nfsd4: don't close read-write opens too soon
[opensuse:kernel.git] / fs / nfsd / nfs4state.c
1 /*
2 *  Copyright (c) 2001 The Regents of the University of Michigan.
3 *  All rights reserved.
4 *
5 *  Kendrick Smith <kmsmith@umich.edu>
6 *  Andy Adamson <kandros@umich.edu>
7 *
8 *  Redistribution and use in source and binary forms, with or without
9 *  modification, are permitted provided that the following conditions
10 *  are met:
11 *
12 *  1. Redistributions of source code must retain the above copyright
13 *     notice, this list of conditions and the following disclaimer.
14 *  2. Redistributions in binary form must reproduce the above copyright
15 *     notice, this list of conditions and the following disclaimer in the
16 *     documentation and/or other materials provided with the distribution.
17 *  3. Neither the name of the University nor the names of its
18 *     contributors may be used to endorse or promote products derived
19 *     from this software without specific prior written permission.
20 *
21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 *
33 */
34
35 #include <linux/file.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/namei.h>
39 #include <linux/swap.h>
40 #include <linux/sunrpc/svcauth_gss.h>
41 #include <linux/sunrpc/clnt.h>
42 #include "xdr4.h"
43 #include "vfs.h"
44
45 #define NFSDDBG_FACILITY                NFSDDBG_PROC
46
47 /* Globals */
48 time_t nfsd4_lease = 90;     /* default lease time */
49 time_t nfsd4_grace = 90;
50 static time_t boot_time;
51 static u32 current_ownerid = 1;
52 static u32 current_fileid = 1;
53 static u32 current_delegid = 1;
54 static stateid_t zerostateid;             /* bits all 0 */
55 static stateid_t onestateid;              /* bits all 1 */
56 static u64 current_sessionid = 1;
57
58 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zerostateid, sizeof(stateid_t)))
59 #define ONE_STATEID(stateid)  (!memcmp((stateid), &onestateid, sizeof(stateid_t)))
60
61 /* forward declarations */
62 static struct nfs4_stateid * find_stateid(stateid_t *stid, int flags);
63 static struct nfs4_delegation * find_delegation_stateid(struct inode *ino, stateid_t *stid);
64 static char user_recovery_dirname[PATH_MAX] = "/var/lib/nfs/v4recovery";
65 static void nfs4_set_recdir(char *recdir);
66
67 /* Locking: */
68
69 /* Currently used for almost all code touching nfsv4 state: */
70 static DEFINE_MUTEX(client_mutex);
71
72 /*
73  * Currently used for the del_recall_lru and file hash table.  In an
74  * effort to decrease the scope of the client_mutex, this spinlock may
75  * eventually cover more:
76  */
77 static DEFINE_SPINLOCK(recall_lock);
78
79 static struct kmem_cache *stateowner_slab = NULL;
80 static struct kmem_cache *file_slab = NULL;
81 static struct kmem_cache *stateid_slab = NULL;
82 static struct kmem_cache *deleg_slab = NULL;
83
84 void
85 nfs4_lock_state(void)
86 {
87         mutex_lock(&client_mutex);
88 }
89
90 void
91 nfs4_unlock_state(void)
92 {
93         mutex_unlock(&client_mutex);
94 }
95
96 static inline u32
97 opaque_hashval(const void *ptr, int nbytes)
98 {
99         unsigned char *cptr = (unsigned char *) ptr;
100
101         u32 x = 0;
102         while (nbytes--) {
103                 x *= 37;
104                 x += *cptr++;
105         }
106         return x;
107 }
108
109 static struct list_head del_recall_lru;
110
111 static inline void
112 put_nfs4_file(struct nfs4_file *fi)
113 {
114         if (atomic_dec_and_lock(&fi->fi_ref, &recall_lock)) {
115                 list_del(&fi->fi_hash);
116                 spin_unlock(&recall_lock);
117                 iput(fi->fi_inode);
118                 kmem_cache_free(file_slab, fi);
119         }
120 }
121
122 static inline void
123 get_nfs4_file(struct nfs4_file *fi)
124 {
125         atomic_inc(&fi->fi_ref);
126 }
127
128 static int num_delegations;
129 unsigned int max_delegations;
130
131 /*
132  * Open owner state (share locks)
133  */
134
135 /* hash tables for nfs4_stateowner */
136 #define OWNER_HASH_BITS              8
137 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
138 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
139
140 #define ownerid_hashval(id) \
141         ((id) & OWNER_HASH_MASK)
142 #define ownerstr_hashval(clientid, ownername) \
143         (((clientid) + opaque_hashval((ownername.data), (ownername.len))) & OWNER_HASH_MASK)
144
145 static struct list_head ownerid_hashtbl[OWNER_HASH_SIZE];
146 static struct list_head ownerstr_hashtbl[OWNER_HASH_SIZE];
147
148 /* hash table for nfs4_file */
149 #define FILE_HASH_BITS                   8
150 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
151
152 /* hash table for (open)nfs4_stateid */
153 #define STATEID_HASH_BITS              10
154 #define STATEID_HASH_SIZE              (1 << STATEID_HASH_BITS)
155 #define STATEID_HASH_MASK              (STATEID_HASH_SIZE - 1)
156
157 #define file_hashval(x) \
158         hash_ptr(x, FILE_HASH_BITS)
159 #define stateid_hashval(owner_id, file_id)  \
160         (((owner_id) + (file_id)) & STATEID_HASH_MASK)
161
162 static struct list_head file_hashtbl[FILE_HASH_SIZE];
163 static struct list_head stateid_hashtbl[STATEID_HASH_SIZE];
164
165 static void __nfs4_file_get_access(struct nfs4_file *fp, int oflag)
166 {
167         BUG_ON(!(fp->fi_fds[oflag] || fp->fi_fds[O_RDWR]));
168         atomic_inc(&fp->fi_access[oflag]);
169 }
170
171 static void nfs4_file_get_access(struct nfs4_file *fp, int oflag)
172 {
173         if (oflag == O_RDWR) {
174                 __nfs4_file_get_access(fp, O_RDONLY);
175                 __nfs4_file_get_access(fp, O_WRONLY);
176         } else
177                 __nfs4_file_get_access(fp, oflag);
178 }
179
180 static void nfs4_file_put_fd(struct nfs4_file *fp, int oflag)
181 {
182         if (fp->fi_fds[oflag]) {
183                 fput(fp->fi_fds[oflag]);
184                 fp->fi_fds[oflag] = NULL;
185         }
186 }
187
188 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
189 {
190         if (atomic_dec_and_test(&fp->fi_access[oflag])) {
191                 nfs4_file_put_fd(fp, oflag);
192                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
193                         nfs4_file_put_fd(fp, O_RDWR);
194         }
195 }
196
197 static void nfs4_file_put_access(struct nfs4_file *fp, int oflag)
198 {
199         if (oflag == O_RDWR) {
200                 __nfs4_file_put_access(fp, O_RDONLY);
201                 __nfs4_file_put_access(fp, O_WRONLY);
202         } else
203                 __nfs4_file_put_access(fp, oflag);
204 }
205
206 static struct nfs4_delegation *
207 alloc_init_deleg(struct nfs4_client *clp, struct nfs4_stateid *stp, struct svc_fh *current_fh, u32 type)
208 {
209         struct nfs4_delegation *dp;
210         struct nfs4_file *fp = stp->st_file;
211
212         dprintk("NFSD alloc_init_deleg\n");
213         /*
214          * Major work on the lease subsystem (for example, to support
215          * calbacks on stat) will be required before we can support
216          * write delegations properly.
217          */
218         if (type != NFS4_OPEN_DELEGATE_READ)
219                 return NULL;
220         if (fp->fi_had_conflict)
221                 return NULL;
222         if (num_delegations > max_delegations)
223                 return NULL;
224         dp = kmem_cache_alloc(deleg_slab, GFP_KERNEL);
225         if (dp == NULL)
226                 return dp;
227         num_delegations++;
228         INIT_LIST_HEAD(&dp->dl_perfile);
229         INIT_LIST_HEAD(&dp->dl_perclnt);
230         INIT_LIST_HEAD(&dp->dl_recall_lru);
231         dp->dl_client = clp;
232         get_nfs4_file(fp);
233         dp->dl_file = fp;
234         dp->dl_type = type;
235         dp->dl_stateid.si_boot = boot_time;
236         dp->dl_stateid.si_stateownerid = current_delegid++;
237         dp->dl_stateid.si_fileid = 0;
238         dp->dl_stateid.si_generation = 0;
239         fh_copy_shallow(&dp->dl_fh, &current_fh->fh_handle);
240         dp->dl_time = 0;
241         atomic_set(&dp->dl_count, 1);
242         INIT_WORK(&dp->dl_recall.cb_work, nfsd4_do_callback_rpc);
243         return dp;
244 }
245
246 void
247 nfs4_put_delegation(struct nfs4_delegation *dp)
248 {
249         if (atomic_dec_and_test(&dp->dl_count)) {
250                 dprintk("NFSD: freeing dp %p\n",dp);
251                 put_nfs4_file(dp->dl_file);
252                 kmem_cache_free(deleg_slab, dp);
253                 num_delegations--;
254         }
255 }
256
257 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
258 {
259         if (atomic_dec_and_test(&fp->fi_delegees)) {
260                 vfs_setlease(fp->fi_deleg_file, F_UNLCK, &fp->fi_lease);
261                 fp->fi_lease = NULL;
262                 fput(fp->fi_deleg_file);
263                 fp->fi_deleg_file = NULL;
264         }
265 }
266
267 /* Called under the state lock. */
268 static void
269 unhash_delegation(struct nfs4_delegation *dp)
270 {
271         list_del_init(&dp->dl_perclnt);
272         spin_lock(&recall_lock);
273         list_del_init(&dp->dl_perfile);
274         list_del_init(&dp->dl_recall_lru);
275         spin_unlock(&recall_lock);
276         nfs4_put_deleg_lease(dp->dl_file);
277         nfs4_put_delegation(dp);
278 }
279
280 /* 
281  * SETCLIENTID state 
282  */
283
284 /* client_lock protects the client lru list and session hash table */
285 static DEFINE_SPINLOCK(client_lock);
286
287 /* Hash tables for nfs4_clientid state */
288 #define CLIENT_HASH_BITS                 4
289 #define CLIENT_HASH_SIZE                (1 << CLIENT_HASH_BITS)
290 #define CLIENT_HASH_MASK                (CLIENT_HASH_SIZE - 1)
291
292 #define clientid_hashval(id) \
293         ((id) & CLIENT_HASH_MASK)
294 #define clientstr_hashval(name) \
295         (opaque_hashval((name), 8) & CLIENT_HASH_MASK)
296 /*
297  * reclaim_str_hashtbl[] holds known client info from previous reset/reboot
298  * used in reboot/reset lease grace period processing
299  *
300  * conf_id_hashtbl[], and conf_str_hashtbl[] hold confirmed
301  * setclientid_confirmed info. 
302  *
303  * unconf_str_hastbl[] and unconf_id_hashtbl[] hold unconfirmed 
304  * setclientid info.
305  *
306  * client_lru holds client queue ordered by nfs4_client.cl_time
307  * for lease renewal.
308  *
309  * close_lru holds (open) stateowner queue ordered by nfs4_stateowner.so_time
310  * for last close replay.
311  */
312 static struct list_head reclaim_str_hashtbl[CLIENT_HASH_SIZE];
313 static int reclaim_str_hashtbl_size = 0;
314 static struct list_head conf_id_hashtbl[CLIENT_HASH_SIZE];
315 static struct list_head conf_str_hashtbl[CLIENT_HASH_SIZE];
316 static struct list_head unconf_str_hashtbl[CLIENT_HASH_SIZE];
317 static struct list_head unconf_id_hashtbl[CLIENT_HASH_SIZE];
318 static struct list_head client_lru;
319 static struct list_head close_lru;
320
321 /*
322  * We store the NONE, READ, WRITE, and BOTH bits separately in the
323  * st_{access,deny}_bmap field of the stateid, in order to track not
324  * only what share bits are currently in force, but also what
325  * combinations of share bits previous opens have used.  This allows us
326  * to enforce the recommendation of rfc 3530 14.2.19 that the server
327  * return an error if the client attempt to downgrade to a combination
328  * of share bits not explicable by closing some of its previous opens.
329  *
330  * XXX: This enforcement is actually incomplete, since we don't keep
331  * track of access/deny bit combinations; so, e.g., we allow:
332  *
333  *      OPEN allow read, deny write
334  *      OPEN allow both, deny none
335  *      DOWNGRADE allow read, deny none
336  *
337  * which we should reject.
338  */
339 static void
340 set_access(unsigned int *access, unsigned long bmap) {
341         int i;
342
343         *access = 0;
344         for (i = 1; i < 4; i++) {
345                 if (test_bit(i, &bmap))
346                         *access |= i;
347         }
348 }
349
350 static void
351 set_deny(unsigned int *deny, unsigned long bmap) {
352         int i;
353
354         *deny = 0;
355         for (i = 0; i < 4; i++) {
356                 if (test_bit(i, &bmap))
357                         *deny |= i ;
358         }
359 }
360
361 static int
362 test_share(struct nfs4_stateid *stp, struct nfsd4_open *open) {
363         unsigned int access, deny;
364
365         set_access(&access, stp->st_access_bmap);
366         set_deny(&deny, stp->st_deny_bmap);
367         if ((access & open->op_share_deny) || (deny & open->op_share_access))
368                 return 0;
369         return 1;
370 }
371
372 static int nfs4_access_to_omode(u32 access)
373 {
374         switch (access & NFS4_SHARE_ACCESS_BOTH) {
375         case NFS4_SHARE_ACCESS_READ:
376                 return O_RDONLY;
377         case NFS4_SHARE_ACCESS_WRITE:
378                 return O_WRONLY;
379         case NFS4_SHARE_ACCESS_BOTH:
380                 return O_RDWR;
381         }
382         BUG();
383 }
384
385 static void unhash_generic_stateid(struct nfs4_stateid *stp)
386 {
387         list_del(&stp->st_hash);
388         list_del(&stp->st_perfile);
389         list_del(&stp->st_perstateowner);
390 }
391
392 static void free_generic_stateid(struct nfs4_stateid *stp)
393 {
394         int i;
395
396         if (stp->st_access_bmap) {
397                 for (i = 1; i < 4; i++) {
398                         if (test_bit(i, &stp->st_access_bmap))
399                                 nfs4_file_put_access(stp->st_file,
400                                                 nfs4_access_to_omode(i));
401                 }
402         }
403         put_nfs4_file(stp->st_file);
404         kmem_cache_free(stateid_slab, stp);
405 }
406
407 static void release_lock_stateid(struct nfs4_stateid *stp)
408 {
409         struct file *file;
410
411         unhash_generic_stateid(stp);
412         file = find_any_file(stp->st_file);
413         if (file)
414                 locks_remove_posix(file, (fl_owner_t)stp->st_stateowner);
415         free_generic_stateid(stp);
416 }
417
418 static void unhash_lockowner(struct nfs4_stateowner *sop)
419 {
420         struct nfs4_stateid *stp;
421
422         list_del(&sop->so_idhash);
423         list_del(&sop->so_strhash);
424         list_del(&sop->so_perstateid);
425         while (!list_empty(&sop->so_stateids)) {
426                 stp = list_first_entry(&sop->so_stateids,
427                                 struct nfs4_stateid, st_perstateowner);
428                 release_lock_stateid(stp);
429         }
430 }
431
432 static void release_lockowner(struct nfs4_stateowner *sop)
433 {
434         unhash_lockowner(sop);
435         nfs4_put_stateowner(sop);
436 }
437
438 static void
439 release_stateid_lockowners(struct nfs4_stateid *open_stp)
440 {
441         struct nfs4_stateowner *lock_sop;
442
443         while (!list_empty(&open_stp->st_lockowners)) {
444                 lock_sop = list_entry(open_stp->st_lockowners.next,
445                                 struct nfs4_stateowner, so_perstateid);
446                 /* list_del(&open_stp->st_lockowners);  */
447                 BUG_ON(lock_sop->so_is_open_owner);
448                 release_lockowner(lock_sop);
449         }
450 }
451
452 static void release_open_stateid(struct nfs4_stateid *stp)
453 {
454         unhash_generic_stateid(stp);
455         release_stateid_lockowners(stp);
456         free_generic_stateid(stp);
457 }
458
459 static void unhash_openowner(struct nfs4_stateowner *sop)
460 {
461         struct nfs4_stateid *stp;
462
463         list_del(&sop->so_idhash);
464         list_del(&sop->so_strhash);
465         list_del(&sop->so_perclient);
466         list_del(&sop->so_perstateid); /* XXX: necessary? */
467         while (!list_empty(&sop->so_stateids)) {
468                 stp = list_first_entry(&sop->so_stateids,
469                                 struct nfs4_stateid, st_perstateowner);
470                 release_open_stateid(stp);
471         }
472 }
473
474 static void release_openowner(struct nfs4_stateowner *sop)
475 {
476         unhash_openowner(sop);
477         list_del(&sop->so_close_lru);
478         nfs4_put_stateowner(sop);
479 }
480
481 #define SESSION_HASH_SIZE       512
482 static struct list_head sessionid_hashtbl[SESSION_HASH_SIZE];
483
484 static inline int
485 hash_sessionid(struct nfs4_sessionid *sessionid)
486 {
487         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
488
489         return sid->sequence % SESSION_HASH_SIZE;
490 }
491
492 static inline void
493 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
494 {
495         u32 *ptr = (u32 *)(&sessionid->data[0]);
496         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
497 }
498
499 static void
500 gen_sessionid(struct nfsd4_session *ses)
501 {
502         struct nfs4_client *clp = ses->se_client;
503         struct nfsd4_sessionid *sid;
504
505         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
506         sid->clientid = clp->cl_clientid;
507         sid->sequence = current_sessionid++;
508         sid->reserved = 0;
509 }
510
511 /*
512  * The protocol defines ca_maxresponssize_cached to include the size of
513  * the rpc header, but all we need to cache is the data starting after
514  * the end of the initial SEQUENCE operation--the rest we regenerate
515  * each time.  Therefore we can advertise a ca_maxresponssize_cached
516  * value that is the number of bytes in our cache plus a few additional
517  * bytes.  In order to stay on the safe side, and not promise more than
518  * we can cache, those additional bytes must be the minimum possible: 24
519  * bytes of rpc header (xid through accept state, with AUTH_NULL
520  * verifier), 12 for the compound header (with zero-length tag), and 44
521  * for the SEQUENCE op response:
522  */
523 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
524
525 static void
526 free_session_slots(struct nfsd4_session *ses)
527 {
528         int i;
529
530         for (i = 0; i < ses->se_fchannel.maxreqs; i++)
531                 kfree(ses->se_slots[i]);
532 }
533
534 /*
535  * We don't actually need to cache the rpc and session headers, so we
536  * can allocate a little less for each slot:
537  */
538 static inline int slot_bytes(struct nfsd4_channel_attrs *ca)
539 {
540         return ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
541 }
542
543 static int nfsd4_sanitize_slot_size(u32 size)
544 {
545         size -= NFSD_MIN_HDR_SEQ_SZ; /* We don't cache the rpc header */
546         size = min_t(u32, size, NFSD_SLOT_CACHE_SIZE);
547
548         return size;
549 }
550
551 /*
552  * XXX: If we run out of reserved DRC memory we could (up to a point)
553  * re-negotiate active sessions and reduce their slot usage to make
554  * rooom for new connections. For now we just fail the create session.
555  */
556 static int nfsd4_get_drc_mem(int slotsize, u32 num)
557 {
558         int avail;
559
560         num = min_t(u32, num, NFSD_MAX_SLOTS_PER_SESSION);
561
562         spin_lock(&nfsd_drc_lock);
563         avail = min_t(int, NFSD_MAX_MEM_PER_SESSION,
564                         nfsd_drc_max_mem - nfsd_drc_mem_used);
565         num = min_t(int, num, avail / slotsize);
566         nfsd_drc_mem_used += num * slotsize;
567         spin_unlock(&nfsd_drc_lock);
568
569         return num;
570 }
571
572 static void nfsd4_put_drc_mem(int slotsize, int num)
573 {
574         spin_lock(&nfsd_drc_lock);
575         nfsd_drc_mem_used -= slotsize * num;
576         spin_unlock(&nfsd_drc_lock);
577 }
578
579 static struct nfsd4_session *alloc_session(int slotsize, int numslots)
580 {
581         struct nfsd4_session *new;
582         int mem, i;
583
584         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
585                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
586         mem = numslots * sizeof(struct nfsd4_slot *);
587
588         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
589         if (!new)
590                 return NULL;
591         /* allocate each struct nfsd4_slot and data cache in one piece */
592         for (i = 0; i < numslots; i++) {
593                 mem = sizeof(struct nfsd4_slot) + slotsize;
594                 new->se_slots[i] = kzalloc(mem, GFP_KERNEL);
595                 if (!new->se_slots[i])
596                         goto out_free;
597         }
598         return new;
599 out_free:
600         while (i--)
601                 kfree(new->se_slots[i]);
602         kfree(new);
603         return NULL;
604 }
605
606 static void init_forechannel_attrs(struct nfsd4_channel_attrs *new, struct nfsd4_channel_attrs *req, int numslots, int slotsize)
607 {
608         u32 maxrpc = nfsd_serv->sv_max_mesg;
609
610         new->maxreqs = numslots;
611         new->maxresp_cached = min_t(u32, req->maxresp_cached,
612                                         slotsize + NFSD_MIN_HDR_SEQ_SZ);
613         new->maxreq_sz = min_t(u32, req->maxreq_sz, maxrpc);
614         new->maxresp_sz = min_t(u32, req->maxresp_sz, maxrpc);
615         new->maxops = min_t(u32, req->maxops, NFSD_MAX_OPS_PER_COMPOUND);
616 }
617
618 static void free_conn(struct nfsd4_conn *c)
619 {
620         svc_xprt_put(c->cn_xprt);
621         kfree(c);
622 }
623
624 static void nfsd4_conn_lost(struct svc_xpt_user *u)
625 {
626         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
627         struct nfs4_client *clp = c->cn_session->se_client;
628
629         spin_lock(&clp->cl_lock);
630         if (!list_empty(&c->cn_persession)) {
631                 list_del(&c->cn_persession);
632                 free_conn(c);
633         }
634         spin_unlock(&clp->cl_lock);
635         nfsd4_probe_callback(clp);
636 }
637
638 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
639 {
640         struct nfsd4_conn *conn;
641
642         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
643         if (!conn)
644                 return NULL;
645         svc_xprt_get(rqstp->rq_xprt);
646         conn->cn_xprt = rqstp->rq_xprt;
647         conn->cn_flags = flags;
648         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
649         return conn;
650 }
651
652 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
653 {
654         conn->cn_session = ses;
655         list_add(&conn->cn_persession, &ses->se_conns);
656 }
657
658 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
659 {
660         struct nfs4_client *clp = ses->se_client;
661
662         spin_lock(&clp->cl_lock);
663         __nfsd4_hash_conn(conn, ses);
664         spin_unlock(&clp->cl_lock);
665 }
666
667 static int nfsd4_register_conn(struct nfsd4_conn *conn)
668 {
669         conn->cn_xpt_user.callback = nfsd4_conn_lost;
670         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
671 }
672
673 static __be32 nfsd4_new_conn(struct svc_rqst *rqstp, struct nfsd4_session *ses, u32 dir)
674 {
675         struct nfsd4_conn *conn;
676         int ret;
677
678         conn = alloc_conn(rqstp, dir);
679         if (!conn)
680                 return nfserr_jukebox;
681         nfsd4_hash_conn(conn, ses);
682         ret = nfsd4_register_conn(conn);
683         if (ret)
684                 /* oops; xprt is already down: */
685                 nfsd4_conn_lost(&conn->cn_xpt_user);
686         return nfs_ok;
687 }
688
689 static __be32 nfsd4_new_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_session *ses)
690 {
691         u32 dir = NFS4_CDFC4_FORE;
692
693         if (ses->se_flags & SESSION4_BACK_CHAN)
694                 dir |= NFS4_CDFC4_BACK;
695
696         return nfsd4_new_conn(rqstp, ses, dir);
697 }
698
699 /* must be called under client_lock */
700 static void nfsd4_del_conns(struct nfsd4_session *s)
701 {
702         struct nfs4_client *clp = s->se_client;
703         struct nfsd4_conn *c;
704
705         spin_lock(&clp->cl_lock);
706         while (!list_empty(&s->se_conns)) {
707                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
708                 list_del_init(&c->cn_persession);
709                 spin_unlock(&clp->cl_lock);
710
711                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
712                 free_conn(c);
713
714                 spin_lock(&clp->cl_lock);
715         }
716         spin_unlock(&clp->cl_lock);
717 }
718
719 void free_session(struct kref *kref)
720 {
721         struct nfsd4_session *ses;
722         int mem;
723
724         ses = container_of(kref, struct nfsd4_session, se_ref);
725         nfsd4_del_conns(ses);
726         spin_lock(&nfsd_drc_lock);
727         mem = ses->se_fchannel.maxreqs * slot_bytes(&ses->se_fchannel);
728         nfsd_drc_mem_used -= mem;
729         spin_unlock(&nfsd_drc_lock);
730         free_session_slots(ses);
731         kfree(ses);
732 }
733
734 static struct nfsd4_session *alloc_init_session(struct svc_rqst *rqstp, struct nfs4_client *clp, struct nfsd4_create_session *cses)
735 {
736         struct nfsd4_session *new;
737         struct nfsd4_channel_attrs *fchan = &cses->fore_channel;
738         int numslots, slotsize;
739         int status;
740         int idx;
741
742         /*
743          * Note decreasing slot size below client's request may
744          * make it difficult for client to function correctly, whereas
745          * decreasing the number of slots will (just?) affect
746          * performance.  When short on memory we therefore prefer to
747          * decrease number of slots instead of their size.
748          */
749         slotsize = nfsd4_sanitize_slot_size(fchan->maxresp_cached);
750         numslots = nfsd4_get_drc_mem(slotsize, fchan->maxreqs);
751         if (numslots < 1)
752                 return NULL;
753
754         new = alloc_session(slotsize, numslots);
755         if (!new) {
756                 nfsd4_put_drc_mem(slotsize, fchan->maxreqs);
757                 return NULL;
758         }
759         init_forechannel_attrs(&new->se_fchannel, fchan, numslots, slotsize);
760
761         new->se_client = clp;
762         gen_sessionid(new);
763
764         INIT_LIST_HEAD(&new->se_conns);
765
766         new->se_cb_seq_nr = 1;
767         new->se_flags = cses->flags;
768         new->se_cb_prog = cses->callback_prog;
769         kref_init(&new->se_ref);
770         idx = hash_sessionid(&new->se_sessionid);
771         spin_lock(&client_lock);
772         list_add(&new->se_hash, &sessionid_hashtbl[idx]);
773         spin_lock(&clp->cl_lock);
774         list_add(&new->se_perclnt, &clp->cl_sessions);
775         spin_unlock(&clp->cl_lock);
776         spin_unlock(&client_lock);
777
778         status = nfsd4_new_conn_from_crses(rqstp, new);
779         /* whoops: benny points out, status is ignored! (err, or bogus) */
780         if (status) {
781                 free_session(&new->se_ref);
782                 return NULL;
783         }
784         if (cses->flags & SESSION4_BACK_CHAN) {
785                 struct sockaddr *sa = svc_addr(rqstp);
786                 /*
787                  * This is a little silly; with sessions there's no real
788                  * use for the callback address.  Use the peer address
789                  * as a reasonable default for now, but consider fixing
790                  * the rpc client not to require an address in the
791                  * future:
792                  */
793                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
794                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
795         }
796         nfsd4_probe_callback(clp);
797         return new;
798 }
799
800 /* caller must hold client_lock */
801 static struct nfsd4_session *
802 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid)
803 {
804         struct nfsd4_session *elem;
805         int idx;
806
807         dump_sessionid(__func__, sessionid);
808         idx = hash_sessionid(sessionid);
809         /* Search in the appropriate list */
810         list_for_each_entry(elem, &sessionid_hashtbl[idx], se_hash) {
811                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
812                             NFS4_MAX_SESSIONID_LEN)) {
813                         return elem;
814                 }
815         }
816
817         dprintk("%s: session not found\n", __func__);
818         return NULL;
819 }
820
821 /* caller must hold client_lock */
822 static void
823 unhash_session(struct nfsd4_session *ses)
824 {
825         list_del(&ses->se_hash);
826         spin_lock(&ses->se_client->cl_lock);
827         list_del(&ses->se_perclnt);
828         spin_unlock(&ses->se_client->cl_lock);
829 }
830
831 /* must be called under the client_lock */
832 static inline void
833 renew_client_locked(struct nfs4_client *clp)
834 {
835         if (is_client_expired(clp)) {
836                 dprintk("%s: client (clientid %08x/%08x) already expired\n",
837                         __func__,
838                         clp->cl_clientid.cl_boot,
839                         clp->cl_clientid.cl_id);
840                 return;
841         }
842
843         /*
844         * Move client to the end to the LRU list.
845         */
846         dprintk("renewing client (clientid %08x/%08x)\n", 
847                         clp->cl_clientid.cl_boot, 
848                         clp->cl_clientid.cl_id);
849         list_move_tail(&clp->cl_lru, &client_lru);
850         clp->cl_time = get_seconds();
851 }
852
853 static inline void
854 renew_client(struct nfs4_client *clp)
855 {
856         spin_lock(&client_lock);
857         renew_client_locked(clp);
858         spin_unlock(&client_lock);
859 }
860
861 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
862 static int
863 STALE_CLIENTID(clientid_t *clid)
864 {
865         if (clid->cl_boot == boot_time)
866                 return 0;
867         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
868                 clid->cl_boot, clid->cl_id, boot_time);
869         return 1;
870 }
871
872 /* 
873  * XXX Should we use a slab cache ?
874  * This type of memory management is somewhat inefficient, but we use it
875  * anyway since SETCLIENTID is not a common operation.
876  */
877 static struct nfs4_client *alloc_client(struct xdr_netobj name)
878 {
879         struct nfs4_client *clp;
880
881         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
882         if (clp == NULL)
883                 return NULL;
884         clp->cl_name.data = kmalloc(name.len, GFP_KERNEL);
885         if (clp->cl_name.data == NULL) {
886                 kfree(clp);
887                 return NULL;
888         }
889         memcpy(clp->cl_name.data, name.data, name.len);
890         clp->cl_name.len = name.len;
891         return clp;
892 }
893
894 static inline void
895 free_client(struct nfs4_client *clp)
896 {
897         while (!list_empty(&clp->cl_sessions)) {
898                 struct nfsd4_session *ses;
899                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
900                                 se_perclnt);
901                 list_del(&ses->se_perclnt);
902                 nfsd4_put_session(ses);
903         }
904         if (clp->cl_cred.cr_group_info)
905                 put_group_info(clp->cl_cred.cr_group_info);
906         kfree(clp->cl_principal);
907         kfree(clp->cl_name.data);
908         kfree(clp);
909 }
910
911 void
912 release_session_client(struct nfsd4_session *session)
913 {
914         struct nfs4_client *clp = session->se_client;
915
916         if (!atomic_dec_and_lock(&clp->cl_refcount, &client_lock))
917                 return;
918         if (is_client_expired(clp)) {
919                 free_client(clp);
920                 session->se_client = NULL;
921         } else
922                 renew_client_locked(clp);
923         spin_unlock(&client_lock);
924 }
925
926 /* must be called under the client_lock */
927 static inline void
928 unhash_client_locked(struct nfs4_client *clp)
929 {
930         struct nfsd4_session *ses;
931
932         mark_client_expired(clp);
933         list_del(&clp->cl_lru);
934         spin_lock(&clp->cl_lock);
935         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
936                 list_del_init(&ses->se_hash);
937         spin_unlock(&clp->cl_lock);
938 }
939
940 static void
941 expire_client(struct nfs4_client *clp)
942 {
943         struct nfs4_stateowner *sop;
944         struct nfs4_delegation *dp;
945         struct list_head reaplist;
946
947         INIT_LIST_HEAD(&reaplist);
948         spin_lock(&recall_lock);
949         while (!list_empty(&clp->cl_delegations)) {
950                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
951                 list_del_init(&dp->dl_perclnt);
952                 list_move(&dp->dl_recall_lru, &reaplist);
953         }
954         spin_unlock(&recall_lock);
955         while (!list_empty(&reaplist)) {
956                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
957                 list_del_init(&dp->dl_recall_lru);
958                 unhash_delegation(dp);
959         }
960         while (!list_empty(&clp->cl_openowners)) {
961                 sop = list_entry(clp->cl_openowners.next, struct nfs4_stateowner, so_perclient);
962                 release_openowner(sop);
963         }
964         nfsd4_shutdown_callback(clp);
965         if (clp->cl_cb_conn.cb_xprt)
966                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
967         list_del(&clp->cl_idhash);
968         list_del(&clp->cl_strhash);
969         spin_lock(&client_lock);
970         unhash_client_locked(clp);
971         if (atomic_read(&clp->cl_refcount) == 0)
972                 free_client(clp);
973         spin_unlock(&client_lock);
974 }
975
976 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
977 {
978         memcpy(target->cl_verifier.data, source->data,
979                         sizeof(target->cl_verifier.data));
980 }
981
982 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
983 {
984         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot; 
985         target->cl_clientid.cl_id = source->cl_clientid.cl_id; 
986 }
987
988 static void copy_cred(struct svc_cred *target, struct svc_cred *source)
989 {
990         target->cr_uid = source->cr_uid;
991         target->cr_gid = source->cr_gid;
992         target->cr_group_info = source->cr_group_info;
993         get_group_info(target->cr_group_info);
994 }
995
996 static int same_name(const char *n1, const char *n2)
997 {
998         return 0 == memcmp(n1, n2, HEXDIR_LEN);
999 }
1000
1001 static int
1002 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1003 {
1004         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1005 }
1006
1007 static int
1008 same_clid(clientid_t *cl1, clientid_t *cl2)
1009 {
1010         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1011 }
1012
1013 /* XXX what about NGROUP */
1014 static int
1015 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
1016 {
1017         return cr1->cr_uid == cr2->cr_uid;
1018 }
1019
1020 static void gen_clid(struct nfs4_client *clp)
1021 {
1022         static u32 current_clientid = 1;
1023
1024         clp->cl_clientid.cl_boot = boot_time;
1025         clp->cl_clientid.cl_id = current_clientid++; 
1026 }
1027
1028 static void gen_confirm(struct nfs4_client *clp)
1029 {
1030         static u32 i;
1031         u32 *p;
1032
1033         p = (u32 *)clp->cl_confirm.data;
1034         *p++ = get_seconds();
1035         *p++ = i++;
1036 }
1037
1038 static struct nfs4_client *create_client(struct xdr_netobj name, char *recdir,
1039                 struct svc_rqst *rqstp, nfs4_verifier *verf)
1040 {
1041         struct nfs4_client *clp;
1042         struct sockaddr *sa = svc_addr(rqstp);
1043         char *princ;
1044
1045         clp = alloc_client(name);
1046         if (clp == NULL)
1047                 return NULL;
1048
1049         INIT_LIST_HEAD(&clp->cl_sessions);
1050
1051         princ = svc_gss_principal(rqstp);
1052         if (princ) {
1053                 clp->cl_principal = kstrdup(princ, GFP_KERNEL);
1054                 if (clp->cl_principal == NULL) {
1055                         free_client(clp);
1056                         return NULL;
1057                 }
1058         }
1059
1060         memcpy(clp->cl_recdir, recdir, HEXDIR_LEN);
1061         atomic_set(&clp->cl_refcount, 0);
1062         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1063         INIT_LIST_HEAD(&clp->cl_idhash);
1064         INIT_LIST_HEAD(&clp->cl_strhash);
1065         INIT_LIST_HEAD(&clp->cl_openowners);
1066         INIT_LIST_HEAD(&clp->cl_delegations);
1067         INIT_LIST_HEAD(&clp->cl_lru);
1068         INIT_LIST_HEAD(&clp->cl_callbacks);
1069         spin_lock_init(&clp->cl_lock);
1070         INIT_WORK(&clp->cl_cb_null.cb_work, nfsd4_do_callback_rpc);
1071         clp->cl_time = get_seconds();
1072         clear_bit(0, &clp->cl_cb_slot_busy);
1073         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1074         copy_verf(clp, verf);
1075         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
1076         clp->cl_flavor = rqstp->rq_flavor;
1077         copy_cred(&clp->cl_cred, &rqstp->rq_cred);
1078         gen_confirm(clp);
1079         clp->cl_cb_session = NULL;
1080         return clp;
1081 }
1082
1083 static int check_name(struct xdr_netobj name)
1084 {
1085         if (name.len == 0) 
1086                 return 0;
1087         if (name.len > NFS4_OPAQUE_LIMIT) {
1088                 dprintk("NFSD: check_name: name too long(%d)!\n", name.len);
1089                 return 0;
1090         }
1091         return 1;
1092 }
1093
1094 static void
1095 add_to_unconfirmed(struct nfs4_client *clp, unsigned int strhashval)
1096 {
1097         unsigned int idhashval;
1098
1099         list_add(&clp->cl_strhash, &unconf_str_hashtbl[strhashval]);
1100         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1101         list_add(&clp->cl_idhash, &unconf_id_hashtbl[idhashval]);
1102         renew_client(clp);
1103 }
1104
1105 static void
1106 move_to_confirmed(struct nfs4_client *clp)
1107 {
1108         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
1109         unsigned int strhashval;
1110
1111         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
1112         list_move(&clp->cl_idhash, &conf_id_hashtbl[idhashval]);
1113         strhashval = clientstr_hashval(clp->cl_recdir);
1114         list_move(&clp->cl_strhash, &conf_str_hashtbl[strhashval]);
1115         renew_client(clp);
1116 }
1117
1118 static struct nfs4_client *
1119 find_confirmed_client(clientid_t *clid)
1120 {
1121         struct nfs4_client *clp;
1122         unsigned int idhashval = clientid_hashval(clid->cl_id);
1123
1124         list_for_each_entry(clp, &conf_id_hashtbl[idhashval], cl_idhash) {
1125                 if (same_clid(&clp->cl_clientid, clid))
1126                         return clp;
1127         }
1128         return NULL;
1129 }
1130
1131 static struct nfs4_client *
1132 find_unconfirmed_client(clientid_t *clid)
1133 {
1134         struct nfs4_client *clp;
1135         unsigned int idhashval = clientid_hashval(clid->cl_id);
1136
1137         list_for_each_entry(clp, &unconf_id_hashtbl[idhashval], cl_idhash) {
1138                 if (same_clid(&clp->cl_clientid, clid))
1139                         return clp;
1140         }
1141         return NULL;
1142 }
1143
1144 static bool clp_used_exchangeid(struct nfs4_client *clp)
1145 {
1146         return clp->cl_exchange_flags != 0;
1147
1148
1149 static struct nfs4_client *
1150 find_confirmed_client_by_str(const char *dname, unsigned int hashval)
1151 {
1152         struct nfs4_client *clp;
1153
1154         list_for_each_entry(clp, &conf_str_hashtbl[hashval], cl_strhash) {
1155                 if (same_name(clp->cl_recdir, dname))
1156                         return clp;
1157         }
1158         return NULL;
1159 }
1160
1161 static struct nfs4_client *
1162 find_unconfirmed_client_by_str(const char *dname, unsigned int hashval)
1163 {
1164         struct nfs4_client *clp;
1165
1166         list_for_each_entry(clp, &unconf_str_hashtbl[hashval], cl_strhash) {
1167                 if (same_name(clp->cl_recdir, dname))
1168                         return clp;
1169         }
1170         return NULL;
1171 }
1172
1173 static void rpc_svcaddr2sockaddr(struct sockaddr *sa, unsigned short family, union svc_addr_u *svcaddr)
1174 {
1175         switch (family) {
1176         case AF_INET:
1177                 ((struct sockaddr_in *)sa)->sin_family = AF_INET;
1178                 ((struct sockaddr_in *)sa)->sin_addr = svcaddr->addr;
1179                 return;
1180         case AF_INET6:
1181                 ((struct sockaddr_in6 *)sa)->sin6_family = AF_INET6;
1182                 ((struct sockaddr_in6 *)sa)->sin6_addr = svcaddr->addr6;
1183                 return;
1184         }
1185 }
1186
1187 static void
1188 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
1189 {
1190         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
1191         struct sockaddr *sa = svc_addr(rqstp);
1192         u32 scopeid = rpc_get_scope_id(sa);
1193         unsigned short expected_family;
1194
1195         /* Currently, we only support tcp and tcp6 for the callback channel */
1196         if (se->se_callback_netid_len == 3 &&
1197             !memcmp(se->se_callback_netid_val, "tcp", 3))
1198                 expected_family = AF_INET;
1199         else if (se->se_callback_netid_len == 4 &&
1200                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
1201                 expected_family = AF_INET6;
1202         else
1203                 goto out_err;
1204
1205         conn->cb_addrlen = rpc_uaddr2sockaddr(se->se_callback_addr_val,
1206                                             se->se_callback_addr_len,
1207                                             (struct sockaddr *)&conn->cb_addr,
1208                                             sizeof(conn->cb_addr));
1209
1210         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
1211                 goto out_err;
1212
1213         if (conn->cb_addr.ss_family == AF_INET6)
1214                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
1215
1216         conn->cb_prog = se->se_callback_prog;
1217         conn->cb_ident = se->se_callback_ident;
1218         rpc_svcaddr2sockaddr((struct sockaddr *)&conn->cb_saddr, expected_family, &rqstp->rq_daddr);
1219         return;
1220 out_err:
1221         conn->cb_addr.ss_family = AF_UNSPEC;
1222         conn->cb_addrlen = 0;
1223         dprintk(KERN_INFO "NFSD: this client (clientid %08x/%08x) "
1224                 "will not receive delegations\n",
1225                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
1226
1227         return;
1228 }
1229
1230 /*
1231  * Cache a reply. nfsd4_check_drc_limit() has bounded the cache size.
1232  */
1233 void
1234 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
1235 {
1236         struct nfsd4_slot *slot = resp->cstate.slot;
1237         unsigned int base;
1238
1239         dprintk("--> %s slot %p\n", __func__, slot);
1240
1241         slot->sl_opcnt = resp->opcnt;
1242         slot->sl_status = resp->cstate.status;
1243
1244         if (nfsd4_not_cached(resp)) {
1245                 slot->sl_datalen = 0;
1246                 return;
1247         }
1248         slot->sl_datalen = (char *)resp->p - (char *)resp->cstate.datap;
1249         base = (char *)resp->cstate.datap -
1250                                         (char *)resp->xbuf->head[0].iov_base;
1251         if (read_bytes_from_xdr_buf(resp->xbuf, base, slot->sl_data,
1252                                     slot->sl_datalen))
1253                 WARN("%s: sessions DRC could not cache compound\n", __func__);
1254         return;
1255 }
1256
1257 /*
1258  * Encode the replay sequence operation from the slot values.
1259  * If cachethis is FALSE encode the uncached rep error on the next
1260  * operation which sets resp->p and increments resp->opcnt for
1261  * nfs4svc_encode_compoundres.
1262  *
1263  */
1264 static __be32
1265 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
1266                           struct nfsd4_compoundres *resp)
1267 {
1268         struct nfsd4_op *op;
1269         struct nfsd4_slot *slot = resp->cstate.slot;
1270
1271         dprintk("--> %s resp->opcnt %d cachethis %u \n", __func__,
1272                 resp->opcnt, resp->cstate.slot->sl_cachethis);
1273
1274         /* Encode the replayed sequence operation */
1275         op = &args->ops[resp->opcnt - 1];
1276         nfsd4_encode_operation(resp, op);
1277
1278         /* Return nfserr_retry_uncached_rep in next operation. */
1279         if (args->opcnt > 1 && slot->sl_cachethis == 0) {
1280                 op = &args->ops[resp->opcnt++];
1281                 op->status = nfserr_retry_uncached_rep;
1282                 nfsd4_encode_operation(resp, op);
1283         }
1284         return op->status;
1285 }
1286
1287 /*
1288  * The sequence operation is not cached because we can use the slot and
1289  * session values.
1290  */
1291 __be32
1292 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
1293                          struct nfsd4_sequence *seq)
1294 {
1295         struct nfsd4_slot *slot = resp->cstate.slot;
1296         __be32 status;
1297
1298         dprintk("--> %s slot %p\n", __func__, slot);
1299
1300         /* Either returns 0 or nfserr_retry_uncached */
1301         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
1302         if (status == nfserr_retry_uncached_rep)
1303                 return status;
1304
1305         /* The sequence operation has been encoded, cstate->datap set. */
1306         memcpy(resp->cstate.datap, slot->sl_data, slot->sl_datalen);
1307
1308         resp->opcnt = slot->sl_opcnt;
1309         resp->p = resp->cstate.datap + XDR_QUADLEN(slot->sl_datalen);
1310         status = slot->sl_status;
1311
1312         return status;
1313 }
1314
1315 /*
1316  * Set the exchange_id flags returned by the server.
1317  */
1318 static void
1319 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
1320 {
1321         /* pNFS is not supported */
1322         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
1323
1324         /* Referrals are supported, Migration is not. */
1325         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
1326
1327         /* set the wire flags to return to client. */
1328         clid->flags = new->cl_exchange_flags;
1329 }
1330
1331 __be32
1332 nfsd4_exchange_id(struct svc_rqst *rqstp,
1333                   struct nfsd4_compound_state *cstate,
1334                   struct nfsd4_exchange_id *exid)
1335 {
1336         struct nfs4_client *unconf, *conf, *new;
1337         int status;
1338         unsigned int            strhashval;
1339         char                    dname[HEXDIR_LEN];
1340         char                    addr_str[INET6_ADDRSTRLEN];
1341         nfs4_verifier           verf = exid->verifier;
1342         struct sockaddr         *sa = svc_addr(rqstp);
1343
1344         rpc_ntop(sa, addr_str, sizeof(addr_str));
1345         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
1346                 "ip_addr=%s flags %x, spa_how %d\n",
1347                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
1348                 addr_str, exid->flags, exid->spa_how);
1349
1350         if (!check_name(exid->clname) || (exid->flags & ~EXCHGID4_FLAG_MASK_A))
1351                 return nfserr_inval;
1352
1353         /* Currently only support SP4_NONE */
1354         switch (exid->spa_how) {
1355         case SP4_NONE:
1356                 break;
1357         case SP4_SSV:
1358                 return nfserr_serverfault;
1359         default:
1360                 BUG();                          /* checked by xdr code */
1361         case SP4_MACH_CRED:
1362                 return nfserr_serverfault;      /* no excuse :-/ */
1363         }
1364
1365         status = nfs4_make_rec_clidname(dname, &exid->clname);
1366
1367         if (status)
1368                 goto error;
1369
1370         strhashval = clientstr_hashval(dname);
1371
1372         nfs4_lock_state();
1373         status = nfs_ok;
1374
1375         conf = find_confirmed_client_by_str(dname, strhashval);
1376         if (conf) {
1377                 if (!clp_used_exchangeid(conf)) {
1378                         status = nfserr_clid_inuse; /* XXX: ? */
1379                         goto out;
1380                 }
1381                 if (!same_verf(&verf, &conf->cl_verifier)) {
1382                         /* 18.35.4 case 8 */
1383                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1384                                 status = nfserr_not_same;
1385                                 goto out;
1386                         }
1387                         /* Client reboot: destroy old state */
1388                         expire_client(conf);
1389                         goto out_new;
1390                 }
1391                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1392                         /* 18.35.4 case 9 */
1393                         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1394                                 status = nfserr_perm;
1395                                 goto out;
1396                         }
1397                         expire_client(conf);
1398                         goto out_new;
1399                 }
1400                 /*
1401                  * Set bit when the owner id and verifier map to an already
1402                  * confirmed client id (18.35.3).
1403                  */
1404                 exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
1405
1406                 /*
1407                  * Falling into 18.35.4 case 2, possible router replay.
1408                  * Leave confirmed record intact and return same result.
1409                  */
1410                 copy_verf(conf, &verf);
1411                 new = conf;
1412                 goto out_copy;
1413         }
1414
1415         /* 18.35.4 case 7 */
1416         if (exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A) {
1417                 status = nfserr_noent;
1418                 goto out;
1419         }
1420
1421         unconf  = find_unconfirmed_client_by_str(dname, strhashval);
1422         if (unconf) {
1423                 /*
1424                  * Possible retry or client restart.  Per 18.35.4 case 4,
1425                  * a new unconfirmed record should be generated regardless
1426                  * of whether any properties have changed.
1427                  */
1428                 expire_client(unconf);
1429         }
1430
1431 out_new:
1432         /* Normal case */
1433         new = create_client(exid->clname, dname, rqstp, &verf);
1434         if (new == NULL) {
1435                 status = nfserr_jukebox;
1436                 goto out;
1437         }
1438
1439         gen_clid(new);
1440         add_to_unconfirmed(new, strhashval);
1441 out_copy:
1442         exid->clientid.cl_boot = new->cl_clientid.cl_boot;
1443         exid->clientid.cl_id = new->cl_clientid.cl_id;
1444
1445         exid->seqid = 1;
1446         nfsd4_set_ex_flags(new, exid);
1447
1448         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
1449                 new->cl_cs_slot.sl_seqid, new->cl_exchange_flags);
1450         status = nfs_ok;
1451
1452 out:
1453         nfs4_unlock_state();
1454 error:
1455         dprintk("nfsd4_exchange_id returns %d\n", ntohl(status));
1456         return status;
1457 }
1458
1459 static int
1460 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
1461 {
1462         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
1463                 slot_seqid);
1464
1465         /* The slot is in use, and no response has been sent. */
1466         if (slot_inuse) {
1467                 if (seqid == slot_seqid)
1468                         return nfserr_jukebox;
1469                 else
1470                         return nfserr_seq_misordered;
1471         }
1472         /* Normal */
1473         if (likely(seqid == slot_seqid + 1))
1474                 return nfs_ok;
1475         /* Replay */
1476         if (seqid == slot_seqid)
1477                 return nfserr_replay_cache;
1478         /* Wraparound */
1479         if (seqid == 1 && (slot_seqid + 1) == 0)
1480                 return nfs_ok;
1481         /* Misordered replay or misordered new request */
1482         return nfserr_seq_misordered;
1483 }
1484
1485 /*
1486  * Cache the create session result into the create session single DRC
1487  * slot cache by saving the xdr structure. sl_seqid has been set.
1488  * Do this for solo or embedded create session operations.
1489  */
1490 static void
1491 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
1492                            struct nfsd4_clid_slot *slot, int nfserr)
1493 {
1494         slot->sl_status = nfserr;
1495         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
1496 }
1497
1498 static __be32
1499 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
1500                             struct nfsd4_clid_slot *slot)
1501 {
1502         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
1503         return slot->sl_status;
1504 }
1505
1506 __be32
1507 nfsd4_create_session(struct svc_rqst *rqstp,
1508                      struct nfsd4_compound_state *cstate,
1509                      struct nfsd4_create_session *cr_ses)
1510 {
1511         struct sockaddr *sa = svc_addr(rqstp);
1512         struct nfs4_client *conf, *unconf;
1513         struct nfsd4_session *new;
1514         struct nfsd4_clid_slot *cs_slot = NULL;
1515         bool confirm_me = false;
1516         int status = 0;
1517
1518         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
1519                 return nfserr_inval;
1520
1521         nfs4_lock_state();
1522         unconf = find_unconfirmed_client(&cr_ses->clientid);
1523         conf = find_confirmed_client(&cr_ses->clientid);
1524
1525         if (conf) {
1526                 cs_slot = &conf->cl_cs_slot;
1527                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1528                 if (status == nfserr_replay_cache) {
1529                         dprintk("Got a create_session replay! seqid= %d\n",
1530                                 cs_slot->sl_seqid);
1531                         /* Return the cached reply status */
1532                         status = nfsd4_replay_create_session(cr_ses, cs_slot);
1533                         goto out;
1534                 } else if (cr_ses->seqid != cs_slot->sl_seqid + 1) {
1535                         status = nfserr_seq_misordered;
1536                         dprintk("Sequence misordered!\n");
1537                         dprintk("Expected seqid= %d but got seqid= %d\n",
1538                                 cs_slot->sl_seqid, cr_ses->seqid);
1539                         goto out;
1540                 }
1541         } else if (unconf) {
1542                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
1543                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
1544                         status = nfserr_clid_inuse;
1545                         goto out;
1546                 }
1547
1548                 cs_slot = &unconf->cl_cs_slot;
1549                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
1550                 if (status) {
1551                         /* an unconfirmed replay returns misordered */
1552                         status = nfserr_seq_misordered;
1553                         goto out;
1554                 }
1555
1556                 confirm_me = true;
1557                 conf = unconf;
1558         } else {
1559                 status = nfserr_stale_clientid;
1560                 goto out;
1561         }
1562
1563         /*
1564          * XXX: we should probably set this at creation time, and check
1565          * for consistent minorversion use throughout:
1566          */
1567         conf->cl_minorversion = 1;
1568         /*
1569          * We do not support RDMA or persistent sessions
1570          */
1571         cr_ses->flags &= ~SESSION4_PERSIST;
1572         cr_ses->flags &= ~SESSION4_RDMA;
1573
1574         status = nfserr_jukebox;
1575         new = alloc_init_session(rqstp, conf, cr_ses);
1576         if (!new)
1577                 goto out;
1578         status = nfs_ok;
1579         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
1580                NFS4_MAX_SESSIONID_LEN);
1581         memcpy(&cr_ses->fore_channel, &new->se_fchannel,
1582                 sizeof(struct nfsd4_channel_attrs));
1583         cs_slot->sl_seqid++;
1584         cr_ses->seqid = cs_slot->sl_seqid;
1585
1586         /* cache solo and embedded create sessions under the state lock */
1587         nfsd4_cache_create_session(cr_ses, cs_slot, status);
1588         if (confirm_me)
1589                 move_to_confirmed(conf);
1590 out:
1591         nfs4_unlock_state();
1592         dprintk("%s returns %d\n", __func__, ntohl(status));
1593         return status;
1594 }
1595
1596 static bool nfsd4_last_compound_op(struct svc_rqst *rqstp)
1597 {
1598         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1599         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
1600
1601         return argp->opcnt == resp->opcnt;
1602 }
1603
1604 static __be32 nfsd4_map_bcts_dir(u32 *dir)
1605 {
1606         switch (*dir) {
1607         case NFS4_CDFC4_FORE:
1608         case NFS4_CDFC4_BACK:
1609                 return nfs_ok;
1610         case NFS4_CDFC4_FORE_OR_BOTH:
1611         case NFS4_CDFC4_BACK_OR_BOTH:
1612                 *dir = NFS4_CDFC4_BOTH;
1613                 return nfs_ok;
1614         };
1615         return nfserr_inval;
1616 }
1617
1618 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
1619                      struct nfsd4_compound_state *cstate,
1620                      struct nfsd4_bind_conn_to_session *bcts)
1621 {
1622         __be32 status;
1623
1624         if (!nfsd4_last_compound_op(rqstp))
1625                 return nfserr_not_only_op;
1626         spin_lock(&client_lock);
1627         cstate->session = find_in_sessionid_hashtbl(&bcts->sessionid);
1628         /* Sorta weird: we only need the refcnt'ing because new_conn acquires
1629          * client_lock iself: */
1630         if (cstate->session) {
1631                 nfsd4_get_session(cstate->session);
1632                 atomic_inc(&cstate->session->se_client->cl_refcount);
1633         }
1634         spin_unlock(&client_lock);
1635         if (!cstate->session)
1636                 return nfserr_badsession;
1637
1638         status = nfsd4_map_bcts_dir(&bcts->dir);
1639         if (!status)
1640                 nfsd4_new_conn(rqstp, cstate->session, bcts->dir);
1641         return status;
1642 }
1643
1644 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
1645 {
1646         if (!session)
1647                 return 0;
1648         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
1649 }
1650
1651 __be32
1652 nfsd4_destroy_session(struct svc_rqst *r,
1653                       struct nfsd4_compound_state *cstate,
1654                       struct nfsd4_destroy_session *sessionid)
1655 {
1656         struct nfsd4_session *ses;
1657         u32 status = nfserr_badsession;
1658
1659         /* Notes:
1660          * - The confirmed nfs4_client->cl_sessionid holds destroyed sessinid
1661          * - Should we return nfserr_back_chan_busy if waiting for
1662          *   callbacks on to-be-destroyed session?
1663          * - Do we need to clear any callback info from previous session?
1664          */
1665
1666         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
1667                 if (!nfsd4_last_compound_op(r))
1668                         return nfserr_not_only_op;
1669         }
1670         dump_sessionid(__func__, &sessionid->sessionid);
1671         spin_lock(&client_lock);
1672         ses = find_in_sessionid_hashtbl(&sessionid->sessionid);
1673         if (!ses) {
1674                 spin_unlock(&client_lock);
1675                 goto out;
1676         }
1677
1678         unhash_session(ses);
1679         spin_unlock(&client_lock);
1680
1681         nfs4_lock_state();
1682         nfsd4_probe_callback_sync(ses->se_client);
1683         nfs4_unlock_state();
1684
1685         nfsd4_del_conns(ses);
1686
1687         nfsd4_put_session(ses);
1688         status = nfs_ok;
1689 out:
1690         dprintk("%s returns %d\n", __func__, ntohl(status));
1691         return status;
1692 }
1693
1694 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
1695 {
1696         struct nfsd4_conn *c;
1697
1698         list_for_each_entry(c, &s->se_conns, cn_persession) {
1699                 if (c->cn_xprt == xpt) {
1700                         return c;
1701                 }
1702         }
1703         return NULL;
1704 }
1705
1706 static void nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
1707 {
1708         struct nfs4_client *clp = ses->se_client;
1709         struct nfsd4_conn *c;
1710         int ret;
1711
1712         spin_lock(&clp->cl_lock);
1713         c = __nfsd4_find_conn(new->cn_xprt, ses);
1714         if (c) {
1715                 spin_unlock(&clp->cl_lock);
1716                 free_conn(new);
1717                 return;
1718         }
1719         __nfsd4_hash_conn(new, ses);
1720         spin_unlock(&clp->cl_lock);
1721         ret = nfsd4_register_conn(new);
1722         if (ret)
1723                 /* oops; xprt is already down: */
1724                 nfsd4_conn_lost(&new->cn_xpt_user);
1725         return;
1726 }
1727
1728 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
1729 {
1730         struct nfsd4_compoundargs *args = rqstp->rq_argp;
1731
1732         return args->opcnt > session->se_fchannel.maxops;
1733 }
1734
1735 __be32
1736 nfsd4_sequence(struct svc_rqst *rqstp,
1737                struct nfsd4_compound_state *cstate,
1738                struct nfsd4_sequence *seq)
1739 {
1740         struct nfsd4_compoundres *resp = rqstp->rq_resp;
1741         struct nfsd4_session *session;
1742         struct nfsd4_slot *slot;
1743         struct nfsd4_conn *conn;
1744         int status;
1745
1746         if (resp->opcnt != 1)
1747                 return nfserr_sequence_pos;
1748
1749         /*
1750          * Will be either used or freed by nfsd4_sequence_check_conn
1751          * below.
1752          */
1753         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
1754         if (!conn)
1755                 return nfserr_jukebox;
1756
1757         spin_lock(&client_lock);
1758         status = nfserr_badsession;
1759         session = find_in_sessionid_hashtbl(&seq->sessionid);
1760         if (!session)
1761                 goto out;
1762
1763         status = nfserr_too_many_ops;
1764         if (nfsd4_session_too_many_ops(rqstp, session))
1765                 goto out;
1766
1767         status = nfserr_badslot;
1768         if (seq->slotid >= session->se_fchannel.maxreqs)
1769                 goto out;
1770
1771         slot = session->se_slots[seq->slotid];
1772         dprintk("%s: slotid %d\n", __func__, seq->slotid);
1773
1774         /* We do not negotiate the number of slots yet, so set the
1775          * maxslots to the session maxreqs which is used to encode
1776          * sr_highest_slotid and the sr_target_slot id to maxslots */
1777         seq->maxslots = session->se_fchannel.maxreqs;
1778
1779         status = check_slot_seqid(seq->seqid, slot->sl_seqid, slot->sl_inuse);
1780         if (status == nfserr_replay_cache) {
1781                 cstate->slot = slot;
1782                 cstate->session = session;
1783                 /* Return the cached reply status and set cstate->status
1784                  * for nfsd4_proc_compound processing */
1785                 status = nfsd4_replay_cache_entry(resp, seq);
1786                 cstate->status = nfserr_replay_cache;
1787                 goto out;
1788         }
1789         if (status)
1790                 goto out;
1791
1792         nfsd4_sequence_check_conn(conn, session);
1793         conn = NULL;
1794
1795         /* Success! bump slot seqid */
1796         slot->sl_inuse = true;
1797         slot->sl_seqid = seq->seqid;
1798         slot->sl_cachethis = seq->cachethis;
1799
1800         cstate->slot = slot;
1801         cstate->session = session;
1802
1803 out:
1804         /* Hold a session reference until done processing the compound. */
1805         if (cstate->session) {
1806                 struct nfs4_client *clp = session->se_client;
1807
1808                 nfsd4_get_session(cstate->session);
1809                 atomic_inc(&clp->cl_refcount);
1810                 if (clp->cl_cb_state == NFSD4_CB_DOWN)
1811                         seq->status_flags |= SEQ4_STATUS_CB_PATH_DOWN;
1812         }
1813         kfree(conn);
1814         spin_unlock(&client_lock);
1815         dprintk("%s: return %d\n", __func__, ntohl(status));
1816         return status;
1817 }
1818
1819 __be32
1820 nfsd4_reclaim_complete(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate, struct nfsd4_reclaim_complete *rc)
1821 {
1822         int status = 0;
1823
1824         if (rc->rca_one_fs) {
1825                 if (!cstate->current_fh.fh_dentry)
1826                         return nfserr_nofilehandle;
1827                 /*
1828                  * We don't take advantage of the rca_one_fs case.
1829                  * That's OK, it's optional, we can safely ignore it.
1830                  */
1831                  return nfs_ok;
1832         }
1833
1834         nfs4_lock_state();
1835         status = nfserr_complete_already;
1836         if (cstate->session->se_client->cl_firststate)
1837                 goto out;
1838
1839         status = nfserr_stale_clientid;
1840         if (is_client_expired(cstate->session->se_client))
1841                 /*
1842                  * The following error isn't really legal.
1843                  * But we only get here if the client just explicitly
1844                  * destroyed the client.  Surely it no longer cares what
1845                  * error it gets back on an operation for the dead
1846                  * client.
1847                  */
1848                 goto out;
1849
1850         status = nfs_ok;
1851         nfsd4_create_clid_dir(cstate->session->se_client);
1852 out:
1853         nfs4_unlock_state();
1854         return status;
1855 }
1856
1857 __be32
1858 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
1859                   struct nfsd4_setclientid *setclid)
1860 {
1861         struct xdr_netobj       clname = { 
1862                 .len = setclid->se_namelen,
1863                 .data = setclid->se_name,
1864         };
1865         nfs4_verifier           clverifier = setclid->se_verf;
1866         unsigned int            strhashval;
1867         struct nfs4_client      *conf, *unconf, *new;
1868         __be32                  status;
1869         char                    dname[HEXDIR_LEN];
1870         
1871         if (!check_name(clname))
1872                 return nfserr_inval;
1873
1874         status = nfs4_make_rec_clidname(dname, &clname);
1875         if (status)
1876                 return status;
1877
1878         /* 
1879          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1880          * We get here on a DRC miss.
1881          */
1882
1883         strhashval = clientstr_hashval(dname);
1884
1885         nfs4_lock_state();
1886         conf = find_confirmed_client_by_str(dname, strhashval);
1887         if (conf) {
1888                 /* RFC 3530 14.2.33 CASE 0: */
1889                 status = nfserr_clid_inuse;
1890                 if (clp_used_exchangeid(conf))
1891                         goto out;
1892                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
1893                         char addr_str[INET6_ADDRSTRLEN];
1894                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
1895                                  sizeof(addr_str));
1896                         dprintk("NFSD: setclientid: string in use by client "
1897                                 "at %s\n", addr_str);
1898                         goto out;
1899                 }
1900         }
1901         /*
1902          * section 14.2.33 of RFC 3530 (under the heading "IMPLEMENTATION")
1903          * has a description of SETCLIENTID request processing consisting
1904          * of 5 bullet points, labeled as CASE0 - CASE4 below.
1905          */
1906         unconf = find_unconfirmed_client_by_str(dname, strhashval);
1907         status = nfserr_jukebox;
1908         if (!conf) {
1909                 /*
1910                  * RFC 3530 14.2.33 CASE 4:
1911                  * placed first, because it is the normal case
1912                  */
1913                 if (unconf)
1914                         expire_client(unconf);
1915                 new = create_client(clname, dname, rqstp, &clverifier);
1916                 if (new == NULL)
1917                         goto out;
1918                 gen_clid(new);
1919         } else if (same_verf(&conf->cl_verifier, &clverifier)) {
1920                 /*
1921                  * RFC 3530 14.2.33 CASE 1:
1922                  * probable callback update
1923                  */
1924                 if (unconf) {
1925                         /* Note this is removing unconfirmed {*x***},
1926                          * which is stronger than RFC recommended {vxc**}.
1927                          * This has the advantage that there is at most
1928                          * one {*x***} in either list at any time.
1929                          */
1930                         expire_client(unconf);
1931                 }
1932                 new = create_client(clname, dname, rqstp, &clverifier);
1933                 if (new == NULL)
1934                         goto out;
1935                 copy_clid(new, conf);
1936         } else if (!unconf) {
1937                 /*
1938                  * RFC 3530 14.2.33 CASE 2:
1939                  * probable client reboot; state will be removed if
1940                  * confirmed.
1941                  */
1942                 new = create_client(clname, dname, rqstp, &clverifier);
1943                 if (new == NULL)
1944                         goto out;
1945                 gen_clid(new);
1946         } else {
1947                 /*
1948                  * RFC 3530 14.2.33 CASE 3:
1949                  * probable client reboot; state will be removed if
1950                  * confirmed.
1951                  */
1952                 expire_client(unconf);
1953                 new = create_client(clname, dname, rqstp, &clverifier);
1954                 if (new == NULL)
1955                         goto out;
1956                 gen_clid(new);
1957         }
1958         /*
1959          * XXX: we should probably set this at creation time, and check
1960          * for consistent minorversion use throughout:
1961          */
1962         new->cl_minorversion = 0;
1963         gen_callback(new, setclid, rqstp);
1964         add_to_unconfirmed(new, strhashval);
1965         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
1966         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
1967         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
1968         status = nfs_ok;
1969 out:
1970         nfs4_unlock_state();
1971         return status;
1972 }
1973
1974
1975 /*
1976  * Section 14.2.34 of RFC 3530 (under the heading "IMPLEMENTATION") has
1977  * a description of SETCLIENTID_CONFIRM request processing consisting of 4
1978  * bullets, labeled as CASE1 - CASE4 below.
1979  */
1980 __be32
1981 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
1982                          struct nfsd4_compound_state *cstate,
1983                          struct nfsd4_setclientid_confirm *setclientid_confirm)
1984 {
1985         struct sockaddr *sa = svc_addr(rqstp);
1986         struct nfs4_client *conf, *unconf;
1987         nfs4_verifier confirm = setclientid_confirm->sc_confirm; 
1988         clientid_t * clid = &setclientid_confirm->sc_clientid;
1989         __be32 status;
1990
1991         if (STALE_CLIENTID(clid))
1992                 return nfserr_stale_clientid;
1993         /* 
1994          * XXX The Duplicate Request Cache (DRC) has been checked (??)
1995          * We get here on a DRC miss.
1996          */
1997
1998         nfs4_lock_state();
1999
2000         conf = find_confirmed_client(clid);
2001         unconf = find_unconfirmed_client(clid);
2002
2003         status = nfserr_clid_inuse;
2004         if (conf && !rpc_cmp_addr((struct sockaddr *) &conf->cl_addr, sa))
2005                 goto out;
2006         if (unconf && !rpc_cmp_addr((struct sockaddr *) &unconf->cl_addr, sa))
2007                 goto out;
2008
2009         /*
2010          * section 14.2.34 of RFC 3530 has a description of
2011          * SETCLIENTID_CONFIRM request processing consisting
2012          * of 4 bullet points, labeled as CASE1 - CASE4 below.
2013          */
2014         if (conf && unconf && same_verf(&confirm, &unconf->cl_confirm)) {
2015                 /*
2016                  * RFC 3530 14.2.34 CASE 1:
2017                  * callback update
2018                  */
2019                 if (!same_creds(&conf->cl_cred, &unconf->cl_cred))
2020                         status = nfserr_clid_inuse;
2021                 else {
2022                         nfsd4_change_callback(conf, &unconf->cl_cb_conn);
2023                         nfsd4_probe_callback(conf);
2024                         expire_client(unconf);
2025                         status = nfs_ok;
2026
2027                 }
2028         } else if (conf && !unconf) {
2029                 /*
2030                  * RFC 3530 14.2.34 CASE 2:
2031                  * probable retransmitted request; play it safe and
2032                  * do nothing.
2033                  */
2034                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred))
2035                         status = nfserr_clid_inuse;
2036                 else
2037                         status = nfs_ok;
2038         } else if (!conf && unconf
2039                         && same_verf(&unconf->cl_confirm, &confirm)) {
2040                 /*
2041                  * RFC 3530 14.2.34 CASE 3:
2042                  * Normal case; new or rebooted client:
2043                  */
2044                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred)) {
2045                         status = nfserr_clid_inuse;
2046                 } else {
2047                         unsigned int hash =
2048                                 clientstr_hashval(unconf->cl_recdir);
2049                         conf = find_confirmed_client_by_str(unconf->cl_recdir,
2050                                                             hash);
2051                         if (conf) {
2052                                 nfsd4_remove_clid_dir(conf);
2053                                 expire_client(conf);
2054                         }
2055                         move_to_confirmed(unconf);
2056                         conf = unconf;
2057                         nfsd4_probe_callback(conf);
2058                         status = nfs_ok;
2059                 }
2060         } else if ((!conf || (conf && !same_verf(&conf->cl_confirm, &confirm)))
2061             && (!unconf || (unconf && !same_verf(&unconf->cl_confirm,
2062                                                                 &confirm)))) {
2063                 /*
2064                  * RFC 3530 14.2.34 CASE 4:
2065                  * Client probably hasn't noticed that we rebooted yet.
2066                  */
2067                 status = nfserr_stale_clientid;
2068         } else {
2069                 /* check that we have hit one of the cases...*/
2070                 status = nfserr_clid_inuse;
2071         }
2072 out:
2073         nfs4_unlock_state();
2074         return status;
2075 }
2076
2077 /* OPEN Share state helper functions */
2078 static inline struct nfs4_file *
2079 alloc_init_file(struct inode *ino)
2080 {
2081         struct nfs4_file *fp;
2082         unsigned int hashval = file_hashval(ino);
2083
2084         fp = kmem_cache_alloc(file_slab, GFP_KERNEL);
2085         if (fp) {
2086                 atomic_set(&fp->fi_ref, 1);
2087                 INIT_LIST_HEAD(&fp->fi_hash);
2088                 INIT_LIST_HEAD(&fp->fi_stateids);
2089                 INIT_LIST_HEAD(&fp->fi_delegations);
2090                 fp->fi_inode = igrab(ino);
2091                 fp->fi_id = current_fileid++;
2092                 fp->fi_had_conflict = false;
2093                 fp->fi_lease = NULL;
2094                 memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
2095                 memset(fp->fi_access, 0, sizeof(fp->fi_access));
2096                 spin_lock(&recall_lock);
2097                 list_add(&fp->fi_hash, &file_hashtbl[hashval]);
2098                 spin_unlock(&recall_lock);
2099                 return fp;
2100         }
2101         return NULL;
2102 }
2103
2104 static void
2105 nfsd4_free_slab(struct kmem_cache **slab)
2106 {
2107         if (*slab == NULL)
2108                 return;
2109         kmem_cache_destroy(*slab);
2110         *slab = NULL;
2111 }
2112
2113 void
2114 nfsd4_free_slabs(void)
2115 {
2116         nfsd4_free_slab(&stateowner_slab);
2117         nfsd4_free_slab(&file_slab);
2118         nfsd4_free_slab(&stateid_slab);
2119         nfsd4_free_slab(&deleg_slab);
2120 }
2121
2122 static int
2123 nfsd4_init_slabs(void)
2124 {
2125         stateowner_slab = kmem_cache_create("nfsd4_stateowners",
2126                         sizeof(struct nfs4_stateowner), 0, 0, NULL);
2127         if (stateowner_slab == NULL)
2128                 goto out_nomem;
2129         file_slab = kmem_cache_create("nfsd4_files",
2130                         sizeof(struct nfs4_file), 0, 0, NULL);
2131         if (file_slab == NULL)
2132                 goto out_nomem;
2133         stateid_slab = kmem_cache_create("nfsd4_stateids",
2134                         sizeof(struct nfs4_stateid), 0, 0, NULL);
2135         if (stateid_slab == NULL)
2136                 goto out_nomem;
2137         deleg_slab = kmem_cache_create("nfsd4_delegations",
2138                         sizeof(struct nfs4_delegation), 0, 0, NULL);
2139         if (deleg_slab == NULL)
2140                 goto out_nomem;
2141         return 0;
2142 out_nomem:
2143         nfsd4_free_slabs();
2144         dprintk("nfsd4: out of memory while initializing nfsv4\n");
2145         return -ENOMEM;
2146 }
2147
2148 void
2149 nfs4_free_stateowner(struct kref *kref)
2150 {
2151         struct nfs4_stateowner *sop =
2152                 container_of(kref, struct nfs4_stateowner, so_ref);
2153         kfree(sop->so_owner.data);
2154         kmem_cache_free(stateowner_slab, sop);
2155 }
2156
2157 static inline struct nfs4_stateowner *
2158 alloc_stateowner(struct xdr_netobj *owner)
2159 {
2160         struct nfs4_stateowner *sop;
2161
2162         if ((sop = kmem_cache_alloc(stateowner_slab, GFP_KERNEL))) {
2163                 if ((sop->so_owner.data = kmalloc(owner->len, GFP_KERNEL))) {
2164                         memcpy(sop->so_owner.data, owner->data, owner->len);
2165                         sop->so_owner.len = owner->len;
2166                         kref_init(&sop->so_ref);
2167                         return sop;
2168                 } 
2169                 kmem_cache_free(stateowner_slab, sop);
2170         }
2171         return NULL;
2172 }
2173
2174 static struct nfs4_stateowner *
2175 alloc_init_open_stateowner(unsigned int strhashval, struct nfs4_client *clp, struct nfsd4_open *open) {
2176         struct nfs4_stateowner *sop;
2177         struct nfs4_replay *rp;
2178         unsigned int idhashval;
2179
2180         if (!(sop = alloc_stateowner(&open->op_owner)))
2181                 return NULL;
2182         idhashval = ownerid_hashval(current_ownerid);
2183         INIT_LIST_HEAD(&sop->so_idhash);
2184         INIT_LIST_HEAD(&sop->so_strhash);
2185         INIT_LIST_HEAD(&sop->so_perclient);
2186         INIT_LIST_HEAD(&sop->so_stateids);
2187         INIT_LIST_HEAD(&sop->so_perstateid);  /* not used */
2188         INIT_LIST_HEAD(&sop->so_close_lru);
2189         sop->so_time = 0;
2190         list_add(&sop->so_idhash, &ownerid_hashtbl[idhashval]);
2191         list_add(&sop->so_strhash, &ownerstr_hashtbl[strhashval]);
2192         list_add(&sop->so_perclient, &clp->cl_openowners);
2193         sop->so_is_open_owner = 1;
2194         sop->so_id = current_ownerid++;
2195         sop->so_client = clp;
2196         sop->so_seqid = open->op_seqid;
2197         sop->so_confirmed = 0;
2198         rp = &sop->so_replay;
2199         rp->rp_status = nfserr_serverfault;
2200         rp->rp_buflen = 0;
2201         rp->rp_buf = rp->rp_ibuf;
2202         return sop;
2203 }
2204
2205 static inline void
2206 init_stateid(struct nfs4_stateid *stp, struct nfs4_file *fp, struct nfsd4_open *open) {
2207         struct nfs4_stateowner *sop = open->op_stateowner;
2208         unsigned int hashval = stateid_hashval(sop->so_id, fp->fi_id);
2209
2210         INIT_LIST_HEAD(&stp->st_hash);
2211         INIT_LIST_HEAD(&stp->st_perstateowner);
2212         INIT_LIST_HEAD(&stp->st_lockowners);
2213         INIT_LIST_HEAD(&stp->st_perfile);
2214         list_add(&stp->st_hash, &stateid_hashtbl[hashval]);
2215         list_add(&stp->st_perstateowner, &sop->so_stateids);
2216         list_add(&stp->st_perfile, &fp->fi_stateids);
2217         stp->st_stateowner = sop;
2218         get_nfs4_file(fp);
2219         stp->st_file = fp;
2220         stp->st_stateid.si_boot = boot_time;
2221         stp->st_stateid.si_stateownerid = sop->so_id;
2222         stp->st_stateid.si_fileid = fp->fi_id;
2223         stp->st_stateid.si_generation = 0;
2224         stp->st_access_bmap = 0;
2225         stp->st_deny_bmap = 0;
2226         __set_bit(open->op_share_access & ~NFS4_SHARE_WANT_MASK,
2227                   &stp->st_access_bmap);
2228         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2229         stp->st_openstp = NULL;
2230 }
2231
2232 static void
2233 move_to_close_lru(struct nfs4_stateowner *sop)
2234 {
2235         dprintk("NFSD: move_to_close_lru nfs4_stateowner %p\n", sop);
2236
2237         list_move_tail(&sop->so_close_lru, &close_lru);
2238         sop->so_time = get_seconds();
2239 }
2240
2241 static int
2242 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner,
2243                                                         clientid_t *clid)
2244 {
2245         return (sop->so_owner.len == owner->len) &&
2246                 0 == memcmp(sop->so_owner.data, owner->data, owner->len) &&
2247                 (sop->so_client->cl_clientid.cl_id == clid->cl_id);
2248 }
2249
2250 static struct nfs4_stateowner *
2251 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open)
2252 {
2253         struct nfs4_stateowner *so = NULL;
2254
2255         list_for_each_entry(so, &ownerstr_hashtbl[hashval], so_strhash) {
2256                 if (same_owner_str(so, &open->op_owner, &open->op_clientid))
2257                         return so;
2258         }
2259         return NULL;
2260 }
2261
2262 /* search file_hashtbl[] for file */
2263 static struct nfs4_file *
2264 find_file(struct inode *ino)
2265 {
2266         unsigned int hashval = file_hashval(ino);
2267         struct nfs4_file *fp;
2268
2269         spin_lock(&recall_lock);
2270         list_for_each_entry(fp, &file_hashtbl[hashval], fi_hash) {
2271                 if (fp->fi_inode == ino) {
2272                         get_nfs4_file(fp);
2273                         spin_unlock(&recall_lock);
2274                         return fp;
2275                 }
2276         }
2277         spin_unlock(&recall_lock);
2278         return NULL;
2279 }
2280
2281 static inline int access_valid(u32 x, u32 minorversion)
2282 {
2283         if ((x & NFS4_SHARE_ACCESS_MASK) < NFS4_SHARE_ACCESS_READ)
2284                 return 0;
2285         if ((x & NFS4_SHARE_ACCESS_MASK) > NFS4_SHARE_ACCESS_BOTH)
2286                 return 0;
2287         x &= ~NFS4_SHARE_ACCESS_MASK;
2288         if (minorversion && x) {
2289                 if ((x & NFS4_SHARE_WANT_MASK) > NFS4_SHARE_WANT_CANCEL)
2290                         return 0;
2291                 if ((x & NFS4_SHARE_WHEN_MASK) > NFS4_SHARE_PUSH_DELEG_WHEN_UNCONTENDED)
2292                         return 0;
2293                 x &= ~(NFS4_SHARE_WANT_MASK | NFS4_SHARE_WHEN_MASK);
2294         }
2295         if (x)
2296                 return 0;
2297         return 1;
2298 }
2299
2300 static inline int deny_valid(u32 x)
2301 {
2302         /* Note: unlike access bits, deny bits may be zero. */
2303         return x <= NFS4_SHARE_DENY_BOTH;
2304 }
2305
2306 /*
2307  * Called to check deny when READ with all zero stateid or
2308  * WRITE with all zero or all one stateid
2309  */
2310 static __be32
2311 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
2312 {
2313         struct inode *ino = current_fh->fh_dentry->d_inode;
2314         struct nfs4_file *fp;
2315         struct nfs4_stateid *stp;
2316         __be32 ret;
2317
2318         dprintk("NFSD: nfs4_share_conflict\n");
2319
2320         fp = find_file(ino);
2321         if (!fp)
2322                 return nfs_ok;
2323         ret = nfserr_locked;
2324         /* Search for conflicting share reservations */
2325         list_for_each_entry(stp, &fp->fi_stateids, st_perfile) {
2326                 if (test_bit(deny_type, &stp->st_deny_bmap) ||
2327                     test_bit(NFS4_SHARE_DENY_BOTH, &stp->st_deny_bmap))
2328                         goto out;
2329         }
2330         ret = nfs_ok;
2331 out:
2332         put_nfs4_file(fp);
2333         return ret;
2334 }
2335
2336 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
2337 {
2338         /* We're assuming the state code never drops its reference
2339          * without first removing the lease.  Since we're in this lease
2340          * callback (and since the lease code is serialized by the kernel
2341          * lock) we know the server hasn't removed the lease yet, we know
2342          * it's safe to take a reference: */
2343         atomic_inc(&dp->dl_count);
2344
2345         list_add_tail(&dp->dl_recall_lru, &del_recall_lru);
2346
2347         /* only place dl_time is set. protected by lock_flocks*/
2348         dp->dl_time = get_seconds();
2349
2350         nfsd4_cb_recall(dp);
2351 }
2352
2353 /* Called from break_lease() with lock_flocks() held. */
2354 static void nfsd_break_deleg_cb(struct file_lock *fl)
2355 {
2356         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
2357         struct nfs4_delegation *dp;
2358
2359         BUG_ON(!fp);
2360         /* We assume break_lease is only called once per lease: */
2361         BUG_ON(fp->fi_had_conflict);
2362         /*
2363          * We don't want the locks code to timeout the lease for us;
2364          * we'll remove it ourself if a delegation isn't returned
2365          * in time:
2366          */
2367         fl->fl_break_time = 0;
2368
2369         spin_lock(&recall_lock);
2370         fp->fi_had_conflict = true;
2371         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2372                 nfsd_break_one_deleg(dp);
2373         spin_unlock(&recall_lock);
2374 }
2375
2376 static
2377 int nfsd_change_deleg_cb(struct file_lock **onlist, int arg)
2378 {
2379         if (arg & F_UNLCK)
2380                 return lease_modify(onlist, arg);
2381         else
2382                 return -EAGAIN;
2383 }
2384
2385 static const struct lock_manager_operations nfsd_lease_mng_ops = {
2386         .fl_break = nfsd_break_deleg_cb,
2387         .fl_change = nfsd_change_deleg_cb,
2388 };
2389
2390
2391 __be32
2392 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
2393                     struct nfsd4_open *open)
2394 {
2395         clientid_t *clientid = &open->op_clientid;
2396         struct nfs4_client *clp = NULL;
2397         unsigned int strhashval;
2398         struct nfs4_stateowner *sop = NULL;
2399
2400         if (!check_name(open->op_owner))
2401                 return nfserr_inval;
2402
2403         if (STALE_CLIENTID(&open->op_clientid))
2404                 return nfserr_stale_clientid;
2405
2406         strhashval = ownerstr_hashval(clientid->cl_id, open->op_owner);
2407         sop = find_openstateowner_str(strhashval, open);
2408         open->op_stateowner = sop;
2409         if (!sop) {
2410                 /* Make sure the client's lease hasn't expired. */
2411                 clp = find_confirmed_client(clientid);
2412                 if (clp == NULL)
2413                         return nfserr_expired;
2414                 goto renew;
2415         }
2416         /* When sessions are used, skip open sequenceid processing */
2417         if (nfsd4_has_session(cstate))
2418                 goto renew;
2419         if (!sop->so_confirmed) {
2420                 /* Replace unconfirmed owners without checking for replay. */
2421                 clp = sop->so_client;
2422                 release_openowner(sop);
2423                 open->op_stateowner = NULL;
2424                 goto renew;
2425         }
2426         if (open->op_seqid == sop->so_seqid - 1) {
2427                 if (sop->so_replay.rp_buflen)
2428                         return nfserr_replay_me;
2429                 /* The original OPEN failed so spectacularly
2430                  * that we don't even have replay data saved!
2431                  * Therefore, we have no choice but to continue
2432                  * processing this OPEN; presumably, we'll
2433                  * fail again for the same reason.
2434                  */
2435                 dprintk("nfsd4_process_open1: replay with no replay cache\n");
2436                 goto renew;
2437         }
2438         if (open->op_seqid != sop->so_seqid)
2439                 return nfserr_bad_seqid;
2440 renew:
2441         if (open->op_stateowner == NULL) {
2442                 sop = alloc_init_open_stateowner(strhashval, clp, open);
2443                 if (sop == NULL)
2444                         return nfserr_jukebox;
2445                 open->op_stateowner = sop;
2446         }
2447         list_del_init(&sop->so_close_lru);
2448         renew_client(sop->so_client);
2449         return nfs_ok;
2450 }
2451
2452 static inline __be32
2453 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
2454 {
2455         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
2456                 return nfserr_openmode;
2457         else
2458                 return nfs_ok;
2459 }
2460
2461 static struct nfs4_delegation *
2462 find_delegation_file(struct nfs4_file *fp, stateid_t *stid)
2463 {
2464         struct nfs4_delegation *dp;
2465
2466         spin_lock(&recall_lock);
2467         list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
2468                 if (dp->dl_stateid.si_stateownerid == stid->si_stateownerid) {
2469                         spin_unlock(&recall_lock);
2470                         return dp;
2471                 }
2472         spin_unlock(&recall_lock);
2473         return NULL;
2474 }
2475
2476 static int share_access_to_flags(u32 share_access)
2477 {
2478         share_access &= ~NFS4_SHARE_WANT_MASK;
2479
2480         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
2481 }
2482
2483 static __be32
2484 nfs4_check_deleg(struct nfs4_file *fp, struct nfsd4_open *open,
2485                 struct nfs4_delegation **dp)
2486 {
2487         int flags;
2488         __be32 status = nfserr_bad_stateid;
2489
2490         *dp = find_delegation_file(fp, &open->op_delegate_stateid);
2491         if (*dp == NULL)
2492                 goto out;
2493         flags = share_access_to_flags(open->op_share_access);
2494         status = nfs4_check_delegmode(*dp, flags);
2495         if (status)
2496                 *dp = NULL;
2497 out:
2498         if (open->op_claim_type != NFS4_OPEN_CLAIM_DELEGATE_CUR)
2499                 return nfs_ok;
2500         if (status)
2501                 return status;
2502         open->op_stateowner->so_confirmed = 1;
2503         return nfs_ok;
2504 }
2505
2506 static __be32
2507 nfs4_check_open(struct nfs4_file *fp, struct nfsd4_open *open, struct nfs4_stateid **stpp)
2508 {
2509         struct nfs4_stateid *local;
2510         __be32 status = nfserr_share_denied;
2511         struct nfs4_stateowner *sop = open->op_stateowner;
2512
2513         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
2514                 /* ignore lock owners */
2515                 if (local->st_stateowner->so_is_open_owner == 0)
2516                         continue;
2517                 /* remember if we have seen this open owner */
2518                 if (local->st_stateowner == sop)
2519                         *stpp = local;
2520                 /* check for conflicting share reservations */
2521                 if (!test_share(local, open))
2522                         goto out;
2523         }
2524         status = 0;
2525 out:
2526         return status;
2527 }
2528
2529 static inline struct nfs4_stateid *
2530 nfs4_alloc_stateid(void)
2531 {
2532         return kmem_cache_alloc(stateid_slab, GFP_KERNEL);
2533 }
2534
2535 static inline int nfs4_access_to_access(u32 nfs4_access)
2536 {
2537         int flags = 0;
2538
2539         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
2540                 flags |= NFSD_MAY_READ;
2541         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
2542                 flags |= NFSD_MAY_WRITE;
2543         return flags;
2544 }
2545
2546 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
2547                 struct svc_fh *cur_fh, struct nfsd4_open *open)
2548 {
2549         __be32 status;
2550         int oflag = nfs4_access_to_omode(open->op_share_access);
2551         int access = nfs4_access_to_access(open->op_share_access);
2552
2553         /* CLAIM_DELEGATE_CUR is used in response to a broken lease;
2554          * allowing it to break the lease and return EAGAIN leaves the
2555          * client unable to make progress in returning the delegation */
2556         if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2557                 access |= NFSD_MAY_NOT_BREAK_LEASE;
2558
2559         if (!fp->fi_fds[oflag]) {
2560                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access,
2561                         &fp->fi_fds[oflag]);
2562                 if (status)
2563                         return status;
2564         }
2565         nfs4_file_get_access(fp, oflag);
2566
2567         return nfs_ok;
2568 }
2569
2570 static __be32
2571 nfs4_new_open(struct svc_rqst *rqstp, struct nfs4_stateid **stpp,
2572                 struct nfs4_file *fp, struct svc_fh *cur_fh,
2573                 struct nfsd4_open *open)
2574 {
2575         struct nfs4_stateid *stp;
2576         __be32 status;
2577
2578         stp = nfs4_alloc_stateid();
2579         if (stp == NULL)
2580                 return nfserr_jukebox;
2581
2582         status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2583         if (status) {
2584                 kmem_cache_free(stateid_slab, stp);
2585                 return status;
2586         }
2587         *stpp = stp;
2588         return 0;
2589 }
2590
2591 static inline __be32
2592 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
2593                 struct nfsd4_open *open)
2594 {
2595         struct iattr iattr = {
2596                 .ia_valid = ATTR_SIZE,
2597                 .ia_size = 0,
2598         };
2599         if (!open->op_truncate)
2600                 return 0;
2601         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
2602                 return nfserr_inval;
2603         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
2604 }
2605
2606 static __be32
2607 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_stateid *stp, struct nfsd4_open *open)
2608 {
2609         u32 op_share_access = open->op_share_access & ~NFS4_SHARE_WANT_MASK;
2610         bool new_access;
2611         __be32 status;
2612
2613         new_access = !test_bit(op_share_access, &stp->st_access_bmap);
2614         if (new_access) {
2615                 status = nfs4_get_vfs_file(rqstp, fp, cur_fh, open);
2616                 if (status)
2617                         return status;
2618         }
2619         status = nfsd4_truncate(rqstp, cur_fh, open);
2620         if (status) {
2621                 if (new_access) {
2622                         int oflag = nfs4_access_to_omode(op_share_access);
2623                         nfs4_file_put_access(fp, oflag);
2624                 }
2625                 return status;
2626         }
2627         /* remember the open */
2628         __set_bit(op_share_access, &stp->st_access_bmap);
2629         __set_bit(open->op_share_deny, &stp->st_deny_bmap);
2630
2631         return nfs_ok;
2632 }
2633
2634
2635 static void
2636 nfs4_set_claim_prev(struct nfsd4_open *open)
2637 {
2638         open->op_stateowner->so_confirmed = 1;
2639         open->op_stateowner->so_client->cl_firststate = 1;
2640 }
2641
2642 /* Should we give out recallable state?: */
2643 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
2644 {
2645         if (clp->cl_cb_state == NFSD4_CB_UP)
2646                 return true;
2647         /*
2648          * In the sessions case, since we don't have to establish a
2649          * separate connection for callbacks, we assume it's OK
2650          * until we hear otherwise:
2651          */
2652         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
2653 }
2654
2655 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_delegation *dp, int flag)
2656 {
2657         struct file_lock *fl;
2658
2659         fl = locks_alloc_lock();
2660         if (!fl)
2661                 return NULL;
2662         locks_init_lock(fl);
2663         fl->fl_lmops = &nfsd_lease_mng_ops;
2664         fl->fl_flags = FL_LEASE;
2665         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
2666         fl->fl_end = OFFSET_MAX;
2667         fl->fl_owner = (fl_owner_t)(dp->dl_file);
2668         fl->fl_pid = current->tgid;
2669         return fl;
2670 }
2671
2672 static int nfs4_setlease(struct nfs4_delegation *dp, int flag)
2673 {
2674         struct nfs4_file *fp = dp->dl_file;
2675         struct file_lock *fl;
2676         int status;
2677
2678         fl = nfs4_alloc_init_lease(dp, flag);
2679         if (!fl)
2680                 return -ENOMEM;
2681         fl->fl_file = find_readable_file(fp);
2682         list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2683         status = vfs_setlease(fl->fl_file, fl->fl_type, &fl);
2684         if (status) {
2685                 list_del_init(&dp->dl_perclnt);
2686                 locks_free_lock(fl);
2687                 return -ENOMEM;
2688         }
2689         fp->fi_lease = fl;
2690         fp->fi_deleg_file = fl->fl_file;
2691         get_file(fp->fi_deleg_file);
2692         atomic_set(&fp->fi_delegees, 1);
2693         list_add(&dp->dl_perfile, &fp->fi_delegations);
2694         return 0;
2695 }
2696
2697 static int nfs4_set_delegation(struct nfs4_delegation *dp, int flag)
2698 {
2699         struct nfs4_file *fp = dp->dl_file;
2700
2701         if (!fp->fi_lease)
2702                 return nfs4_setlease(dp, flag);
2703         spin_lock(&recall_lock);
2704         if (fp->fi_had_conflict) {
2705                 spin_unlock(&recall_lock);
2706                 return -EAGAIN;
2707         }
2708         atomic_inc(&fp->fi_delegees);
2709         list_add(&dp->dl_perfile, &fp->fi_delegations);
2710         spin_unlock(&recall_lock);
2711         list_add(&dp->dl_perclnt, &dp->dl_client->cl_delegations);
2712         return 0;
2713 }
2714
2715 /*
2716  * Attempt to hand out a delegation.
2717  */
2718 static void
2719 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open, struct nfs4_stateid *stp)
2720 {
2721         struct nfs4_delegation *dp;
2722         struct nfs4_stateowner *sop = stp->st_stateowner;
2723         int cb_up;
2724         int status, flag = 0;
2725
2726         cb_up = nfsd4_cb_channel_good(sop->so_client);
2727         flag = NFS4_OPEN_DELEGATE_NONE;
2728         open->op_recall = 0;
2729         switch (open->op_claim_type) {
2730                 case NFS4_OPEN_CLAIM_PREVIOUS:
2731                         if (!cb_up)
2732                                 open->op_recall = 1;
2733                         flag = open->op_delegate_type;
2734                         if (flag == NFS4_OPEN_DELEGATE_NONE)
2735                                 goto out;
2736                         break;
2737                 case NFS4_OPEN_CLAIM_NULL:
2738                         /* Let's not give out any delegations till everyone's
2739                          * had the chance to reclaim theirs.... */
2740                         if (locks_in_grace())
2741                                 goto out;
2742                         if (!cb_up || !sop->so_confirmed)
2743                                 goto out;
2744                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
2745                                 flag = NFS4_OPEN_DELEGATE_WRITE;
2746                         else
2747                                 flag = NFS4_OPEN_DELEGATE_READ;
2748                         break;
2749                 default:
2750                         goto out;
2751         }
2752
2753         dp = alloc_init_deleg(sop->so_client, stp, fh, flag);
2754         if (dp == NULL)
2755                 goto out_no_deleg;
2756         status = nfs4_set_delegation(dp, flag);
2757         if (status)
2758                 goto out_free;
2759
2760         memcpy(&open->op_delegate_stateid, &dp->dl_stateid, sizeof(dp->dl_stateid));
2761
2762         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
2763                 STATEID_VAL(&dp->dl_stateid));
2764 out:
2765         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS
2766                         && flag == NFS4_OPEN_DELEGATE_NONE
2767                         && open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE)
2768                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
2769         open->op_delegate_type = flag;
2770         return;
2771 out_free:
2772         nfs4_put_delegation(dp);
2773 out_no_deleg:
2774         flag = NFS4_OPEN_DELEGATE_NONE;
2775         goto out;
2776 }
2777
2778 /*
2779  * called with nfs4_lock_state() held.
2780  */
2781 __be32
2782 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
2783 {
2784         struct nfsd4_compoundres *resp = rqstp->rq_resp;
2785         struct nfs4_file *fp = NULL;
2786         struct inode *ino = current_fh->fh_dentry->d_inode;
2787         struct nfs4_stateid *stp = NULL;
2788         struct nfs4_delegation *dp = NULL;
2789         __be32 status;
2790
2791         status = nfserr_inval;
2792         if (!access_valid(open->op_share_access, resp->cstate.minorversion)
2793                         || !deny_valid(open->op_share_deny))
2794                 goto out;
2795         /*
2796          * Lookup file; if found, lookup stateid and check open request,
2797          * and check for delegations in the process of being recalled.
2798          * If not found, create the nfs4_file struct
2799          */
2800         fp = find_file(ino);
2801         if (fp) {
2802                 if ((status = nfs4_check_open(fp, open, &stp)))
2803                         goto out;
2804                 status = nfs4_check_deleg(fp, open, &dp);
2805                 if (status)
2806                         goto out;
2807         } else {
2808                 status = nfserr_bad_stateid;
2809                 if (open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR)
2810                         goto out;
2811                 status = nfserr_jukebox;
2812                 fp = alloc_init_file(ino);
2813                 if (fp == NULL)
2814                         goto out;
2815         }
2816
2817         /*
2818          * OPEN the file, or upgrade an existing OPEN.
2819          * If truncate fails, the OPEN fails.
2820          */
2821         if (stp) {
2822                 /* Stateid was found, this is an OPEN upgrade */
2823                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
2824                 if (status)
2825                         goto out;
2826                 update_stateid(&stp->st_stateid);
2827         } else {
2828                 status = nfs4_new_open(rqstp, &stp, fp, current_fh, open);
2829                 if (status)
2830                         goto out;
2831                 init_stateid(stp, fp, open);
2832                 status = nfsd4_truncate(rqstp, current_fh, open);
2833                 if (status) {
2834                         release_open_stateid(stp);
2835                         goto out;
2836                 }
2837                 if (nfsd4_has_session(&resp->cstate))
2838                         update_stateid(&stp->st_stateid);
2839         }
2840         memcpy(&open->op_stateid, &stp->st_stateid, sizeof(stateid_t));
2841
2842         if (nfsd4_has_session(&resp->cstate))
2843                 open->op_stateowner->so_confirmed = 1;
2844
2845         /*
2846         * Attempt to hand out a delegation. No error return, because the
2847         * OPEN succeeds even if we fail.
2848         */
2849         nfs4_open_delegation(current_fh, open, stp);
2850
2851         status = nfs_ok;
2852
2853         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
2854                 STATEID_VAL(&stp->st_stateid));
2855 out:
2856         if (fp)
2857                 put_nfs4_file(fp);
2858         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
2859                 nfs4_set_claim_prev(open);
2860         /*
2861         * To finish the open response, we just need to set the rflags.
2862         */
2863         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
2864         if (!open->op_stateowner->so_confirmed &&
2865             !nfsd4_has_session(&resp->cstate))
2866                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
2867
2868         return status;
2869 }
2870
2871 __be32
2872 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2873             clientid_t *clid)
2874 {
2875         struct nfs4_client *clp;
2876         __be32 status;
2877
2878         nfs4_lock_state();
2879         dprintk("process_renew(%08x/%08x): starting\n", 
2880                         clid->cl_boot, clid->cl_id);
2881         status = nfserr_stale_clientid;
2882         if (STALE_CLIENTID(clid))
2883                 goto out;
2884         clp = find_confirmed_client(clid);
2885         status = nfserr_expired;
2886         if (clp == NULL) {
2887                 /* We assume the client took too long to RENEW. */
2888                 dprintk("nfsd4_renew: clientid not found!\n");
2889                 goto out;
2890         }
2891         renew_client(clp);
2892         status = nfserr_cb_path_down;
2893         if (!list_empty(&clp->cl_delegations)
2894                         && clp->cl_cb_state != NFSD4_CB_UP)
2895                 goto out;
2896         status = nfs_ok;
2897 out:
2898         nfs4_unlock_state();
2899         return status;
2900 }