
2197                  * We don't take advantage of the rca_one_fs case.
2198                  * That's OK, it's optional, we can safely ignore it.
2199                  */
2200                  return nfs_ok;
2201         }
2202
2203         nfs4_lock_state();
2204         status = nfserr_complete_already;
2205         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2206                              &cstate->session->se_client->cl_flags))
2207                 goto out;
2208
2209         status = nfserr_stale_clientid;
2210         if (is_client_expired(cstate->session->se_client))
2211                 /*
2212                  * The following error isn't really legal.
2213                  * But we only get here if the client just explicitly
2214                  * destroyed the client.  Surely it no longer cares what
2215                  * error it gets back on an operation for the dead
2216                  * client.
2217                  */
2218                 goto out;
2219
2220         status = nfs_ok;
2221         nfsd4_client_record_create(cstate->session->se_client);
2222 out:
2223         nfs4_unlock_state();
2224         return status;
2225 }
2226
2227 __be32
2228 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2229                   struct nfsd4_setclientid *setclid)
2230 {
2231         struct xdr_netobj       clname = setclid->se_name;
2232         nfs4_verifier           clverifier = setclid->se_verf;
2233         struct nfs4_client      *conf, *unconf, *new;
2234         __be32                  status;
2235         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2236
2237         /* Cases below refer to rfc 3530 section 14.2.33: */
2238         nfs4_lock_state();
2239         conf = find_confirmed_client_by_name(&clname, nn);
2240         if (conf) {
2241                 /* case 0: */
2242                 status = nfserr_clid_inuse;
2243                 if (clp_used_exchangeid(conf))
2244                         goto out;
2245                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2246                         char addr_str[INET6_ADDRSTRLEN];
2247                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2248                                  sizeof(addr_str));
2249                         dprintk("NFSD: setclientid: string in use by client "
2250                                 "at %s\n", addr_str);
2251                         goto out;
2252                 }
2253         }
2254         unconf = find_unconfirmed_client_by_name(&clname, nn);
2255         if (unconf)
2256                 expire_client(unconf);
2257         status = nfserr_jukebox;
2258         new = create_client(clname, rqstp, &clverifier);
2259         if (new == NULL)
2260                 goto out;
2261         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2262                 /* case 1: probable callback update */
2263                 copy_clid(new, conf);
2264         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2265                 gen_clid(new, nn);
2266         new->cl_minorversion = 0;
2267         gen_callback(new, setclid, rqstp);
2268         add_to_unconfirmed(new);
2269         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2270         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2271         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2272         status = nfs_ok;
2273 out:
2274         nfs4_unlock_state();
2275         return status;
2276 }
2277
2278
2279 __be32
2280 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2281                          struct nfsd4_compound_state *cstate,
2282                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2283 {
2284         struct nfs4_client *conf, *unconf;
2285         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
2286         clientid_t * clid = &setclientid_confirm->sc_clientid;
2287         __be32 status;
2288         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2289
2290         if (STALE_CLIENTID(clid, nn))
2291                 return nfserr_stale_clientid;
2292         nfs4_lock_state();
2293
2294         conf = find_confirmed_client(clid, false, nn);
2295         unconf = find_unconfirmed_client(clid, false, nn);
2296         /*
2297          * We try hard to give out unique clientid's, so if we get an
2298          * attempt to confirm the same clientid with a different cred,
2299          * there's a bug somewhere.  Let's charitably assume it's our
2300          * bug.
2301          */
2302         status = nfserr_serverfault;
2303         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2304                 goto out;
2305         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2306                 goto out;
2307         /* cases below refer to rfc 3530 section 14.2.34: */
2308         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2309                 if (conf && !unconf) /* case 2: probable retransmit */
2310                         status = nfs_ok;
2311                 else /* case 4: client hasn't noticed we rebooted yet? */
2312                         status = nfserr_stale_clientid;
2313                 goto out;
2314         }
2315         status = nfs_ok;
2316         if (conf) { /* case 1: callback update */
2317                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2318                 nfsd4_probe_callback(conf);
2319                 expire_client(unconf);
2320         } else { /* case 3: normal case; new or rebooted client */
2321                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2322                 if (conf)
2323                         expire_client(conf);
2324                 move_to_confirmed(unconf);
2325                 nfsd4_probe_callback(unconf);
2326         }
2327 out:
2328         nfs4_unlock_state();
2329         return status;
2330 }
2331
2332 static struct nfs4_file *nfsd4_alloc_file(void)
2333 {
2334         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2335 }
2336
2337 /* OPEN Share state helper functions */
2338 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2339 {
2340         unsigned int hashval = file_hashval(ino);
2341
2342         atomic_set(&fp->fi_ref, 1);
2343         INIT_LIST_HEAD(&fp->fi_hash);
2344         INIT_LIST_HEAD(&fp->fi_stateids);
2345         INIT_LIST_HEAD(&fp->fi_delegations);
2346         fp->fi_inode = igrab(ino);
2347         fp->fi_had_conflict = false;
2348         fp->fi_lease = NULL;
2349         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2350         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2351         spin_lock(&recall_lock);
2352         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2353         spin_unlock(&recall_lock);
2354 }
2355
2356 static void
2357 nfsd4_free_slab(struct kmem_cache **slab)
2358 {
2359         if (*slab == NULL)
2360                 return;
2361         kmem_cache_destroy(*slab);
2362         *slab = NULL;
2363 }
2364
2365 void
2366 nfsd4_free_slabs(void)
2367 {
2368         nfsd4_free_slab(&openowner_slab);
2369         nfsd4_free_slab(&lockowner_slab);
2370         nfsd4_free_slab(&file_slab);
2371         nfsd4_free_slab(&stateid_slab);
2372         nfsd4_free_slab(&deleg_slab);
2373 }
2374
2375 int
2376 nfsd4_init_slabs(void)
2377 {
2378         openowner_slab = kmem_cache_create("nfsd4_openowners",
2379                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2380         if (openowner_slab == NULL)
2381                 goto out_nomem;
2382         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2383                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2384         if (lockowner_slab == NULL)
2385                 goto out_nomem;
2386         file_slab = kmem_cache_create("nfsd4_files",
2387                         sizeof(struct nfs4_file), 0, 0, NULL);
2388         if (file_slab == NULL)
2389                 goto out_nomem;
2390         stateid_slab = kmem_cache_create("nfsd4_stateids",
2391                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2392         if (stateid_slab == NULL)
2393                 goto out_nomem;
2394         deleg_slab = kmem_cache_create("nfsd4_delegations",
2395                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2396         if (deleg_slab == NULL)
2397                 goto out_nomem;
2398         return 0;
2399 out_nomem:
2400         nfsd4_free_slabs();
2401         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2402         return -ENOMEM;
2403 }
2404
2405 void nfs4_free_openowner(struct nfs4_openowner *oo)
2406 {
2407         kfree(oo->oo_owner.so_owner.data);
2408         kmem_cache_free(openowner_slab, oo);
2409 }
2410
2411 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2412 {
2413         kfree(lo->lo_owner.so_owner.data);
2414         kmem_cache_free(lockowner_slab, lo);
2415 }
2416
2417 static void init_nfs4_replay(struct nfs4_replay *rp)
2418 {
2419         rp->rp_status = nfserr_serverfault;
2420         rp->rp_buflen = 0;
2421         rp->rp_buf = rp->rp_ibuf;
2422 }
2423
2424 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2425 {
2426         struct nfs4_stateowner *sop;
2427
2428         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2429         if (!sop)
2430                 return NULL;
2431
2432         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2433         if (!sop->so_owner.data) {
2434                 kmem_cache_free(slab, sop);
2435                 return NULL;
2436         }
2437         sop->so_owner.len = owner->len;
2438
2439         INIT_LIST_HEAD(&sop->so_stateids);
2440         sop->so_client = clp;
2441         init_nfs4_replay(&sop->so_replay);
2442         return sop;
2443 }
2444
2445 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2446 {
2447         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2448
2449         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2450         list_add(&oo->oo_perclient, &clp->cl_openowners);
2451 }
2452
2453 static struct nfs4_openowner *
2454 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2455         struct nfs4_openowner *oo;
2456
2457         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2458         if (!oo)
2459                 return NULL;
2460         oo->oo_owner.so_is_open_owner = 1;
2461         oo->oo_owner.so_seqid = open->op_seqid;
2462         oo->oo_flags = NFS4_OO_NEW;
2463         oo->oo_time = 0;
2464         oo->oo_last_closed_stid = NULL;
2465         INIT_LIST_HEAD(&oo->oo_close_lru);
2466         hash_openowner(oo, clp, strhashval);
2467         return oo;
2468 }
2469
2470 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2471         struct nfs4_openowner *oo = open->op_openowner;
2472
2473         stp->st_stid.sc_type = NFS4_OPEN_STID;
2474         INIT_LIST_HEAD(&stp->st_lockowners);
2475         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2476         list_add(&stp->st_perfile, &fp->fi_stateids);
2477         stp->st_stateowner = &oo->oo_owner;
2478         get_nfs4_file(fp);
2479         stp->st_file = fp;
2480         stp->st_access_bmap = 0;
2481         stp->st_deny_bmap = 0;
2482         set_access(open->op_share_access, stp);
2483         set_deny(open->op_share_deny, stp);
2484         stp->st_openstp = NULL;
2485 }
2486
2487 static void
2488 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2489 {
2490         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2491
2492         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2493
2494         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2495         oo->oo_time = get_seconds();
2496 }
2497
2498 static int
2499 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2500                                                         clientid_t *clid)
2501 {
2502         return (sop->so_owner.len == owner->len) &&
2503                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2504                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2505 }
2506
2507 static struct nfs4_openowner *
2508 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2509                         bool sessions, struct nfsd_net *nn)
2510 {
2511         struct nfs4_stateowner *so;
2512         struct nfs4_openowner *oo;
2513         struct nfs4_client *clp;
2514
2515         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2516                 if (!so->so_is_open_owner)
2517                         continue;
2518                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2519                         oo = openowner(so);
2520                         clp = oo->oo_owner.so_client;
2521                         if ((bool)clp->cl_minorversion != sessions)
2522                                 return NULL;
2523                         renew_client(oo->oo_owner.so_client);
2524                         return oo;
2525                 }
2526         }
2527         return NULL;
2528 }
2529
2530 /* search file_hashtbl[] for file */
2531 static struct nfs4_file *
2532 find_file(struct inode *ino)
2533 {
2534         unsigned int hashval = file_hashval(ino);
2535         struct nfs4_file *fp;
2536
2537         spin_lock(&recall_lock);
2538         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2539                 if (fp->fi_inode == ino) {
2540                         get_nfs4_file(fp);
2541                         spin_unlock(&recall_lock);
2542                         return fp;
2543                 }
2544         }
2545         spin_unlock(&recall_lock);
2546         return NULL;
2547 }
2548
2549 /*
2550  * Called to check deny when READ with all zero stateid or
2551  * WRITE with all zero or all one stateid
2552  */
2553 static __be32
2554 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2555 {
2556         struct inode *ino = current_fh->fh_dentry->d_inode;
2557         struct nfs4_file *fp;
2558         struct nfs4_ol_stateid *stp;
2559         __be32 ret;
2560
2561         dprintk("NFSD: nfs4_share_conflict\n");
2562
2563         fp = find_file(ino);
2564         if (!fp)
2565                 return nfs_ok;
2566         ret = nfserr_locked;
2567         /* Search for conflicting share reservations */
2568         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2569                 if (test_deny(deny_type, stp) ||
2570                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2571                         goto out;
2572         }
2573         ret = nfs_ok;
2574 out:
2575         put_nfs4_file(fp);
2576         return ret;
2577 }
2578
2579 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2580 {
2581         /* We're assuming the state code never drops its reference
2582          * without first removing the lease.  Since we're in this lease
2583          * callback (and since the lease code is serialized by the kernel
2584          * lock) we know the server hasn't removed the lease yet, we know
2585          * it's safe to take a reference: */
2586         atomic_inc(&dp->dl_count);
2587
2588         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2589
2590         /* only place dl_time is set. protected by lock_flocks*/
2591         dp->dl_time = get_seconds();
2592
2593         nfsd4_cb_recall(dp);
2594 }
2595
2596 /* Called from break_lease() with lock_flocks() held. */
2597 static void nfsd_break_deleg_cb(struct file_lock *fl)
2598 {
2599         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2600         struct nfs4_delegation *dp;
2601
2602         if (!fp) {
2603                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2604                 return;
2605         }
2606         if (fp->fi_had_conflict) {
2607                 WARN(1, "duplicate break on %p\n", fp);
2608                 return;
2609         }
2610         /*
2611          * We don't want the locks code to timeout the lease for us;
2612          * we'll remove it ourself if a delegation isn't returned
2613          * in time:
2614          */
2615         fl->fl_break_time = 0;
2616
2617         spin_lock(&recall_lock);
2618         fp->fi_had_conflict = true;
2619         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2620                 nfsd_break_one_deleg(dp);
2621         spin_unlock(&recall_lock);
2622 }
2623
2624 static
2625 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2626 {
2627         if (arg & F_UNLCK)
2628                 return lease_modify(onlist, arg);
2629         else
2630                 return -EAGAIN;
2631 }
2632
2633 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2634         .lm_break = nfsd_break_deleg_cb,
2635         .lm_change = nfsd_change_deleg_cb,
2636 };
2637
2638 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2639 {
2640         if (nfsd4_has_session(cstate))
2641                 return nfs_ok;
2642         if (seqid == so->so_seqid - 1)
2643                 return nfserr_replay_me;
2644         if (seqid == so->so_seqid)
2645                 return nfs_ok;
2646         return nfserr_bad_seqid;
2647 }
2648
2649 __be32
2650 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2651                     struct nfsd4_open *open, struct nfsd_net *nn)
2652 {
2653         clientid_t *clientid = &open->op_clientid;
2654         struct nfs4_client *clp = NULL;
2655         unsigned int strhashval;
2656         struct nfs4_openowner *oo = NULL;
2657         __be32 status;
2658
2659         if (STALE_CLIENTID(&open->op_clientid, nn))
2660                 return nfserr_stale_clientid;
2661         /*
2662          * In case we need it later, after we've already created the
2663          * file and don't want to risk a further failure:
2664          */
2665         open->op_file = nfsd4_alloc_file();
2666         if (open->op_file == NULL)
2667                 return nfserr_jukebox;
2668
2669         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2670         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2671         open->op_openowner = oo;
2672         if (!oo) {
2673                 clp = find_confirmed_client(clientid, cstate->minorversion,
2674                                             nn);
2675                 if (clp == NULL)
2676                         return nfserr_expired;
2677                 goto new_owner;
2678         }
2679         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2680                 /* Replace unconfirmed owners without checking for replay. */
2681                 clp = oo->oo_owner.so_client;
2682                 release_openowner(oo);
2683                 open->op_openowner = NULL;
2684                 goto new_owner;
2685         }
2686         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2687         if (status)
2688                 return status;
2689         clp = oo->oo_owner.so_client;
2690         goto alloc_stateid;
2691 new_owner:
2692         oo = alloc_init_open_stateowner(strhashval, clp, open);
2693         if (oo == NULL)
2694                 return nfserr_jukebox;
2695         open->op_openowner = oo;
2696 alloc_stateid:
2697         open->op_stp = nfs4_alloc_stateid(clp);
2698         if (!open->op_stp)
2699                 return nfserr_jukebox;
2700         return nfs_ok;
2701 }
2702
2703 static inline __be32
2704 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2705 {
2706         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2707                 return nfserr_openmode;
2708         else
2709                 return nfs_ok;
2710 }
2711
2712 static int share_access_to_flags(u32 share_access)
2713 {
2714         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2715 }
2716
2717 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2718 {
2719         struct nfs4_stid *ret;
2720
2721         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2722         if (!ret)
2723                 return NULL;
2724         return delegstateid(ret);
2725 }
2726
2727 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2728 {
2729         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2730                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2731 }
2732
2733 static __be32
2734 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2735                 struct nfs4_delegation **dp)
2736 {
2737         int flags;
2738         __be32 status = nfserr_bad_stateid;
2739
2740         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2741         if (*dp == NULL)
2742                 goto out;
2743         flags = share_access_to_flags(open->op_share_access);
2744         status = nfs4_check_delegmode(*dp, flags);
2745         if (status)
2746                 *dp = NULL;
2747 out:
2748         if (!nfsd4_is_deleg_cur(open))
2749                 return nfs_ok;
2750         if (status)
2751                 return status;
2752         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2753         return nfs_ok;
2754 }
2755
2756 static __be32
2757 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2758 {
2759         struct nfs4_ol_stateid *local;
2760         struct nfs4_openowner *oo = open->op_openowner;
2761
2762         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2763                 /* ignore lock owners */
2764                 if (local->st_stateowner->so_is_open_owner == 0)
2765                         continue;
2766                 /* remember if we have seen this open owner */
2767                 if (local->st_stateowner == &oo->oo_owner)
2768                         *stpp = local;
2769                 /* check for conflicting share reservations */
2770                 if (!test_share(local, open))
2771                         return nfserr_share_denied;
2772         }
2773         return nfs_ok;
2774 }
2775
2776 static inline int nfs4_access_to_access(u32 nfs4_access)
2777 {
2778         int flags = 0;
2779
2780         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2781                 flags |= NFSD_MAY_READ;
2782         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2783                 flags |= NFSD_MAY_WRITE;
2784         return flags;
2785 }
2786
2787 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2788                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2789 {
2790         __be32 status;
2791         int oflag = nfs4_access_to_omode(open->op_share_access);
2792         int access = nfs4_access_to_access(open->op_share_access);
2793
2794         if (!fp->fi_fds[oflag]) {
2795                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2796                         &fp->fi_fds[oflag]);
2797                 if (status)
2798                         return status;
2799         }
2800         nfs4_file_get_access(fp, oflag);
2801
2802         return nfs_ok;
2803 }
2804
2805 static inline __be32
2806 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2807                 struct nfsd4_open *open)
2808 {
2809         struct iattr iattr = {
2810                 .ia_valid = ATTR_SIZE,
2811                 .ia_size = 0,
2812         };
2813         if (!open->op_truncate)
2814                 return 0;
2815         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2816                 return nfserr_inval;
2817         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2818 }
2819
2820 static __be32
2821 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)
2822 {
2823         u32 op_share_access = open->op_share_access;
2824         bool new_access;
2825         __be32 status;
2826
2827         new_access = !test_access(op_share_access, stp);
2828         if (new_access) {
2829                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2830                 if (status)
2831                         return status;
2832         }
2833         status = nfsd4_truncate(rqstp, cur_fh, open);
2834         if (status) {
2835                 if (new_access) {
2836                         int oflag = nfs4_access_to_omode(op_share_access);
2837                         nfs4_file_put_access(fp, oflag);
2838                 }
2839                 return status;
2840         }
2841         /* remember the open */
2842         set_access(op_share_access, stp);
2843         set_deny(open->op_share_deny, stp);
2844
2845         return nfs_ok;
2846 }
2847
2848
2849 static void
2850 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2851 {
2852         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2853 }
2854
2855 /* Should we give out recallable state?: */
2856 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2857 {
2858         if (clp->cl_cb_state == NFSD4_CB_UP)
2859                 return true;
2860         /*
2861          * In the sessions case, since we don't have to establish a
2862          * separate connection for callbacks, we assume it's OK
2863          * until we hear otherwise:
2864          */
2865         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2866 }
2867
2868 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2869 {
2870         struct file_lock *fl;
2871
2872         fl = locks_alloc_lock();
2873         if (!fl)
2874                 return NULL;
2875         locks_init_lock(fl);
2876         fl->fl_lmops = &nfsd_lease_mng_ops;
2877         fl->fl_flags = FL_LEASE;
2878         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2879         fl->fl_end = OFFSET_MAX;
2880         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2881         fl->fl_pid = current->tgid;
2882         return fl;
2883 }
2884
2885 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2886 {
2887         struct nfs4_file *fp = dp->dl_file;
2888         struct file_lock *fl;
2889         int status;
2890
2891         fl = nfs4_alloc_init_lease(dp, flag);
2892         if (!fl)
2893                 return -ENOMEM;
2894         fl->fl_file = find_readable_file(fp);
2895         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2896         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2897         if (status) {
2898                 list_del_init(&dp->dl_perclnt);
2899                 locks_free_lock(fl);
2900                 return -ENOMEM;
2901         }
2902         fp->fi_lease = fl;
2903         fp->fi_deleg_file = get_file(fl->fl_file);
2904         atomic_set(&fp->fi_delegees, 1);
2905         list_add(&dp->dl_perfile, &fp->fi_delegations);
2906         return 0;
2907 }
2908
2909 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2910 {
2911         struct nfs4_file *fp = dp->dl_file;
2912
2913         if (!fp->fi_lease)
2914                 return nfs4_setlease(dp, flag);
2915         spin_lock(&recall_lock);
2916         if (fp->fi_had_conflict) {
2917                 spin_unlock(&recall_lock);
2918                 return -EAGAIN;
2919         }
2920         atomic_inc(&fp->fi_delegees);
2921         list_add(&dp->dl_perfile, &fp->fi_delegations);
2922         spin_unlock(&recall_lock);
2923         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2924         return 0;
2925 }
2926
2927 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2928 {
2929         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2930         if (status == -EAGAIN)
2931                 open->op_why_no_deleg = WND4_CONTENTION;
2932         else {
2933                 open->op_why_no_deleg = WND4_RESOURCE;
2934                 switch (open->op_deleg_want) {
2935                 case NFS4_SHARE_WANT_READ_DELEG:
2936                 case NFS4_SHARE_WANT_WRITE_DELEG:
2937                 case NFS4_SHARE_WANT_ANY_DELEG:
2938                         break;
2939                 case NFS4_SHARE_WANT_CANCEL:
2940                         open->op_why_no_deleg = WND4_CANCELLED;
2941                         break;
2942                 case NFS4_SHARE_WANT_NO_DELEG:
2943                         WARN_ON_ONCE(1);
2944                 }
2945         }
2946 }
2947
2948 /*
2949  * Attempt to hand out a delegation.
2950  */
2951 static void
2952 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2953                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2954 {
2955         struct nfs4_delegation *dp;
2956         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2957         int cb_up;
2958         int status = 0, flag = 0;
2959
2960         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2961         flag = NFS4_OPEN_DELEGATE_NONE;
2962         open->op_recall = 0;
2963         switch (open->op_claim_type) {
2964                 case NFS4_OPEN_CLAIM_PREVIOUS:
2965                         if (!cb_up)
2966                                 open->op_recall = 1;
2967                         flag = open->op_delegate_type;
2968                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2969                                 goto out;
2970                         break;
2971                 case NFS4_OPEN_CLAIM_NULL:
2972                         /* Let's not give out any delegations till everyone's
2973                          * had the chance to reclaim theirs.... */
2974                         if (locks_in_grace(net))
2975                                 goto out;
2976                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2977                                 goto out;
2978                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2979                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2980                         else
2981                                 flag = NFS4_OPEN_DELEGATE_READ;
2982                         break;
2983                 default:
2984                         goto out;
2985         }
2986
2987         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2988         if (dp == NULL)
2989                 goto out_no_deleg;
2990         status = nfs4_set_delegation(dp, flag);
2991         if (status)
2992                 goto out_free;
2993
2994         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2995
2996         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2997                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2998 out:
2999         open->op_delegate_type = flag;
3000         if (flag == NFS4_OPEN_DELEGATE_NONE) {
3001                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
3002                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
3003                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
3004
3005                 /* 4.1 client asking for a delegation? */
3006                 if (open->op_deleg_want)
3007                         nfsd4_open_deleg_none_ext(open, status);
3008         }
3009         return;
3010 out_free:
3011         unhash_stid(&dp->dl_stid);
3012         nfs4_put_delegation(dp);
3013 out_no_deleg:
3014         flag = NFS4_OPEN_DELEGATE_NONE;
3015         goto out;
3016 }
3017
3018 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
3019                                         struct nfs4_delegation *dp)
3020 {
3021         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
3022             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3023                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3024                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3025         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3026                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3027                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3028                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3029         }
3030         /* Otherwise the client must be confused wanting a delegation
3031          * it already has, therefore we don't return
3032          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3033          */
3034 }
3035
3036 /*
3037  * called with nfs4_lock_state() held.
3038  */
3039 __be32
3040 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3041 {
3042         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3043         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3044         struct nfs4_file *fp = NULL;
3045         struct inode *ino = current_fh->fh_dentry->d_inode;
3046         struct nfs4_ol_stateid *stp = NULL;
3047         struct nfs4_delegation *dp = NULL;
3048         __be32 status;
3049
3050         /*
3051          * Lookup file; if found, lookup stateid and check open request,
3052          * and check for delegations in the process of being recalled.
3053          * If not found, create the nfs4_file struct
3054          */
3055         fp = find_file(ino);
3056         if (fp) {
3057                 if ((status = nfs4_check_open(fp, open, &stp)))
3058                         goto out;
3059                 status = nfs4_check_deleg(cl, fp, open, &dp);
3060                 if (status)
3061                         goto out;
3062         } else {
3063                 status = nfserr_bad_stateid;
3064                 if (nfsd4_is_deleg_cur(open))
3065                         goto out;
3066                 status = nfserr_jukebox;
3067                 fp = open->op_file;
3068                 open->op_file = NULL;
3069                 nfsd4_init_file(fp, ino);
3070         }
3071
3072         /*
3073          * OPEN the file, or upgrade an existing OPEN.
3074          * If truncate fails, the OPEN fails.
3075          */
3076         if (stp) {
3077                 /* Stateid was found, this is an OPEN upgrade */
3078                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3079                 if (status)
3080                         goto out;
3081         } else {
3082                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3083                 if (status)
3084                         goto out;
3085                 status = nfsd4_truncate(rqstp, current_fh, open);
3086                 if (status)
3087                         goto out;
3088                 stp = open->op_stp;
3089                 open->op_stp = NULL;
3090                 init_open_stateid(stp, fp, open);
3091         }
3092         update_stateid(&stp->st_stid.sc_stateid);
3093         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3094
3095         if (nfsd4_has_session(&resp->cstate)) {
3096                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3097
3098                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3099                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3100                         open->op_why_no_deleg = WND4_NOT_WANTED;
3101                         goto nodeleg;
3102                 }
3103         }
3104
3105         /*
3106         * Attempt to hand out a delegation. No error return, because the
3107         * OPEN succeeds even if we fail.
3108         */
3109         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3110 nodeleg:
3111         status = nfs_ok;
3112
3113         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3114                 STATEID_VAL(&stp->st_stid.sc_stateid));
3115 out:
3116         /* 4.1 client trying to upgrade/downgrade delegation? */
3117         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3118             open->op_deleg_want)
3119                 nfsd4_deleg_xgrade_none_ext(open, dp);
3120
3121         if (fp)
3122                 put_nfs4_file(fp);
3123         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3124                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3125         /*
3126         * To finish the open response, we just need to set the rflags.
3127         */
3128         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3129         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3130             !nfsd4_has_session(&resp->cstate))
3131                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3132
3133         return status;
3134 }
3135
3136 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3137 {
3138         if (open->op_openowner) {
3139                 struct nfs4_openowner *oo = open->op_openowner;
3140
3141                 if (!list_empty(&oo->oo_owner.so_stateids))
3142                         list_del_init(&oo->oo_close_lru);
3143                 if (oo->oo_flags & NFS4_OO_NEW) {
3144                         if (status) {
3145                                 release_openowner(oo);
3146                                 open->op_openowner = NULL;
3147                         } else
3148                                 oo->oo_flags &= ~NFS4_OO_NEW;
3149                 }
3150         }
3151         if (open->op_file)
3152                 nfsd4_free_file(open->op_file);
3153         if (open->op_stp)
3154                 free_generic_stateid(open->op_stp);
3155 }
3156
3157 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3158 {
3159         struct nfs4_client *found;
3160
3161         if (STALE_CLIENTID(clid, nn))
3162                 return nfserr_stale_clientid;
3163         found = find_confirmed_client(clid, session, nn);
3164         if (clp)
3165                 *clp = found;
3166         return found ? nfs_ok : nfserr_expired;
3167 }
3168
3169 __be32
3170 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3171             clientid_t *clid)
3172 {
3173         struct nfs4_client *clp;
3174         __be32 status;
3175         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3176
3177         nfs4_lock_state();
3178         dprintk("process_renew(%08x/%08x): starting\n", 
3179                         clid->cl_boot, clid->cl_id);
3180         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3181         if (status)
3182                 goto out;
3183         status = nfserr_cb_path_down;
3184         if (!list_empty(&clp->cl_delegations)
3185                         && clp->cl_cb_state != NFSD4_CB_UP)
3186                 goto out;
3187         status = nfs_ok;
3188 out:
3189         nfs4_unlock_state();
3190         return status;
3191 }
3192
3193 static void
3194 nfsd4_end_grace(struct nfsd_net *nn)
3195 {
3196         /* do nothing if grace period already ended */
3197         if (nn->grace_ended)
3198                 return;
3199
3200         dprintk("NFSD: end of grace period\n");
3201         nn->grace_ended = true;
3202         nfsd4_record_grace_done(nn, nn->boot_time);
3203         locks_end_grace(&nn->nfsd4_manager);
3204         /*
3205          * Now that every NFSv4 client has had the chance to recover and
3206          * to see the (possibly new, possibly shorter) lease time, we
3207          * can safely set the next grace time to the current lease time:
3208          */
3209         nn->nfsd4_grace = nn->nfsd4_lease;
3210 }
3211
3212 static time_t
3213 nfs4_laundromat(struct nfsd_net *nn)
3214 {
3215         struct nfs4_client *clp;
3216         struct nfs4_openowner *oo;
3217         struct nfs4_delegation *dp;
3218         struct list_head *pos, *next, reaplist;
3219         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3220         time_t t, clientid_val = nn->nfsd4_lease;
3221         time_t u, test_val = nn->nfsd4_lease;
3222
3223         nfs4_lock_state();
3224
3225         dprintk("NFSD: laundromat service - starting\n");
3226         nfsd4_end_grace(nn);
3227         INIT_LIST_HEAD(&reaplist);
3228         spin_lock(&nn->client_lock);
3229         list_for_each_safe(pos, next, &nn->client_lru) {
3230                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3231                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3232                         t = clp->cl_time - cutoff;
3233                         if (clientid_val > t)
3234                                 clientid_val = t;
3235                         break;
3236                 }
3237                 if (atomic_read(&clp->cl_refcount)) {
3238                         dprintk("NFSD: client in use (clientid %08x)\n",
3239                                 clp->cl_clientid.cl_id);
3240                         continue;
3241                 }
3242                 unhash_client_locked(clp);
3243                 list_add(&clp->cl_lru, &reaplist);
3244         }
3245         spin_unlock(&nn->client_lock);
3246         list_for_each_safe(pos, next, &reaplist) {
3247                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3248                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3249                         clp->cl_clientid.cl_id);
3250                 expire_client(clp);
3251         }
3252         spin_lock(&recall_lock);
3253         list_for_each_safe(pos, next, &del_recall_lru) {
3254                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3255                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3256                         continue;
3257                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3258                         u = dp->dl_time - cutoff;
3259                         if (test_val > u)
3260                                 test_val = u;
3261                         break;
3262                 }
3263                 list_move(&dp->dl_recall_lru, &reaplist);
3264         }
3265         spin_unlock(&recall_lock);
3266         list_for_each_safe(pos, next, &reaplist) {
3267                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3268                 unhash_delegation(dp);
3269         }
3270         test_val = nn->nfsd4_lease;
3271         list_for_each_safe(pos, next, &nn->close_lru) {
3272                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3273                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3274                         u = oo->oo_time - cutoff;
3275                         if (test_val > u)
3276                                 test_val = u;
3277                         break;
3278                 }
3279                 release_openowner(oo);
3280         }
3281         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3282                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3283         nfs4_unlock_state();
3284         return clientid_val;
3285 }
3286
3287 static struct workqueue_struct *laundry_wq;
3288 static void laundromat_main(struct work_struct *);
3289
3290 static void
3291 laundromat_main(struct work_struct *laundry)
3292 {
3293         time_t t;
3294         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3295                                                   work);
3296         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3297                                            laundromat_work);
3298
3299         t = nfs4_laundromat(nn);
3300         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3301         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3302 }
3303
3304 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3305 {
3306         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3307                 return nfserr_bad_stateid;
3308         return nfs_ok;
3309 }
3310
3311 static int
3312 STALE_STATEID(stateid_t *stateid, struct nfsd_net *nn)
3313 {
3314         if (stateid->si_opaque.so_clid.cl_boot == nn->boot_time)
3315                 return 0;
3316         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3317                 STATEID_VAL(stateid));
3318         return 1;
3319 }
3320
3321 static inline int
3322 access_permit_read(struct nfs4_ol_stateid *stp)
3323 {
3324         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3325                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3326                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3327 }
3328
3329 static inline int
3330 access_permit_write(struct nfs4_ol_stateid *stp)
3331 {
3332         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3333                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3334 }
3335
3336 static
3337 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3338 {
3339         __be32 status = nfserr_openmode;
3340
3341         /* For lock stateid's, we test the parent open, not the lock: */
3342         if (stp->st_openstp)
3343                 stp = stp->st_openstp;
3344         if ((flags & WR_STATE) && !access_permit_write(stp))
3345                 goto out;
3346         if ((flags & RD_STATE) && !access_permit_read(stp))
3347                 goto out;
3348         status = nfs_ok;
3349 out:
3350         return status;
3351 }
3352
3353 static inline __be32
3354 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3355 {
3356         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3357                 return nfs_ok;
3358         else if (locks_in_grace(net)) {
3359                 /* Answer in remaining cases depends on existence of
3360                  * conflicting state; so we must wait out the grace period. */
3361                 return nfserr_grace;
3362         } else if (flags & WR_STATE)
3363                 return nfs4_share_conflict(current_fh,
3364                                 NFS4_SHARE_DENY_WRITE);
3365         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3366                 return nfs4_share_conflict(current_fh,
3367                                 NFS4_SHARE_DENY_READ);
3368 }
3369
3370 /*
3371  * Allow READ/WRITE during grace period on recovered state only for files
3372  * that are not able to provide mandatory locking.
3373  */
3374 static inline int
3375 grace_disallows_io(struct net *net, struct inode *inode)
3376 {
3377         return locks_in_grace(net) && mandatory_lock(inode);
3378 }
3379
3380 /* Returns true iff a is later than b: */
3381 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3382 {
3383         return (s32)a->si_generation - (s32)b->si_generation > 0;
3384 }
3385
3386 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3387 {
3388         /*
3389          * When sessions are used the stateid generation number is ignored
3390          * when it is zero.
3391          */
3392         if (has_session && in->si_generation == 0)
3393                 return nfs_ok;
3394
3395         if (in->si_generation == ref->si_generation)
3396                 return nfs_ok;
3397
3398         /* If the client sends us a stateid from the future, it's buggy: */
3399         if (stateid_generation_after(in, ref))
3400                 return nfserr_bad_stateid;
3401         /*
3402          * However, we could see a stateid from the past, even from a
3403          * non-buggy client.  For example, if the client sends a lock
3404          * while some IO is outstanding, the lock may bump si_generation
3405          * while the IO is still in flight.  The client could avoid that
3406          * situation by waiting for responses on all the IO requests,
3407          * but better performance may result in retrying IO that
3408          * receives an old_stateid error if requests are rarely
3409          * reordered in flight:
3410          */
3411         return nfserr_old_stateid;
3412 }
3413
3414 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3415 {
3416         struct nfs4_stid *s;
3417         struct nfs4_ol_stateid *ols;
3418         __be32 status;
3419
3420         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3421                 return nfserr_bad_stateid;
3422         /* Client debugging aid. */
3423         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3424                 char addr_str[INET6_ADDRSTRLEN];
3425                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3426                                  sizeof(addr_str));
3427                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3428                                         "with incorrect client ID\n", addr_str);
3429                 return nfserr_bad_stateid;
3430         }
3431         s = find_stateid(cl, stateid);
3432         if (!s)
3433                 return nfserr_bad_stateid;
3434         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3435         if (status)
3436                 return status;
3437         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3438                 return nfs_ok;
3439         ols = openlockstateid(s);
3440         if (ols->st_stateowner->so_is_open_owner
3441             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3442                 return nfserr_bad_stateid;
3443         return nfs_ok;
3444 }
3445
3446 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3447                                    struct nfs4_stid **s, bool sessions,
3448                                    struct nfsd_net *nn)
3449 {
3450         struct nfs4_client *cl;
3451
3452         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3453                 return nfserr_bad_stateid;
3454         if (STALE_STATEID(stateid, nn))
3455                 return nfserr_stale_stateid;
3456         cl = find_confirmed_client(&stateid->si_opaque.so_clid, sessions, nn);
3457         if (!cl)
3458                 return nfserr_expired;
3459         *s = find_stateid_by_type(cl, stateid, typemask);
3460         if (!*s)
3461                 return nfserr_bad_stateid;
3462         return nfs_ok;
3463
3464 }
3465
3466 /*
3467 * Checks for stateid operations
3468 */
3469 __be32
3470 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3471                            stateid_t *stateid, int flags, struct file **filpp)
3472 {
3473         struct nfs4_stid *s;
3474         struct nfs4_ol_stateid *stp = NULL;
3475         struct nfs4_delegation *dp = NULL;
3476         struct svc_fh *current_fh = &cstate->current_fh;
3477         struct inode *ino = current_fh->fh_dentry->d_inode;
3478         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3479         __be32 status;
3480
3481         if (filpp)
3482                 *filpp = NULL;
3483
3484         if (grace_disallows_io(net, ino))
3485                 return nfserr_grace;
3486
3487         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3488                 return check_special_stateids(net, current_fh, stateid, flags);
3489
3490         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3491                                       &s, cstate->minorversion, nn);
3492         if (status)
3493                 return status;
3494         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3495         if (status)
3496                 goto out;
3497         switch (s->sc_type) {
3498         case NFS4_DELEG_STID:
3499                 dp = delegstateid(s);
3500                 status = nfs4_check_delegmode(dp, flags);
3501                 if (status)
3502                         goto out;
3503                 if (filpp) {
3504                         *filpp = dp->dl_file->fi_deleg_file;
3505                         if (!*filpp) {
3506                                 WARN_ON_ONCE(1);
3507                                 status = nfserr_serverfault;
3508                                 goto out;
3509                         }
3510                 }
3511                 break;
3512         case NFS4_OPEN_STID:
3513         case NFS4_LOCK_STID:
3514                 stp = openlockstateid(s);
3515                 status = nfs4_check_fh(current_fh, stp);
3516                 if (status)
3517                         goto out;
3518                 if (stp->st_stateowner->so_is_open_owner
3519                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3520                         goto out;
3521                 status = nfs4_check_openmode(stp, flags);
3522                 if (status)
3523                         goto out;
3524                 if (filpp) {
3525                         if (flags & RD_STATE)
3526                                 *filpp = find_readable_file(stp->st_file);
3527                         else
3528                                 *filpp = find_writeable_file(stp->st_file);
3529                 }
3530                 break;
3531         default:
3532                 return nfserr_bad_stateid;
3533         }
3534         status = nfs_ok;
3535 out:
3536         return status;
3537 }
3538
3539 static __be32
3540 nfsd4_free_lock_stateid(struct nfs4_ol_stateid *stp)
3541 {
3542         if (check_for_locks(stp->st_file, lockowner(stp->st_stateowner)))
3543                 return nfserr_locks_held;
3544         release_lock_stateid(stp);
3545         return nfs_ok;
3546 }
3547
3548 /*
3549  * Test if the stateid is valid
3550  */
3551 __be32
3552 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3553                    struct nfsd4_test_stateid *test_stateid)
3554 {
3555         struct nfsd4_test_stateid_id *stateid;
3556         struct nfs4_client *cl = cstate->session->se_client;
3557
3558         nfs4_lock_state();
3559         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
3560                 stateid->ts_id_status =
3561                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
3562         nfs4_unlock_state();
3563
3564         return nfs_ok;
3565 }
3566
3567 __be32
3568 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3569                    struct nfsd4_free_stateid *free_stateid)
3570 {
3571         stateid_t *stateid = &free_stateid->fr_stateid;
3572         struct nfs4_stid *s;
3573         struct nfs4_client *cl = cstate->session->se_client;
3574         __be32 ret = nfserr_bad_stateid;
3575
3576         nfs4_lock_state();
3577         s = find_stateid(cl, stateid);
3578         if (!s)
3579                 goto out;
3580         switch (s->sc_type) {
3581         case NFS4_DELEG_STID:
3582                 ret = nfserr_locks_held;
3583                 goto out;
3584         case NFS4_OPEN_STID:
3585         case NFS4_LOCK_STID:
3586                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
3587                 if (ret)
3588                         goto out;
3589                 if (s->sc_type == NFS4_LOCK_STID)
3590                         ret = nfsd4_free_lock_stateid(openlockstateid(s));
3591                 else
3592                         ret = nfserr_locks_held;
3593                 break;
3594         default:
3595                 ret = nfserr_bad_stateid;
3596         }
3597 out:
3598         nfs4_unlock_state();
3599         return ret;
3600 }
3601
3602 static inline int
3603 setlkflg (int type)
3604 {
3605         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
3606                 RD_STATE : WR_STATE;
3607 }
3608
3609 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
3610 {
3611         struct svc_fh *current_fh = &cstate->current_fh;
3612         struct nfs4_stateowner *sop = stp->st_stateowner;
3613         __be32 status;
3614
3615         status = nfsd4_check_seqid(cstate, sop, seqid);
3616         if (status)
3617                 return status;
3618         if (stp->st_stid.sc_type == NFS4_CLOSED_STID)
3619                 /*
3620                  * "Closed" stateid's exist *only* to return
3621                  * nfserr_replay_me from the previous step.
3622                  */
3623                 return nfserr_bad_stateid;
3624         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
3625         if (status)
3626                 return status;
3627         return nfs4_check_fh(current_fh, stp);
3628 }
3629
3630 /* 
3631  * Checks for sequence id mutating operations. 
3632  */
3633 static __be32
3634 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3635                          stateid_t *stateid, char typemask,
3636                          struct nfs4_ol_stateid **stpp,
3637                          struct nfsd_net *nn)
3638 {
3639         __be32 status;
3640         struct nfs4_stid *s;
3641
3642         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
3643                 seqid, STATEID_VAL(stateid));
3644
3645         *stpp = NULL;
3646         status = nfsd4_lookup_stateid(stateid, typemask, &s,
3647                                       cstate->minorversion, nn);
3648         if (status)
3649                 return status;
3650         *stpp = openlockstateid(s);
3651         cstate->replay_owner = (*stpp)->st_stateowner;
3652
3653         return nfs4_seqid_op_checks(cstate, stateid, seqid, *stpp);
3654 }
3655
3656 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
3657                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
3658 {
3659         __be32 status;
3660         struct nfs4_openowner *oo;
3661
3662         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
3663                                                 NFS4_OPEN_STID, stpp, nn);
3664         if (status)
3665                 return status;
3666         oo = openowner((*stpp)->st_stateowner);
3667         if (!(oo->oo_flags & NFS4_OO_CONFIRMED))
3668                 return nfserr_bad_stateid;
3669         return nfs_ok;
3670 }
3671
3672 __be32
3673 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3674                    struct nfsd4_open_confirm *oc)
3675 {
3676         __be32 status;
3677         struct nfs4_openowner *oo;
3678         struct nfs4_ol_stateid *stp;
3679         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3680
3681         dprintk("NFSD: nfsd4_open_confirm on file %.*s\n",
3682                         (int)cstate->current_fh.fh_dentry->d_name.len,
3683                         cstate->current_fh.fh_dentry->d_name.name);
3684
3685         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
3686         if (status)
3687                 return status;
3688
3689         nfs4_lock_state();
3690
3691         status = nfs4_preprocess_seqid_op(cstate,
3692                                         oc->oc_seqid, &oc->oc_req_stateid,
3693                                         NFS4_OPEN_STID, &stp, nn);
3694         if (status)
3695                 goto out;
3696         oo = openowner(stp->st_stateowner);
3697         status = nfserr_bad_stateid;
3698         if (oo->oo_flags & NFS4_OO_CONFIRMED)
3699                 goto out;
3700         oo->oo_flags |= NFS4_OO_CONFIRMED;
3701         update_stateid(&stp->st_stid.sc_stateid);
3702         memcpy(&oc->oc_resp_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3703         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
3704                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
3705
3706         nfsd4_client_record_create(oo->oo_owner.so_client);
3707         status = nfs_ok;
3708 out:
3709         if (!cstate->replay_owner)
3710                 nfs4_unlock_state();
3711         return status;
3712 }
3713
3714 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
3715 {
3716         if (!test_access(access, stp))
3717                 return;
3718         nfs4_file_put_access(stp->st_file, nfs4_access_to_omode(access));
3719         clear_access(access, stp);
3720 }
3721
3722 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
3723 {
3724         switch (to_access) {
3725         case NFS4_SHARE_ACCESS_READ:
3726                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
3727                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3728                 break;
3729         case NFS4_SHARE_ACCESS_WRITE:
3730                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
3731                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
3732                 break;
3733         case NFS4_SHARE_ACCESS_BOTH:
3734                 break;
3735         default:
3736                 WARN_ON_ONCE(1);
3737         }
3738 }
3739
3740 static void
3741 reset_union_bmap_deny(unsigned long deny, struct nfs4_ol_stateid *stp)
3742 {
3743         int i;
3744         for (i = 0; i < 4; i++) {
3745                 if ((i & deny) != i)
3746                         clear_deny(i, stp);
3747         }
3748 }
3749
3750 __be32
3751 nfsd4_open_downgrade(struct svc_rqst *rqstp,
3752                      struct nfsd4_compound_state *cstate,
3753                      struct nfsd4_open_downgrade *od)
3754 {
3755         __be32 status;
3756         struct nfs4_ol_stateid *stp;
3757         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3758
3759         dprintk("NFSD: nfsd4_open_downgrade on file %.*s\n", 
3760                         (int)cstate->current_fh.fh_dentry->d_name.len,
3761                         cstate->current_fh.fh_dentry->d_name.name);
3762
3763         /* We don't yet support WANT bits: */
3764         if (od->od_deleg_want)
3765                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
3766                         od->od_deleg_want);
3767
3768         nfs4_lock_state();
3769         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
3770                                         &od->od_stateid, &stp, nn);
3771         if (status)
3772                 goto out; 
3773         status = nfserr_inval;
3774         if (!test_access(od->od_share_access, stp)) {
3775                 dprintk("NFSD: access not a subset current bitmap: 0x%lx, input access=%08x\n",
3776                         stp->st_access_bmap, od->od_share_access);
3777                 goto out;
3778         }
3779         if (!test_deny(od->od_share_deny, stp)) {
3780                 dprintk("NFSD:deny not a subset current bitmap: 0x%lx, input deny=%08x\n",
3781                         stp->st_deny_bmap, od->od_share_deny);
3782                 goto out;
3783         }
3784         nfs4_stateid_downgrade(stp, od->od_share_access);
3785
3786         reset_union_bmap_deny(od->od_share_deny, stp);
3787
3788         update_stateid(&stp->st_stid.sc_stateid);
3789         memcpy(&od->od_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3790         status = nfs_ok;
3791 out:
3792         if (!cstate->replay_owner)
3793                 nfs4_unlock_state();
3794         return status;
3795 }
3796
3797 void nfsd4_purge_closed_stateid(struct nfs4_stateowner *so)
3798 {
3799         struct nfs4_openowner *oo;
3800         struct nfs4_ol_stateid *s;
3801
3802         if (!so->so_is_open_owner)
3803                 return;
3804         oo = openowner(so);
3805         s = oo->oo_last_closed_stid;
3806         if (!s)
3807                 return;
3808         if (!(oo->oo_flags & NFS4_OO_PURGE_CLOSE)) {
3809                 /* Release the last_closed_stid on the next seqid bump: */
3810                 oo->oo_flags |= NFS4_OO_PURGE_CLOSE;
3811                 return;
3812         }
3813         oo->oo_flags &= ~NFS4_OO_PURGE_CLOSE;
3814         release_last_closed_stateid(oo);
3815 }
3816
3817 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
3818 {
3819         unhash_open_stateid(s);
3820         s->st_stid.sc_type = NFS4_CLOSED_STID;
3821 }
3822
3823 /*
3824  * nfs4_unlock_state() called after encode
3825  */
3826 __be32
3827 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3828             struct nfsd4_close *close)
3829 {
3830         __be32 status;
3831         struct nfs4_openowner *oo;
3832         struct nfs4_ol_stateid *stp;
3833         struct net *net = SVC_NET(rqstp);
3834         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3835
3836         dprintk("NFSD: nfsd4_close on file %.*s\n", 
3837                         (int)cstate->current_fh.fh_dentry->d_name.len,
3838                         cstate->current_fh.fh_dentry->d_name.name);
3839
3840         nfs4_lock_state();
3841         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
3842                                         &close->cl_stateid,
3843                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
3844                                         &stp, nn);
3845         if (status)
3846                 goto out; 
3847         oo = openowner(stp->st_stateowner);
3848         status = nfs_ok;
3849         update_stateid(&stp->st_stid.sc_stateid);
3850         memcpy(&close->cl_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3851
3852         nfsd4_close_open_stateid(stp);
3853         release_last_closed_stateid(oo);
3854         oo->oo_last_closed_stid = stp;
3855
3856         if (list_empty(&oo->oo_owner.so_stateids)) {
3857                 if (cstate->minorversion) {
3858                         release_openowner(oo);
3859                         cstate->replay_owner = NULL;
3860                 } else {
3861                         /*
3862                          * In the 4.0 case we need to keep the owners around a
3863                          * little while to handle CLOSE replay.
3864                          */
3865                         if (list_empty(&oo->oo_owner.so_stateids))
3866                                 move_to_close_lru(oo, SVC_NET(rqstp));
3867                 }
3868         }
3869 out:
3870         if (!cstate->replay_owner)
3871                 nfs4_unlock_state();
3872         return status;
3873 }
3874
3875 __be32
3876 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3877                   struct nfsd4_delegreturn *dr)
3878 {
3879         struct nfs4_delegation *dp;
3880         stateid_t *stateid = &dr->dr_stateid;
3881         struct nfs4_stid *s;
3882         __be32 status;
3883         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3884
3885         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
3886                 return status;
3887
3888         nfs4_lock_state();
3889         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID, &s,
3890                                       cstate->minorversion, nn);
3891         if (status)
3892                 goto out;
3893         dp = delegstateid(s);
3894         status = check_stateid_generation(stateid, &dp->dl_stid.sc_stateid, nfsd4_has_session(cstate));
3895         if (status)
3896                 goto out;
3897
3898         unhash_delegation(dp);
3899 out:
3900         nfs4_unlock_state();
3901
3902         return status;
3903 }
3904
3905
3906 #define LOFF_OVERFLOW(start, len)      ((u64)(len) > ~(u64)(start))
3907
3908 #define LOCKOWNER_INO_HASH_MASK (LOCKOWNER_INO_HASH_SIZE - 1)
3909
3910 static inline u64
3911 end_offset(u64 start, u64 len)
3912 {
3913         u64 end;
3914
3915         end = start + len;
3916         return end >= start ? end: NFS4_MAX_UINT64;
3917 }
3918
3919 /* last octet in a range */
3920 static inline u64
3921 last_byte_offset(u64 start, u64 len)
3922 {
3923         u64 end;
3924
3925         WARN_ON_ONCE(!len);
3926         end = start + len;
3927         return end > start ? end - 1: NFS4_MAX_UINT64;
3928 }
3929
3930 static unsigned int lockowner_ino_hashval(struct inode *inode, u32 cl_id, struct xdr_netobj *ownername)
3931 {
3932         return (file_hashval(inode) + cl_id
3933                         + opaque_hashval(ownername->data, ownername->len))
3934                 & LOCKOWNER_INO_HASH_MASK;
3935 }
3936
3937 /*
3938  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
3939  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
3940  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
3941  * locking, this prevents us from being completely protocol-compliant.  The
3942  * real solution to this problem is to start using unsigned file offsets in
3943  * the VFS, but this is a very deep change!
3944  */
3945 static inline void
3946 nfs4_transform_lock_offset(struct file_lock *lock)
3947 {
3948         if (lock->fl_start < 0)
3949                 lock->fl_start = OFFSET_MAX;
3950         if (lock->fl_end < 0)
3951                 lock->fl_end = OFFSET_MAX;
3952 }
3953
3954 /* Hack!: For now, we're defining this just so we can use a pointer to it
3955  * as a unique cookie to identify our (NFSv4's) posix locks. */
3956 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
3957 };
3958
3959 static inline void
3960 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
3961 {
3962         struct nfs4_lockowner *lo;
3963
3964         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
3965                 lo = (struct nfs4_lockowner *) fl->fl_owner;
3966                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
3967                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
3968                 if (!deny->ld_owner.data)
3969                         /* We just don't care that much */
3970                         goto nevermind;
3971                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
3972                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
3973         } else {
3974 nevermind:
3975                 deny->ld_owner.len = 0;
3976                 deny->ld_owner.data = NULL;
3977                 deny->ld_clientid.cl_boot = 0;
3978                 deny->ld_clientid.cl_id = 0;
3979         }
3980         deny->ld_start = fl->fl_start;
3981         deny->ld_length = NFS4_MAX_UINT64;
3982         if (fl->fl_end != NFS4_MAX_UINT64)
3983                 deny->ld_length = fl->fl_end - fl->fl_start + 1;        
3984         deny->ld_type = NFS4_READ_LT;
3985         if (fl->fl_type != F_RDLCK)
3986                 deny->ld_type = NFS4_WRITE_LT;
3987 }
3988
3989 static bool same_lockowner_ino(struct nfs4_lockowner *lo, struct inode *inode, clientid_t *clid, struct xdr_netobj *owner)
3990 {
3991         struct nfs4_ol_stateid *lst;
3992
3993         if (!same_owner_str(&lo->lo_owner, owner, clid))
3994                 return false;
3995         lst = list_first_entry(&lo->lo_owner.so_stateids,
3996                                struct nfs4_ol_stateid, st_perstateowner);
3997         return lst->st_file->fi_inode == inode;
3998 }
3999
4000 static struct nfs4_lockowner *
4001 find_lockowner_str(struct inode *inode, clientid_t *clid,
4002                    struct xdr_netobj *owner, struct nfsd_net *nn)
4003 {
4004         unsigned int hashval = lockowner_ino_hashval(inode, clid->cl_id, owner);
4005         struct nfs4_lockowner *lo;
4006
4007         list_for_each_entry(lo, &nn->lockowner_ino_hashtbl[hashval], lo_owner_ino_hash) {
4008                 if (same_lockowner_ino(lo, inode, clid, owner))
4009                         return lo;
4010         }
4011         return NULL;
4012 }
4013
4014 static void hash_lockowner(struct nfs4_lockowner *lo, unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp)
4015 {
4016         struct inode *inode = open_stp->st_file->fi_inode;
4017         unsigned int inohash = lockowner_ino_hashval(inode,
4018                         clp->cl_clientid.cl_id, &lo->lo_owner.so_owner);
4019         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
4020
4021         list_add(&lo->lo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
4022         list_add(&lo->lo_owner_ino_hash, &nn->lockowner_ino_hashtbl[inohash]);
4023         list_add(&lo->lo_perstateid, &open_stp->st_lockowners);
4024 }
4025
4026 /*
4027  * Alloc a lock owner structure.
4028  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has 
4029  * occurred. 
4030  *
4031  * strhashval = ownerstr_hashval
4032  */
4033
4034 static struct nfs4_lockowner *
4035 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfs4_ol_stateid *open_stp, struct nfsd4_lock *lock) {
4036         struct nfs4_lockowner *lo;
4037
4038         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
4039         if (!lo)
4040                 return NULL;
4041         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
4042         lo->lo_owner.so_is_open_owner = 0;
4043         /* It is the openowner seqid that will be incremented in encode in the
4044          * case of new lockowners; so increment the lock seqid manually: */
4045         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid + 1;
4046         hash_lockowner(lo, strhashval, clp, open_stp);
4047         return lo;
4048 }
4049
4050 static struct nfs4_ol_stateid *
4051 alloc_init_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp, struct nfs4_ol_stateid *open_stp)
4052 {
4053         struct nfs4_ol_stateid *stp;
4054         struct nfs4_client *clp = lo->lo_owner.so_client;
4055
4056         stp = nfs4_alloc_stateid(clp);
4057         if (stp == NULL)
4058                 return NULL;
4059         stp->st_stid.sc_type = NFS4_LOCK_STID;
4060         list_add(&stp->st_perfile, &fp->fi_stateids);
4061         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
4062         stp->st_stateowner = &lo->lo_owner;
4063         get_nfs4_file(fp);
4064         stp->st_file = fp;
4065         stp->st_access_bmap = 0;
4066         stp->st_deny_bmap = open_stp->st_deny_bmap;
4067         stp->st_openstp = open_stp;
4068         return stp;
4069 }
4070
4071 static int
4072 check_lock_length(u64 offset, u64 length)
4073 {
4074         return ((length == 0)  || ((length != NFS4_MAX_UINT64) &&
4075              LOFF_OVERFLOW(offset, length)));
4076 }
4077
4078 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
4079 {
4080         struct nfs4_file *fp = lock_stp->st_file;
4081         int oflag = nfs4_access_to_omode(access);
4082
4083         if (test_access(access, lock_stp))
4084                 return;
4085         nfs4_file_get_access(fp, oflag);
4086         set_access(access, lock_stp);
4087 }
4088
4089 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)
4090 {
4091         struct nfs4_file *fi = ost->st_file;
4092         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
4093         struct nfs4_client *cl = oo->oo_owner.so_client;
4094         struct nfs4_lockowner *lo;
4095         unsigned int strhashval;
4096         struct nfsd_net *nn = net_generic(cl->net, nfsd_net_id);
4097
4098         lo = find_lockowner_str(fi->fi_inode, &cl->cl_clientid,
4099                                 &lock->v.new.owner, nn);
4100         if (lo) {
4101                 if (!cstate->minorversion)
4102                         return nfserr_bad_seqid;
4103                 /* XXX: a lockowner always has exactly one stateid: */
4104                 *lst = list_first_entry(&lo->lo_owner.so_stateids,
4105                                 struct nfs4_ol_stateid, st_perstateowner);
4106                 return nfs_ok;
4107         }
4108         strhashval = ownerstr_hashval(cl->cl_clientid.cl_id,
4109                         &lock->v.new.owner);
4110         lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
4111         if (lo == NULL)
4112                 return nfserr_jukebox;
4113         *lst = alloc_init_lock_stateid(lo, fi, ost);
4114         if (*lst == NULL) {
4115                 release_lockowner(lo);
4116                 return nfserr_jukebox;
4117         }
4118         *new = true;
4119         return nfs_ok;
4120 }
4121
4122 /*
4123  *  LOCK operation 
4124  */
4125 __be32
4126 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4127            struct nfsd4_lock *lock)
4128 {
4129         struct nfs4_openowner *open_sop = NULL;
4130         struct nfs4_lockowner *lock_sop = NULL;
4131         struct nfs4_ol_stateid *lock_stp;
4132         struct file *filp = NULL;
4133         struct file_lock *file_lock = NULL;
4134         struct file_lock *conflock = NULL;
4135         __be32 status = 0;
4136         bool new_state = false;
4137         int lkflg;
4138         int err;
4139         struct net *net = SVC_NET(rqstp);
4140         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4141
4142         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
4143                 (long long) lock->lk_offset,
4144                 (long long) lock->lk_length);
4145
4146         if (check_lock_length(lock->lk_offset, lock->lk_length))
4147                  return nfserr_inval;
4148
4149         if ((status = fh_verify(rqstp, &cstate->current_fh,
4150                                 S_IFREG, NFSD_MAY_LOCK))) {
4151                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
4152                 return status;
4153         }
4154
4155         nfs4_lock_state();
4156
4157         if (lock->lk_is_new) {
4158                 struct nfs4_ol_stateid *open_stp = NULL;
4159
4160                 if (nfsd4_has_session(cstate))
4161                         /* See rfc 5661 18.10.3: given clientid is ignored: */
4162                         memcpy(&lock->v.new.clientid,
4163                                 &cstate->session->se_client->cl_clientid,
4164                                 sizeof(clientid_t));
4165
4166                 status = nfserr_stale_clientid;
4167                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
4168                         goto out;
4169
4170                 /* validate and update open stateid and open seqid */
4171                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
4172                                         lock->lk_new_open_seqid,
4173                                         &lock->lk_new_open_stateid,
4174                                         &open_stp, nn);
4175                 if (status)
4176                         goto out;
4177                 open_sop = openowner(open_stp->st_stateowner);
4178                 status = nfserr_bad_stateid;
4179                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
4180                                                 &lock->v.new.clientid))
4181                         goto out;
4182                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
4183                                                         &lock_stp, &new_state);
4184         } else
4185                 status = nfs4_preprocess_seqid_op(cstate,
4186                                        lock->lk_old_lock_seqid,
4187                                        &lock->lk_old_lock_stateid,
4188                                        NFS4_LOCK_STID, &lock_stp, nn);
4189         if (status)
4190                 goto out;
4191         lock_sop = lockowner(lock_stp->st_stateowner);
4192
4193         lkflg = setlkflg(lock->lk_type);
4194         status = nfs4_check_openmode(lock_stp, lkflg);
4195         if (status)
4196                 goto out;
4197
4198         status = nfserr_grace;
4199         if (locks_in_grace(net) && !lock->lk_reclaim)
4200                 goto out;
4201         status = nfserr_no_grace;
4202         if (!locks_in_grace(net) && lock->lk_reclaim)
4203                 goto out;
4204
4205         file_lock = locks_alloc_lock();
4206         if (!file_lock) {
4207                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4208                 status = nfserr_jukebox;
4209                 goto out;
4210         }
4211
4212         locks_init_lock(file_lock);
4213         switch (lock->lk_type) {
4214                 case NFS4_READ_LT:
4215                 case NFS4_READW_LT:
4216                         filp = find_readable_file(lock_stp->st_file);
4217                         if (filp)
4218                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
4219                         file_lock->fl_type = F_RDLCK;
4220                         break;
4221                 case NFS4_WRITE_LT:
4222                 case NFS4_WRITEW_LT:
4223                         filp = find_writeable_file(lock_stp->st_file);
4224                         if (filp)
4225                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
4226                         file_lock->fl_type = F_WRLCK;
4227                         break;
4228                 default:
4229                         status = nfserr_inval;
4230                 goto out;
4231         }
4232         if (!filp) {
4233                 status = nfserr_openmode;
4234                 goto out;
4235         }
4236         file_lock->fl_owner = (fl_owner_t)lock_sop;
4237         file_lock->fl_pid = current->tgid;
4238         file_lock->fl_file = filp;
4239         file_lock->fl_flags = FL_POSIX;
4240         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4241         file_lock->fl_start = lock->lk_offset;
4242         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
4243         nfs4_transform_lock_offset(file_lock);
4244
4245         conflock = locks_alloc_lock();
4246         if (!conflock) {
4247                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4248                 status = nfserr_jukebox;
4249                 goto out;
4250         }
4251
4252         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
4253         switch (-err) {
4254         case 0: /* success! */
4255                 update_stateid(&lock_stp->st_stid.sc_stateid);
4256                 memcpy(&lock->lk_resp_stateid, &lock_stp->st_stid.sc_stateid, 
4257                                 sizeof(stateid_t));
4258                 status = 0;
4259                 break;
4260         case (EAGAIN):          /* conflock holds conflicting lock */
4261                 status = nfserr_denied;
4262                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
4263                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
4264                 break;
4265         case (EDEADLK):
4266                 status = nfserr_deadlock;
4267                 break;
4268         default:
4269                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
4270                 status = nfserrno(err);
4271                 break;
4272         }
4273 out:
4274         if (status && new_state)
4275                 release_lockowner(lock_sop);
4276         if (!cstate->replay_owner)
4277                 nfs4_unlock_state();
4278         if (file_lock)
4279                 locks_free_lock(file_lock);
4280         if (conflock)
4281                 locks_free_lock(conflock);
4282         return status;
4283 }
4284
4285 /*
4286  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
4287  * so we do a temporary open here just to get an open file to pass to
4288  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
4289  * inode operation.)
4290  */
4291 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
4292 {
4293         struct file *file;
4294         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
4295         if (!err) {
4296                 err = nfserrno(vfs_test_lock(file, lock));
4297                 nfsd_close(file);
4298         }
4299         return err;
4300 }
4301
4302 /*
4303  * LOCKT operation
4304  */
4305 __be32
4306 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4307             struct nfsd4_lockt *lockt)
4308 {
4309         struct inode *inode;
4310         struct file_lock *file_lock = NULL;
4311         struct nfs4_lockowner *lo;
4312         __be32 status;
4313         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4314
4315         if (locks_in_grace(SVC_NET(rqstp)))
4316                 return nfserr_grace;
4317
4318         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
4319                  return nfserr_inval;
4320
4321         nfs4_lock_state();
4322
4323         if (!nfsd4_has_session(cstate)) {
4324                 status = lookup_clientid(&lockt->lt_clientid, false, nn, NULL);
4325                 if (status)
4326                         goto out;
4327         }
4328
4329         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
4330                 goto out;
4331
4332         inode = cstate->current_fh.fh_dentry->d_inode;
4333         file_lock = locks_alloc_lock();
4334         if (!file_lock) {
4335                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4336                 status = nfserr_jukebox;
4337                 goto out;
4338         }
4339         locks_init_lock(file_lock);
4340         switch (lockt->lt_type) {
4341                 case NFS4_READ_LT:
4342                 case NFS4_READW_LT:
4343                         file_lock->fl_type = F_RDLCK;
4344                 break;
4345                 case NFS4_WRITE_LT:
4346                 case NFS4_WRITEW_LT:
4347                         file_lock->fl_type = F_WRLCK;
4348                 break;
4349                 default:
4350                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
4351                         status = nfserr_inval;
4352                 goto out;
4353         }
4354
4355         lo = find_lockowner_str(inode, &lockt->lt_clientid, &lockt->lt_owner, nn);
4356         if (lo)
4357                 file_lock->fl_owner = (fl_owner_t)lo;
4358         file_lock->fl_pid = current->tgid;
4359         file_lock->fl_flags = FL_POSIX;
4360
4361         file_lock->fl_start = lockt->lt_offset;
4362         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
4363
4364         nfs4_transform_lock_offset(file_lock);
4365
4366         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
4367         if (status)
4368                 goto out;
4369
4370         if (file_lock->fl_type != F_UNLCK) {
4371                 status = nfserr_denied;
4372                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
4373         }
4374 out:
4375         nfs4_unlock_state();
4376         if (file_lock)
4377                 locks_free_lock(file_lock);
4378         return status;
4379 }
4380
4381 __be32
4382 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4383             struct nfsd4_locku *locku)
4384 {
4385         struct nfs4_ol_stateid *stp;
4386         struct file *filp = NULL;
4387         struct file_lock *file_lock = NULL;
4388         __be32 status;
4389         int err;
4390         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4391
4392         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
4393                 (long long) locku->lu_offset,
4394                 (long long) locku->lu_length);
4395
4396         if (check_lock_length(locku->lu_offset, locku->lu_length))
4397                  return nfserr_inval;
4398
4399         nfs4_lock_state();
4400                                                                                 
4401         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
4402                                         &locku->lu_stateid, NFS4_LOCK_STID,
4403                                         &stp, nn);
4404         if (status)
4405                 goto out;
4406         filp = find_any_file(stp->st_file);
4407         if (!filp) {
4408                 status = nfserr_lock_range;
4409                 goto out;
4410         }
4411         file_lock = locks_alloc_lock();
4412         if (!file_lock) {
4413                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
4414                 status = nfserr_jukebox;
4415                 goto out;
4416         }
4417         locks_init_lock(file_lock);
4418         file_lock->fl_type = F_UNLCK;
4419         file_lock->fl_owner = (fl_owner_t)lockowner(stp->st_stateowner);
4420         file_lock->fl_pid = current->tgid;
4421         file_lock->fl_file = filp;
4422         file_lock->fl_flags = FL_POSIX;
4423         file_lock->fl_lmops = &nfsd_posix_mng_ops;
4424         file_lock->fl_start = locku->lu_offset;
4425
4426         file_lock->fl_end = last_byte_offset(locku->lu_offset,
4427                                                 locku->lu_length);
4428         nfs4_transform_lock_offset(file_lock);
4429
4430         /*
4431         *  Try to unlock the file in the VFS.
4432         */
4433         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
4434         if (err) {
4435                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
4436                 goto out_nfserr;
4437         }
4438         /*
4439         * OK, unlock succeeded; the only thing left to do is update the stateid.
4440         */
4441         update_stateid(&stp->st_stid.sc_stateid);
4442         memcpy(&locku->lu_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
4443
4444 out:
4445         if (!cstate->replay_owner)
4446                 nfs4_unlock_state();
4447         if (file_lock)
4448                 locks_free_lock(file_lock);
4449         return status;
4450
4451 out_nfserr:
4452         status = nfserrno(err);
4453         goto out;
4454 }
4455
4456 /*
4457  * returns
4458  *      1: locks held by lockowner
4459  *      0: no locks held by lockowner
4460  */
4461 static int
4462 check_for_locks(struct nfs4_file *filp, struct nfs4_lockowner *lowner)
4463 {
4464         struct file_lock **flpp;
4465         struct inode *inode = filp->fi_inode;
4466         int status = 0;
4467
4468         lock_flocks();
4469         for (flpp = &inode->i_flock; *flpp != NULL; flpp = &(*flpp)->fl_next) {
4470                 if ((*flpp)->fl_owner == (fl_owner_t)lowner) {
4471                         status = 1;
4472                         goto out;
4473                 }
4474         }
4475 out:
4476         unlock_flocks();
4477         return status;
4478 }
4479
4480 __be32
4481 nfsd4_release_lockowner(struct svc_rqst *rqstp,
4482                         struct nfsd4_compound_state *cstate,
4483                         struct nfsd4_release_lockowner *rlockowner)
4484 {
4485         clientid_t *clid = &rlockowner->rl_clientid;
4486         struct nfs4_stateowner *sop;
4487         struct nfs4_lockowner *lo;
4488         struct nfs4_ol_stateid *stp;
4489         struct xdr_netobj *owner = &rlockowner->rl_owner;
4490         struct list_head matches;
4491         unsigned int hashval = ownerstr_hashval(clid->cl_id, owner);
4492         __be32 status;
4493         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4494
4495         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
4496                 clid->cl_boot, clid->cl_id);
4497
4498         nfs4_lock_state();
4499
4500         status = lookup_clientid(clid, cstate->minorversion, nn, NULL);
4501         if (status)
4502                 goto out;
4503
4504         status = nfserr_locks_held;
4505         INIT_LIST_HEAD(&matches);
4506
4507         list_for_each_entry(sop, &nn->ownerstr_hashtbl[hashval], so_strhash) {
4508                 if (sop->so_is_open_owner)
4509                         continue;
4510                 if (!same_owner_str(sop, owner, clid))
4511                         continue;
4512                 list_for_each_entry(stp, &sop->so_stateids,
4513                                 st_perstateowner) {
4514                         lo = lockowner(sop);
4515                         if (check_for_locks(stp->st_file, lo))
4516                                 goto out;
4517                         list_add(&lo->lo_list, &matches);
4518                 }
4519         }
4520         /* Clients probably won't expect us to return with some (but not all)
4521          * of the lockowner state released; so don't release any until all
4522          * have been checked. */
4523         status = nfs_ok;
4524         while (!list_empty(&matches)) {
4525                 lo = list_entry(matches.next, struct nfs4_lockowner,
4526                                                                 lo_list);
4527                 /* unhash_stateowner deletes so_perclient only
4528                  * for openowners. */
4529                 list_del(&lo->lo_list);
4530                 release_lockowner(lo);
4531         }
4532 out:
4533         nfs4_unlock_state();
4534         return status;
4535 }
4536
4537 static inline struct nfs4_client_reclaim *
4538 alloc_reclaim(void)
4539 {
4540         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
4541 }
4542
4543 bool
4544 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
4545 {
4546         struct nfs4_client_reclaim *crp;
4547
4548         crp = nfsd4_find_reclaim_client(name, nn);
4549         return (crp && crp->cr_clp);
4550 }
4551
4552 /*
4553  * failure => all reset bets are off, nfserr_no_grace...
4554  */
4555 struct nfs4_client_reclaim *
4556 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
4557 {
4558         unsigned int strhashval;
4559         struct nfs4_client_reclaim *crp;
4560
4561         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
4562         crp = alloc_reclaim();
4563         if (crp) {
4564                 strhashval = clientstr_hashval(name);
4565                 INIT_LIST_HEAD(&crp->cr_strhash);
4566                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
4567                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
4568                 crp->cr_clp = NULL;
4569                 nn->reclaim_str_hashtbl_size++;
4570         }
4571         return crp;
4572 }
4573
4574 void
4575 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
4576 {
4577         list_del(&crp->cr_strhash);
4578         kfree(crp);
4579         nn->reclaim_str_hashtbl_size--;
4580 }
4581
4582 void
4583 nfs4_release_reclaim(struct nfsd_net *nn)
4584 {
4585         struct nfs4_client_reclaim *crp = NULL;
4586         int i;
4587
4588         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4589                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
4590                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
4591                                         struct nfs4_client_reclaim, cr_strhash);
4592                         nfs4_remove_reclaim_record(crp, nn);
4593                 }
4594         }
4595         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
4596 }
4597
4598 /*
4599  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
4600 struct nfs4_client_reclaim *
4601 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
4602 {
4603         unsigned int strhashval;
4604         struct nfs4_client_reclaim *crp = NULL;
4605
4606         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
4607
4608         strhashval = clientstr_hashval(recdir);
4609         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
4610                 if (same_name(crp->cr_recdir, recdir)) {
4611                         return crp;
4612                 }
4613         }
4614         return NULL;
4615 }
4616
4617 /*
4618 * Called from OPEN. Look for clientid in reclaim list.
4619 */
4620 __be32
4621 nfs4_check_open_reclaim(clientid_t *clid, bool sessions, struct nfsd_net *nn)
4622 {
4623         struct nfs4_client *clp;
4624
4625         /* find clientid in conf_id_hashtbl */
4626         clp = find_confirmed_client(clid, sessions, nn);
4627         if (clp == NULL)
4628                 return nfserr_reclaim_bad;
4629
4630         return nfsd4_client_record_check(clp) ? nfserr_reclaim_bad : nfs_ok;
4631 }
4632
4633 #ifdef CONFIG_NFSD_FAULT_INJECTION
4634
4635 u64 nfsd_forget_client(struct nfs4_client *clp, u64 max)
4636 {
4637         expire_client(clp);
4638         return 1;
4639 }
4640
4641 u64 nfsd_print_client(struct nfs4_client *clp, u64 num)
4642 {
4643         char buf[INET6_ADDRSTRLEN];
4644         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4645         printk(KERN_INFO "NFS Client: %s\n", buf);
4646         return 1;
4647 }
4648
4649 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
4650                              const char *type)
4651 {
4652         char buf[INET6_ADDRSTRLEN];
4653         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
4654         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
4655 }
4656
4657 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_lockowner *))
4658 {
4659         struct nfs4_openowner *oop;
4660         struct nfs4_lockowner *lop, *lo_next;
4661         struct nfs4_ol_stateid *stp, *st_next;
4662         u64 count = 0;
4663
4664         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
4665                 list_for_each_entry_safe(stp, st_next, &oop->oo_owner.so_stateids, st_perstateowner) {
4666                         list_for_each_entry_safe(lop, lo_next, &stp->st_lockowners, lo_perstateid) {
4667                                 if (func)
4668                                         func(lop);
4669                                 if (++count == max)
4670                                         return count;
4671                         }
4672                 }
4673         }
4674
4675         return count;
4676 }
4677
4678 u64 nfsd_forget_client_locks(struct nfs4_client *clp, u64 max)
4679 {
4680         return nfsd_foreach_client_lock(clp, max, release_lockowner);
4681 }
4682
4683 u64 nfsd_print_client_locks(struct nfs4_client *clp, u64 max)
4684 {
4685         u64 count = nfsd_foreach_client_lock(clp, max, NULL);
4686         nfsd_print_count(clp, count, "locked files");
4687         return count;
4688 }
4689
4690 static u64 nfsd_foreach_client_open(struct nfs4_client *clp, u64 max, void (*func)(struct nfs4_openowner *))
4691 {
4692         struct nfs4_openowner *oop, *next;
4693         u64 count = 0;
4694
4695         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
4696                 if (func)
4697                         func(oop);
4698                 if (++count == max)
4699                         break;
4700         }
4701
4702         return count;
4703 }
4704
4705 u64 nfsd_forget_client_openowners(struct nfs4_client *clp, u64 max)
4706 {
4707         return nfsd_foreach_client_open(clp, max, release_openowner);
4708 }
4709
4710 u64 nfsd_print_client_openowners(struct nfs4_client *clp, u64 max)
4711 {
4712         u64 count = nfsd_foreach_client_open(clp, max, NULL);
4713         nfsd_print_count(clp, count, "open files");
4714         return count;
4715 }
4716
4717 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
4718                                      struct list_head *victims)
4719 {
4720         struct nfs4_delegation *dp, *next;
4721         u64 count = 0;
4722
4723         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
4724                 if (victims)
4725                         list_move(&dp->dl_recall_lru, victims);
4726                 if (++count == max)
4727                         break;
4728         }
4729         return count;
4730 }
4731
4732 u64 nfsd_forget_client_delegations(struct nfs4_client *clp, u64 max)
4733 {
4734         struct nfs4_delegation *dp, *next;
4735         LIST_HEAD(victims);
4736         u64 count;
4737
4738         spin_lock(&recall_lock);
4739         count = nfsd_find_all_delegations(clp, max, &victims);
4740         spin_unlock(&recall_lock);
4741
4742         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4743                 unhash_delegation(dp);
4744
4745         return count;
4746 }
4747
4748 u64 nfsd_recall_client_delegations(struct nfs4_client *clp, u64 max)
4749 {
4750         struct nfs4_delegation *dp, *next;
4751         LIST_HEAD(victims);
4752         u64 count;
4753
4754         spin_lock(&recall_lock);
4755         count = nfsd_find_all_delegations(clp, max, &victims);
4756         list_for_each_entry_safe(dp, next, &victims, dl_recall_lru)
4757                 nfsd_break_one_deleg(dp);
4758         spin_unlock(&recall_lock);
4759
4760         return count;
4761 }
4762
4763 u64 nfsd_print_client_delegations(struct nfs4_client *clp, u64 max)
4764 {
4765         u64 count = 0;
4766
4767         spin_lock(&recall_lock);
4768         count = nfsd_find_all_delegations(clp, max, NULL);
4769         spin_unlock(&recall_lock);
4770
4771         nfsd_print_count(clp, count, "delegations");
4772         return count;
4773 }
4774
4775 u64 nfsd_for_n_state(u64 max, u64 (*func)(struct nfs4_client *, u64))
4776 {
4777         struct nfs4_client *clp, *next;
4778         u64 count = 0;
4779         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4780
4781         if (!nfsd_netns_ready(nn))
4782                 return 0;
4783
4784         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
4785                 count += func(clp, max - count);
4786                 if ((max != 0) && (count >= max))
4787                         break;
4788         }
4789
4790         return count;
4791 }
4792
4793 struct nfs4_client *nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
4794 {
4795         struct nfs4_client *clp;
4796         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns, nfsd_net_id);
4797
4798         if (!nfsd_netns_ready(nn))
4799                 return NULL;
4800
4801         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
4802                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
4803                         return clp;
4804         }
4805         return NULL;
4806 }
4807
4808 #endif /* CONFIG_NFSD_FAULT_INJECTION */
4809
4810 /* initialization to perform at module load time: */
4811
4812 void
4813 nfs4_state_init(void)
4814 {
4815         int i;
4816
4817         for (i = 0; i < FILE_HASH_SIZE; i++) {
4818                 INIT_LIST_HEAD(&file_hashtbl[i]);
4819         }
4820         INIT_LIST_HEAD(&del_recall_lru);
4821 }
4822
4823 /*
4824  * Since the lifetime of a delegation isn't limited to that of an open, a
4825  * client may quite reasonably hang on to a delegation as long as it has
4826  * the inode cached.  This becomes an obvious problem the first time a
4827  * client's inode cache approaches the size of the server's total memory.
4828  *
4829  * For now we avoid this problem by imposing a hard limit on the number
4830  * of delegations, which varies according to the server's memory size.
4831  */
4832 static void
4833 set_max_delegations(void)
4834 {
4835         /*
4836          * Allow at most 4 delegations per megabyte of RAM.  Quick
4837          * estimates suggest that in the worst case (where every delegation
4838          * is for a different inode), a delegation could take about 1.5K,
4839          * giving a worst case usage of about 6% of memory.
4840          */
4841         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
4842 }
4843
4844 static int nfs4_state_create_net(struct net *net)
4845 {
4846         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4847         int i;
4848
4849         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4850                         CLIENT_HASH_SIZE, GFP_KERNEL);
4851         if (!nn->conf_id_hashtbl)
4852                 goto err;
4853         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
4854                         CLIENT_HASH_SIZE, GFP_KERNEL);
4855         if (!nn->unconf_id_hashtbl)
4856                 goto err_unconf_id;
4857         nn->ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
4858                         OWNER_HASH_SIZE, GFP_KERNEL);
4859         if (!nn->ownerstr_hashtbl)
4860                 goto err_ownerstr;
4861         nn->lockowner_ino_hashtbl = kmalloc(sizeof(struct list_head) *
4862                         LOCKOWNER_INO_HASH_SIZE, GFP_KERNEL);
4863         if (!nn->lockowner_ino_hashtbl)
4864                 goto err_lockowner_ino;
4865         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
4866                         SESSION_HASH_SIZE, GFP_KERNEL);
4867         if (!nn->sessionid_hashtbl)
4868                 goto err_sessionid;
4869
4870         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4871                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
4872                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
4873         }
4874         for (i = 0; i < OWNER_HASH_SIZE; i++)
4875                 INIT_LIST_HEAD(&nn->ownerstr_hashtbl[i]);
4876         for (i = 0; i < LOCKOWNER_INO_HASH_SIZE; i++)
4877                 INIT_LIST_HEAD(&nn->lockowner_ino_hashtbl[i]);
4878         for (i = 0; i < SESSION_HASH_SIZE; i++)
4879                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
4880         nn->conf_name_tree = RB_ROOT;
4881         nn->unconf_name_tree = RB_ROOT;
4882         INIT_LIST_HEAD(&nn->client_lru);
4883         INIT_LIST_HEAD(&nn->close_lru);
4884         spin_lock_init(&nn->client_lock);
4885
4886         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
4887         get_net(net);
4888
4889         return 0;
4890
4891 err_sessionid:
4892         kfree(nn->lockowner_ino_hashtbl);
4893 err_lockowner_ino:
4894         kfree(nn->ownerstr_hashtbl);
4895 err_ownerstr:
4896         kfree(nn->unconf_id_hashtbl);
4897 err_unconf_id:
4898         kfree(nn->conf_id_hashtbl);
4899 err:
4900         return -ENOMEM;
4901 }
4902
4903 static void
4904 nfs4_state_destroy_net(struct net *net)
4905 {
4906         int i;
4907         struct nfs4_client *clp = NULL;
4908         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4909         struct rb_node *node, *tmp;
4910
4911         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
4912                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
4913                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
4914                         destroy_client(clp);
4915                 }
4916         }
4917
4918         node = rb_first(&nn->unconf_name_tree);
4919         while (node != NULL) {
4920                 tmp = node;
4921                 node = rb_next(tmp);
4922                 clp = rb_entry(tmp, struct nfs4_client, cl_namenode);
4923                 rb_erase(tmp, &nn->unconf_name_tree);
4924                 destroy_client(clp);
4925         }
4926
4927         kfree(nn->sessionid_hashtbl);
4928         kfree(nn->lockowner_ino_hashtbl);
4929         kfree(nn->ownerstr_hashtbl);
4930         kfree(nn->unconf_id_hashtbl);
4931         kfree(nn->conf_id_hashtbl);
4932         put_net(net);
4933 }
4934
4935 int
4936 nfs4_state_start_net(struct net *net)
4937 {
4938         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4939         int ret;
4940
4941         ret = nfs4_state_create_net(net);
4942         if (ret)
4943                 return ret;
4944         nfsd4_client_tracking_init(net);
4945         nn->boot_time = get_seconds();
4946         locks_start_grace(net, &nn->nfsd4_manager);
4947         nn->grace_ended = false;
4948         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %p)\n",
4949                nn->nfsd4_grace, net);
4950         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
4951         return 0;
4952 }
4953
4954 /* initialization to perform when the nfsd service is started: */
4955
4956 int
4957 nfs4_state_start(void)
4958 {
4959         int ret;
4960
4961         ret = set_callback_cred();
4962         if (ret)
4963                 return -ENOMEM;
4964         laundry_wq = create_singlethread_workqueue("nfsd4");
4965         if (laundry_wq == NULL) {
4966                 ret = -ENOMEM;
4967                 goto out_recovery;
4968         }
4969         ret = nfsd4_create_callback_queue();
4970         if (ret)
4971                 goto out_free_laundry;
4972
4973         set_max_delegations();
4974
4975         return 0;
4976
4977 out_free_laundry:
4978         destroy_workqueue(laundry_wq);
4979 out_recovery:
4980         return ret;
4981 }
4982
4983 /* should be called with the state lock held */
4984 void
4985 nfs4_state_shutdown_net(struct net *net)
4986 {
4987         struct nfs4_delegation *dp = NULL;
4988         struct list_head *pos, *next, reaplist;
4989         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
4990
4991         cancel_delayed_work_sync(&nn->laundromat_work);
4992         locks_end_grace(&nn->nfsd4_manager);
4993
4994         INIT_LIST_HEAD(&reaplist);
4995         spin_lock(&recall_lock);
4996         list_for_each_safe(pos, next, &del_recall_lru) {
4997                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4998                 if (dp->dl_stid.sc_client->net != net)
4999                         continue;
5000                 list_move(&dp->dl_recall_lru, &reaplist);
5001         }
5002         spin_unlock(&recall_lock);
5003         list_for_each_safe(pos, next, &reaplist) {
5004                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
5005                 unhash_delegation(dp);
5006         }
5007
5008         nfsd4_client_tracking_exit(net);
5009         nfs4_state_destroy_net(net);
5010 }
5011
5012 void
5013 nfs4_state_shutdown(void)
5014 {
5015         destroy_workqueue(laundry_wq);
5016         nfsd4_destroy_callback_queue();
5017 }
5018
5019 static void
5020 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5021 {
5022         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
5023                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
5024 }
5025
5026 static void
5027 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
5028 {
5029         if (cstate->minorversion) {
5030                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
5031                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5032         }
5033 }
5034
5035 void
5036 clear_current_stateid(struct nfsd4_compound_state *cstate)
5037 {
5038         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
5039 }
5040
5041 /*
5042  * functions to set current state id
5043  */
5044 void
5045 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5046 {
5047         put_stateid(cstate, &odp->od_stateid);
5048 }
5049
5050 void
5051 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate, struct nfsd4_open *open)
5052 {
5053         put_stateid(cstate, &open->op_stateid);
5054 }
5055
5056 void
5057 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5058 {
5059         put_stateid(cstate, &close->cl_stateid);
5060 }
5061
5062 void
5063 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate, struct nfsd4_lock *lock)
5064 {
5065         put_stateid(cstate, &lock->lk_resp_stateid);
5066 }
5067
5068 /*
5069  * functions to consume current state id
5070  */
5071
5072 void
5073 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate, struct nfsd4_open_downgrade *odp)
5074 {
5075         get_stateid(cstate, &odp->od_stateid);
5076 }
5077
5078 void
5079 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate, struct nfsd4_delegreturn *drp)
5080 {
5081         get_stateid(cstate, &drp->dr_stateid);
5082 }
5083
5084 void
5085 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate, struct nfsd4_free_stateid *fsp)
5086 {
5087         get_stateid(cstate, &fsp->fr_stateid);
5088 }
5089
5090 void
5091 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate, struct nfsd4_setattr *setattr)
5092 {
5093         get_stateid(cstate, &setattr->sa_stateid);
5094 }
5095
5096 void
5097 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate, struct nfsd4_close *close)
5098 {
5099         get_stateid(cstate, &close->cl_stateid);
5100 }
5101
5102 void
5103 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate, struct nfsd4_locku *locku)
5104 {
5105         get_stateid(cstate, &locku->lu_stateid);
5106 }
5107
5108 void
5109 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate, struct nfsd4_read *read)
5110 {
5111         get_stateid(cstate, &read->rd_stateid);
5112 }
5113
5114 void
5115 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate, struct nfsd4_write *write)
5116 {
5117         get_stateid(cstate, &write->wr_stateid);
5118 }