nfs: allow blocking locks to be awoken by lock callbacks
[muen/linux.git] / fs / nfs / nfs4proc.c
1 /*
2  *  fs/nfs/nfs4proc.c
3  *
4  *  Client-side procedure declarations for NFSv4.
5  *
6  *  Copyright (c) 2002 The Regents of the University of Michigan.
7  *  All rights reserved.
8  *
9  *  Kendrick Smith <kmsmith@umich.edu>
10  *  Andy Adamson   <andros@umich.edu>
11  *
12  *  Redistribution and use in source and binary forms, with or without
13  *  modification, are permitted provided that the following conditions
14  *  are met:
15  *
16  *  1. Redistributions of source code must retain the above copyright
17  *     notice, this list of conditions and the following disclaimer.
18  *  2. Redistributions in binary form must reproduce the above copyright
19  *     notice, this list of conditions and the following disclaimer in the
20  *     documentation and/or other materials provided with the distribution.
21  *  3. Neither the name of the University nor the names of its
22  *     contributors may be used to endorse or promote products derived
23  *     from this software without specific prior written permission.
24  *
25  *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
26  *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
27  *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
28  *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
30  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
31  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
32  *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
33  *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
34  *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
35  *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36  */
37
38 #include <linux/mm.h>
39 #include <linux/delay.h>
40 #include <linux/errno.h>
41 #include <linux/file.h>
42 #include <linux/string.h>
43 #include <linux/ratelimit.h>
44 #include <linux/printk.h>
45 #include <linux/slab.h>
46 #include <linux/sunrpc/clnt.h>
47 #include <linux/nfs.h>
48 #include <linux/nfs4.h>
49 #include <linux/nfs_fs.h>
50 #include <linux/nfs_page.h>
51 #include <linux/nfs_mount.h>
52 #include <linux/namei.h>
53 #include <linux/mount.h>
54 #include <linux/module.h>
55 #include <linux/xattr.h>
56 #include <linux/utsname.h>
57 #include <linux/freezer.h>
58
59 #include "nfs4_fs.h"
60 #include "delegation.h"
61 #include "internal.h"
62 #include "iostat.h"
63 #include "callback.h"
64 #include "pnfs.h"
65 #include "netns.h"
66 #include "nfs4idmap.h"
67 #include "nfs4session.h"
68 #include "fscache.h"
69
70 #include "nfs4trace.h"
71
72 #define NFSDBG_FACILITY         NFSDBG_PROC
73
74 #define NFS4_POLL_RETRY_MIN     (HZ/10)
75 #define NFS4_POLL_RETRY_MAX     (15*HZ)
76
77 /* file attributes which can be mapped to nfs attributes */
78 #define NFS4_VALID_ATTRS (ATTR_MODE \
79         | ATTR_UID \
80         | ATTR_GID \
81         | ATTR_SIZE \
82         | ATTR_ATIME \
83         | ATTR_MTIME \
84         | ATTR_CTIME \
85         | ATTR_ATIME_SET \
86         | ATTR_MTIME_SET)
87
88 struct nfs4_opendata;
89 static int _nfs4_proc_open(struct nfs4_opendata *data);
90 static int _nfs4_recover_proc_open(struct nfs4_opendata *data);
91 static int nfs4_do_fsinfo(struct nfs_server *, struct nfs_fh *, struct nfs_fsinfo *);
92 static void nfs_fixup_referral_attributes(struct nfs_fattr *fattr);
93 static int nfs4_proc_getattr(struct nfs_server *, struct nfs_fh *, struct nfs_fattr *, struct nfs4_label *label);
94 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle, struct nfs_fattr *fattr, struct nfs4_label *label);
95 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
96                             struct nfs_fattr *fattr, struct iattr *sattr,
97                             struct nfs4_state *state, struct nfs4_label *ilabel,
98                             struct nfs4_label *olabel);
99 #ifdef CONFIG_NFS_V4_1
100 static int nfs41_test_stateid(struct nfs_server *, nfs4_stateid *,
101                 struct rpc_cred *);
102 static int nfs41_free_stateid(struct nfs_server *, nfs4_stateid *,
103                 struct rpc_cred *);
104 #endif
105
106 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
107 static inline struct nfs4_label *
108 nfs4_label_init_security(struct inode *dir, struct dentry *dentry,
109         struct iattr *sattr, struct nfs4_label *label)
110 {
111         int err;
112
113         if (label == NULL)
114                 return NULL;
115
116         if (nfs_server_capable(dir, NFS_CAP_SECURITY_LABEL) == 0)
117                 return NULL;
118
119         err = security_dentry_init_security(dentry, sattr->ia_mode,
120                                 &dentry->d_name, (void **)&label->label, &label->len);
121         if (err == 0)
122                 return label;
123
124         return NULL;
125 }
126 static inline void
127 nfs4_label_release_security(struct nfs4_label *label)
128 {
129         if (label)
130                 security_release_secctx(label->label, label->len);
131 }
132 static inline u32 *nfs4_bitmask(struct nfs_server *server, struct nfs4_label *label)
133 {
134         if (label)
135                 return server->attr_bitmask;
136
137         return server->attr_bitmask_nl;
138 }
139 #else
140 static inline struct nfs4_label *
141 nfs4_label_init_security(struct inode *dir, struct dentry *dentry,
142         struct iattr *sattr, struct nfs4_label *l)
143 { return NULL; }
144 static inline void
145 nfs4_label_release_security(struct nfs4_label *label)
146 { return; }
147 static inline u32 *
148 nfs4_bitmask(struct nfs_server *server, struct nfs4_label *label)
149 { return server->attr_bitmask; }
150 #endif
151
152 /* Prevent leaks of NFSv4 errors into userland */
153 static int nfs4_map_errors(int err)
154 {
155         if (err >= -1000)
156                 return err;
157         switch (err) {
158         case -NFS4ERR_RESOURCE:
159         case -NFS4ERR_LAYOUTTRYLATER:
160         case -NFS4ERR_RECALLCONFLICT:
161                 return -EREMOTEIO;
162         case -NFS4ERR_WRONGSEC:
163         case -NFS4ERR_WRONG_CRED:
164                 return -EPERM;
165         case -NFS4ERR_BADOWNER:
166         case -NFS4ERR_BADNAME:
167                 return -EINVAL;
168         case -NFS4ERR_SHARE_DENIED:
169                 return -EACCES;
170         case -NFS4ERR_MINOR_VERS_MISMATCH:
171                 return -EPROTONOSUPPORT;
172         case -NFS4ERR_FILE_OPEN:
173                 return -EBUSY;
174         default:
175                 dprintk("%s could not handle NFSv4 error %d\n",
176                                 __func__, -err);
177                 break;
178         }
179         return -EIO;
180 }
181
182 /*
183  * This is our standard bitmap for GETATTR requests.
184  */
185 const u32 nfs4_fattr_bitmap[3] = {
186         FATTR4_WORD0_TYPE
187         | FATTR4_WORD0_CHANGE
188         | FATTR4_WORD0_SIZE
189         | FATTR4_WORD0_FSID
190         | FATTR4_WORD0_FILEID,
191         FATTR4_WORD1_MODE
192         | FATTR4_WORD1_NUMLINKS
193         | FATTR4_WORD1_OWNER
194         | FATTR4_WORD1_OWNER_GROUP
195         | FATTR4_WORD1_RAWDEV
196         | FATTR4_WORD1_SPACE_USED
197         | FATTR4_WORD1_TIME_ACCESS
198         | FATTR4_WORD1_TIME_METADATA
199         | FATTR4_WORD1_TIME_MODIFY
200         | FATTR4_WORD1_MOUNTED_ON_FILEID,
201 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
202         FATTR4_WORD2_SECURITY_LABEL
203 #endif
204 };
205
206 static const u32 nfs4_pnfs_open_bitmap[3] = {
207         FATTR4_WORD0_TYPE
208         | FATTR4_WORD0_CHANGE
209         | FATTR4_WORD0_SIZE
210         | FATTR4_WORD0_FSID
211         | FATTR4_WORD0_FILEID,
212         FATTR4_WORD1_MODE
213         | FATTR4_WORD1_NUMLINKS
214         | FATTR4_WORD1_OWNER
215         | FATTR4_WORD1_OWNER_GROUP
216         | FATTR4_WORD1_RAWDEV
217         | FATTR4_WORD1_SPACE_USED
218         | FATTR4_WORD1_TIME_ACCESS
219         | FATTR4_WORD1_TIME_METADATA
220         | FATTR4_WORD1_TIME_MODIFY,
221         FATTR4_WORD2_MDSTHRESHOLD
222 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
223         | FATTR4_WORD2_SECURITY_LABEL
224 #endif
225 };
226
227 static const u32 nfs4_open_noattr_bitmap[3] = {
228         FATTR4_WORD0_TYPE
229         | FATTR4_WORD0_CHANGE
230         | FATTR4_WORD0_FILEID,
231 };
232
233 const u32 nfs4_statfs_bitmap[3] = {
234         FATTR4_WORD0_FILES_AVAIL
235         | FATTR4_WORD0_FILES_FREE
236         | FATTR4_WORD0_FILES_TOTAL,
237         FATTR4_WORD1_SPACE_AVAIL
238         | FATTR4_WORD1_SPACE_FREE
239         | FATTR4_WORD1_SPACE_TOTAL
240 };
241
242 const u32 nfs4_pathconf_bitmap[3] = {
243         FATTR4_WORD0_MAXLINK
244         | FATTR4_WORD0_MAXNAME,
245         0
246 };
247
248 const u32 nfs4_fsinfo_bitmap[3] = { FATTR4_WORD0_MAXFILESIZE
249                         | FATTR4_WORD0_MAXREAD
250                         | FATTR4_WORD0_MAXWRITE
251                         | FATTR4_WORD0_LEASE_TIME,
252                         FATTR4_WORD1_TIME_DELTA
253                         | FATTR4_WORD1_FS_LAYOUT_TYPES,
254                         FATTR4_WORD2_LAYOUT_BLKSIZE
255                         | FATTR4_WORD2_CLONE_BLKSIZE
256 };
257
258 const u32 nfs4_fs_locations_bitmap[3] = {
259         FATTR4_WORD0_TYPE
260         | FATTR4_WORD0_CHANGE
261         | FATTR4_WORD0_SIZE
262         | FATTR4_WORD0_FSID
263         | FATTR4_WORD0_FILEID
264         | FATTR4_WORD0_FS_LOCATIONS,
265         FATTR4_WORD1_MODE
266         | FATTR4_WORD1_NUMLINKS
267         | FATTR4_WORD1_OWNER
268         | FATTR4_WORD1_OWNER_GROUP
269         | FATTR4_WORD1_RAWDEV
270         | FATTR4_WORD1_SPACE_USED
271         | FATTR4_WORD1_TIME_ACCESS
272         | FATTR4_WORD1_TIME_METADATA
273         | FATTR4_WORD1_TIME_MODIFY
274         | FATTR4_WORD1_MOUNTED_ON_FILEID,
275 };
276
277 static void nfs4_setup_readdir(u64 cookie, __be32 *verifier, struct dentry *dentry,
278                 struct nfs4_readdir_arg *readdir)
279 {
280         __be32 *start, *p;
281
282         if (cookie > 2) {
283                 readdir->cookie = cookie;
284                 memcpy(&readdir->verifier, verifier, sizeof(readdir->verifier));
285                 return;
286         }
287
288         readdir->cookie = 0;
289         memset(&readdir->verifier, 0, sizeof(readdir->verifier));
290         if (cookie == 2)
291                 return;
292         
293         /*
294          * NFSv4 servers do not return entries for '.' and '..'
295          * Therefore, we fake these entries here.  We let '.'
296          * have cookie 0 and '..' have cookie 1.  Note that
297          * when talking to the server, we always send cookie 0
298          * instead of 1 or 2.
299          */
300         start = p = kmap_atomic(*readdir->pages);
301         
302         if (cookie == 0) {
303                 *p++ = xdr_one;                                  /* next */
304                 *p++ = xdr_zero;                   /* cookie, first word */
305                 *p++ = xdr_one;                   /* cookie, second word */
306                 *p++ = xdr_one;                             /* entry len */
307                 memcpy(p, ".\0\0\0", 4);                        /* entry */
308                 p++;
309                 *p++ = xdr_one;                         /* bitmap length */
310                 *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
311                 *p++ = htonl(8);              /* attribute buffer length */
312                 p = xdr_encode_hyper(p, NFS_FILEID(d_inode(dentry)));
313         }
314         
315         *p++ = xdr_one;                                  /* next */
316         *p++ = xdr_zero;                   /* cookie, first word */
317         *p++ = xdr_two;                   /* cookie, second word */
318         *p++ = xdr_two;                             /* entry len */
319         memcpy(p, "..\0\0", 4);                         /* entry */
320         p++;
321         *p++ = xdr_one;                         /* bitmap length */
322         *p++ = htonl(FATTR4_WORD0_FILEID);             /* bitmap */
323         *p++ = htonl(8);              /* attribute buffer length */
324         p = xdr_encode_hyper(p, NFS_FILEID(d_inode(dentry->d_parent)));
325
326         readdir->pgbase = (char *)p - (char *)start;
327         readdir->count -= readdir->pgbase;
328         kunmap_atomic(start);
329 }
330
331 static long nfs4_update_delay(long *timeout)
332 {
333         long ret;
334         if (!timeout)
335                 return NFS4_POLL_RETRY_MAX;
336         if (*timeout <= 0)
337                 *timeout = NFS4_POLL_RETRY_MIN;
338         if (*timeout > NFS4_POLL_RETRY_MAX)
339                 *timeout = NFS4_POLL_RETRY_MAX;
340         ret = *timeout;
341         *timeout <<= 1;
342         return ret;
343 }
344
345 static int nfs4_delay(struct rpc_clnt *clnt, long *timeout)
346 {
347         int res = 0;
348
349         might_sleep();
350
351         freezable_schedule_timeout_killable_unsafe(
352                 nfs4_update_delay(timeout));
353         if (fatal_signal_pending(current))
354                 res = -ERESTARTSYS;
355         return res;
356 }
357
358 /* This is the error handling routine for processes that are allowed
359  * to sleep.
360  */
361 static int nfs4_do_handle_exception(struct nfs_server *server,
362                 int errorcode, struct nfs4_exception *exception)
363 {
364         struct nfs_client *clp = server->nfs_client;
365         struct nfs4_state *state = exception->state;
366         const nfs4_stateid *stateid = exception->stateid;
367         struct inode *inode = exception->inode;
368         int ret = errorcode;
369
370         exception->delay = 0;
371         exception->recovering = 0;
372         exception->retry = 0;
373         switch(errorcode) {
374                 case 0:
375                         return 0;
376                 case -NFS4ERR_OPENMODE:
377                 case -NFS4ERR_DELEG_REVOKED:
378                 case -NFS4ERR_ADMIN_REVOKED:
379                 case -NFS4ERR_BAD_STATEID:
380                         if (inode) {
381                                 int err;
382
383                                 err = nfs_async_inode_return_delegation(inode,
384                                                 stateid);
385                                 if (err == 0)
386                                         goto wait_on_recovery;
387                                 if (stateid != NULL && stateid->type == NFS4_DELEGATION_STATEID_TYPE) {
388                                         exception->retry = 1;
389                                         break;
390                                 }
391                         }
392                         if (state == NULL)
393                                 break;
394                         ret = nfs4_schedule_stateid_recovery(server, state);
395                         if (ret < 0)
396                                 break;
397                         goto wait_on_recovery;
398                 case -NFS4ERR_EXPIRED:
399                         if (state != NULL) {
400                                 ret = nfs4_schedule_stateid_recovery(server, state);
401                                 if (ret < 0)
402                                         break;
403                         }
404                 case -NFS4ERR_STALE_STATEID:
405                 case -NFS4ERR_STALE_CLIENTID:
406                         nfs4_schedule_lease_recovery(clp);
407                         goto wait_on_recovery;
408                 case -NFS4ERR_MOVED:
409                         ret = nfs4_schedule_migration_recovery(server);
410                         if (ret < 0)
411                                 break;
412                         goto wait_on_recovery;
413                 case -NFS4ERR_LEASE_MOVED:
414                         nfs4_schedule_lease_moved_recovery(clp);
415                         goto wait_on_recovery;
416 #if defined(CONFIG_NFS_V4_1)
417                 case -NFS4ERR_BADSESSION:
418                 case -NFS4ERR_BADSLOT:
419                 case -NFS4ERR_BAD_HIGH_SLOT:
420                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
421                 case -NFS4ERR_DEADSESSION:
422                 case -NFS4ERR_SEQ_FALSE_RETRY:
423                 case -NFS4ERR_SEQ_MISORDERED:
424                         dprintk("%s ERROR: %d Reset session\n", __func__,
425                                 errorcode);
426                         nfs4_schedule_session_recovery(clp->cl_session, errorcode);
427                         goto wait_on_recovery;
428 #endif /* defined(CONFIG_NFS_V4_1) */
429                 case -NFS4ERR_FILE_OPEN:
430                         if (exception->timeout > HZ) {
431                                 /* We have retried a decent amount, time to
432                                  * fail
433                                  */
434                                 ret = -EBUSY;
435                                 break;
436                         }
437                 case -NFS4ERR_DELAY:
438                         nfs_inc_server_stats(server, NFSIOS_DELAY);
439                 case -NFS4ERR_GRACE:
440                 case -NFS4ERR_LAYOUTTRYLATER:
441                 case -NFS4ERR_RECALLCONFLICT:
442                         exception->delay = 1;
443                         return 0;
444
445                 case -NFS4ERR_RETRY_UNCACHED_REP:
446                 case -NFS4ERR_OLD_STATEID:
447                         exception->retry = 1;
448                         break;
449                 case -NFS4ERR_BADOWNER:
450                         /* The following works around a Linux server bug! */
451                 case -NFS4ERR_BADNAME:
452                         if (server->caps & NFS_CAP_UIDGID_NOMAP) {
453                                 server->caps &= ~NFS_CAP_UIDGID_NOMAP;
454                                 exception->retry = 1;
455                                 printk(KERN_WARNING "NFS: v4 server %s "
456                                                 "does not accept raw "
457                                                 "uid/gids. "
458                                                 "Reenabling the idmapper.\n",
459                                                 server->nfs_client->cl_hostname);
460                         }
461         }
462         /* We failed to handle the error */
463         return nfs4_map_errors(ret);
464 wait_on_recovery:
465         exception->recovering = 1;
466         return 0;
467 }
468
469 /* This is the error handling routine for processes that are allowed
470  * to sleep.
471  */
472 int nfs4_handle_exception(struct nfs_server *server, int errorcode, struct nfs4_exception *exception)
473 {
474         struct nfs_client *clp = server->nfs_client;
475         int ret;
476
477         ret = nfs4_do_handle_exception(server, errorcode, exception);
478         if (exception->delay) {
479                 ret = nfs4_delay(server->client, &exception->timeout);
480                 goto out_retry;
481         }
482         if (exception->recovering) {
483                 ret = nfs4_wait_clnt_recover(clp);
484                 if (test_bit(NFS_MIG_FAILED, &server->mig_status))
485                         return -EIO;
486                 goto out_retry;
487         }
488         return ret;
489 out_retry:
490         if (ret == 0)
491                 exception->retry = 1;
492         return ret;
493 }
494
495 static int
496 nfs4_async_handle_exception(struct rpc_task *task, struct nfs_server *server,
497                 int errorcode, struct nfs4_exception *exception)
498 {
499         struct nfs_client *clp = server->nfs_client;
500         int ret;
501
502         ret = nfs4_do_handle_exception(server, errorcode, exception);
503         if (exception->delay) {
504                 rpc_delay(task, nfs4_update_delay(&exception->timeout));
505                 goto out_retry;
506         }
507         if (exception->recovering) {
508                 rpc_sleep_on(&clp->cl_rpcwaitq, task, NULL);
509                 if (test_bit(NFS4CLNT_MANAGER_RUNNING, &clp->cl_state) == 0)
510                         rpc_wake_up_queued_task(&clp->cl_rpcwaitq, task);
511                 goto out_retry;
512         }
513         if (test_bit(NFS_MIG_FAILED, &server->mig_status))
514                 ret = -EIO;
515         return ret;
516 out_retry:
517         if (ret == 0)
518                 exception->retry = 1;
519         return ret;
520 }
521
522 static int
523 nfs4_async_handle_error(struct rpc_task *task, struct nfs_server *server,
524                         struct nfs4_state *state, long *timeout)
525 {
526         struct nfs4_exception exception = {
527                 .state = state,
528         };
529
530         if (task->tk_status >= 0)
531                 return 0;
532         if (timeout)
533                 exception.timeout = *timeout;
534         task->tk_status = nfs4_async_handle_exception(task, server,
535                         task->tk_status,
536                         &exception);
537         if (exception.delay && timeout)
538                 *timeout = exception.timeout;
539         if (exception.retry)
540                 return -EAGAIN;
541         return 0;
542 }
543
544 /*
545  * Return 'true' if 'clp' is using an rpc_client that is integrity protected
546  * or 'false' otherwise.
547  */
548 static bool _nfs4_is_integrity_protected(struct nfs_client *clp)
549 {
550         rpc_authflavor_t flavor = clp->cl_rpcclient->cl_auth->au_flavor;
551
552         if (flavor == RPC_AUTH_GSS_KRB5I ||
553             flavor == RPC_AUTH_GSS_KRB5P)
554                 return true;
555
556         return false;
557 }
558
559 static void do_renew_lease(struct nfs_client *clp, unsigned long timestamp)
560 {
561         spin_lock(&clp->cl_lock);
562         if (time_before(clp->cl_last_renewal,timestamp))
563                 clp->cl_last_renewal = timestamp;
564         spin_unlock(&clp->cl_lock);
565 }
566
567 static void renew_lease(const struct nfs_server *server, unsigned long timestamp)
568 {
569         struct nfs_client *clp = server->nfs_client;
570
571         if (!nfs4_has_session(clp))
572                 do_renew_lease(clp, timestamp);
573 }
574
575 struct nfs4_call_sync_data {
576         const struct nfs_server *seq_server;
577         struct nfs4_sequence_args *seq_args;
578         struct nfs4_sequence_res *seq_res;
579 };
580
581 void nfs4_init_sequence(struct nfs4_sequence_args *args,
582                         struct nfs4_sequence_res *res, int cache_reply)
583 {
584         args->sa_slot = NULL;
585         args->sa_cache_this = cache_reply;
586         args->sa_privileged = 0;
587
588         res->sr_slot = NULL;
589 }
590
591 static void nfs4_set_sequence_privileged(struct nfs4_sequence_args *args)
592 {
593         args->sa_privileged = 1;
594 }
595
596 int nfs40_setup_sequence(struct nfs4_slot_table *tbl,
597                          struct nfs4_sequence_args *args,
598                          struct nfs4_sequence_res *res,
599                          struct rpc_task *task)
600 {
601         struct nfs4_slot *slot;
602
603         /* slot already allocated? */
604         if (res->sr_slot != NULL)
605                 goto out_start;
606
607         spin_lock(&tbl->slot_tbl_lock);
608         if (nfs4_slot_tbl_draining(tbl) && !args->sa_privileged)
609                 goto out_sleep;
610
611         slot = nfs4_alloc_slot(tbl);
612         if (IS_ERR(slot)) {
613                 if (slot == ERR_PTR(-ENOMEM))
614                         task->tk_timeout = HZ >> 2;
615                 goto out_sleep;
616         }
617         spin_unlock(&tbl->slot_tbl_lock);
618
619         args->sa_slot = slot;
620         res->sr_slot = slot;
621
622 out_start:
623         rpc_call_start(task);
624         return 0;
625
626 out_sleep:
627         if (args->sa_privileged)
628                 rpc_sleep_on_priority(&tbl->slot_tbl_waitq, task,
629                                 NULL, RPC_PRIORITY_PRIVILEGED);
630         else
631                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
632         spin_unlock(&tbl->slot_tbl_lock);
633         return -EAGAIN;
634 }
635 EXPORT_SYMBOL_GPL(nfs40_setup_sequence);
636
637 static void nfs40_sequence_free_slot(struct nfs4_sequence_res *res)
638 {
639         struct nfs4_slot *slot = res->sr_slot;
640         struct nfs4_slot_table *tbl;
641
642         tbl = slot->table;
643         spin_lock(&tbl->slot_tbl_lock);
644         if (!nfs41_wake_and_assign_slot(tbl, slot))
645                 nfs4_free_slot(tbl, slot);
646         spin_unlock(&tbl->slot_tbl_lock);
647
648         res->sr_slot = NULL;
649 }
650
651 static int nfs40_sequence_done(struct rpc_task *task,
652                                struct nfs4_sequence_res *res)
653 {
654         if (res->sr_slot != NULL)
655                 nfs40_sequence_free_slot(res);
656         return 1;
657 }
658
659 #if defined(CONFIG_NFS_V4_1)
660
661 static void nfs41_sequence_free_slot(struct nfs4_sequence_res *res)
662 {
663         struct nfs4_session *session;
664         struct nfs4_slot_table *tbl;
665         struct nfs4_slot *slot = res->sr_slot;
666         bool send_new_highest_used_slotid = false;
667
668         tbl = slot->table;
669         session = tbl->session;
670
671         /* Bump the slot sequence number */
672         if (slot->seq_done)
673                 slot->seq_nr++;
674         slot->seq_done = 0;
675
676         spin_lock(&tbl->slot_tbl_lock);
677         /* Be nice to the server: try to ensure that the last transmitted
678          * value for highest_user_slotid <= target_highest_slotid
679          */
680         if (tbl->highest_used_slotid > tbl->target_highest_slotid)
681                 send_new_highest_used_slotid = true;
682
683         if (nfs41_wake_and_assign_slot(tbl, slot)) {
684                 send_new_highest_used_slotid = false;
685                 goto out_unlock;
686         }
687         nfs4_free_slot(tbl, slot);
688
689         if (tbl->highest_used_slotid != NFS4_NO_SLOT)
690                 send_new_highest_used_slotid = false;
691 out_unlock:
692         spin_unlock(&tbl->slot_tbl_lock);
693         res->sr_slot = NULL;
694         if (send_new_highest_used_slotid)
695                 nfs41_notify_server(session->clp);
696         if (waitqueue_active(&tbl->slot_waitq))
697                 wake_up_all(&tbl->slot_waitq);
698 }
699
700 static int nfs41_sequence_process(struct rpc_task *task,
701                 struct nfs4_sequence_res *res)
702 {
703         struct nfs4_session *session;
704         struct nfs4_slot *slot = res->sr_slot;
705         struct nfs_client *clp;
706         bool interrupted = false;
707         int ret = 1;
708
709         if (slot == NULL)
710                 goto out_noaction;
711         /* don't increment the sequence number if the task wasn't sent */
712         if (!RPC_WAS_SENT(task))
713                 goto out;
714
715         session = slot->table->session;
716
717         if (slot->interrupted) {
718                 slot->interrupted = 0;
719                 interrupted = true;
720         }
721
722         trace_nfs4_sequence_done(session, res);
723         /* Check the SEQUENCE operation status */
724         switch (res->sr_status) {
725         case 0:
726                 /* Update the slot's sequence and clientid lease timer */
727                 slot->seq_done = 1;
728                 clp = session->clp;
729                 do_renew_lease(clp, res->sr_timestamp);
730                 /* Check sequence flags */
731                 nfs41_handle_sequence_flag_errors(clp, res->sr_status_flags);
732                 nfs41_update_target_slotid(slot->table, slot, res);
733                 break;
734         case 1:
735                 /*
736                  * sr_status remains 1 if an RPC level error occurred.
737                  * The server may or may not have processed the sequence
738                  * operation..
739                  * Mark the slot as having hosted an interrupted RPC call.
740                  */
741                 slot->interrupted = 1;
742                 goto out;
743         case -NFS4ERR_DELAY:
744                 /* The server detected a resend of the RPC call and
745                  * returned NFS4ERR_DELAY as per Section 2.10.6.2
746                  * of RFC5661.
747                  */
748                 dprintk("%s: slot=%u seq=%u: Operation in progress\n",
749                         __func__,
750                         slot->slot_nr,
751                         slot->seq_nr);
752                 goto out_retry;
753         case -NFS4ERR_BADSLOT:
754                 /*
755                  * The slot id we used was probably retired. Try again
756                  * using a different slot id.
757                  */
758                 goto retry_nowait;
759         case -NFS4ERR_SEQ_MISORDERED:
760                 /*
761                  * Was the last operation on this sequence interrupted?
762                  * If so, retry after bumping the sequence number.
763                  */
764                 if (interrupted) {
765                         ++slot->seq_nr;
766                         goto retry_nowait;
767                 }
768                 /*
769                  * Could this slot have been previously retired?
770                  * If so, then the server may be expecting seq_nr = 1!
771                  */
772                 if (slot->seq_nr != 1) {
773                         slot->seq_nr = 1;
774                         goto retry_nowait;
775                 }
776                 break;
777         case -NFS4ERR_SEQ_FALSE_RETRY:
778                 ++slot->seq_nr;
779                 goto retry_nowait;
780         default:
781                 /* Just update the slot sequence no. */
782                 slot->seq_done = 1;
783         }
784 out:
785         /* The session may be reset by one of the error handlers. */
786         dprintk("%s: Error %d free the slot \n", __func__, res->sr_status);
787 out_noaction:
788         return ret;
789 retry_nowait:
790         if (rpc_restart_call_prepare(task)) {
791                 nfs41_sequence_free_slot(res);
792                 task->tk_status = 0;
793                 ret = 0;
794         }
795         goto out;
796 out_retry:
797         if (!rpc_restart_call(task))
798                 goto out;
799         rpc_delay(task, NFS4_POLL_RETRY_MAX);
800         return 0;
801 }
802
803 int nfs41_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
804 {
805         if (!nfs41_sequence_process(task, res))
806                 return 0;
807         if (res->sr_slot != NULL)
808                 nfs41_sequence_free_slot(res);
809         return 1;
810
811 }
812 EXPORT_SYMBOL_GPL(nfs41_sequence_done);
813
814 static int nfs4_sequence_process(struct rpc_task *task, struct nfs4_sequence_res *res)
815 {
816         if (res->sr_slot == NULL)
817                 return 1;
818         if (res->sr_slot->table->session != NULL)
819                 return nfs41_sequence_process(task, res);
820         return nfs40_sequence_done(task, res);
821 }
822
823 static void nfs4_sequence_free_slot(struct nfs4_sequence_res *res)
824 {
825         if (res->sr_slot != NULL) {
826                 if (res->sr_slot->table->session != NULL)
827                         nfs41_sequence_free_slot(res);
828                 else
829                         nfs40_sequence_free_slot(res);
830         }
831 }
832
833 int nfs4_sequence_done(struct rpc_task *task, struct nfs4_sequence_res *res)
834 {
835         if (res->sr_slot == NULL)
836                 return 1;
837         if (!res->sr_slot->table->session)
838                 return nfs40_sequence_done(task, res);
839         return nfs41_sequence_done(task, res);
840 }
841 EXPORT_SYMBOL_GPL(nfs4_sequence_done);
842
843 int nfs41_setup_sequence(struct nfs4_session *session,
844                                 struct nfs4_sequence_args *args,
845                                 struct nfs4_sequence_res *res,
846                                 struct rpc_task *task)
847 {
848         struct nfs4_slot *slot;
849         struct nfs4_slot_table *tbl;
850
851         dprintk("--> %s\n", __func__);
852         /* slot already allocated? */
853         if (res->sr_slot != NULL)
854                 goto out_success;
855
856         tbl = &session->fc_slot_table;
857
858         task->tk_timeout = 0;
859
860         spin_lock(&tbl->slot_tbl_lock);
861         if (test_bit(NFS4_SLOT_TBL_DRAINING, &tbl->slot_tbl_state) &&
862             !args->sa_privileged) {
863                 /* The state manager will wait until the slot table is empty */
864                 dprintk("%s session is draining\n", __func__);
865                 goto out_sleep;
866         }
867
868         slot = nfs4_alloc_slot(tbl);
869         if (IS_ERR(slot)) {
870                 /* If out of memory, try again in 1/4 second */
871                 if (slot == ERR_PTR(-ENOMEM))
872                         task->tk_timeout = HZ >> 2;
873                 dprintk("<-- %s: no free slots\n", __func__);
874                 goto out_sleep;
875         }
876         spin_unlock(&tbl->slot_tbl_lock);
877
878         args->sa_slot = slot;
879
880         dprintk("<-- %s slotid=%u seqid=%u\n", __func__,
881                         slot->slot_nr, slot->seq_nr);
882
883         res->sr_slot = slot;
884         res->sr_timestamp = jiffies;
885         res->sr_status_flags = 0;
886         /*
887          * sr_status is only set in decode_sequence, and so will remain
888          * set to 1 if an rpc level failure occurs.
889          */
890         res->sr_status = 1;
891         trace_nfs4_setup_sequence(session, args);
892 out_success:
893         rpc_call_start(task);
894         return 0;
895 out_sleep:
896         /* Privileged tasks are queued with top priority */
897         if (args->sa_privileged)
898                 rpc_sleep_on_priority(&tbl->slot_tbl_waitq, task,
899                                 NULL, RPC_PRIORITY_PRIVILEGED);
900         else
901                 rpc_sleep_on(&tbl->slot_tbl_waitq, task, NULL);
902         spin_unlock(&tbl->slot_tbl_lock);
903         return -EAGAIN;
904 }
905 EXPORT_SYMBOL_GPL(nfs41_setup_sequence);
906
907 static int nfs4_setup_sequence(const struct nfs_server *server,
908                                struct nfs4_sequence_args *args,
909                                struct nfs4_sequence_res *res,
910                                struct rpc_task *task)
911 {
912         struct nfs4_session *session = nfs4_get_session(server);
913         int ret = 0;
914
915         if (!session)
916                 return nfs40_setup_sequence(server->nfs_client->cl_slot_tbl,
917                                             args, res, task);
918
919         dprintk("--> %s clp %p session %p sr_slot %u\n",
920                 __func__, session->clp, session, res->sr_slot ?
921                         res->sr_slot->slot_nr : NFS4_NO_SLOT);
922
923         ret = nfs41_setup_sequence(session, args, res, task);
924
925         dprintk("<-- %s status=%d\n", __func__, ret);
926         return ret;
927 }
928
929 static void nfs41_call_sync_prepare(struct rpc_task *task, void *calldata)
930 {
931         struct nfs4_call_sync_data *data = calldata;
932         struct nfs4_session *session = nfs4_get_session(data->seq_server);
933
934         dprintk("--> %s data->seq_server %p\n", __func__, data->seq_server);
935
936         nfs41_setup_sequence(session, data->seq_args, data->seq_res, task);
937 }
938
939 static void nfs41_call_sync_done(struct rpc_task *task, void *calldata)
940 {
941         struct nfs4_call_sync_data *data = calldata;
942
943         nfs41_sequence_done(task, data->seq_res);
944 }
945
946 static const struct rpc_call_ops nfs41_call_sync_ops = {
947         .rpc_call_prepare = nfs41_call_sync_prepare,
948         .rpc_call_done = nfs41_call_sync_done,
949 };
950
951 #else   /* !CONFIG_NFS_V4_1 */
952
953 static int nfs4_setup_sequence(const struct nfs_server *server,
954                                struct nfs4_sequence_args *args,
955                                struct nfs4_sequence_res *res,
956                                struct rpc_task *task)
957 {
958         return nfs40_setup_sequence(server->nfs_client->cl_slot_tbl,
959                                     args, res, task);
960 }
961
962 static int nfs4_sequence_process(struct rpc_task *task, struct nfs4_sequence_res *res)
963 {
964         return nfs40_sequence_done(task, res);
965 }
966
967 static void nfs4_sequence_free_slot(struct nfs4_sequence_res *res)
968 {
969         if (res->sr_slot != NULL)
970                 nfs40_sequence_free_slot(res);
971 }
972
973 int nfs4_sequence_done(struct rpc_task *task,
974                        struct nfs4_sequence_res *res)
975 {
976         return nfs40_sequence_done(task, res);
977 }
978 EXPORT_SYMBOL_GPL(nfs4_sequence_done);
979
980 #endif  /* !CONFIG_NFS_V4_1 */
981
982 static void nfs40_call_sync_prepare(struct rpc_task *task, void *calldata)
983 {
984         struct nfs4_call_sync_data *data = calldata;
985         nfs4_setup_sequence(data->seq_server,
986                                 data->seq_args, data->seq_res, task);
987 }
988
989 static void nfs40_call_sync_done(struct rpc_task *task, void *calldata)
990 {
991         struct nfs4_call_sync_data *data = calldata;
992         nfs4_sequence_done(task, data->seq_res);
993 }
994
995 static const struct rpc_call_ops nfs40_call_sync_ops = {
996         .rpc_call_prepare = nfs40_call_sync_prepare,
997         .rpc_call_done = nfs40_call_sync_done,
998 };
999
1000 static int nfs4_call_sync_sequence(struct rpc_clnt *clnt,
1001                                    struct nfs_server *server,
1002                                    struct rpc_message *msg,
1003                                    struct nfs4_sequence_args *args,
1004                                    struct nfs4_sequence_res *res)
1005 {
1006         int ret;
1007         struct rpc_task *task;
1008         struct nfs_client *clp = server->nfs_client;
1009         struct nfs4_call_sync_data data = {
1010                 .seq_server = server,
1011                 .seq_args = args,
1012                 .seq_res = res,
1013         };
1014         struct rpc_task_setup task_setup = {
1015                 .rpc_client = clnt,
1016                 .rpc_message = msg,
1017                 .callback_ops = clp->cl_mvops->call_sync_ops,
1018                 .callback_data = &data
1019         };
1020
1021         task = rpc_run_task(&task_setup);
1022         if (IS_ERR(task))
1023                 ret = PTR_ERR(task);
1024         else {
1025                 ret = task->tk_status;
1026                 rpc_put_task(task);
1027         }
1028         return ret;
1029 }
1030
1031 int nfs4_call_sync(struct rpc_clnt *clnt,
1032                    struct nfs_server *server,
1033                    struct rpc_message *msg,
1034                    struct nfs4_sequence_args *args,
1035                    struct nfs4_sequence_res *res,
1036                    int cache_reply)
1037 {
1038         nfs4_init_sequence(args, res, cache_reply);
1039         return nfs4_call_sync_sequence(clnt, server, msg, args, res);
1040 }
1041
1042 static void update_changeattr(struct inode *dir, struct nfs4_change_info *cinfo)
1043 {
1044         struct nfs_inode *nfsi = NFS_I(dir);
1045
1046         spin_lock(&dir->i_lock);
1047         nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1048         if (!cinfo->atomic || cinfo->before != dir->i_version)
1049                 nfs_force_lookup_revalidate(dir);
1050         dir->i_version = cinfo->after;
1051         nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1052         nfs_fscache_invalidate(dir);
1053         spin_unlock(&dir->i_lock);
1054 }
1055
1056 struct nfs4_opendata {
1057         struct kref kref;
1058         struct nfs_openargs o_arg;
1059         struct nfs_openres o_res;
1060         struct nfs_open_confirmargs c_arg;
1061         struct nfs_open_confirmres c_res;
1062         struct nfs4_string owner_name;
1063         struct nfs4_string group_name;
1064         struct nfs4_label *a_label;
1065         struct nfs_fattr f_attr;
1066         struct nfs4_label *f_label;
1067         struct dentry *dir;
1068         struct dentry *dentry;
1069         struct nfs4_state_owner *owner;
1070         struct nfs4_state *state;
1071         struct iattr attrs;
1072         unsigned long timestamp;
1073         unsigned int rpc_done : 1;
1074         unsigned int file_created : 1;
1075         unsigned int is_recover : 1;
1076         int rpc_status;
1077         int cancelled;
1078 };
1079
1080 static bool nfs4_clear_cap_atomic_open_v1(struct nfs_server *server,
1081                 int err, struct nfs4_exception *exception)
1082 {
1083         if (err != -EINVAL)
1084                 return false;
1085         if (!(server->caps & NFS_CAP_ATOMIC_OPEN_V1))
1086                 return false;
1087         server->caps &= ~NFS_CAP_ATOMIC_OPEN_V1;
1088         exception->retry = 1;
1089         return true;
1090 }
1091
1092 static u32
1093 nfs4_map_atomic_open_share(struct nfs_server *server,
1094                 fmode_t fmode, int openflags)
1095 {
1096         u32 res = 0;
1097
1098         switch (fmode & (FMODE_READ | FMODE_WRITE)) {
1099         case FMODE_READ:
1100                 res = NFS4_SHARE_ACCESS_READ;
1101                 break;
1102         case FMODE_WRITE:
1103                 res = NFS4_SHARE_ACCESS_WRITE;
1104                 break;
1105         case FMODE_READ|FMODE_WRITE:
1106                 res = NFS4_SHARE_ACCESS_BOTH;
1107         }
1108         if (!(server->caps & NFS_CAP_ATOMIC_OPEN_V1))
1109                 goto out;
1110         /* Want no delegation if we're using O_DIRECT */
1111         if (openflags & O_DIRECT)
1112                 res |= NFS4_SHARE_WANT_NO_DELEG;
1113 out:
1114         return res;
1115 }
1116
1117 static enum open_claim_type4
1118 nfs4_map_atomic_open_claim(struct nfs_server *server,
1119                 enum open_claim_type4 claim)
1120 {
1121         if (server->caps & NFS_CAP_ATOMIC_OPEN_V1)
1122                 return claim;
1123         switch (claim) {
1124         default:
1125                 return claim;
1126         case NFS4_OPEN_CLAIM_FH:
1127                 return NFS4_OPEN_CLAIM_NULL;
1128         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
1129                 return NFS4_OPEN_CLAIM_DELEGATE_CUR;
1130         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
1131                 return NFS4_OPEN_CLAIM_DELEGATE_PREV;
1132         }
1133 }
1134
1135 static void nfs4_init_opendata_res(struct nfs4_opendata *p)
1136 {
1137         p->o_res.f_attr = &p->f_attr;
1138         p->o_res.f_label = p->f_label;
1139         p->o_res.seqid = p->o_arg.seqid;
1140         p->c_res.seqid = p->c_arg.seqid;
1141         p->o_res.server = p->o_arg.server;
1142         p->o_res.access_request = p->o_arg.access;
1143         nfs_fattr_init(&p->f_attr);
1144         nfs_fattr_init_names(&p->f_attr, &p->owner_name, &p->group_name);
1145 }
1146
1147 static struct nfs4_opendata *nfs4_opendata_alloc(struct dentry *dentry,
1148                 struct nfs4_state_owner *sp, fmode_t fmode, int flags,
1149                 const struct iattr *attrs,
1150                 struct nfs4_label *label,
1151                 enum open_claim_type4 claim,
1152                 gfp_t gfp_mask)
1153 {
1154         struct dentry *parent = dget_parent(dentry);
1155         struct inode *dir = d_inode(parent);
1156         struct nfs_server *server = NFS_SERVER(dir);
1157         struct nfs_seqid *(*alloc_seqid)(struct nfs_seqid_counter *, gfp_t);
1158         struct nfs4_opendata *p;
1159
1160         p = kzalloc(sizeof(*p), gfp_mask);
1161         if (p == NULL)
1162                 goto err;
1163
1164         p->f_label = nfs4_label_alloc(server, gfp_mask);
1165         if (IS_ERR(p->f_label))
1166                 goto err_free_p;
1167
1168         p->a_label = nfs4_label_alloc(server, gfp_mask);
1169         if (IS_ERR(p->a_label))
1170                 goto err_free_f;
1171
1172         alloc_seqid = server->nfs_client->cl_mvops->alloc_seqid;
1173         p->o_arg.seqid = alloc_seqid(&sp->so_seqid, gfp_mask);
1174         if (IS_ERR(p->o_arg.seqid))
1175                 goto err_free_label;
1176         nfs_sb_active(dentry->d_sb);
1177         p->dentry = dget(dentry);
1178         p->dir = parent;
1179         p->owner = sp;
1180         atomic_inc(&sp->so_count);
1181         p->o_arg.open_flags = flags;
1182         p->o_arg.fmode = fmode & (FMODE_READ|FMODE_WRITE);
1183         p->o_arg.share_access = nfs4_map_atomic_open_share(server,
1184                         fmode, flags);
1185         /* don't put an ACCESS op in OPEN compound if O_EXCL, because ACCESS
1186          * will return permission denied for all bits until close */
1187         if (!(flags & O_EXCL)) {
1188                 /* ask server to check for all possible rights as results
1189                  * are cached */
1190                 p->o_arg.access = NFS4_ACCESS_READ | NFS4_ACCESS_MODIFY |
1191                                   NFS4_ACCESS_EXTEND | NFS4_ACCESS_EXECUTE;
1192         }
1193         p->o_arg.clientid = server->nfs_client->cl_clientid;
1194         p->o_arg.id.create_time = ktime_to_ns(sp->so_seqid.create_time);
1195         p->o_arg.id.uniquifier = sp->so_seqid.owner_id;
1196         p->o_arg.name = &dentry->d_name;
1197         p->o_arg.server = server;
1198         p->o_arg.bitmask = nfs4_bitmask(server, label);
1199         p->o_arg.open_bitmap = &nfs4_fattr_bitmap[0];
1200         p->o_arg.label = nfs4_label_copy(p->a_label, label);
1201         p->o_arg.claim = nfs4_map_atomic_open_claim(server, claim);
1202         switch (p->o_arg.claim) {
1203         case NFS4_OPEN_CLAIM_NULL:
1204         case NFS4_OPEN_CLAIM_DELEGATE_CUR:
1205         case NFS4_OPEN_CLAIM_DELEGATE_PREV:
1206                 p->o_arg.fh = NFS_FH(dir);
1207                 break;
1208         case NFS4_OPEN_CLAIM_PREVIOUS:
1209         case NFS4_OPEN_CLAIM_FH:
1210         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
1211         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
1212                 p->o_arg.fh = NFS_FH(d_inode(dentry));
1213         }
1214         if (attrs != NULL && attrs->ia_valid != 0) {
1215                 __u32 verf[2];
1216
1217                 p->o_arg.u.attrs = &p->attrs;
1218                 memcpy(&p->attrs, attrs, sizeof(p->attrs));
1219
1220                 verf[0] = jiffies;
1221                 verf[1] = current->pid;
1222                 memcpy(p->o_arg.u.verifier.data, verf,
1223                                 sizeof(p->o_arg.u.verifier.data));
1224         }
1225         p->c_arg.fh = &p->o_res.fh;
1226         p->c_arg.stateid = &p->o_res.stateid;
1227         p->c_arg.seqid = p->o_arg.seqid;
1228         nfs4_init_opendata_res(p);
1229         kref_init(&p->kref);
1230         return p;
1231
1232 err_free_label:
1233         nfs4_label_free(p->a_label);
1234 err_free_f:
1235         nfs4_label_free(p->f_label);
1236 err_free_p:
1237         kfree(p);
1238 err:
1239         dput(parent);
1240         return NULL;
1241 }
1242
1243 static void nfs4_opendata_free(struct kref *kref)
1244 {
1245         struct nfs4_opendata *p = container_of(kref,
1246                         struct nfs4_opendata, kref);
1247         struct super_block *sb = p->dentry->d_sb;
1248
1249         nfs_free_seqid(p->o_arg.seqid);
1250         nfs4_sequence_free_slot(&p->o_res.seq_res);
1251         if (p->state != NULL)
1252                 nfs4_put_open_state(p->state);
1253         nfs4_put_state_owner(p->owner);
1254
1255         nfs4_label_free(p->a_label);
1256         nfs4_label_free(p->f_label);
1257
1258         dput(p->dir);
1259         dput(p->dentry);
1260         nfs_sb_deactive(sb);
1261         nfs_fattr_free_names(&p->f_attr);
1262         kfree(p->f_attr.mdsthreshold);
1263         kfree(p);
1264 }
1265
1266 static void nfs4_opendata_put(struct nfs4_opendata *p)
1267 {
1268         if (p != NULL)
1269                 kref_put(&p->kref, nfs4_opendata_free);
1270 }
1271
1272 static int nfs4_wait_for_completion_rpc_task(struct rpc_task *task)
1273 {
1274         int ret;
1275
1276         ret = rpc_wait_for_completion_task(task);
1277         return ret;
1278 }
1279
1280 static bool nfs4_mode_match_open_stateid(struct nfs4_state *state,
1281                 fmode_t fmode)
1282 {
1283         switch(fmode & (FMODE_READ|FMODE_WRITE)) {
1284         case FMODE_READ|FMODE_WRITE:
1285                 return state->n_rdwr != 0;
1286         case FMODE_WRITE:
1287                 return state->n_wronly != 0;
1288         case FMODE_READ:
1289                 return state->n_rdonly != 0;
1290         }
1291         WARN_ON_ONCE(1);
1292         return false;
1293 }
1294
1295 static int can_open_cached(struct nfs4_state *state, fmode_t mode, int open_mode)
1296 {
1297         int ret = 0;
1298
1299         if (open_mode & (O_EXCL|O_TRUNC))
1300                 goto out;
1301         switch (mode & (FMODE_READ|FMODE_WRITE)) {
1302                 case FMODE_READ:
1303                         ret |= test_bit(NFS_O_RDONLY_STATE, &state->flags) != 0
1304                                 && state->n_rdonly != 0;
1305                         break;
1306                 case FMODE_WRITE:
1307                         ret |= test_bit(NFS_O_WRONLY_STATE, &state->flags) != 0
1308                                 && state->n_wronly != 0;
1309                         break;
1310                 case FMODE_READ|FMODE_WRITE:
1311                         ret |= test_bit(NFS_O_RDWR_STATE, &state->flags) != 0
1312                                 && state->n_rdwr != 0;
1313         }
1314 out:
1315         return ret;
1316 }
1317
1318 static int can_open_delegated(struct nfs_delegation *delegation, fmode_t fmode,
1319                 enum open_claim_type4 claim)
1320 {
1321         if (delegation == NULL)
1322                 return 0;
1323         if ((delegation->type & fmode) != fmode)
1324                 return 0;
1325         if (test_bit(NFS_DELEGATION_RETURNING, &delegation->flags))
1326                 return 0;
1327         switch (claim) {
1328         case NFS4_OPEN_CLAIM_NULL:
1329         case NFS4_OPEN_CLAIM_FH:
1330                 break;
1331         case NFS4_OPEN_CLAIM_PREVIOUS:
1332                 if (!test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags))
1333                         break;
1334         default:
1335                 return 0;
1336         }
1337         nfs_mark_delegation_referenced(delegation);
1338         return 1;
1339 }
1340
1341 static void update_open_stateflags(struct nfs4_state *state, fmode_t fmode)
1342 {
1343         switch (fmode) {
1344                 case FMODE_WRITE:
1345                         state->n_wronly++;
1346                         break;
1347                 case FMODE_READ:
1348                         state->n_rdonly++;
1349                         break;
1350                 case FMODE_READ|FMODE_WRITE:
1351                         state->n_rdwr++;
1352         }
1353         nfs4_state_set_mode_locked(state, state->state | fmode);
1354 }
1355
1356 static void nfs_test_and_clear_all_open_stateid(struct nfs4_state *state)
1357 {
1358         struct nfs_client *clp = state->owner->so_server->nfs_client;
1359         bool need_recover = false;
1360
1361         if (test_and_clear_bit(NFS_O_RDONLY_STATE, &state->flags) && state->n_rdonly)
1362                 need_recover = true;
1363         if (test_and_clear_bit(NFS_O_WRONLY_STATE, &state->flags) && state->n_wronly)
1364                 need_recover = true;
1365         if (test_and_clear_bit(NFS_O_RDWR_STATE, &state->flags) && state->n_rdwr)
1366                 need_recover = true;
1367         if (need_recover)
1368                 nfs4_state_mark_reclaim_nograce(clp, state);
1369 }
1370
1371 static bool nfs_need_update_open_stateid(struct nfs4_state *state,
1372                 nfs4_stateid *stateid)
1373 {
1374         if (test_and_set_bit(NFS_OPEN_STATE, &state->flags) == 0)
1375                 return true;
1376         if (!nfs4_stateid_match_other(stateid, &state->open_stateid)) {
1377                 nfs_test_and_clear_all_open_stateid(state);
1378                 return true;
1379         }
1380         if (nfs4_stateid_is_newer(stateid, &state->open_stateid))
1381                 return true;
1382         return false;
1383 }
1384
1385 static void nfs_resync_open_stateid_locked(struct nfs4_state *state)
1386 {
1387         if (!(state->n_wronly || state->n_rdonly || state->n_rdwr))
1388                 return;
1389         if (state->n_wronly)
1390                 set_bit(NFS_O_WRONLY_STATE, &state->flags);
1391         if (state->n_rdonly)
1392                 set_bit(NFS_O_RDONLY_STATE, &state->flags);
1393         if (state->n_rdwr)
1394                 set_bit(NFS_O_RDWR_STATE, &state->flags);
1395         set_bit(NFS_OPEN_STATE, &state->flags);
1396 }
1397
1398 static void nfs_clear_open_stateid_locked(struct nfs4_state *state,
1399                 nfs4_stateid *arg_stateid,
1400                 nfs4_stateid *stateid, fmode_t fmode)
1401 {
1402         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1403         switch (fmode & (FMODE_READ|FMODE_WRITE)) {
1404         case FMODE_WRITE:
1405                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1406                 break;
1407         case FMODE_READ:
1408                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1409                 break;
1410         case 0:
1411                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1412                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1413                 clear_bit(NFS_OPEN_STATE, &state->flags);
1414         }
1415         if (stateid == NULL)
1416                 return;
1417         /* Handle races with OPEN */
1418         if (!nfs4_stateid_match_other(arg_stateid, &state->open_stateid) ||
1419             (nfs4_stateid_match_other(stateid, &state->open_stateid) &&
1420             !nfs4_stateid_is_newer(stateid, &state->open_stateid))) {
1421                 nfs_resync_open_stateid_locked(state);
1422                 return;
1423         }
1424         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1425                 nfs4_stateid_copy(&state->stateid, stateid);
1426         nfs4_stateid_copy(&state->open_stateid, stateid);
1427 }
1428
1429 static void nfs_clear_open_stateid(struct nfs4_state *state,
1430         nfs4_stateid *arg_stateid,
1431         nfs4_stateid *stateid, fmode_t fmode)
1432 {
1433         write_seqlock(&state->seqlock);
1434         nfs_clear_open_stateid_locked(state, arg_stateid, stateid, fmode);
1435         write_sequnlock(&state->seqlock);
1436         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags))
1437                 nfs4_schedule_state_manager(state->owner->so_server->nfs_client);
1438 }
1439
1440 static void nfs_set_open_stateid_locked(struct nfs4_state *state, nfs4_stateid *stateid, fmode_t fmode)
1441 {
1442         switch (fmode) {
1443                 case FMODE_READ:
1444                         set_bit(NFS_O_RDONLY_STATE, &state->flags);
1445                         break;
1446                 case FMODE_WRITE:
1447                         set_bit(NFS_O_WRONLY_STATE, &state->flags);
1448                         break;
1449                 case FMODE_READ|FMODE_WRITE:
1450                         set_bit(NFS_O_RDWR_STATE, &state->flags);
1451         }
1452         if (!nfs_need_update_open_stateid(state, stateid))
1453                 return;
1454         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1455                 nfs4_stateid_copy(&state->stateid, stateid);
1456         nfs4_stateid_copy(&state->open_stateid, stateid);
1457 }
1458
1459 static void __update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, const nfs4_stateid *deleg_stateid, fmode_t fmode)
1460 {
1461         /*
1462          * Protect the call to nfs4_state_set_mode_locked and
1463          * serialise the stateid update
1464          */
1465         spin_lock(&state->owner->so_lock);
1466         write_seqlock(&state->seqlock);
1467         if (deleg_stateid != NULL) {
1468                 nfs4_stateid_copy(&state->stateid, deleg_stateid);
1469                 set_bit(NFS_DELEGATED_STATE, &state->flags);
1470         }
1471         if (open_stateid != NULL)
1472                 nfs_set_open_stateid_locked(state, open_stateid, fmode);
1473         write_sequnlock(&state->seqlock);
1474         update_open_stateflags(state, fmode);
1475         spin_unlock(&state->owner->so_lock);
1476 }
1477
1478 static int update_open_stateid(struct nfs4_state *state, nfs4_stateid *open_stateid, nfs4_stateid *delegation, fmode_t fmode)
1479 {
1480         struct nfs_inode *nfsi = NFS_I(state->inode);
1481         struct nfs_delegation *deleg_cur;
1482         int ret = 0;
1483
1484         fmode &= (FMODE_READ|FMODE_WRITE);
1485
1486         rcu_read_lock();
1487         deleg_cur = rcu_dereference(nfsi->delegation);
1488         if (deleg_cur == NULL)
1489                 goto no_delegation;
1490
1491         spin_lock(&deleg_cur->lock);
1492         if (rcu_dereference(nfsi->delegation) != deleg_cur ||
1493            test_bit(NFS_DELEGATION_RETURNING, &deleg_cur->flags) ||
1494             (deleg_cur->type & fmode) != fmode)
1495                 goto no_delegation_unlock;
1496
1497         if (delegation == NULL)
1498                 delegation = &deleg_cur->stateid;
1499         else if (!nfs4_stateid_match(&deleg_cur->stateid, delegation))
1500                 goto no_delegation_unlock;
1501
1502         nfs_mark_delegation_referenced(deleg_cur);
1503         __update_open_stateid(state, open_stateid, &deleg_cur->stateid, fmode);
1504         ret = 1;
1505 no_delegation_unlock:
1506         spin_unlock(&deleg_cur->lock);
1507 no_delegation:
1508         rcu_read_unlock();
1509
1510         if (!ret && open_stateid != NULL) {
1511                 __update_open_stateid(state, open_stateid, NULL, fmode);
1512                 ret = 1;
1513         }
1514         if (test_bit(NFS_STATE_RECLAIM_NOGRACE, &state->flags))
1515                 nfs4_schedule_state_manager(state->owner->so_server->nfs_client);
1516
1517         return ret;
1518 }
1519
1520 static bool nfs4_update_lock_stateid(struct nfs4_lock_state *lsp,
1521                 const nfs4_stateid *stateid)
1522 {
1523         struct nfs4_state *state = lsp->ls_state;
1524         bool ret = false;
1525
1526         spin_lock(&state->state_lock);
1527         if (!nfs4_stateid_match_other(stateid, &lsp->ls_stateid))
1528                 goto out_noupdate;
1529         if (!nfs4_stateid_is_newer(stateid, &lsp->ls_stateid))
1530                 goto out_noupdate;
1531         nfs4_stateid_copy(&lsp->ls_stateid, stateid);
1532         ret = true;
1533 out_noupdate:
1534         spin_unlock(&state->state_lock);
1535         return ret;
1536 }
1537
1538 static void nfs4_return_incompatible_delegation(struct inode *inode, fmode_t fmode)
1539 {
1540         struct nfs_delegation *delegation;
1541
1542         rcu_read_lock();
1543         delegation = rcu_dereference(NFS_I(inode)->delegation);
1544         if (delegation == NULL || (delegation->type & fmode) == fmode) {
1545                 rcu_read_unlock();
1546                 return;
1547         }
1548         rcu_read_unlock();
1549         nfs4_inode_return_delegation(inode);
1550 }
1551
1552 static struct nfs4_state *nfs4_try_open_cached(struct nfs4_opendata *opendata)
1553 {
1554         struct nfs4_state *state = opendata->state;
1555         struct nfs_inode *nfsi = NFS_I(state->inode);
1556         struct nfs_delegation *delegation;
1557         int open_mode = opendata->o_arg.open_flags;
1558         fmode_t fmode = opendata->o_arg.fmode;
1559         enum open_claim_type4 claim = opendata->o_arg.claim;
1560         nfs4_stateid stateid;
1561         int ret = -EAGAIN;
1562
1563         for (;;) {
1564                 spin_lock(&state->owner->so_lock);
1565                 if (can_open_cached(state, fmode, open_mode)) {
1566                         update_open_stateflags(state, fmode);
1567                         spin_unlock(&state->owner->so_lock);
1568                         goto out_return_state;
1569                 }
1570                 spin_unlock(&state->owner->so_lock);
1571                 rcu_read_lock();
1572                 delegation = rcu_dereference(nfsi->delegation);
1573                 if (!can_open_delegated(delegation, fmode, claim)) {
1574                         rcu_read_unlock();
1575                         break;
1576                 }
1577                 /* Save the delegation */
1578                 nfs4_stateid_copy(&stateid, &delegation->stateid);
1579                 rcu_read_unlock();
1580                 nfs_release_seqid(opendata->o_arg.seqid);
1581                 if (!opendata->is_recover) {
1582                         ret = nfs_may_open(state->inode, state->owner->so_cred, open_mode);
1583                         if (ret != 0)
1584                                 goto out;
1585                 }
1586                 ret = -EAGAIN;
1587
1588                 /* Try to update the stateid using the delegation */
1589                 if (update_open_stateid(state, NULL, &stateid, fmode))
1590                         goto out_return_state;
1591         }
1592 out:
1593         return ERR_PTR(ret);
1594 out_return_state:
1595         atomic_inc(&state->count);
1596         return state;
1597 }
1598
1599 static void
1600 nfs4_opendata_check_deleg(struct nfs4_opendata *data, struct nfs4_state *state)
1601 {
1602         struct nfs_client *clp = NFS_SERVER(state->inode)->nfs_client;
1603         struct nfs_delegation *delegation;
1604         int delegation_flags = 0;
1605
1606         rcu_read_lock();
1607         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
1608         if (delegation)
1609                 delegation_flags = delegation->flags;
1610         rcu_read_unlock();
1611         switch (data->o_arg.claim) {
1612         default:
1613                 break;
1614         case NFS4_OPEN_CLAIM_DELEGATE_CUR:
1615         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
1616                 pr_err_ratelimited("NFS: Broken NFSv4 server %s is "
1617                                    "returning a delegation for "
1618                                    "OPEN(CLAIM_DELEGATE_CUR)\n",
1619                                    clp->cl_hostname);
1620                 return;
1621         }
1622         if ((delegation_flags & 1UL<<NFS_DELEGATION_NEED_RECLAIM) == 0)
1623                 nfs_inode_set_delegation(state->inode,
1624                                          data->owner->so_cred,
1625                                          &data->o_res);
1626         else
1627                 nfs_inode_reclaim_delegation(state->inode,
1628                                              data->owner->so_cred,
1629                                              &data->o_res);
1630 }
1631
1632 /*
1633  * Check the inode attributes against the CLAIM_PREVIOUS returned attributes
1634  * and update the nfs4_state.
1635  */
1636 static struct nfs4_state *
1637 _nfs4_opendata_reclaim_to_nfs4_state(struct nfs4_opendata *data)
1638 {
1639         struct inode *inode = data->state->inode;
1640         struct nfs4_state *state = data->state;
1641         int ret;
1642
1643         if (!data->rpc_done) {
1644                 if (data->rpc_status) {
1645                         ret = data->rpc_status;
1646                         goto err;
1647                 }
1648                 /* cached opens have already been processed */
1649                 goto update;
1650         }
1651
1652         ret = nfs_refresh_inode(inode, &data->f_attr);
1653         if (ret)
1654                 goto err;
1655
1656         if (data->o_res.delegation_type != 0)
1657                 nfs4_opendata_check_deleg(data, state);
1658 update:
1659         update_open_stateid(state, &data->o_res.stateid, NULL,
1660                             data->o_arg.fmode);
1661         atomic_inc(&state->count);
1662
1663         return state;
1664 err:
1665         return ERR_PTR(ret);
1666
1667 }
1668
1669 static struct nfs4_state *
1670 _nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data)
1671 {
1672         struct inode *inode;
1673         struct nfs4_state *state = NULL;
1674         int ret;
1675
1676         if (!data->rpc_done) {
1677                 state = nfs4_try_open_cached(data);
1678                 trace_nfs4_cached_open(data->state);
1679                 goto out;
1680         }
1681
1682         ret = -EAGAIN;
1683         if (!(data->f_attr.valid & NFS_ATTR_FATTR))
1684                 goto err;
1685         inode = nfs_fhget(data->dir->d_sb, &data->o_res.fh, &data->f_attr, data->f_label);
1686         ret = PTR_ERR(inode);
1687         if (IS_ERR(inode))
1688                 goto err;
1689         ret = -ENOMEM;
1690         state = nfs4_get_open_state(inode, data->owner);
1691         if (state == NULL)
1692                 goto err_put_inode;
1693         if (data->o_res.delegation_type != 0)
1694                 nfs4_opendata_check_deleg(data, state);
1695         update_open_stateid(state, &data->o_res.stateid, NULL,
1696                         data->o_arg.fmode);
1697         iput(inode);
1698 out:
1699         nfs_release_seqid(data->o_arg.seqid);
1700         return state;
1701 err_put_inode:
1702         iput(inode);
1703 err:
1704         return ERR_PTR(ret);
1705 }
1706
1707 static struct nfs4_state *
1708 nfs4_opendata_to_nfs4_state(struct nfs4_opendata *data)
1709 {
1710         struct nfs4_state *ret;
1711
1712         if (data->o_arg.claim == NFS4_OPEN_CLAIM_PREVIOUS)
1713                 ret =_nfs4_opendata_reclaim_to_nfs4_state(data);
1714         else
1715                 ret = _nfs4_opendata_to_nfs4_state(data);
1716         nfs4_sequence_free_slot(&data->o_res.seq_res);
1717         return ret;
1718 }
1719
1720 static struct nfs_open_context *nfs4_state_find_open_context(struct nfs4_state *state)
1721 {
1722         struct nfs_inode *nfsi = NFS_I(state->inode);
1723         struct nfs_open_context *ctx;
1724
1725         spin_lock(&state->inode->i_lock);
1726         list_for_each_entry(ctx, &nfsi->open_files, list) {
1727                 if (ctx->state != state)
1728                         continue;
1729                 get_nfs_open_context(ctx);
1730                 spin_unlock(&state->inode->i_lock);
1731                 return ctx;
1732         }
1733         spin_unlock(&state->inode->i_lock);
1734         return ERR_PTR(-ENOENT);
1735 }
1736
1737 static struct nfs4_opendata *nfs4_open_recoverdata_alloc(struct nfs_open_context *ctx,
1738                 struct nfs4_state *state, enum open_claim_type4 claim)
1739 {
1740         struct nfs4_opendata *opendata;
1741
1742         opendata = nfs4_opendata_alloc(ctx->dentry, state->owner, 0, 0,
1743                         NULL, NULL, claim, GFP_NOFS);
1744         if (opendata == NULL)
1745                 return ERR_PTR(-ENOMEM);
1746         opendata->state = state;
1747         atomic_inc(&state->count);
1748         return opendata;
1749 }
1750
1751 static int nfs4_open_recover_helper(struct nfs4_opendata *opendata,
1752                 fmode_t fmode)
1753 {
1754         struct nfs4_state *newstate;
1755         int ret;
1756
1757         if (!nfs4_mode_match_open_stateid(opendata->state, fmode))
1758                 return 0;
1759         opendata->o_arg.open_flags = 0;
1760         opendata->o_arg.fmode = fmode;
1761         opendata->o_arg.share_access = nfs4_map_atomic_open_share(
1762                         NFS_SB(opendata->dentry->d_sb),
1763                         fmode, 0);
1764         memset(&opendata->o_res, 0, sizeof(opendata->o_res));
1765         memset(&opendata->c_res, 0, sizeof(opendata->c_res));
1766         nfs4_init_opendata_res(opendata);
1767         ret = _nfs4_recover_proc_open(opendata);
1768         if (ret != 0)
1769                 return ret; 
1770         newstate = nfs4_opendata_to_nfs4_state(opendata);
1771         if (IS_ERR(newstate))
1772                 return PTR_ERR(newstate);
1773         if (newstate != opendata->state)
1774                 ret = -ESTALE;
1775         nfs4_close_state(newstate, fmode);
1776         return ret;
1777 }
1778
1779 static int nfs4_open_recover(struct nfs4_opendata *opendata, struct nfs4_state *state)
1780 {
1781         int ret;
1782
1783         /* Don't trigger recovery in nfs_test_and_clear_all_open_stateid */
1784         clear_bit(NFS_O_RDWR_STATE, &state->flags);
1785         clear_bit(NFS_O_WRONLY_STATE, &state->flags);
1786         clear_bit(NFS_O_RDONLY_STATE, &state->flags);
1787         /* memory barrier prior to reading state->n_* */
1788         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1789         clear_bit(NFS_OPEN_STATE, &state->flags);
1790         smp_rmb();
1791         ret = nfs4_open_recover_helper(opendata, FMODE_READ|FMODE_WRITE);
1792         if (ret != 0)
1793                 return ret;
1794         ret = nfs4_open_recover_helper(opendata, FMODE_WRITE);
1795         if (ret != 0)
1796                 return ret;
1797         ret = nfs4_open_recover_helper(opendata, FMODE_READ);
1798         if (ret != 0)
1799                 return ret;
1800         /*
1801          * We may have performed cached opens for all three recoveries.
1802          * Check if we need to update the current stateid.
1803          */
1804         if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0 &&
1805             !nfs4_stateid_match(&state->stateid, &state->open_stateid)) {
1806                 write_seqlock(&state->seqlock);
1807                 if (test_bit(NFS_DELEGATED_STATE, &state->flags) == 0)
1808                         nfs4_stateid_copy(&state->stateid, &state->open_stateid);
1809                 write_sequnlock(&state->seqlock);
1810         }
1811         return 0;
1812 }
1813
1814 /*
1815  * OPEN_RECLAIM:
1816  *      reclaim state on the server after a reboot.
1817  */
1818 static int _nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1819 {
1820         struct nfs_delegation *delegation;
1821         struct nfs4_opendata *opendata;
1822         fmode_t delegation_type = 0;
1823         int status;
1824
1825         opendata = nfs4_open_recoverdata_alloc(ctx, state,
1826                         NFS4_OPEN_CLAIM_PREVIOUS);
1827         if (IS_ERR(opendata))
1828                 return PTR_ERR(opendata);
1829         rcu_read_lock();
1830         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
1831         if (delegation != NULL && test_bit(NFS_DELEGATION_NEED_RECLAIM, &delegation->flags) != 0)
1832                 delegation_type = delegation->type;
1833         rcu_read_unlock();
1834         opendata->o_arg.u.delegation_type = delegation_type;
1835         status = nfs4_open_recover(opendata, state);
1836         nfs4_opendata_put(opendata);
1837         return status;
1838 }
1839
1840 static int nfs4_do_open_reclaim(struct nfs_open_context *ctx, struct nfs4_state *state)
1841 {
1842         struct nfs_server *server = NFS_SERVER(state->inode);
1843         struct nfs4_exception exception = { };
1844         int err;
1845         do {
1846                 err = _nfs4_do_open_reclaim(ctx, state);
1847                 trace_nfs4_open_reclaim(ctx, 0, err);
1848                 if (nfs4_clear_cap_atomic_open_v1(server, err, &exception))
1849                         continue;
1850                 if (err != -NFS4ERR_DELAY)
1851                         break;
1852                 nfs4_handle_exception(server, err, &exception);
1853         } while (exception.retry);
1854         return err;
1855 }
1856
1857 static int nfs4_open_reclaim(struct nfs4_state_owner *sp, struct nfs4_state *state)
1858 {
1859         struct nfs_open_context *ctx;
1860         int ret;
1861
1862         ctx = nfs4_state_find_open_context(state);
1863         if (IS_ERR(ctx))
1864                 return -EAGAIN;
1865         ret = nfs4_do_open_reclaim(ctx, state);
1866         put_nfs_open_context(ctx);
1867         return ret;
1868 }
1869
1870 static int nfs4_handle_delegation_recall_error(struct nfs_server *server, struct nfs4_state *state, const nfs4_stateid *stateid, int err)
1871 {
1872         switch (err) {
1873                 default:
1874                         printk(KERN_ERR "NFS: %s: unhandled error "
1875                                         "%d.\n", __func__, err);
1876                 case 0:
1877                 case -ENOENT:
1878                 case -EAGAIN:
1879                 case -ESTALE:
1880                         break;
1881                 case -NFS4ERR_BADSESSION:
1882                 case -NFS4ERR_BADSLOT:
1883                 case -NFS4ERR_BAD_HIGH_SLOT:
1884                 case -NFS4ERR_CONN_NOT_BOUND_TO_SESSION:
1885                 case -NFS4ERR_DEADSESSION:
1886                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1887                         nfs4_schedule_session_recovery(server->nfs_client->cl_session, err);
1888                         return -EAGAIN;
1889                 case -NFS4ERR_STALE_CLIENTID:
1890                 case -NFS4ERR_STALE_STATEID:
1891                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1892                 case -NFS4ERR_EXPIRED:
1893                         /* Don't recall a delegation if it was lost */
1894                         nfs4_schedule_lease_recovery(server->nfs_client);
1895                         return -EAGAIN;
1896                 case -NFS4ERR_MOVED:
1897                         nfs4_schedule_migration_recovery(server);
1898                         return -EAGAIN;
1899                 case -NFS4ERR_LEASE_MOVED:
1900                         nfs4_schedule_lease_moved_recovery(server->nfs_client);
1901                         return -EAGAIN;
1902                 case -NFS4ERR_DELEG_REVOKED:
1903                 case -NFS4ERR_ADMIN_REVOKED:
1904                 case -NFS4ERR_BAD_STATEID:
1905                 case -NFS4ERR_OPENMODE:
1906                         nfs_inode_find_state_and_recover(state->inode,
1907                                         stateid);
1908                         nfs4_schedule_stateid_recovery(server, state);
1909                         return -EAGAIN;
1910                 case -NFS4ERR_DELAY:
1911                 case -NFS4ERR_GRACE:
1912                         set_bit(NFS_DELEGATED_STATE, &state->flags);
1913                         ssleep(1);
1914                         return -EAGAIN;
1915                 case -ENOMEM:
1916                 case -NFS4ERR_DENIED:
1917                         /* kill_proc(fl->fl_pid, SIGLOST, 1); */
1918                         return 0;
1919         }
1920         return err;
1921 }
1922
1923 int nfs4_open_delegation_recall(struct nfs_open_context *ctx,
1924                 struct nfs4_state *state, const nfs4_stateid *stateid,
1925                 fmode_t type)
1926 {
1927         struct nfs_server *server = NFS_SERVER(state->inode);
1928         struct nfs4_opendata *opendata;
1929         int err = 0;
1930
1931         opendata = nfs4_open_recoverdata_alloc(ctx, state,
1932                         NFS4_OPEN_CLAIM_DELEG_CUR_FH);
1933         if (IS_ERR(opendata))
1934                 return PTR_ERR(opendata);
1935         nfs4_stateid_copy(&opendata->o_arg.u.delegation, stateid);
1936         write_seqlock(&state->seqlock);
1937         nfs4_stateid_copy(&state->stateid, &state->open_stateid);
1938         write_sequnlock(&state->seqlock);
1939         clear_bit(NFS_DELEGATED_STATE, &state->flags);
1940         switch (type & (FMODE_READ|FMODE_WRITE)) {
1941         case FMODE_READ|FMODE_WRITE:
1942         case FMODE_WRITE:
1943                 err = nfs4_open_recover_helper(opendata, FMODE_READ|FMODE_WRITE);
1944                 if (err)
1945                         break;
1946                 err = nfs4_open_recover_helper(opendata, FMODE_WRITE);
1947                 if (err)
1948                         break;
1949         case FMODE_READ:
1950                 err = nfs4_open_recover_helper(opendata, FMODE_READ);
1951         }
1952         nfs4_opendata_put(opendata);
1953         return nfs4_handle_delegation_recall_error(server, state, stateid, err);
1954 }
1955
1956 static void nfs4_open_confirm_prepare(struct rpc_task *task, void *calldata)
1957 {
1958         struct nfs4_opendata *data = calldata;
1959
1960         nfs40_setup_sequence(data->o_arg.server->nfs_client->cl_slot_tbl,
1961                              &data->c_arg.seq_args, &data->c_res.seq_res, task);
1962 }
1963
1964 static void nfs4_open_confirm_done(struct rpc_task *task, void *calldata)
1965 {
1966         struct nfs4_opendata *data = calldata;
1967
1968         nfs40_sequence_done(task, &data->c_res.seq_res);
1969
1970         data->rpc_status = task->tk_status;
1971         if (data->rpc_status == 0) {
1972                 nfs4_stateid_copy(&data->o_res.stateid, &data->c_res.stateid);
1973                 nfs_confirm_seqid(&data->owner->so_seqid, 0);
1974                 renew_lease(data->o_res.server, data->timestamp);
1975                 data->rpc_done = 1;
1976         }
1977 }
1978
1979 static void nfs4_open_confirm_release(void *calldata)
1980 {
1981         struct nfs4_opendata *data = calldata;
1982         struct nfs4_state *state = NULL;
1983
1984         /* If this request hasn't been cancelled, do nothing */
1985         if (data->cancelled == 0)
1986                 goto out_free;
1987         /* In case of error, no cleanup! */
1988         if (!data->rpc_done)
1989                 goto out_free;
1990         state = nfs4_opendata_to_nfs4_state(data);
1991         if (!IS_ERR(state))
1992                 nfs4_close_state(state, data->o_arg.fmode);
1993 out_free:
1994         nfs4_opendata_put(data);
1995 }
1996
1997 static const struct rpc_call_ops nfs4_open_confirm_ops = {
1998         .rpc_call_prepare = nfs4_open_confirm_prepare,
1999         .rpc_call_done = nfs4_open_confirm_done,
2000         .rpc_release = nfs4_open_confirm_release,
2001 };
2002
2003 /*
2004  * Note: On error, nfs4_proc_open_confirm will free the struct nfs4_opendata
2005  */
2006 static int _nfs4_proc_open_confirm(struct nfs4_opendata *data)
2007 {
2008         struct nfs_server *server = NFS_SERVER(d_inode(data->dir));
2009         struct rpc_task *task;
2010         struct  rpc_message msg = {
2011                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_CONFIRM],
2012                 .rpc_argp = &data->c_arg,
2013                 .rpc_resp = &data->c_res,
2014                 .rpc_cred = data->owner->so_cred,
2015         };
2016         struct rpc_task_setup task_setup_data = {
2017                 .rpc_client = server->client,
2018                 .rpc_message = &msg,
2019                 .callback_ops = &nfs4_open_confirm_ops,
2020                 .callback_data = data,
2021                 .workqueue = nfsiod_workqueue,
2022                 .flags = RPC_TASK_ASYNC,
2023         };
2024         int status;
2025
2026         nfs4_init_sequence(&data->c_arg.seq_args, &data->c_res.seq_res, 1);
2027         kref_get(&data->kref);
2028         data->rpc_done = 0;
2029         data->rpc_status = 0;
2030         data->timestamp = jiffies;
2031         if (data->is_recover)
2032                 nfs4_set_sequence_privileged(&data->c_arg.seq_args);
2033         task = rpc_run_task(&task_setup_data);
2034         if (IS_ERR(task))
2035                 return PTR_ERR(task);
2036         status = nfs4_wait_for_completion_rpc_task(task);
2037         if (status != 0) {
2038                 data->cancelled = 1;
2039                 smp_wmb();
2040         } else
2041                 status = data->rpc_status;
2042         rpc_put_task(task);
2043         return status;
2044 }
2045
2046 static void nfs4_open_prepare(struct rpc_task *task, void *calldata)
2047 {
2048         struct nfs4_opendata *data = calldata;
2049         struct nfs4_state_owner *sp = data->owner;
2050         struct nfs_client *clp = sp->so_server->nfs_client;
2051         enum open_claim_type4 claim = data->o_arg.claim;
2052
2053         if (nfs_wait_on_sequence(data->o_arg.seqid, task) != 0)
2054                 goto out_wait;
2055         /*
2056          * Check if we still need to send an OPEN call, or if we can use
2057          * a delegation instead.
2058          */
2059         if (data->state != NULL) {
2060                 struct nfs_delegation *delegation;
2061
2062                 if (can_open_cached(data->state, data->o_arg.fmode, data->o_arg.open_flags))
2063                         goto out_no_action;
2064                 rcu_read_lock();
2065                 delegation = rcu_dereference(NFS_I(data->state->inode)->delegation);
2066                 if (can_open_delegated(delegation, data->o_arg.fmode, claim))
2067                         goto unlock_no_action;
2068                 rcu_read_unlock();
2069         }
2070         /* Update client id. */
2071         data->o_arg.clientid = clp->cl_clientid;
2072         switch (claim) {
2073         default:
2074                 break;
2075         case NFS4_OPEN_CLAIM_PREVIOUS:
2076         case NFS4_OPEN_CLAIM_DELEG_CUR_FH:
2077         case NFS4_OPEN_CLAIM_DELEG_PREV_FH:
2078                 data->o_arg.open_bitmap = &nfs4_open_noattr_bitmap[0];
2079         case NFS4_OPEN_CLAIM_FH:
2080                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_NOATTR];
2081                 nfs_copy_fh(&data->o_res.fh, data->o_arg.fh);
2082         }
2083         data->timestamp = jiffies;
2084         if (nfs4_setup_sequence(data->o_arg.server,
2085                                 &data->o_arg.seq_args,
2086                                 &data->o_res.seq_res,
2087                                 task) != 0)
2088                 nfs_release_seqid(data->o_arg.seqid);
2089
2090         /* Set the create mode (note dependency on the session type) */
2091         data->o_arg.createmode = NFS4_CREATE_UNCHECKED;
2092         if (data->o_arg.open_flags & O_EXCL) {
2093                 data->o_arg.createmode = NFS4_CREATE_EXCLUSIVE;
2094                 if (nfs4_has_persistent_session(clp))
2095                         data->o_arg.createmode = NFS4_CREATE_GUARDED;
2096                 else if (clp->cl_mvops->minor_version > 0)
2097                         data->o_arg.createmode = NFS4_CREATE_EXCLUSIVE4_1;
2098         }
2099         return;
2100 unlock_no_action:
2101         trace_nfs4_cached_open(data->state);
2102         rcu_read_unlock();
2103 out_no_action:
2104         task->tk_action = NULL;
2105 out_wait:
2106         nfs4_sequence_done(task, &data->o_res.seq_res);
2107 }
2108
2109 static void nfs4_open_done(struct rpc_task *task, void *calldata)
2110 {
2111         struct nfs4_opendata *data = calldata;
2112
2113         data->rpc_status = task->tk_status;
2114
2115         if (!nfs4_sequence_process(task, &data->o_res.seq_res))
2116                 return;
2117
2118         if (task->tk_status == 0) {
2119                 if (data->o_res.f_attr->valid & NFS_ATTR_FATTR_TYPE) {
2120                         switch (data->o_res.f_attr->mode & S_IFMT) {
2121                         case S_IFREG:
2122                                 break;
2123                         case S_IFLNK:
2124                                 data->rpc_status = -ELOOP;
2125                                 break;
2126                         case S_IFDIR:
2127                                 data->rpc_status = -EISDIR;
2128                                 break;
2129                         default:
2130                                 data->rpc_status = -ENOTDIR;
2131                         }
2132                 }
2133                 renew_lease(data->o_res.server, data->timestamp);
2134                 if (!(data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM))
2135                         nfs_confirm_seqid(&data->owner->so_seqid, 0);
2136         }
2137         data->rpc_done = 1;
2138 }
2139
2140 static void nfs4_open_release(void *calldata)
2141 {
2142         struct nfs4_opendata *data = calldata;
2143         struct nfs4_state *state = NULL;
2144
2145         /* If this request hasn't been cancelled, do nothing */
2146         if (data->cancelled == 0)
2147                 goto out_free;
2148         /* In case of error, no cleanup! */
2149         if (data->rpc_status != 0 || !data->rpc_done)
2150                 goto out_free;
2151         /* In case we need an open_confirm, no cleanup! */
2152         if (data->o_res.rflags & NFS4_OPEN_RESULT_CONFIRM)
2153                 goto out_free;
2154         state = nfs4_opendata_to_nfs4_state(data);
2155         if (!IS_ERR(state))
2156                 nfs4_close_state(state, data->o_arg.fmode);
2157 out_free:
2158         nfs4_opendata_put(data);
2159 }
2160
2161 static const struct rpc_call_ops nfs4_open_ops = {
2162         .rpc_call_prepare = nfs4_open_prepare,
2163         .rpc_call_done = nfs4_open_done,
2164         .rpc_release = nfs4_open_release,
2165 };
2166
2167 static int nfs4_run_open_task(struct nfs4_opendata *data, int isrecover)
2168 {
2169         struct inode *dir = d_inode(data->dir);
2170         struct nfs_server *server = NFS_SERVER(dir);
2171         struct nfs_openargs *o_arg = &data->o_arg;
2172         struct nfs_openres *o_res = &data->o_res;
2173         struct rpc_task *task;
2174         struct rpc_message msg = {
2175                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN],
2176                 .rpc_argp = o_arg,
2177                 .rpc_resp = o_res,
2178                 .rpc_cred = data->owner->so_cred,
2179         };
2180         struct rpc_task_setup task_setup_data = {
2181                 .rpc_client = server->client,
2182                 .rpc_message = &msg,
2183                 .callback_ops = &nfs4_open_ops,
2184                 .callback_data = data,
2185                 .workqueue = nfsiod_workqueue,
2186                 .flags = RPC_TASK_ASYNC,
2187         };
2188         int status;
2189
2190         nfs4_init_sequence(&o_arg->seq_args, &o_res->seq_res, 1);
2191         kref_get(&data->kref);
2192         data->rpc_done = 0;
2193         data->rpc_status = 0;
2194         data->cancelled = 0;
2195         data->is_recover = 0;
2196         if (isrecover) {
2197                 nfs4_set_sequence_privileged(&o_arg->seq_args);
2198                 data->is_recover = 1;
2199         }
2200         task = rpc_run_task(&task_setup_data);
2201         if (IS_ERR(task))
2202                 return PTR_ERR(task);
2203         status = nfs4_wait_for_completion_rpc_task(task);
2204         if (status != 0) {
2205                 data->cancelled = 1;
2206                 smp_wmb();
2207         } else
2208                 status = data->rpc_status;
2209         rpc_put_task(task);
2210
2211         return status;
2212 }
2213
2214 static int _nfs4_recover_proc_open(struct nfs4_opendata *data)
2215 {
2216         struct inode *dir = d_inode(data->dir);
2217         struct nfs_openres *o_res = &data->o_res;
2218         int status;
2219
2220         status = nfs4_run_open_task(data, 1);
2221         if (status != 0 || !data->rpc_done)
2222                 return status;
2223
2224         nfs_fattr_map_and_free_names(NFS_SERVER(dir), &data->f_attr);
2225
2226         if (o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
2227                 status = _nfs4_proc_open_confirm(data);
2228                 if (status != 0)
2229                         return status;
2230         }
2231
2232         return status;
2233 }
2234
2235 /*
2236  * Additional permission checks in order to distinguish between an
2237  * open for read, and an open for execute. This works around the
2238  * fact that NFSv4 OPEN treats read and execute permissions as being
2239  * the same.
2240  * Note that in the non-execute case, we want to turn off permission
2241  * checking if we just created a new file (POSIX open() semantics).
2242  */
2243 static int nfs4_opendata_access(struct rpc_cred *cred,
2244                                 struct nfs4_opendata *opendata,
2245                                 struct nfs4_state *state, fmode_t fmode,
2246                                 int openflags)
2247 {
2248         struct nfs_access_entry cache;
2249         u32 mask;
2250
2251         /* access call failed or for some reason the server doesn't
2252          * support any access modes -- defer access call until later */
2253         if (opendata->o_res.access_supported == 0)
2254                 return 0;
2255
2256         mask = 0;
2257         /*
2258          * Use openflags to check for exec, because fmode won't
2259          * always have FMODE_EXEC set when file open for exec.
2260          */
2261         if (openflags & __FMODE_EXEC) {
2262                 /* ONLY check for exec rights */
2263                 mask = MAY_EXEC;
2264         } else if ((fmode & FMODE_READ) && !opendata->file_created)
2265                 mask = MAY_READ;
2266
2267         cache.cred = cred;
2268         cache.jiffies = jiffies;
2269         nfs_access_set_mask(&cache, opendata->o_res.access_result);
2270         nfs_access_add_cache(state->inode, &cache);
2271
2272         if ((mask & ~cache.mask & (MAY_READ | MAY_EXEC)) == 0)
2273                 return 0;
2274
2275         /* even though OPEN succeeded, access is denied. Close the file */
2276         nfs4_close_state(state, fmode);
2277         return -EACCES;
2278 }
2279
2280 /*
2281  * Note: On error, nfs4_proc_open will free the struct nfs4_opendata
2282  */
2283 static int _nfs4_proc_open(struct nfs4_opendata *data)
2284 {
2285         struct inode *dir = d_inode(data->dir);
2286         struct nfs_server *server = NFS_SERVER(dir);
2287         struct nfs_openargs *o_arg = &data->o_arg;
2288         struct nfs_openres *o_res = &data->o_res;
2289         int status;
2290
2291         status = nfs4_run_open_task(data, 0);
2292         if (!data->rpc_done)
2293                 return status;
2294         if (status != 0) {
2295                 if (status == -NFS4ERR_BADNAME &&
2296                                 !(o_arg->open_flags & O_CREAT))
2297                         return -ENOENT;
2298                 return status;
2299         }
2300
2301         nfs_fattr_map_and_free_names(server, &data->f_attr);
2302
2303         if (o_arg->open_flags & O_CREAT) {
2304                 update_changeattr(dir, &o_res->cinfo);
2305                 if (o_arg->open_flags & O_EXCL)
2306                         data->file_created = 1;
2307                 else if (o_res->cinfo.before != o_res->cinfo.after)
2308                         data->file_created = 1;
2309         }
2310         if ((o_res->rflags & NFS4_OPEN_RESULT_LOCKTYPE_POSIX) == 0)
2311                 server->caps &= ~NFS_CAP_POSIX_LOCK;
2312         if(o_res->rflags & NFS4_OPEN_RESULT_CONFIRM) {
2313                 status = _nfs4_proc_open_confirm(data);
2314                 if (status != 0)
2315                         return status;
2316         }
2317         if (!(o_res->f_attr->valid & NFS_ATTR_FATTR))
2318                 nfs4_proc_getattr(server, &o_res->fh, o_res->f_attr, o_res->f_label);
2319         return 0;
2320 }
2321
2322 static int nfs4_recover_expired_lease(struct nfs_server *server)
2323 {
2324         return nfs4_client_recover_expired_lease(server->nfs_client);
2325 }
2326
2327 /*
2328  * OPEN_EXPIRED:
2329  *      reclaim state on the server after a network partition.
2330  *      Assumes caller holds the appropriate lock
2331  */
2332 static int _nfs4_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
2333 {
2334         struct nfs4_opendata *opendata;
2335         int ret;
2336
2337         opendata = nfs4_open_recoverdata_alloc(ctx, state,
2338                         NFS4_OPEN_CLAIM_FH);
2339         if (IS_ERR(opendata))
2340                 return PTR_ERR(opendata);
2341         ret = nfs4_open_recover(opendata, state);
2342         if (ret == -ESTALE)
2343                 d_drop(ctx->dentry);
2344         nfs4_opendata_put(opendata);
2345         return ret;
2346 }
2347
2348 static int nfs4_do_open_expired(struct nfs_open_context *ctx, struct nfs4_state *state)
2349 {
2350         struct nfs_server *server = NFS_SERVER(state->inode);
2351         struct nfs4_exception exception = { };
2352         int err;
2353
2354         do {
2355                 err = _nfs4_open_expired(ctx, state);
2356                 trace_nfs4_open_expired(ctx, 0, err);
2357                 if (nfs4_clear_cap_atomic_open_v1(server, err, &exception))
2358                         continue;
2359                 switch (err) {
2360                 default:
2361                         goto out;
2362                 case -NFS4ERR_GRACE:
2363                 case -NFS4ERR_DELAY:
2364                         nfs4_handle_exception(server, err, &exception);
2365                         err = 0;
2366                 }
2367         } while (exception.retry);
2368 out:
2369         return err;
2370 }
2371
2372 static int nfs4_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
2373 {
2374         struct nfs_open_context *ctx;
2375         int ret;
2376
2377         ctx = nfs4_state_find_open_context(state);
2378         if (IS_ERR(ctx))
2379                 return -EAGAIN;
2380         ret = nfs4_do_open_expired(ctx, state);
2381         put_nfs_open_context(ctx);
2382         return ret;
2383 }
2384
2385 static void nfs_finish_clear_delegation_stateid(struct nfs4_state *state)
2386 {
2387         nfs_remove_bad_delegation(state->inode);
2388         write_seqlock(&state->seqlock);
2389         nfs4_stateid_copy(&state->stateid, &state->open_stateid);
2390         write_sequnlock(&state->seqlock);
2391         clear_bit(NFS_DELEGATED_STATE, &state->flags);
2392 }
2393
2394 static void nfs40_clear_delegation_stateid(struct nfs4_state *state)
2395 {
2396         if (rcu_access_pointer(NFS_I(state->inode)->delegation) != NULL)
2397                 nfs_finish_clear_delegation_stateid(state);
2398 }
2399
2400 static int nfs40_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
2401 {
2402         /* NFSv4.0 doesn't allow for delegation recovery on open expire */
2403         nfs40_clear_delegation_stateid(state);
2404         return nfs4_open_expired(sp, state);
2405 }
2406
2407 #if defined(CONFIG_NFS_V4_1)
2408 static void nfs41_check_delegation_stateid(struct nfs4_state *state)
2409 {
2410         struct nfs_server *server = NFS_SERVER(state->inode);
2411         nfs4_stateid stateid;
2412         struct nfs_delegation *delegation;
2413         struct rpc_cred *cred;
2414         int status;
2415
2416         /* Get the delegation credential for use by test/free_stateid */
2417         rcu_read_lock();
2418         delegation = rcu_dereference(NFS_I(state->inode)->delegation);
2419         if (delegation == NULL) {
2420                 rcu_read_unlock();
2421                 return;
2422         }
2423
2424         nfs4_stateid_copy(&stateid, &delegation->stateid);
2425         cred = get_rpccred(delegation->cred);
2426         rcu_read_unlock();
2427         status = nfs41_test_stateid(server, &stateid, cred);
2428         trace_nfs4_test_delegation_stateid(state, NULL, status);
2429
2430         if (status != NFS_OK) {
2431                 /* Free the stateid unless the server explicitly
2432                  * informs us the stateid is unrecognized. */
2433                 if (status != -NFS4ERR_BAD_STATEID)
2434                         nfs41_free_stateid(server, &stateid, cred);
2435                 nfs_finish_clear_delegation_stateid(state);
2436         }
2437
2438         put_rpccred(cred);
2439 }
2440
2441 /**
2442  * nfs41_check_open_stateid - possibly free an open stateid
2443  *
2444  * @state: NFSv4 state for an inode
2445  *
2446  * Returns NFS_OK if recovery for this stateid is now finished.
2447  * Otherwise a negative NFS4ERR value is returned.
2448  */
2449 static int nfs41_check_open_stateid(struct nfs4_state *state)
2450 {
2451         struct nfs_server *server = NFS_SERVER(state->inode);
2452         nfs4_stateid *stateid = &state->open_stateid;
2453         struct rpc_cred *cred = state->owner->so_cred;
2454         int status;
2455
2456         /* If a state reset has been done, test_stateid is unneeded */
2457         if ((test_bit(NFS_O_RDONLY_STATE, &state->flags) == 0) &&
2458             (test_bit(NFS_O_WRONLY_STATE, &state->flags) == 0) &&
2459             (test_bit(NFS_O_RDWR_STATE, &state->flags) == 0))
2460                 return -NFS4ERR_BAD_STATEID;
2461
2462         status = nfs41_test_stateid(server, stateid, cred);
2463         trace_nfs4_test_open_stateid(state, NULL, status);
2464         if (status != NFS_OK) {
2465                 /* Free the stateid unless the server explicitly
2466                  * informs us the stateid is unrecognized. */
2467                 if (status != -NFS4ERR_BAD_STATEID)
2468                         nfs41_free_stateid(server, stateid, cred);
2469
2470                 clear_bit(NFS_O_RDONLY_STATE, &state->flags);
2471                 clear_bit(NFS_O_WRONLY_STATE, &state->flags);
2472                 clear_bit(NFS_O_RDWR_STATE, &state->flags);
2473                 clear_bit(NFS_OPEN_STATE, &state->flags);
2474         }
2475         return status;
2476 }
2477
2478 static int nfs41_open_expired(struct nfs4_state_owner *sp, struct nfs4_state *state)
2479 {
2480         int status;
2481
2482         nfs41_check_delegation_stateid(state);
2483         status = nfs41_check_open_stateid(state);
2484         if (status != NFS_OK)
2485                 status = nfs4_open_expired(sp, state);
2486         return status;
2487 }
2488 #endif
2489
2490 /*
2491  * on an EXCLUSIVE create, the server should send back a bitmask with FATTR4-*
2492  * fields corresponding to attributes that were used to store the verifier.
2493  * Make sure we clobber those fields in the later setattr call
2494  */
2495 static inline void nfs4_exclusive_attrset(struct nfs4_opendata *opendata,
2496                                 struct iattr *sattr, struct nfs4_label **label)
2497 {
2498         const u32 *attrset = opendata->o_res.attrset;
2499
2500         if ((attrset[1] & FATTR4_WORD1_TIME_ACCESS) &&
2501             !(sattr->ia_valid & ATTR_ATIME_SET))
2502                 sattr->ia_valid |= ATTR_ATIME;
2503
2504         if ((attrset[1] & FATTR4_WORD1_TIME_MODIFY) &&
2505             !(sattr->ia_valid & ATTR_MTIME_SET))
2506                 sattr->ia_valid |= ATTR_MTIME;
2507
2508         /* Except MODE, it seems harmless of setting twice. */
2509         if ((attrset[1] & FATTR4_WORD1_MODE))
2510                 sattr->ia_valid &= ~ATTR_MODE;
2511
2512         if (attrset[2] & FATTR4_WORD2_SECURITY_LABEL)
2513                 *label = NULL;
2514 }
2515
2516 static int _nfs4_open_and_get_state(struct nfs4_opendata *opendata,
2517                 fmode_t fmode,
2518                 int flags,
2519                 struct nfs_open_context *ctx)
2520 {
2521         struct nfs4_state_owner *sp = opendata->owner;
2522         struct nfs_server *server = sp->so_server;
2523         struct dentry *dentry;
2524         struct nfs4_state *state;
2525         unsigned int seq;
2526         int ret;
2527
2528         seq = raw_seqcount_begin(&sp->so_reclaim_seqcount);
2529
2530         ret = _nfs4_proc_open(opendata);
2531         if (ret != 0)
2532                 goto out;
2533
2534         state = nfs4_opendata_to_nfs4_state(opendata);
2535         ret = PTR_ERR(state);
2536         if (IS_ERR(state))
2537                 goto out;
2538         if (server->caps & NFS_CAP_POSIX_LOCK)
2539                 set_bit(NFS_STATE_POSIX_LOCKS, &state->flags);
2540         if (opendata->o_res.rflags & NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK)
2541                 set_bit(NFS_STATE_MAY_NOTIFY_LOCK, &state->flags);
2542
2543         dentry = opendata->dentry;
2544         if (d_really_is_negative(dentry)) {
2545                 struct dentry *alias;
2546                 d_drop(dentry);
2547                 alias = d_exact_alias(dentry, state->inode);
2548                 if (!alias)
2549                         alias = d_splice_alias(igrab(state->inode), dentry);
2550                 /* d_splice_alias() can't fail here - it's a non-directory */
2551                 if (alias) {
2552                         dput(ctx->dentry);
2553                         ctx->dentry = dentry = alias;
2554                 }
2555                 nfs_set_verifier(dentry,
2556                                 nfs_save_change_attribute(d_inode(opendata->dir)));
2557         }
2558
2559         ret = nfs4_opendata_access(sp->so_cred, opendata, state, fmode, flags);
2560         if (ret != 0)
2561                 goto out;
2562
2563         ctx->state = state;
2564         if (d_inode(dentry) == state->inode) {
2565                 nfs_inode_attach_open_context(ctx);
2566                 if (read_seqcount_retry(&sp->so_reclaim_seqcount, seq))
2567                         nfs4_schedule_stateid_recovery(server, state);
2568         }
2569 out:
2570         return ret;
2571 }
2572
2573 /*
2574  * Returns a referenced nfs4_state
2575  */
2576 static int _nfs4_do_open(struct inode *dir,
2577                         struct nfs_open_context *ctx,
2578                         int flags,
2579                         struct iattr *sattr,
2580                         struct nfs4_label *label,
2581                         int *opened)
2582 {
2583         struct nfs4_state_owner  *sp;
2584         struct nfs4_state     *state = NULL;
2585         struct nfs_server       *server = NFS_SERVER(dir);
2586         struct nfs4_opendata *opendata;
2587         struct dentry *dentry = ctx->dentry;
2588         struct rpc_cred *cred = ctx->cred;
2589         struct nfs4_threshold **ctx_th = &ctx->mdsthreshold;
2590         fmode_t fmode = ctx->mode & (FMODE_READ|FMODE_WRITE|FMODE_EXEC);
2591         enum open_claim_type4 claim = NFS4_OPEN_CLAIM_NULL;
2592         struct nfs4_label *olabel = NULL;
2593         int status;
2594
2595         /* Protect against reboot recovery conflicts */
2596         status = -ENOMEM;
2597         sp = nfs4_get_state_owner(server, cred, GFP_KERNEL);
2598         if (sp == NULL) {
2599                 dprintk("nfs4_do_open: nfs4_get_state_owner failed!\n");
2600                 goto out_err;
2601         }
2602         status = nfs4_recover_expired_lease(server);
2603         if (status != 0)
2604                 goto err_put_state_owner;
2605         if (d_really_is_positive(dentry))
2606                 nfs4_return_incompatible_delegation(d_inode(dentry), fmode);
2607         status = -ENOMEM;
2608         if (d_really_is_positive(dentry))
2609                 claim = NFS4_OPEN_CLAIM_FH;
2610         opendata = nfs4_opendata_alloc(dentry, sp, fmode, flags, sattr,
2611                         label, claim, GFP_KERNEL);
2612         if (opendata == NULL)
2613                 goto err_put_state_owner;
2614
2615         if (label) {
2616                 olabel = nfs4_label_alloc(server, GFP_KERNEL);
2617                 if (IS_ERR(olabel)) {
2618                         status = PTR_ERR(olabel);
2619                         goto err_opendata_put;
2620                 }
2621         }
2622
2623         if (server->attr_bitmask[2] & FATTR4_WORD2_MDSTHRESHOLD) {
2624                 if (!opendata->f_attr.mdsthreshold) {
2625                         opendata->f_attr.mdsthreshold = pnfs_mdsthreshold_alloc();
2626                         if (!opendata->f_attr.mdsthreshold)
2627                                 goto err_free_label;
2628                 }
2629                 opendata->o_arg.open_bitmap = &nfs4_pnfs_open_bitmap[0];
2630         }
2631         if (d_really_is_positive(dentry))
2632                 opendata->state = nfs4_get_open_state(d_inode(dentry), sp);
2633
2634         status = _nfs4_open_and_get_state(opendata, fmode, flags, ctx);
2635         if (status != 0)
2636                 goto err_free_label;
2637         state = ctx->state;
2638
2639         if ((opendata->o_arg.open_flags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL) &&
2640             (opendata->o_arg.createmode != NFS4_CREATE_GUARDED)) {
2641                 nfs4_exclusive_attrset(opendata, sattr, &label);
2642                 /*
2643                  * send create attributes which was not set by open
2644                  * with an extra setattr.
2645                  */
2646                 if (sattr->ia_valid & NFS4_VALID_ATTRS) {
2647                         nfs_fattr_init(opendata->o_res.f_attr);
2648                         status = nfs4_do_setattr(state->inode, cred,
2649                                         opendata->o_res.f_attr, sattr,
2650                                         state, label, olabel);
2651                         if (status == 0) {
2652                                 nfs_setattr_update_inode(state->inode, sattr,
2653                                                 opendata->o_res.f_attr);
2654                                 nfs_setsecurity(state->inode, opendata->o_res.f_attr, olabel);
2655                         }
2656                 }
2657         }
2658         if (opened && opendata->file_created)
2659                 *opened |= FILE_CREATED;
2660
2661         if (pnfs_use_threshold(ctx_th, opendata->f_attr.mdsthreshold, server)) {
2662                 *ctx_th = opendata->f_attr.mdsthreshold;
2663                 opendata->f_attr.mdsthreshold = NULL;
2664         }
2665
2666         nfs4_label_free(olabel);
2667
2668         nfs4_opendata_put(opendata);
2669         nfs4_put_state_owner(sp);
2670         return 0;
2671 err_free_label:
2672         nfs4_label_free(olabel);
2673 err_opendata_put:
2674         nfs4_opendata_put(opendata);
2675 err_put_state_owner:
2676         nfs4_put_state_owner(sp);
2677 out_err:
2678         return status;
2679 }
2680
2681
2682 static struct nfs4_state *nfs4_do_open(struct inode *dir,
2683                                         struct nfs_open_context *ctx,
2684                                         int flags,
2685                                         struct iattr *sattr,
2686                                         struct nfs4_label *label,
2687                                         int *opened)
2688 {
2689         struct nfs_server *server = NFS_SERVER(dir);
2690         struct nfs4_exception exception = { };
2691         struct nfs4_state *res;
2692         int status;
2693
2694         do {
2695                 status = _nfs4_do_open(dir, ctx, flags, sattr, label, opened);
2696                 res = ctx->state;
2697                 trace_nfs4_open_file(ctx, flags, status);
2698                 if (status == 0)
2699                         break;
2700                 /* NOTE: BAD_SEQID means the server and client disagree about the
2701                  * book-keeping w.r.t. state-changing operations
2702                  * (OPEN/CLOSE/LOCK/LOCKU...)
2703                  * It is actually a sign of a bug on the client or on the server.
2704                  *
2705                  * If we receive a BAD_SEQID error in the particular case of
2706                  * doing an OPEN, we assume that nfs_increment_open_seqid() will
2707                  * have unhashed the old state_owner for us, and that we can
2708                  * therefore safely retry using a new one. We should still warn
2709                  * the user though...
2710                  */
2711                 if (status == -NFS4ERR_BAD_SEQID) {
2712                         pr_warn_ratelimited("NFS: v4 server %s "
2713                                         " returned a bad sequence-id error!\n",
2714                                         NFS_SERVER(dir)->nfs_client->cl_hostname);
2715                         exception.retry = 1;
2716                         continue;
2717                 }
2718                 /*
2719                  * BAD_STATEID on OPEN means that the server cancelled our
2720                  * state before it received the OPEN_CONFIRM.
2721                  * Recover by retrying the request as per the discussion
2722                  * on Page 181 of RFC3530.
2723                  */
2724                 if (status == -NFS4ERR_BAD_STATEID) {
2725                         exception.retry = 1;
2726                         continue;
2727                 }
2728                 if (status == -EAGAIN) {
2729                         /* We must have found a delegation */
2730                         exception.retry = 1;
2731                         continue;
2732                 }
2733                 if (nfs4_clear_cap_atomic_open_v1(server, status, &exception))
2734                         continue;
2735                 res = ERR_PTR(nfs4_handle_exception(server,
2736                                         status, &exception));
2737         } while (exception.retry);
2738         return res;
2739 }
2740
2741 static int _nfs4_do_setattr(struct inode *inode,
2742                             struct nfs_setattrargs *arg,
2743                             struct nfs_setattrres *res,
2744                             struct rpc_cred *cred,
2745                             struct nfs4_state *state)
2746 {
2747         struct nfs_server *server = NFS_SERVER(inode);
2748         struct rpc_message msg = {
2749                 .rpc_proc       = &nfs4_procedures[NFSPROC4_CLNT_SETATTR],
2750                 .rpc_argp       = arg,
2751                 .rpc_resp       = res,
2752                 .rpc_cred       = cred,
2753         };
2754         struct rpc_cred *delegation_cred = NULL;
2755         unsigned long timestamp = jiffies;
2756         fmode_t fmode;
2757         bool truncate;
2758         int status;
2759
2760         nfs_fattr_init(res->fattr);
2761
2762         /* Servers should only apply open mode checks for file size changes */
2763         truncate = (arg->iap->ia_valid & ATTR_SIZE) ? true : false;
2764         fmode = truncate ? FMODE_WRITE : FMODE_READ;
2765
2766         if (nfs4_copy_delegation_stateid(inode, fmode, &arg->stateid, &delegation_cred)) {
2767                 /* Use that stateid */
2768         } else if (truncate && state != NULL) {
2769                 struct nfs_lockowner lockowner = {
2770                         .l_owner = current->files,
2771                         .l_pid = current->tgid,
2772                 };
2773                 if (!nfs4_valid_open_stateid(state))
2774                         return -EBADF;
2775                 if (nfs4_select_rw_stateid(state, FMODE_WRITE, &lockowner,
2776                                 &arg->stateid, &delegation_cred) == -EIO)
2777                         return -EBADF;
2778         } else
2779                 nfs4_stateid_copy(&arg->stateid, &zero_stateid);
2780         if (delegation_cred)
2781                 msg.rpc_cred = delegation_cred;
2782
2783         status = nfs4_call_sync(server->client, server, &msg, &arg->seq_args, &res->seq_res, 1);
2784
2785         put_rpccred(delegation_cred);
2786         if (status == 0 && state != NULL)
2787                 renew_lease(server, timestamp);
2788         trace_nfs4_setattr(inode, &arg->stateid, status);
2789         return status;
2790 }
2791
2792 static int nfs4_do_setattr(struct inode *inode, struct rpc_cred *cred,
2793                            struct nfs_fattr *fattr, struct iattr *sattr,
2794                            struct nfs4_state *state, struct nfs4_label *ilabel,
2795                            struct nfs4_label *olabel)
2796 {
2797         struct nfs_server *server = NFS_SERVER(inode);
2798         struct nfs_setattrargs  arg = {
2799                 .fh             = NFS_FH(inode),
2800                 .iap            = sattr,
2801                 .server         = server,
2802                 .bitmask = server->attr_bitmask,
2803                 .label          = ilabel,
2804         };
2805         struct nfs_setattrres  res = {
2806                 .fattr          = fattr,
2807                 .label          = olabel,
2808                 .server         = server,
2809         };
2810         struct nfs4_exception exception = {
2811                 .state = state,
2812                 .inode = inode,
2813                 .stateid = &arg.stateid,
2814         };
2815         int err;
2816
2817         arg.bitmask = nfs4_bitmask(server, ilabel);
2818         if (ilabel)
2819                 arg.bitmask = nfs4_bitmask(server, olabel);
2820
2821         do {
2822                 err = _nfs4_do_setattr(inode, &arg, &res, cred, state);
2823                 switch (err) {
2824                 case -NFS4ERR_OPENMODE:
2825                         if (!(sattr->ia_valid & ATTR_SIZE)) {
2826                                 pr_warn_once("NFSv4: server %s is incorrectly "
2827                                                 "applying open mode checks to "
2828                                                 "a SETATTR that is not "
2829                                                 "changing file size.\n",
2830                                                 server->nfs_client->cl_hostname);
2831                         }
2832                         if (state && !(state->state & FMODE_WRITE)) {
2833                                 err = -EBADF;
2834                                 if (sattr->ia_valid & ATTR_OPEN)
2835                                         err = -EACCES;
2836                                 goto out;
2837                         }
2838                 }
2839                 err = nfs4_handle_exception(server, err, &exception);
2840         } while (exception.retry);
2841 out:
2842         return err;
2843 }
2844
2845 static bool
2846 nfs4_wait_on_layoutreturn(struct inode *inode, struct rpc_task *task)
2847 {
2848         if (inode == NULL || !nfs_have_layout(inode))
2849                 return false;
2850
2851         return pnfs_wait_on_layoutreturn(inode, task);
2852 }
2853
2854 struct nfs4_closedata {
2855         struct inode *inode;
2856         struct nfs4_state *state;
2857         struct nfs_closeargs arg;
2858         struct nfs_closeres res;
2859         struct nfs_fattr fattr;
2860         unsigned long timestamp;
2861         bool roc;
2862         u32 roc_barrier;
2863 };
2864
2865 static void nfs4_free_closedata(void *data)
2866 {
2867         struct nfs4_closedata *calldata = data;
2868         struct nfs4_state_owner *sp = calldata->state->owner;
2869         struct super_block *sb = calldata->state->inode->i_sb;
2870
2871         if (calldata->roc)
2872                 pnfs_roc_release(calldata->state->inode);
2873         nfs4_put_open_state(calldata->state);
2874         nfs_free_seqid(calldata->arg.seqid);
2875         nfs4_put_state_owner(sp);
2876         nfs_sb_deactive(sb);
2877         kfree(calldata);
2878 }
2879
2880 static void nfs4_close_done(struct rpc_task *task, void *data)
2881 {
2882         struct nfs4_closedata *calldata = data;
2883         struct nfs4_state *state = calldata->state;
2884         struct nfs_server *server = NFS_SERVER(calldata->inode);
2885         nfs4_stateid *res_stateid = NULL;
2886
2887         dprintk("%s: begin!\n", __func__);
2888         if (!nfs4_sequence_done(task, &calldata->res.seq_res))
2889                 return;
2890         trace_nfs4_close(state, &calldata->arg, &calldata->res, task->tk_status);
2891         /* hmm. we are done with the inode, and in the process of freeing
2892          * the state_owner. we keep this around to process errors
2893          */
2894         switch (task->tk_status) {
2895                 case 0:
2896                         res_stateid = &calldata->res.stateid;
2897                         if (calldata->roc)
2898                                 pnfs_roc_set_barrier(state->inode,
2899                                                      calldata->roc_barrier);
2900                         renew_lease(server, calldata->timestamp);
2901                         break;
2902                 case -NFS4ERR_ADMIN_REVOKED:
2903                 case -NFS4ERR_STALE_STATEID:
2904                 case -NFS4ERR_OLD_STATEID:
2905                 case -NFS4ERR_BAD_STATEID:
2906                 case -NFS4ERR_EXPIRED:
2907                         if (!nfs4_stateid_match(&calldata->arg.stateid,
2908                                                 &state->open_stateid)) {
2909                                 rpc_restart_call_prepare(task);
2910                                 goto out_release;
2911                         }
2912                         if (calldata->arg.fmode == 0)
2913                                 break;
2914                 default:
2915                         if (nfs4_async_handle_error(task, server, state, NULL) == -EAGAIN) {
2916                                 rpc_restart_call_prepare(task);
2917                                 goto out_release;
2918                         }
2919         }
2920         nfs_clear_open_stateid(state, &calldata->arg.stateid,
2921                         res_stateid, calldata->arg.fmode);
2922 out_release:
2923         nfs_release_seqid(calldata->arg.seqid);
2924         nfs_refresh_inode(calldata->inode, calldata->res.fattr);
2925         dprintk("%s: done, ret = %d!\n", __func__, task->tk_status);
2926 }
2927
2928 static void nfs4_close_prepare(struct rpc_task *task, void *data)
2929 {
2930         struct nfs4_closedata *calldata = data;
2931         struct nfs4_state *state = calldata->state;
2932         struct inode *inode = calldata->inode;
2933         bool is_rdonly, is_wronly, is_rdwr;
2934         int call_close = 0;
2935
2936         dprintk("%s: begin!\n", __func__);
2937         if (nfs_wait_on_sequence(calldata->arg.seqid, task) != 0)
2938                 goto out_wait;
2939
2940         task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_OPEN_DOWNGRADE];
2941         spin_lock(&state->owner->so_lock);
2942         is_rdwr = test_bit(NFS_O_RDWR_STATE, &state->flags);
2943         is_rdonly = test_bit(NFS_O_RDONLY_STATE, &state->flags);
2944         is_wronly = test_bit(NFS_O_WRONLY_STATE, &state->flags);
2945         nfs4_stateid_copy(&calldata->arg.stateid, &state->open_stateid);
2946         /* Calculate the change in open mode */
2947         calldata->arg.fmode = 0;
2948         if (state->n_rdwr == 0) {
2949                 if (state->n_rdonly == 0)
2950                         call_close |= is_rdonly;
2951                 else if (is_rdonly)
2952                         calldata->arg.fmode |= FMODE_READ;
2953                 if (state->n_wronly == 0)
2954                         call_close |= is_wronly;
2955                 else if (is_wronly)
2956                         calldata->arg.fmode |= FMODE_WRITE;
2957                 if (calldata->arg.fmode != (FMODE_READ|FMODE_WRITE))
2958                         call_close |= is_rdwr;
2959         } else if (is_rdwr)
2960                 calldata->arg.fmode |= FMODE_READ|FMODE_WRITE;
2961
2962         if (!nfs4_valid_open_stateid(state))
2963                 call_close = 0;
2964         spin_unlock(&state->owner->so_lock);
2965
2966         if (!call_close) {
2967                 /* Note: exit _without_ calling nfs4_close_done */
2968                 goto out_no_action;
2969         }
2970
2971         if (nfs4_wait_on_layoutreturn(inode, task)) {
2972                 nfs_release_seqid(calldata->arg.seqid);
2973                 goto out_wait;
2974         }
2975
2976         if (calldata->arg.fmode == 0)
2977                 task->tk_msg.rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE];
2978         if (calldata->roc)
2979                 pnfs_roc_get_barrier(inode, &calldata->roc_barrier);
2980
2981         calldata->arg.share_access =
2982                 nfs4_map_atomic_open_share(NFS_SERVER(inode),
2983                                 calldata->arg.fmode, 0);
2984
2985         nfs_fattr_init(calldata->res.fattr);
2986         calldata->timestamp = jiffies;
2987         if (nfs4_setup_sequence(NFS_SERVER(inode),
2988                                 &calldata->arg.seq_args,
2989                                 &calldata->res.seq_res,
2990                                 task) != 0)
2991                 nfs_release_seqid(calldata->arg.seqid);
2992         dprintk("%s: done!\n", __func__);
2993         return;
2994 out_no_action:
2995         task->tk_action = NULL;
2996 out_wait:
2997         nfs4_sequence_done(task, &calldata->res.seq_res);
2998 }
2999
3000 static const struct rpc_call_ops nfs4_close_ops = {
3001         .rpc_call_prepare = nfs4_close_prepare,
3002         .rpc_call_done = nfs4_close_done,
3003         .rpc_release = nfs4_free_closedata,
3004 };
3005
3006 static bool nfs4_roc(struct inode *inode)
3007 {
3008         if (!nfs_have_layout(inode))
3009                 return false;
3010         return pnfs_roc(inode);
3011 }
3012
3013 /* 
3014  * It is possible for data to be read/written from a mem-mapped file 
3015  * after the sys_close call (which hits the vfs layer as a flush).
3016  * This means that we can't safely call nfsv4 close on a file until 
3017  * the inode is cleared. This in turn means that we are not good
3018  * NFSv4 citizens - we do not indicate to the server to update the file's 
3019  * share state even when we are done with one of the three share 
3020  * stateid's in the inode.
3021  *
3022  * NOTE: Caller must be holding the sp->so_owner semaphore!
3023  */
3024 int nfs4_do_close(struct nfs4_state *state, gfp_t gfp_mask, int wait)
3025 {
3026         struct nfs_server *server = NFS_SERVER(state->inode);
3027         struct nfs_seqid *(*alloc_seqid)(struct nfs_seqid_counter *, gfp_t);
3028         struct nfs4_closedata *calldata;
3029         struct nfs4_state_owner *sp = state->owner;
3030         struct rpc_task *task;
3031         struct rpc_message msg = {
3032                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_CLOSE],
3033                 .rpc_cred = state->owner->so_cred,
3034         };
3035         struct rpc_task_setup task_setup_data = {
3036                 .rpc_client = server->client,
3037                 .rpc_message = &msg,
3038                 .callback_ops = &nfs4_close_ops,
3039                 .workqueue = nfsiod_workqueue,
3040                 .flags = RPC_TASK_ASYNC,
3041         };
3042         int status = -ENOMEM;
3043
3044         nfs4_state_protect(server->nfs_client, NFS_SP4_MACH_CRED_CLEANUP,
3045                 &task_setup_data.rpc_client, &msg);
3046
3047         calldata = kzalloc(sizeof(*calldata), gfp_mask);
3048         if (calldata == NULL)
3049                 goto out;
3050         nfs4_init_sequence(&calldata->arg.seq_args, &calldata->res.seq_res, 1);
3051         calldata->inode = state->inode;
3052         calldata->state = state;
3053         calldata->arg.fh = NFS_FH(state->inode);
3054         /* Serialization for the sequence id */
3055         alloc_seqid = server->nfs_client->cl_mvops->alloc_seqid;
3056         calldata->arg.seqid = alloc_seqid(&state->owner->so_seqid, gfp_mask);
3057         if (IS_ERR(calldata->arg.seqid))
3058                 goto out_free_calldata;
3059         calldata->arg.fmode = 0;
3060         calldata->arg.bitmask = server->cache_consistency_bitmask;
3061         calldata->res.fattr = &calldata->fattr;
3062         calldata->res.seqid = calldata->arg.seqid;
3063         calldata->res.server = server;
3064         calldata->roc = nfs4_roc(state->inode);
3065         nfs_sb_active(calldata->inode->i_sb);
3066
3067         msg.rpc_argp = &calldata->arg;
3068         msg.rpc_resp = &calldata->res;
3069         task_setup_data.callback_data = calldata;
3070         task = rpc_run_task(&task_setup_data);
3071         if (IS_ERR(task))
3072                 return PTR_ERR(task);
3073         status = 0;
3074         if (wait)
3075                 status = rpc_wait_for_completion_task(task);
3076         rpc_put_task(task);
3077         return status;
3078 out_free_calldata:
3079         kfree(calldata);
3080 out:
3081         nfs4_put_open_state(state);
3082         nfs4_put_state_owner(sp);
3083         return status;
3084 }
3085
3086 static struct inode *
3087 nfs4_atomic_open(struct inode *dir, struct nfs_open_context *ctx,
3088                 int open_flags, struct iattr *attr, int *opened)
3089 {
3090         struct nfs4_state *state;
3091         struct nfs4_label l = {0, 0, 0, NULL}, *label = NULL;
3092
3093         label = nfs4_label_init_security(dir, ctx->dentry, attr, &l);
3094
3095         /* Protect against concurrent sillydeletes */
3096         state = nfs4_do_open(dir, ctx, open_flags, attr, label, opened);
3097
3098         nfs4_label_release_security(label);
3099
3100         if (IS_ERR(state))
3101                 return ERR_CAST(state);
3102         return state->inode;
3103 }
3104
3105 static void nfs4_close_context(struct nfs_open_context *ctx, int is_sync)
3106 {
3107         if (ctx->state == NULL)
3108                 return;
3109         if (is_sync)
3110                 nfs4_close_sync(ctx->state, ctx->mode);
3111         else
3112                 nfs4_close_state(ctx->state, ctx->mode);
3113 }
3114
3115 #define FATTR4_WORD1_NFS40_MASK (2*FATTR4_WORD1_MOUNTED_ON_FILEID - 1UL)
3116 #define FATTR4_WORD2_NFS41_MASK (2*FATTR4_WORD2_SUPPATTR_EXCLCREAT - 1UL)
3117 #define FATTR4_WORD2_NFS42_MASK (2*FATTR4_WORD2_SECURITY_LABEL - 1UL)
3118
3119 static int _nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
3120 {
3121         u32 bitmask[3] = {}, minorversion = server->nfs_client->cl_minorversion;
3122         struct nfs4_server_caps_arg args = {
3123                 .fhandle = fhandle,
3124                 .bitmask = bitmask,
3125         };
3126         struct nfs4_server_caps_res res = {};
3127         struct rpc_message msg = {
3128                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_SERVER_CAPS],
3129                 .rpc_argp = &args,
3130                 .rpc_resp = &res,
3131         };
3132         int status;
3133
3134         bitmask[0] = FATTR4_WORD0_SUPPORTED_ATTRS |
3135                      FATTR4_WORD0_FH_EXPIRE_TYPE |
3136                      FATTR4_WORD0_LINK_SUPPORT |
3137                      FATTR4_WORD0_SYMLINK_SUPPORT |
3138                      FATTR4_WORD0_ACLSUPPORT;
3139         if (minorversion)
3140                 bitmask[2] = FATTR4_WORD2_SUPPATTR_EXCLCREAT;
3141
3142         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3143         if (status == 0) {
3144                 /* Sanity check the server answers */
3145                 switch (minorversion) {
3146                 case 0:
3147                         res.attr_bitmask[1] &= FATTR4_WORD1_NFS40_MASK;
3148                         res.attr_bitmask[2] = 0;
3149                         break;
3150                 case 1:
3151                         res.attr_bitmask[2] &= FATTR4_WORD2_NFS41_MASK;
3152                         break;
3153                 case 2:
3154                         res.attr_bitmask[2] &= FATTR4_WORD2_NFS42_MASK;
3155                 }
3156                 memcpy(server->attr_bitmask, res.attr_bitmask, sizeof(server->attr_bitmask));
3157                 server->caps &= ~(NFS_CAP_ACLS|NFS_CAP_HARDLINKS|
3158                                 NFS_CAP_SYMLINKS|NFS_CAP_FILEID|
3159                                 NFS_CAP_MODE|NFS_CAP_NLINK|NFS_CAP_OWNER|
3160                                 NFS_CAP_OWNER_GROUP|NFS_CAP_ATIME|
3161                                 NFS_CAP_CTIME|NFS_CAP_MTIME|
3162                                 NFS_CAP_SECURITY_LABEL);
3163                 if (res.attr_bitmask[0] & FATTR4_WORD0_ACL &&
3164                                 res.acl_bitmask & ACL4_SUPPORT_ALLOW_ACL)
3165                         server->caps |= NFS_CAP_ACLS;
3166                 if (res.has_links != 0)
3167                         server->caps |= NFS_CAP_HARDLINKS;
3168                 if (res.has_symlinks != 0)
3169                         server->caps |= NFS_CAP_SYMLINKS;
3170                 if (res.attr_bitmask[0] & FATTR4_WORD0_FILEID)
3171                         server->caps |= NFS_CAP_FILEID;
3172                 if (res.attr_bitmask[1] & FATTR4_WORD1_MODE)
3173                         server->caps |= NFS_CAP_MODE;
3174                 if (res.attr_bitmask[1] & FATTR4_WORD1_NUMLINKS)
3175                         server->caps |= NFS_CAP_NLINK;
3176                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER)
3177                         server->caps |= NFS_CAP_OWNER;
3178                 if (res.attr_bitmask[1] & FATTR4_WORD1_OWNER_GROUP)
3179                         server->caps |= NFS_CAP_OWNER_GROUP;
3180                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_ACCESS)
3181                         server->caps |= NFS_CAP_ATIME;
3182                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_METADATA)
3183                         server->caps |= NFS_CAP_CTIME;
3184                 if (res.attr_bitmask[1] & FATTR4_WORD1_TIME_MODIFY)
3185                         server->caps |= NFS_CAP_MTIME;
3186 #ifdef CONFIG_NFS_V4_SECURITY_LABEL
3187                 if (res.attr_bitmask[2] & FATTR4_WORD2_SECURITY_LABEL)
3188                         server->caps |= NFS_CAP_SECURITY_LABEL;
3189 #endif
3190                 memcpy(server->attr_bitmask_nl, res.attr_bitmask,
3191                                 sizeof(server->attr_bitmask));
3192                 server->attr_bitmask_nl[2] &= ~FATTR4_WORD2_SECURITY_LABEL;
3193
3194                 memcpy(server->cache_consistency_bitmask, res.attr_bitmask, sizeof(server->cache_consistency_bitmask));
3195                 server->cache_consistency_bitmask[0] &= FATTR4_WORD0_CHANGE|FATTR4_WORD0_SIZE;
3196                 server->cache_consistency_bitmask[1] &= FATTR4_WORD1_TIME_METADATA|FATTR4_WORD1_TIME_MODIFY;
3197                 server->cache_consistency_bitmask[2] = 0;
3198                 memcpy(server->exclcreat_bitmask, res.exclcreat_bitmask,
3199                         sizeof(server->exclcreat_bitmask));
3200                 server->acl_bitmask = res.acl_bitmask;
3201                 server->fh_expire_type = res.fh_expire_type;
3202         }
3203
3204         return status;
3205 }
3206
3207 int nfs4_server_capabilities(struct nfs_server *server, struct nfs_fh *fhandle)
3208 {
3209         struct nfs4_exception exception = { };
3210         int err;
3211         do {
3212                 err = nfs4_handle_exception(server,
3213                                 _nfs4_server_capabilities(server, fhandle),
3214                                 &exception);
3215         } while (exception.retry);
3216         return err;
3217 }
3218
3219 static int _nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
3220                 struct nfs_fsinfo *info)
3221 {
3222         u32 bitmask[3];
3223         struct nfs4_lookup_root_arg args = {
3224                 .bitmask = bitmask,
3225         };
3226         struct nfs4_lookup_res res = {
3227                 .server = server,
3228                 .fattr = info->fattr,
3229                 .fh = fhandle,
3230         };
3231         struct rpc_message msg = {
3232                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOOKUP_ROOT],
3233                 .rpc_argp = &args,
3234                 .rpc_resp = &res,
3235         };
3236
3237         bitmask[0] = nfs4_fattr_bitmap[0];
3238         bitmask[1] = nfs4_fattr_bitmap[1];
3239         /*
3240          * Process the label in the upcoming getfattr
3241          */
3242         bitmask[2] = nfs4_fattr_bitmap[2] & ~FATTR4_WORD2_SECURITY_LABEL;
3243
3244         nfs_fattr_init(info->fattr);
3245         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3246 }
3247
3248 static int nfs4_lookup_root(struct nfs_server *server, struct nfs_fh *fhandle,
3249                 struct nfs_fsinfo *info)
3250 {
3251         struct nfs4_exception exception = { };
3252         int err;
3253         do {
3254                 err = _nfs4_lookup_root(server, fhandle, info);
3255                 trace_nfs4_lookup_root(server, fhandle, info->fattr, err);
3256                 switch (err) {
3257                 case 0:
3258                 case -NFS4ERR_WRONGSEC:
3259                         goto out;
3260                 default:
3261                         err = nfs4_handle_exception(server, err, &exception);
3262                 }
3263         } while (exception.retry);
3264 out:
3265         return err;
3266 }
3267
3268 static int nfs4_lookup_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
3269                                 struct nfs_fsinfo *info, rpc_authflavor_t flavor)
3270 {
3271         struct rpc_auth_create_args auth_args = {
3272                 .pseudoflavor = flavor,
3273         };
3274         struct rpc_auth *auth;
3275         int ret;
3276
3277         auth = rpcauth_create(&auth_args, server->client);
3278         if (IS_ERR(auth)) {
3279                 ret = -EACCES;
3280                 goto out;
3281         }
3282         ret = nfs4_lookup_root(server, fhandle, info);
3283 out:
3284         return ret;
3285 }
3286
3287 /*
3288  * Retry pseudoroot lookup with various security flavors.  We do this when:
3289  *
3290  *   NFSv4.0: the PUTROOTFH operation returns NFS4ERR_WRONGSEC
3291  *   NFSv4.1: the server does not support the SECINFO_NO_NAME operation
3292  *
3293  * Returns zero on success, or a negative NFS4ERR value, or a
3294  * negative errno value.
3295  */
3296 static int nfs4_find_root_sec(struct nfs_server *server, struct nfs_fh *fhandle,
3297                               struct nfs_fsinfo *info)
3298 {
3299         /* Per 3530bis 15.33.5 */
3300         static const rpc_authflavor_t flav_array[] = {
3301                 RPC_AUTH_GSS_KRB5P,
3302                 RPC_AUTH_GSS_KRB5I,
3303                 RPC_AUTH_GSS_KRB5,
3304                 RPC_AUTH_UNIX,                  /* courtesy */
3305                 RPC_AUTH_NULL,
3306         };
3307         int status = -EPERM;
3308         size_t i;
3309
3310         if (server->auth_info.flavor_len > 0) {
3311                 /* try each flavor specified by user */
3312                 for (i = 0; i < server->auth_info.flavor_len; i++) {
3313                         status = nfs4_lookup_root_sec(server, fhandle, info,
3314                                                 server->auth_info.flavors[i]);
3315                         if (status == -NFS4ERR_WRONGSEC || status == -EACCES)
3316                                 continue;
3317                         break;
3318                 }
3319         } else {
3320                 /* no flavors specified by user, try default list */
3321                 for (i = 0; i < ARRAY_SIZE(flav_array); i++) {
3322                         status = nfs4_lookup_root_sec(server, fhandle, info,
3323                                                       flav_array[i]);
3324                         if (status == -NFS4ERR_WRONGSEC || status == -EACCES)
3325                                 continue;
3326                         break;
3327                 }
3328         }
3329
3330         /*
3331          * -EACCESS could mean that the user doesn't have correct permissions
3332          * to access the mount.  It could also mean that we tried to mount
3333          * with a gss auth flavor, but rpc.gssd isn't running.  Either way,
3334          * existing mount programs don't handle -EACCES very well so it should
3335          * be mapped to -EPERM instead.
3336          */
3337         if (status == -EACCES)
3338                 status = -EPERM;
3339         return status;
3340 }
3341
3342 /**
3343  * nfs4_proc_get_rootfh - get file handle for server's pseudoroot
3344  * @server: initialized nfs_server handle
3345  * @fhandle: we fill in the pseudo-fs root file handle
3346  * @info: we fill in an FSINFO struct
3347  * @auth_probe: probe the auth flavours
3348  *
3349  * Returns zero on success, or a negative errno.
3350  */
3351 int nfs4_proc_get_rootfh(struct nfs_server *server, struct nfs_fh *fhandle,
3352                          struct nfs_fsinfo *info,
3353                          bool auth_probe)
3354 {
3355         int status = 0;
3356
3357         if (!auth_probe)
3358                 status = nfs4_lookup_root(server, fhandle, info);
3359
3360         if (auth_probe || status == NFS4ERR_WRONGSEC)
3361                 status = server->nfs_client->cl_mvops->find_root_sec(server,
3362                                 fhandle, info);
3363
3364         if (status == 0)
3365                 status = nfs4_server_capabilities(server, fhandle);
3366         if (status == 0)
3367                 status = nfs4_do_fsinfo(server, fhandle, info);
3368
3369         return nfs4_map_errors(status);
3370 }
3371
3372 static int nfs4_proc_get_root(struct nfs_server *server, struct nfs_fh *mntfh,
3373                               struct nfs_fsinfo *info)
3374 {
3375         int error;
3376         struct nfs_fattr *fattr = info->fattr;
3377         struct nfs4_label *label = NULL;
3378
3379         error = nfs4_server_capabilities(server, mntfh);
3380         if (error < 0) {
3381                 dprintk("nfs4_get_root: getcaps error = %d\n", -error);
3382                 return error;
3383         }
3384
3385         label = nfs4_label_alloc(server, GFP_KERNEL);
3386         if (IS_ERR(label))
3387                 return PTR_ERR(label);
3388
3389         error = nfs4_proc_getattr(server, mntfh, fattr, label);
3390         if (error < 0) {
3391                 dprintk("nfs4_get_root: getattr error = %d\n", -error);
3392                 goto err_free_label;
3393         }
3394
3395         if (fattr->valid & NFS_ATTR_FATTR_FSID &&
3396             !nfs_fsid_equal(&server->fsid, &fattr->fsid))
3397                 memcpy(&server->fsid, &fattr->fsid, sizeof(server->fsid));
3398
3399 err_free_label:
3400         nfs4_label_free(label);
3401
3402         return error;
3403 }
3404
3405 /*
3406  * Get locations and (maybe) other attributes of a referral.
3407  * Note that we'll actually follow the referral later when
3408  * we detect fsid mismatch in inode revalidation
3409  */
3410 static int nfs4_get_referral(struct rpc_clnt *client, struct inode *dir,
3411                              const struct qstr *name, struct nfs_fattr *fattr,
3412                              struct nfs_fh *fhandle)
3413 {
3414         int status = -ENOMEM;
3415         struct page *page = NULL;
3416         struct nfs4_fs_locations *locations = NULL;
3417
3418         page = alloc_page(GFP_KERNEL);
3419         if (page == NULL)
3420                 goto out;
3421         locations = kmalloc(sizeof(struct nfs4_fs_locations), GFP_KERNEL);
3422         if (locations == NULL)
3423                 goto out;
3424
3425         status = nfs4_proc_fs_locations(client, dir, name, locations, page);
3426         if (status != 0)
3427                 goto out;
3428
3429         /*
3430          * If the fsid didn't change, this is a migration event, not a
3431          * referral.  Cause us to drop into the exception handler, which
3432          * will kick off migration recovery.
3433          */
3434         if (nfs_fsid_equal(&NFS_SERVER(dir)->fsid, &locations->fattr.fsid)) {
3435                 dprintk("%s: server did not return a different fsid for"
3436                         " a referral at %s\n", __func__, name->name);
3437                 status = -NFS4ERR_MOVED;
3438                 goto out;
3439         }
3440         /* Fixup attributes for the nfs_lookup() call to nfs_fhget() */
3441         nfs_fixup_referral_attributes(&locations->fattr);
3442
3443         /* replace the lookup nfs_fattr with the locations nfs_fattr */
3444         memcpy(fattr, &locations->fattr, sizeof(struct nfs_fattr));
3445         memset(fhandle, 0, sizeof(struct nfs_fh));
3446 out:
3447         if (page)
3448                 __free_page(page);
3449         kfree(locations);
3450         return status;
3451 }
3452
3453 static int _nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle,
3454                                 struct nfs_fattr *fattr, struct nfs4_label *label)
3455 {
3456         struct nfs4_getattr_arg args = {
3457                 .fh = fhandle,
3458                 .bitmask = server->attr_bitmask,
3459         };
3460         struct nfs4_getattr_res res = {
3461                 .fattr = fattr,
3462                 .label = label,
3463                 .server = server,
3464         };
3465         struct rpc_message msg = {
3466                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETATTR],
3467                 .rpc_argp = &args,
3468                 .rpc_resp = &res,
3469         };
3470
3471         args.bitmask = nfs4_bitmask(server, label);
3472
3473         nfs_fattr_init(fattr);
3474         return nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3475 }
3476
3477 static int nfs4_proc_getattr(struct nfs_server *server, struct nfs_fh *fhandle,
3478                                 struct nfs_fattr *fattr, struct nfs4_label *label)
3479 {
3480         struct nfs4_exception exception = { };
3481         int err;
3482         do {
3483                 err = _nfs4_proc_getattr(server, fhandle, fattr, label);
3484                 trace_nfs4_getattr(server, fhandle, fattr, err);
3485                 err = nfs4_handle_exception(server, err,
3486                                 &exception);
3487         } while (exception.retry);
3488         return err;
3489 }
3490
3491 /* 
3492  * The file is not closed if it is opened due to the a request to change
3493  * the size of the file. The open call will not be needed once the
3494  * VFS layer lookup-intents are implemented.
3495  *
3496  * Close is called when the inode is destroyed.
3497  * If we haven't opened the file for O_WRONLY, we
3498  * need to in the size_change case to obtain a stateid.
3499  *
3500  * Got race?
3501  * Because OPEN is always done by name in nfsv4, it is
3502  * possible that we opened a different file by the same
3503  * name.  We can recognize this race condition, but we
3504  * can't do anything about it besides returning an error.
3505  *
3506  * This will be fixed with VFS changes (lookup-intent).
3507  */
3508 static int
3509 nfs4_proc_setattr(struct dentry *dentry, struct nfs_fattr *fattr,
3510                   struct iattr *sattr)
3511 {
3512         struct inode *inode = d_inode(dentry);
3513         struct rpc_cred *cred = NULL;
3514         struct nfs4_state *state = NULL;
3515         struct nfs4_label *label = NULL;
3516         int status;
3517
3518         if (pnfs_ld_layoutret_on_setattr(inode) &&
3519             sattr->ia_valid & ATTR_SIZE &&
3520             sattr->ia_size < i_size_read(inode))
3521                 pnfs_commit_and_return_layout(inode);
3522
3523         nfs_fattr_init(fattr);
3524         
3525         /* Deal with open(O_TRUNC) */
3526         if (sattr->ia_valid & ATTR_OPEN)
3527                 sattr->ia_valid &= ~(ATTR_MTIME|ATTR_CTIME);
3528
3529         /* Optimization: if the end result is no change, don't RPC */
3530         if ((sattr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
3531                 return 0;
3532
3533         /* Search for an existing open(O_WRITE) file */
3534         if (sattr->ia_valid & ATTR_FILE) {
3535                 struct nfs_open_context *ctx;
3536
3537                 ctx = nfs_file_open_context(sattr->ia_file);
3538                 if (ctx) {
3539                         cred = ctx->cred;
3540                         state = ctx->state;
3541                 }
3542         }
3543
3544         label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
3545         if (IS_ERR(label))
3546                 return PTR_ERR(label);
3547
3548         status = nfs4_do_setattr(inode, cred, fattr, sattr, state, NULL, label);
3549         if (status == 0) {
3550                 nfs_setattr_update_inode(inode, sattr, fattr);
3551                 nfs_setsecurity(inode, fattr, label);
3552         }
3553         nfs4_label_free(label);
3554         return status;
3555 }
3556
3557 static int _nfs4_proc_lookup(struct rpc_clnt *clnt, struct inode *dir,
3558                 const struct qstr *name, struct nfs_fh *fhandle,
3559                 struct nfs_fattr *fattr, struct nfs4_label *label)
3560 {
3561         struct nfs_server *server = NFS_SERVER(dir);
3562         int                    status;
3563         struct nfs4_lookup_arg args = {
3564                 .bitmask = server->attr_bitmask,
3565                 .dir_fh = NFS_FH(dir),
3566                 .name = name,
3567         };
3568         struct nfs4_lookup_res res = {
3569                 .server = server,
3570                 .fattr = fattr,
3571                 .label = label,
3572                 .fh = fhandle,
3573         };
3574         struct rpc_message msg = {
3575                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LOOKUP],
3576                 .rpc_argp = &args,
3577                 .rpc_resp = &res,
3578         };
3579
3580         args.bitmask = nfs4_bitmask(server, label);
3581
3582         nfs_fattr_init(fattr);
3583
3584         dprintk("NFS call  lookup %s\n", name->name);
3585         status = nfs4_call_sync(clnt, server, &msg, &args.seq_args, &res.seq_res, 0);
3586         dprintk("NFS reply lookup: %d\n", status);
3587         return status;
3588 }
3589
3590 static void nfs_fixup_secinfo_attributes(struct nfs_fattr *fattr)
3591 {
3592         fattr->valid |= NFS_ATTR_FATTR_TYPE | NFS_ATTR_FATTR_MODE |
3593                 NFS_ATTR_FATTR_NLINK | NFS_ATTR_FATTR_MOUNTPOINT;
3594         fattr->mode = S_IFDIR | S_IRUGO | S_IXUGO;
3595         fattr->nlink = 2;
3596 }
3597
3598 static int nfs4_proc_lookup_common(struct rpc_clnt **clnt, struct inode *dir,
3599                                    const struct qstr *name, struct nfs_fh *fhandle,
3600                                    struct nfs_fattr *fattr, struct nfs4_label *label)
3601 {
3602         struct nfs4_exception exception = { };
3603         struct rpc_clnt *client = *clnt;
3604         int err;
3605         do {
3606                 err = _nfs4_proc_lookup(client, dir, name, fhandle, fattr, label);
3607                 trace_nfs4_lookup(dir, name, err);
3608                 switch (err) {
3609                 case -NFS4ERR_BADNAME:
3610                         err = -ENOENT;
3611                         goto out;
3612                 case -NFS4ERR_MOVED:
3613                         err = nfs4_get_referral(client, dir, name, fattr, fhandle);
3614                         if (err == -NFS4ERR_MOVED)
3615                                 err = nfs4_handle_exception(NFS_SERVER(dir), err, &exception);
3616                         goto out;
3617                 case -NFS4ERR_WRONGSEC:
3618                         err = -EPERM;
3619                         if (client != *clnt)
3620                                 goto out;
3621                         client = nfs4_negotiate_security(client, dir, name);
3622                         if (IS_ERR(client))
3623                                 return PTR_ERR(client);
3624
3625                         exception.retry = 1;
3626                         break;
3627                 default:
3628                         err = nfs4_handle_exception(NFS_SERVER(dir), err, &exception);
3629                 }
3630         } while (exception.retry);
3631
3632 out:
3633         if (err == 0)
3634                 *clnt = client;
3635         else if (client != *clnt)
3636                 rpc_shutdown_client(client);
3637
3638         return err;
3639 }
3640
3641 static int nfs4_proc_lookup(struct inode *dir, const struct qstr *name,
3642                             struct nfs_fh *fhandle, struct nfs_fattr *fattr,
3643                             struct nfs4_label *label)
3644 {
3645         int status;
3646         struct rpc_clnt *client = NFS_CLIENT(dir);
3647
3648         status = nfs4_proc_lookup_common(&client, dir, name, fhandle, fattr, label);
3649         if (client != NFS_CLIENT(dir)) {
3650                 rpc_shutdown_client(client);
3651                 nfs_fixup_secinfo_attributes(fattr);
3652         }
3653         return status;
3654 }
3655
3656 struct rpc_clnt *
3657 nfs4_proc_lookup_mountpoint(struct inode *dir, const struct qstr *name,
3658                             struct nfs_fh *fhandle, struct nfs_fattr *fattr)
3659 {
3660         struct rpc_clnt *client = NFS_CLIENT(dir);
3661         int status;
3662
3663         status = nfs4_proc_lookup_common(&client, dir, name, fhandle, fattr, NULL);
3664         if (status < 0)
3665                 return ERR_PTR(status);
3666         return (client == NFS_CLIENT(dir)) ? rpc_clone_client(client) : client;
3667 }
3668
3669 static int _nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry)
3670 {
3671         struct nfs_server *server = NFS_SERVER(inode);
3672         struct nfs4_accessargs args = {
3673                 .fh = NFS_FH(inode),
3674                 .bitmask = server->cache_consistency_bitmask,
3675         };
3676         struct nfs4_accessres res = {
3677                 .server = server,
3678         };
3679         struct rpc_message msg = {
3680                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_ACCESS],
3681                 .rpc_argp = &args,
3682                 .rpc_resp = &res,
3683                 .rpc_cred = entry->cred,
3684         };
3685         int mode = entry->mask;
3686         int status = 0;
3687
3688         /*
3689          * Determine which access bits we want to ask for...
3690          */
3691         if (mode & MAY_READ)
3692                 args.access |= NFS4_ACCESS_READ;
3693         if (S_ISDIR(inode->i_mode)) {
3694                 if (mode & MAY_WRITE)
3695                         args.access |= NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE;
3696                 if (mode & MAY_EXEC)
3697                         args.access |= NFS4_ACCESS_LOOKUP;
3698         } else {
3699                 if (mode & MAY_WRITE)
3700                         args.access |= NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND;
3701                 if (mode & MAY_EXEC)
3702                         args.access |= NFS4_ACCESS_EXECUTE;
3703         }
3704
3705         res.fattr = nfs_alloc_fattr();
3706         if (res.fattr == NULL)
3707                 return -ENOMEM;
3708
3709         status = nfs4_call_sync(server->client, server, &msg, &args.seq_args, &res.seq_res, 0);
3710         if (!status) {
3711                 nfs_access_set_mask(entry, res.access);
3712                 nfs_refresh_inode(inode, res.fattr);
3713         }
3714         nfs_free_fattr(res.fattr);
3715         return status;
3716 }
3717
3718 static int nfs4_proc_access(struct inode *inode, struct nfs_access_entry *entry)
3719 {
3720         struct nfs4_exception exception = { };
3721         int err;
3722         do {
3723                 err = _nfs4_proc_access(inode, entry);
3724                 trace_nfs4_access(inode, err);
3725                 err = nfs4_handle_exception(NFS_SERVER(inode), err,
3726                                 &exception);
3727         } while (exception.retry);
3728         return err;
3729 }
3730
3731 /*
3732  * TODO: For the time being, we don't try to get any attributes
3733  * along with any of the zero-copy operations READ, READDIR,
3734  * READLINK, WRITE.
3735  *
3736  * In the case of the first three, we want to put the GETATTR
3737  * after the read-type operation -- this is because it is hard
3738  * to predict the length of a GETATTR response in v4, and thus
3739  * align the READ data correctly.  This means that the GETATTR
3740  * may end up partially falling into the page cache, and we should
3741  * shift it into the 'tail' of the xdr_buf before processing.
3742  * To do this efficiently, we need to know the total length
3743  * of data received, which doesn't seem to be available outside
3744  * of the RPC layer.
3745  *
3746  * In the case of WRITE, we also want to put the GETATTR after
3747  * the operation -- in this case because we want to make sure
3748  * we get the post-operation mtime and size.
3749  *
3750  * Both of these changes to the XDR layer would in fact be quite
3751  * minor, but I decided to leave them for a subsequent patch.
3752  */
3753 static int _nfs4_proc_readlink(struct inode *inode, struct page *page,
3754                 unsigned int pgbase, unsigned int pglen)
3755 {
3756         struct nfs4_readlink args = {
3757                 .fh       = NFS_FH(inode),
3758                 .pgbase   = pgbase,
3759                 .pglen    = pglen,
3760                 .pages    = &page,
3761         };
3762         struct nfs4_readlink_res res;
3763         struct rpc_message msg = {
3764                 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_READLINK],
3765                 .rpc_argp = &args,
3766                 .rpc_resp = &res,
3767         };
3768
3769         return nfs4_call_sync(NFS_SERVER(inode)->client, NFS_SERVER(inode), &msg, &args.seq_args, &res.seq_res, 0);
3770 }
3771
3772 static int nfs4_proc_readlink(struct inode *inode, struct page *page,
3773                 unsigned int pgbase, unsigned int pglen)
3774 {
3775         struct nfs4_exception exception = { };
3776         int err;
3777         do {
3778                 err = _nfs4_proc_readlink(inode, page, pgbase, pglen);
3779                 trace_nfs4_readlink(inode, err);
3780                 err = nfs4_handle_exception(NFS_SERVER(inode), err,
3781                                 &exception);
3782         } while (exception.retry);
3783         return err;
3784 }
3785
3786 /*
3787  * This is just for mknod.  open(O_CREAT) will always do ->open_context().
3788  */
3789 static int
3790 nfs4_proc_create(struct inode *dir, struct dentry *dentry, struct iattr *sattr,
3791                  int flags)
3792 {
3793         struct nfs4_label l, *ilabel = NULL;
3794         struct nfs_open_context *ctx;
3795         struct nfs4_state *state;
3796         int status = 0;
3797
3798         ctx = alloc_nfs_open_context(dentry, FMODE_READ);
3799         if (IS_ERR(ctx))
3800                 return PTR_ERR(ctx);
3801
3802         ilabel = nfs4_label_init_security(dir, dentry, sattr, &l);
3803
3804         sattr->ia_mode &= ~current_umask();
3805    &nbs