NFS: Add mount options for controlling the lookup cache
[linux-2.6-block.git] / fs / nfs / inode.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/nfs/inode.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs inode and superblock handling functions
7 *
8 * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some
9 * experimental NFS changes. Modularisation taken straight from SYS5 fs.
10 *
11 * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts.
12 * J.S.Peatfield@damtp.cam.ac.uk
13 *
14 */
15
1da177e4
LT
16#include <linux/module.h>
17#include <linux/init.h>
e8edc6e0 18#include <linux/sched.h>
1da177e4
LT
19#include <linux/time.h>
20#include <linux/kernel.h>
21#include <linux/mm.h>
22#include <linux/string.h>
23#include <linux/stat.h>
24#include <linux/errno.h>
25#include <linux/unistd.h>
26#include <linux/sunrpc/clnt.h>
27#include <linux/sunrpc/stats.h>
4ece3a2d 28#include <linux/sunrpc/metrics.h>
1da177e4
LT
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/nfs4_mount.h>
32#include <linux/lockd/bind.h>
33#include <linux/smp_lock.h>
34#include <linux/seq_file.h>
35#include <linux/mount.h>
36#include <linux/nfs_idmap.h>
37#include <linux/vfs.h>
9cdb3883
MN
38#include <linux/inet.h>
39#include <linux/nfs_xdr.h>
1da177e4
LT
40
41#include <asm/system.h>
42#include <asm/uaccess.h>
43
4ce79717 44#include "nfs4_fs.h"
a72b4422 45#include "callback.h"
1da177e4 46#include "delegation.h"
d9ef5a8c 47#include "iostat.h"
f7b422b1 48#include "internal.h"
1da177e4
LT
49
50#define NFSDBG_FACILITY NFSDBG_VFS
1da177e4 51
f43bf0be
TM
52#define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
53
54/* Default is to see 64-bit inode numbers */
55static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
56
1da177e4 57static void nfs_invalidate_inode(struct inode *);
24aa1fe6 58static int nfs_update_inode(struct inode *, struct nfs_fattr *);
1da177e4 59
e18b890b 60static struct kmem_cache * nfs_inode_cachep;
b7fa0554 61
1da177e4
LT
62static inline unsigned long
63nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
64{
65 return nfs_fileid_to_ino_t(fattr->fileid);
66}
67
f43bf0be
TM
68/**
69 * nfs_compat_user_ino64 - returns the user-visible inode number
70 * @fileid: 64-bit fileid
71 *
72 * This function returns a 32-bit inode number if the boot parameter
73 * nfs.enable_ino64 is zero.
74 */
75u64 nfs_compat_user_ino64(u64 fileid)
76{
77 int ino;
78
79 if (enable_ino64)
80 return fileid;
81 ino = fileid;
82 if (sizeof(ino) < sizeof(fileid))
83 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
84 return ino;
85}
86
f7b422b1 87int nfs_write_inode(struct inode *inode, int sync)
1da177e4 88{
1da177e4
LT
89 int ret;
90
d30c8348
TM
91 if (sync) {
92 ret = filemap_fdatawait(inode->i_mapping);
93 if (ret == 0)
94 ret = nfs_commit_inode(inode, FLUSH_SYNC);
95 } else
96 ret = nfs_commit_inode(inode, 0);
97 if (ret >= 0)
98 return 0;
99 __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
100 return ret;
1da177e4
LT
101}
102
f7b422b1 103void nfs_clear_inode(struct inode *inode)
1da177e4 104{
da6d503a
TM
105 /*
106 * The following should never happen...
107 */
108 BUG_ON(nfs_have_writebacks(inode));
1c3c07e9 109 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
ada70d94 110 nfs_zap_acl_cache(inode);
1c3c07e9 111 nfs_access_zap_cache(inode);
1da177e4
LT
112}
113
29884df0
TM
114/**
115 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
116 */
117int nfs_sync_mapping(struct address_space *mapping)
118{
119 int ret;
120
121 if (mapping->nrpages == 0)
122 return 0;
123 unmap_mapping_range(mapping, 0, 0, 0);
28fd1298 124 ret = filemap_write_and_wait(mapping);
29884df0
TM
125 if (ret != 0)
126 goto out;
127 ret = nfs_wb_all(mapping->host);
128out:
129 return ret;
130}
131
1da177e4
LT
132/*
133 * Invalidate the local caches
134 */
b37b03b7 135static void nfs_zap_caches_locked(struct inode *inode)
1da177e4
LT
136{
137 struct nfs_inode *nfsi = NFS_I(inode);
138 int mode = inode->i_mode;
139
91d5b470
CL
140 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
141
c7c20973
TM
142 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
143 nfsi->attrtimeo_timestamp = jiffies;
1da177e4
LT
144
145 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
146 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
55296809 147 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
1da177e4 148 else
55296809 149 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
b37b03b7 150}
dc59250c 151
b37b03b7
TM
152void nfs_zap_caches(struct inode *inode)
153{
154 spin_lock(&inode->i_lock);
155 nfs_zap_caches_locked(inode);
dc59250c 156 spin_unlock(&inode->i_lock);
ada70d94
TM
157}
158
cd9ae2b6
TM
159void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
160{
161 if (mapping->nrpages != 0) {
162 spin_lock(&inode->i_lock);
163 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
164 spin_unlock(&inode->i_lock);
165 }
166}
167
f41f7418 168void nfs_zap_acl_cache(struct inode *inode)
ada70d94
TM
169{
170 void (*clear_acl_cache)(struct inode *);
171
172 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
173 if (clear_acl_cache != NULL)
174 clear_acl_cache(inode);
dc59250c 175 spin_lock(&inode->i_lock);
55296809 176 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
dc59250c 177 spin_unlock(&inode->i_lock);
1da177e4
LT
178}
179
c4812998
TM
180void nfs_invalidate_atime(struct inode *inode)
181{
182 spin_lock(&inode->i_lock);
183 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
184 spin_unlock(&inode->i_lock);
185}
186
1da177e4 187/*
b37b03b7
TM
188 * Invalidate, but do not unhash, the inode.
189 * NB: must be called with inode->i_lock held!
1da177e4 190 */
b37b03b7 191static void nfs_invalidate_inode(struct inode *inode)
1da177e4 192{
3a10c30a 193 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
b37b03b7 194 nfs_zap_caches_locked(inode);
1da177e4
LT
195}
196
197struct nfs_find_desc {
198 struct nfs_fh *fh;
199 struct nfs_fattr *fattr;
200};
201
202/*
203 * In NFSv3 we can have 64bit inode numbers. In order to support
204 * this, and re-exported directories (also seen in NFSv2)
205 * we are forced to allow 2 different inodes to have the same
206 * i_ino.
207 */
208static int
209nfs_find_actor(struct inode *inode, void *opaque)
210{
211 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
212 struct nfs_fh *fh = desc->fh;
213 struct nfs_fattr *fattr = desc->fattr;
214
215 if (NFS_FILEID(inode) != fattr->fileid)
216 return 0;
217 if (nfs_compare_fh(NFS_FH(inode), fh))
218 return 0;
219 if (is_bad_inode(inode) || NFS_STALE(inode))
220 return 0;
221 return 1;
222}
223
224static int
225nfs_init_locked(struct inode *inode, void *opaque)
226{
227 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
228 struct nfs_fattr *fattr = desc->fattr;
229
99fadcd7 230 set_nfs_fileid(inode, fattr->fileid);
1da177e4
LT
231 nfs_copy_fh(NFS_FH(inode), desc->fh);
232 return 0;
233}
234
235/* Don't use READDIRPLUS on directories that we believe are too large */
236#define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE)
237
238/*
239 * This is our front-end to iget that looks up inodes by file handle
240 * instead of inode number.
241 */
242struct inode *
243nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
244{
245 struct nfs_find_desc desc = {
246 .fh = fh,
247 .fattr = fattr
248 };
03f28e3a 249 struct inode *inode = ERR_PTR(-ENOENT);
1da177e4
LT
250 unsigned long hash;
251
252 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
253 goto out_no_inode;
254
255 if (!fattr->nlink) {
256 printk("NFS: Buggy server - nlink == 0!\n");
257 goto out_no_inode;
258 }
259
260 hash = nfs_fattr_to_ino_t(fattr);
261
03f28e3a
TM
262 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
263 if (inode == NULL) {
264 inode = ERR_PTR(-ENOMEM);
1da177e4 265 goto out_no_inode;
03f28e3a 266 }
1da177e4
LT
267
268 if (inode->i_state & I_NEW) {
269 struct nfs_inode *nfsi = NFS_I(inode);
b0c4fddc 270 unsigned long now = jiffies;
1da177e4
LT
271
272 /* We set i_ino for the few things that still rely on it,
273 * such as stat(2) */
274 inode->i_ino = hash;
275
276 /* We can't support update_atime(), since the server will reset it */
277 inode->i_flags |= S_NOATIME|S_NOCMTIME;
278 inode->i_mode = fattr->mode;
279 /* Why so? Because we want revalidate for devices/FIFOs, and
280 * that's precisely what we have in nfs_file_inode_operations.
281 */
8fa5c000 282 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
1da177e4
LT
283 if (S_ISREG(inode->i_mode)) {
284 inode->i_fop = &nfs_file_operations;
285 inode->i_data.a_ops = &nfs_file_aops;
286 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
287 } else if (S_ISDIR(inode->i_mode)) {
8fa5c000 288 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
1da177e4
LT
289 inode->i_fop = &nfs_dir_operations;
290 if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS)
291 && fattr->size <= NFS_LIMIT_READDIRPLUS)
3a10c30a 292 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
55a97593
TM
293 /* Deal with crossing mountpoints */
294 if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) {
6b97fd3d
MN
295 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
296 inode->i_op = &nfs_referral_inode_operations;
297 else
298 inode->i_op = &nfs_mountpoint_inode_operations;
55a97593 299 inode->i_fop = NULL;
c37dcd33 300 set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags);
55a97593 301 }
1da177e4
LT
302 } else if (S_ISLNK(inode->i_mode))
303 inode->i_op = &nfs_symlink_inode_operations;
304 else
305 init_special_inode(inode, inode->i_mode, fattr->rdev);
306
33801147 307 nfsi->read_cache_jiffies = fattr->time_start;
b0c4fddc
TM
308 nfsi->last_updated = now;
309 nfsi->cache_change_attribute = now;
1da177e4
LT
310 inode->i_atime = fattr->atime;
311 inode->i_mtime = fattr->mtime;
312 inode->i_ctime = fattr->ctime;
313 if (fattr->valid & NFS_ATTR_FATTR_V4)
314 nfsi->change_attr = fattr->change_attr;
315 inode->i_size = nfs_size_to_loff_t(fattr->size);
316 inode->i_nlink = fattr->nlink;
317 inode->i_uid = fattr->uid;
318 inode->i_gid = fattr->gid;
319 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
320 /*
321 * report the blocks in 512byte units
322 */
323 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1da177e4
LT
324 } else {
325 inode->i_blocks = fattr->du.nfs2.blocks;
1da177e4
LT
326 }
327 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
b0c4fddc 328 nfsi->attrtimeo_timestamp = now;
1da177e4 329 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
1c3c07e9 330 nfsi->access_cache = RB_ROOT;
1da177e4
LT
331
332 unlock_new_inode(inode);
333 } else
334 nfs_refresh_inode(inode, fattr);
335 dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n",
336 inode->i_sb->s_id,
337 (long long)NFS_FILEID(inode),
338 atomic_read(&inode->i_count));
339
340out:
341 return inode;
342
343out_no_inode:
03f28e3a 344 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
1da177e4
LT
345 goto out;
346}
347
659bfcd6 348#define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE)
1da177e4
LT
349
350int
351nfs_setattr(struct dentry *dentry, struct iattr *attr)
352{
353 struct inode *inode = dentry->d_inode;
354 struct nfs_fattr fattr;
355 int error;
356
91d5b470
CL
357 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
358
188b95dd
JL
359 /* skip mode change if it's just for clearing setuid/setgid */
360 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
361 attr->ia_valid &= ~ATTR_MODE;
362
1da177e4
LT
363 if (attr->ia_valid & ATTR_SIZE) {
364 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
365 attr->ia_valid &= ~ATTR_SIZE;
366 }
367
368 /* Optimization: if the end result is no change, don't RPC */
369 attr->ia_valid &= NFS_VALID_ATTRS;
659bfcd6 370 if ((attr->ia_valid & ~ATTR_FILE) == 0)
1da177e4
LT
371 return 0;
372
755c1e20 373 /* Write all dirty data */
e1552e19
TM
374 if (S_ISREG(inode->i_mode)) {
375 filemap_write_and_wait(inode->i_mapping);
376 nfs_wb_all(inode);
377 }
642ac549
TM
378 /*
379 * Return any delegations if we're going to change ACLs
380 */
381 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
382 nfs_inode_return_delegation(inode);
1da177e4 383 error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr);
65e4308d 384 if (error == 0)
1da177e4 385 nfs_refresh_inode(inode, &fattr);
65e4308d
TM
386 return error;
387}
388
a3d01454
TM
389/**
390 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
391 * @inode: inode of the file used
392 * @offset: file offset to start truncating
393 *
394 * This is a copy of the common vmtruncate, but with the locking
395 * corrected to take into account the fact that NFS requires
396 * inode->i_size to be updated under the inode->i_lock.
397 */
398static int nfs_vmtruncate(struct inode * inode, loff_t offset)
399{
400 if (i_size_read(inode) < offset) {
401 unsigned long limit;
402
403 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
404 if (limit != RLIM_INFINITY && offset > limit)
405 goto out_sig;
406 if (offset > inode->i_sb->s_maxbytes)
407 goto out_big;
408 spin_lock(&inode->i_lock);
409 i_size_write(inode, offset);
410 spin_unlock(&inode->i_lock);
411 } else {
412 struct address_space *mapping = inode->i_mapping;
413
414 /*
415 * truncation of in-use swapfiles is disallowed - it would
416 * cause subsequent swapout to scribble on the now-freed
417 * blocks.
418 */
419 if (IS_SWAPFILE(inode))
420 return -ETXTBSY;
421 spin_lock(&inode->i_lock);
422 i_size_write(inode, offset);
423 spin_unlock(&inode->i_lock);
424
425 /*
426 * unmap_mapping_range is called twice, first simply for
427 * efficiency so that truncate_inode_pages does fewer
428 * single-page unmaps. However after this first call, and
429 * before truncate_inode_pages finishes, it is possible for
430 * private pages to be COWed, which remain after
431 * truncate_inode_pages finishes, hence the second
432 * unmap_mapping_range call must be made for correctness.
433 */
434 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
435 truncate_inode_pages(mapping, offset);
436 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
437 }
438 return 0;
439out_sig:
440 send_sig(SIGXFSZ, current, 0);
441out_big:
442 return -EFBIG;
443}
444
65e4308d
TM
445/**
446 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
447 * @inode: pointer to struct inode
448 * @attr: pointer to struct iattr
449 *
450 * Note: we do this in the *proc.c in order to ensure that
451 * it works for things like exclusive creates too.
452 */
453void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
454{
455 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
1da177e4 456 if ((attr->ia_valid & ATTR_MODE) != 0) {
65e4308d
TM
457 int mode = attr->ia_mode & S_IALLUGO;
458 mode |= inode->i_mode & ~S_IALLUGO;
1da177e4
LT
459 inode->i_mode = mode;
460 }
461 if ((attr->ia_valid & ATTR_UID) != 0)
462 inode->i_uid = attr->ia_uid;
463 if ((attr->ia_valid & ATTR_GID) != 0)
464 inode->i_gid = attr->ia_gid;
dc59250c 465 spin_lock(&inode->i_lock);
55296809 466 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
dc59250c 467 spin_unlock(&inode->i_lock);
65e4308d
TM
468 }
469 if ((attr->ia_valid & ATTR_SIZE) != 0) {
91d5b470 470 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
a3d01454 471 nfs_vmtruncate(inode, attr->ia_size);
65e4308d 472 }
1da177e4
LT
473}
474
412d582e
CL
475static int nfs_wait_schedule(void *word)
476{
477 if (signal_pending(current))
478 return -ERESTARTSYS;
479 schedule();
480 return 0;
481}
482
1da177e4
LT
483/*
484 * Wait for the inode to get unlocked.
1da177e4 485 */
412d582e 486static int nfs_wait_on_inode(struct inode *inode)
1da177e4 487{
1da177e4 488 struct nfs_inode *nfsi = NFS_I(inode);
1da177e4 489 int error;
412d582e 490
412d582e 491 error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING,
150030b7 492 nfs_wait_schedule, TASK_KILLABLE);
412d582e 493
1da177e4
LT
494 return error;
495}
496
412d582e
CL
497static void nfs_wake_up_inode(struct inode *inode)
498{
499 struct nfs_inode *nfsi = NFS_I(inode);
500
501 clear_bit(NFS_INO_REVALIDATING, &nfsi->flags);
502 smp_mb__after_clear_bit();
503 wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING);
504}
505
1da177e4
LT
506int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
507{
508 struct inode *inode = dentry->d_inode;
55296809 509 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1da177e4
LT
510 int err;
511
28c494c5
CL
512 /*
513 * Flush out writes to the server in order to update c/mtime.
514 *
515 * Hold the i_mutex to suspend application writes temporarily;
516 * this prevents long-running writing applications from blocking
517 * nfs_wb_nocommit.
518 */
519 if (S_ISREG(inode->i_mode)) {
520 mutex_lock(&inode->i_mutex);
ed90ef51 521 nfs_wb_nocommit(inode);
28c494c5
CL
522 mutex_unlock(&inode->i_mutex);
523 }
fc33a7bb
CH
524
525 /*
526 * We may force a getattr if the user cares about atime.
527 *
528 * Note that we only have to check the vfsmount flags here:
529 * - NFS always sets S_NOATIME by so checking it would give a
530 * bogus result
531 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
532 * no point in checking those.
533 */
534 if ((mnt->mnt_flags & MNT_NOATIME) ||
535 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1da177e4 536 need_atime = 0;
fc33a7bb 537
1da177e4
LT
538 if (need_atime)
539 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
540 else
541 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
4e769b93 542 if (!err) {
1da177e4 543 generic_fillattr(inode, stat);
f43bf0be 544 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
4e769b93 545 }
1da177e4
LT
546 return err;
547}
548
b92dccf6 549static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1da177e4
LT
550{
551 struct nfs_open_context *ctx;
552
f52720ca 553 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1da177e4 554 if (ctx != NULL) {
88be9f99
TM
555 ctx->path.dentry = dget(dentry);
556 ctx->path.mnt = mntget(mnt);
1da177e4
LT
557 ctx->cred = get_rpccred(cred);
558 ctx->state = NULL;
559 ctx->lockowner = current->files;
66d3aac0 560 ctx->flags = 0;
1da177e4 561 ctx->error = 0;
00a92642 562 ctx->dir_cookie = 0;
5e11934d 563 atomic_set(&ctx->count, 1);
1da177e4
LT
564 }
565 return ctx;
566}
567
568struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
569{
570 if (ctx != NULL)
5e11934d 571 atomic_inc(&ctx->count);
1da177e4
LT
572 return ctx;
573}
574
a49c3c77 575static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
1da177e4 576{
383ba719 577 struct inode *inode;
3bec63db 578
383ba719
TM
579 if (ctx == NULL)
580 return;
581
582 inode = ctx->path.dentry->d_inode;
5e11934d
TM
583 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
584 return;
585 list_del(&ctx->list);
586 spin_unlock(&inode->i_lock);
a49c3c77
TM
587 if (ctx->state != NULL) {
588 if (wait)
589 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
590 else
591 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
592 }
3bec63db
TM
593 if (ctx->cred != NULL)
594 put_rpccred(ctx->cred);
31f31db1 595 path_put(&ctx->path);
3bec63db
TM
596 kfree(ctx);
597}
598
a49c3c77
TM
599void put_nfs_open_context(struct nfs_open_context *ctx)
600{
601 __put_nfs_open_context(ctx, 0);
602}
603
604static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
605{
606 __put_nfs_open_context(ctx, 1);
607}
608
1da177e4
LT
609/*
610 * Ensure that mmap has a recent RPC credential for use when writing out
611 * shared pages
612 */
b92dccf6 613static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1da177e4 614{
01cce933 615 struct inode *inode = filp->f_path.dentry->d_inode;
1da177e4
LT
616 struct nfs_inode *nfsi = NFS_I(inode);
617
618 filp->private_data = get_nfs_open_context(ctx);
619 spin_lock(&inode->i_lock);
620 list_add(&ctx->list, &nfsi->open_files);
621 spin_unlock(&inode->i_lock);
622}
623
d530838b
TM
624/*
625 * Given an inode, search for an open context with the desired characteristics
626 */
627struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1da177e4
LT
628{
629 struct nfs_inode *nfsi = NFS_I(inode);
630 struct nfs_open_context *pos, *ctx = NULL;
631
632 spin_lock(&inode->i_lock);
633 list_for_each_entry(pos, &nfsi->open_files, list) {
d530838b
TM
634 if (cred != NULL && pos->cred != cred)
635 continue;
1da177e4
LT
636 if ((pos->mode & mode) == mode) {
637 ctx = get_nfs_open_context(pos);
638 break;
639 }
640 }
641 spin_unlock(&inode->i_lock);
642 return ctx;
643}
644
b92dccf6 645static void nfs_file_clear_open_context(struct file *filp)
1da177e4 646{
01cce933 647 struct inode *inode = filp->f_path.dentry->d_inode;
cd3758e3 648 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1da177e4
LT
649
650 if (ctx) {
651 filp->private_data = NULL;
652 spin_lock(&inode->i_lock);
653 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
654 spin_unlock(&inode->i_lock);
a49c3c77 655 put_nfs_open_context_sync(ctx);
1da177e4
LT
656 }
657}
658
659/*
660 * These allocate and release file read/write context information.
661 */
662int nfs_open(struct inode *inode, struct file *filp)
663{
664 struct nfs_open_context *ctx;
665 struct rpc_cred *cred;
666
98a8e323 667 cred = rpc_lookup_cred();
1da177e4
LT
668 if (IS_ERR(cred))
669 return PTR_ERR(cred);
01cce933 670 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
1da177e4
LT
671 put_rpccred(cred);
672 if (ctx == NULL)
673 return -ENOMEM;
674 ctx->mode = filp->f_mode;
675 nfs_file_set_open_context(filp, ctx);
676 put_nfs_open_context(ctx);
1da177e4
LT
677 return 0;
678}
679
680int nfs_release(struct inode *inode, struct file *filp)
681{
1da177e4
LT
682 nfs_file_clear_open_context(filp);
683 return 0;
684}
685
686/*
687 * This function is called whenever some part of NFS notices that
688 * the cached attributes have to be refreshed.
689 */
690int
691__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
692{
693 int status = -ESTALE;
694 struct nfs_fattr fattr;
695 struct nfs_inode *nfsi = NFS_I(inode);
1da177e4
LT
696
697 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
698 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
699
1842bfb4 700 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
85233a7a 701 if (is_bad_inode(inode))
1da177e4
LT
702 goto out_nowait;
703 if (NFS_STALE(inode))
704 goto out_nowait;
705
412d582e
CL
706 status = nfs_wait_on_inode(inode);
707 if (status < 0)
708 goto out;
7fdc49c4
TM
709
710 status = -ESTALE;
711 if (NFS_STALE(inode))
712 goto out;
1da177e4 713
1da177e4
LT
714 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
715 if (status != 0) {
716 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
717 inode->i_sb->s_id,
718 (long long)NFS_FILEID(inode), status);
719 if (status == -ESTALE) {
720 nfs_zap_caches(inode);
721 if (!S_ISDIR(inode->i_mode))
3a10c30a 722 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1da177e4
LT
723 }
724 goto out;
725 }
726
33801147 727 spin_lock(&inode->i_lock);
24aa1fe6 728 status = nfs_update_inode(inode, &fattr);
1da177e4 729 if (status) {
33801147 730 spin_unlock(&inode->i_lock);
1da177e4
LT
731 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
732 inode->i_sb->s_id,
733 (long long)NFS_FILEID(inode), status);
734 goto out;
735 }
dc59250c 736 spin_unlock(&inode->i_lock);
55296809 737
24aa1fe6 738 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
ada70d94 739 nfs_zap_acl_cache(inode);
55296809 740
1da177e4
LT
741 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
742 inode->i_sb->s_id,
743 (long long)NFS_FILEID(inode));
744
412d582e
CL
745 out:
746 nfs_wake_up_inode(inode);
747
1da177e4 748 out_nowait:
1da177e4
LT
749 return status;
750}
751
752int nfs_attribute_timeout(struct inode *inode)
753{
754 struct nfs_inode *nfsi = NFS_I(inode);
755
756 if (nfs_have_delegation(inode, FMODE_READ))
757 return 0;
b0b53973
TM
758 /*
759 * Special case: if the attribute timeout is set to 0, then always
760 * treat the cache as having expired (unless holding
761 * a delegation).
762 */
763 if (nfsi->attrtimeo == 0)
764 return 1;
c7e15961 765 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1da177e4
LT
766}
767
768/**
769 * nfs_revalidate_inode - Revalidate the inode attributes
770 * @server - pointer to nfs_server struct
771 * @inode - pointer to inode struct
772 *
773 * Updates inode attribute information by retrieving the data from the server.
774 */
775int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
776{
44b11874 777 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
1da177e4
LT
778 && !nfs_attribute_timeout(inode))
779 return NFS_STALE(inode) ? -ESTALE : 0;
780 return __nfs_revalidate_inode(server, inode);
781}
782
717d44e8
TM
783static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
784{
785 struct nfs_inode *nfsi = NFS_I(inode);
786
787 if (mapping->nrpages != 0) {
788 int ret = invalidate_inode_pages2(mapping);
789 if (ret < 0)
790 return ret;
791 }
792 spin_lock(&inode->i_lock);
793 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
2f78e431 794 if (S_ISDIR(inode->i_mode))
717d44e8 795 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
717d44e8
TM
796 spin_unlock(&inode->i_lock);
797 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
798 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
799 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
800 return 0;
801}
802
803static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
804{
805 int ret = 0;
806
807 mutex_lock(&inode->i_mutex);
808 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
809 ret = nfs_sync_mapping(mapping);
810 if (ret == 0)
811 ret = nfs_invalidate_mapping_nolock(inode, mapping);
812 }
813 mutex_unlock(&inode->i_mutex);
814 return ret;
815}
816
7d52e862 817/**
717d44e8 818 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
7d52e862
TM
819 * @inode - pointer to host inode
820 * @mapping - pointer to mapping
821 */
717d44e8 822int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
7d52e862
TM
823{
824 struct nfs_inode *nfsi = NFS_I(inode);
44b11874
TM
825 int ret = 0;
826
44b11874 827 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
717d44e8 828 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
44b11874 829 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
717d44e8
TM
830 if (ret < 0)
831 goto out;
832 }
833 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
834 ret = nfs_invalidate_mapping_nolock(inode, mapping);
835out:
836 return ret;
837}
7d52e862 838
717d44e8
TM
839/**
840 * nfs_revalidate_mapping - Revalidate the pagecache
841 * @inode - pointer to host inode
842 * @mapping - pointer to mapping
843 *
844 * This version of the function will take the inode->i_mutex and attempt to
845 * flush out all dirty data if it needs to invalidate the page cache.
846 */
847int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
848{
849 struct nfs_inode *nfsi = NFS_I(inode);
850 int ret = 0;
dc59250c 851
717d44e8
TM
852 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
853 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
854 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
855 if (ret < 0)
856 goto out;
7d52e862 857 }
717d44e8
TM
858 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
859 ret = nfs_invalidate_mapping(inode, mapping);
cd9ae2b6 860out:
44b11874 861 return ret;
7d52e862
TM
862}
863
a895b4a1
TM
864static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
865{
866 struct nfs_inode *nfsi = NFS_I(inode);
867
70ca8852
TM
868 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
869 nfsi->change_attr == fattr->pre_change_attr) {
870 nfsi->change_attr = fattr->change_attr;
871 if (S_ISDIR(inode->i_mode))
872 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
873 }
a895b4a1
TM
874 /* If we have atomic WCC data, we may update some attributes */
875 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
17cadc95 876 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
a895b4a1 877 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
a895b4a1
TM
878 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
879 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
e323ea46
TM
880 if (S_ISDIR(inode->i_mode))
881 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
a895b4a1 882 }
a3d01454 883 if (i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) &&
8a8c74bf 884 nfsi->npages == 0)
a3d01454 885 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
a895b4a1
TM
886 }
887}
888
1da177e4 889/**
33801147 890 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1da177e4
LT
891 * @inode - pointer to inode
892 * @fattr - updated attributes
893 *
894 * Verifies the attribute cache. If we have just changed the attributes,
895 * so that fattr carries weak cache consistency data, then it may
896 * also update the ctime/mtime/change_attribute.
897 */
33801147 898static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1da177e4
LT
899{
900 struct nfs_inode *nfsi = NFS_I(inode);
901 loff_t cur_size, new_isize;
2a3f5fd4 902 unsigned long invalid = 0;
1da177e4 903
dc59250c 904
ca62b9c3
TM
905 /* Has the inode gone and changed behind our back? */
906 if (nfsi->fileid != fattr->fileid
907 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
908 return -EIO;
909 }
910
a895b4a1
TM
911 /* Do atomic weak cache consistency updates */
912 nfs_wcc_update_inode(inode, fattr);
1da177e4 913
9d1e9232 914 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
f1bb0b92 915 nfsi->change_attr != fattr->change_attr)
2a3f5fd4 916 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1da177e4 917
1da177e4 918 /* Verify a few of the more important attributes */
f1bb0b92 919 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
2a3f5fd4 920 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
ca62b9c3
TM
921
922 cur_size = i_size_read(inode);
923 new_isize = nfs_size_to_loff_t(fattr->size);
fb374d24 924 if (cur_size != new_isize && nfsi->npages == 0)
2a3f5fd4 925 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1da177e4
LT
926
927 /* Have any file permissions changed? */
928 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
929 || inode->i_uid != fattr->uid
930 || inode->i_gid != fattr->gid)
2a3f5fd4 931 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1da177e4
LT
932
933 /* Has the link count changed? */
934 if (inode->i_nlink != fattr->nlink)
2a3f5fd4 935 invalid |= NFS_INO_INVALID_ATTR;
1da177e4
LT
936
937 if (!timespec_equal(&inode->i_atime, &fattr->atime))
2a3f5fd4
TM
938 invalid |= NFS_INO_INVALID_ATIME;
939
940 if (invalid != 0)
941 nfsi->cache_validity |= invalid;
942 else
943 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
944 | NFS_INO_INVALID_ATIME
945 | NFS_INO_REVAL_PAGECACHE);
1da177e4 946
33801147 947 nfsi->read_cache_jiffies = fattr->time_start;
1da177e4
LT
948 return 0;
949}
950
33801147
TM
951/**
952 * nfs_refresh_inode - try to update the inode attribute cache
953 * @inode - pointer to inode
954 * @fattr - updated attributes
955 *
956 * Check that an RPC call that returned attributes has not overlapped with
957 * other recent updates of the inode metadata, then decide whether it is
958 * safe to do a full update of the inode attributes, or whether just to
959 * call nfs_check_inode_attributes.
960 */
961int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
962{
963 struct nfs_inode *nfsi = NFS_I(inode);
964 int status;
965
966 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
967 return 0;
968 spin_lock(&inode->i_lock);
33801147 969 if (time_after(fattr->time_start, nfsi->last_updated))
24aa1fe6 970 status = nfs_update_inode(inode, fattr);
33801147
TM
971 else
972 status = nfs_check_inode_attributes(inode, fattr);
973
974 spin_unlock(&inode->i_lock);
975 return status;
976}
977
decf491f
TM
978/**
979 * nfs_post_op_update_inode - try to update the inode attribute cache
980 * @inode - pointer to inode
981 * @fattr - updated attributes
982 *
983 * After an operation that has changed the inode metadata, mark the
984 * attribute cache as being invalid, then try to update it.
f551e44f
CL
985 *
986 * NB: if the server didn't return any post op attributes, this
987 * function will force the retrieval of attributes before the next
988 * NFS request. Thus it should be used only for operations that
989 * are expected to change one or more attributes, to avoid
990 * unnecessary NFS requests and trips through nfs_update_inode().
decf491f
TM
991 */
992int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
993{
994 struct nfs_inode *nfsi = NFS_I(inode);
decf491f 995
7668fdbe 996 spin_lock(&inode->i_lock);
2a3f5fd4 997 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
40d24704
TM
998 if (S_ISDIR(inode->i_mode))
999 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
7668fdbe 1000 spin_unlock(&inode->i_lock);
2a3f5fd4 1001 return nfs_refresh_inode(inode, fattr);
decf491f
TM
1002}
1003
70ca8852
TM
1004/**
1005 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1006 * @inode - pointer to inode
1007 * @fattr - updated attributes
1008 *
1009 * After an operation that has changed the inode metadata, mark the
1010 * attribute cache as being invalid, then try to update it. Fake up
1011 * weak cache consistency data, if none exist.
1012 *
1013 * This function is mainly designed to be used by the ->write_done() functions.
1014 */
1015int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1016{
1017 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1018 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
1019 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1020 fattr->valid |= NFS_ATTR_WCC_V4;
1021 }
1022 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
1023 (fattr->valid & NFS_ATTR_WCC) == 0) {
1024 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1025 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
a3d01454 1026 fattr->pre_size = i_size_read(inode);
70ca8852
TM
1027 fattr->valid |= NFS_ATTR_WCC;
1028 }
1029 return nfs_post_op_update_inode(inode, fattr);
1030}
1031
1da177e4
LT
1032/*
1033 * Many nfs protocol calls return the new file attributes after
1034 * an operation. Here we update the inode to reflect the state
1035 * of the server's inode.
1036 *
1037 * This is a bit tricky because we have to make sure all dirty pages
1038 * have been sent off to the server before calling invalidate_inode_pages.
1039 * To make sure no other process adds more write requests while we try
1040 * our best to flush them, we make them sleep during the attribute refresh.
1041 *
1042 * A very similar scenario holds for the dir cache.
1043 */
24aa1fe6 1044static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1da177e4 1045{
8b4bdcf8 1046 struct nfs_server *server;
1da177e4 1047 struct nfs_inode *nfsi = NFS_I(inode);
951a143b 1048 loff_t cur_isize, new_isize;
2a3f5fd4 1049 unsigned long invalid = 0;
3e7d950a 1050 unsigned long now = jiffies;
1da177e4
LT
1051
1052 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
3110ff80 1053 __func__, inode->i_sb->s_id, inode->i_ino,
1da177e4
LT
1054 atomic_read(&inode->i_count), fattr->valid);
1055
325cfed9
CL
1056 if (nfsi->fileid != fattr->fileid)
1057 goto out_fileid;
1da177e4
LT
1058
1059 /*
1060 * Make sure the inode's type hasn't changed.
1061 */
33801147 1062 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1da177e4
LT
1063 goto out_changed;
1064
8b4bdcf8 1065 server = NFS_SERVER(inode);
a0356862 1066 /* Update the fsid? */
c37dcd33
TM
1067 if (S_ISDIR(inode->i_mode) &&
1068 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1069 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
8b4bdcf8
TM
1070 server->fsid = fattr->fsid;
1071
1da177e4
LT
1072 /*
1073 * Update the read time so we don't revalidate too often.
1074 */
33801147 1075 nfsi->read_cache_jiffies = fattr->time_start;
3e7d950a 1076
f2115dc9
TM
1077 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1078 | NFS_INO_REVAL_PAGECACHE);
1da177e4 1079
a895b4a1
TM
1080 /* Do atomic weak cache consistency updates */
1081 nfs_wcc_update_inode(inode, fattr);
1082
47aabaa7
TM
1083 /* More cache consistency checks */
1084 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1085 /* NFSv2/v3: Check if the mtime agrees */
1086 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1087 dprintk("NFS: mtime change on server for file %s/%ld\n",
1088 inode->i_sb->s_id, inode->i_ino);
1089 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
bfc69a45
TM
1090 if (S_ISDIR(inode->i_mode))
1091 nfs_force_lookup_revalidate(inode);
47aabaa7
TM
1092 }
1093 /* If ctime has changed we should definitely clear access+acl caches */
12b373eb 1094 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
47aabaa7 1095 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
47aabaa7
TM
1096 } else if (nfsi->change_attr != fattr->change_attr) {
1097 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1098 inode->i_sb->s_id, inode->i_ino);
1099 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
bfc69a45
TM
1100 if (S_ISDIR(inode->i_mode))
1101 nfs_force_lookup_revalidate(inode);
47aabaa7
TM
1102 }
1103
951a143b 1104 /* Check if our cached file size is stale */
1da177e4
LT
1105 new_isize = nfs_size_to_loff_t(fattr->size);
1106 cur_isize = i_size_read(inode);
951a143b 1107 if (new_isize != cur_isize) {
b64e8a5e
TM
1108 /* Do we perhaps have any outstanding writes, or has
1109 * the file grown beyond our last write? */
1110 if (nfsi->npages == 0 || new_isize > cur_isize) {
a3d01454 1111 i_size_write(inode, new_isize);
1da177e4
LT
1112 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1113 }
951a143b
TM
1114 dprintk("NFS: isize change on server for file %s/%ld\n",
1115 inode->i_sb->s_id, inode->i_ino);
1da177e4
LT
1116 }
1117
1da177e4 1118
47aabaa7
TM
1119 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1120 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1da177e4 1121 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
47aabaa7 1122 nfsi->change_attr = fattr->change_attr;
1da177e4
LT
1123
1124 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1125 inode->i_uid != fattr->uid ||
1126 inode->i_gid != fattr->gid)
ada70d94 1127 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1da177e4
LT
1128
1129 inode->i_mode = fattr->mode;
1130 inode->i_nlink = fattr->nlink;
1131 inode->i_uid = fattr->uid;
1132 inode->i_gid = fattr->gid;
1133
1134 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1135 /*
1136 * report the blocks in 512byte units
1137 */
1138 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1da177e4
LT
1139 } else {
1140 inode->i_blocks = fattr->du.nfs2.blocks;
1da177e4
LT
1141 }
1142
1143 /* Update attrtimeo value if we're out of the unstable period */
1144 if (invalid & NFS_INO_INVALID_ATTR) {
91d5b470 1145 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1da177e4 1146 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
3e7d950a 1147 nfsi->attrtimeo_timestamp = now;
6d2b2966
TM
1148 nfsi->last_updated = now;
1149 } else {
1150 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1151 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1152 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1153 nfsi->attrtimeo_timestamp = now;
1154 }
1155 /*
1156 * Avoid jiffy wraparound issues with nfsi->last_updated
1157 */
1158 if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now))
1159 nfsi->last_updated = nfsi->read_cache_jiffies;
1da177e4 1160 }
f2115dc9 1161 invalid &= ~NFS_INO_INVALID_ATTR;
1da177e4
LT
1162 /* Don't invalidate the data if we were to blame */
1163 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1164 || S_ISLNK(inode->i_mode)))
1165 invalid &= ~NFS_INO_INVALID_DATA;
412c77ce
TM
1166 if (!nfs_have_delegation(inode, FMODE_READ) ||
1167 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
55296809 1168 nfsi->cache_validity |= invalid;
412c77ce 1169 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1da177e4
LT
1170
1171 return 0;
1172 out_changed:
1173 /*
1174 * Big trouble! The inode has become a different object.
1175 */
1da177e4 1176 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
3110ff80 1177 __func__, inode->i_ino, inode->i_mode, fattr->mode);
b37b03b7 1178 out_err:
1da177e4
LT
1179 /*
1180 * No need to worry about unhashing the dentry, as the
1181 * lookup validation will know that the inode is bad.
1182 * (But we fall through to invalidate the caches.)
1183 */
1184 nfs_invalidate_inode(inode);
1da177e4 1185 return -ESTALE;
325cfed9
CL
1186
1187 out_fileid:
1188 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1189 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
54ceac45 1190 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
325cfed9
CL
1191 (long long)nfsi->fileid, (long long)fattr->fileid);
1192 goto out_err;
1da177e4
LT
1193}
1194
55a97593 1195
1da177e4
LT
1196#ifdef CONFIG_NFS_V4
1197
1da177e4
LT
1198/*
1199 * Clean out any remaining NFSv4 state that might be left over due
1200 * to open() calls that passed nfs_atomic_lookup, but failed to call
1201 * nfs_open().
1202 */
f7b422b1 1203void nfs4_clear_inode(struct inode *inode)
1da177e4 1204{
1da177e4 1205 /* If we are holding a delegation, return it! */
e6f81075 1206 nfs_inode_return_delegation_noreclaim(inode);
1da177e4
LT
1207 /* First call standard NFS clear_inode() code */
1208 nfs_clear_inode(inode);
1da177e4 1209}
1da177e4
LT
1210#endif
1211
f7b422b1 1212struct inode *nfs_alloc_inode(struct super_block *sb)
1da177e4
LT
1213{
1214 struct nfs_inode *nfsi;
e94b1766 1215 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1da177e4
LT
1216 if (!nfsi)
1217 return NULL;
55296809
CL
1218 nfsi->flags = 0UL;
1219 nfsi->cache_validity = 0UL;
458818ed
TM
1220#ifdef CONFIG_NFS_V3_ACL
1221 nfsi->acl_access = ERR_PTR(-EAGAIN);
1222 nfsi->acl_default = ERR_PTR(-EAGAIN);
1223#endif
e50a1c2e
BF
1224#ifdef CONFIG_NFS_V4
1225 nfsi->nfs4_acl = NULL;
1226#endif /* CONFIG_NFS_V4 */
1da177e4
LT
1227 return &nfsi->vfs_inode;
1228}
1229
f7b422b1 1230void nfs_destroy_inode(struct inode *inode)
1da177e4
LT
1231{
1232 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1233}
1234
d75d5414
AM
1235static inline void nfs4_init_once(struct nfs_inode *nfsi)
1236{
1237#ifdef CONFIG_NFS_V4
1238 INIT_LIST_HEAD(&nfsi->open_states);
1239 nfsi->delegation = NULL;
1240 nfsi->delegation_state = 0;
1241 init_rwsem(&nfsi->rwsem);
1242#endif
1243}
f7b422b1 1244
51cc5068 1245static void init_once(void *foo)
1da177e4
LT
1246{
1247 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1248
a35afb83 1249 inode_init_once(&nfsi->vfs_inode);
a35afb83
CL
1250 INIT_LIST_HEAD(&nfsi->open_files);
1251 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1252 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1253 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
a35afb83
CL
1254 nfsi->ncommit = 0;
1255 nfsi->npages = 0;
565277f6
TM
1256 atomic_set(&nfsi->silly_count, 1);
1257 INIT_HLIST_HEAD(&nfsi->silly_list);
1258 init_waitqueue_head(&nfsi->waitqueue);
a35afb83 1259 nfs4_init_once(nfsi);
1da177e4 1260}
20c2df83 1261
f7b422b1 1262static int __init nfs_init_inodecache(void)
1da177e4
LT
1263{
1264 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1265 sizeof(struct nfs_inode),
fffb60f9
PJ
1266 0, (SLAB_RECLAIM_ACCOUNT|
1267 SLAB_MEM_SPREAD),
20c2df83 1268 init_once);
1da177e4
LT
1269 if (nfs_inode_cachep == NULL)
1270 return -ENOMEM;
1271
1272 return 0;
1273}
1274
266bee88 1275static void nfs_destroy_inodecache(void)
1da177e4 1276{
1a1d92c1 1277 kmem_cache_destroy(nfs_inode_cachep);
1da177e4
LT
1278}
1279
5746006f
TM
1280struct workqueue_struct *nfsiod_workqueue;
1281
1282/*
1283 * start up the nfsiod workqueue
1284 */
1285static int nfsiod_start(void)
1286{
1287 struct workqueue_struct *wq;
1288 dprintk("RPC: creating workqueue nfsiod\n");
1289 wq = create_singlethread_workqueue("nfsiod");
1290 if (wq == NULL)
1291 return -ENOMEM;
1292 nfsiod_workqueue = wq;
1293 return 0;
1294}
1295
1296/*
1297 * Destroy the nfsiod workqueue
1298 */
1299static void nfsiod_stop(void)
1300{
1301 struct workqueue_struct *wq;
1302
1303 wq = nfsiod_workqueue;
1304 if (wq == NULL)
1305 return;
1306 nfsiod_workqueue = NULL;
1307 destroy_workqueue(wq);
1308}
1309
1da177e4
LT
1310/*
1311 * Initialize NFS
1312 */
1313static int __init init_nfs_fs(void)
1314{
1315 int err;
1316
5746006f
TM
1317 err = nfsiod_start();
1318 if (err)
1319 goto out6;
1320
6aaca566
DH
1321 err = nfs_fs_proc_init();
1322 if (err)
1323 goto out5;
1324
1da177e4
LT
1325 err = nfs_init_nfspagecache();
1326 if (err)
1327 goto out4;
1328
1329 err = nfs_init_inodecache();
1330 if (err)
1331 goto out3;
1332
1333 err = nfs_init_readpagecache();
1334 if (err)
1335 goto out2;
1336
1337 err = nfs_init_writepagecache();
1338 if (err)
1339 goto out1;
1340
1da177e4
LT
1341 err = nfs_init_directcache();
1342 if (err)
1343 goto out0;
1da177e4
LT
1344
1345#ifdef CONFIG_PROC_FS
1346 rpc_proc_register(&nfs_rpcstat);
1347#endif
f7b422b1 1348 if ((err = register_nfs_fs()) != 0)
1da177e4
LT
1349 goto out;
1350 return 0;
1351out:
1352#ifdef CONFIG_PROC_FS
1353 rpc_proc_unregister("nfs");
1354#endif
1da177e4 1355 nfs_destroy_directcache();
6b59a754 1356out0:
6b59a754 1357 nfs_destroy_writepagecache();
1da177e4
LT
1358out1:
1359 nfs_destroy_readpagecache();
1360out2:
1361 nfs_destroy_inodecache();
1362out3:
1363 nfs_destroy_nfspagecache();
1364out4:
6aaca566
DH
1365 nfs_fs_proc_exit();
1366out5:
5746006f
TM
1367 nfsiod_stop();
1368out6:
1da177e4
LT
1369 return err;
1370}
1371
1372static void __exit exit_nfs_fs(void)
1373{
1da177e4 1374 nfs_destroy_directcache();
1da177e4
LT
1375 nfs_destroy_writepagecache();
1376 nfs_destroy_readpagecache();
1377 nfs_destroy_inodecache();
1378 nfs_destroy_nfspagecache();
1379#ifdef CONFIG_PROC_FS
1380 rpc_proc_unregister("nfs");
1381#endif
f7b422b1 1382 unregister_nfs_fs();
6aaca566 1383 nfs_fs_proc_exit();
5746006f 1384 nfsiod_stop();
1da177e4
LT
1385}
1386
1387/* Not quite true; I just maintain it */
1388MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1389MODULE_LICENSE("GPL");
f43bf0be 1390module_param(enable_ino64, bool, 0644);
1da177e4
LT
1391
1392module_init(init_nfs_fs)
1393module_exit(exit_nfs_fs)