?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
nfs: disable data cache revalidation for swapfiles
[linux/fpc-iii.git] / fs / nfs / inode.c
blob1d57fe9f49a9e6180e535c8b71e8040d223ffd2d
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@lxorguk.ukuu.org.uk>, 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
16 #include <linux/module.h>
17 #include <linux/init.h>
18 #include <linux/sched.h>
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>
28 #include <linux/sunrpc/metrics.h>
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/seq_file.h>
34 #include <linux/mount.h>
35 #include <linux/vfs.h>
36 #include <linux/inet.h>
37 #include <linux/nfs_xdr.h>
38 #include <linux/slab.h>
39 #include <linux/compat.h>
40 #include <linux/freezer.h>
41 #include <linux/crc32.h>
42
43 #include <asm/uaccess.h>
44
45 #include "nfs4_fs.h"
46 #include "callback.h"
47 #include "delegation.h"
48 #include "iostat.h"
49 #include "internal.h"
50 #include "fscache.h"
51 #include "dns_resolve.h"
52 #include "pnfs.h"
53 #include "netns.h"
54
55 #define NFSDBG_FACILITY NFSDBG_VFS
56
57 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
58
59 /* Default is to see 64-bit inode numbers */
60 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
61
62 static void nfs_invalidate_inode(struct inode *);
63 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
64
65 static struct kmem_cache * nfs_inode_cachep;
66
67 static inline unsigned long
68 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
69 {
70 return nfs_fileid_to_ino_t(fattr->fileid);
71 }
72
73 /**
74 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
75 * @word: long word containing the bit lock
76 */
77 int nfs_wait_bit_killable(void *word)
78 {
79 if (fatal_signal_pending(current))
80 return -ERESTARTSYS;
81 freezable_schedule();
82 return 0;
83 }
84
85 /**
86 * nfs_compat_user_ino64 - returns the user-visible inode number
87 * @fileid: 64-bit fileid
88 *
89 * This function returns a 32-bit inode number if the boot parameter
90 * nfs.enable_ino64 is zero.
91 */
92 u64 nfs_compat_user_ino64(u64 fileid)
93 {
94 #ifdef CONFIG_COMPAT
95 compat_ulong_t ino;
96 #else
97 unsigned long ino;
98 #endif
99
100 if (enable_ino64)
101 return fileid;
102 ino = fileid;
103 if (sizeof(ino) < sizeof(fileid))
104 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
105 return ino;
106 }
107
108 static void nfs_clear_inode(struct inode *inode)
109 {
110 /*
111 * The following should never happen...
112 */
113 BUG_ON(nfs_have_writebacks(inode));
114 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
115 nfs_zap_acl_cache(inode);
116 nfs_access_zap_cache(inode);
117 nfs_fscache_release_inode_cookie(inode);
118 }
119
120 void nfs_evict_inode(struct inode *inode)
121 {
122 truncate_inode_pages(&inode->i_data, 0);
123 clear_inode(inode);
124 nfs_clear_inode(inode);
125 }
126
127 /**
128 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
129 */
130 int nfs_sync_mapping(struct address_space *mapping)
131 {
132 int ret = 0;
133
134 if (mapping->nrpages != 0) {
135 unmap_mapping_range(mapping, 0, 0, 0);
136 ret = nfs_wb_all(mapping->host);
137 }
138 return ret;
139 }
140
141 /*
142 * Invalidate the local caches
143 */
144 static void nfs_zap_caches_locked(struct inode *inode)
145 {
146 struct nfs_inode *nfsi = NFS_I(inode);
147 int mode = inode->i_mode;
148
149 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
150
151 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
152 nfsi->attrtimeo_timestamp = jiffies;
153
154 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
155 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
156 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
157 else
158 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
159 }
160
161 void nfs_zap_caches(struct inode *inode)
162 {
163 spin_lock(&inode->i_lock);
164 nfs_zap_caches_locked(inode);
165 spin_unlock(&inode->i_lock);
166 }
167
168 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
169 {
170 if (mapping->nrpages != 0) {
171 spin_lock(&inode->i_lock);
172 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
173 spin_unlock(&inode->i_lock);
174 }
175 }
176
177 void nfs_zap_acl_cache(struct inode *inode)
178 {
179 void (*clear_acl_cache)(struct inode *);
180
181 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
182 if (clear_acl_cache != NULL)
183 clear_acl_cache(inode);
184 spin_lock(&inode->i_lock);
185 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
186 spin_unlock(&inode->i_lock);
187 }
188
189 void nfs_invalidate_atime(struct inode *inode)
190 {
191 spin_lock(&inode->i_lock);
192 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
193 spin_unlock(&inode->i_lock);
194 }
195
196 /*
197 * Invalidate, but do not unhash, the inode.
198 * NB: must be called with inode->i_lock held!
199 */
200 static void nfs_invalidate_inode(struct inode *inode)
201 {
202 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
203 nfs_zap_caches_locked(inode);
204 }
205
206 struct nfs_find_desc {
207 struct nfs_fh *fh;
208 struct nfs_fattr *fattr;
209 };
210
211 /*
212 * In NFSv3 we can have 64bit inode numbers. In order to support
213 * this, and re-exported directories (also seen in NFSv2)
214 * we are forced to allow 2 different inodes to have the same
215 * i_ino.
216 */
217 static int
218 nfs_find_actor(struct inode *inode, void *opaque)
219 {
220 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
221 struct nfs_fh *fh = desc->fh;
222 struct nfs_fattr *fattr = desc->fattr;
223
224 if (NFS_FILEID(inode) != fattr->fileid)
225 return 0;
226 if (nfs_compare_fh(NFS_FH(inode), fh))
227 return 0;
228 if (is_bad_inode(inode) || NFS_STALE(inode))
229 return 0;
230 return 1;
231 }
232
233 static int
234 nfs_init_locked(struct inode *inode, void *opaque)
235 {
236 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
237 struct nfs_fattr *fattr = desc->fattr;
238
239 set_nfs_fileid(inode, fattr->fileid);
240 nfs_copy_fh(NFS_FH(inode), desc->fh);
241 return 0;
242 }
243
244 /*
245 * This is our front-end to iget that looks up inodes by file handle
246 * instead of inode number.
247 */
248 struct inode *
249 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
250 {
251 struct nfs_find_desc desc = {
252 .fh = fh,
253 .fattr = fattr
254 };
255 struct inode *inode = ERR_PTR(-ENOENT);
256 unsigned long hash;
257
258 nfs_attr_check_mountpoint(sb, fattr);
259
260 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
261 !nfs_attr_use_mounted_on_fileid(fattr))
262 goto out_no_inode;
263 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
264 goto out_no_inode;
265
266 hash = nfs_fattr_to_ino_t(fattr);
267
268 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
269 if (inode == NULL) {
270 inode = ERR_PTR(-ENOMEM);
271 goto out_no_inode;
272 }
273
274 if (inode->i_state & I_NEW) {
275 struct nfs_inode *nfsi = NFS_I(inode);
276 unsigned long now = jiffies;
277
278 /* We set i_ino for the few things that still rely on it,
279 * such as stat(2) */
280 inode->i_ino = hash;
281
282 /* We can't support update_atime(), since the server will reset it */
283 inode->i_flags |= S_NOATIME|S_NOCMTIME;
284 inode->i_mode = fattr->mode;
285 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
286 && nfs_server_capable(inode, NFS_CAP_MODE))
287 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
288 /* Why so? Because we want revalidate for devices/FIFOs, and
289 * that's precisely what we have in nfs_file_inode_operations.
290 */
291 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
292 if (S_ISREG(inode->i_mode)) {
293 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
294 inode->i_data.a_ops = &nfs_file_aops;
295 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
296 } else if (S_ISDIR(inode->i_mode)) {
297 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
298 inode->i_fop = &nfs_dir_operations;
299 inode->i_data.a_ops = &nfs_dir_aops;
300 /* Deal with crossing mountpoints */
301 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
302 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
303 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
304 inode->i_op = &nfs_referral_inode_operations;
305 else
306 inode->i_op = &nfs_mountpoint_inode_operations;
307 inode->i_fop = NULL;
308 inode->i_flags |= S_AUTOMOUNT;
309 }
310 } else if (S_ISLNK(inode->i_mode))
311 inode->i_op = &nfs_symlink_inode_operations;
312 else
313 init_special_inode(inode, inode->i_mode, fattr->rdev);
314
315 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
316 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
317 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
318 inode->i_version = 0;
319 inode->i_size = 0;
320 clear_nlink(inode);
321 inode->i_uid = -2;
322 inode->i_gid = -2;
323 inode->i_blocks = 0;
324 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
325 nfsi->write_io = 0;
326 nfsi->read_io = 0;
327
328 nfsi->read_cache_jiffies = fattr->time_start;
329 nfsi->attr_gencount = fattr->gencount;
330 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
331 inode->i_atime = fattr->atime;
332 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
333 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
334 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
335 inode->i_mtime = fattr->mtime;
336 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
337 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
338 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
339 inode->i_ctime = fattr->ctime;
340 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
341 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
342 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
343 inode->i_version = fattr->change_attr;
344 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
345 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
346 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
347 inode->i_size = nfs_size_to_loff_t(fattr->size);
348 else
349 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
350 | NFS_INO_REVAL_PAGECACHE;
351 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
352 set_nlink(inode, fattr->nlink);
353 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
354 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
355 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
356 inode->i_uid = fattr->uid;
357 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
358 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
359 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
360 inode->i_gid = fattr->gid;
361 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
362 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
363 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
364 inode->i_blocks = fattr->du.nfs2.blocks;
365 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
366 /*
367 * report the blocks in 512byte units
368 */
369 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
370 }
371 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
372 nfsi->attrtimeo_timestamp = now;
373 nfsi->access_cache = RB_ROOT;
374
375 nfs_fscache_init_inode_cookie(inode);
376
377 unlock_new_inode(inode);
378 } else
379 nfs_refresh_inode(inode, fattr);
380 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
381 inode->i_sb->s_id,
382 (long long)NFS_FILEID(inode),
383 nfs_display_fhandle_hash(fh),
384 atomic_read(&inode->i_count));
385
386 out:
387 return inode;
388
389 out_no_inode:
390 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
391 goto out;
392 }
393
394 #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET|ATTR_FILE|ATTR_OPEN)
395
396 int
397 nfs_setattr(struct dentry *dentry, struct iattr *attr)
398 {
399 struct inode *inode = dentry->d_inode;
400 struct nfs_fattr *fattr;
401 int error = -ENOMEM;
402
403 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
404
405 /* skip mode change if it's just for clearing setuid/setgid */
406 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
407 attr->ia_valid &= ~ATTR_MODE;
408
409 if (attr->ia_valid & ATTR_SIZE) {
410 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
411 attr->ia_valid &= ~ATTR_SIZE;
412 }
413
414 /* Optimization: if the end result is no change, don't RPC */
415 attr->ia_valid &= NFS_VALID_ATTRS;
416 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
417 return 0;
418
419 /* Write all dirty data */
420 if (S_ISREG(inode->i_mode)) {
421 nfs_inode_dio_wait(inode);
422 nfs_wb_all(inode);
423 }
424
425 fattr = nfs_alloc_fattr();
426 if (fattr == NULL)
427 goto out;
428 /*
429 * Return any delegations if we're going to change ACLs
430 */
431 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
432 NFS_PROTO(inode)->return_delegation(inode);
433 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
434 if (error == 0)
435 nfs_refresh_inode(inode, fattr);
436 nfs_free_fattr(fattr);
437 out:
438 return error;
439 }
440
441 /**
442 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
443 * @inode: inode of the file used
444 * @offset: file offset to start truncating
445 *
446 * This is a copy of the common vmtruncate, but with the locking
447 * corrected to take into account the fact that NFS requires
448 * inode->i_size to be updated under the inode->i_lock.
449 */
450 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
451 {
452 loff_t oldsize;
453 int err;
454
455 err = inode_newsize_ok(inode, offset);
456 if (err)
457 goto out;
458
459 spin_lock(&inode->i_lock);
460 oldsize = inode->i_size;
461 i_size_write(inode, offset);
462 spin_unlock(&inode->i_lock);
463
464 truncate_pagecache(inode, oldsize, offset);
465 out:
466 return err;
467 }
468
469 /**
470 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
471 * @inode: pointer to struct inode
472 * @attr: pointer to struct iattr
473 *
474 * Note: we do this in the *proc.c in order to ensure that
475 * it works for things like exclusive creates too.
476 */
477 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
478 {
479 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
480 spin_lock(&inode->i_lock);
481 if ((attr->ia_valid & ATTR_MODE) != 0) {
482 int mode = attr->ia_mode & S_IALLUGO;
483 mode |= inode->i_mode & ~S_IALLUGO;
484 inode->i_mode = mode;
485 }
486 if ((attr->ia_valid & ATTR_UID) != 0)
487 inode->i_uid = attr->ia_uid;
488 if ((attr->ia_valid & ATTR_GID) != 0)
489 inode->i_gid = attr->ia_gid;
490 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
491 spin_unlock(&inode->i_lock);
492 }
493 if ((attr->ia_valid & ATTR_SIZE) != 0) {
494 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
495 nfs_vmtruncate(inode, attr->ia_size);
496 }
497 }
498
499 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
500 {
501 struct inode *inode = dentry->d_inode;
502 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
503 int err;
504
505 /* Flush out writes to the server in order to update c/mtime. */
506 if (S_ISREG(inode->i_mode)) {
507 nfs_inode_dio_wait(inode);
508 err = filemap_write_and_wait(inode->i_mapping);
509 if (err)
510 goto out;
511 }
512
513 /*
514 * We may force a getattr if the user cares about atime.
515 *
516 * Note that we only have to check the vfsmount flags here:
517 * - NFS always sets S_NOATIME by so checking it would give a
518 * bogus result
519 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
520 * no point in checking those.
521 */
522 if ((mnt->mnt_flags & MNT_NOATIME) ||
523 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
524 need_atime = 0;
525
526 if (need_atime)
527 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
528 else
529 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
530 if (!err) {
531 generic_fillattr(inode, stat);
532 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
533 }
534 out:
535 return err;
536 }
537
538 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
539 {
540 atomic_set(&l_ctx->count, 1);
541 l_ctx->lockowner = current->files;
542 l_ctx->pid = current->tgid;
543 INIT_LIST_HEAD(&l_ctx->list);
544 }
545
546 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
547 {
548 struct nfs_lock_context *pos;
549
550 list_for_each_entry(pos, &ctx->lock_context.list, list) {
551 if (pos->lockowner != current->files)
552 continue;
553 if (pos->pid != current->tgid)
554 continue;
555 atomic_inc(&pos->count);
556 return pos;
557 }
558 return NULL;
559 }
560
561 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
562 {
563 struct nfs_lock_context *res, *new = NULL;
564 struct inode *inode = ctx->dentry->d_inode;
565
566 spin_lock(&inode->i_lock);
567 res = __nfs_find_lock_context(ctx);
568 if (res == NULL) {
569 spin_unlock(&inode->i_lock);
570 new = kmalloc(sizeof(*new), GFP_KERNEL);
571 if (new == NULL)
572 return NULL;
573 nfs_init_lock_context(new);
574 spin_lock(&inode->i_lock);
575 res = __nfs_find_lock_context(ctx);
576 if (res == NULL) {
577 list_add_tail(&new->list, &ctx->lock_context.list);
578 new->open_context = ctx;
579 res = new;
580 new = NULL;
581 }
582 }
583 spin_unlock(&inode->i_lock);
584 kfree(new);
585 return res;
586 }
587
588 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
589 {
590 struct nfs_open_context *ctx = l_ctx->open_context;
591 struct inode *inode = ctx->dentry->d_inode;
592
593 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
594 return;
595 list_del(&l_ctx->list);
596 spin_unlock(&inode->i_lock);
597 kfree(l_ctx);
598 }
599
600 /**
601 * nfs_close_context - Common close_context() routine NFSv2/v3
602 * @ctx: pointer to context
603 * @is_sync: is this a synchronous close
604 *
605 * always ensure that the attributes are up to date if we're mounted
606 * with close-to-open semantics
607 */
608 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
609 {
610 struct inode *inode;
611 struct nfs_server *server;
612
613 if (!(ctx->mode & FMODE_WRITE))
614 return;
615 if (!is_sync)
616 return;
617 inode = ctx->dentry->d_inode;
618 if (!list_empty(&NFS_I(inode)->open_files))
619 return;
620 server = NFS_SERVER(inode);
621 if (server->flags & NFS_MOUNT_NOCTO)
622 return;
623 nfs_revalidate_inode(server, inode);
624 }
625
626 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
627 {
628 struct nfs_open_context *ctx;
629 struct rpc_cred *cred = rpc_lookup_cred();
630 if (IS_ERR(cred))
631 return ERR_CAST(cred);
632
633 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
634 if (!ctx) {
635 put_rpccred(cred);
636 return ERR_PTR(-ENOMEM);
637 }
638 nfs_sb_active(dentry->d_sb);
639 ctx->dentry = dget(dentry);
640 ctx->cred = cred;
641 ctx->state = NULL;
642 ctx->mode = f_mode;
643 ctx->flags = 0;
644 ctx->error = 0;
645 nfs_init_lock_context(&ctx->lock_context);
646 ctx->lock_context.open_context = ctx;
647 INIT_LIST_HEAD(&ctx->list);
648 ctx->mdsthreshold = NULL;
649 return ctx;
650 }
651
652 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
653 {
654 if (ctx != NULL)
655 atomic_inc(&ctx->lock_context.count);
656 return ctx;
657 }
658
659 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
660 {
661 struct inode *inode = ctx->dentry->d_inode;
662 struct super_block *sb = ctx->dentry->d_sb;
663
664 if (!list_empty(&ctx->list)) {
665 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
666 return;
667 list_del(&ctx->list);
668 spin_unlock(&inode->i_lock);
669 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
670 return;
671 if (inode != NULL)
672 NFS_PROTO(inode)->close_context(ctx, is_sync);
673 if (ctx->cred != NULL)
674 put_rpccred(ctx->cred);
675 dput(ctx->dentry);
676 nfs_sb_deactive(sb);
677 kfree(ctx->mdsthreshold);
678 kfree(ctx);
679 }
680
681 void put_nfs_open_context(struct nfs_open_context *ctx)
682 {
683 __put_nfs_open_context(ctx, 0);
684 }
685
686 /*
687 * Ensure that mmap has a recent RPC credential for use when writing out
688 * shared pages
689 */
690 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
691 {
692 struct inode *inode = filp->f_path.dentry->d_inode;
693 struct nfs_inode *nfsi = NFS_I(inode);
694
695 filp->private_data = get_nfs_open_context(ctx);
696 spin_lock(&inode->i_lock);
697 list_add(&ctx->list, &nfsi->open_files);
698 spin_unlock(&inode->i_lock);
699 }
700
701 /*
702 * Given an inode, search for an open context with the desired characteristics
703 */
704 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
705 {
706 struct nfs_inode *nfsi = NFS_I(inode);
707 struct nfs_open_context *pos, *ctx = NULL;
708
709 spin_lock(&inode->i_lock);
710 list_for_each_entry(pos, &nfsi->open_files, list) {
711 if (cred != NULL && pos->cred != cred)
712 continue;
713 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
714 continue;
715 ctx = get_nfs_open_context(pos);
716 break;
717 }
718 spin_unlock(&inode->i_lock);
719 return ctx;
720 }
721
722 static void nfs_file_clear_open_context(struct file *filp)
723 {
724 struct inode *inode = filp->f_path.dentry->d_inode;
725 struct nfs_open_context *ctx = nfs_file_open_context(filp);
726
727 if (ctx) {
728 filp->private_data = NULL;
729 spin_lock(&inode->i_lock);
730 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
731 spin_unlock(&inode->i_lock);
732 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
733 }
734 }
735
736 /*
737 * These allocate and release file read/write context information.
738 */
739 int nfs_open(struct inode *inode, struct file *filp)
740 {
741 struct nfs_open_context *ctx;
742
743 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
744 if (IS_ERR(ctx))
745 return PTR_ERR(ctx);
746 nfs_file_set_open_context(filp, ctx);
747 put_nfs_open_context(ctx);
748 nfs_fscache_set_inode_cookie(inode, filp);
749 return 0;
750 }
751
752 int nfs_release(struct inode *inode, struct file *filp)
753 {
754 nfs_file_clear_open_context(filp);
755 return 0;
756 }
757
758 /*
759 * This function is called whenever some part of NFS notices that
760 * the cached attributes have to be refreshed.
761 */
762 int
763 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
764 {
765 int status = -ESTALE;
766 struct nfs_fattr *fattr = NULL;
767 struct nfs_inode *nfsi = NFS_I(inode);
768
769 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
770 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
771
772 if (is_bad_inode(inode))
773 goto out;
774 if (NFS_STALE(inode))
775 goto out;
776
777 status = -ENOMEM;
778 fattr = nfs_alloc_fattr();
779 if (fattr == NULL)
780 goto out;
781
782 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
783 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
784 if (status != 0) {
785 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
786 inode->i_sb->s_id,
787 (long long)NFS_FILEID(inode), status);
788 if (status == -ESTALE) {
789 nfs_zap_caches(inode);
790 if (!S_ISDIR(inode->i_mode))
791 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
792 }
793 goto out;
794 }
795
796 status = nfs_refresh_inode(inode, fattr);
797 if (status) {
798 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
799 inode->i_sb->s_id,
800 (long long)NFS_FILEID(inode), status);
801 goto out;
802 }
803
804 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
805 nfs_zap_acl_cache(inode);
806
807 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
808 inode->i_sb->s_id,
809 (long long)NFS_FILEID(inode));
810
811 out:
812 nfs_free_fattr(fattr);
813 return status;
814 }
815
816 int nfs_attribute_timeout(struct inode *inode)
817 {
818 struct nfs_inode *nfsi = NFS_I(inode);
819
820 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
821 }
822
823 static int nfs_attribute_cache_expired(struct inode *inode)
824 {
825 if (nfs_have_delegated_attributes(inode))
826 return 0;
827 return nfs_attribute_timeout(inode);
828 }
829
830 /**
831 * nfs_revalidate_inode - Revalidate the inode attributes
832 * @server - pointer to nfs_server struct
833 * @inode - pointer to inode struct
834 *
835 * Updates inode attribute information by retrieving the data from the server.
836 */
837 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
838 {
839 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
840 && !nfs_attribute_cache_expired(inode))
841 return NFS_STALE(inode) ? -ESTALE : 0;
842 return __nfs_revalidate_inode(server, inode);
843 }
844
845 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
846 {
847 struct nfs_inode *nfsi = NFS_I(inode);
848
849 if (mapping->nrpages != 0) {
850 int ret = invalidate_inode_pages2(mapping);
851 if (ret < 0)
852 return ret;
853 }
854 spin_lock(&inode->i_lock);
855 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
856 if (S_ISDIR(inode->i_mode))
857 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
858 spin_unlock(&inode->i_lock);
859 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
860 nfs_fscache_reset_inode_cookie(inode);
861 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
862 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
863 return 0;
864 }
865
866 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
867 {
868 if (nfs_have_delegated_attributes(inode))
869 return false;
870 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
871 || nfs_attribute_timeout(inode)
872 || NFS_STALE(inode);
873 }
874
875 /**
876 * nfs_revalidate_mapping - Revalidate the pagecache
877 * @inode - pointer to host inode
878 * @mapping - pointer to mapping
879 */
880 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
881 {
882 struct nfs_inode *nfsi = NFS_I(inode);
883 int ret = 0;
884
885 /* swapfiles are not supposed to be shared. */
886 if (IS_SWAPFILE(inode))
887 goto out;
888
889 if (nfs_mapping_need_revalidate_inode(inode)) {
890 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
891 if (ret < 0)
892 goto out;
893 }
894 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
895 ret = nfs_invalidate_mapping(inode, mapping);
896 out:
897 return ret;
898 }
899
900 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
901 {
902 struct nfs_inode *nfsi = NFS_I(inode);
903 unsigned long ret = 0;
904
905 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
906 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
907 && inode->i_version == fattr->pre_change_attr) {
908 inode->i_version = fattr->change_attr;
909 if (S_ISDIR(inode->i_mode))
910 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
911 ret |= NFS_INO_INVALID_ATTR;
912 }
913 /* If we have atomic WCC data, we may update some attributes */
914 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
915 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
916 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
917 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
918 ret |= NFS_INO_INVALID_ATTR;
919 }
920
921 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
922 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
923 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
924 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
925 if (S_ISDIR(inode->i_mode))
926 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
927 ret |= NFS_INO_INVALID_ATTR;
928 }
929 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
930 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
931 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
932 && nfsi->npages == 0) {
933 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
934 ret |= NFS_INO_INVALID_ATTR;
935 }
936 return ret;
937 }
938
939 /**
940 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
941 * @inode - pointer to inode
942 * @fattr - updated attributes
943 *
944 * Verifies the attribute cache. If we have just changed the attributes,
945 * so that fattr carries weak cache consistency data, then it may
946 * also update the ctime/mtime/change_attribute.
947 */
948 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
949 {
950 struct nfs_inode *nfsi = NFS_I(inode);
951 loff_t cur_size, new_isize;
952 unsigned long invalid = 0;
953
954
955 if (nfs_have_delegated_attributes(inode))
956 return 0;
957 /* Has the inode gone and changed behind our back? */
958 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
959 return -EIO;
960 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
961 return -EIO;
962
963 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
964 inode->i_version != fattr->change_attr)
965 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
966
967 /* Verify a few of the more important attributes */
968 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
969 invalid |= NFS_INO_INVALID_ATTR;
970
971 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
972 cur_size = i_size_read(inode);
973 new_isize = nfs_size_to_loff_t(fattr->size);
974 if (cur_size != new_isize && nfsi->npages == 0)
975 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
976 }
977
978 /* Have any file permissions changed? */
979 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
980 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
981 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
982 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
983 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
984 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
985
986 /* Has the link count changed? */
987 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
988 invalid |= NFS_INO_INVALID_ATTR;
989
990 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
991 invalid |= NFS_INO_INVALID_ATIME;
992
993 if (invalid != 0)
994 nfsi->cache_validity |= invalid;
995
996 nfsi->read_cache_jiffies = fattr->time_start;
997 return 0;
998 }
999
1000 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1001 {
1002 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
1003 return 0;
1004 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
1005 }
1006
1007 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1008 {
1009 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1010 return 0;
1011 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1012 }
1013
1014 static atomic_long_t nfs_attr_generation_counter;
1015
1016 static unsigned long nfs_read_attr_generation_counter(void)
1017 {
1018 return atomic_long_read(&nfs_attr_generation_counter);
1019 }
1020
1021 unsigned long nfs_inc_attr_generation_counter(void)
1022 {
1023 return atomic_long_inc_return(&nfs_attr_generation_counter);
1024 }
1025
1026 void nfs_fattr_init(struct nfs_fattr *fattr)
1027 {
1028 fattr->valid = 0;
1029 fattr->time_start = jiffies;
1030 fattr->gencount = nfs_inc_attr_generation_counter();
1031 fattr->owner_name = NULL;
1032 fattr->group_name = NULL;
1033 }
1034
1035 struct nfs_fattr *nfs_alloc_fattr(void)
1036 {
1037 struct nfs_fattr *fattr;
1038
1039 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1040 if (fattr != NULL)
1041 nfs_fattr_init(fattr);
1042 return fattr;
1043 }
1044
1045 struct nfs_fh *nfs_alloc_fhandle(void)
1046 {
1047 struct nfs_fh *fh;
1048
1049 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1050 if (fh != NULL)
1051 fh->size = 0;
1052 return fh;
1053 }
1054
1055 #ifdef NFS_DEBUG
1056 /*
1057 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1058 * in the same way that wireshark does
1059 *
1060 * @fh: file handle
1061 *
1062 * For debugging only.
1063 */
1064 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1065 {
1066 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1067 * not on the result */
1068 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size);
1069 }
1070
1071 /*
1072 * _nfs_display_fhandle - display an NFS file handle on the console
1073 *
1074 * @fh: file handle to display
1075 * @caption: display caption
1076 *
1077 * For debugging only.
1078 */
1079 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1080 {
1081 unsigned short i;
1082
1083 if (fh == NULL || fh->size == 0) {
1084 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1085 return;
1086 }
1087
1088 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1089 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1090 for (i = 0; i < fh->size; i += 16) {
1091 __be32 *pos = (__be32 *)&fh->data[i];
1092
1093 switch ((fh->size - i - 1) >> 2) {
1094 case 0:
1095 printk(KERN_DEFAULT " %08x\n",
1096 be32_to_cpup(pos));
1097 break;
1098 case 1:
1099 printk(KERN_DEFAULT " %08x %08x\n",
1100 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1101 break;
1102 case 2:
1103 printk(KERN_DEFAULT " %08x %08x %08x\n",
1104 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1105 be32_to_cpup(pos + 2));
1106 break;
1107 default:
1108 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1109 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1110 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1111 }
1112 }
1113 }
1114 #endif
1115
1116 /**
1117 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1118 * @inode - pointer to inode
1119 * @fattr - attributes
1120 *
1121 * Attempt to divine whether or not an RPC call reply carrying stale
1122 * attributes got scheduled after another call carrying updated ones.
1123 *
1124 * To do so, the function first assumes that a more recent ctime means
1125 * that the attributes in fattr are newer, however it also attempt to
1126 * catch the case where ctime either didn't change, or went backwards
1127 * (if someone reset the clock on the server) by looking at whether
1128 * or not this RPC call was started after the inode was last updated.
1129 * Note also the check for wraparound of 'attr_gencount'
1130 *
1131 * The function returns 'true' if it thinks the attributes in 'fattr' are
1132 * more recent than the ones cached in the inode.
1133 *
1134 */
1135 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1136 {
1137 const struct nfs_inode *nfsi = NFS_I(inode);
1138
1139 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1140 nfs_ctime_need_update(inode, fattr) ||
1141 nfs_size_need_update(inode, fattr) ||
1142 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1143 }
1144
1145 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1146 {
1147 if (nfs_inode_attrs_need_update(inode, fattr))
1148 return nfs_update_inode(inode, fattr);
1149 return nfs_check_inode_attributes(inode, fattr);
1150 }
1151
1152 /**
1153 * nfs_refresh_inode - try to update the inode attribute cache
1154 * @inode - pointer to inode
1155 * @fattr - updated attributes
1156 *
1157 * Check that an RPC call that returned attributes has not overlapped with
1158 * other recent updates of the inode metadata, then decide whether it is
1159 * safe to do a full update of the inode attributes, or whether just to
1160 * call nfs_check_inode_attributes.
1161 */
1162 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1163 {
1164 int status;
1165
1166 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1167 return 0;
1168 spin_lock(&inode->i_lock);
1169 status = nfs_refresh_inode_locked(inode, fattr);
1170 spin_unlock(&inode->i_lock);
1171
1172 return status;
1173 }
1174
1175 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1176 {
1177 struct nfs_inode *nfsi = NFS_I(inode);
1178
1179 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1180 if (S_ISDIR(inode->i_mode))
1181 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1182 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1183 return 0;
1184 return nfs_refresh_inode_locked(inode, fattr);
1185 }
1186
1187 /**
1188 * nfs_post_op_update_inode - try to update the inode attribute cache
1189 * @inode - pointer to inode
1190 * @fattr - updated attributes
1191 *
1192 * After an operation that has changed the inode metadata, mark the
1193 * attribute cache as being invalid, then try to update it.
1194 *
1195 * NB: if the server didn't return any post op attributes, this
1196 * function will force the retrieval of attributes before the next
1197 * NFS request. Thus it should be used only for operations that
1198 * are expected to change one or more attributes, to avoid
1199 * unnecessary NFS requests and trips through nfs_update_inode().
1200 */
1201 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1202 {
1203 int status;
1204
1205 spin_lock(&inode->i_lock);
1206 status = nfs_post_op_update_inode_locked(inode, fattr);
1207 spin_unlock(&inode->i_lock);
1208 return status;
1209 }
1210
1211 /**
1212 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1213 * @inode - pointer to inode
1214 * @fattr - updated attributes
1215 *
1216 * After an operation that has changed the inode metadata, mark the
1217 * attribute cache as being invalid, then try to update it. Fake up
1218 * weak cache consistency data, if none exist.
1219 *
1220 * This function is mainly designed to be used by the ->write_done() functions.
1221 */
1222 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1223 {
1224 int status;
1225
1226 spin_lock(&inode->i_lock);
1227 /* Don't do a WCC update if these attributes are already stale */
1228 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1229 !nfs_inode_attrs_need_update(inode, fattr)) {
1230 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1231 | NFS_ATTR_FATTR_PRESIZE
1232 | NFS_ATTR_FATTR_PREMTIME
1233 | NFS_ATTR_FATTR_PRECTIME);
1234 goto out_noforce;
1235 }
1236 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1237 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1238 fattr->pre_change_attr = inode->i_version;
1239 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1240 }
1241 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1242 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1243 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1244 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1245 }
1246 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1247 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1248 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1249 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1250 }
1251 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1252 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1253 fattr->pre_size = i_size_read(inode);
1254 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1255 }
1256 out_noforce:
1257 status = nfs_post_op_update_inode_locked(inode, fattr);
1258 spin_unlock(&inode->i_lock);
1259 return status;
1260 }
1261
1262 /*
1263 * Many nfs protocol calls return the new file attributes after
1264 * an operation. Here we update the inode to reflect the state
1265 * of the server's inode.
1266 *
1267 * This is a bit tricky because we have to make sure all dirty pages
1268 * have been sent off to the server before calling invalidate_inode_pages.
1269 * To make sure no other process adds more write requests while we try
1270 * our best to flush them, we make them sleep during the attribute refresh.
1271 *
1272 * A very similar scenario holds for the dir cache.
1273 */
1274 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1275 {
1276 struct nfs_server *server;
1277 struct nfs_inode *nfsi = NFS_I(inode);
1278 loff_t cur_isize, new_isize;
1279 unsigned long invalid = 0;
1280 unsigned long now = jiffies;
1281 unsigned long save_cache_validity;
1282
1283 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1284 __func__, inode->i_sb->s_id, inode->i_ino,
1285 nfs_display_fhandle_hash(NFS_FH(inode)),
1286 atomic_read(&inode->i_count), fattr->valid);
1287
1288 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1289 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1290 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1291 NFS_SERVER(inode)->nfs_client->cl_hostname,
1292 inode->i_sb->s_id, (long long)nfsi->fileid,
1293 (long long)fattr->fileid);
1294 goto out_err;
1295 }
1296
1297 /*
1298 * Make sure the inode's type hasn't changed.
1299 */
1300 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1301 /*
1302 * Big trouble! The inode has become a different object.
1303 */
1304 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1305 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1306 goto out_err;
1307 }
1308
1309 server = NFS_SERVER(inode);
1310 /* Update the fsid? */
1311 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1312 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1313 !IS_AUTOMOUNT(inode))
1314 server->fsid = fattr->fsid;
1315
1316 /*
1317 * Update the read time so we don't revalidate too often.
1318 */
1319 nfsi->read_cache_jiffies = fattr->time_start;
1320
1321 save_cache_validity = nfsi->cache_validity;
1322 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1323 | NFS_INO_INVALID_ATIME
1324 | NFS_INO_REVAL_FORCED
1325 | NFS_INO_REVAL_PAGECACHE);
1326
1327 /* Do atomic weak cache consistency updates */
1328 invalid |= nfs_wcc_update_inode(inode, fattr);
1329
1330 /* More cache consistency checks */
1331 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1332 if (inode->i_version != fattr->change_attr) {
1333 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1334 inode->i_sb->s_id, inode->i_ino);
1335 invalid |= NFS_INO_INVALID_ATTR
1336 | NFS_INO_INVALID_DATA
1337 | NFS_INO_INVALID_ACCESS
1338 | NFS_INO_INVALID_ACL
1339 | NFS_INO_REVAL_PAGECACHE;
1340 if (S_ISDIR(inode->i_mode))
1341 nfs_force_lookup_revalidate(inode);
1342 inode->i_version = fattr->change_attr;
1343 }
1344 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1345 invalid |= save_cache_validity;
1346
1347 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1348 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1349 } else if (server->caps & NFS_CAP_MTIME)
1350 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1351 | NFS_INO_REVAL_FORCED);
1352
1353 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1354 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1355 } else if (server->caps & NFS_CAP_CTIME)
1356 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1357 | NFS_INO_REVAL_FORCED);
1358
1359 /* Check if our cached file size is stale */
1360 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1361 new_isize = nfs_size_to_loff_t(fattr->size);
1362 cur_isize = i_size_read(inode);
1363 if (new_isize != cur_isize) {
1364 /* Do we perhaps have any outstanding writes, or has
1365 * the file grown beyond our last write? */
1366 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1367 new_isize > cur_isize) {
1368 i_size_write(inode, new_isize);
1369 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1370 }
1371 dprintk("NFS: isize change on server for file %s/%ld "
1372 "(%Ld to %Ld)\n",
1373 inode->i_sb->s_id,
1374 inode->i_ino,
1375 (long long)cur_isize,
1376 (long long)new_isize);
1377 }
1378 } else
1379 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1380 | NFS_INO_REVAL_PAGECACHE
1381 | NFS_INO_REVAL_FORCED);
1382
1383
1384 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1385 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1386 else if (server->caps & NFS_CAP_ATIME)
1387 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1388 | NFS_INO_REVAL_FORCED);
1389
1390 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1391 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1392 umode_t newmode = inode->i_mode & S_IFMT;
1393 newmode |= fattr->mode & S_IALLUGO;
1394 inode->i_mode = newmode;
1395 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1396 }
1397 } else if (server->caps & NFS_CAP_MODE)
1398 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1399 | NFS_INO_INVALID_ACCESS
1400 | NFS_INO_INVALID_ACL
1401 | NFS_INO_REVAL_FORCED);
1402
1403 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1404 if (inode->i_uid != fattr->uid) {
1405 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1406 inode->i_uid = fattr->uid;
1407 }
1408 } else if (server->caps & NFS_CAP_OWNER)
1409 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1410 | NFS_INO_INVALID_ACCESS
1411 | NFS_INO_INVALID_ACL
1412 | NFS_INO_REVAL_FORCED);
1413
1414 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1415 if (inode->i_gid != fattr->gid) {
1416 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1417 inode->i_gid = fattr->gid;
1418 }
1419 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1420 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1421 | NFS_INO_INVALID_ACCESS
1422 | NFS_INO_INVALID_ACL
1423 | NFS_INO_REVAL_FORCED);
1424
1425 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1426 if (inode->i_nlink != fattr->nlink) {
1427 invalid |= NFS_INO_INVALID_ATTR;
1428 if (S_ISDIR(inode->i_mode))
1429 invalid |= NFS_INO_INVALID_DATA;
1430 set_nlink(inode, fattr->nlink);
1431 }
1432 } else if (server->caps & NFS_CAP_NLINK)
1433 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1434 | NFS_INO_REVAL_FORCED);
1435
1436 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1437 /*
1438 * report the blocks in 512byte units
1439 */
1440 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1441 }
1442 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1443 inode->i_blocks = fattr->du.nfs2.blocks;
1444
1445 /* Update attrtimeo value if we're out of the unstable period */
1446 if (invalid & NFS_INO_INVALID_ATTR) {
1447 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1448 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1449 nfsi->attrtimeo_timestamp = now;
1450 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1451 } else {
1452 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1453 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1454 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1455 nfsi->attrtimeo_timestamp = now;
1456 }
1457 }
1458 invalid &= ~NFS_INO_INVALID_ATTR;
1459 /* Don't invalidate the data if we were to blame */
1460 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1461 || S_ISLNK(inode->i_mode)))
1462 invalid &= ~NFS_INO_INVALID_DATA;
1463 if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ) ||
1464 (save_cache_validity & NFS_INO_REVAL_FORCED))
1465 nfsi->cache_validity |= invalid;
1466
1467 return 0;
1468 out_err:
1469 /*
1470 * No need to worry about unhashing the dentry, as the
1471 * lookup validation will know that the inode is bad.
1472 * (But we fall through to invalidate the caches.)
1473 */
1474 nfs_invalidate_inode(inode);
1475 return -ESTALE;
1476 }
1477
1478
1479 #ifdef CONFIG_NFS_V4
1480
1481 /*
1482 * Clean out any remaining NFSv4 state that might be left over due
1483 * to open() calls that passed nfs_atomic_lookup, but failed to call
1484 * nfs_open().
1485 */
1486 void nfs4_evict_inode(struct inode *inode)
1487 {
1488 truncate_inode_pages(&inode->i_data, 0);
1489 clear_inode(inode);
1490 pnfs_return_layout(inode);
1491 pnfs_destroy_layout(NFS_I(inode));
1492 /* If we are holding a delegation, return it! */
1493 nfs_inode_return_delegation_noreclaim(inode);
1494 /* First call standard NFS clear_inode() code */
1495 nfs_clear_inode(inode);
1496 }
1497 #endif
1498
1499 struct inode *nfs_alloc_inode(struct super_block *sb)
1500 {
1501 struct nfs_inode *nfsi;
1502 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1503 if (!nfsi)
1504 return NULL;
1505 nfsi->flags = 0UL;
1506 nfsi->cache_validity = 0UL;
1507 #ifdef CONFIG_NFS_V3_ACL
1508 nfsi->acl_access = ERR_PTR(-EAGAIN);
1509 nfsi->acl_default = ERR_PTR(-EAGAIN);
1510 #endif
1511 #ifdef CONFIG_NFS_V4
1512 nfsi->nfs4_acl = NULL;
1513 #endif /* CONFIG_NFS_V4 */
1514 return &nfsi->vfs_inode;
1515 }
1516
1517 static void nfs_i_callback(struct rcu_head *head)
1518 {
1519 struct inode *inode = container_of(head, struct inode, i_rcu);
1520 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1521 }
1522
1523 void nfs_destroy_inode(struct inode *inode)
1524 {
1525 call_rcu(&inode->i_rcu, nfs_i_callback);
1526 }
1527
1528 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1529 {
1530 #ifdef CONFIG_NFS_V4
1531 INIT_LIST_HEAD(&nfsi->open_states);
1532 nfsi->delegation = NULL;
1533 nfsi->delegation_state = 0;
1534 init_rwsem(&nfsi->rwsem);
1535 nfsi->layout = NULL;
1536 #endif
1537 }
1538
1539 static void init_once(void *foo)
1540 {
1541 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1542
1543 inode_init_once(&nfsi->vfs_inode);
1544 INIT_LIST_HEAD(&nfsi->open_files);
1545 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1546 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1547 INIT_LIST_HEAD(&nfsi->commit_info.list);
1548 nfsi->npages = 0;
1549 nfsi->commit_info.ncommit = 0;
1550 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1551 atomic_set(&nfsi->silly_count, 1);
1552 INIT_HLIST_HEAD(&nfsi->silly_list);
1553 init_waitqueue_head(&nfsi->waitqueue);
1554 nfs4_init_once(nfsi);
1555 }
1556
1557 static int __init nfs_init_inodecache(void)
1558 {
1559 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1560 sizeof(struct nfs_inode),
1561 0, (SLAB_RECLAIM_ACCOUNT|
1562 SLAB_MEM_SPREAD),
1563 init_once);
1564 if (nfs_inode_cachep == NULL)
1565 return -ENOMEM;
1566
1567 return 0;
1568 }
1569
1570 static void nfs_destroy_inodecache(void)
1571 {
1572 kmem_cache_destroy(nfs_inode_cachep);
1573 }
1574
1575 struct workqueue_struct *nfsiod_workqueue;
1576
1577 /*
1578 * start up the nfsiod workqueue
1579 */
1580 static int nfsiod_start(void)
1581 {
1582 struct workqueue_struct *wq;
1583 dprintk("RPC: creating workqueue nfsiod\n");
1584 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1585 if (wq == NULL)
1586 return -ENOMEM;
1587 nfsiod_workqueue = wq;
1588 return 0;
1589 }
1590
1591 /*
1592 * Destroy the nfsiod workqueue
1593 */
1594 static void nfsiod_stop(void)
1595 {
1596 struct workqueue_struct *wq;
1597
1598 wq = nfsiod_workqueue;
1599 if (wq == NULL)
1600 return;
1601 nfsiod_workqueue = NULL;
1602 destroy_workqueue(wq);
1603 }
1604
1605 int nfs_net_id;
1606 EXPORT_SYMBOL_GPL(nfs_net_id);
1607
1608 static int nfs_net_init(struct net *net)
1609 {
1610 nfs_clients_init(net);
1611 return nfs_dns_resolver_cache_init(net);
1612 }
1613
1614 static void nfs_net_exit(struct net *net)
1615 {
1616 nfs_dns_resolver_cache_destroy(net);
1617 nfs_cleanup_cb_ident_idr(net);
1618 }
1619
1620 static struct pernet_operations nfs_net_ops = {
1621 .init = nfs_net_init,
1622 .exit = nfs_net_exit,
1623 .id = &nfs_net_id,
1624 .size = sizeof(struct nfs_net),
1625 };
1626
1627 /*
1628 * Initialize NFS
1629 */
1630 static int __init init_nfs_fs(void)
1631 {
1632 int err;
1633
1634 err = nfs_dns_resolver_init();
1635 if (err < 0)
1636 goto out11;
1637
1638 err = register_pernet_subsys(&nfs_net_ops);
1639 if (err < 0)
1640 goto out10;
1641
1642 err = nfs_fscache_register();
1643 if (err < 0)
1644 goto out9;
1645
1646 err = nfsiod_start();
1647 if (err)
1648 goto out8;
1649
1650 err = nfs_fs_proc_init();
1651 if (err)
1652 goto out7;
1653
1654 err = nfs_init_nfspagecache();
1655 if (err)
1656 goto out6;
1657
1658 err = nfs_init_inodecache();
1659 if (err)
1660 goto out5;
1661
1662 err = nfs_init_readpagecache();
1663 if (err)
1664 goto out4;
1665
1666 err = nfs_init_writepagecache();
1667 if (err)
1668 goto out3;
1669
1670 err = nfs_init_directcache();
1671 if (err)
1672 goto out2;
1673
1674 #ifdef CONFIG_PROC_FS
1675 rpc_proc_register(&init_net, &nfs_rpcstat);
1676 #endif
1677
1678 #ifdef CONFIG_NFS_V4
1679 err = init_nfs_v4();
1680 if (err)
1681 goto out1;
1682 #endif
1683
1684 if ((err = register_nfs_fs()) != 0)
1685 goto out0;
1686
1687 return 0;
1688 out0:
1689 #ifdef CONFIG_NFS_V4
1690 exit_nfs_v4();
1691 out1:
1692 #endif
1693 #ifdef CONFIG_PROC_FS
1694 rpc_proc_unregister(&init_net, "nfs");
1695 #endif
1696 nfs_destroy_directcache();
1697 out2:
1698 nfs_destroy_writepagecache();
1699 out3:
1700 nfs_destroy_readpagecache();
1701 out4:
1702 nfs_destroy_inodecache();
1703 out5:
1704 nfs_destroy_nfspagecache();
1705 out6:
1706 nfs_fs_proc_exit();
1707 out7:
1708 nfsiod_stop();
1709 out8:
1710 nfs_fscache_unregister();
1711 out9:
1712 unregister_pernet_subsys(&nfs_net_ops);
1713 out10:
1714 nfs_dns_resolver_destroy();
1715 out11:
1716 return err;
1717 }
1718
1719 static void __exit exit_nfs_fs(void)
1720 {
1721 #ifdef CONFIG_NFS_V4
1722 exit_nfs_v4();
1723 #endif
1724 nfs_destroy_directcache();
1725 nfs_destroy_writepagecache();
1726 nfs_destroy_readpagecache();
1727 nfs_destroy_inodecache();
1728 nfs_destroy_nfspagecache();
1729 nfs_fscache_unregister();
1730 unregister_pernet_subsys(&nfs_net_ops);
1731 nfs_dns_resolver_destroy();
1732 #ifdef CONFIG_PROC_FS
1733 rpc_proc_unregister(&init_net, "nfs");
1734 #endif
1735 unregister_nfs_fs();
1736 nfs_fs_proc_exit();
1737 nfsiod_stop();
1738 }
1739
1740 /* Not quite true; I just maintain it */
1741 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1742 MODULE_LICENSE("GPL");
1743 module_param(enable_ino64, bool, 0644);
1744
1745 module_init(init_nfs_fs)
1746 module_exit(exit_nfs_fs)
Linux with added FPC-III support