Merge tag 'nfs-for-3.5-1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / nfs / inode.c
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/nfs_idmap.h>
36 #include <linux/vfs.h>
37 #include <linux/inet.h>
38 #include <linux/nfs_xdr.h>
39 #include <linux/slab.h>
40 #include <linux/compat.h>
41 #include <linux/freezer.h>
42 #include <linux/crc32.h>
43
44 #include <asm/uaccess.h>
45
46 #include "nfs4_fs.h"
47 #include "callback.h"
48 #include "delegation.h"
49 #include "iostat.h"
50 #include "internal.h"
51 #include "fscache.h"
52 #include "dns_resolve.h"
53 #include "pnfs.h"
54 #include "netns.h"
55
56 #define NFSDBG_FACILITY NFSDBG_VFS
57
58 #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1
59
60 /* Default is to see 64-bit inode numbers */
61 static bool enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED;
62
63 static void nfs_invalidate_inode(struct inode *);
64 static int nfs_update_inode(struct inode *, struct nfs_fattr *);
65
66 static struct kmem_cache * nfs_inode_cachep;
67
68 static inline unsigned long
69 nfs_fattr_to_ino_t(struct nfs_fattr *fattr)
70 {
71 return nfs_fileid_to_ino_t(fattr->fileid);
72 }
73
74 /**
75 * nfs_wait_bit_killable - helper for functions that are sleeping on bit locks
76 * @word: long word containing the bit lock
77 */
78 int nfs_wait_bit_killable(void *word)
79 {
80 if (fatal_signal_pending(current))
81 return -ERESTARTSYS;
82 freezable_schedule();
83 return 0;
84 }
85
86 /**
87 * nfs_compat_user_ino64 - returns the user-visible inode number
88 * @fileid: 64-bit fileid
89 *
90 * This function returns a 32-bit inode number if the boot parameter
91 * nfs.enable_ino64 is zero.
92 */
93 u64 nfs_compat_user_ino64(u64 fileid)
94 {
95 #ifdef CONFIG_COMPAT
96 compat_ulong_t ino;
97 #else
98 unsigned long ino;
99 #endif
100
101 if (enable_ino64)
102 return fileid;
103 ino = fileid;
104 if (sizeof(ino) < sizeof(fileid))
105 ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8;
106 return ino;
107 }
108
109 static void nfs_clear_inode(struct inode *inode)
110 {
111 /*
112 * The following should never happen...
113 */
114 BUG_ON(nfs_have_writebacks(inode));
115 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
116 nfs_zap_acl_cache(inode);
117 nfs_access_zap_cache(inode);
118 nfs_fscache_release_inode_cookie(inode);
119 }
120
121 void nfs_evict_inode(struct inode *inode)
122 {
123 truncate_inode_pages(&inode->i_data, 0);
124 clear_inode(inode);
125 nfs_clear_inode(inode);
126 }
127
128 /**
129 * nfs_sync_mapping - helper to flush all mmapped dirty data to disk
130 */
131 int nfs_sync_mapping(struct address_space *mapping)
132 {
133 int ret = 0;
134
135 if (mapping->nrpages != 0) {
136 unmap_mapping_range(mapping, 0, 0, 0);
137 ret = nfs_wb_all(mapping->host);
138 }
139 return ret;
140 }
141
142 /*
143 * Invalidate the local caches
144 */
145 static void nfs_zap_caches_locked(struct inode *inode)
146 {
147 struct nfs_inode *nfsi = NFS_I(inode);
148 int mode = inode->i_mode;
149
150 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
151
152 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
153 nfsi->attrtimeo_timestamp = jiffies;
154
155 memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode)));
156 if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))
157 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
158 else
159 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE;
160 }
161
162 void nfs_zap_caches(struct inode *inode)
163 {
164 spin_lock(&inode->i_lock);
165 nfs_zap_caches_locked(inode);
166 spin_unlock(&inode->i_lock);
167 }
168
169 void nfs_zap_mapping(struct inode *inode, struct address_space *mapping)
170 {
171 if (mapping->nrpages != 0) {
172 spin_lock(&inode->i_lock);
173 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
174 spin_unlock(&inode->i_lock);
175 }
176 }
177
178 void nfs_zap_acl_cache(struct inode *inode)
179 {
180 void (*clear_acl_cache)(struct inode *);
181
182 clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache;
183 if (clear_acl_cache != NULL)
184 clear_acl_cache(inode);
185 spin_lock(&inode->i_lock);
186 NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL;
187 spin_unlock(&inode->i_lock);
188 }
189
190 void nfs_invalidate_atime(struct inode *inode)
191 {
192 spin_lock(&inode->i_lock);
193 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
194 spin_unlock(&inode->i_lock);
195 }
196
197 /*
198 * Invalidate, but do not unhash, the inode.
199 * NB: must be called with inode->i_lock held!
200 */
201 static void nfs_invalidate_inode(struct inode *inode)
202 {
203 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
204 nfs_zap_caches_locked(inode);
205 }
206
207 struct nfs_find_desc {
208 struct nfs_fh *fh;
209 struct nfs_fattr *fattr;
210 };
211
212 /*
213 * In NFSv3 we can have 64bit inode numbers. In order to support
214 * this, and re-exported directories (also seen in NFSv2)
215 * we are forced to allow 2 different inodes to have the same
216 * i_ino.
217 */
218 static int
219 nfs_find_actor(struct inode *inode, void *opaque)
220 {
221 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
222 struct nfs_fh *fh = desc->fh;
223 struct nfs_fattr *fattr = desc->fattr;
224
225 if (NFS_FILEID(inode) != fattr->fileid)
226 return 0;
227 if (nfs_compare_fh(NFS_FH(inode), fh))
228 return 0;
229 if (is_bad_inode(inode) || NFS_STALE(inode))
230 return 0;
231 return 1;
232 }
233
234 static int
235 nfs_init_locked(struct inode *inode, void *opaque)
236 {
237 struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque;
238 struct nfs_fattr *fattr = desc->fattr;
239
240 set_nfs_fileid(inode, fattr->fileid);
241 nfs_copy_fh(NFS_FH(inode), desc->fh);
242 return 0;
243 }
244
245 /*
246 * This is our front-end to iget that looks up inodes by file handle
247 * instead of inode number.
248 */
249 struct inode *
250 nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
251 {
252 struct nfs_find_desc desc = {
253 .fh = fh,
254 .fattr = fattr
255 };
256 struct inode *inode = ERR_PTR(-ENOENT);
257 unsigned long hash;
258
259 nfs_attr_check_mountpoint(sb, fattr);
260
261 if (((fattr->valid & NFS_ATTR_FATTR_FILEID) == 0) &&
262 !nfs_attr_use_mounted_on_fileid(fattr))
263 goto out_no_inode;
264 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) == 0)
265 goto out_no_inode;
266
267 hash = nfs_fattr_to_ino_t(fattr);
268
269 inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc);
270 if (inode == NULL) {
271 inode = ERR_PTR(-ENOMEM);
272 goto out_no_inode;
273 }
274
275 if (inode->i_state & I_NEW) {
276 struct nfs_inode *nfsi = NFS_I(inode);
277 unsigned long now = jiffies;
278
279 /* We set i_ino for the few things that still rely on it,
280 * such as stat(2) */
281 inode->i_ino = hash;
282
283 /* We can't support update_atime(), since the server will reset it */
284 inode->i_flags |= S_NOATIME|S_NOCMTIME;
285 inode->i_mode = fattr->mode;
286 if ((fattr->valid & NFS_ATTR_FATTR_MODE) == 0
287 && nfs_server_capable(inode, NFS_CAP_MODE))
288 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
289 /* Why so? Because we want revalidate for devices/FIFOs, and
290 * that's precisely what we have in nfs_file_inode_operations.
291 */
292 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops;
293 if (S_ISREG(inode->i_mode)) {
294 inode->i_fop = NFS_SB(sb)->nfs_client->rpc_ops->file_ops;
295 inode->i_data.a_ops = &nfs_file_aops;
296 inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info;
297 } else if (S_ISDIR(inode->i_mode)) {
298 inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops;
299 inode->i_fop = &nfs_dir_operations;
300 inode->i_data.a_ops = &nfs_dir_aops;
301 /* Deal with crossing mountpoints */
302 if (fattr->valid & NFS_ATTR_FATTR_MOUNTPOINT ||
303 fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) {
304 if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL)
305 inode->i_op = &nfs_referral_inode_operations;
306 else
307 inode->i_op = &nfs_mountpoint_inode_operations;
308 inode->i_fop = NULL;
309 inode->i_flags |= S_AUTOMOUNT;
310 }
311 } else if (S_ISLNK(inode->i_mode))
312 inode->i_op = &nfs_symlink_inode_operations;
313 else
314 init_special_inode(inode, inode->i_mode, fattr->rdev);
315
316 memset(&inode->i_atime, 0, sizeof(inode->i_atime));
317 memset(&inode->i_mtime, 0, sizeof(inode->i_mtime));
318 memset(&inode->i_ctime, 0, sizeof(inode->i_ctime));
319 inode->i_version = 0;
320 inode->i_size = 0;
321 clear_nlink(inode);
322 inode->i_uid = -2;
323 inode->i_gid = -2;
324 inode->i_blocks = 0;
325 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
326 nfsi->write_io = 0;
327 nfsi->read_io = 0;
328
329 nfsi->read_cache_jiffies = fattr->time_start;
330 nfsi->attr_gencount = fattr->gencount;
331 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
332 inode->i_atime = fattr->atime;
333 else if (nfs_server_capable(inode, NFS_CAP_ATIME))
334 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
335 if (fattr->valid & NFS_ATTR_FATTR_MTIME)
336 inode->i_mtime = fattr->mtime;
337 else if (nfs_server_capable(inode, NFS_CAP_MTIME))
338 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
339 if (fattr->valid & NFS_ATTR_FATTR_CTIME)
340 inode->i_ctime = fattr->ctime;
341 else if (nfs_server_capable(inode, NFS_CAP_CTIME))
342 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
343 if (fattr->valid & NFS_ATTR_FATTR_CHANGE)
344 inode->i_version = fattr->change_attr;
345 else if (nfs_server_capable(inode, NFS_CAP_CHANGE_ATTR))
346 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
347 if (fattr->valid & NFS_ATTR_FATTR_SIZE)
348 inode->i_size = nfs_size_to_loff_t(fattr->size);
349 else
350 nfsi->cache_validity |= NFS_INO_INVALID_ATTR
351 | NFS_INO_REVAL_PAGECACHE;
352 if (fattr->valid & NFS_ATTR_FATTR_NLINK)
353 set_nlink(inode, fattr->nlink);
354 else if (nfs_server_capable(inode, NFS_CAP_NLINK))
355 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
356 if (fattr->valid & NFS_ATTR_FATTR_OWNER)
357 inode->i_uid = fattr->uid;
358 else if (nfs_server_capable(inode, NFS_CAP_OWNER))
359 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
360 if (fattr->valid & NFS_ATTR_FATTR_GROUP)
361 inode->i_gid = fattr->gid;
362 else if (nfs_server_capable(inode, NFS_CAP_OWNER_GROUP))
363 nfsi->cache_validity |= NFS_INO_INVALID_ATTR;
364 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
365 inode->i_blocks = fattr->du.nfs2.blocks;
366 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
367 /*
368 * report the blocks in 512byte units
369 */
370 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
371 }
372 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
373 nfsi->attrtimeo_timestamp = now;
374 nfsi->access_cache = RB_ROOT;
375
376 nfs_fscache_init_inode_cookie(inode);
377
378 unlock_new_inode(inode);
379 } else
380 nfs_refresh_inode(inode, fattr);
381 dprintk("NFS: nfs_fhget(%s/%Ld fh_crc=0x%08x ct=%d)\n",
382 inode->i_sb->s_id,
383 (long long)NFS_FILEID(inode),
384 nfs_display_fhandle_hash(fh),
385 atomic_read(&inode->i_count));
386
387 out:
388 return inode;
389
390 out_no_inode:
391 dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode));
392 goto out;
393 }
394
395 #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)
396
397 int
398 nfs_setattr(struct dentry *dentry, struct iattr *attr)
399 {
400 struct inode *inode = dentry->d_inode;
401 struct nfs_fattr *fattr;
402 int error = -ENOMEM;
403
404 nfs_inc_stats(inode, NFSIOS_VFSSETATTR);
405
406 /* skip mode change if it's just for clearing setuid/setgid */
407 if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID))
408 attr->ia_valid &= ~ATTR_MODE;
409
410 if (attr->ia_valid & ATTR_SIZE) {
411 if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode))
412 attr->ia_valid &= ~ATTR_SIZE;
413 }
414
415 /* Optimization: if the end result is no change, don't RPC */
416 attr->ia_valid &= NFS_VALID_ATTRS;
417 if ((attr->ia_valid & ~(ATTR_FILE|ATTR_OPEN)) == 0)
418 return 0;
419
420 /* Write all dirty data */
421 if (S_ISREG(inode->i_mode))
422 nfs_wb_all(inode);
423
424 fattr = nfs_alloc_fattr();
425 if (fattr == NULL)
426 goto out;
427 /*
428 * Return any delegations if we're going to change ACLs
429 */
430 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0)
431 nfs_inode_return_delegation(inode);
432 error = NFS_PROTO(inode)->setattr(dentry, fattr, attr);
433 if (error == 0)
434 nfs_refresh_inode(inode, fattr);
435 nfs_free_fattr(fattr);
436 out:
437 return error;
438 }
439
440 /**
441 * nfs_vmtruncate - unmap mappings "freed" by truncate() syscall
442 * @inode: inode of the file used
443 * @offset: file offset to start truncating
444 *
445 * This is a copy of the common vmtruncate, but with the locking
446 * corrected to take into account the fact that NFS requires
447 * inode->i_size to be updated under the inode->i_lock.
448 */
449 static int nfs_vmtruncate(struct inode * inode, loff_t offset)
450 {
451 loff_t oldsize;
452 int err;
453
454 err = inode_newsize_ok(inode, offset);
455 if (err)
456 goto out;
457
458 spin_lock(&inode->i_lock);
459 oldsize = inode->i_size;
460 i_size_write(inode, offset);
461 spin_unlock(&inode->i_lock);
462
463 truncate_pagecache(inode, oldsize, offset);
464 out:
465 return err;
466 }
467
468 /**
469 * nfs_setattr_update_inode - Update inode metadata after a setattr call.
470 * @inode: pointer to struct inode
471 * @attr: pointer to struct iattr
472 *
473 * Note: we do this in the *proc.c in order to ensure that
474 * it works for things like exclusive creates too.
475 */
476 void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr)
477 {
478 if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) {
479 spin_lock(&inode->i_lock);
480 if ((attr->ia_valid & ATTR_MODE) != 0) {
481 int mode = attr->ia_mode & S_IALLUGO;
482 mode |= inode->i_mode & ~S_IALLUGO;
483 inode->i_mode = mode;
484 }
485 if ((attr->ia_valid & ATTR_UID) != 0)
486 inode->i_uid = attr->ia_uid;
487 if ((attr->ia_valid & ATTR_GID) != 0)
488 inode->i_gid = attr->ia_gid;
489 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
490 spin_unlock(&inode->i_lock);
491 }
492 if ((attr->ia_valid & ATTR_SIZE) != 0) {
493 nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC);
494 nfs_vmtruncate(inode, attr->ia_size);
495 }
496 }
497
498 int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat)
499 {
500 struct inode *inode = dentry->d_inode;
501 int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME;
502 int err;
503
504 /* Flush out writes to the server in order to update c/mtime. */
505 if (S_ISREG(inode->i_mode)) {
506 err = filemap_write_and_wait(inode->i_mapping);
507 if (err)
508 goto out;
509 }
510
511 /*
512 * We may force a getattr if the user cares about atime.
513 *
514 * Note that we only have to check the vfsmount flags here:
515 * - NFS always sets S_NOATIME by so checking it would give a
516 * bogus result
517 * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is
518 * no point in checking those.
519 */
520 if ((mnt->mnt_flags & MNT_NOATIME) ||
521 ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode)))
522 need_atime = 0;
523
524 if (need_atime)
525 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
526 else
527 err = nfs_revalidate_inode(NFS_SERVER(inode), inode);
528 if (!err) {
529 generic_fillattr(inode, stat);
530 stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode));
531 }
532 out:
533 return err;
534 }
535
536 static void nfs_init_lock_context(struct nfs_lock_context *l_ctx)
537 {
538 atomic_set(&l_ctx->count, 1);
539 l_ctx->lockowner = current->files;
540 l_ctx->pid = current->tgid;
541 INIT_LIST_HEAD(&l_ctx->list);
542 }
543
544 static struct nfs_lock_context *__nfs_find_lock_context(struct nfs_open_context *ctx)
545 {
546 struct nfs_lock_context *pos;
547
548 list_for_each_entry(pos, &ctx->lock_context.list, list) {
549 if (pos->lockowner != current->files)
550 continue;
551 if (pos->pid != current->tgid)
552 continue;
553 atomic_inc(&pos->count);
554 return pos;
555 }
556 return NULL;
557 }
558
559 struct nfs_lock_context *nfs_get_lock_context(struct nfs_open_context *ctx)
560 {
561 struct nfs_lock_context *res, *new = NULL;
562 struct inode *inode = ctx->dentry->d_inode;
563
564 spin_lock(&inode->i_lock);
565 res = __nfs_find_lock_context(ctx);
566 if (res == NULL) {
567 spin_unlock(&inode->i_lock);
568 new = kmalloc(sizeof(*new), GFP_KERNEL);
569 if (new == NULL)
570 return NULL;
571 nfs_init_lock_context(new);
572 spin_lock(&inode->i_lock);
573 res = __nfs_find_lock_context(ctx);
574 if (res == NULL) {
575 list_add_tail(&new->list, &ctx->lock_context.list);
576 new->open_context = ctx;
577 res = new;
578 new = NULL;
579 }
580 }
581 spin_unlock(&inode->i_lock);
582 kfree(new);
583 return res;
584 }
585
586 void nfs_put_lock_context(struct nfs_lock_context *l_ctx)
587 {
588 struct nfs_open_context *ctx = l_ctx->open_context;
589 struct inode *inode = ctx->dentry->d_inode;
590
591 if (!atomic_dec_and_lock(&l_ctx->count, &inode->i_lock))
592 return;
593 list_del(&l_ctx->list);
594 spin_unlock(&inode->i_lock);
595 kfree(l_ctx);
596 }
597
598 /**
599 * nfs_close_context - Common close_context() routine NFSv2/v3
600 * @ctx: pointer to context
601 * @is_sync: is this a synchronous close
602 *
603 * always ensure that the attributes are up to date if we're mounted
604 * with close-to-open semantics
605 */
606 void nfs_close_context(struct nfs_open_context *ctx, int is_sync)
607 {
608 struct inode *inode;
609 struct nfs_server *server;
610
611 if (!(ctx->mode & FMODE_WRITE))
612 return;
613 if (!is_sync)
614 return;
615 inode = ctx->dentry->d_inode;
616 if (!list_empty(&NFS_I(inode)->open_files))
617 return;
618 server = NFS_SERVER(inode);
619 if (server->flags & NFS_MOUNT_NOCTO)
620 return;
621 nfs_revalidate_inode(server, inode);
622 }
623
624 struct nfs_open_context *alloc_nfs_open_context(struct dentry *dentry, fmode_t f_mode)
625 {
626 struct nfs_open_context *ctx;
627 struct rpc_cred *cred = rpc_lookup_cred();
628 if (IS_ERR(cred))
629 return ERR_CAST(cred);
630
631 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
632 if (!ctx) {
633 put_rpccred(cred);
634 return ERR_PTR(-ENOMEM);
635 }
636 nfs_sb_active(dentry->d_sb);
637 ctx->dentry = dget(dentry);
638 ctx->cred = cred;
639 ctx->state = NULL;
640 ctx->mode = f_mode;
641 ctx->flags = 0;
642 ctx->error = 0;
643 nfs_init_lock_context(&ctx->lock_context);
644 ctx->lock_context.open_context = ctx;
645 INIT_LIST_HEAD(&ctx->list);
646 ctx->mdsthreshold = NULL;
647 return ctx;
648 }
649
650 struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx)
651 {
652 if (ctx != NULL)
653 atomic_inc(&ctx->lock_context.count);
654 return ctx;
655 }
656
657 static void __put_nfs_open_context(struct nfs_open_context *ctx, int is_sync)
658 {
659 struct inode *inode = ctx->dentry->d_inode;
660 struct super_block *sb = ctx->dentry->d_sb;
661
662 if (!list_empty(&ctx->list)) {
663 if (!atomic_dec_and_lock(&ctx->lock_context.count, &inode->i_lock))
664 return;
665 list_del(&ctx->list);
666 spin_unlock(&inode->i_lock);
667 } else if (!atomic_dec_and_test(&ctx->lock_context.count))
668 return;
669 if (inode != NULL)
670 NFS_PROTO(inode)->close_context(ctx, is_sync);
671 if (ctx->cred != NULL)
672 put_rpccred(ctx->cred);
673 dput(ctx->dentry);
674 nfs_sb_deactive(sb);
675 kfree(ctx->mdsthreshold);
676 kfree(ctx);
677 }
678
679 void put_nfs_open_context(struct nfs_open_context *ctx)
680 {
681 __put_nfs_open_context(ctx, 0);
682 }
683
684 /*
685 * Ensure that mmap has a recent RPC credential for use when writing out
686 * shared pages
687 */
688 void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx)
689 {
690 struct inode *inode = filp->f_path.dentry->d_inode;
691 struct nfs_inode *nfsi = NFS_I(inode);
692
693 filp->private_data = get_nfs_open_context(ctx);
694 spin_lock(&inode->i_lock);
695 list_add(&ctx->list, &nfsi->open_files);
696 spin_unlock(&inode->i_lock);
697 }
698
699 /*
700 * Given an inode, search for an open context with the desired characteristics
701 */
702 struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, fmode_t mode)
703 {
704 struct nfs_inode *nfsi = NFS_I(inode);
705 struct nfs_open_context *pos, *ctx = NULL;
706
707 spin_lock(&inode->i_lock);
708 list_for_each_entry(pos, &nfsi->open_files, list) {
709 if (cred != NULL && pos->cred != cred)
710 continue;
711 if ((pos->mode & (FMODE_READ|FMODE_WRITE)) != mode)
712 continue;
713 ctx = get_nfs_open_context(pos);
714 break;
715 }
716 spin_unlock(&inode->i_lock);
717 return ctx;
718 }
719
720 static void nfs_file_clear_open_context(struct file *filp)
721 {
722 struct inode *inode = filp->f_path.dentry->d_inode;
723 struct nfs_open_context *ctx = nfs_file_open_context(filp);
724
725 if (ctx) {
726 filp->private_data = NULL;
727 spin_lock(&inode->i_lock);
728 list_move_tail(&ctx->list, &NFS_I(inode)->open_files);
729 spin_unlock(&inode->i_lock);
730 __put_nfs_open_context(ctx, filp->f_flags & O_DIRECT ? 0 : 1);
731 }
732 }
733
734 /*
735 * These allocate and release file read/write context information.
736 */
737 int nfs_open(struct inode *inode, struct file *filp)
738 {
739 struct nfs_open_context *ctx;
740
741 ctx = alloc_nfs_open_context(filp->f_path.dentry, filp->f_mode);
742 if (IS_ERR(ctx))
743 return PTR_ERR(ctx);
744 nfs_file_set_open_context(filp, ctx);
745 put_nfs_open_context(ctx);
746 nfs_fscache_set_inode_cookie(inode, filp);
747 return 0;
748 }
749
750 int nfs_release(struct inode *inode, struct file *filp)
751 {
752 nfs_file_clear_open_context(filp);
753 return 0;
754 }
755
756 /*
757 * This function is called whenever some part of NFS notices that
758 * the cached attributes have to be refreshed.
759 */
760 int
761 __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
762 {
763 int status = -ESTALE;
764 struct nfs_fattr *fattr = NULL;
765 struct nfs_inode *nfsi = NFS_I(inode);
766
767 dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n",
768 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
769
770 if (is_bad_inode(inode))
771 goto out;
772 if (NFS_STALE(inode))
773 goto out;
774
775 status = -ENOMEM;
776 fattr = nfs_alloc_fattr();
777 if (fattr == NULL)
778 goto out;
779
780 nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE);
781 status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), fattr);
782 if (status != 0) {
783 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n",
784 inode->i_sb->s_id,
785 (long long)NFS_FILEID(inode), status);
786 if (status == -ESTALE) {
787 nfs_zap_caches(inode);
788 if (!S_ISDIR(inode->i_mode))
789 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
790 }
791 goto out;
792 }
793
794 status = nfs_refresh_inode(inode, fattr);
795 if (status) {
796 dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n",
797 inode->i_sb->s_id,
798 (long long)NFS_FILEID(inode), status);
799 goto out;
800 }
801
802 if (nfsi->cache_validity & NFS_INO_INVALID_ACL)
803 nfs_zap_acl_cache(inode);
804
805 dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n",
806 inode->i_sb->s_id,
807 (long long)NFS_FILEID(inode));
808
809 out:
810 nfs_free_fattr(fattr);
811 return status;
812 }
813
814 int nfs_attribute_timeout(struct inode *inode)
815 {
816 struct nfs_inode *nfsi = NFS_I(inode);
817
818 return !time_in_range_open(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo);
819 }
820
821 static int nfs_attribute_cache_expired(struct inode *inode)
822 {
823 if (nfs_have_delegated_attributes(inode))
824 return 0;
825 return nfs_attribute_timeout(inode);
826 }
827
828 /**
829 * nfs_revalidate_inode - Revalidate the inode attributes
830 * @server - pointer to nfs_server struct
831 * @inode - pointer to inode struct
832 *
833 * Updates inode attribute information by retrieving the data from the server.
834 */
835 int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode)
836 {
837 if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR)
838 && !nfs_attribute_cache_expired(inode))
839 return NFS_STALE(inode) ? -ESTALE : 0;
840 return __nfs_revalidate_inode(server, inode);
841 }
842
843 static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping)
844 {
845 struct nfs_inode *nfsi = NFS_I(inode);
846
847 if (mapping->nrpages != 0) {
848 int ret = invalidate_inode_pages2(mapping);
849 if (ret < 0)
850 return ret;
851 }
852 spin_lock(&inode->i_lock);
853 nfsi->cache_validity &= ~NFS_INO_INVALID_DATA;
854 if (S_ISDIR(inode->i_mode))
855 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
856 spin_unlock(&inode->i_lock);
857 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
858 nfs_fscache_reset_inode_cookie(inode);
859 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
860 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
861 return 0;
862 }
863
864 static bool nfs_mapping_need_revalidate_inode(struct inode *inode)
865 {
866 if (nfs_have_delegated_attributes(inode))
867 return false;
868 return (NFS_I(inode)->cache_validity & NFS_INO_REVAL_PAGECACHE)
869 || nfs_attribute_timeout(inode)
870 || NFS_STALE(inode);
871 }
872
873 /**
874 * nfs_revalidate_mapping - Revalidate the pagecache
875 * @inode - pointer to host inode
876 * @mapping - pointer to mapping
877 */
878 int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping)
879 {
880 struct nfs_inode *nfsi = NFS_I(inode);
881 int ret = 0;
882
883 if (nfs_mapping_need_revalidate_inode(inode)) {
884 ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
885 if (ret < 0)
886 goto out;
887 }
888 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
889 ret = nfs_invalidate_mapping(inode, mapping);
890 out:
891 return ret;
892 }
893
894 static unsigned long nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr)
895 {
896 struct nfs_inode *nfsi = NFS_I(inode);
897 unsigned long ret = 0;
898
899 if ((fattr->valid & NFS_ATTR_FATTR_PRECHANGE)
900 && (fattr->valid & NFS_ATTR_FATTR_CHANGE)
901 && inode->i_version == fattr->pre_change_attr) {
902 inode->i_version = fattr->change_attr;
903 if (S_ISDIR(inode->i_mode))
904 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
905 ret |= NFS_INO_INVALID_ATTR;
906 }
907 /* If we have atomic WCC data, we may update some attributes */
908 if ((fattr->valid & NFS_ATTR_FATTR_PRECTIME)
909 && (fattr->valid & NFS_ATTR_FATTR_CTIME)
910 && timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) {
911 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
912 ret |= NFS_INO_INVALID_ATTR;
913 }
914
915 if ((fattr->valid & NFS_ATTR_FATTR_PREMTIME)
916 && (fattr->valid & NFS_ATTR_FATTR_MTIME)
917 && timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) {
918 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
919 if (S_ISDIR(inode->i_mode))
920 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
921 ret |= NFS_INO_INVALID_ATTR;
922 }
923 if ((fattr->valid & NFS_ATTR_FATTR_PRESIZE)
924 && (fattr->valid & NFS_ATTR_FATTR_SIZE)
925 && i_size_read(inode) == nfs_size_to_loff_t(fattr->pre_size)
926 && nfsi->npages == 0) {
927 i_size_write(inode, nfs_size_to_loff_t(fattr->size));
928 ret |= NFS_INO_INVALID_ATTR;
929 }
930 return ret;
931 }
932
933 /**
934 * nfs_check_inode_attributes - verify consistency of the inode attribute cache
935 * @inode - pointer to inode
936 * @fattr - updated attributes
937 *
938 * Verifies the attribute cache. If we have just changed the attributes,
939 * so that fattr carries weak cache consistency data, then it may
940 * also update the ctime/mtime/change_attribute.
941 */
942 static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr)
943 {
944 struct nfs_inode *nfsi = NFS_I(inode);
945 loff_t cur_size, new_isize;
946 unsigned long invalid = 0;
947
948
949 if (nfs_have_delegated_attributes(inode))
950 return 0;
951 /* Has the inode gone and changed behind our back? */
952 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid)
953 return -EIO;
954 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT))
955 return -EIO;
956
957 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
958 inode->i_version != fattr->change_attr)
959 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
960
961 /* Verify a few of the more important attributes */
962 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) && !timespec_equal(&inode->i_mtime, &fattr->mtime))
963 invalid |= NFS_INO_INVALID_ATTR;
964
965 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
966 cur_size = i_size_read(inode);
967 new_isize = nfs_size_to_loff_t(fattr->size);
968 if (cur_size != new_isize && nfsi->npages == 0)
969 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
970 }
971
972 /* Have any file permissions changed? */
973 if ((fattr->valid & NFS_ATTR_FATTR_MODE) && (inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO))
974 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
975 if ((fattr->valid & NFS_ATTR_FATTR_OWNER) && inode->i_uid != fattr->uid)
976 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
977 if ((fattr->valid & NFS_ATTR_FATTR_GROUP) && inode->i_gid != fattr->gid)
978 invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL;
979
980 /* Has the link count changed? */
981 if ((fattr->valid & NFS_ATTR_FATTR_NLINK) && inode->i_nlink != fattr->nlink)
982 invalid |= NFS_INO_INVALID_ATTR;
983
984 if ((fattr->valid & NFS_ATTR_FATTR_ATIME) && !timespec_equal(&inode->i_atime, &fattr->atime))
985 invalid |= NFS_INO_INVALID_ATIME;
986
987 if (invalid != 0)
988 nfsi->cache_validity |= invalid;
989
990 nfsi->read_cache_jiffies = fattr->time_start;
991 return 0;
992 }
993
994 static int nfs_ctime_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
995 {
996 if (!(fattr->valid & NFS_ATTR_FATTR_CTIME))
997 return 0;
998 return timespec_compare(&fattr->ctime, &inode->i_ctime) > 0;
999 }
1000
1001 static int nfs_size_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1002 {
1003 if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1004 return 0;
1005 return nfs_size_to_loff_t(fattr->size) > i_size_read(inode);
1006 }
1007
1008 static atomic_long_t nfs_attr_generation_counter;
1009
1010 static unsigned long nfs_read_attr_generation_counter(void)
1011 {
1012 return atomic_long_read(&nfs_attr_generation_counter);
1013 }
1014
1015 unsigned long nfs_inc_attr_generation_counter(void)
1016 {
1017 return atomic_long_inc_return(&nfs_attr_generation_counter);
1018 }
1019
1020 void nfs_fattr_init(struct nfs_fattr *fattr)
1021 {
1022 fattr->valid = 0;
1023 fattr->time_start = jiffies;
1024 fattr->gencount = nfs_inc_attr_generation_counter();
1025 fattr->owner_name = NULL;
1026 fattr->group_name = NULL;
1027 }
1028
1029 struct nfs_fattr *nfs_alloc_fattr(void)
1030 {
1031 struct nfs_fattr *fattr;
1032
1033 fattr = kmalloc(sizeof(*fattr), GFP_NOFS);
1034 if (fattr != NULL)
1035 nfs_fattr_init(fattr);
1036 return fattr;
1037 }
1038
1039 struct nfs_fh *nfs_alloc_fhandle(void)
1040 {
1041 struct nfs_fh *fh;
1042
1043 fh = kmalloc(sizeof(struct nfs_fh), GFP_NOFS);
1044 if (fh != NULL)
1045 fh->size = 0;
1046 return fh;
1047 }
1048
1049 #ifdef NFS_DEBUG
1050 /*
1051 * _nfs_display_fhandle_hash - calculate the crc32 hash for the filehandle
1052 * in the same way that wireshark does
1053 *
1054 * @fh: file handle
1055 *
1056 * For debugging only.
1057 */
1058 u32 _nfs_display_fhandle_hash(const struct nfs_fh *fh)
1059 {
1060 /* wireshark uses 32-bit AUTODIN crc and does a bitwise
1061 * not on the result */
1062 return ~crc32(0xFFFFFFFF, &fh->data[0], fh->size);
1063 }
1064
1065 /*
1066 * _nfs_display_fhandle - display an NFS file handle on the console
1067 *
1068 * @fh: file handle to display
1069 * @caption: display caption
1070 *
1071 * For debugging only.
1072 */
1073 void _nfs_display_fhandle(const struct nfs_fh *fh, const char *caption)
1074 {
1075 unsigned short i;
1076
1077 if (fh == NULL || fh->size == 0) {
1078 printk(KERN_DEFAULT "%s at %p is empty\n", caption, fh);
1079 return;
1080 }
1081
1082 printk(KERN_DEFAULT "%s at %p is %u bytes, crc: 0x%08x:\n",
1083 caption, fh, fh->size, _nfs_display_fhandle_hash(fh));
1084 for (i = 0; i < fh->size; i += 16) {
1085 __be32 *pos = (__be32 *)&fh->data[i];
1086
1087 switch ((fh->size - i - 1) >> 2) {
1088 case 0:
1089 printk(KERN_DEFAULT " %08x\n",
1090 be32_to_cpup(pos));
1091 break;
1092 case 1:
1093 printk(KERN_DEFAULT " %08x %08x\n",
1094 be32_to_cpup(pos), be32_to_cpup(pos + 1));
1095 break;
1096 case 2:
1097 printk(KERN_DEFAULT " %08x %08x %08x\n",
1098 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1099 be32_to_cpup(pos + 2));
1100 break;
1101 default:
1102 printk(KERN_DEFAULT " %08x %08x %08x %08x\n",
1103 be32_to_cpup(pos), be32_to_cpup(pos + 1),
1104 be32_to_cpup(pos + 2), be32_to_cpup(pos + 3));
1105 }
1106 }
1107 }
1108 #endif
1109
1110 /**
1111 * nfs_inode_attrs_need_update - check if the inode attributes need updating
1112 * @inode - pointer to inode
1113 * @fattr - attributes
1114 *
1115 * Attempt to divine whether or not an RPC call reply carrying stale
1116 * attributes got scheduled after another call carrying updated ones.
1117 *
1118 * To do so, the function first assumes that a more recent ctime means
1119 * that the attributes in fattr are newer, however it also attempt to
1120 * catch the case where ctime either didn't change, or went backwards
1121 * (if someone reset the clock on the server) by looking at whether
1122 * or not this RPC call was started after the inode was last updated.
1123 * Note also the check for wraparound of 'attr_gencount'
1124 *
1125 * The function returns 'true' if it thinks the attributes in 'fattr' are
1126 * more recent than the ones cached in the inode.
1127 *
1128 */
1129 static int nfs_inode_attrs_need_update(const struct inode *inode, const struct nfs_fattr *fattr)
1130 {
1131 const struct nfs_inode *nfsi = NFS_I(inode);
1132
1133 return ((long)fattr->gencount - (long)nfsi->attr_gencount) > 0 ||
1134 nfs_ctime_need_update(inode, fattr) ||
1135 nfs_size_need_update(inode, fattr) ||
1136 ((long)nfsi->attr_gencount - (long)nfs_read_attr_generation_counter() > 0);
1137 }
1138
1139 static int nfs_refresh_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1140 {
1141 if (nfs_inode_attrs_need_update(inode, fattr))
1142 return nfs_update_inode(inode, fattr);
1143 return nfs_check_inode_attributes(inode, fattr);
1144 }
1145
1146 /**
1147 * nfs_refresh_inode - try to update the inode attribute cache
1148 * @inode - pointer to inode
1149 * @fattr - updated attributes
1150 *
1151 * Check that an RPC call that returned attributes has not overlapped with
1152 * other recent updates of the inode metadata, then decide whether it is
1153 * safe to do a full update of the inode attributes, or whether just to
1154 * call nfs_check_inode_attributes.
1155 */
1156 int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1157 {
1158 int status;
1159
1160 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1161 return 0;
1162 spin_lock(&inode->i_lock);
1163 status = nfs_refresh_inode_locked(inode, fattr);
1164 spin_unlock(&inode->i_lock);
1165
1166 return status;
1167 }
1168
1169 static int nfs_post_op_update_inode_locked(struct inode *inode, struct nfs_fattr *fattr)
1170 {
1171 struct nfs_inode *nfsi = NFS_I(inode);
1172
1173 nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE;
1174 if (S_ISDIR(inode->i_mode))
1175 nfsi->cache_validity |= NFS_INO_INVALID_DATA;
1176 if ((fattr->valid & NFS_ATTR_FATTR) == 0)
1177 return 0;
1178 return nfs_refresh_inode_locked(inode, fattr);
1179 }
1180
1181 /**
1182 * nfs_post_op_update_inode - try to update the inode attribute cache
1183 * @inode - pointer to inode
1184 * @fattr - updated attributes
1185 *
1186 * After an operation that has changed the inode metadata, mark the
1187 * attribute cache as being invalid, then try to update it.
1188 *
1189 * NB: if the server didn't return any post op attributes, this
1190 * function will force the retrieval of attributes before the next
1191 * NFS request. Thus it should be used only for operations that
1192 * are expected to change one or more attributes, to avoid
1193 * unnecessary NFS requests and trips through nfs_update_inode().
1194 */
1195 int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1196 {
1197 int status;
1198
1199 spin_lock(&inode->i_lock);
1200 status = nfs_post_op_update_inode_locked(inode, fattr);
1201 spin_unlock(&inode->i_lock);
1202 return status;
1203 }
1204
1205 /**
1206 * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache
1207 * @inode - pointer to inode
1208 * @fattr - updated attributes
1209 *
1210 * After an operation that has changed the inode metadata, mark the
1211 * attribute cache as being invalid, then try to update it. Fake up
1212 * weak cache consistency data, if none exist.
1213 *
1214 * This function is mainly designed to be used by the ->write_done() functions.
1215 */
1216 int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr)
1217 {
1218 int status;
1219
1220 spin_lock(&inode->i_lock);
1221 /* Don't do a WCC update if these attributes are already stale */
1222 if ((fattr->valid & NFS_ATTR_FATTR) == 0 ||
1223 !nfs_inode_attrs_need_update(inode, fattr)) {
1224 fattr->valid &= ~(NFS_ATTR_FATTR_PRECHANGE
1225 | NFS_ATTR_FATTR_PRESIZE
1226 | NFS_ATTR_FATTR_PREMTIME
1227 | NFS_ATTR_FATTR_PRECTIME);
1228 goto out_noforce;
1229 }
1230 if ((fattr->valid & NFS_ATTR_FATTR_CHANGE) != 0 &&
1231 (fattr->valid & NFS_ATTR_FATTR_PRECHANGE) == 0) {
1232 fattr->pre_change_attr = inode->i_version;
1233 fattr->valid |= NFS_ATTR_FATTR_PRECHANGE;
1234 }
1235 if ((fattr->valid & NFS_ATTR_FATTR_CTIME) != 0 &&
1236 (fattr->valid & NFS_ATTR_FATTR_PRECTIME) == 0) {
1237 memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime));
1238 fattr->valid |= NFS_ATTR_FATTR_PRECTIME;
1239 }
1240 if ((fattr->valid & NFS_ATTR_FATTR_MTIME) != 0 &&
1241 (fattr->valid & NFS_ATTR_FATTR_PREMTIME) == 0) {
1242 memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime));
1243 fattr->valid |= NFS_ATTR_FATTR_PREMTIME;
1244 }
1245 if ((fattr->valid & NFS_ATTR_FATTR_SIZE) != 0 &&
1246 (fattr->valid & NFS_ATTR_FATTR_PRESIZE) == 0) {
1247 fattr->pre_size = i_size_read(inode);
1248 fattr->valid |= NFS_ATTR_FATTR_PRESIZE;
1249 }
1250 out_noforce:
1251 status = nfs_post_op_update_inode_locked(inode, fattr);
1252 spin_unlock(&inode->i_lock);
1253 return status;
1254 }
1255
1256 /*
1257 * Many nfs protocol calls return the new file attributes after
1258 * an operation. Here we update the inode to reflect the state
1259 * of the server's inode.
1260 *
1261 * This is a bit tricky because we have to make sure all dirty pages
1262 * have been sent off to the server before calling invalidate_inode_pages.
1263 * To make sure no other process adds more write requests while we try
1264 * our best to flush them, we make them sleep during the attribute refresh.
1265 *
1266 * A very similar scenario holds for the dir cache.
1267 */
1268 static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr)
1269 {
1270 struct nfs_server *server;
1271 struct nfs_inode *nfsi = NFS_I(inode);
1272 loff_t cur_isize, new_isize;
1273 unsigned long invalid = 0;
1274 unsigned long now = jiffies;
1275 unsigned long save_cache_validity;
1276
1277 dfprintk(VFS, "NFS: %s(%s/%ld fh_crc=0x%08x ct=%d info=0x%x)\n",
1278 __func__, inode->i_sb->s_id, inode->i_ino,
1279 nfs_display_fhandle_hash(NFS_FH(inode)),
1280 atomic_read(&inode->i_count), fattr->valid);
1281
1282 if ((fattr->valid & NFS_ATTR_FATTR_FILEID) && nfsi->fileid != fattr->fileid) {
1283 printk(KERN_ERR "NFS: server %s error: fileid changed\n"
1284 "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n",
1285 NFS_SERVER(inode)->nfs_client->cl_hostname,
1286 inode->i_sb->s_id, (long long)nfsi->fileid,
1287 (long long)fattr->fileid);
1288 goto out_err;
1289 }
1290
1291 /*
1292 * Make sure the inode's type hasn't changed.
1293 */
1294 if ((fattr->valid & NFS_ATTR_FATTR_TYPE) && (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) {
1295 /*
1296 * Big trouble! The inode has become a different object.
1297 */
1298 printk(KERN_DEBUG "NFS: %s: inode %ld mode changed, %07o to %07o\n",
1299 __func__, inode->i_ino, inode->i_mode, fattr->mode);
1300 goto out_err;
1301 }
1302
1303 server = NFS_SERVER(inode);
1304 /* Update the fsid? */
1305 if (S_ISDIR(inode->i_mode) && (fattr->valid & NFS_ATTR_FATTR_FSID) &&
1306 !nfs_fsid_equal(&server->fsid, &fattr->fsid) &&
1307 !IS_AUTOMOUNT(inode))
1308 server->fsid = fattr->fsid;
1309
1310 /*
1311 * Update the read time so we don't revalidate too often.
1312 */
1313 nfsi->read_cache_jiffies = fattr->time_start;
1314
1315 save_cache_validity = nfsi->cache_validity;
1316 nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR
1317 | NFS_INO_INVALID_ATIME
1318 | NFS_INO_REVAL_FORCED
1319 | NFS_INO_REVAL_PAGECACHE);
1320
1321 /* Do atomic weak cache consistency updates */
1322 invalid |= nfs_wcc_update_inode(inode, fattr);
1323
1324 /* More cache consistency checks */
1325 if (fattr->valid & NFS_ATTR_FATTR_CHANGE) {
1326 if (inode->i_version != fattr->change_attr) {
1327 dprintk("NFS: change_attr change on server for file %s/%ld\n",
1328 inode->i_sb->s_id, inode->i_ino);
1329 invalid |= NFS_INO_INVALID_ATTR
1330 | NFS_INO_INVALID_DATA
1331 | NFS_INO_INVALID_ACCESS
1332 | NFS_INO_INVALID_ACL
1333 | NFS_INO_REVAL_PAGECACHE;
1334 if (S_ISDIR(inode->i_mode))
1335 nfs_force_lookup_revalidate(inode);
1336 inode->i_version = fattr->change_attr;
1337 }
1338 } else if (server->caps & NFS_CAP_CHANGE_ATTR)
1339 invalid |= save_cache_validity;
1340
1341 if (fattr->valid & NFS_ATTR_FATTR_MTIME) {
1342 memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime));
1343 } else if (server->caps & NFS_CAP_MTIME)
1344 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1345 | NFS_INO_REVAL_FORCED);
1346
1347 if (fattr->valid & NFS_ATTR_FATTR_CTIME) {
1348 memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime));
1349 } else if (server->caps & NFS_CAP_CTIME)
1350 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1351 | NFS_INO_REVAL_FORCED);
1352
1353 /* Check if our cached file size is stale */
1354 if (fattr->valid & NFS_ATTR_FATTR_SIZE) {
1355 new_isize = nfs_size_to_loff_t(fattr->size);
1356 cur_isize = i_size_read(inode);
1357 if (new_isize != cur_isize) {
1358 /* Do we perhaps have any outstanding writes, or has
1359 * the file grown beyond our last write? */
1360 if ((nfsi->npages == 0 && !test_bit(NFS_INO_LAYOUTCOMMIT, &nfsi->flags)) ||
1361 new_isize > cur_isize) {
1362 i_size_write(inode, new_isize);
1363 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA;
1364 }
1365 dprintk("NFS: isize change on server for file %s/%ld "
1366 "(%Ld to %Ld)\n",
1367 inode->i_sb->s_id,
1368 inode->i_ino,
1369 (long long)cur_isize,
1370 (long long)new_isize);
1371 }
1372 } else
1373 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1374 | NFS_INO_REVAL_PAGECACHE
1375 | NFS_INO_REVAL_FORCED);
1376
1377
1378 if (fattr->valid & NFS_ATTR_FATTR_ATIME)
1379 memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime));
1380 else if (server->caps & NFS_CAP_ATIME)
1381 invalid |= save_cache_validity & (NFS_INO_INVALID_ATIME
1382 | NFS_INO_REVAL_FORCED);
1383
1384 if (fattr->valid & NFS_ATTR_FATTR_MODE) {
1385 if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO)) {
1386 umode_t newmode = inode->i_mode & S_IFMT;
1387 newmode |= fattr->mode & S_IALLUGO;
1388 inode->i_mode = newmode;
1389 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1390 }
1391 } else if (server->caps & NFS_CAP_MODE)
1392 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1393 | NFS_INO_INVALID_ACCESS
1394 | NFS_INO_INVALID_ACL
1395 | NFS_INO_REVAL_FORCED);
1396
1397 if (fattr->valid & NFS_ATTR_FATTR_OWNER) {
1398 if (inode->i_uid != fattr->uid) {
1399 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1400 inode->i_uid = fattr->uid;
1401 }
1402 } else if (server->caps & NFS_CAP_OWNER)
1403 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1404 | NFS_INO_INVALID_ACCESS
1405 | NFS_INO_INVALID_ACL
1406 | NFS_INO_REVAL_FORCED);
1407
1408 if (fattr->valid & NFS_ATTR_FATTR_GROUP) {
1409 if (inode->i_gid != fattr->gid) {
1410 invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL;
1411 inode->i_gid = fattr->gid;
1412 }
1413 } else if (server->caps & NFS_CAP_OWNER_GROUP)
1414 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1415 | NFS_INO_INVALID_ACCESS
1416 | NFS_INO_INVALID_ACL
1417 | NFS_INO_REVAL_FORCED);
1418
1419 if (fattr->valid & NFS_ATTR_FATTR_NLINK) {
1420 if (inode->i_nlink != fattr->nlink) {
1421 invalid |= NFS_INO_INVALID_ATTR;
1422 if (S_ISDIR(inode->i_mode))
1423 invalid |= NFS_INO_INVALID_DATA;
1424 set_nlink(inode, fattr->nlink);
1425 }
1426 } else if (server->caps & NFS_CAP_NLINK)
1427 invalid |= save_cache_validity & (NFS_INO_INVALID_ATTR
1428 | NFS_INO_REVAL_FORCED);
1429
1430 if (fattr->valid & NFS_ATTR_FATTR_SPACE_USED) {
1431 /*
1432 * report the blocks in 512byte units
1433 */
1434 inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used);
1435 }
1436 if (fattr->valid & NFS_ATTR_FATTR_BLOCKS_USED)
1437 inode->i_blocks = fattr->du.nfs2.blocks;
1438
1439 /* Update attrtimeo value if we're out of the unstable period */
1440 if (invalid & NFS_INO_INVALID_ATTR) {
1441 nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE);
1442 nfsi->attrtimeo = NFS_MINATTRTIMEO(inode);
1443 nfsi->attrtimeo_timestamp = now;
1444 nfsi->attr_gencount = nfs_inc_attr_generation_counter();
1445 } else {
1446 if (!time_in_range_open(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) {
1447 if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode))
1448 nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode);
1449 nfsi->attrtimeo_timestamp = now;
1450 }
1451 }
1452 invalid &= ~NFS_INO_INVALID_ATTR;
1453 /* Don't invalidate the data if we were to blame */
1454 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
1455 || S_ISLNK(inode->i_mode)))
1456 invalid &= ~NFS_INO_INVALID_DATA;
1457 if (!nfs_have_delegation(inode, FMODE_READ) ||
1458 (save_cache_validity & NFS_INO_REVAL_FORCED))
1459 nfsi->cache_validity |= invalid;
1460
1461 return 0;
1462 out_err:
1463 /*
1464 * No need to worry about unhashing the dentry, as the
1465 * lookup validation will know that the inode is bad.
1466 * (But we fall through to invalidate the caches.)
1467 */
1468 nfs_invalidate_inode(inode);
1469 return -ESTALE;
1470 }
1471
1472
1473 #ifdef CONFIG_NFS_V4
1474
1475 /*
1476 * Clean out any remaining NFSv4 state that might be left over due
1477 * to open() calls that passed nfs_atomic_lookup, but failed to call
1478 * nfs_open().
1479 */
1480 void nfs4_evict_inode(struct inode *inode)
1481 {
1482 truncate_inode_pages(&inode->i_data, 0);
1483 clear_inode(inode);
1484 pnfs_return_layout(inode);
1485 pnfs_destroy_layout(NFS_I(inode));
1486 /* If we are holding a delegation, return it! */
1487 nfs_inode_return_delegation_noreclaim(inode);
1488 /* First call standard NFS clear_inode() code */
1489 nfs_clear_inode(inode);
1490 }
1491 #endif
1492
1493 struct inode *nfs_alloc_inode(struct super_block *sb)
1494 {
1495 struct nfs_inode *nfsi;
1496 nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL);
1497 if (!nfsi)
1498 return NULL;
1499 nfsi->flags = 0UL;
1500 nfsi->cache_validity = 0UL;
1501 #ifdef CONFIG_NFS_V3_ACL
1502 nfsi->acl_access = ERR_PTR(-EAGAIN);
1503 nfsi->acl_default = ERR_PTR(-EAGAIN);
1504 #endif
1505 #ifdef CONFIG_NFS_V4
1506 nfsi->nfs4_acl = NULL;
1507 #endif /* CONFIG_NFS_V4 */
1508 return &nfsi->vfs_inode;
1509 }
1510
1511 static void nfs_i_callback(struct rcu_head *head)
1512 {
1513 struct inode *inode = container_of(head, struct inode, i_rcu);
1514 kmem_cache_free(nfs_inode_cachep, NFS_I(inode));
1515 }
1516
1517 void nfs_destroy_inode(struct inode *inode)
1518 {
1519 call_rcu(&inode->i_rcu, nfs_i_callback);
1520 }
1521
1522 static inline void nfs4_init_once(struct nfs_inode *nfsi)
1523 {
1524 #ifdef CONFIG_NFS_V4
1525 INIT_LIST_HEAD(&nfsi->open_states);
1526 nfsi->delegation = NULL;
1527 nfsi->delegation_state = 0;
1528 init_rwsem(&nfsi->rwsem);
1529 nfsi->layout = NULL;
1530 atomic_set(&nfsi->commit_info.rpcs_out, 0);
1531 #endif
1532 }
1533
1534 static void init_once(void *foo)
1535 {
1536 struct nfs_inode *nfsi = (struct nfs_inode *) foo;
1537
1538 inode_init_once(&nfsi->vfs_inode);
1539 INIT_LIST_HEAD(&nfsi->open_files);
1540 INIT_LIST_HEAD(&nfsi->access_cache_entry_lru);
1541 INIT_LIST_HEAD(&nfsi->access_cache_inode_lru);
1542 INIT_LIST_HEAD(&nfsi->commit_info.list);
1543 nfsi->npages = 0;
1544 nfsi->commit_info.ncommit = 0;
1545 atomic_set(&nfsi->silly_count, 1);
1546 INIT_HLIST_HEAD(&nfsi->silly_list);
1547 init_waitqueue_head(&nfsi->waitqueue);
1548 nfs4_init_once(nfsi);
1549 }
1550
1551 static int __init nfs_init_inodecache(void)
1552 {
1553 nfs_inode_cachep = kmem_cache_create("nfs_inode_cache",
1554 sizeof(struct nfs_inode),
1555 0, (SLAB_RECLAIM_ACCOUNT|
1556 SLAB_MEM_SPREAD),
1557 init_once);
1558 if (nfs_inode_cachep == NULL)
1559 return -ENOMEM;
1560
1561 return 0;
1562 }
1563
1564 static void nfs_destroy_inodecache(void)
1565 {
1566 kmem_cache_destroy(nfs_inode_cachep);
1567 }
1568
1569 struct workqueue_struct *nfsiod_workqueue;
1570
1571 /*
1572 * start up the nfsiod workqueue
1573 */
1574 static int nfsiod_start(void)
1575 {
1576 struct workqueue_struct *wq;
1577 dprintk("RPC: creating workqueue nfsiod\n");
1578 wq = alloc_workqueue("nfsiod", WQ_MEM_RECLAIM, 0);
1579 if (wq == NULL)
1580 return -ENOMEM;
1581 nfsiod_workqueue = wq;
1582 return 0;
1583 }
1584
1585 /*
1586 * Destroy the nfsiod workqueue
1587 */
1588 static void nfsiod_stop(void)
1589 {
1590 struct workqueue_struct *wq;
1591
1592 wq = nfsiod_workqueue;
1593 if (wq == NULL)
1594 return;
1595 nfsiod_workqueue = NULL;
1596 destroy_workqueue(wq);
1597 }
1598
1599 int nfs_net_id;
1600 EXPORT_SYMBOL_GPL(nfs_net_id);
1601
1602 static int nfs_net_init(struct net *net)
1603 {
1604 nfs_clients_init(net);
1605 return nfs_dns_resolver_cache_init(net);
1606 }
1607
1608 static void nfs_net_exit(struct net *net)
1609 {
1610 nfs_dns_resolver_cache_destroy(net);
1611 nfs_cleanup_cb_ident_idr(net);
1612 }
1613
1614 static struct pernet_operations nfs_net_ops = {
1615 .init = nfs_net_init,
1616 .exit = nfs_net_exit,
1617 .id = &nfs_net_id,
1618 .size = sizeof(struct nfs_net),
1619 };
1620
1621 /*
1622 * Initialize NFS
1623 */
1624 static int __init init_nfs_fs(void)
1625 {
1626 int err;
1627
1628 err = nfs_idmap_init();
1629 if (err < 0)
1630 goto out10;
1631
1632 err = nfs_dns_resolver_init();
1633 if (err < 0)
1634 goto out9;
1635
1636 err = register_pernet_subsys(&nfs_net_ops);
1637 if (err < 0)
1638 goto out8;
1639
1640 err = nfs_fscache_register();
1641 if (err < 0)
1642 goto out7;
1643
1644 err = nfsiod_start();
1645 if (err)
1646 goto out6;
1647
1648 err = nfs_fs_proc_init();
1649 if (err)
1650 goto out5;
1651
1652 err = nfs_init_nfspagecache();
1653 if (err)
1654 goto out4;
1655
1656 err = nfs_init_inodecache();
1657 if (err)
1658 goto out3;
1659
1660 err = nfs_init_readpagecache();
1661 if (err)
1662 goto out2;
1663
1664 err = nfs_init_writepagecache();
1665 if (err)
1666 goto out1;
1667
1668 err = nfs_init_directcache();
1669 if (err)
1670 goto out0;
1671
1672 #ifdef CONFIG_PROC_FS
1673 rpc_proc_register(&init_net, &nfs_rpcstat);
1674 #endif
1675 if ((err = register_nfs_fs()) != 0)
1676 goto out;
1677 return 0;
1678 out:
1679 #ifdef CONFIG_PROC_FS
1680 rpc_proc_unregister(&init_net, "nfs");
1681 #endif
1682 nfs_destroy_directcache();
1683 out0:
1684 nfs_destroy_writepagecache();
1685 out1:
1686 nfs_destroy_readpagecache();
1687 out2:
1688 nfs_destroy_inodecache();
1689 out3:
1690 nfs_destroy_nfspagecache();
1691 out4:
1692 nfs_fs_proc_exit();
1693 out5:
1694 nfsiod_stop();
1695 out6:
1696 nfs_fscache_unregister();
1697 out7:
1698 unregister_pernet_subsys(&nfs_net_ops);
1699 out8:
1700 nfs_dns_resolver_destroy();
1701 out9:
1702 nfs_idmap_quit();
1703 out10:
1704 return err;
1705 }
1706
1707 static void __exit exit_nfs_fs(void)
1708 {
1709 nfs_destroy_directcache();
1710 nfs_destroy_writepagecache();
1711 nfs_destroy_readpagecache();
1712 nfs_destroy_inodecache();
1713 nfs_destroy_nfspagecache();
1714 nfs_fscache_unregister();
1715 unregister_pernet_subsys(&nfs_net_ops);
1716 nfs_dns_resolver_destroy();
1717 nfs_idmap_quit();
1718 #ifdef CONFIG_PROC_FS
1719 rpc_proc_unregister(&init_net, "nfs");
1720 #endif
1721 unregister_nfs_fs();
1722 nfs_fs_proc_exit();
1723 nfsiod_stop();
1724 }
1725
1726 /* Not quite true; I just maintain it */
1727 MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>");
1728 MODULE_LICENSE("GPL");
1729 module_param(enable_ino64, bool, 0644);
1730
1731 module_init(init_nfs_fs)
1732 module_exit(exit_nfs_fs)