RPC/RDMA: ensure connection attempt is complete before signalling.
[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) {
2f28ea61 456 spin_lock(&inode->i_lock);
1da177e4 457 if ((attr->ia_valid & ATTR_MODE) != 0) {
65e4308d
TM
458 int mode = attr->ia_mode & S_IALLUGO;
459 mode |= inode->i_mode & ~S_IALLUGO;
1da177e4
LT
460 inode->i_mode = mode;
461 }
462 if ((attr->ia_valid & ATTR_UID) != 0)
463 inode->i_uid = attr->ia_uid;
464 if ((attr->ia_valid & ATTR_GID) != 0)
465 inode->i_gid = attr->ia_gid;
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
1da177e4
LT
475int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
476{
477 struct inode *inode = dentry->d_inode;
55296809 478 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
1da177e4
LT
479 int err;
480
28c494c5
CL
481 /*
482 * Flush out writes to the server in order to update c/mtime.
483 *
484 * Hold the i_mutex to suspend application writes temporarily;
485 * this prevents long-running writing applications from blocking
486 * nfs_wb_nocommit.
487 */
488 if (S_ISREG(inode->i_mode)) {
489 mutex_lock(&inode->i_mutex);
ed90ef51 490 nfs_wb_nocommit(inode);
28c494c5
CL
491 mutex_unlock(&inode->i_mutex);
492 }
fc33a7bb
CH
493
494 /*
495 * We may force a getattr if the user cares about atime.
496 *
497 * Note that we only have to check the vfsmount flags here:
498 * - NFS always sets S_NOATIME by so checking it would give a
499 * bogus result
500 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
501 * no point in checking those.
502 */
503 if ((mnt->mnt_flags & MNT_NOATIME) ||
504 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
1da177e4 505 need_atime = 0;
fc33a7bb 506
1da177e4
LT
507 if (need_atime)
508 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
509 else
510 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
4e769b93 511 if (!err) {
1da177e4 512 generic_fillattr(inode, stat);
f43bf0be 513 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
4e769b93 514 }
1da177e4
LT
515 return err;
516}
517
b92dccf6 518static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred)
1da177e4
LT
519{
520 struct nfs_open_context *ctx;
521
f52720ca 522 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
1da177e4 523 if (ctx != NULL) {
88be9f99
TM
524 ctx->path.dentry = dget(dentry);
525 ctx->path.mnt = mntget(mnt);
1da177e4
LT
526 ctx->cred = get_rpccred(cred);
527 ctx->state = NULL;
528 ctx->lockowner = current->files;
66d3aac0 529 ctx->flags = 0;
1da177e4 530 ctx->error = 0;
00a92642 531 ctx->dir_cookie = 0;
5e11934d 532 atomic_set(&ctx->count, 1);
1da177e4
LT
533 }
534 return ctx;
535}
536
537struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
538{
539 if (ctx != NULL)
5e11934d 540 atomic_inc(&ctx->count);
1da177e4
LT
541 return ctx;
542}
543
a49c3c77 544static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait)
1da177e4 545{
383ba719 546 struct inode *inode;
3bec63db 547
383ba719
TM
548 if (ctx == NULL)
549 return;
550
551 inode = ctx->path.dentry->d_inode;
5e11934d
TM
552 if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock))
553 return;
554 list_del(&ctx->list);
555 spin_unlock(&inode->i_lock);
a49c3c77
TM
556 if (ctx->state != NULL) {
557 if (wait)
558 nfs4_close_sync(&ctx->path, ctx->state, ctx->mode);
559 else
560 nfs4_close_state(&ctx->path, ctx->state, ctx->mode);
561 }
3bec63db
TM
562 if (ctx->cred != NULL)
563 put_rpccred(ctx->cred);
31f31db1 564 path_put(&ctx->path);
3bec63db
TM
565 kfree(ctx);
566}
567
a49c3c77
TM
568void put_nfs_open_context(struct nfs_open_context *ctx)
569{
570 __put_nfs_open_context(ctx, 0);
571}
572
573static void put_nfs_open_context_sync(struct nfs_open_context *ctx)
574{
575 __put_nfs_open_context(ctx, 1);
576}
577
1da177e4
LT
578/*
579 * Ensure that mmap has a recent RPC credential for use when writing out
580 * shared pages
581 */
b92dccf6 582static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
1da177e4 583{
01cce933 584 struct inode *inode = filp->f_path.dentry->d_inode;
1da177e4
LT
585 struct nfs_inode *nfsi = NFS_I(inode);
586
587 filp->private_data = get_nfs_open_context(ctx);
588 spin_lock(&inode->i_lock);
589 list_add(&ctx->list, &nfsi->open_files);
590 spin_unlock(&inode->i_lock);
591}
592
d530838b
TM
593/*
594 * Given an inode, search for an open context with the desired characteristics
595 */
596struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode)
1da177e4
LT
597{
598 struct nfs_inode *nfsi = NFS_I(inode);
599 struct nfs_open_context *pos, *ctx = NULL;
600
601 spin_lock(&inode->i_lock);
602 list_for_each_entry(pos, &nfsi->open_files, list) {
d530838b
TM
603 if (cred != NULL && pos->cred != cred)
604 continue;
1da177e4
LT
605 if ((pos->mode & mode) == mode) {
606 ctx = get_nfs_open_context(pos);
607 break;
608 }
609 }
610 spin_unlock(&inode->i_lock);
611 return ctx;
612}
613
b92dccf6 614static void nfs_file_clear_open_context(struct file *filp)
1da177e4 615{
01cce933 616 struct inode *inode = filp->f_path.dentry->d_inode;
cd3758e3 617 struct nfs_open_context *ctx = nfs_file_open_context(filp);
1da177e4
LT
618
619 if (ctx) {
620 filp->private_data = NULL;
621 spin_lock(&inode->i_lock);
622 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
623 spin_unlock(&inode->i_lock);
a49c3c77 624 put_nfs_open_context_sync(ctx);
1da177e4
LT
625 }
626}
627
628/*
629 * These allocate and release file read/write context information.
630 */
631int nfs_open(struct inode *inode, struct file *filp)
632{
633 struct nfs_open_context *ctx;
634 struct rpc_cred *cred;
635
98a8e323 636 cred = rpc_lookup_cred();
1da177e4
LT
637 if (IS_ERR(cred))
638 return PTR_ERR(cred);
01cce933 639 ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred);
1da177e4
LT
640 put_rpccred(cred);
641 if (ctx == NULL)
642 return -ENOMEM;
643 ctx->mode = filp->f_mode;
644 nfs_file_set_open_context(filp, ctx);
645 put_nfs_open_context(ctx);
1da177e4
LT
646 return 0;
647}
648
649int nfs_release(struct inode *inode, struct file *filp)
650{
1da177e4
LT
651 nfs_file_clear_open_context(filp);
652 return 0;
653}
654
655/*
656 * This function is called whenever some part of NFS notices that
657 * the cached attributes have to be refreshed.
658 */
659int
660__nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
661{
662 int status = -ESTALE;
663 struct nfs_fattr fattr;
664 struct nfs_inode *nfsi = NFS_I(inode);
1da177e4
LT
665
666 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
667 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
668
85233a7a 669 if (is_bad_inode(inode))
691beb13 670 goto out;
1da177e4 671 if (NFS_STALE(inode))
412d582e 672 goto out;
7fdc49c4 673
7fdc49c4
TM
674 if (NFS_STALE(inode))
675 goto out;
1da177e4 676
691beb13 677 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
1da177e4
LT
678 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr);
679 if (status != 0) {
680 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
681 inode->i_sb->s_id,
682 (long long)NFS_FILEID(inode), status);
683 if (status == -ESTALE) {
684 nfs_zap_caches(inode);
685 if (!S_ISDIR(inode->i_mode))
3a10c30a 686 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1da177e4
LT
687 }
688 goto out;
689 }
690
4dc05efb 691 status = nfs_refresh_inode(inode, &fattr);
1da177e4
LT
692 if (status) {
693 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
694 inode->i_sb->s_id,
695 (long long)NFS_FILEID(inode), status);
696 goto out;
697 }
55296809 698
24aa1fe6 699 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
ada70d94 700 nfs_zap_acl_cache(inode);
55296809 701
1da177e4
LT
702 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
703 inode->i_sb->s_id,
704 (long long)NFS_FILEID(inode));
705
412d582e 706 out:
1da177e4
LT
707 return status;
708}
709
710int nfs_attribute_timeout(struct inode *inode)
711{
712 struct nfs_inode *nfsi = NFS_I(inode);
713
714 if (nfs_have_delegation(inode, FMODE_READ))
715 return 0;
b0b53973
TM
716 /*
717 * Special case: if the attribute timeout is set to 0, then always
718 * treat the cache as having expired (unless holding
719 * a delegation).
720 */
721 if (nfsi->attrtimeo == 0)
722 return 1;
c7e15961 723 return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
1da177e4
LT
724}
725
726/**
727 * nfs_revalidate_inode - Revalidate the inode attributes
728 * @server - pointer to nfs_server struct
729 * @inode - pointer to inode struct
730 *
731 * Updates inode attribute information by retrieving the data from the server.
732 */
733int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
734{
44b11874 735 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
1da177e4
LT
736 && !nfs_attribute_timeout(inode))
737 return NFS_STALE(inode) ? -ESTALE : 0;
738 return __nfs_revalidate_inode(server, inode);
739}
740
717d44e8
TM
741static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
742{
743 struct nfs_inode *nfsi = NFS_I(inode);
744
745 if (mapping->nrpages != 0) {
746 int ret = invalidate_inode_pages2(mapping);
747 if (ret < 0)
748 return ret;
749 }
750 spin_lock(&inode->i_lock);
751 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
2f78e431 752 if (S_ISDIR(inode->i_mode))
717d44e8 753 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
717d44e8
TM
754 spin_unlock(&inode->i_lock);
755 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
756 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
757 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
758 return 0;
759}
760
761static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
762{
763 int ret = 0;
764
765 mutex_lock(&inode->i_mutex);
766 if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) {
767 ret = nfs_sync_mapping(mapping);
768 if (ret == 0)
769 ret = nfs_invalidate_mapping_nolock(inode, mapping);
770 }
771 mutex_unlock(&inode->i_mutex);
772 return ret;
773}
774
7d52e862 775/**
717d44e8 776 * nfs_revalidate_mapping_nolock - Revalidate the pagecache
7d52e862
TM
777 * @inode - pointer to host inode
778 * @mapping - pointer to mapping
779 */
717d44e8 780int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping)
7d52e862
TM
781{
782 struct nfs_inode *nfsi = NFS_I(inode);
44b11874
TM
783 int ret = 0;
784
44b11874 785 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
717d44e8 786 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
44b11874 787 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
717d44e8
TM
788 if (ret < 0)
789 goto out;
790 }
791 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
792 ret = nfs_invalidate_mapping_nolock(inode, mapping);
793out:
794 return ret;
795}
7d52e862 796
717d44e8
TM
797/**
798 * nfs_revalidate_mapping - Revalidate the pagecache
799 * @inode - pointer to host inode
800 * @mapping - pointer to mapping
801 *
802 * This version of the function will take the inode->i_mutex and attempt to
803 * flush out all dirty data if it needs to invalidate the page cache.
804 */
805int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
806{
807 struct nfs_inode *nfsi = NFS_I(inode);
808 int ret = 0;
dc59250c 809
717d44e8
TM
810 if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
811 || nfs_attribute_timeout(inode) || NFS_STALE(inode)) {
812 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
813 if (ret < 0)
814 goto out;
7d52e862 815 }
717d44e8
TM
816 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
817 ret = nfs_invalidate_mapping(inode, mapping);
cd9ae2b6 818out:
44b11874 819 return ret;
7d52e862
TM
820}
821
a895b4a1
TM
822static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
823{
824 struct nfs_inode *nfsi = NFS_I(inode);
825
70ca8852
TM
826 if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 &&
827 nfsi->change_attr == fattr->pre_change_attr) {
828 nfsi->change_attr = fattr->change_attr;
829 if (S_ISDIR(inode->i_mode))
830 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
831 }
a895b4a1
TM
832 /* If we have atomic WCC data, we may update some attributes */
833 if ((fattr->valid & NFS_ATTR_WCC) != 0) {
17cadc95 834 if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime))
a895b4a1 835 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
a895b4a1
TM
836 if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
837 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
e323ea46
TM
838 if (S_ISDIR(inode->i_mode))
839 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
a895b4a1 840 }
a3d01454 841 if (i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size) &&
8a8c74bf 842 nfsi->npages == 0)
a3d01454 843 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
a895b4a1
TM
844 }
845}
846
1da177e4 847/**
33801147 848 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
1da177e4
LT
849 * @inode - pointer to inode
850 * @fattr - updated attributes
851 *
852 * Verifies the attribute cache. If we have just changed the attributes,
853 * so that fattr carries weak cache consistency data, then it may
854 * also update the ctime/mtime/change_attribute.
855 */
33801147 856static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
1da177e4
LT
857{
858 struct nfs_inode *nfsi = NFS_I(inode);
859 loff_t cur_size, new_isize;
2a3f5fd4 860 unsigned long invalid = 0;
1da177e4 861
dc59250c 862
ca62b9c3
TM
863 /* Has the inode gone and changed behind our back? */
864 if (nfsi->fileid != fattr->fileid
865 || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
866 return -EIO;
867 }
868
9d1e9232 869 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
f1bb0b92 870 nfsi->change_attr != fattr->change_attr)
2a3f5fd4 871 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1da177e4 872
1da177e4 873 /* Verify a few of the more important attributes */
f1bb0b92 874 if (!timespec_equal(&inode->i_mtime, &fattr->mtime))
2a3f5fd4 875 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
ca62b9c3
TM
876
877 cur_size = i_size_read(inode);
878 new_isize = nfs_size_to_loff_t(fattr->size);
fb374d24 879 if (cur_size != new_isize && nfsi->npages == 0)
2a3f5fd4 880 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1da177e4
LT
881
882 /* Have any file permissions changed? */
883 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)
884 || inode->i_uid != fattr->uid
885 || inode->i_gid != fattr->gid)
2a3f5fd4 886 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
1da177e4
LT
887
888 /* Has the link count changed? */
889 if (inode->i_nlink != fattr->nlink)
2a3f5fd4 890 invalid |= NFS_INO_INVALID_ATTR;
1da177e4
LT
891
892 if (!timespec_equal(&inode->i_atime, &fattr->atime))
2a3f5fd4
TM
893 invalid |= NFS_INO_INVALID_ATIME;
894
895 if (invalid != 0)
896 nfsi->cache_validity |= invalid;
1da177e4 897
33801147 898 nfsi->read_cache_jiffies = fattr->time_start;
1da177e4
LT
899 return 0;
900}
901
a10ad176 902static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
870a5be8 903{
a10ad176
TM
904 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
905}
906
907static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
908{
909 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
910}
870a5be8 911
a10ad176
TM
912/**
913 * nfs_inode_attrs_need_update - check if the inode attributes need updating
914 * @inode - pointer to inode
915 * @fattr - attributes
916 *
917 * Attempt to divine whether or not an RPC call reply carrying stale
918 * attributes got scheduled after another call carrying updated ones.
919 *
920 * To do so, the function first assumes that a more recent ctime means
921 * that the attributes in fattr are newer, however it also attempt to
922 * catch the case where ctime either didn't change, or went backwards
923 * (if someone reset the clock on the server) by looking at whether
924 * or not this RPC call was started after the inode was last updated.
925 * Note also the check for jiffy wraparound if the last_updated timestamp
926 * is later than 'jiffies'.
927 *
928 * The function returns 'true' if it thinks the attributes in 'fattr' are
929 * more recent than the ones cached in the inode.
930 *
931 */
932static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
933{
934 const struct nfs_inode *nfsi = NFS_I(inode);
935
03254e65
TM
936 return time_after(fattr->time_start, nfsi->last_updated) ||
937 nfs_ctime_need_update(inode, fattr) ||
938 nfs_size_need_update(inode, fattr) ||
939 time_after(nfsi->last_updated, jiffies);
a10ad176
TM
940}
941
942static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
943{
944 if (nfs_inode_attrs_need_update(inode, fattr))
870a5be8
TM
945 return nfs_update_inode(inode, fattr);
946 return nfs_check_inode_attributes(inode, fattr);
947}
948
33801147
TM
949/**
950 * nfs_refresh_inode - try to update the inode attribute cache
951 * @inode - pointer to inode
952 * @fattr - updated attributes
953 *
954 * Check that an RPC call that returned attributes has not overlapped with
955 * other recent updates of the inode metadata, then decide whether it is
956 * safe to do a full update of the inode attributes, or whether just to
957 * call nfs_check_inode_attributes.
958 */
959int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
960{
33801147
TM
961 int status;
962
963 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
964 return 0;
965 spin_lock(&inode->i_lock);
870a5be8 966 status = nfs_refresh_inode_locked(inode, fattr);
33801147
TM
967 spin_unlock(&inode->i_lock);
968 return status;
969}
970
d65f557f
TM
971static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
972{
973 struct nfs_inode *nfsi = NFS_I(inode);
974
975 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
976 if (S_ISDIR(inode->i_mode))
977 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
978 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
979 return 0;
980 return nfs_refresh_inode_locked(inode, fattr);
981}
982
decf491f
TM
983/**
984 * nfs_post_op_update_inode - try to update the inode attribute cache
985 * @inode - pointer to inode
986 * @fattr - updated attributes
987 *
988 * After an operation that has changed the inode metadata, mark the
989 * attribute cache as being invalid, then try to update it.
f551e44f
CL
990 *
991 * NB: if the server didn't return any post op attributes, this
992 * function will force the retrieval of attributes before the next
993 * NFS request. Thus it should be used only for operations that
994 * are expected to change one or more attributes, to avoid
995 * unnecessary NFS requests and trips through nfs_update_inode().
decf491f
TM
996 */
997int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
998{
d65f557f 999 int status;
decf491f 1000
7668fdbe 1001 spin_lock(&inode->i_lock);
d65f557f 1002 status = nfs_post_op_update_inode_locked(inode, fattr);
7668fdbe 1003 spin_unlock(&inode->i_lock);
870a5be8 1004 return status;
decf491f
TM
1005}
1006
70ca8852
TM
1007/**
1008 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1009 * @inode - pointer to inode
1010 * @fattr - updated attributes
1011 *
1012 * After an operation that has changed the inode metadata, mark the
1013 * attribute cache as being invalid, then try to update it. Fake up
1014 * weak cache consistency data, if none exist.
1015 *
1016 * This function is mainly designed to be used by the ->write_done() functions.
1017 */
1018int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1019{
d65f557f
TM
1020 int status;
1021
1022 spin_lock(&inode->i_lock);
1023 /* Don't do a WCC update if these attributes are already stale */
1024 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1025 !nfs_inode_attrs_need_update(inode, fattr)) {
1026 fattr->valid &= ~(NFS_ATTR_WCC_V4|NFS_ATTR_WCC);
1027 goto out_noforce;
1028 }
70ca8852
TM
1029 if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 &&
1030 (fattr->valid & NFS_ATTR_WCC_V4) == 0) {
1031 fattr->pre_change_attr = NFS_I(inode)->change_attr;
1032 fattr->valid |= NFS_ATTR_WCC_V4;
1033 }
1034 if ((fattr->valid & NFS_ATTR_FATTR) != 0 &&
1035 (fattr->valid & NFS_ATTR_WCC) == 0) {
1036 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1037 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
a3d01454 1038 fattr->pre_size = i_size_read(inode);
70ca8852
TM
1039 fattr->valid |= NFS_ATTR_WCC;
1040 }
d65f557f
TM
1041out_noforce:
1042 status = nfs_post_op_update_inode_locked(inode, fattr);
1043 spin_unlock(&inode->i_lock);
1044 return status;
70ca8852
TM
1045}
1046
1da177e4
LT
1047/*
1048 * Many nfs protocol calls return the new file attributes after
1049 * an operation. Here we update the inode to reflect the state
1050 * of the server's inode.
1051 *
1052 * This is a bit tricky because we have to make sure all dirty pages
1053 * have been sent off to the server before calling invalidate_inode_pages.
1054 * To make sure no other process adds more write requests while we try
1055 * our best to flush them, we make them sleep during the attribute refresh.
1056 *
1057 * A very similar scenario holds for the dir cache.
1058 */
24aa1fe6 1059static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1da177e4 1060{
8b4bdcf8 1061 struct nfs_server *server;
1da177e4 1062 struct nfs_inode *nfsi = NFS_I(inode);
951a143b 1063 loff_t cur_isize, new_isize;
2a3f5fd4 1064 unsigned long invalid = 0;
3e7d950a 1065 unsigned long now = jiffies;
1da177e4
LT
1066
1067 dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n",
3110ff80 1068 __func__, inode->i_sb->s_id, inode->i_ino,
1da177e4
LT
1069 atomic_read(&inode->i_count), fattr->valid);
1070
325cfed9
CL
1071 if (nfsi->fileid != fattr->fileid)
1072 goto out_fileid;
1da177e4
LT
1073
1074 /*
1075 * Make sure the inode's type hasn't changed.
1076 */
33801147 1077 if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
1da177e4
LT
1078 goto out_changed;
1079
8b4bdcf8 1080 server = NFS_SERVER(inode);
a0356862 1081 /* Update the fsid? */
c37dcd33
TM
1082 if (S_ISDIR(inode->i_mode) &&
1083 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1084 !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags))
8b4bdcf8
TM
1085 server->fsid = fattr->fsid;
1086
1da177e4
LT
1087 /*
1088 * Update the read time so we don't revalidate too often.
1089 */
33801147 1090 nfsi->read_cache_jiffies = fattr->time_start;
3e7d950a 1091
f2115dc9
TM
1092 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME
1093 | NFS_INO_REVAL_PAGECACHE);
1da177e4 1094
a895b4a1
TM
1095 /* Do atomic weak cache consistency updates */
1096 nfs_wcc_update_inode(inode, fattr);
1097
47aabaa7
TM
1098 /* More cache consistency checks */
1099 if (!(fattr->valid & NFS_ATTR_FATTR_V4)) {
1100 /* NFSv2/v3: Check if the mtime agrees */
1101 if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) {
1102 dprintk("NFS: mtime change on server for file %s/%ld\n",
1103 inode->i_sb->s_id, inode->i_ino);
1104 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
bfc69a45
TM
1105 if (S_ISDIR(inode->i_mode))
1106 nfs_force_lookup_revalidate(inode);
47aabaa7
TM
1107 }
1108 /* If ctime has changed we should definitely clear access+acl caches */
12b373eb 1109 if (!timespec_equal(&inode->i_ctime, &fattr->ctime))
47aabaa7 1110 invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
47aabaa7
TM
1111 } else if (nfsi->change_attr != fattr->change_attr) {
1112 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1113 inode->i_sb->s_id, inode->i_ino);
1114 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
bfc69a45
TM
1115 if (S_ISDIR(inode->i_mode))
1116 nfs_force_lookup_revalidate(inode);
47aabaa7
TM
1117 }
1118
951a143b 1119 /* Check if our cached file size is stale */
1da177e4
LT
1120 new_isize = nfs_size_to_loff_t(fattr->size);
1121 cur_isize = i_size_read(inode);
951a143b 1122 if (new_isize != cur_isize) {
b64e8a5e
TM
1123 /* Do we perhaps have any outstanding writes, or has
1124 * the file grown beyond our last write? */
1125 if (nfsi->npages == 0 || new_isize > cur_isize) {
a3d01454 1126 i_size_write(inode, new_isize);
1da177e4
LT
1127 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1128 }
951a143b
TM
1129 dprintk("NFS: isize change on server for file %s/%ld\n",
1130 inode->i_sb->s_id, inode->i_ino);
1da177e4
LT
1131 }
1132
1da177e4 1133
47aabaa7
TM
1134 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1135 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1da177e4 1136 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
47aabaa7 1137 nfsi->change_attr = fattr->change_attr;
1da177e4
LT
1138
1139 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) ||
1140 inode->i_uid != fattr->uid ||
1141 inode->i_gid != fattr->gid)
ada70d94 1142 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1da177e4
LT
1143
1144 inode->i_mode = fattr->mode;
1145 inode->i_nlink = fattr->nlink;
1146 inode->i_uid = fattr->uid;
1147 inode->i_gid = fattr->gid;
1148
1149 if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) {
1150 /*
1151 * report the blocks in 512byte units
1152 */
1153 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1da177e4
LT
1154 } else {
1155 inode->i_blocks = fattr->du.nfs2.blocks;
1da177e4
LT
1156 }
1157
1158 /* Update attrtimeo value if we're out of the unstable period */
1159 if (invalid & NFS_INO_INVALID_ATTR) {
91d5b470 1160 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1da177e4 1161 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
3e7d950a 1162 nfsi->attrtimeo_timestamp = now;
6d2b2966
TM
1163 nfsi->last_updated = now;
1164 } else {
1165 if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1166 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1167 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1168 nfsi->attrtimeo_timestamp = now;
1169 }
1da177e4 1170 }
f2115dc9 1171 invalid &= ~NFS_INO_INVALID_ATTR;
1da177e4
LT
1172 /* Don't invalidate the data if we were to blame */
1173 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1174 || S_ISLNK(inode->i_mode)))
1175 invalid &= ~NFS_INO_INVALID_DATA;
412c77ce
TM
1176 if (!nfs_have_delegation(inode, FMODE_READ) ||
1177 (nfsi->cache_validity & NFS_INO_REVAL_FORCED))
55296809 1178 nfsi->cache_validity |= invalid;
412c77ce 1179 nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED;
1da177e4
LT
1180
1181 return 0;
1182 out_changed:
1183 /*
1184 * Big trouble! The inode has become a different object.
1185 */
1da177e4 1186 printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n",
3110ff80 1187 __func__, inode->i_ino, inode->i_mode, fattr->mode);
b37b03b7 1188 out_err:
1da177e4
LT
1189 /*
1190 * No need to worry about unhashing the dentry, as the
1191 * lookup validation will know that the inode is bad.
1192 * (But we fall through to invalidate the caches.)
1193 */
1194 nfs_invalidate_inode(inode);
1da177e4 1195 return -ESTALE;
325cfed9
CL
1196
1197 out_fileid:
1198 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1199 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
54ceac45 1200 NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id,
325cfed9
CL
1201 (long long)nfsi->fileid, (long long)fattr->fileid);
1202 goto out_err;
1da177e4
LT
1203}
1204
55a97593 1205
1da177e4
LT
1206#ifdef CONFIG_NFS_V4
1207
1da177e4
LT
1208/*
1209 * Clean out any remaining NFSv4 state that might be left over due
1210 * to open() calls that passed nfs_atomic_lookup, but failed to call
1211 * nfs_open().
1212 */
f7b422b1 1213void nfs4_clear_inode(struct inode *inode)
1da177e4 1214{
1da177e4 1215 /* If we are holding a delegation, return it! */
e6f81075 1216 nfs_inode_return_delegation_noreclaim(inode);
1da177e4
LT
1217 /* First call standard NFS clear_inode() code */
1218 nfs_clear_inode(inode);
1da177e4 1219}
1da177e4
LT
1220#endif
1221
f7b422b1 1222struct inode *nfs_alloc_inode(struct super_block *sb)
1da177e4
LT
1223{
1224 struct nfs_inode *nfsi;
e94b1766 1225 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1da177e4
LT
1226 if (!nfsi)
1227 return NULL;
55296809
CL
1228 nfsi->flags = 0UL;
1229 nfsi->cache_validity = 0UL;
458818ed
TM
1230#ifdef CONFIG_NFS_V3_ACL
1231 nfsi->acl_access = ERR_PTR(-EAGAIN);
1232 nfsi->acl_default = ERR_PTR(-EAGAIN);
1233#endif
e50a1c2e
BF
1234#ifdef CONFIG_NFS_V4
1235 nfsi->nfs4_acl = NULL;
1236#endif /* CONFIG_NFS_V4 */
1da177e4
LT
1237 return &nfsi->vfs_inode;
1238}
1239
f7b422b1 1240void nfs_destroy_inode(struct inode *inode)
1da177e4
LT
1241{
1242 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1243}
1244
d75d5414
AM
1245static inline void nfs4_init_once(struct nfs_inode *nfsi)
1246{
1247#ifdef CONFIG_NFS_V4
1248 INIT_LIST_HEAD(&nfsi->open_states);
1249 nfsi->delegation = NULL;
1250 nfsi->delegation_state = 0;
1251 init_rwsem(&nfsi->rwsem);
1252#endif
1253}
f7b422b1 1254
51cc5068 1255static void init_once(void *foo)
1da177e4
LT
1256{
1257 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1258
a35afb83 1259 inode_init_once(&nfsi->vfs_inode);
a35afb83
CL
1260 INIT_LIST_HEAD(&nfsi->open_files);
1261 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1262 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1263 INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC);
a35afb83
CL
1264 nfsi->ncommit = 0;
1265 nfsi->npages = 0;
565277f6
TM
1266 atomic_set(&nfsi->silly_count, 1);
1267 INIT_HLIST_HEAD(&nfsi->silly_list);
1268 init_waitqueue_head(&nfsi->waitqueue);
a35afb83 1269 nfs4_init_once(nfsi);
1da177e4 1270}
20c2df83 1271
f7b422b1 1272static int __init nfs_init_inodecache(void)
1da177e4
LT
1273{
1274 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1275 sizeof(struct nfs_inode),
fffb60f9
PJ
1276 0, (SLAB_RECLAIM_ACCOUNT|
1277 SLAB_MEM_SPREAD),
20c2df83 1278 init_once);
1da177e4
LT
1279 if (nfs_inode_cachep == NULL)
1280 return -ENOMEM;
1281
1282 return 0;
1283}
1284
266bee88 1285static void nfs_destroy_inodecache(void)
1da177e4 1286{
1a1d92c1 1287 kmem_cache_destroy(nfs_inode_cachep);
1da177e4
LT
1288}
1289
5746006f
TM
1290struct workqueue_struct *nfsiod_workqueue;
1291
1292/*
1293 * start up the nfsiod workqueue
1294 */
1295static int nfsiod_start(void)
1296{
1297 struct workqueue_struct *wq;
1298 dprintk("RPC: creating workqueue nfsiod\n");
1299 wq = create_singlethread_workqueue("nfsiod");
1300 if (wq == NULL)
1301 return -ENOMEM;
1302 nfsiod_workqueue = wq;
1303 return 0;
1304}
1305
1306/*
1307 * Destroy the nfsiod workqueue
1308 */
1309static void nfsiod_stop(void)
1310{
1311 struct workqueue_struct *wq;
1312
1313 wq = nfsiod_workqueue;
1314 if (wq == NULL)
1315 return;
1316 nfsiod_workqueue = NULL;
1317 destroy_workqueue(wq);
1318}
1319
1da177e4
LT
1320/*
1321 * Initialize NFS
1322 */
1323static int __init init_nfs_fs(void)
1324{
1325 int err;
1326
5746006f
TM
1327 err = nfsiod_start();
1328 if (err)
1329 goto out6;
1330
6aaca566
DH
1331 err = nfs_fs_proc_init();
1332 if (err)
1333 goto out5;
1334
1da177e4
LT
1335 err = nfs_init_nfspagecache();
1336 if (err)
1337 goto out4;
1338
1339 err = nfs_init_inodecache();
1340 if (err)
1341 goto out3;
1342
1343 err = nfs_init_readpagecache();
1344 if (err)
1345 goto out2;
1346
1347 err = nfs_init_writepagecache();
1348 if (err)
1349 goto out1;
1350
1da177e4
LT
1351 err = nfs_init_directcache();
1352 if (err)
1353 goto out0;
1da177e4
LT
1354
1355#ifdef CONFIG_PROC_FS
1356 rpc_proc_register(&nfs_rpcstat);
1357#endif
f7b422b1 1358 if ((err = register_nfs_fs()) != 0)
1da177e4
LT
1359 goto out;
1360 return 0;
1361out:
1362#ifdef CONFIG_PROC_FS
1363 rpc_proc_unregister("nfs");
1364#endif
1da177e4 1365 nfs_destroy_directcache();
6b59a754 1366out0:
6b59a754 1367 nfs_destroy_writepagecache();
1da177e4
LT
1368out1:
1369 nfs_destroy_readpagecache();
1370out2:
1371 nfs_destroy_inodecache();
1372out3:
1373 nfs_destroy_nfspagecache();
1374out4:
6aaca566
DH
1375 nfs_fs_proc_exit();
1376out5:
5746006f
TM
1377 nfsiod_stop();
1378out6:
1da177e4
LT
1379 return err;
1380}
1381
1382static void __exit exit_nfs_fs(void)
1383{
1da177e4 1384 nfs_destroy_directcache();
1da177e4
LT
1385 nfs_destroy_writepagecache();
1386 nfs_destroy_readpagecache();
1387 nfs_destroy_inodecache();
1388 nfs_destroy_nfspagecache();
1389#ifdef CONFIG_PROC_FS
1390 rpc_proc_unregister("nfs");
1391#endif
f7b422b1 1392 unregister_nfs_fs();
6aaca566 1393 nfs_fs_proc_exit();
5746006f 1394 nfsiod_stop();
1da177e4
LT
1395}
1396
1397/* Not quite true; I just maintain it */
1398MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1399MODULE_LICENSE("GPL");
f43bf0be 1400module_param(enable_ino64, bool, 0644);
1da177e4
LT
1401
1402module_init(init_nfs_fs)
1403module_exit(exit_nfs_fs)