Merge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm...
[~shefty/rdma-dev.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/pagemap.h>
41 #include <linux/ratelimit.h>
42 #include <linux/sunrpc/svcauth_gss.h>
43 #include <linux/sunrpc/clnt.h>
44 #include "xdr4.h"
45 #include "vfs.h"
46 #include "current_stateid.h"
47
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 void init_stid(struct nfs4_stid *stid, struct nfs4_client *cl, unsigned char type)
265 {
266         stateid_t *s = &stid->sc_stateid;
267         int new_id;
268
269         stid->sc_type = type;
270         stid->sc_client = cl;
271         s->si_opaque.so_clid = cl->cl_clientid;
272         new_id = get_new_stid(stid);
273         s->si_opaque.so_id = (u32)new_id;
274         /* Will be incremented before return to client: */
275         s->si_generation = 0;
276 }
277
278 static struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab)
279 {
280         struct idr *stateids = &cl->cl_stateids;
281
282         if (!idr_pre_get(stateids, GFP_KERNEL))
283                 return NULL;
284         /*
285          * Note: if we fail here (or any time between now and the time
286          * we actually get the new idr), we won't need to undo the idr
287          * preallocation, since the idr code caps the number of
288          * preallocated entries.
289          */
290         return kmem_cache_alloc(slab, GFP_KERNEL);
291 }
292
293 static struct nfs4_ol_stateid * nfs4_alloc_stateid(struct nfs4_client *clp)
294 {
295         return openlockstateid(nfs4_alloc_stid(clp, stateid_slab));
296 }
297
298 static struct nfs4_delegation *
299 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_ol_stateid *stp, struct svc_fh *current_fh, u32 type)
300 {
301         struct nfs4_delegation *dp;
302         struct nfs4_file *fp = stp->st_file;
303
304         dprintk("NFSD alloc_init_deleg\n");
305         /*
306          * Major work on the lease subsystem (for example, to support
307          * calbacks on stat) will be required before we can support
308          * write delegations properly.
309          */
310         if (type != NFS4_OPEN_DELEGATE_READ)
311                 return NULL;
312         if (fp->fi_had_conflict)
313                 return NULL;
314         if (num_delegations > max_delegations)
315                 return NULL;
316         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab));
317         if (dp == NULL)
318                 return dp;
319         init_stid(&dp->dl_stid, clp, NFS4_DELEG_STID);
320         /*
321          * delegation seqid's are never incremented.  The 4.1 special
322          * meaning of seqid 0 isn't meaningful, really, but let's avoid
323          * 0 anyway just for consistency and use 1:
324          */
325         dp->dl_stid.sc_stateid.si_generation = 1;
326         num_delegations++;
327         INIT_LIST_HEAD(&dp->dl_perfile);
328         INIT_LIST_HEAD(&dp->dl_perclnt);
329         INIT_LIST_HEAD(&dp->dl_recall_lru);
330         get_nfs4_file(fp);
331         dp->dl_file = fp;
332         dp->dl_type = type;
333         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
334         dp->dl_time = 0;
335         atomic_set(&dp->dl_count, 1);
336         nfsd4_init_callback(&dp->dl_recall);
337         return dp;
338 }
339
340 void
341 nfs4_put_delegation(struct nfs4_delegation *dp)
342 {
343         if (atomic_dec_and_test(&dp->dl_count)) {
344                 dprintk("NFSD: freeing dp %p\n",dp);
345                 put_nfs4_file(dp->dl_file);
346                 kmem_cache_free(deleg_slab, dp);
347                 num_delegations--;
348         }
349 }
350
351 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
352 {
353         if (atomic_dec_and_test(&fp->fi_delegees)) {
354                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
355                 fp->fi_lease = NULL;
356                 fput(fp->fi_deleg_file);
357                 fp->fi_deleg_file = NULL;
358         }
359 }
360
361 static void unhash_stid(struct nfs4_stid *s)
362 {
363         struct idr *stateids = &s->sc_client->cl_stateids;
364
365         idr_remove(stateids, s->sc_stateid.si_opaque.so_id);
366 }
367
368 /* Called under the state lock. */
369 static void
370 unhash_delegation(struct nfs4_delegation *dp)
371 {
372         unhash_stid(&dp->dl_stid);
373         list_del_init(&dp->dl_perclnt);
374         spin_lock(&recall_lock);
375         list_del_init(&dp->dl_perfile);
376         list_del_init(&dp->dl_recall_lru);
377         spin_unlock(&recall_lock);
378         nfs4_put_deleg_lease(dp->dl_file);
379         nfs4_put_delegation(dp);
380 }
381
382 /* 
383  * SETCLIENTID state 
384  */
385
386 static unsigned int clientid_hashval(u32 id)
387 {
388         return id & CLIENT_HASH_MASK;
389 }
390
391 static unsigned int clientstr_hashval(const char *name)
392 {
393         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
394 }
395
396 /*
397  * We store the NONE, READ, WRITE, and BOTH bits separately in the
398  * st_{access,deny}_bmap field of the stateid, in order to track not
399  * only what share bits are currently in force, but also what
400  * combinations of share bits previous opens have used.  This allows us
401  * to enforce the recommendation of rfc 3530 14.2.19 that the server
402  * return an error if the client attempt to downgrade to a combination
403  * of share bits not explicable by closing some of its previous opens.
404  *
405  * XXX: This enforcement is actually incomplete, since we don't keep
406  * track of access/deny bit combinations; so, e.g., we allow:
407  *
408  *      OPEN allow read, deny write
409  *      OPEN allow both, deny none
410  *      DOWNGRADE allow read, deny none
411  *
412  * which we should reject.
413  */
414 static unsigned int
415 bmap_to_share_mode(unsigned long bmap) {
416         int i;
417         unsigned int access = 0;
418
419         for (i = 1; i < 4; i++) {
420                 if (test_bit(i, &bmap))
421                         access |= i;
422         }
423         return access;
424 }
425
426 static bool
427 test_share(struct nfs4_ol_stateid *stp, struct nfsd4_open *open) {
428         unsigned int access, deny;
429
430         access = bmap_to_share_mode(stp->st_access_bmap);
431         deny = bmap_to_share_mode(stp->st_deny_bmap);
432         if ((access & open->op_share_deny) || (deny & open->op_share_access))
433                 return false;
434         return true;
435 }
436
437 /* set share access for a given stateid */
438 static inline void
439 set_access(u32 access, struct nfs4_ol_stateid *stp)
440 {
441         __set_bit(access, &stp->st_access_bmap);
442 }
443
444 /* clear share access for a given stateid */
445 static inline void
446 clear_access(u32 access, struct nfs4_ol_stateid *stp)
447 {
448         __clear_bit(access, &stp->st_access_bmap);
449 }
450
451 /* test whether a given stateid has access */
452 static inline bool
453 test_access(u32 access, struct nfs4_ol_stateid *stp)
454 {
455         return test_bit(access, &stp->st_access_bmap);
456 }
457
458 /* set share deny for a given stateid */
459 static inline void
460 set_deny(u32 access, struct nfs4_ol_stateid *stp)
461 {
462         __set_bit(access, &stp->st_deny_bmap);
463 }
464
465 /* clear share deny for a given stateid */
466 static inline void
467 clear_deny(u32 access, struct nfs4_ol_stateid *stp)
468 {
469         __clear_bit(access, &stp->st_deny_bmap);
470 }
471
472 /* test whether a given stateid is denying specific access */
473 static inline bool
474 test_deny(u32 access, struct nfs4_ol_stateid *stp)
475 {
476         return test_bit(access, &stp->st_deny_bmap);
477 }
478
479 static int nfs4_access_to_omode(u32 access)
480 {
481         switch (access & NFS4_SHARE_ACCESS_BOTH) {
482         case NFS4_SHARE_ACCESS_READ:
483                 return O_RDONLY;
484         case NFS4_SHARE_ACCESS_WRITE:
485                 return O_WRONLY;
486         case NFS4_SHARE_ACCESS_BOTH:
487                 return O_RDWR;
488         }
489         WARN_ON_ONCE(1);
490         return O_RDONLY;
491 }
492
493 /* release all access and file references for a given stateid */
494 static void
495 release_all_access(struct nfs4_ol_stateid *stp)
496 {
497         int i;
498
499         for (i = 1; i < 4; i++) {
500                 if (test_access(i, stp))
501                         nfs4_file_put_access(stp->st_file,
502                                              nfs4_access_to_omode(i));
503                 clear_access(i, stp);
504         }
505 }
506
507 static void unhash_generic_stateid(struct nfs4_ol_stateid *stp)
508 {
509         list_del(&stp->st_perfile);
510         list_del(&stp->st_perstateowner);
511 }
512
513 static void close_generic_stateid(struct nfs4_ol_stateid *stp)
514 {
515         release_all_access(stp);
516         put_nfs4_file(stp->st_file);
517         stp->st_file = NULL;
518 }
519
520 static void free_generic_stateid(struct nfs4_ol_stateid *stp)
521 {
522         kmem_cache_free(stateid_slab, stp);
523 }
524
525 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
526 {
527         struct file *file;
528
529         unhash_generic_stateid(stp);
530         unhash_stid(&stp->st_stid);
531         file = find_any_file(stp->st_file);
532         if (file)
533                 locks_remove_posix(file, (fl_owner_t)lockowner(stp->st_stateowner));
534         close_generic_stateid(stp);
535         free_generic_stateid(stp);
536 }
537
538 static void unhash_lockowner(struct nfs4_lockowner *lo)
539 {
540         struct nfs4_ol_stateid *stp;
541
542         list_del(&lo->lo_owner.so_strhash);
543         list_del(&lo->lo_perstateid);
544         list_del(&lo->lo_owner_ino_hash);
545         while (!list_empty(&lo->lo_owner.so_stateids)) {
546                 stp = list_first_entry(&lo->lo_owner.so_stateids,
547                                 struct nfs4_ol_stateid, st_perstateowner);
548                 release_lock_stateid(stp);
549         }
550 }
551
552 static void release_lockowner(struct nfs4_lockowner *lo)
553 {
554         unhash_lockowner(lo);
555         nfs4_free_lockowner(lo);
556 }
557
558 static void
559 release_stateid_lockowners(struct nfs4_ol_stateid *open_stp)
560 {
561         struct nfs4_lockowner *lo;
562
563         while (!list_empty(&open_stp->st_lockowners)) {
564                 lo = list_entry(open_stp->st_lockowners.next,
565                                 struct nfs4_lockowner, lo_perstateid);
566                 release_lockowner(lo);
567         }
568 }
569
570 static void unhash_open_stateid(struct nfs4_ol_stateid *stp)
571 {
572         unhash_generic_stateid(stp);
573         release_stateid_lockowners(stp);
574         close_generic_stateid(stp);
575 }
576
577 static void release_open_stateid(struct nfs4_ol_stateid *stp)
578 {
579         unhash_open_stateid(stp);
580         unhash_stid(&stp->st_stid);
581         free_generic_stateid(stp);
582 }
583
584 static void unhash_openowner(struct nfs4_openowner *oo)
585 {
586         struct nfs4_ol_stateid *stp;
587
588         list_del(&oo->oo_owner.so_strhash);
589         list_del(&oo->oo_perclient);
590         while (!list_empty(&oo->oo_owner.so_stateids)) {
591                 stp = list_first_entry(&oo->oo_owner.so_stateids,
592                                 struct nfs4_ol_stateid, st_perstateowner);
593                 release_open_stateid(stp);
594         }
595 }
596
597 static void release_last_closed_stateid(struct nfs4_openowner *oo)
598 {
599         struct nfs4_ol_stateid *s = oo->oo_last_closed_stid;
600
601         if (s) {
602                 unhash_stid(&s->st_stid);
603                 free_generic_stateid(s);
604                 oo->oo_last_closed_stid = NULL;
605         }
606 }
607
608 static void release_openowner(struct nfs4_openowner *oo)
609 {
610         unhash_openowner(oo);
611         list_del(&oo->oo_close_lru);
612         release_last_closed_stateid(oo);
613         nfs4_free_openowner(oo);
614 }
615
616 static inline int
617 hash_sessionid(struct nfs4_sessionid *sessionid)
618 {
619         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
620
621         return sid->sequence % SESSION_HASH_SIZE;
622 }
623
624 #ifdef NFSD_DEBUG
625 static inline void
626 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
627 {
628         u32 *ptr = (u32 *)(&sessionid->data[0]);
629         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
630 }
631 #else
632 static inline void
633 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
634 {
635 }
636 #endif
637
638
639 static void
640 gen_sessionid(struct nfsd4_session *ses)
641 {
642         struct nfs4_client *clp = ses->se_client;
643         struct nfsd4_sessionid *sid;
644
645         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
646         sid->clientid = clp->cl_clientid;
647         sid->sequence = current_sessionid++;
648         sid->reserved = 0;
649 }
650
651 /*
652  * The protocol defines ca_maxresponssize_cached to include the size of
653  * the rpc header, but all we need to cache is the data starting after
654  * the end of the initial SEQUENCE operation--the rest we regenerate
655  * each time.  Therefore we can advertise a ca_maxresponssize_cached
656  * value that is the number of bytes in our cache plus a few additional
657  * bytes.  In order to stay on the safe side, and not promise more than
658  * we can cache, those additional bytes must be the minimum possible: 24
659  * bytes of rpc header (xid through accept state, with AUTH_NULL
660  * verifier), 12 for the compound header (with zero-length tag), and 44
661  * for the SEQUENCE op response:
662  */
663 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
664
665 static void
666 free_session_slots(struct nfsd4_session *ses)
667 {
668         int i;
669
670         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
671                 kfree(ses->se_slots[i]);
672 }
673
674 /*
675  * We don't actually need to cache the rpc and session headers, so we
676  * can allocate a little less for each slot:
677  */
678 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
679 {
680         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
681 }
682
683 static int nfsd4_sanitize_slot_size(u32 size)
684 {
685         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
686         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
687
688         return size;
689 }
690
691 /*
692  * XXX: If we run out of reserved DRC memory we could (up to a point)
693  * re-negotiate active sessions and reduce their slot usage to make
694  * room for new connections. For now we just fail the create session.
695  */
696 static int nfsd4_get_drc_mem(int slotsize, u32 num)
697 {
698         int avail;
699
700         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
701
702         spin_lock(&nfsd_drc_lock);
703         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
704                     nfsd_drc_max_mem - nfsd_drc_mem_used);
705         num = min_t(int, num, avail / slotsize);
706         nfsd_drc_mem_used += num * slotsize;
707         spin_unlock(&nfsd_drc_lock);
708
709         return num;
710 }
711
712 static void nfsd4_put_drc_mem(int slotsize, int num)
713 {
714         spin_lock(&nfsd_drc_lock);
715         nfsd_drc_mem_used -= slotsize * num;
716         spin_unlock(&nfsd_drc_lock);
717 }
718
719 static struct nfsd4_session *__alloc_session(int slotsize, int numslots)
720 {
721         struct nfsd4_session *new;
722         int mem, i;
723
724         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
725                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
726         mem = numslots * sizeof(struct nfsd4_slot *);
727
728         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
729         if (!new)
730                 return NULL;
731         /* allocate each struct nfsd4_slot and data cache in one piece */
732         for (i = 0; i < numslots; i++) {
733                 mem = sizeof(struct nfsd4_slot) + slotsize;
734                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
735                 if (!new->se_slots[i])
736                         goto out_free;
737         }
738         return new;
739 out_free:
740         while (i--)
741                 kfree(new->se_slots[i]);
742         kfree(new);
743         return NULL;
744 }
745
746 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new,
747                                    struct nfsd4_channel_attrs *req,
748                                    int numslots, int slotsize,
749                                    struct nfsd_net *nn)
750 {
751         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
752
753         new->maxreqs = numslots;
754         new->maxresp_cached = min_t(u32, req->maxresp_cached,
755                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
756         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
757         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
758         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
759 }
760
761 static void free_conn(struct nfsd4_conn *c)
762 {
763         svc_xprt_put(c->cn_xprt);
764         kfree(c);
765 }
766
767 static void nfsd4_conn_lost(struct svc_xpt_user *u)
768 {
769         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
770         struct nfs4_client *clp = c->cn_session->se_client;
771
772         spin_lock(&clp->cl_lock);
773         if (!list_empty(&c->cn_persession)) {
774                 list_del(&c->cn_persession);
775                 free_conn(c);
776         }
777         spin_unlock(&clp->cl_lock);
778         nfsd4_probe_callback(clp);
779 }
780
781 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
782 {
783         struct nfsd4_conn *conn;
784
785         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
786         if (!conn)
787                 return NULL;
788         svc_xprt_get(rqstp->rq_xprt);
789         conn->cn_xprt = rqstp->rq_xprt;
790         conn->cn_flags = flags;
791         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
792         return conn;
793 }
794
795 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
796 {
797         conn->cn_session = ses;
798         list_add(&conn->cn_persession, &ses->se_conns);
799 }
800
801 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
802 {
803         struct nfs4_client *clp = ses->se_client;
804
805         spin_lock(&clp->cl_lock);
806         __nfsd4_hash_conn(conn, ses);
807         spin_unlock(&clp->cl_lock);
808 }
809
810 static int nfsd4_register_conn(struct nfsd4_conn *conn)
811 {
812         conn->cn_xpt_user.callback = nfsd4_conn_lost;
813         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
814 }
815
816 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
817 {
818         int ret;
819
820         nfsd4_hash_conn(conn, ses);
821         ret = nfsd4_register_conn(conn);
822         if (ret)
823                 /* oops; xprt is already down: */
824                 nfsd4_conn_lost(&conn->cn_xpt_user);
825         if (conn->cn_flags & NFS4_CDFC4_BACK) {
826                 /* callback channel may be back up */
827                 nfsd4_probe_callback(ses->se_client);
828         }
829 }
830
831 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
832 {
833         u32 dir = NFS4_CDFC4_FORE;
834
835         if (cses->flags & SESSION4_BACK_CHAN)
836                 dir |= NFS4_CDFC4_BACK;
837         return alloc_conn(rqstp, dir);
838 }
839
840 /* must be called under client_lock */
841 static void nfsd4_del_conns(struct nfsd4_session *s)
842 {
843         struct nfs4_client *clp = s->se_client;
844         struct nfsd4_conn *c;
845
846         spin_lock(&clp->cl_lock);
847         while (!list_empty(&s->se_conns)) {
848                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
849                 list_del_init(&c->cn_persession);
850                 spin_unlock(&clp->cl_lock);
851
852                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
853                 free_conn(c);
854
855                 spin_lock(&clp->cl_lock);
856         }
857         spin_unlock(&clp->cl_lock);
858 }
859
860 static void __free_session(struct nfsd4_session *ses)
861 {
862         nfsd4_put_drc_mem(slot_bytes(&ses->se_fchannel), ses->se_fchannel.maxreqs);
863         free_session_slots(ses);
864         kfree(ses);
865 }
866
867 static void free_session(struct kref *kref)
868 {
869         struct nfsd4_session *ses;
870         struct nfsd_net *nn;
871
872         ses = container_of(kref, struct nfsd4_session, se_ref);
873         nn = net_generic(ses->se_client->net, nfsd_net_id);
874
875         lockdep_assert_held(&nn->client_lock);
876         nfsd4_del_conns(ses);
877         __free_session(ses);
878 }
879
880 void nfsd4_put_session(struct nfsd4_session *ses)
881 {
882         struct nfsd_net *nn = net_generic(ses->se_client->net, nfsd_net_id);
883
884         spin_lock(&nn->client_lock);
885         nfsd4_put_session_locked(ses);
886         spin_unlock(&nn->client_lock);
887 }
888
889 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fchan,
890                                            struct nfsd_net *nn)
891 {
892         struct nfsd4_session *new;
893         int numslots, slotsize;
894         /*
895          * Note decreasing slot size below client's request may
896          * make it difficult for client to function correctly, whereas
897          * decreasing the number of slots will (just?) affect
898          * performance.  When short on memory we therefore prefer to
899          * decrease number of slots instead of their size.
900          */
901         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
902         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
903         if (numslots < 1)
904                 return NULL;
905
906         new = __alloc_session(slotsize, numslots);
907         if (!new) {
908                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
909                 return NULL;
910         }
911         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize, nn);
912         return new;
913 }
914
915 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
916 {
917         int idx;
918         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
919
920         new->se_client = clp;
921         gen_sessionid(new);
922
923         INIT_LIST_HEAD(&new->se_conns);
924
925         new->se_cb_seq_nr = 1;
926         new->se_flags = cses->flags;
927         new->se_cb_prog = cses->callback_prog;
928         new->se_cb_sec = cses->cb_sec;
929         kref_init(&new->se_ref);
930         idx = hash_sessionid(&new->se_sessionid);
931         spin_lock(&nn->client_lock);
932         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
933         spin_lock(&clp->cl_lock);
934         list_add(&new->se_perclnt, &clp->cl_sessions);
935         spin_unlock(&clp->cl_lock);
936         spin_unlock(&nn->client_lock);
937
938         if (cses->flags & SESSION4_BACK_CHAN) {
939                 struct sockaddr *sa = svc_addr(rqstp);
940                 /*
941                  * This is a little silly; with sessions there's no real
942                  * use for the callback address.  Use the peer address
943                  * as a reasonable default for now, but consider fixing
944                  * the rpc client not to require an address in the
945                  * future:
946                  */
947                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
948                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
949         }
950 }
951
952 /* caller must hold client_lock */
953 static struct nfsd4_session *
954 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
955 {
956         struct nfsd4_session *elem;
957         int idx;
958         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
959
960         dump_sessionid(__func__, sessionid);
961         idx = hash_sessionid(sessionid);
962         /* Search in the appropriate list */
963         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
964                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
965                             NFS4_MAX_SESSIONID_LEN)) {
966                         return elem;
967                 }
968         }
969
970         dprintk("%s: session not found\n", __func__);
971         return NULL;
972 }
973
974 /* caller must hold client_lock */
975 static void
976 unhash_session(struct nfsd4_session *ses)
977 {
978         list_del(&ses->se_hash);
979         spin_lock(&ses->se_client->cl_lock);
980         list_del(&ses->se_perclnt);
981         spin_unlock(&ses->se_client->cl_lock);
982 }
983
984 /* must be called under the client_lock */
985 static inline void
986 renew_client_locked(struct nfs4_client *clp)
987 {
988         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
989
990         if (is_client_expired(clp)) {
991                 WARN_ON(1);
992                 printk("%s: client (clientid %08x/%08x) already expired\n",
993                         __func__,
994                         clp->cl_clientid.cl_boot,
995                         clp->cl_clientid.cl_id);
996                 return;
997         }
998
999         dprintk("renewing client (clientid %08x/%08x)\n", 
1000                         clp->cl_clientid.cl_boot, 
1001                         clp->cl_clientid.cl_id);
1002         list_move_tail(&clp->cl_lru, &nn->client_lru);
1003         clp->cl_time = get_seconds();
1004 }
1005
1006 static inline void
1007 renew_client(struct nfs4_client *clp)
1008 {
1009         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1010
1011         spin_lock(&nn->client_lock);
1012         renew_client_locked(clp);
1013         spin_unlock(&nn->client_lock);
1014 }
1015
1016 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1017 static int
1018 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1019 {
1020         if (clid->cl_boot == nn->boot_time)
1021                 return 0;
1022         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1023                 clid->cl_boot, clid->cl_id, nn->boot_time);
1024         return 1;
1025 }
1026
1027 /* 
1028  * XXX Should we use a slab cache ?
1029  * This type of memory management is somewhat inefficient, but we use it
1030  * anyway since SETCLIENTID is not a common operation.
1031  */
1032 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1033 {
1034         struct nfs4_client *clp;
1035
1036         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1037         if (clp == NULL)
1038                 return NULL;
1039         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1040         if (clp->cl_name.data == NULL) {
1041                 kfree(clp);
1042                 return NULL;
1043         }
1044         clp->cl_name.len = name.len;
1045         return clp;
1046 }
1047
1048 static inline void
1049 free_client(struct nfs4_client *clp)
1050 {
1051         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1052
1053         lockdep_assert_held(&nn->client_lock);
1054         while (!list_empty(&clp->cl_sessions)) {
1055                 struct nfsd4_session *ses;
1056                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1057                                 se_perclnt);
1058                 list_del(&ses->se_perclnt);
1059                 nfsd4_put_session_locked(ses);
1060         }
1061         free_svc_cred(&clp->cl_cred);
1062         kfree(clp->cl_name.data);
1063         kfree(clp);
1064 }
1065
1066 void
1067 release_session_client(struct nfsd4_session *session)
1068 {
1069         struct nfs4_client *clp = session->se_client;
1070         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1071
1072         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
1073                 return;
1074         if (is_client_expired(clp)) {
1075                 free_client(clp);
1076                 session->se_client = NULL;
1077         } else
1078                 renew_client_locked(clp);
1079         spin_unlock(&nn->client_lock);
1080 }
1081
1082 /* must be called under the client_lock */
1083 static inline void
1084 unhash_client_locked(struct nfs4_client *clp)
1085 {
1086         struct nfsd4_session *ses;
1087
1088         mark_client_expired(clp);
1089         list_del(&clp->cl_lru);
1090         spin_lock(&clp->cl_lock);
1091         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1092                 list_del_init(&ses->se_hash);
1093         spin_unlock(&clp->cl_lock);
1094 }
1095
1096 static void
1097 destroy_client(struct nfs4_client *clp)
1098 {
1099         struct nfs4_openowner *oo;
1100         struct nfs4_delegation *dp;
1101         struct list_head reaplist;
1102         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1103
1104         INIT_LIST_HEAD(&reaplist);
1105         spin_lock(&recall_lock);
1106         while (!list_empty(&clp->cl_delegations)) {
1107                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1108                 list_del_init(&dp->dl_perclnt);
1109                 list_move(&dp->dl_recall_lru, &reaplist);
1110         }
1111         spin_unlock(&recall_lock);
1112         while (!list_empty(&reaplist)) {
1113                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1114                 unhash_delegation(dp);
1115         }
1116         while (!list_empty(&clp->cl_openowners)) {
1117                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1118                 release_openowner(oo);
1119         }
1120         nfsd4_shutdown_callback(clp);
1121         if (clp->cl_cb_conn.cb_xprt)
1122                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1123         list_del(&clp->cl_idhash);
1124         if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1125                 rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1126         else
1127                 rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1128         spin_lock(&nn->client_lock);
1129         unhash_client_locked(clp);
1130         if (atomic_read(&clp->cl_refcount) == 0)
1131                 free_client(clp);
1132         spin_unlock(&nn->client_lock);
1133 }
1134
1135 static void expire_client(struct nfs4_client *clp)
1136 {
1137         nfsd4_client_record_remove(clp);
1138         destroy_client(clp);
1139 }
1140
1141 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1142 {
1143         memcpy(target->cl_verifier.data, source->data,
1144                         sizeof(target->cl_verifier.data));
1145 }
1146
1147 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1148 {
1149         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
1150         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
1151 }
1152
1153 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1154 {
1155         if (source->cr_principal) {
1156                 target->cr_principal =
1157                                 kstrdup(source->cr_principal, GFP_KERNEL);
1158                 if (target->cr_principal == NULL)
1159                         return -ENOMEM;
1160         } else
1161                 target->cr_principal = NULL;
1162         target->cr_flavor = source->cr_flavor;
1163         target->cr_uid = source->cr_uid;
1164         target->cr_gid = source->cr_gid;
1165         target->cr_group_info = source->cr_group_info;
1166         get_group_info(target->cr_group_info);
1167         return 0;
1168 }
1169
1170 static long long
1171 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1172 {
1173         long long res;
1174
1175         res = o1->len - o2->len;
1176         if (res)
1177                 return res;
1178         return (long long)memcmp(o1->data, o2->data, o1->len);
1179 }
1180
1181 static int same_name(const char *n1, const char *n2)
1182 {
1183         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1184 }
1185
1186 static int
1187 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1188 {
1189         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1190 }
1191
1192 static int
1193 same_clid(clientid_t *cl1, clientid_t *cl2)
1194 {
1195         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1196 }
1197
1198 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1199 {
1200         int i;
1201
1202         if (g1->ngroups != g2->ngroups)
1203                 return false;
1204         for (i=0; i<g1->ngroups; i++)
1205                 if (!gid_eq(GROUP_AT(g1, i), GROUP_AT(g2, i)))
1206                         return false;
1207         return true;
1208 }
1209
1210 /*
1211  * RFC 3530 language requires clid_inuse be returned when the
1212  * "principal" associated with a requests differs from that previously
1213  * used.  We use uid, gid's, and gss principal string as our best
1214  * approximation.  We also don't want to allow non-gss use of a client
1215  * established using gss: in theory cr_principal should catch that
1216  * change, but in practice cr_principal can be null even in the gss case
1217  * since gssd doesn't always pass down a principal string.
1218  */
1219 static bool is_gss_cred(struct svc_cred *cr)
1220 {
1221         /* Is cr_flavor one of the gss "pseudoflavors"?: */
1222         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
1223 }
1224
1225
1226 static bool
1227 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1228 {
1229         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
1230                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
1231                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
1232                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
1233                 return false;
1234         if (cr1->cr_principal == cr2->cr_principal)
1235                 return true;
1236         if (!cr1->cr_principal || !cr2->cr_principal)
1237                 return false;
1238         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
1239 }
1240
1241 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
1242 {
1243         static u32 current_clientid = 1;
1244
1245         clp->cl_clientid.cl_boot = nn->boot_time;
1246         clp->cl_clientid.cl_id = current_clientid++; 
1247 }
1248
1249 static void gen_confirm(struct nfs4_client *clp)
1250 {
1251         __be32 verf[2];
1252         static u32 i;
1253
1254         verf[0] = (__be32)get_seconds();
1255         verf[1] = (__be32)i++;
1256         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
1257 }
1258
1259 static struct nfs4_stid *find_stateid(struct nfs4_client *cl, stateid_t *t)
1260 {
1261         return idr_find(&cl->cl_stateids, t->si_opaque.so_id);
1262 }
1263
1264 static struct nfs4_stid *find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
1265 {
1266         struct nfs4_stid *s;
1267
1268         s = find_stateid(cl, t);
1269         if (!s)
1270                 return NULL;
1271         if (typemask & s->sc_type)
1272                 return s;
1273         return NULL;
1274 }
1275
1276 static struct nfs4_client *create_client(struct xdr_netobj name,
1277                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1278 {
1279         struct nfs4_client *clp;
1280         struct sockaddr *sa = svc_addr(rqstp);
1281         int ret;
1282         struct net *net = SVC_NET(rqstp);
1283         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1284
1285         clp = alloc_client(name);
1286         if (clp == NULL)
1287                 return NULL;
1288
1289         INIT_LIST_HEAD(&clp->cl_sessions);
1290         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1291         if (ret) {
1292                 spin_lock(&nn->client_lock);
1293                 free_client(clp);
1294                 spin_unlock(&nn->client_lock);
1295                 return NULL;
1296         }
1297         idr_init(&clp->cl_stateids);
1298         atomic_set(&clp->cl_refcount, 0);
1299         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1300         INIT_LIST_HEAD(&clp->cl_idhash);
1301         INIT_LIST_HEAD(&clp->cl_openowners);
1302         INIT_LIST_HEAD(&clp->cl_delegations);
1303         INIT_LIST_HEAD(&clp->cl_lru);
1304         INIT_LIST_HEAD(&clp->cl_callbacks);
1305         spin_lock_init(&clp->cl_lock);
1306         nfsd4_init_callback(&clp->cl_cb_null);
1307         clp->cl_time = get_seconds();
1308         clear_bit(0, &clp->cl_cb_slot_busy);
1309         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1310         copy_verf(clp, verf);
1311         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1312         gen_confirm(clp);
1313         clp->cl_cb_session = NULL;
1314         clp->net = net;
1315         return clp;
1316 }
1317
1318 static void
1319 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
1320 {
1321         struct rb_node **new = &(root->rb_node), *parent = NULL;
1322         struct nfs4_client *clp;
1323
1324         while (*new) {
1325                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
1326                 parent = *new;
1327
1328                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
1329                         new = &((*new)->rb_left);
1330                 else
1331                         new = &((*new)->rb_right);
1332         }
1333
1334         rb_link_node(&new_clp->cl_namenode, parent, new);
1335         rb_insert_color(&new_clp->cl_namenode, root);
1336 }
1337
1338 static struct nfs4_client *
1339 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
1340 {
1341         long long cmp;
1342         struct rb_node *node = root->rb_node;
1343         struct nfs4_client *clp;
1344
1345         while (node) {
1346                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
1347                 cmp = compare_blob(&clp->cl_name, name);
1348                 if (cmp > 0)
1349                         node = node->rb_left;
1350                 else if (cmp < 0)
1351                         node = node->rb_right;
1352                 else
1353                         return clp;
1354         }
1355         return NULL;
1356 }
1357
1358 static void
1359 add_to_unconfirmed(struct nfs4_client *clp)
1360 {
1361         unsigned int idhashval;
1362         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1363
1364         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1365         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
1366         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1367         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
1368         renew_client(clp);
1369 }
1370
1371 static void
1372 move_to_confirmed(struct nfs4_client *clp)
1373 {
1374         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1375         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1376
1377         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1378         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
1379         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1380         add_clp_to_name_tree(clp, &nn->conf_name_tree);
1381         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
1382         renew_client(clp);
1383 }
1384
1385 static struct nfs4_client *
1386 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1387 {
1388         struct nfs4_client *clp;
1389         unsigned int idhashval = clientid_hashval(clid->cl_id);
1390
1391         list_for_each_entry(clp, &nn->conf_id_hashtbl[idhashval], cl_idhash) {
1392                 if (same_clid(&clp->cl_clientid, clid)) {
1393                         if ((bool)clp->cl_minorversion != sessions)
1394                                 return NULL;
1395                         renew_client(clp);
1396                         return clp;
1397                 }
1398         }
1399         return NULL;
1400 }
1401
1402 static struct nfs4_client *
1403 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
1404 {
1405         struct nfs4_client *clp;
1406         unsigned int idhashval = clientid_hashval(clid->cl_id);
1407
1408         list_for_each_entry(clp, &nn->unconf_id_hashtbl[idhashval], cl_idhash) {
1409                 if (same_clid(&clp->cl_clientid, clid)) {
1410                         if ((bool)clp->cl_minorversion != sessions)
1411                                 return NULL;
1412                         return clp;
1413                 }
1414         }
1415         return NULL;
1416 }
1417
1418 static bool clp_used_exchangeid(struct nfs4_client *clp)
1419 {
1420         return clp->cl_exchange_flags != 0;
1421
1422
1423 static struct nfs4_client *
1424 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1425 {
1426         return find_clp_in_name_tree(name, &nn->conf_name_tree);
1427 }
1428
1429 static struct nfs4_client *
1430 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
1431 {
1432         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
1433 }
1434
1435 static void
1436 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1437 {
1438         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1439         struct sockaddr *sa = svc_addr(rqstp);
1440         u32 scopeid = rpc_get_scope_id(sa);
1441         unsigned short expected_family;
1442
1443         /* Currently, we only support tcp and tcp6 for the callback channel */
1444         if (se->se_callback_netid_len == 3 &&
1445             !memcmp(se->se_callback_netid_val, "tcp", 3))
1446                 expected_family = AF_INET;
1447         else if (se->se_callback_netid_len == 4 &&
1448                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1449                 expected_family = AF_INET6;
1450         else
1451                 goto out_err;
1452
1453         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
1454                                             se->se_callback_addr_len,
1455                                             (struct sockaddr *)&conn->cb_addr,
1456                                             sizeof(conn->cb_addr));
1457
1458         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1459                 goto out_err;
1460
1461         if (conn->cb_addr.ss_family == AF_INET6)
1462                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1463
1464         conn->cb_prog = se->se_callback_prog;
1465         conn->cb_ident = se->se_callback_ident;
1466         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
1467         return;
1468 out_err:
1469         conn->cb_addr.ss_family = AF_UNSPEC;
1470         conn->cb_addrlen = 0;
1471         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1472                 "will not receive delegations\n",
1473                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1474
1475         return;
1476 }
1477
1478 /*
1479  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1480  */
1481 void
1482 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1483 {
1484         struct nfsd4_slot *slot = resp->cstate.slot;
1485         unsigned int base;
1486
1487         dprintk("--> %s slot %p\n", __func__, slot);
1488
1489         slot->sl_opcnt = resp->opcnt;
1490         slot->sl_status = resp->cstate.status;
1491
1492         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
1493         if (nfsd4_not_cached(resp)) {
1494                 slot->sl_datalen = 0;
1495                 return;
1496         }
1497         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1498         base = (char *)resp->cstate.datap -
1499                                         (char *)resp->xbuf->head[0].iov_base;
1500         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1501                                     slot->sl_datalen))
1502                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1503         return;
1504 }
1505
1506 /*
1507  * Encode the replay sequence operation from the slot values.
1508  * If cachethis is FALSE encode the uncached rep error on the next
1509  * operation which sets resp->p and increments resp->opcnt for
1510  * nfs4svc_encode_compoundres.
1511  *
1512  */
1513 static __be32
1514 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1515                           struct nfsd4_compoundres *resp)
1516 {
1517         struct nfsd4_op *op;
1518         struct nfsd4_slot *slot = resp->cstate.slot;
1519
1520         /* Encode the replayed sequence operation */
1521         op = &args->ops[resp->opcnt - 1];
1522         nfsd4_encode_operation(resp, op);
1523
1524         /* Return nfserr_retry_uncached_rep in next operation. */
1525         if (args->opcnt > 1 && !(slot->sl_flags & NFSD4_SLOT_CACHETHIS)) {
1526                 op = &args->ops[resp->opcnt++];
1527                 op->status = nfserr_retry_uncached_rep;
1528                 nfsd4_encode_operation(resp, op);
1529         }
1530         return op->status;
1531 }
1532
1533 /*
1534  * The sequence operation is not cached because we can use the slot and
1535  * session values.
1536  */
1537 __be32
1538 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1539                          struct nfsd4_sequence *seq)
1540 {
1541         struct nfsd4_slot *slot = resp->cstate.slot;
1542         __be32 status;
1543
1544         dprintk("--> %s slot %p\n", __func__, slot);
1545
1546         /* Either returns 0 or nfserr_retry_uncached */
1547         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1548         if (status == nfserr_retry_uncached_rep)
1549                 return status;
1550
1551         /* The sequence operation has been encoded, cstate->datap set. */
1552         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1553
1554         resp->opcnt = slot->sl_opcnt;
1555         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1556         status = slot->sl_status;
1557
1558         return status;
1559 }
1560
1561 /*
1562  * Set the exchange_id flags returned by the server.
1563  */
1564 static void
1565 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1566 {
1567         /* pNFS is not supported */
1568         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1569
1570         /* Referrals are supported, Migration is not. */
1571         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1572
1573         /* set the wire flags to return to client. */
1574         clid->flags = new->cl_exchange_flags;
1575 }
1576
1577 static bool client_has_state(struct nfs4_client *clp)
1578 {
1579         /*
1580          * Note clp->cl_openowners check isn't quite right: there's no
1581          * need to count owners without stateid's.
1582          *
1583          * Also note we should probably be using this in 4.0 case too.
1584          */
1585         return !list_empty(&clp->cl_openowners)
1586                 || !list_empty(&clp->cl_delegations)
1587                 || !list_empty(&clp->cl_sessions);
1588 }
1589
1590 __be32
1591 nfsd4_exchange_id(struct svc_rqst *rqstp,
1592                   struct nfsd4_compound_state *cstate,
1593                   struct nfsd4_exchange_id *exid)
1594 {
1595         struct nfs4_client *unconf, *conf, *new;
1596         __be32 status;
1597         char                    addr_str[INET6_ADDRSTRLEN];
1598         nfs4_verifier           verf = exid->verifier;
1599         struct sockaddr         *sa = svc_addr(rqstp);
1600         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
1601         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1602
1603         rpc_ntop(sa, addr_str, sizeof(addr_str));
1604         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1605                 "ip_addr=%s flags %x, spa_how %d\n",
1606                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1607                 addr_str, exid->flags, exid->spa_how);
1608
1609         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
1610                 return nfserr_inval;
1611
1612         /* Currently only support SP4_NONE */
1613         switch (exid->spa_how) {
1614         case SP4_NONE:
1615                 break;
1616         default:                                /* checked by xdr code */
1617                 WARN_ON_ONCE(1);
1618         case SP4_SSV:
1619         case SP4_MACH_CRED:
1620                 return nfserr_serverfault;      /* no excuse :-/ */
1621         }
1622
1623         /* Cases below refer to rfc 5661 section 18.35.4: */
1624         nfs4_lock_state();
1625         conf = find_confirmed_client_by_name(&exid->clname, nn);
1626         if (conf) {
1627                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
1628                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
1629
1630                 if (update) {
1631                         if (!clp_used_exchangeid(conf)) { /* buggy client */
1632                                 status = nfserr_inval;
1633                                 goto out;
1634                         }
1635                         if (!creds_match) { /* case 9 */
1636                                 status = nfserr_perm;
1637                                 goto out;
1638                         }
1639                         if (!verfs_match) { /* case 8 */
1640                                 status = nfserr_not_same;
1641                                 goto out;
1642                         }
1643                         /* case 6 */
1644                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1645                         new = conf;
1646                         goto out_copy;
1647                 }
1648                 if (!creds_match) { /* case 3 */
1649                         if (client_has_state(conf)) {
1650                                 status = nfserr_clid_inuse;
1651                                 goto out;
1652                         }
1653                         expire_client(conf);
1654                         goto out_new;
1655                 }
1656                 if (verfs_match) { /* case 2 */
1657                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
1658                         new = conf;
1659                         goto out_copy;
1660                 }
1661                 /* case 5, client reboot */
1662                 goto out_new;
1663         }
1664
1665         if (update) { /* case 7 */
1666                 status = nfserr_noent;
1667                 goto out;
1668         }
1669
1670         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
1671         if (unconf) /* case 4, possible retry or client restart */
1672                 expire_client(unconf);
1673
1674         /* case 1 (normal case) */
1675 out_new:
1676         new = create_client(exid->clname, rqstp, &verf);
1677         if (new == NULL) {
1678                 status = nfserr_jukebox;
1679                 goto out;
1680         }
1681         new->cl_minorversion = 1;
1682
1683         gen_clid(new, nn);
1684         add_to_unconfirmed(new);
1685 out_copy:
1686         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1687         exid->clientid.cl_id = new->cl_clientid.cl_id;
1688
1689         exid->seqid = new->cl_cs_slot.sl_seqid + 1;
1690         nfsd4_set_ex_flags(new, exid);
1691
1692         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1693                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1694         status = nfs_ok;
1695
1696 out:
1697         nfs4_unlock_state();
1698         return status;
1699 }
1700
1701 static __be32
1702 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1703 {
1704         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1705                 slot_seqid);
1706
1707         /* The slot is in use, and no response has been sent. */
1708         if (slot_inuse) {
1709                 if (seqid == slot_seqid)
1710                         return nfserr_jukebox;
1711                 else
1712                         return nfserr_seq_misordered;
1713         }
1714         /* Note unsigned 32-bit arithmetic handles wraparound: */
1715         if (likely(seqid == slot_seqid + 1))
1716                 return nfs_ok;
1717         if (seqid == slot_seqid)
1718                 return nfserr_replay_cache;
1719         return nfserr_seq_misordered;
1720 }
1721
1722 /*
1723  * Cache the create session result into the create session single DRC
1724  * slot cache by saving the xdr structure. sl_seqid has been set.
1725  * Do this for solo or embedded create session operations.
1726  */
1727 static void
1728 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1729                            struct nfsd4_clid_slot *slot, __be32 nfserr)
1730 {
1731         slot->sl_status = nfserr;
1732         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1733 }
1734
1735 static __be32
1736 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1737                             struct nfsd4_clid_slot *slot)
1738 {
1739         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1740         return slot->sl_status;
1741 }
1742
1743 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
1744                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
1745                         1 +     /* MIN tag is length with zero, only length */ \
1746                         3 +     /* version, opcount, opcode */ \
1747                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1748                                 /* seqid, slotID, slotID, cache */ \
1749                         4 ) * sizeof(__be32))
1750
1751 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
1752                         2 +     /* verifier: AUTH_NULL, length 0 */\
1753                         1 +     /* status */ \
1754                         1 +     /* MIN tag is length with zero, only length */ \
1755                         3 +     /* opcount, opcode, opstatus*/ \
1756                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
1757                                 /* seqid, slotID, slotID, slotID, status */ \
1758                         5 ) * sizeof(__be32))
1759
1760 static bool check_forechannel_attrs(struct nfsd4_channel_attrs fchannel)
1761 {
1762         return fchannel.maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ
1763                 || fchannel.maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ;
1764 }
1765
1766 __be32
1767 nfsd4_create_session(struct svc_rqst *rqstp,
1768                      struct nfsd4_compound_state *cstate,
1769                      struct nfsd4_create_session *cr_ses)
1770 {
1771         struct sockaddr *sa = svc_addr(rqstp);
1772         struct nfs4_client *conf, *unconf;
1773         struct nfsd4_session *new;
1774         struct nfsd4_conn *conn;
1775         struct nfsd4_clid_slot *cs_slot = NULL;
1776         __be32 status = 0;
1777         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1778
1779         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1780                 return nfserr_inval;
1781         if (check_forechannel_attrs(cr_ses->fore_channel))
1782                 return nfserr_toosmall;
1783         new = alloc_session(&cr_ses->fore_channel, nn);
1784         if (!new)
1785                 return nfserr_jukebox;
1786         status = nfserr_jukebox;
1787         conn = alloc_conn_from_crses(rqstp, cr_ses);
1788         if (!conn)
1789                 goto out_free_session;
1790
1791         nfs4_lock_state();
1792         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
1793         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
1794
1795         if (conf) {
1796                 cs_slot = &conf->cl_cs_slot;
1797                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1798                 if (status == nfserr_replay_cache) {
1799                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1800                         goto out_free_conn;
1801                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1802                         status = nfserr_seq_misordered;
1803                         goto out_free_conn;
1804                 }
1805         } else if (unconf) {
1806                 struct nfs4_client *old;
1807                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1808                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1809                         status = nfserr_clid_inuse;
1810                         goto out_free_conn;
1811                 }
1812                 cs_slot = &unconf->cl_cs_slot;
1813                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1814                 if (status) {
1815                         /* an unconfirmed replay returns misordered */
1816                         status = nfserr_seq_misordered;
1817                         goto out_free_conn;
1818                 }
1819                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
1820                 if (old)
1821                         expire_client(old);
1822                 move_to_confirmed(unconf);
1823                 conf = unconf;
1824         } else {
1825                 status = nfserr_stale_clientid;
1826                 goto out_free_conn;
1827         }
1828         status = nfs_ok;
1829         /*
1830          * We do not support RDMA or persistent sessions
1831          */
1832         cr_ses->flags &= ~SESSION4_PERSIST;
1833         cr_ses->flags &= ~SESSION4_RDMA;
1834
1835         init_session(rqstp, new, conf, cr_ses);
1836         nfsd4_init_conn(rqstp, conn, new);
1837
1838         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1839                NFS4_MAX_SESSIONID_LEN);
1840         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1841                 sizeof(struct nfsd4_channel_attrs));
1842         cs_slot->sl_seqid++;
1843         cr_ses->seqid = cs_slot->sl_seqid;
1844
1845         /* cache solo and embedded create sessions under the state lock */
1846         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1847 out:
1848         nfs4_unlock_state();
1849         dprintk("%s returns %d\n", __func__, ntohl(status));
1850         return status;
1851 out_free_conn:
1852         free_conn(conn);
1853 out_free_session:
1854         __free_session(new);
1855         goto out;
1856 }
1857
1858 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1859 {
1860         switch (*dir) {
1861         case NFS4_CDFC4_FORE:
1862         case NFS4_CDFC4_BACK:
1863                 return nfs_ok;
1864         case NFS4_CDFC4_FORE_OR_BOTH:
1865         case NFS4_CDFC4_BACK_OR_BOTH:
1866                 *dir = NFS4_CDFC4_BOTH;
1867                 return nfs_ok;
1868         };
1869         return nfserr_inval;
1870 }
1871
1872 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_backchannel_ctl *bc)
1873 {
1874         struct nfsd4_session *session = cstate->session;
1875         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1876
1877         spin_lock(&nn->client_lock);
1878         session->se_cb_prog = bc->bc_cb_program;
1879         session->se_cb_sec = bc->bc_cb_sec;
1880         spin_unlock(&nn->client_lock);
1881
1882         nfsd4_probe_callback(session->se_client);
1883
1884         return nfs_ok;
1885 }
1886
1887 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1888                      struct nfsd4_compound_state *cstate,
1889                      struct nfsd4_bind_conn_to_session *bcts)
1890 {
1891         __be32 status;
1892         struct nfsd4_conn *conn;
1893         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1894
1895         if (!nfsd4_last_compound_op(rqstp))
1896                 return nfserr_not_only_op;
1897         spin_lock(&nn->client_lock);
1898         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid, SVC_NET(rqstp));
1899         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1900          * client_lock iself: */
1901         if (cstate->session) {
1902                 nfsd4_get_session(cstate->session);
1903                 atomic_inc(&cstate->session->se_client->cl_refcount);
1904         }
1905         spin_unlock(&nn->client_lock);
1906         if (!cstate->session)
1907                 return nfserr_badsession;
1908
1909         status = nfsd4_map_bcts_dir(&bcts->dir);
1910         if (status)
1911                 return status;
1912         conn = alloc_conn(rqstp, bcts->dir);
1913         if (!conn)
1914                 return nfserr_jukebox;
1915         nfsd4_init_conn(rqstp, conn, cstate->session);
1916         return nfs_ok;
1917 }
1918
1919 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1920 {
1921         if (!session)
1922                 return 0;
1923         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1924 }
1925
1926 __be32
1927 nfsd4_destroy_session(struct svc_rqst *r,
1928                       struct nfsd4_compound_state *cstate,
1929                       struct nfsd4_destroy_session *sessionid)
1930 {
1931         struct nfsd4_session *ses;
1932         __be32 status = nfserr_badsession;
1933         struct nfsd_net *nn = net_generic(SVC_NET(r), nfsd_net_id);
1934
1935         /* Notes:
1936          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1937          * - Should we return nfserr_back_chan_busy if waiting for
1938          *   callbacks on to-be-destroyed session?
1939          * - Do we need to clear any callback info from previous session?
1940          */
1941
1942         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1943                 if (!nfsd4_last_compound_op(r))
1944                         return nfserr_not_only_op;
1945         }
1946         dump_sessionid(__func__, &sessionid->sessionid);
1947         spin_lock(&nn->client_lock);
1948         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, SVC_NET(r));
1949         if (!ses) {
1950                 spin_unlock(&nn->client_lock);
1951                 goto out;
1952         }
1953
1954         unhash_session(ses);
1955         spin_unlock(&nn->client_lock);
1956
1957         nfs4_lock_state();
1958         nfsd4_probe_callback_sync(ses->se_client);
1959         nfs4_unlock_state();
1960
1961         spin_lock(&nn->client_lock);
1962         nfsd4_del_conns(ses);
1963         nfsd4_put_session_locked(ses);
1964         spin_unlock(&nn->client_lock);
1965         status = nfs_ok;
1966 out:
1967         dprintk("%s returns %d\n", __func__, ntohl(status));
1968         return status;
1969 }
1970
1971 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1972 {
1973         struct nfsd4_conn *c;
1974
1975         list_for_each_entry(c, &s->se_conns, cn_persession) {
1976                 if (c->cn_xprt == xpt) {
1977                         return c;
1978                 }
1979         }
1980         return NULL;
1981 }
1982
1983 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1984 {
1985         struct nfs4_client *clp = ses->se_client;
1986         struct nfsd4_conn *c;
1987         int ret;
1988
1989         spin_lock(&clp->cl_lock);
1990         c = __nfsd4_find_conn(new->cn_xprt, ses);
1991         if (c) {
1992                 spin_unlock(&clp->cl_lock);
1993                 free_conn(new);
1994                 return;
1995         }
1996         __nfsd4_hash_conn(new, ses);
1997         spin_unlock(&clp->cl_lock);
1998         ret = nfsd4_register_conn(new);
1999         if (ret)
2000                 /* oops; xprt is already down: */
2001                 nfsd4_conn_lost(&new->cn_xpt_user);
2002         return;
2003 }
2004
2005 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
2006 {
2007         struct nfsd4_compoundargs *args = rqstp->rq_argp;
2008
2009         return args->opcnt > session->se_fchannel.maxops;
2010 }
2011
2012 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
2013                                   struct nfsd4_session *session)
2014 {
2015         struct xdr_buf *xb = &rqstp->rq_arg;
2016
2017         return xb->len > session->se_fchannel.maxreq_sz;
2018 }
2019
2020 __be32
2021 nfsd4_sequence(struct svc_rqst *rqstp,
2022                struct nfsd4_compound_state *cstate,
2023                struct nfsd4_sequence *seq)
2024 {
2025         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2026         struct nfsd4_session *session;
2027         struct nfsd4_slot *slot;
2028         struct nfsd4_conn *conn;
2029         __be32 status;
2030         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2031
2032         if (resp->opcnt != 1)
2033                 return nfserr_sequence_pos;
2034
2035         /*
2036          * Will be either used or freed by nfsd4_sequence_check_conn
2037          * below.
2038          */
2039         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
2040         if (!conn)
2041                 return nfserr_jukebox;
2042
2043         spin_lock(&nn->client_lock);
2044         status = nfserr_badsession;
2045         session = find_in_sessionid_hashtbl(&seq->sessionid, SVC_NET(rqstp));
2046         if (!session)
2047                 goto out;
2048
2049         status = nfserr_too_many_ops;
2050         if (nfsd4_session_too_many_ops(rqstp, session))
2051                 goto out;
2052
2053         status = nfserr_req_too_big;
2054         if (nfsd4_request_too_big(rqstp, session))
2055                 goto out;
2056
2057         status = nfserr_badslot;
2058         if (seq->slotid >= session->se_fchannel.maxreqs)
2059                 goto out;
2060
2061         slot = session->se_slots[seq->slotid];
2062         dprintk("%s: slotid %d\n", __func__, seq->slotid);
2063
2064         /* We do not negotiate the number of slots yet, so set the
2065          * maxslots to the session maxreqs which is used to encode
2066          * sr_highest_slotid and the sr_target_slot id to maxslots */
2067         seq->maxslots = session->se_fchannel.maxreqs;
2068
2069         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
2070                                         slot->sl_flags & NFSD4_SLOT_INUSE);
2071         if (status == nfserr_replay_cache) {
2072                 status = nfserr_seq_misordered;
2073                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
2074                         goto out;
2075                 cstate->slot = slot;
2076                 cstate->session = session;
2077                 /* Return the cached reply status and set cstate->status
2078                  * for nfsd4_proc_compound processing */
2079                 status = nfsd4_replay_cache_entry(resp, seq);
2080                 cstate->status = nfserr_replay_cache;
2081                 goto out;
2082         }
2083         if (status)
2084                 goto out;
2085
2086         nfsd4_sequence_check_conn(conn, session);
2087         conn = NULL;
2088
2089         /* Success! bump slot seqid */
2090         slot->sl_seqid = seq->seqid;
2091         slot->sl_flags |= NFSD4_SLOT_INUSE;
2092         if (seq->cachethis)
2093                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
2094         else
2095                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
2096
2097         cstate->slot = slot;
2098         cstate->session = session;
2099
2100 out:
2101         /* Hold a session reference until done processing the compound. */
2102         if (cstate->session) {
2103                 struct nfs4_client *clp = session->se_client;
2104
2105                 nfsd4_get_session(cstate->session);
2106                 atomic_inc(&clp->cl_refcount);
2107                 switch (clp->cl_cb_state) {
2108                 case NFSD4_CB_DOWN:
2109                         seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
2110                         break;
2111                 case NFSD4_CB_FAULT:
2112                         seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
2113                         break;
2114                 default:
2115                         seq->status_flags = 0;
2116                 }
2117         }
2118         kfree(conn);
2119         spin_unlock(&nn->client_lock);
2120         dprintk("%s: return %d\n", __func__, ntohl(status));
2121         return status;
2122 }
2123
2124 __be32
2125 nfsd4_destroy_clientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_destroy_clientid *dc)
2126 {
2127         struct nfs4_client *conf, *unconf, *clp;
2128         __be32 status = 0;
2129         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2130
2131         nfs4_lock_state();
2132         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
2133         conf = find_confirmed_client(&dc->clientid, true, nn);
2134
2135         if (conf) {
2136                 clp = conf;
2137
2138                 if (!is_client_expired(conf) && client_has_state(conf)) {
2139                         status = nfserr_clientid_busy;
2140                         goto out;
2141                 }
2142
2143                 /* rfc5661 18.50.3 */
2144                 if (cstate->session && conf == cstate->session->se_client) {
2145                         status = nfserr_clientid_busy;
2146                         goto out;
2147                 }
2148         } else if (unconf)
2149                 clp = unconf;
2150         else {
2151                 status = nfserr_stale_clientid;
2152                 goto out;
2153         }
2154
2155         expire_client(clp);
2156 out:
2157         nfs4_unlock_state();
2158         dprintk("%s return %d\n", __func__, ntohl(status));
2159         return status;
2160 }
2161
2162 __be32
2163 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
2164 {
2165         __be32 status = 0;
2166
2167         if (rc->rca_one_fs) {
2168                 if (!cstate->current_fh.fh_dentry)
2169                         return nfserr_nofilehandle;
2170                 /*
2171                  * We don't take advantage of the rca_one_fs case.
2172                  * That's OK, it's optional, we can safely ignore it.
2173                  */
2174                  return nfs_ok;
2175         }
2176
2177         nfs4_lock_state();
2178         status = nfserr_complete_already;
2179         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
2180                              &cstate->session->se_client->cl_flags))
2181                 goto out;
2182
2183         status = nfserr_stale_clientid;
2184         if (is_client_expired(cstate->session->se_client))
2185                 /*
2186                  * The following error isn't really legal.
2187                  * But we only get here if the client just explicitly
2188                  * destroyed the client.  Surely it no longer cares what
2189                  * error it gets back on an operation for the dead
2190                  * client.
2191                  */
2192                 goto out;
2193
2194         status = nfs_ok;
2195         nfsd4_client_record_create(cstate->session->se_client);
2196 out:
2197         nfs4_unlock_state();
2198         return status;
2199 }
2200
2201 __be32
2202 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2203                   struct nfsd4_setclientid *setclid)
2204 {
2205         struct xdr_netobj       clname = setclid->se_name;
2206         nfs4_verifier           clverifier = setclid->se_verf;
2207         struct nfs4_client      *conf, *unconf, *new;
2208         __be32                  status;
2209         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2210
2211         /* Cases below refer to rfc 3530 section 14.2.33: */
2212         nfs4_lock_state();
2213         conf = find_confirmed_client_by_name(&clname, nn);
2214         if (conf) {
2215                 /* case 0: */
2216                 status = nfserr_clid_inuse;
2217                 if (clp_used_exchangeid(conf))
2218                         goto out;
2219                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
2220                         char addr_str[INET6_ADDRSTRLEN];
2221                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
2222                                  sizeof(addr_str));
2223                         dprintk("NFSD: setclientid: string in use by client "
2224                                 "at %s\n", addr_str);
2225                         goto out;
2226                 }
2227         }
2228         unconf = find_unconfirmed_client_by_name(&clname, nn);
2229         if (unconf)
2230                 expire_client(unconf);
2231         status = nfserr_jukebox;
2232         new = create_client(clname, rqstp, &clverifier);
2233         if (new == NULL)
2234                 goto out;
2235         if (conf && same_verf(&conf->cl_verifier, &clverifier))
2236                 /* case 1: probable callback update */
2237                 copy_clid(new, conf);
2238         else /* case 4 (new client) or cases 2, 3 (client reboot): */
2239                 gen_clid(new, nn);
2240         new->cl_minorversion = 0;
2241         gen_callback(new, setclid, rqstp);
2242         add_to_unconfirmed(new);
2243         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
2244         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
2245         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
2246         status = nfs_ok;
2247 out:
2248         nfs4_unlock_state();
2249         return status;
2250 }
2251
2252
2253 __be32
2254 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
2255                          struct nfsd4_compound_state *cstate,
2256                          struct nfsd4_setclientid_confirm *setclientid_confirm)
2257 {
2258         struct nfs4_client *conf, *unconf;
2259         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
2260         clientid_t * clid = &setclientid_confirm->sc_clientid;
2261         __be32 status;
2262         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2263
2264         if (STALE_CLIENTID(clid, nn))
2265                 return nfserr_stale_clientid;
2266         nfs4_lock_state();
2267
2268         conf = find_confirmed_client(clid, false, nn);
2269         unconf = find_unconfirmed_client(clid, false, nn);
2270         /*
2271          * We try hard to give out unique clientid's, so if we get an
2272          * attempt to confirm the same clientid with a different cred,
2273          * there's a bug somewhere.  Let's charitably assume it's our
2274          * bug.
2275          */
2276         status = nfserr_serverfault;
2277         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
2278                 goto out;
2279         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
2280                 goto out;
2281         /* cases below refer to rfc 3530 section 14.2.34: */
2282         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
2283                 if (conf && !unconf) /* case 2: probable retransmit */
2284                         status = nfs_ok;
2285                 else /* case 4: client hasn't noticed we rebooted yet? */
2286                         status = nfserr_stale_clientid;
2287                 goto out;
2288         }
2289         status = nfs_ok;
2290         if (conf) { /* case 1: callback update */
2291                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2292                 nfsd4_probe_callback(conf);
2293                 expire_client(unconf);
2294         } else { /* case 3: normal case; new or rebooted client */
2295                 conf = find_confirmed_client_by_name(&unconf->cl_name, nn);
2296                 if (conf)
2297                         expire_client(conf);
2298                 move_to_confirmed(unconf);
2299                 nfsd4_probe_callback(unconf);
2300         }
2301 out:
2302         nfs4_unlock_state();
2303         return status;
2304 }
2305
2306 static struct nfs4_file *nfsd4_alloc_file(void)
2307 {
2308         return kmem_cache_alloc(file_slab, GFP_KERNEL);
2309 }
2310
2311 /* OPEN Share state helper functions */
2312 static void nfsd4_init_file(struct nfs4_file *fp, struct inode *ino)
2313 {
2314         unsigned int hashval = file_hashval(ino);
2315
2316         atomic_set(&fp->fi_ref, 1);
2317         INIT_LIST_HEAD(&fp->fi_hash);
2318         INIT_LIST_HEAD(&fp->fi_stateids);
2319         INIT_LIST_HEAD(&fp->fi_delegations);
2320         fp->fi_inode = igrab(ino);
2321         fp->fi_had_conflict = false;
2322         fp->fi_lease = NULL;
2323         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2324         memset(fp->fi_access, 0, sizeof(fp->fi_access));
2325         spin_lock(&recall_lock);
2326         list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2327         spin_unlock(&recall_lock);
2328 }
2329
2330 static void
2331 nfsd4_free_slab(struct kmem_cache **slab)
2332 {
2333         if (*slab == NULL)
2334                 return;
2335         kmem_cache_destroy(*slab);
2336         *slab = NULL;
2337 }
2338
2339 void
2340 nfsd4_free_slabs(void)
2341 {
2342         nfsd4_free_slab(&openowner_slab);
2343         nfsd4_free_slab(&lockowner_slab);
2344         nfsd4_free_slab(&file_slab);
2345         nfsd4_free_slab(&stateid_slab);
2346         nfsd4_free_slab(&deleg_slab);
2347 }
2348
2349 int
2350 nfsd4_init_slabs(void)
2351 {
2352         openowner_slab = kmem_cache_create("nfsd4_openowners",
2353                         sizeof(struct nfs4_openowner), 0, 0, NULL);
2354         if (openowner_slab == NULL)
2355                 goto out_nomem;
2356         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
2357                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
2358         if (lockowner_slab == NULL)
2359                 goto out_nomem;
2360         file_slab = kmem_cache_create("nfsd4_files",
2361                         sizeof(struct nfs4_file), 0, 0, NULL);
2362         if (file_slab == NULL)
2363                 goto out_nomem;
2364         stateid_slab = kmem_cache_create("nfsd4_stateids",
2365                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
2366         if (stateid_slab == NULL)
2367                 goto out_nomem;
2368         deleg_slab = kmem_cache_create("nfsd4_delegations",
2369                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2370         if (deleg_slab == NULL)
2371                 goto out_nomem;
2372         return 0;
2373 out_nomem:
2374         nfsd4_free_slabs();
2375         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2376         return -ENOMEM;
2377 }
2378
2379 void nfs4_free_openowner(struct nfs4_openowner *oo)
2380 {
2381         kfree(oo->oo_owner.so_owner.data);
2382         kmem_cache_free(openowner_slab, oo);
2383 }
2384
2385 void nfs4_free_lockowner(struct nfs4_lockowner *lo)
2386 {
2387         kfree(lo->lo_owner.so_owner.data);
2388         kmem_cache_free(lockowner_slab, lo);
2389 }
2390
2391 static void init_nfs4_replay(struct nfs4_replay *rp)
2392 {
2393         rp->rp_status = nfserr_serverfault;
2394         rp->rp_buflen = 0;
2395         rp->rp_buf = rp->rp_ibuf;
2396 }
2397
2398 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
2399 {
2400         struct nfs4_stateowner *sop;
2401
2402         sop = kmem_cache_alloc(slab, GFP_KERNEL);
2403         if (!sop)
2404                 return NULL;
2405
2406         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
2407         if (!sop->so_owner.data) {
2408                 kmem_cache_free(slab, sop);
2409                 return NULL;
2410         }
2411         sop->so_owner.len = owner->len;
2412
2413         INIT_LIST_HEAD(&sop->so_stateids);
2414         sop->so_client = clp;
2415         init_nfs4_replay(&sop->so_replay);
2416         return sop;
2417 }
2418
2419 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
2420 {
2421         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2422
2423         list_add(&oo->oo_owner.so_strhash, &nn->ownerstr_hashtbl[strhashval]);
2424         list_add(&oo->oo_perclient, &clp->cl_openowners);
2425 }
2426
2427 static struct nfs4_openowner *
2428 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2429         struct nfs4_openowner *oo;
2430
2431         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
2432         if (!oo)
2433                 return NULL;
2434         oo->oo_owner.so_is_open_owner = 1;
2435         oo->oo_owner.so_seqid = open->op_seqid;
2436         oo->oo_flags = NFS4_OO_NEW;
2437         oo->oo_time = 0;
2438         oo->oo_last_closed_stid = NULL;
2439         INIT_LIST_HEAD(&oo->oo_close_lru);
2440         hash_openowner(oo, clp, strhashval);
2441         return oo;
2442 }
2443
2444 static void init_open_stateid(struct nfs4_ol_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2445         struct nfs4_openowner *oo = open->op_openowner;
2446         struct nfs4_client *clp = oo->oo_owner.so_client;
2447
2448         init_stid(&stp->st_stid, clp, NFS4_OPEN_STID);
2449         INIT_LIST_HEAD(&stp->st_lockowners);
2450         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
2451         list_add(&stp->st_perfile, &fp->fi_stateids);
2452         stp->st_stateowner = &oo->oo_owner;
2453         get_nfs4_file(fp);
2454         stp->st_file = fp;
2455         stp->st_access_bmap = 0;
2456         stp->st_deny_bmap = 0;
2457         set_access(open->op_share_access, stp);
2458         set_deny(open->op_share_deny, stp);
2459         stp->st_openstp = NULL;
2460 }
2461
2462 static void
2463 move_to_close_lru(struct nfs4_openowner *oo, struct net *net)
2464 {
2465         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2466
2467         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
2468
2469         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
2470         oo->oo_time = get_seconds();
2471 }
2472
2473 static int
2474 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2475                                                         clientid_t *clid)
2476 {
2477         return (sop->so_owner.len == owner->len) &&
2478                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2479                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2480 }
2481
2482 static struct nfs4_openowner *
2483 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
2484                         bool sessions, struct nfsd_net *nn)
2485 {
2486         struct nfs4_stateowner *so;
2487         struct nfs4_openowner *oo;
2488         struct nfs4_client *clp;
2489
2490         list_for_each_entry(so, &nn->ownerstr_hashtbl[hashval], so_strhash) {
2491                 if (!so->so_is_open_owner)
2492                         continue;
2493                 if (same_owner_str(so, &open->op_owner, &open->op_clientid)) {
2494                         oo = openowner(so);
2495                         clp = oo->oo_owner.so_client;
2496                         if ((bool)clp->cl_minorversion != sessions)
2497                                 return NULL;
2498                         renew_client(oo->oo_owner.so_client);
2499                         return oo;
2500                 }
2501         }
2502         return NULL;
2503 }
2504
2505 /* search file_hashtbl[] for file */
2506 static struct nfs4_file *
2507 find_file(struct inode *ino)
2508 {
2509         unsigned int hashval = file_hashval(ino);
2510         struct nfs4_file *fp;
2511
2512         spin_lock(&recall_lock);
2513         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2514                 if (fp->fi_inode == ino) {
2515                         get_nfs4_file(fp);
2516                         spin_unlock(&recall_lock);
2517                         return fp;
2518                 }
2519         }
2520         spin_unlock(&recall_lock);
2521         return NULL;
2522 }
2523
2524 /*
2525  * Called to check deny when READ with all zero stateid or
2526  * WRITE with all zero or all one stateid
2527  */
2528 static __be32
2529 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2530 {
2531         struct inode *ino = current_fh->fh_dentry->d_inode;
2532         struct nfs4_file *fp;
2533         struct nfs4_ol_stateid *stp;
2534         __be32 ret;
2535
2536         dprintk("NFSD: nfs4_share_conflict\n");
2537
2538         fp = find_file(ino);
2539         if (!fp)
2540                 return nfs_ok;
2541         ret = nfserr_locked;
2542         /* Search for conflicting share reservations */
2543         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2544                 if (test_deny(deny_type, stp) ||
2545                     test_deny(NFS4_SHARE_DENY_BOTH, stp))
2546                         goto out;
2547         }
2548         ret = nfs_ok;
2549 out:
2550         put_nfs4_file(fp);
2551         return ret;
2552 }
2553
2554 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2555 {
2556         /* We're assuming the state code never drops its reference
2557          * without first removing the lease.  Since we're in this lease
2558          * callback (and since the lease code is serialized by the kernel
2559          * lock) we know the server hasn't removed the lease yet, we know
2560          * it's safe to take a reference: */
2561         atomic_inc(&dp->dl_count);
2562
2563         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2564
2565         /* only place dl_time is set. protected by lock_flocks*/
2566         dp->dl_time = get_seconds();
2567
2568         nfsd4_cb_recall(dp);
2569 }
2570
2571 /* Called from break_lease() with lock_flocks() held. */
2572 static void nfsd_break_deleg_cb(struct file_lock *fl)
2573 {
2574         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2575         struct nfs4_delegation *dp;
2576
2577         if (!fp) {
2578                 WARN(1, "(%p)->fl_owner NULL\n", fl);
2579                 return;
2580         }
2581         if (fp->fi_had_conflict) {
2582                 WARN(1, "duplicate break on %p\n", fp);
2583                 return;
2584         }
2585         /*
2586          * We don't want the locks code to timeout the lease for us;
2587          * we'll remove it ourself if a delegation isn't returned
2588          * in time:
2589          */
2590         fl->fl_break_time = 0;
2591
2592         spin_lock(&recall_lock);
2593         fp->fi_had_conflict = true;
2594         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2595                 nfsd_break_one_deleg(dp);
2596         spin_unlock(&recall_lock);
2597 }
2598
2599 static
2600 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2601 {
2602         if (arg & F_UNLCK)
2603                 return lease_modify(onlist, arg);
2604         else
2605                 return -EAGAIN;
2606 }
2607
2608 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2609         .lm_break = nfsd_break_deleg_cb,
2610         .lm_change = nfsd_change_deleg_cb,
2611 };
2612
2613 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
2614 {
2615         if (nfsd4_has_session(cstate))
2616                 return nfs_ok;
2617         if (seqid == so->so_seqid - 1)
2618                 return nfserr_replay_me;
2619         if (seqid == so->so_seqid)
2620                 return nfs_ok;
2621         return nfserr_bad_seqid;
2622 }
2623
2624 __be32
2625 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2626                     struct nfsd4_open *open, struct nfsd_net *nn)
2627 {
2628         clientid_t *clientid = &open->op_clientid;
2629         struct nfs4_client *clp = NULL;
2630         unsigned int strhashval;
2631         struct nfs4_openowner *oo = NULL;
2632         __be32 status;
2633
2634         if (STALE_CLIENTID(&open->op_clientid, nn))
2635                 return nfserr_stale_clientid;
2636         /*
2637          * In case we need it later, after we've already created the
2638          * file and don't want to risk a further failure:
2639          */
2640         open->op_file = nfsd4_alloc_file();
2641         if (open->op_file == NULL)
2642                 return nfserr_jukebox;
2643
2644         strhashval = ownerstr_hashval(clientid->cl_id, &open->op_owner);
2645         oo = find_openstateowner_str(strhashval, open, cstate->minorversion, nn);
2646         open->op_openowner = oo;
2647         if (!oo) {
2648                 clp = find_confirmed_client(clientid, cstate->minorversion,
2649                                             nn);
2650                 if (clp == NULL)
2651                         return nfserr_expired;
2652                 goto new_owner;
2653         }
2654         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
2655                 /* Replace unconfirmed owners without checking for replay. */
2656                 clp = oo->oo_owner.so_client;
2657                 release_openowner(oo);
2658                 open->op_openowner = NULL;
2659                 goto new_owner;
2660         }
2661         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
2662         if (status)
2663                 return status;
2664         clp = oo->oo_owner.so_client;
2665         goto alloc_stateid;
2666 new_owner:
2667         oo = alloc_init_open_stateowner(strhashval, clp, open);
2668         if (oo == NULL)
2669                 return nfserr_jukebox;
2670         open->op_openowner = oo;
2671 alloc_stateid:
2672         open->op_stp = nfs4_alloc_stateid(clp);
2673         if (!open->op_stp)
2674                 return nfserr_jukebox;
2675         return nfs_ok;
2676 }
2677
2678 static inline __be32
2679 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2680 {
2681         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2682                 return nfserr_openmode;
2683         else
2684                 return nfs_ok;
2685 }
2686
2687 static int share_access_to_flags(u32 share_access)
2688 {
2689         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2690 }
2691
2692 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
2693 {
2694         struct nfs4_stid *ret;
2695
2696         ret = find_stateid_by_type(cl, s, NFS4_DELEG_STID);
2697         if (!ret)
2698                 return NULL;
2699         return delegstateid(ret);
2700 }
2701
2702 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
2703 {
2704         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
2705                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
2706 }
2707
2708 static __be32
2709 nfs4_check_deleg(struct nfs4_client *cl, struct nfs4_file *fp, struct nfsd4_open *open,
2710                 struct nfs4_delegation **dp)
2711 {
2712         int flags;
2713         __be32 status = nfserr_bad_stateid;
2714
2715         *dp = find_deleg_stateid(cl, &open->op_delegate_stateid);
2716         if (*dp == NULL)
2717                 goto out;
2718         flags = share_access_to_flags(open->op_share_access);
2719         status = nfs4_check_delegmode(*dp, flags);
2720         if (status)
2721                 *dp = NULL;
2722 out:
2723         if (!nfsd4_is_deleg_cur(open))
2724                 return nfs_ok;
2725         if (status)
2726                 return status;
2727         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2728         return nfs_ok;
2729 }
2730
2731 static __be32
2732 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_ol_stateid **stpp)
2733 {
2734         struct nfs4_ol_stateid *local;
2735         struct nfs4_openowner *oo = open->op_openowner;
2736
2737         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2738                 /* ignore lock owners */
2739                 if (local->st_stateowner->so_is_open_owner == 0)
2740                         continue;
2741                 /* remember if we have seen this open owner */
2742                 if (local->st_stateowner == &oo->oo_owner)
2743                         *stpp = local;
2744                 /* check for conflicting share reservations */
2745                 if (!test_share(local, open))
2746                         return nfserr_share_denied;
2747         }
2748         return nfs_ok;
2749 }
2750
2751 static inline int nfs4_access_to_access(u32 nfs4_access)
2752 {
2753         int flags = 0;
2754
2755         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2756                 flags |= NFSD_MAY_READ;
2757         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2758                 flags |= NFSD_MAY_WRITE;
2759         return flags;
2760 }
2761
2762 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2763                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2764 {
2765         __be32 status;
2766         int oflag = nfs4_access_to_omode(open->op_share_access);
2767         int access = nfs4_access_to_access(open->op_share_access);
2768
2769         if (!fp->fi_fds[oflag]) {
2770                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2771                         &fp->fi_fds[oflag]);
2772                 if (status)
2773                         return status;
2774         }
2775         nfs4_file_get_access(fp, oflag);
2776
2777         return nfs_ok;
2778 }
2779
2780 static inline __be32
2781 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2782                 struct nfsd4_open *open)
2783 {
2784         struct iattr iattr = {
2785                 .ia_valid = ATTR_SIZE,
2786                 .ia_size = 0,
2787         };
2788         if (!open->op_truncate)
2789                 return 0;
2790         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2791                 return nfserr_inval;
2792         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2793 }
2794
2795 static __be32
2796 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)
2797 {
2798         u32 op_share_access = open->op_share_access;
2799         bool new_access;
2800         __be32 status;
2801
2802         new_access = !test_access(op_share_access, stp);
2803         if (new_access) {
2804                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2805                 if (status)
2806                         return status;
2807         }
2808         status = nfsd4_truncate(rqstp, cur_fh, open);
2809         if (status) {
2810                 if (new_access) {
2811                         int oflag = nfs4_access_to_omode(op_share_access);
2812                         nfs4_file_put_access(fp, oflag);
2813                 }
2814                 return status;
2815         }
2816         /* remember the open */
2817         set_access(op_share_access, stp);
2818         set_deny(open->op_share_deny, stp);
2819
2820         return nfs_ok;
2821 }
2822
2823
2824 static void
2825 nfs4_set_claim_prev(struct nfsd4_open *open, bool has_session)
2826 {
2827         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
2828 }
2829
2830 /* Should we give out recallable state?: */
2831 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2832 {
2833         if (clp->cl_cb_state == NFSD4_CB_UP)
2834                 return true;
2835         /*
2836          * In the sessions case, since we don't have to establish a
2837          * separate connection for callbacks, we assume it's OK
2838          * until we hear otherwise:
2839          */
2840         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2841 }
2842
2843 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2844 {
2845         struct file_lock *fl;
2846
2847         fl = locks_alloc_lock();
2848         if (!fl)
2849                 return NULL;
2850         locks_init_lock(fl);
2851         fl->fl_lmops = &nfsd_lease_mng_ops;
2852         fl->fl_flags = FL_LEASE;
2853         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2854         fl->fl_end = OFFSET_MAX;
2855         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2856         fl->fl_pid = current->tgid;
2857         return fl;
2858 }
2859
2860 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2861 {
2862         struct nfs4_file *fp = dp->dl_file;
2863         struct file_lock *fl;
2864         int status;
2865
2866         fl = nfs4_alloc_init_lease(dp, flag);
2867         if (!fl)
2868                 return -ENOMEM;
2869         fl->fl_file = find_readable_file(fp);
2870         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2871         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2872         if (status) {
2873                 list_del_init(&dp->dl_perclnt);
2874                 locks_free_lock(fl);
2875                 return -ENOMEM;
2876         }
2877         fp->fi_lease = fl;
2878         fp->fi_deleg_file = get_file(fl->fl_file);
2879         atomic_set(&fp->fi_delegees, 1);
2880         list_add(&dp->dl_perfile, &fp->fi_delegations);
2881         return 0;
2882 }
2883
2884 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2885 {
2886         struct nfs4_file *fp = dp->dl_file;
2887
2888         if (!fp->fi_lease)
2889                 return nfs4_setlease(dp, flag);
2890         spin_lock(&recall_lock);
2891         if (fp->fi_had_conflict) {
2892                 spin_unlock(&recall_lock);
2893                 return -EAGAIN;
2894         }
2895         atomic_inc(&fp->fi_delegees);
2896         list_add(&dp->dl_perfile, &fp->fi_delegations);
2897         spin_unlock(&recall_lock);
2898         list_add(&dp->dl_perclnt, &dp->dl_stid.sc_client->cl_delegations);
2899         return 0;
2900 }
2901
2902 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
2903 {
2904         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2905         if (status == -EAGAIN)
2906                 open->op_why_no_deleg = WND4_CONTENTION;
2907         else {
2908                 open->op_why_no_deleg = WND4_RESOURCE;
2909                 switch (open->op_deleg_want) {
2910                 case NFS4_SHARE_WANT_READ_DELEG:
2911                 case NFS4_SHARE_WANT_WRITE_DELEG:
2912                 case NFS4_SHARE_WANT_ANY_DELEG:
2913                         break;
2914                 case NFS4_SHARE_WANT_CANCEL:
2915                         open->op_why_no_deleg = WND4_CANCELLED;
2916                         break;
2917                 case NFS4_SHARE_WANT_NO_DELEG:
2918                         WARN_ON_ONCE(1);
2919                 }
2920         }
2921 }
2922
2923 /*
2924  * Attempt to hand out a delegation.
2925  */
2926 static void
2927 nfs4_open_delegation(struct net *net, struct svc_fh *fh,
2928                      struct nfsd4_open *open, struct nfs4_ol_stateid *stp)
2929 {
2930         struct nfs4_delegation *dp;
2931         struct nfs4_openowner *oo = container_of(stp->st_stateowner, struct nfs4_openowner, oo_owner);
2932         int cb_up;
2933         int status = 0, flag = 0;
2934
2935         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
2936         flag = NFS4_OPEN_DELEGATE_NONE;
2937         open->op_recall = 0;
2938         switch (open->op_claim_type) {
2939                 case NFS4_OPEN_CLAIM_PREVIOUS:
2940                         if (!cb_up)
2941                                 open->op_recall = 1;
2942                         flag = open->op_delegate_type;
2943                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2944                                 goto out;
2945                         break;
2946                 case NFS4_OPEN_CLAIM_NULL:
2947                         /* Let's not give out any delegations till everyone's
2948                          * had the chance to reclaim theirs.... */
2949                         if (locks_in_grace(net))
2950                                 goto out;
2951                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
2952                                 goto out;
2953                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2954                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2955                         else
2956                                 flag = NFS4_OPEN_DELEGATE_READ;
2957                         break;
2958                 default:
2959                         goto out;
2960         }
2961
2962         dp = alloc_init_deleg(oo->oo_owner.so_client, stp, fh, flag);
2963         if (dp == NULL)
2964                 goto out_no_deleg;
2965         status = nfs4_set_delegation(dp, flag);
2966         if (status)
2967                 goto out_free;
2968
2969         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
2970
2971         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2972                 STATEID_VAL(&dp->dl_stid.sc_stateid));
2973 out:
2974         open->op_delegate_type = flag;
2975         if (flag == NFS4_OPEN_DELEGATE_NONE) {
2976                 if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
2977                     open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2978                         dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2979
2980                 /* 4.1 client asking for a delegation? */
2981                 if (open->op_deleg_want)
2982                         nfsd4_open_deleg_none_ext(open, status);
2983         }
2984         return;
2985 out_free:
2986         unhash_stid(&dp->dl_stid);
2987         nfs4_put_delegation(dp);
2988 out_no_deleg:
2989         flag = NFS4_OPEN_DELEGATE_NONE;
2990         goto out;
2991 }
2992
2993 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
2994                                         struct nfs4_delegation *dp)
2995 {
2996         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
2997             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
2998                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
2999                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
3000         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
3001                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
3002                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3003                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
3004         }
3005         /* Otherwise the client must be confused wanting a delegation
3006          * it already has, therefore we don't return
3007          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
3008          */
3009 }
3010
3011 /*
3012  * called with nfs4_lock_state() held.
3013  */
3014 __be32
3015 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
3016 {
3017         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3018         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
3019         struct nfs4_file *fp = NULL;
3020         struct inode *ino = current_fh->fh_dentry->d_inode;
3021         struct nfs4_ol_stateid *stp = NULL;
3022         struct nfs4_delegation *dp = NULL;
3023         __be32 status;
3024
3025         /*
3026          * Lookup file; if found, lookup stateid and check open request,
3027          * and check for delegations in the process of being recalled.
3028          * If not found, create the nfs4_file struct
3029          */
3030         fp = find_file(ino);
3031         if (fp) {
3032                 if ((status = nfs4_check_open(fp, open, &stp)))
3033                         goto out;
3034                 status = nfs4_check_deleg(cl, fp, open, &dp);
3035                 if (status)
3036                         goto out;
3037         } else {
3038                 status = nfserr_bad_stateid;
3039                 if (nfsd4_is_deleg_cur(open))
3040                         goto out;
3041                 status = nfserr_jukebox;
3042                 fp = open->op_file;
3043                 open->op_file = NULL;
3044                 nfsd4_init_file(fp, ino);
3045         }
3046
3047         /*
3048          * OPEN the file, or upgrade an existing OPEN.
3049          * If truncate fails, the OPEN fails.
3050          */
3051         if (stp) {
3052                 /* Stateid was found, this is an OPEN upgrade */
3053                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
3054                 if (status)
3055                         goto out;
3056         } else {
3057                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, open);
3058                 if (status)
3059                         goto out;
3060                 status = nfsd4_truncate(rqstp, current_fh, open);
3061                 if (status)
3062                         goto out;
3063                 stp = open->op_stp;
3064                 open->op_stp = NULL;
3065                 init_open_stateid(stp, fp, open);
3066         }
3067         update_stateid(&stp->st_stid.sc_stateid);
3068         memcpy(&open->op_stateid, &stp->st_stid.sc_stateid, sizeof(stateid_t));
3069
3070         if (nfsd4_has_session(&resp->cstate)) {
3071                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
3072
3073                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
3074                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
3075                         open->op_why_no_deleg = WND4_NOT_WANTED;
3076                         goto nodeleg;
3077                 }
3078         }
3079
3080         /*
3081         * Attempt to hand out a delegation. No error return, because the
3082         * OPEN succeeds even if we fail.
3083         */
3084         nfs4_open_delegation(SVC_NET(rqstp), current_fh, open, stp);
3085 nodeleg:
3086         status = nfs_ok;
3087
3088         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
3089                 STATEID_VAL(&stp->st_stid.sc_stateid));
3090 out:
3091         /* 4.1 client trying to upgrade/downgrade delegation? */
3092         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
3093             open->op_deleg_want)
3094                 nfsd4_deleg_xgrade_none_ext(open, dp);
3095
3096         if (fp)
3097                 put_nfs4_file(fp);
3098         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
3099                 nfs4_set_claim_prev(open, nfsd4_has_session(&resp->cstate));
3100         /*
3101         * To finish the open response, we just need to set the rflags.
3102         */
3103         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
3104         if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED) &&
3105             !nfsd4_has_session(&resp->cstate))
3106                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
3107
3108         return status;
3109 }
3110
3111 void nfsd4_cleanup_open_state(struct nfsd4_open *open, __be32 status)
3112 {
3113         if (open->op_openowner) {
3114                 struct nfs4_openowner *oo = open->op_openowner;
3115
3116                 if (!list_empty(&oo->oo_owner.so_stateids))
3117                         list_del_init(&oo->oo_close_lru);
3118                 if (oo->oo_flags & NFS4_OO_NEW) {
3119                         if (status) {
3120                                 release_openowner(oo);
3121                                 open->op_openowner = NULL;
3122                         } else
3123                                 oo->oo_flags &= ~NFS4_OO_NEW;
3124                 }
3125         }
3126         if (open->op_file)
3127                 nfsd4_free_file(open->op_file);
3128         if (open->op_stp)
3129                 free_generic_stateid(open->op_stp);
3130 }
3131
3132 static __be32 lookup_clientid(clientid_t *clid, bool session, struct nfsd_net *nn, struct nfs4_client **clp)
3133 {
3134         struct nfs4_client *found;
3135
3136         if (STALE_CLIENTID(clid, nn))
3137                 return nfserr_stale_clientid;
3138         found = find_confirmed_client(clid, session, nn);
3139         if (clp)
3140                 *clp = found;
3141         return found ? nfs_ok : nfserr_expired;
3142 }
3143
3144 __be32
3145 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3146             clientid_t *clid)
3147 {
3148         struct nfs4_client *clp;
3149         __be32 status;
3150         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3151
3152         nfs4_lock_state();
3153         dprintk("process_renew(%08x/%08x): starting\n", 
3154                         clid->cl_boot, clid->cl_id);
3155         status = lookup_clientid(clid, cstate->minorversion, nn, &clp);
3156         if (status)
3157                 goto out;
3158         status = nfserr_cb_path_down;
3159         if (!list_empty(&clp->cl_delegations)
3160                         && clp->cl_cb_state != NFSD4_CB_UP)
3161                 goto out;
3162         status = nfs_ok;
3163 out:
3164         nfs4_unlock_state();
3165         return status;
3166 }
3167
3168 static void
3169 nfsd4_end_grace(struct nfsd_net *nn)
3170 {
3171         /* do nothing if grace period already ended */
3172         if (nn->grace_ended)
3173                 return;
3174
3175         dprintk("NFSD: end of grace period\n");
3176         nn->grace_ended = true;
3177         nfsd4_record_grace_done(nn, nn->boot_time);
3178         locks_end_grace(&nn->nfsd4_manager);
3179         /*
3180          * Now that every NFSv4 client has had the chance to recover and
3181          * to see the (possibly new, possibly shorter) lease time, we
3182          * can safely set the next grace time to the current lease time:
3183          */
3184         nn->nfsd4_grace = nn->nfsd4_lease;
3185 }
3186
3187 static time_t
3188 nfs4_laundromat(struct nfsd_net *nn)
3189 {
3190         struct nfs4_client *clp;
3191         struct nfs4_openowner *oo;
3192         struct nfs4_delegation *dp;
3193         struct list_head *pos, *next, reaplist;
3194         time_t cutoff = get_seconds() - nn->nfsd4_lease;
3195         time_t t, clientid_val = nn->nfsd4_lease;
3196         time_t u, test_val = nn->nfsd4_lease;
3197
3198         nfs4_lock_state();
3199
3200         dprintk("NFSD: laundromat service - starting\n");
3201         nfsd4_end_grace(nn);
3202         INIT_LIST_HEAD(&reaplist);
3203         spin_lock(&nn->client_lock);
3204         list_for_each_safe(pos, next, &nn->client_lru) {
3205                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3206                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
3207                         t = clp->cl_time - cutoff;
3208                         if (clientid_val > t)
3209                                 clientid_val = t;
3210                         break;
3211                 }
3212                 if (atomic_read(&clp->cl_refcount)) {
3213                         dprintk("NFSD: client in use (clientid %08x)\n",
3214                                 clp->cl_clientid.cl_id);
3215                         continue;
3216                 }
3217                 unhash_client_locked(clp);
3218                 list_add(&clp->cl_lru, &reaplist);
3219         }
3220         spin_unlock(&nn->client_lock);
3221         list_for_each_safe(pos, next, &reaplist) {
3222                 clp = list_entry(pos, struct nfs4_client, cl_lru);
3223                 dprintk("NFSD: purging unused client (clientid %08x)\n",
3224                         clp->cl_clientid.cl_id);
3225                 expire_client(clp);
3226         }
3227         spin_lock(&recall_lock);
3228         list_for_each_safe(pos, next, &del_recall_lru) {
3229                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3230                 if (net_generic(dp->dl_stid.sc_client->net, nfsd_net_id) != nn)
3231                         continue;
3232                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
3233                         u = dp->dl_time - cutoff;
3234                         if (test_val > u)
3235                                 test_val = u;
3236                         break;
3237                 }
3238                 list_move(&dp->dl_recall_lru, &reaplist);
3239         }
3240         spin_unlock(&recall_lock);
3241         list_for_each_safe(pos, next, &reaplist) {
3242                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
3243                 unhash_delegation(dp);
3244         }
3245         test_val = nn->nfsd4_lease;
3246         list_for_each_safe(pos, next, &nn->close_lru) {
3247                 oo = container_of(pos, struct nfs4_openowner, oo_close_lru);
3248                 if (time_after((unsigned long)oo->oo_time, (unsigned long)cutoff)) {
3249                         u = oo->oo_time - cutoff;
3250                         if (test_val > u)
3251                                 test_val = u;
3252                         break;
3253                 }
3254                 release_openowner(oo);
3255         }
3256         if (clientid_val < NFSD_LAUNDROMAT_MINTIMEOUT)
3257                 clientid_val = NFSD_LAUNDROMAT_MINTIMEOUT;
3258         nfs4_unlock_state();
3259         return clientid_val;
3260 }
3261
3262 static struct workqueue_struct *laundry_wq;
3263 static void laundromat_main(struct work_struct *);
3264
3265 static void
3266 laundromat_main(struct work_struct *laundry)
3267 {
3268         time_t t;
3269         struct delayed_work *dwork = container_of(laundry, struct delayed_work,
3270                                                   work);
3271         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
3272                                            laundromat_work);
3273
3274         t = nfs4_laundromat(nn);
3275         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
3276         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
3277 }
3278
3279 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_ol_stateid *stp)
3280 {
3281         if (fhp->fh_dentry->d_inode != stp->st_file->fi_inode)
3282                 return nfserr_bad_stateid;
3283         return nfs_ok;
3284 }
3285
3286 static int
3287 STALE_STATEID(stateid_t *stateid, struct nfsd_net *nn)
3288 {
3289         if (stateid->si_opaque.so_clid.cl_boot == nn->boot_time)
3290                 return 0;
3291         dprintk("NFSD: stale stateid " STATEID_FMT "!\n",
3292                 STATEID_VAL(stateid));
3293         return 1;
3294 }
3295
3296 static inline int
3297 access_permit_read(struct nfs4_ol_stateid *stp)
3298 {
3299         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
3300                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
3301                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
3302 }
3303
3304 static inline int
3305 access_permit_write(struct nfs4_ol_stateid *stp)
3306 {
3307         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
3308                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
3309 }
3310
3311 static
3312 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
3313 {
3314         __be32 status = nfserr_openmode;
3315
3316         /* For lock stateid's, we test the parent open, not the lock: */
3317         if (stp->st_openstp)
3318                 stp = stp->st_openstp;
3319         if ((flags & WR_STATE) && !access_permit_write(stp))
3320                 goto out;
3321         if ((flags & RD_STATE) && !access_permit_read(stp))
3322                 goto out;
3323         status = nfs_ok;
3324 out:
3325         return status;
3326 }
3327
3328 static inline __be32
3329 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
3330 {
3331         if (ONE_STATEID(stateid) && (flags & RD_STATE))
3332                 return nfs_ok;
3333         else if (locks_in_grace(net)) {
3334                 /* Answer in remaining cases depends on existence of
3335                  * conflicting state; so we must wait out the grace period. */
3336                 return nfserr_grace;
3337         } else if (flags & WR_STATE)
3338                 return nfs4_share_conflict(current_fh,
3339                                 NFS4_SHARE_DENY_WRITE);
3340         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
3341                 return nfs4_share_conflict(current_fh,
3342                                 NFS4_SHARE_DENY_READ);
3343 }
3344
3345 /*
3346  * Allow READ/WRITE during grace period on recovered state only for files
3347  * that are not able to provide mandatory locking.
3348  */
3349 static inline int
3350 grace_disallows_io(struct net *net, struct inode *inode)
3351 {
3352         return locks_in_grace(net) && mandatory_lock(inode);
3353 }
3354
3355 /* Returns true iff a is later than b: */
3356 static bool stateid_generation_after(stateid_t *a, stateid_t *b)
3357 {
3358         return (s32)a->si_generation - (s32)b->si_generation > 0;
3359 }
3360
3361 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
3362 {
3363         /*
3364          * When sessions are used the stateid generation number is ignored
3365          * when it is zero.
3366          */
3367         if (has_session && in->si_generation == 0)
3368                 return nfs_ok;
3369
3370         if (in->si_generation == ref->si_generation)
3371                 return nfs_ok;
3372
3373         /* If the client sends us a stateid from the future, it's buggy: */
3374         if (stateid_generation_after(in, ref))
3375                 return nfserr_bad_stateid;
3376         /*
3377          * However, we could see a stateid from the past, even from a
3378          * non-buggy client.  For example, if the client sends a lock
3379          * while some IO is outstanding, the lock may bump si_generation
3380          * while the IO is still in flight.  The client could avoid that
3381          * situation by waiting for responses on all the IO requests,
3382          * but better performance may result in retrying IO that
3383          * receives an old_stateid error if requests are rarely
3384          * reordered in flight:
3385          */
3386         return nfserr_old_stateid;
3387 }
3388
3389 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
3390 {
3391         struct nfs4_stid *s;
3392         struct nfs4_ol_stateid *ols;
3393         __be32 status;
3394
3395         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3396                 return nfserr_bad_stateid;
3397         /* Client debugging aid. */
3398         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
3399                 char addr_str[INET6_ADDRSTRLEN];
3400                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
3401                                  sizeof(addr_str));
3402                 pr_warn_ratelimited("NFSD: client %s testing state ID "
3403                                         "with incorrect client ID\n", addr_str);
3404                 return nfserr_bad_stateid;
3405         }
3406         s = find_stateid(cl, stateid);
3407         if (!s)
3408                 return nfserr_bad_stateid;
3409         status = check_stateid_generation(stateid, &s->sc_stateid, 1);
3410         if (status)
3411                 return status;
3412         if (!(s->sc_type & (NFS4_OPEN_STID | NFS4_LOCK_STID)))
3413                 return nfs_ok;
3414         ols = openlockstateid(s);
3415         if (ols->st_stateowner->so_is_open_owner
3416             && !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3417                 return nfserr_bad_stateid;
3418         return nfs_ok;
3419 }
3420
3421 static __be32 nfsd4_lookup_stateid(stateid_t *stateid, unsigned char typemask,
3422                                    struct nfs4_stid **s, bool sessions,
3423                                    struct nfsd_net *nn)
3424 {
3425         struct nfs4_client *cl;
3426
3427         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3428                 return nfserr_bad_stateid;
3429         if (STALE_STATEID(stateid, nn))
3430                 return nfserr_stale_stateid;
3431         cl = find_confirmed_client(&stateid->si_opaque.so_clid, sessions, nn);
3432         if (!cl)
3433                 return nfserr_expired;
3434         *s = find_stateid_by_type(cl, stateid, typemask);
3435         if (!*s)
3436                 return nfserr_bad_stateid;
3437         return nfs_ok;
3438
3439 }
3440
3441 /*
3442 * Checks for stateid operations
3443 */
3444 __be32
3445 nfs4_preprocess_stateid_op(struct net *net, struct nfsd4_compound_state *cstate,
3446                            stateid_t *stateid, int flags, struct file **filpp)
3447 {
3448         struct nfs4_stid *s;
3449         struct nfs4_ol_stateid *stp = NULL;
3450         struct nfs4_delegation *dp = NULL;
3451         struct svc_fh *current_fh = &cstate->current_fh;
3452         struct inode *ino = current_fh->fh_dentry->d_inode;
3453         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3454         __be32 status;
3455
3456         if (filpp)
3457                 *filpp = NULL;
3458
3459         if (grace_disallows_io(net, ino))
3460                 return nfserr_grace;
3461
3462         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid))
3463                 return check_special_stateids(net, current_fh, stateid, flags);
3464
3465         status = nfsd4_lookup_stateid(stateid, NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
3466                                       &s, cstate->minorversion, nn);
3467         if (status)
3468                 return status;
3469         status = check_stateid_generation(stateid, &s->sc_stateid, nfsd4_has_session(cstate));
3470         if (status)
3471                 goto out;
3472         switch (s->sc_type) {
3473         case NFS4_DELEG_STID:
3474                 dp = delegstateid(s);
3475                 status = nfs4_check_delegmode(dp, flags);
3476                 if (status)
3477                         goto out;
3478                 if (filpp) {
3479                         *filpp = dp->dl_file->fi_deleg_file;
3480                         if (!*filpp) {
3481                                 WARN_ON_ONCE(1);
3482                                 status = nfserr_serverfault;
3483                                 goto out;
3484                         }
3485                 }
3486                 break;
3487         case NFS4_OPEN_STID:
3488         case NFS4_LOCK_STID:
3489                 stp = openlockstateid(s);
3490                 status = nfs4_check_fh(current_fh, stp);
3491                 if (status)
3492                         goto out;
3493                 if (stp->st_stateowner->so_is_open_owner
3494                     && !(openowner(stp->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
3495                         goto out;
3496                 status = nfs4_check_openmode(stp, flags);
3497                 if (status)
3498                         goto out;
3499                 if (filpp) {
3500                         if (flags & RD_STATE)
3501                                 *filpp = find_readable_file(stp->st_file);
3502                         else
3503                                 *filpp = find_writeable_file(stp->st_file);
3504                 }
3505                 break;