Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/inode.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * nfs inode and superblock handling functions | |
7 | * | |
8 | * Modularised by Alan Cox <Alan.Cox@linux.org>, while hacking some | |
9 | * experimental NFS changes. Modularisation taken straight from SYS5 fs. | |
10 | * | |
11 | * Change to nfs_read_super() to permit NFS mounts to multi-homed hosts. | |
12 | * J.S.Peatfield@damtp.cam.ac.uk | |
13 | * | |
14 | */ | |
15 | ||
1da177e4 LT |
16 | #include <linux/module.h> |
17 | #include <linux/init.h> | |
e8edc6e0 | 18 | #include <linux/sched.h> |
1da177e4 LT |
19 | #include <linux/time.h> |
20 | #include <linux/kernel.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/string.h> | |
23 | #include <linux/stat.h> | |
24 | #include <linux/errno.h> | |
25 | #include <linux/unistd.h> | |
26 | #include <linux/sunrpc/clnt.h> | |
27 | #include <linux/sunrpc/stats.h> | |
4ece3a2d | 28 | #include <linux/sunrpc/metrics.h> |
1da177e4 LT |
29 | #include <linux/nfs_fs.h> |
30 | #include <linux/nfs_mount.h> | |
31 | #include <linux/nfs4_mount.h> | |
32 | #include <linux/lockd/bind.h> | |
33 | #include <linux/smp_lock.h> | |
34 | #include <linux/seq_file.h> | |
35 | #include <linux/mount.h> | |
36 | #include <linux/nfs_idmap.h> | |
37 | #include <linux/vfs.h> | |
9cdb3883 MN |
38 | #include <linux/inet.h> |
39 | #include <linux/nfs_xdr.h> | |
1da177e4 LT |
40 | |
41 | #include <asm/system.h> | |
42 | #include <asm/uaccess.h> | |
43 | ||
4ce79717 | 44 | #include "nfs4_fs.h" |
a72b4422 | 45 | #include "callback.h" |
1da177e4 | 46 | #include "delegation.h" |
d9ef5a8c | 47 | #include "iostat.h" |
f7b422b1 | 48 | #include "internal.h" |
1da177e4 LT |
49 | |
50 | #define NFSDBG_FACILITY NFSDBG_VFS | |
1da177e4 | 51 | |
f43bf0be TM |
52 | #define NFS_64_BIT_INODE_NUMBERS_ENABLED 1 |
53 | ||
54 | /* Default is to see 64-bit inode numbers */ | |
55 | static int enable_ino64 = NFS_64_BIT_INODE_NUMBERS_ENABLED; | |
56 | ||
1da177e4 | 57 | static void nfs_invalidate_inode(struct inode *); |
24aa1fe6 | 58 | static int nfs_update_inode(struct inode *, struct nfs_fattr *); |
1da177e4 | 59 | |
ada70d94 | 60 | static void nfs_zap_acl_cache(struct inode *); |
1da177e4 | 61 | |
e18b890b | 62 | static struct kmem_cache * nfs_inode_cachep; |
b7fa0554 | 63 | |
1da177e4 LT |
64 | static inline unsigned long |
65 | nfs_fattr_to_ino_t(struct nfs_fattr *fattr) | |
66 | { | |
67 | return nfs_fileid_to_ino_t(fattr->fileid); | |
68 | } | |
69 | ||
f43bf0be TM |
70 | /** |
71 | * nfs_compat_user_ino64 - returns the user-visible inode number | |
72 | * @fileid: 64-bit fileid | |
73 | * | |
74 | * This function returns a 32-bit inode number if the boot parameter | |
75 | * nfs.enable_ino64 is zero. | |
76 | */ | |
77 | u64 nfs_compat_user_ino64(u64 fileid) | |
78 | { | |
79 | int ino; | |
80 | ||
81 | if (enable_ino64) | |
82 | return fileid; | |
83 | ino = fileid; | |
84 | if (sizeof(ino) < sizeof(fileid)) | |
85 | ino ^= fileid >> (sizeof(fileid)-sizeof(ino)) * 8; | |
86 | return ino; | |
87 | } | |
88 | ||
f7b422b1 | 89 | int nfs_write_inode(struct inode *inode, int sync) |
1da177e4 | 90 | { |
1da177e4 LT |
91 | int ret; |
92 | ||
d30c8348 TM |
93 | if (sync) { |
94 | ret = filemap_fdatawait(inode->i_mapping); | |
95 | if (ret == 0) | |
96 | ret = nfs_commit_inode(inode, FLUSH_SYNC); | |
97 | } else | |
98 | ret = nfs_commit_inode(inode, 0); | |
99 | if (ret >= 0) | |
100 | return 0; | |
101 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
102 | return ret; | |
1da177e4 LT |
103 | } |
104 | ||
f7b422b1 | 105 | void nfs_clear_inode(struct inode *inode) |
1da177e4 | 106 | { |
da6d503a TM |
107 | /* |
108 | * The following should never happen... | |
109 | */ | |
110 | BUG_ON(nfs_have_writebacks(inode)); | |
1c3c07e9 | 111 | BUG_ON(!list_empty(&NFS_I(inode)->open_files)); |
ada70d94 | 112 | nfs_zap_acl_cache(inode); |
1c3c07e9 | 113 | nfs_access_zap_cache(inode); |
1da177e4 LT |
114 | } |
115 | ||
29884df0 TM |
116 | /** |
117 | * nfs_sync_mapping - helper to flush all mmapped dirty data to disk | |
118 | */ | |
119 | int nfs_sync_mapping(struct address_space *mapping) | |
120 | { | |
121 | int ret; | |
122 | ||
123 | if (mapping->nrpages == 0) | |
124 | return 0; | |
125 | unmap_mapping_range(mapping, 0, 0, 0); | |
28fd1298 | 126 | ret = filemap_write_and_wait(mapping); |
29884df0 TM |
127 | if (ret != 0) |
128 | goto out; | |
129 | ret = nfs_wb_all(mapping->host); | |
130 | out: | |
131 | return ret; | |
132 | } | |
133 | ||
1da177e4 LT |
134 | /* |
135 | * Invalidate the local caches | |
136 | */ | |
b37b03b7 | 137 | static void nfs_zap_caches_locked(struct inode *inode) |
1da177e4 LT |
138 | { |
139 | struct nfs_inode *nfsi = NFS_I(inode); | |
140 | int mode = inode->i_mode; | |
141 | ||
91d5b470 CL |
142 | nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); |
143 | ||
c7c20973 TM |
144 | nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); |
145 | nfsi->attrtimeo_timestamp = jiffies; | |
1da177e4 LT |
146 | |
147 | memset(NFS_COOKIEVERF(inode), 0, sizeof(NFS_COOKIEVERF(inode))); | |
148 | if (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)) | |
55296809 | 149 | nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; |
1da177e4 | 150 | else |
55296809 | 151 | nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL|NFS_INO_REVAL_PAGECACHE; |
b37b03b7 | 152 | } |
dc59250c | 153 | |
b37b03b7 TM |
154 | void nfs_zap_caches(struct inode *inode) |
155 | { | |
156 | spin_lock(&inode->i_lock); | |
157 | nfs_zap_caches_locked(inode); | |
dc59250c | 158 | spin_unlock(&inode->i_lock); |
ada70d94 TM |
159 | } |
160 | ||
cd9ae2b6 TM |
161 | void nfs_zap_mapping(struct inode *inode, struct address_space *mapping) |
162 | { | |
163 | if (mapping->nrpages != 0) { | |
164 | spin_lock(&inode->i_lock); | |
165 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
166 | spin_unlock(&inode->i_lock); | |
167 | } | |
168 | } | |
169 | ||
ada70d94 TM |
170 | static void nfs_zap_acl_cache(struct inode *inode) |
171 | { | |
172 | void (*clear_acl_cache)(struct inode *); | |
173 | ||
174 | clear_acl_cache = NFS_PROTO(inode)->clear_acl_cache; | |
175 | if (clear_acl_cache != NULL) | |
176 | clear_acl_cache(inode); | |
dc59250c | 177 | spin_lock(&inode->i_lock); |
55296809 | 178 | NFS_I(inode)->cache_validity &= ~NFS_INO_INVALID_ACL; |
dc59250c | 179 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
180 | } |
181 | ||
c4812998 TM |
182 | void nfs_invalidate_atime(struct inode *inode) |
183 | { | |
184 | spin_lock(&inode->i_lock); | |
185 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; | |
186 | spin_unlock(&inode->i_lock); | |
187 | } | |
188 | ||
1da177e4 | 189 | /* |
b37b03b7 TM |
190 | * Invalidate, but do not unhash, the inode. |
191 | * NB: must be called with inode->i_lock held! | |
1da177e4 | 192 | */ |
b37b03b7 | 193 | static void nfs_invalidate_inode(struct inode *inode) |
1da177e4 | 194 | { |
3a10c30a | 195 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); |
b37b03b7 | 196 | nfs_zap_caches_locked(inode); |
1da177e4 LT |
197 | } |
198 | ||
199 | struct nfs_find_desc { | |
200 | struct nfs_fh *fh; | |
201 | struct nfs_fattr *fattr; | |
202 | }; | |
203 | ||
204 | /* | |
205 | * In NFSv3 we can have 64bit inode numbers. In order to support | |
206 | * this, and re-exported directories (also seen in NFSv2) | |
207 | * we are forced to allow 2 different inodes to have the same | |
208 | * i_ino. | |
209 | */ | |
210 | static int | |
211 | nfs_find_actor(struct inode *inode, void *opaque) | |
212 | { | |
213 | struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; | |
214 | struct nfs_fh *fh = desc->fh; | |
215 | struct nfs_fattr *fattr = desc->fattr; | |
216 | ||
217 | if (NFS_FILEID(inode) != fattr->fileid) | |
218 | return 0; | |
219 | if (nfs_compare_fh(NFS_FH(inode), fh)) | |
220 | return 0; | |
221 | if (is_bad_inode(inode) || NFS_STALE(inode)) | |
222 | return 0; | |
223 | return 1; | |
224 | } | |
225 | ||
226 | static int | |
227 | nfs_init_locked(struct inode *inode, void *opaque) | |
228 | { | |
229 | struct nfs_find_desc *desc = (struct nfs_find_desc *)opaque; | |
230 | struct nfs_fattr *fattr = desc->fattr; | |
231 | ||
99fadcd7 | 232 | set_nfs_fileid(inode, fattr->fileid); |
1da177e4 LT |
233 | nfs_copy_fh(NFS_FH(inode), desc->fh); |
234 | return 0; | |
235 | } | |
236 | ||
237 | /* Don't use READDIRPLUS on directories that we believe are too large */ | |
238 | #define NFS_LIMIT_READDIRPLUS (8*PAGE_SIZE) | |
239 | ||
240 | /* | |
241 | * This is our front-end to iget that looks up inodes by file handle | |
242 | * instead of inode number. | |
243 | */ | |
244 | struct inode * | |
245 | nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr) | |
246 | { | |
247 | struct nfs_find_desc desc = { | |
248 | .fh = fh, | |
249 | .fattr = fattr | |
250 | }; | |
03f28e3a | 251 | struct inode *inode = ERR_PTR(-ENOENT); |
1da177e4 LT |
252 | unsigned long hash; |
253 | ||
254 | if ((fattr->valid & NFS_ATTR_FATTR) == 0) | |
255 | goto out_no_inode; | |
256 | ||
257 | if (!fattr->nlink) { | |
258 | printk("NFS: Buggy server - nlink == 0!\n"); | |
259 | goto out_no_inode; | |
260 | } | |
261 | ||
262 | hash = nfs_fattr_to_ino_t(fattr); | |
263 | ||
03f28e3a TM |
264 | inode = iget5_locked(sb, hash, nfs_find_actor, nfs_init_locked, &desc); |
265 | if (inode == NULL) { | |
266 | inode = ERR_PTR(-ENOMEM); | |
1da177e4 | 267 | goto out_no_inode; |
03f28e3a | 268 | } |
1da177e4 LT |
269 | |
270 | if (inode->i_state & I_NEW) { | |
271 | struct nfs_inode *nfsi = NFS_I(inode); | |
b0c4fddc | 272 | unsigned long now = jiffies; |
1da177e4 LT |
273 | |
274 | /* We set i_ino for the few things that still rely on it, | |
275 | * such as stat(2) */ | |
276 | inode->i_ino = hash; | |
277 | ||
278 | /* We can't support update_atime(), since the server will reset it */ | |
279 | inode->i_flags |= S_NOATIME|S_NOCMTIME; | |
280 | inode->i_mode = fattr->mode; | |
281 | /* Why so? Because we want revalidate for devices/FIFOs, and | |
282 | * that's precisely what we have in nfs_file_inode_operations. | |
283 | */ | |
8fa5c000 | 284 | inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->file_inode_ops; |
1da177e4 LT |
285 | if (S_ISREG(inode->i_mode)) { |
286 | inode->i_fop = &nfs_file_operations; | |
287 | inode->i_data.a_ops = &nfs_file_aops; | |
288 | inode->i_data.backing_dev_info = &NFS_SB(sb)->backing_dev_info; | |
289 | } else if (S_ISDIR(inode->i_mode)) { | |
8fa5c000 | 290 | inode->i_op = NFS_SB(sb)->nfs_client->rpc_ops->dir_inode_ops; |
1da177e4 LT |
291 | inode->i_fop = &nfs_dir_operations; |
292 | if (nfs_server_capable(inode, NFS_CAP_READDIRPLUS) | |
293 | && fattr->size <= NFS_LIMIT_READDIRPLUS) | |
3a10c30a | 294 | set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
55a97593 TM |
295 | /* Deal with crossing mountpoints */ |
296 | if (!nfs_fsid_equal(&NFS_SB(sb)->fsid, &fattr->fsid)) { | |
6b97fd3d MN |
297 | if (fattr->valid & NFS_ATTR_FATTR_V4_REFERRAL) |
298 | inode->i_op = &nfs_referral_inode_operations; | |
299 | else | |
300 | inode->i_op = &nfs_mountpoint_inode_operations; | |
55a97593 | 301 | inode->i_fop = NULL; |
c37dcd33 | 302 | set_bit(NFS_INO_MOUNTPOINT, &nfsi->flags); |
55a97593 | 303 | } |
1da177e4 LT |
304 | } else if (S_ISLNK(inode->i_mode)) |
305 | inode->i_op = &nfs_symlink_inode_operations; | |
306 | else | |
307 | init_special_inode(inode, inode->i_mode, fattr->rdev); | |
308 | ||
33801147 | 309 | nfsi->read_cache_jiffies = fattr->time_start; |
b0c4fddc TM |
310 | nfsi->last_updated = now; |
311 | nfsi->cache_change_attribute = now; | |
1da177e4 LT |
312 | inode->i_atime = fattr->atime; |
313 | inode->i_mtime = fattr->mtime; | |
314 | inode->i_ctime = fattr->ctime; | |
315 | if (fattr->valid & NFS_ATTR_FATTR_V4) | |
316 | nfsi->change_attr = fattr->change_attr; | |
317 | inode->i_size = nfs_size_to_loff_t(fattr->size); | |
318 | inode->i_nlink = fattr->nlink; | |
319 | inode->i_uid = fattr->uid; | |
320 | inode->i_gid = fattr->gid; | |
321 | if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { | |
322 | /* | |
323 | * report the blocks in 512byte units | |
324 | */ | |
325 | inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); | |
1da177e4 LT |
326 | } else { |
327 | inode->i_blocks = fattr->du.nfs2.blocks; | |
1da177e4 LT |
328 | } |
329 | nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); | |
b0c4fddc | 330 | nfsi->attrtimeo_timestamp = now; |
1da177e4 | 331 | memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); |
1c3c07e9 | 332 | nfsi->access_cache = RB_ROOT; |
1da177e4 LT |
333 | |
334 | unlock_new_inode(inode); | |
335 | } else | |
336 | nfs_refresh_inode(inode, fattr); | |
337 | dprintk("NFS: nfs_fhget(%s/%Ld ct=%d)\n", | |
338 | inode->i_sb->s_id, | |
339 | (long long)NFS_FILEID(inode), | |
340 | atomic_read(&inode->i_count)); | |
341 | ||
342 | out: | |
343 | return inode; | |
344 | ||
345 | out_no_inode: | |
03f28e3a | 346 | dprintk("nfs_fhget: iget failed with error %ld\n", PTR_ERR(inode)); |
1da177e4 LT |
347 | goto out; |
348 | } | |
349 | ||
350 | #define NFS_VALID_ATTRS (ATTR_MODE|ATTR_UID|ATTR_GID|ATTR_SIZE|ATTR_ATIME|ATTR_ATIME_SET|ATTR_MTIME|ATTR_MTIME_SET) | |
351 | ||
352 | int | |
353 | nfs_setattr(struct dentry *dentry, struct iattr *attr) | |
354 | { | |
355 | struct inode *inode = dentry->d_inode; | |
356 | struct nfs_fattr fattr; | |
357 | int error; | |
358 | ||
91d5b470 CL |
359 | nfs_inc_stats(inode, NFSIOS_VFSSETATTR); |
360 | ||
188b95dd JL |
361 | /* skip mode change if it's just for clearing setuid/setgid */ |
362 | if (attr->ia_valid & (ATTR_KILL_SUID | ATTR_KILL_SGID)) | |
363 | attr->ia_valid &= ~ATTR_MODE; | |
364 | ||
1da177e4 LT |
365 | if (attr->ia_valid & ATTR_SIZE) { |
366 | if (!S_ISREG(inode->i_mode) || attr->ia_size == i_size_read(inode)) | |
367 | attr->ia_valid &= ~ATTR_SIZE; | |
368 | } | |
369 | ||
370 | /* Optimization: if the end result is no change, don't RPC */ | |
371 | attr->ia_valid &= NFS_VALID_ATTRS; | |
372 | if (attr->ia_valid == 0) | |
373 | return 0; | |
374 | ||
375 | lock_kernel(); | |
755c1e20 | 376 | /* Write all dirty data */ |
e1552e19 TM |
377 | if (S_ISREG(inode->i_mode)) { |
378 | filemap_write_and_wait(inode->i_mapping); | |
379 | nfs_wb_all(inode); | |
380 | } | |
642ac549 TM |
381 | /* |
382 | * Return any delegations if we're going to change ACLs | |
383 | */ | |
384 | if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) | |
385 | nfs_inode_return_delegation(inode); | |
1da177e4 | 386 | error = NFS_PROTO(inode)->setattr(dentry, &fattr, attr); |
65e4308d | 387 | if (error == 0) |
1da177e4 | 388 | nfs_refresh_inode(inode, &fattr); |
65e4308d TM |
389 | unlock_kernel(); |
390 | return error; | |
391 | } | |
392 | ||
393 | /** | |
394 | * nfs_setattr_update_inode - Update inode metadata after a setattr call. | |
395 | * @inode: pointer to struct inode | |
396 | * @attr: pointer to struct iattr | |
397 | * | |
398 | * Note: we do this in the *proc.c in order to ensure that | |
399 | * it works for things like exclusive creates too. | |
400 | */ | |
401 | void nfs_setattr_update_inode(struct inode *inode, struct iattr *attr) | |
402 | { | |
403 | if ((attr->ia_valid & (ATTR_MODE|ATTR_UID|ATTR_GID)) != 0) { | |
1da177e4 | 404 | if ((attr->ia_valid & ATTR_MODE) != 0) { |
65e4308d TM |
405 | int mode = attr->ia_mode & S_IALLUGO; |
406 | mode |= inode->i_mode & ~S_IALLUGO; | |
1da177e4 LT |
407 | inode->i_mode = mode; |
408 | } | |
409 | if ((attr->ia_valid & ATTR_UID) != 0) | |
410 | inode->i_uid = attr->ia_uid; | |
411 | if ((attr->ia_valid & ATTR_GID) != 0) | |
412 | inode->i_gid = attr->ia_gid; | |
dc59250c | 413 | spin_lock(&inode->i_lock); |
55296809 | 414 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; |
dc59250c | 415 | spin_unlock(&inode->i_lock); |
65e4308d TM |
416 | } |
417 | if ((attr->ia_valid & ATTR_SIZE) != 0) { | |
91d5b470 | 418 | nfs_inc_stats(inode, NFSIOS_SETATTRTRUNC); |
65e4308d TM |
419 | inode->i_size = attr->ia_size; |
420 | vmtruncate(inode, attr->ia_size); | |
421 | } | |
1da177e4 LT |
422 | } |
423 | ||
412d582e CL |
424 | static int nfs_wait_schedule(void *word) |
425 | { | |
426 | if (signal_pending(current)) | |
427 | return -ERESTARTSYS; | |
428 | schedule(); | |
429 | return 0; | |
430 | } | |
431 | ||
1da177e4 LT |
432 | /* |
433 | * Wait for the inode to get unlocked. | |
1da177e4 | 434 | */ |
412d582e | 435 | static int nfs_wait_on_inode(struct inode *inode) |
1da177e4 | 436 | { |
1da177e4 | 437 | struct nfs_inode *nfsi = NFS_I(inode); |
1da177e4 | 438 | int error; |
412d582e | 439 | |
412d582e | 440 | error = wait_on_bit_lock(&nfsi->flags, NFS_INO_REVALIDATING, |
150030b7 | 441 | nfs_wait_schedule, TASK_KILLABLE); |
412d582e | 442 | |
1da177e4 LT |
443 | return error; |
444 | } | |
445 | ||
412d582e CL |
446 | static void nfs_wake_up_inode(struct inode *inode) |
447 | { | |
448 | struct nfs_inode *nfsi = NFS_I(inode); | |
449 | ||
450 | clear_bit(NFS_INO_REVALIDATING, &nfsi->flags); | |
451 | smp_mb__after_clear_bit(); | |
452 | wake_up_bit(&nfsi->flags, NFS_INO_REVALIDATING); | |
453 | } | |
454 | ||
1da177e4 LT |
455 | int nfs_getattr(struct vfsmount *mnt, struct dentry *dentry, struct kstat *stat) |
456 | { | |
457 | struct inode *inode = dentry->d_inode; | |
55296809 | 458 | int need_atime = NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATIME; |
1da177e4 LT |
459 | int err; |
460 | ||
28c494c5 CL |
461 | /* |
462 | * Flush out writes to the server in order to update c/mtime. | |
463 | * | |
464 | * Hold the i_mutex to suspend application writes temporarily; | |
465 | * this prevents long-running writing applications from blocking | |
466 | * nfs_wb_nocommit. | |
467 | */ | |
468 | if (S_ISREG(inode->i_mode)) { | |
469 | mutex_lock(&inode->i_mutex); | |
ed90ef51 | 470 | nfs_wb_nocommit(inode); |
28c494c5 CL |
471 | mutex_unlock(&inode->i_mutex); |
472 | } | |
fc33a7bb CH |
473 | |
474 | /* | |
475 | * We may force a getattr if the user cares about atime. | |
476 | * | |
477 | * Note that we only have to check the vfsmount flags here: | |
478 | * - NFS always sets S_NOATIME by so checking it would give a | |
479 | * bogus result | |
480 | * - NFS never sets MS_NOATIME or MS_NODIRATIME so there is | |
481 | * no point in checking those. | |
482 | */ | |
483 | if ((mnt->mnt_flags & MNT_NOATIME) || | |
484 | ((mnt->mnt_flags & MNT_NODIRATIME) && S_ISDIR(inode->i_mode))) | |
1da177e4 | 485 | need_atime = 0; |
fc33a7bb | 486 | |
1da177e4 LT |
487 | if (need_atime) |
488 | err = __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
489 | else | |
490 | err = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
4e769b93 | 491 | if (!err) { |
1da177e4 | 492 | generic_fillattr(inode, stat); |
f43bf0be | 493 | stat->ino = nfs_compat_user_ino64(NFS_FILEID(inode)); |
4e769b93 | 494 | } |
1da177e4 LT |
495 | return err; |
496 | } | |
497 | ||
b92dccf6 | 498 | static struct nfs_open_context *alloc_nfs_open_context(struct vfsmount *mnt, struct dentry *dentry, struct rpc_cred *cred) |
1da177e4 LT |
499 | { |
500 | struct nfs_open_context *ctx; | |
501 | ||
f52720ca | 502 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); |
1da177e4 | 503 | if (ctx != NULL) { |
88be9f99 TM |
504 | ctx->path.dentry = dget(dentry); |
505 | ctx->path.mnt = mntget(mnt); | |
1da177e4 LT |
506 | ctx->cred = get_rpccred(cred); |
507 | ctx->state = NULL; | |
508 | ctx->lockowner = current->files; | |
66d3aac0 | 509 | ctx->flags = 0; |
1da177e4 | 510 | ctx->error = 0; |
00a92642 | 511 | ctx->dir_cookie = 0; |
5e11934d | 512 | atomic_set(&ctx->count, 1); |
1da177e4 LT |
513 | } |
514 | return ctx; | |
515 | } | |
516 | ||
517 | struct nfs_open_context *get_nfs_open_context(struct nfs_open_context *ctx) | |
518 | { | |
519 | if (ctx != NULL) | |
5e11934d | 520 | atomic_inc(&ctx->count); |
1da177e4 LT |
521 | return ctx; |
522 | } | |
523 | ||
a49c3c77 | 524 | static void __put_nfs_open_context(struct nfs_open_context *ctx, int wait) |
1da177e4 | 525 | { |
383ba719 | 526 | struct inode *inode; |
3bec63db | 527 | |
383ba719 TM |
528 | if (ctx == NULL) |
529 | return; | |
530 | ||
531 | inode = ctx->path.dentry->d_inode; | |
5e11934d TM |
532 | if (!atomic_dec_and_lock(&ctx->count, &inode->i_lock)) |
533 | return; | |
534 | list_del(&ctx->list); | |
535 | spin_unlock(&inode->i_lock); | |
a49c3c77 TM |
536 | if (ctx->state != NULL) { |
537 | if (wait) | |
538 | nfs4_close_sync(&ctx->path, ctx->state, ctx->mode); | |
539 | else | |
540 | nfs4_close_state(&ctx->path, ctx->state, ctx->mode); | |
541 | } | |
3bec63db TM |
542 | if (ctx->cred != NULL) |
543 | put_rpccred(ctx->cred); | |
544 | dput(ctx->path.dentry); | |
545 | mntput(ctx->path.mnt); | |
546 | kfree(ctx); | |
547 | } | |
548 | ||
a49c3c77 TM |
549 | void put_nfs_open_context(struct nfs_open_context *ctx) |
550 | { | |
551 | __put_nfs_open_context(ctx, 0); | |
552 | } | |
553 | ||
554 | static void put_nfs_open_context_sync(struct nfs_open_context *ctx) | |
555 | { | |
556 | __put_nfs_open_context(ctx, 1); | |
557 | } | |
558 | ||
1da177e4 LT |
559 | /* |
560 | * Ensure that mmap has a recent RPC credential for use when writing out | |
561 | * shared pages | |
562 | */ | |
b92dccf6 | 563 | static void nfs_file_set_open_context(struct file *filp, struct nfs_open_context *ctx) |
1da177e4 | 564 | { |
01cce933 | 565 | struct inode *inode = filp->f_path.dentry->d_inode; |
1da177e4 LT |
566 | struct nfs_inode *nfsi = NFS_I(inode); |
567 | ||
568 | filp->private_data = get_nfs_open_context(ctx); | |
569 | spin_lock(&inode->i_lock); | |
570 | list_add(&ctx->list, &nfsi->open_files); | |
571 | spin_unlock(&inode->i_lock); | |
572 | } | |
573 | ||
d530838b TM |
574 | /* |
575 | * Given an inode, search for an open context with the desired characteristics | |
576 | */ | |
577 | struct nfs_open_context *nfs_find_open_context(struct inode *inode, struct rpc_cred *cred, int mode) | |
1da177e4 LT |
578 | { |
579 | struct nfs_inode *nfsi = NFS_I(inode); | |
580 | struct nfs_open_context *pos, *ctx = NULL; | |
581 | ||
582 | spin_lock(&inode->i_lock); | |
583 | list_for_each_entry(pos, &nfsi->open_files, list) { | |
d530838b TM |
584 | if (cred != NULL && pos->cred != cred) |
585 | continue; | |
1da177e4 LT |
586 | if ((pos->mode & mode) == mode) { |
587 | ctx = get_nfs_open_context(pos); | |
588 | break; | |
589 | } | |
590 | } | |
591 | spin_unlock(&inode->i_lock); | |
592 | return ctx; | |
593 | } | |
594 | ||
b92dccf6 | 595 | static void nfs_file_clear_open_context(struct file *filp) |
1da177e4 | 596 | { |
01cce933 | 597 | struct inode *inode = filp->f_path.dentry->d_inode; |
cd3758e3 | 598 | struct nfs_open_context *ctx = nfs_file_open_context(filp); |
1da177e4 LT |
599 | |
600 | if (ctx) { | |
601 | filp->private_data = NULL; | |
602 | spin_lock(&inode->i_lock); | |
603 | list_move_tail(&ctx->list, &NFS_I(inode)->open_files); | |
604 | spin_unlock(&inode->i_lock); | |
a49c3c77 | 605 | put_nfs_open_context_sync(ctx); |
1da177e4 LT |
606 | } |
607 | } | |
608 | ||
609 | /* | |
610 | * These allocate and release file read/write context information. | |
611 | */ | |
612 | int nfs_open(struct inode *inode, struct file *filp) | |
613 | { | |
614 | struct nfs_open_context *ctx; | |
615 | struct rpc_cred *cred; | |
616 | ||
98a8e323 | 617 | cred = rpc_lookup_cred(); |
1da177e4 LT |
618 | if (IS_ERR(cred)) |
619 | return PTR_ERR(cred); | |
01cce933 | 620 | ctx = alloc_nfs_open_context(filp->f_path.mnt, filp->f_path.dentry, cred); |
1da177e4 LT |
621 | put_rpccred(cred); |
622 | if (ctx == NULL) | |
623 | return -ENOMEM; | |
624 | ctx->mode = filp->f_mode; | |
625 | nfs_file_set_open_context(filp, ctx); | |
626 | put_nfs_open_context(ctx); | |
1da177e4 LT |
627 | return 0; |
628 | } | |
629 | ||
630 | int nfs_release(struct inode *inode, struct file *filp) | |
631 | { | |
1da177e4 LT |
632 | nfs_file_clear_open_context(filp); |
633 | return 0; | |
634 | } | |
635 | ||
636 | /* | |
637 | * This function is called whenever some part of NFS notices that | |
638 | * the cached attributes have to be refreshed. | |
639 | */ | |
640 | int | |
641 | __nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) | |
642 | { | |
643 | int status = -ESTALE; | |
644 | struct nfs_fattr fattr; | |
645 | struct nfs_inode *nfsi = NFS_I(inode); | |
1da177e4 LT |
646 | |
647 | dfprintk(PAGECACHE, "NFS: revalidating (%s/%Ld)\n", | |
648 | inode->i_sb->s_id, (long long)NFS_FILEID(inode)); | |
649 | ||
1842bfb4 | 650 | nfs_inc_stats(inode, NFSIOS_INODEREVALIDATE); |
1da177e4 | 651 | lock_kernel(); |
85233a7a | 652 | if (is_bad_inode(inode)) |
1da177e4 LT |
653 | goto out_nowait; |
654 | if (NFS_STALE(inode)) | |
655 | goto out_nowait; | |
656 | ||
412d582e CL |
657 | status = nfs_wait_on_inode(inode); |
658 | if (status < 0) | |
659 | goto out; | |
7fdc49c4 TM |
660 | |
661 | status = -ESTALE; | |
662 | if (NFS_STALE(inode)) | |
663 | goto out; | |
1da177e4 | 664 | |
1da177e4 LT |
665 | status = NFS_PROTO(inode)->getattr(server, NFS_FH(inode), &fattr); |
666 | if (status != 0) { | |
667 | dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) getattr failed, error=%d\n", | |
668 | inode->i_sb->s_id, | |
669 | (long long)NFS_FILEID(inode), status); | |
670 | if (status == -ESTALE) { | |
671 | nfs_zap_caches(inode); | |
672 | if (!S_ISDIR(inode->i_mode)) | |
3a10c30a | 673 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); |
1da177e4 LT |
674 | } |
675 | goto out; | |
676 | } | |
677 | ||
33801147 | 678 | spin_lock(&inode->i_lock); |
24aa1fe6 | 679 | status = nfs_update_inode(inode, &fattr); |
1da177e4 | 680 | if (status) { |
33801147 | 681 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
682 | dfprintk(PAGECACHE, "nfs_revalidate_inode: (%s/%Ld) refresh failed, error=%d\n", |
683 | inode->i_sb->s_id, | |
684 | (long long)NFS_FILEID(inode), status); | |
685 | goto out; | |
686 | } | |
dc59250c | 687 | spin_unlock(&inode->i_lock); |
55296809 | 688 | |
24aa1fe6 | 689 | if (nfsi->cache_validity & NFS_INO_INVALID_ACL) |
ada70d94 | 690 | nfs_zap_acl_cache(inode); |
55296809 | 691 | |
1da177e4 LT |
692 | dfprintk(PAGECACHE, "NFS: (%s/%Ld) revalidation complete\n", |
693 | inode->i_sb->s_id, | |
694 | (long long)NFS_FILEID(inode)); | |
695 | ||
412d582e CL |
696 | out: |
697 | nfs_wake_up_inode(inode); | |
698 | ||
1da177e4 LT |
699 | out_nowait: |
700 | unlock_kernel(); | |
701 | return status; | |
702 | } | |
703 | ||
704 | int nfs_attribute_timeout(struct inode *inode) | |
705 | { | |
706 | struct nfs_inode *nfsi = NFS_I(inode); | |
707 | ||
708 | if (nfs_have_delegation(inode, FMODE_READ)) | |
709 | return 0; | |
b0b53973 TM |
710 | /* |
711 | * Special case: if the attribute timeout is set to 0, then always | |
712 | * treat the cache as having expired (unless holding | |
713 | * a delegation). | |
714 | */ | |
715 | if (nfsi->attrtimeo == 0) | |
716 | return 1; | |
c7e15961 | 717 | return !time_in_range(jiffies, nfsi->read_cache_jiffies, nfsi->read_cache_jiffies + nfsi->attrtimeo); |
1da177e4 LT |
718 | } |
719 | ||
720 | /** | |
721 | * nfs_revalidate_inode - Revalidate the inode attributes | |
722 | * @server - pointer to nfs_server struct | |
723 | * @inode - pointer to inode struct | |
724 | * | |
725 | * Updates inode attribute information by retrieving the data from the server. | |
726 | */ | |
727 | int nfs_revalidate_inode(struct nfs_server *server, struct inode *inode) | |
728 | { | |
44b11874 | 729 | if (!(NFS_I(inode)->cache_validity & NFS_INO_INVALID_ATTR) |
1da177e4 LT |
730 | && !nfs_attribute_timeout(inode)) |
731 | return NFS_STALE(inode) ? -ESTALE : 0; | |
732 | return __nfs_revalidate_inode(server, inode); | |
733 | } | |
734 | ||
717d44e8 TM |
735 | static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_space *mapping) |
736 | { | |
737 | struct nfs_inode *nfsi = NFS_I(inode); | |
738 | ||
739 | if (mapping->nrpages != 0) { | |
740 | int ret = invalidate_inode_pages2(mapping); | |
741 | if (ret < 0) | |
742 | return ret; | |
743 | } | |
744 | spin_lock(&inode->i_lock); | |
745 | nfsi->cache_validity &= ~NFS_INO_INVALID_DATA; | |
2f78e431 | 746 | if (S_ISDIR(inode->i_mode)) |
717d44e8 | 747 | memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); |
717d44e8 TM |
748 | spin_unlock(&inode->i_lock); |
749 | nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); | |
750 | dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", | |
751 | inode->i_sb->s_id, (long long)NFS_FILEID(inode)); | |
752 | return 0; | |
753 | } | |
754 | ||
755 | static int nfs_invalidate_mapping(struct inode *inode, struct address_space *mapping) | |
756 | { | |
757 | int ret = 0; | |
758 | ||
759 | mutex_lock(&inode->i_mutex); | |
760 | if (NFS_I(inode)->cache_validity & NFS_INO_INVALID_DATA) { | |
761 | ret = nfs_sync_mapping(mapping); | |
762 | if (ret == 0) | |
763 | ret = nfs_invalidate_mapping_nolock(inode, mapping); | |
764 | } | |
765 | mutex_unlock(&inode->i_mutex); | |
766 | return ret; | |
767 | } | |
768 | ||
7d52e862 | 769 | /** |
717d44e8 | 770 | * nfs_revalidate_mapping_nolock - Revalidate the pagecache |
7d52e862 TM |
771 | * @inode - pointer to host inode |
772 | * @mapping - pointer to mapping | |
773 | */ | |
717d44e8 | 774 | int nfs_revalidate_mapping_nolock(struct inode *inode, struct address_space *mapping) |
7d52e862 TM |
775 | { |
776 | struct nfs_inode *nfsi = NFS_I(inode); | |
44b11874 TM |
777 | int ret = 0; |
778 | ||
44b11874 | 779 | if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
717d44e8 | 780 | || nfs_attribute_timeout(inode) || NFS_STALE(inode)) { |
44b11874 | 781 | ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); |
717d44e8 TM |
782 | if (ret < 0) |
783 | goto out; | |
784 | } | |
785 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) | |
786 | ret = nfs_invalidate_mapping_nolock(inode, mapping); | |
787 | out: | |
788 | return ret; | |
789 | } | |
7d52e862 | 790 | |
717d44e8 TM |
791 | /** |
792 | * nfs_revalidate_mapping - Revalidate the pagecache | |
793 | * @inode - pointer to host inode | |
794 | * @mapping - pointer to mapping | |
795 | * | |
796 | * This version of the function will take the inode->i_mutex and attempt to | |
797 | * flush out all dirty data if it needs to invalidate the page cache. | |
798 | */ | |
799 | int nfs_revalidate_mapping(struct inode *inode, struct address_space *mapping) | |
800 | { | |
801 | struct nfs_inode *nfsi = NFS_I(inode); | |
802 | int ret = 0; | |
dc59250c | 803 | |
717d44e8 TM |
804 | if ((nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
805 | || nfs_attribute_timeout(inode) || NFS_STALE(inode)) { | |
806 | ret = __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
807 | if (ret < 0) | |
808 | goto out; | |
7d52e862 | 809 | } |
717d44e8 TM |
810 | if (nfsi->cache_validity & NFS_INO_INVALID_DATA) |
811 | ret = nfs_invalidate_mapping(inode, mapping); | |
cd9ae2b6 | 812 | out: |
44b11874 | 813 | return ret; |
7d52e862 TM |
814 | } |
815 | ||
a895b4a1 TM |
816 | static void nfs_wcc_update_inode(struct inode *inode, struct nfs_fattr *fattr) |
817 | { | |
818 | struct nfs_inode *nfsi = NFS_I(inode); | |
819 | ||
70ca8852 TM |
820 | if ((fattr->valid & NFS_ATTR_WCC_V4) != 0 && |
821 | nfsi->change_attr == fattr->pre_change_attr) { | |
822 | nfsi->change_attr = fattr->change_attr; | |
823 | if (S_ISDIR(inode->i_mode)) | |
824 | nfsi->cache_validity |= NFS_INO_INVALID_DATA; | |
825 | } | |
a895b4a1 TM |
826 | /* If we have atomic WCC data, we may update some attributes */ |
827 | if ((fattr->valid & NFS_ATTR_WCC) != 0) { | |
17cadc95 | 828 | if (timespec_equal(&inode->i_ctime, &fattr->pre_ctime)) |
a895b4a1 | 829 | memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); |
a895b4a1 TM |
830 | if (timespec_equal(&inode->i_mtime, &fattr->pre_mtime)) { |
831 | memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); | |
e323ea46 TM |
832 | if (S_ISDIR(inode->i_mode)) |
833 | nfsi->cache_validity |= NFS_INO_INVALID_DATA; | |
a895b4a1 | 834 | } |
8a8c74bf CL |
835 | if (inode->i_size == nfs_size_to_loff_t(fattr->pre_size) && |
836 | nfsi->npages == 0) | |
837 | inode->i_size = nfs_size_to_loff_t(fattr->size); | |
a895b4a1 TM |
838 | } |
839 | } | |
840 | ||
1da177e4 | 841 | /** |
33801147 | 842 | * nfs_check_inode_attributes - verify consistency of the inode attribute cache |
1da177e4 LT |
843 | * @inode - pointer to inode |
844 | * @fattr - updated attributes | |
845 | * | |
846 | * Verifies the attribute cache. If we have just changed the attributes, | |
847 | * so that fattr carries weak cache consistency data, then it may | |
848 | * also update the ctime/mtime/change_attribute. | |
849 | */ | |
33801147 | 850 | static int nfs_check_inode_attributes(struct inode *inode, struct nfs_fattr *fattr) |
1da177e4 LT |
851 | { |
852 | struct nfs_inode *nfsi = NFS_I(inode); | |
853 | loff_t cur_size, new_isize; | |
2a3f5fd4 | 854 | unsigned long invalid = 0; |
1da177e4 | 855 | |
dc59250c | 856 | |
ca62b9c3 TM |
857 | /* Has the inode gone and changed behind our back? */ |
858 | if (nfsi->fileid != fattr->fileid | |
859 | || (inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) { | |
860 | return -EIO; | |
861 | } | |
862 | ||
a895b4a1 TM |
863 | /* Do atomic weak cache consistency updates */ |
864 | nfs_wcc_update_inode(inode, fattr); | |
1da177e4 | 865 | |
9d1e9232 | 866 | if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 && |
f1bb0b92 | 867 | nfsi->change_attr != fattr->change_attr) |
2a3f5fd4 | 868 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; |
1da177e4 | 869 | |
1da177e4 | 870 | /* Verify a few of the more important attributes */ |
f1bb0b92 | 871 | if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) |
2a3f5fd4 | 872 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; |
ca62b9c3 TM |
873 | |
874 | cur_size = i_size_read(inode); | |
875 | new_isize = nfs_size_to_loff_t(fattr->size); | |
fb374d24 | 876 | if (cur_size != new_isize && nfsi->npages == 0) |
2a3f5fd4 | 877 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; |
1da177e4 LT |
878 | |
879 | /* Have any file permissions changed? */ | |
880 | if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) | |
881 | || inode->i_uid != fattr->uid | |
882 | || inode->i_gid != fattr->gid) | |
2a3f5fd4 | 883 | invalid |= NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ACCESS | NFS_INO_INVALID_ACL; |
1da177e4 LT |
884 | |
885 | /* Has the link count changed? */ | |
886 | if (inode->i_nlink != fattr->nlink) | |
2a3f5fd4 | 887 | invalid |= NFS_INO_INVALID_ATTR; |
1da177e4 LT |
888 | |
889 | if (!timespec_equal(&inode->i_atime, &fattr->atime)) | |
2a3f5fd4 TM |
890 | invalid |= NFS_INO_INVALID_ATIME; |
891 | ||
892 | if (invalid != 0) | |
893 | nfsi->cache_validity |= invalid; | |
894 | else | |
895 | nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | |
896 | | NFS_INO_INVALID_ATIME | |
897 | | NFS_INO_REVAL_PAGECACHE); | |
1da177e4 | 898 | |
33801147 | 899 | nfsi->read_cache_jiffies = fattr->time_start; |
1da177e4 LT |
900 | return 0; |
901 | } | |
902 | ||
33801147 TM |
903 | /** |
904 | * nfs_refresh_inode - try to update the inode attribute cache | |
905 | * @inode - pointer to inode | |
906 | * @fattr - updated attributes | |
907 | * | |
908 | * Check that an RPC call that returned attributes has not overlapped with | |
909 | * other recent updates of the inode metadata, then decide whether it is | |
910 | * safe to do a full update of the inode attributes, or whether just to | |
911 | * call nfs_check_inode_attributes. | |
912 | */ | |
913 | int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr) | |
914 | { | |
915 | struct nfs_inode *nfsi = NFS_I(inode); | |
916 | int status; | |
917 | ||
918 | if ((fattr->valid & NFS_ATTR_FATTR) == 0) | |
919 | return 0; | |
920 | spin_lock(&inode->i_lock); | |
33801147 | 921 | if (time_after(fattr->time_start, nfsi->last_updated)) |
24aa1fe6 | 922 | status = nfs_update_inode(inode, fattr); |
33801147 TM |
923 | else |
924 | status = nfs_check_inode_attributes(inode, fattr); | |
925 | ||
926 | spin_unlock(&inode->i_lock); | |
927 | return status; | |
928 | } | |
929 | ||
decf491f TM |
930 | /** |
931 | * nfs_post_op_update_inode - try to update the inode attribute cache | |
932 | * @inode - pointer to inode | |
933 | * @fattr - updated attributes | |
934 | * | |
935 | * After an operation that has changed the inode metadata, mark the | |
936 | * attribute cache as being invalid, then try to update it. | |
f551e44f CL |
937 | * |
938 | * NB: if the server didn't return any post op attributes, this | |
939 | * function will force the retrieval of attributes before the next | |
940 | * NFS request. Thus it should be used only for operations that | |
941 | * are expected to change one or more attributes, to avoid | |
942 | * unnecessary NFS requests and trips through nfs_update_inode(). | |
decf491f TM |
943 | */ |
944 | int nfs_post_op_update_inode(struct inode *inode, struct nfs_fattr *fattr) | |
945 | { | |
946 | struct nfs_inode *nfsi = NFS_I(inode); | |
decf491f | 947 | |
7668fdbe | 948 | spin_lock(&inode->i_lock); |
2a3f5fd4 | 949 | nfsi->cache_validity |= NFS_INO_INVALID_ATTR|NFS_INO_REVAL_PAGECACHE; |
40d24704 TM |
950 | if (S_ISDIR(inode->i_mode)) |
951 | nfsi->cache_validity |= NFS_INO_INVALID_DATA; | |
7668fdbe | 952 | spin_unlock(&inode->i_lock); |
2a3f5fd4 | 953 | return nfs_refresh_inode(inode, fattr); |
decf491f TM |
954 | } |
955 | ||
70ca8852 TM |
956 | /** |
957 | * nfs_post_op_update_inode_force_wcc - try to update the inode attribute cache | |
958 | * @inode - pointer to inode | |
959 | * @fattr - updated attributes | |
960 | * | |
961 | * After an operation that has changed the inode metadata, mark the | |
962 | * attribute cache as being invalid, then try to update it. Fake up | |
963 | * weak cache consistency data, if none exist. | |
964 | * | |
965 | * This function is mainly designed to be used by the ->write_done() functions. | |
966 | */ | |
967 | int nfs_post_op_update_inode_force_wcc(struct inode *inode, struct nfs_fattr *fattr) | |
968 | { | |
969 | if ((fattr->valid & NFS_ATTR_FATTR_V4) != 0 && | |
970 | (fattr->valid & NFS_ATTR_WCC_V4) == 0) { | |
971 | fattr->pre_change_attr = NFS_I(inode)->change_attr; | |
972 | fattr->valid |= NFS_ATTR_WCC_V4; | |
973 | } | |
974 | if ((fattr->valid & NFS_ATTR_FATTR) != 0 && | |
975 | (fattr->valid & NFS_ATTR_WCC) == 0) { | |
976 | memcpy(&fattr->pre_ctime, &inode->i_ctime, sizeof(fattr->pre_ctime)); | |
977 | memcpy(&fattr->pre_mtime, &inode->i_mtime, sizeof(fattr->pre_mtime)); | |
978 | fattr->pre_size = inode->i_size; | |
979 | fattr->valid |= NFS_ATTR_WCC; | |
980 | } | |
981 | return nfs_post_op_update_inode(inode, fattr); | |
982 | } | |
983 | ||
1da177e4 LT |
984 | /* |
985 | * Many nfs protocol calls return the new file attributes after | |
986 | * an operation. Here we update the inode to reflect the state | |
987 | * of the server's inode. | |
988 | * | |
989 | * This is a bit tricky because we have to make sure all dirty pages | |
990 | * have been sent off to the server before calling invalidate_inode_pages. | |
991 | * To make sure no other process adds more write requests while we try | |
992 | * our best to flush them, we make them sleep during the attribute refresh. | |
993 | * | |
994 | * A very similar scenario holds for the dir cache. | |
995 | */ | |
24aa1fe6 | 996 | static int nfs_update_inode(struct inode *inode, struct nfs_fattr *fattr) |
1da177e4 | 997 | { |
8b4bdcf8 | 998 | struct nfs_server *server; |
1da177e4 | 999 | struct nfs_inode *nfsi = NFS_I(inode); |
951a143b | 1000 | loff_t cur_isize, new_isize; |
2a3f5fd4 | 1001 | unsigned long invalid = 0; |
3e7d950a | 1002 | unsigned long now = jiffies; |
1da177e4 LT |
1003 | |
1004 | dfprintk(VFS, "NFS: %s(%s/%ld ct=%d info=0x%x)\n", | |
3110ff80 | 1005 | __func__, inode->i_sb->s_id, inode->i_ino, |
1da177e4 LT |
1006 | atomic_read(&inode->i_count), fattr->valid); |
1007 | ||
325cfed9 CL |
1008 | if (nfsi->fileid != fattr->fileid) |
1009 | goto out_fileid; | |
1da177e4 LT |
1010 | |
1011 | /* | |
1012 | * Make sure the inode's type hasn't changed. | |
1013 | */ | |
33801147 | 1014 | if ((inode->i_mode & S_IFMT) != (fattr->mode & S_IFMT)) |
1da177e4 LT |
1015 | goto out_changed; |
1016 | ||
8b4bdcf8 | 1017 | server = NFS_SERVER(inode); |
a0356862 | 1018 | /* Update the fsid? */ |
c37dcd33 TM |
1019 | if (S_ISDIR(inode->i_mode) && |
1020 | !nfs_fsid_equal(&server->fsid, &fattr->fsid) && | |
1021 | !test_bit(NFS_INO_MOUNTPOINT, &nfsi->flags)) | |
8b4bdcf8 TM |
1022 | server->fsid = fattr->fsid; |
1023 | ||
1da177e4 LT |
1024 | /* |
1025 | * Update the read time so we don't revalidate too often. | |
1026 | */ | |
33801147 | 1027 | nfsi->read_cache_jiffies = fattr->time_start; |
3e7d950a | 1028 | |
f2115dc9 TM |
1029 | nfsi->cache_validity &= ~(NFS_INO_INVALID_ATTR | NFS_INO_INVALID_ATIME |
1030 | | NFS_INO_REVAL_PAGECACHE); | |
1da177e4 | 1031 | |
a895b4a1 TM |
1032 | /* Do atomic weak cache consistency updates */ |
1033 | nfs_wcc_update_inode(inode, fattr); | |
1034 | ||
47aabaa7 TM |
1035 | /* More cache consistency checks */ |
1036 | if (!(fattr->valid & NFS_ATTR_FATTR_V4)) { | |
1037 | /* NFSv2/v3: Check if the mtime agrees */ | |
1038 | if (!timespec_equal(&inode->i_mtime, &fattr->mtime)) { | |
1039 | dprintk("NFS: mtime change on server for file %s/%ld\n", | |
1040 | inode->i_sb->s_id, inode->i_ino); | |
1041 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; | |
bfc69a45 TM |
1042 | if (S_ISDIR(inode->i_mode)) |
1043 | nfs_force_lookup_revalidate(inode); | |
47aabaa7 TM |
1044 | } |
1045 | /* If ctime has changed we should definitely clear access+acl caches */ | |
12b373eb | 1046 | if (!timespec_equal(&inode->i_ctime, &fattr->ctime)) |
47aabaa7 | 1047 | invalid |= NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; |
47aabaa7 TM |
1048 | } else if (nfsi->change_attr != fattr->change_attr) { |
1049 | dprintk("NFS: change_attr change on server for file %s/%ld\n", | |
1050 | inode->i_sb->s_id, inode->i_ino); | |
1051 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; | |
bfc69a45 TM |
1052 | if (S_ISDIR(inode->i_mode)) |
1053 | nfs_force_lookup_revalidate(inode); | |
47aabaa7 TM |
1054 | } |
1055 | ||
951a143b | 1056 | /* Check if our cached file size is stale */ |
1da177e4 LT |
1057 | new_isize = nfs_size_to_loff_t(fattr->size); |
1058 | cur_isize = i_size_read(inode); | |
951a143b | 1059 | if (new_isize != cur_isize) { |
b64e8a5e TM |
1060 | /* Do we perhaps have any outstanding writes, or has |
1061 | * the file grown beyond our last write? */ | |
1062 | if (nfsi->npages == 0 || new_isize > cur_isize) { | |
1da177e4 LT |
1063 | inode->i_size = new_isize; |
1064 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA; | |
1065 | } | |
951a143b TM |
1066 | dprintk("NFS: isize change on server for file %s/%ld\n", |
1067 | inode->i_sb->s_id, inode->i_ino); | |
1da177e4 LT |
1068 | } |
1069 | ||
1da177e4 | 1070 | |
47aabaa7 TM |
1071 | memcpy(&inode->i_mtime, &fattr->mtime, sizeof(inode->i_mtime)); |
1072 | memcpy(&inode->i_ctime, &fattr->ctime, sizeof(inode->i_ctime)); | |
1da177e4 | 1073 | memcpy(&inode->i_atime, &fattr->atime, sizeof(inode->i_atime)); |
47aabaa7 | 1074 | nfsi->change_attr = fattr->change_attr; |
1da177e4 LT |
1075 | |
1076 | if ((inode->i_mode & S_IALLUGO) != (fattr->mode & S_IALLUGO) || | |
1077 | inode->i_uid != fattr->uid || | |
1078 | inode->i_gid != fattr->gid) | |
ada70d94 | 1079 | invalid |= NFS_INO_INVALID_ATTR|NFS_INO_INVALID_ACCESS|NFS_INO_INVALID_ACL; |
1da177e4 LT |
1080 | |
1081 | inode->i_mode = fattr->mode; | |
1082 | inode->i_nlink = fattr->nlink; | |
1083 | inode->i_uid = fattr->uid; | |
1084 | inode->i_gid = fattr->gid; | |
1085 | ||
1086 | if (fattr->valid & (NFS_ATTR_FATTR_V3 | NFS_ATTR_FATTR_V4)) { | |
1087 | /* | |
1088 | * report the blocks in 512byte units | |
1089 | */ | |
1090 | inode->i_blocks = nfs_calc_block_size(fattr->du.nfs3.used); | |
1da177e4 LT |
1091 | } else { |
1092 | inode->i_blocks = fattr->du.nfs2.blocks; | |
1da177e4 LT |
1093 | } |
1094 | ||
1095 | /* Update attrtimeo value if we're out of the unstable period */ | |
1096 | if (invalid & NFS_INO_INVALID_ATTR) { | |
91d5b470 | 1097 | nfs_inc_stats(inode, NFSIOS_ATTRINVALIDATE); |
1da177e4 | 1098 | nfsi->attrtimeo = NFS_MINATTRTIMEO(inode); |
3e7d950a | 1099 | nfsi->attrtimeo_timestamp = now; |
6d2b2966 TM |
1100 | nfsi->last_updated = now; |
1101 | } else { | |
1102 | if (!time_in_range(now, nfsi->attrtimeo_timestamp, nfsi->attrtimeo_timestamp + nfsi->attrtimeo)) { | |
1103 | if ((nfsi->attrtimeo <<= 1) > NFS_MAXATTRTIMEO(inode)) | |
1104 | nfsi->attrtimeo = NFS_MAXATTRTIMEO(inode); | |
1105 | nfsi->attrtimeo_timestamp = now; | |
1106 | } | |
1107 | /* | |
1108 | * Avoid jiffy wraparound issues with nfsi->last_updated | |
1109 | */ | |
1110 | if (!time_in_range(nfsi->last_updated, nfsi->read_cache_jiffies, now)) | |
1111 | nfsi->last_updated = nfsi->read_cache_jiffies; | |
1da177e4 | 1112 | } |
f2115dc9 | 1113 | invalid &= ~NFS_INO_INVALID_ATTR; |
1da177e4 LT |
1114 | /* Don't invalidate the data if we were to blame */ |
1115 | if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) | |
1116 | || S_ISLNK(inode->i_mode))) | |
1117 | invalid &= ~NFS_INO_INVALID_DATA; | |
412c77ce TM |
1118 | if (!nfs_have_delegation(inode, FMODE_READ) || |
1119 | (nfsi->cache_validity & NFS_INO_REVAL_FORCED)) | |
55296809 | 1120 | nfsi->cache_validity |= invalid; |
412c77ce | 1121 | nfsi->cache_validity &= ~NFS_INO_REVAL_FORCED; |
1da177e4 LT |
1122 | |
1123 | return 0; | |
1124 | out_changed: | |
1125 | /* | |
1126 | * Big trouble! The inode has become a different object. | |
1127 | */ | |
1da177e4 | 1128 | printk(KERN_DEBUG "%s: inode %ld mode changed, %07o to %07o\n", |
3110ff80 | 1129 | __func__, inode->i_ino, inode->i_mode, fattr->mode); |
b37b03b7 | 1130 | out_err: |
1da177e4 LT |
1131 | /* |
1132 | * No need to worry about unhashing the dentry, as the | |
1133 | * lookup validation will know that the inode is bad. | |
1134 | * (But we fall through to invalidate the caches.) | |
1135 | */ | |
1136 | nfs_invalidate_inode(inode); | |
1da177e4 | 1137 | return -ESTALE; |
325cfed9 CL |
1138 | |
1139 | out_fileid: | |
1140 | printk(KERN_ERR "NFS: server %s error: fileid changed\n" | |
1141 | "fsid %s: expected fileid 0x%Lx, got 0x%Lx\n", | |
54ceac45 | 1142 | NFS_SERVER(inode)->nfs_client->cl_hostname, inode->i_sb->s_id, |
325cfed9 CL |
1143 | (long long)nfsi->fileid, (long long)fattr->fileid); |
1144 | goto out_err; | |
1da177e4 LT |
1145 | } |
1146 | ||
55a97593 | 1147 | |
1da177e4 LT |
1148 | #ifdef CONFIG_NFS_V4 |
1149 | ||
1da177e4 LT |
1150 | /* |
1151 | * Clean out any remaining NFSv4 state that might be left over due | |
1152 | * to open() calls that passed nfs_atomic_lookup, but failed to call | |
1153 | * nfs_open(). | |
1154 | */ | |
f7b422b1 | 1155 | void nfs4_clear_inode(struct inode *inode) |
1da177e4 | 1156 | { |
1da177e4 | 1157 | /* If we are holding a delegation, return it! */ |
e6f81075 | 1158 | nfs_inode_return_delegation_noreclaim(inode); |
1da177e4 LT |
1159 | /* First call standard NFS clear_inode() code */ |
1160 | nfs_clear_inode(inode); | |
1da177e4 | 1161 | } |
1da177e4 LT |
1162 | #endif |
1163 | ||
f7b422b1 | 1164 | struct inode *nfs_alloc_inode(struct super_block *sb) |
1da177e4 LT |
1165 | { |
1166 | struct nfs_inode *nfsi; | |
e94b1766 | 1167 | nfsi = (struct nfs_inode *)kmem_cache_alloc(nfs_inode_cachep, GFP_KERNEL); |
1da177e4 LT |
1168 | if (!nfsi) |
1169 | return NULL; | |
55296809 CL |
1170 | nfsi->flags = 0UL; |
1171 | nfsi->cache_validity = 0UL; | |
458818ed TM |
1172 | #ifdef CONFIG_NFS_V3_ACL |
1173 | nfsi->acl_access = ERR_PTR(-EAGAIN); | |
1174 | nfsi->acl_default = ERR_PTR(-EAGAIN); | |
1175 | #endif | |
e50a1c2e BF |
1176 | #ifdef CONFIG_NFS_V4 |
1177 | nfsi->nfs4_acl = NULL; | |
1178 | #endif /* CONFIG_NFS_V4 */ | |
1da177e4 LT |
1179 | return &nfsi->vfs_inode; |
1180 | } | |
1181 | ||
f7b422b1 | 1182 | void nfs_destroy_inode(struct inode *inode) |
1da177e4 LT |
1183 | { |
1184 | kmem_cache_free(nfs_inode_cachep, NFS_I(inode)); | |
1185 | } | |
1186 | ||
d75d5414 AM |
1187 | static inline void nfs4_init_once(struct nfs_inode *nfsi) |
1188 | { | |
1189 | #ifdef CONFIG_NFS_V4 | |
1190 | INIT_LIST_HEAD(&nfsi->open_states); | |
1191 | nfsi->delegation = NULL; | |
1192 | nfsi->delegation_state = 0; | |
1193 | init_rwsem(&nfsi->rwsem); | |
1194 | #endif | |
1195 | } | |
f7b422b1 | 1196 | |
4ba9b9d0 | 1197 | static void init_once(struct kmem_cache * cachep, void *foo) |
1da177e4 LT |
1198 | { |
1199 | struct nfs_inode *nfsi = (struct nfs_inode *) foo; | |
1200 | ||
a35afb83 | 1201 | inode_init_once(&nfsi->vfs_inode); |
a35afb83 CL |
1202 | INIT_LIST_HEAD(&nfsi->open_files); |
1203 | INIT_LIST_HEAD(&nfsi->access_cache_entry_lru); | |
1204 | INIT_LIST_HEAD(&nfsi->access_cache_inode_lru); | |
1205 | INIT_RADIX_TREE(&nfsi->nfs_page_tree, GFP_ATOMIC); | |
a35afb83 CL |
1206 | nfsi->ncommit = 0; |
1207 | nfsi->npages = 0; | |
565277f6 TM |
1208 | atomic_set(&nfsi->silly_count, 1); |
1209 | INIT_HLIST_HEAD(&nfsi->silly_list); | |
1210 | init_waitqueue_head(&nfsi->waitqueue); | |
a35afb83 | 1211 | nfs4_init_once(nfsi); |
1da177e4 | 1212 | } |
20c2df83 | 1213 | |
f7b422b1 | 1214 | static int __init nfs_init_inodecache(void) |
1da177e4 LT |
1215 | { |
1216 | nfs_inode_cachep = kmem_cache_create("nfs_inode_cache", | |
1217 | sizeof(struct nfs_inode), | |
fffb60f9 PJ |
1218 | 0, (SLAB_RECLAIM_ACCOUNT| |
1219 | SLAB_MEM_SPREAD), | |
20c2df83 | 1220 | init_once); |
1da177e4 LT |
1221 | if (nfs_inode_cachep == NULL) |
1222 | return -ENOMEM; | |
1223 | ||
1224 | return 0; | |
1225 | } | |
1226 | ||
266bee88 | 1227 | static void nfs_destroy_inodecache(void) |
1da177e4 | 1228 | { |
1a1d92c1 | 1229 | kmem_cache_destroy(nfs_inode_cachep); |
1da177e4 LT |
1230 | } |
1231 | ||
5746006f TM |
1232 | struct workqueue_struct *nfsiod_workqueue; |
1233 | ||
1234 | /* | |
1235 | * start up the nfsiod workqueue | |
1236 | */ | |
1237 | static int nfsiod_start(void) | |
1238 | { | |
1239 | struct workqueue_struct *wq; | |
1240 | dprintk("RPC: creating workqueue nfsiod\n"); | |
1241 | wq = create_singlethread_workqueue("nfsiod"); | |
1242 | if (wq == NULL) | |
1243 | return -ENOMEM; | |
1244 | nfsiod_workqueue = wq; | |
1245 | return 0; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Destroy the nfsiod workqueue | |
1250 | */ | |
1251 | static void nfsiod_stop(void) | |
1252 | { | |
1253 | struct workqueue_struct *wq; | |
1254 | ||
1255 | wq = nfsiod_workqueue; | |
1256 | if (wq == NULL) | |
1257 | return; | |
1258 | nfsiod_workqueue = NULL; | |
1259 | destroy_workqueue(wq); | |
1260 | } | |
1261 | ||
1da177e4 LT |
1262 | /* |
1263 | * Initialize NFS | |
1264 | */ | |
1265 | static int __init init_nfs_fs(void) | |
1266 | { | |
1267 | int err; | |
1268 | ||
5746006f TM |
1269 | err = nfsiod_start(); |
1270 | if (err) | |
1271 | goto out6; | |
1272 | ||
6aaca566 DH |
1273 | err = nfs_fs_proc_init(); |
1274 | if (err) | |
1275 | goto out5; | |
1276 | ||
1da177e4 LT |
1277 | err = nfs_init_nfspagecache(); |
1278 | if (err) | |
1279 | goto out4; | |
1280 | ||
1281 | err = nfs_init_inodecache(); | |
1282 | if (err) | |
1283 | goto out3; | |
1284 | ||
1285 | err = nfs_init_readpagecache(); | |
1286 | if (err) | |
1287 | goto out2; | |
1288 | ||
1289 | err = nfs_init_writepagecache(); | |
1290 | if (err) | |
1291 | goto out1; | |
1292 | ||
1da177e4 LT |
1293 | err = nfs_init_directcache(); |
1294 | if (err) | |
1295 | goto out0; | |
1da177e4 LT |
1296 | |
1297 | #ifdef CONFIG_PROC_FS | |
1298 | rpc_proc_register(&nfs_rpcstat); | |
1299 | #endif | |
f7b422b1 | 1300 | if ((err = register_nfs_fs()) != 0) |
1da177e4 LT |
1301 | goto out; |
1302 | return 0; | |
1303 | out: | |
1304 | #ifdef CONFIG_PROC_FS | |
1305 | rpc_proc_unregister("nfs"); | |
1306 | #endif | |
1da177e4 | 1307 | nfs_destroy_directcache(); |
6b59a754 | 1308 | out0: |
6b59a754 | 1309 | nfs_destroy_writepagecache(); |
1da177e4 LT |
1310 | out1: |
1311 | nfs_destroy_readpagecache(); | |
1312 | out2: | |
1313 | nfs_destroy_inodecache(); | |
1314 | out3: | |
1315 | nfs_destroy_nfspagecache(); | |
1316 | out4: | |
6aaca566 DH |
1317 | nfs_fs_proc_exit(); |
1318 | out5: | |
5746006f TM |
1319 | nfsiod_stop(); |
1320 | out6: | |
1da177e4 LT |
1321 | return err; |
1322 | } | |
1323 | ||
1324 | static void __exit exit_nfs_fs(void) | |
1325 | { | |
1da177e4 | 1326 | nfs_destroy_directcache(); |
1da177e4 LT |
1327 | nfs_destroy_writepagecache(); |
1328 | nfs_destroy_readpagecache(); | |
1329 | nfs_destroy_inodecache(); | |
1330 | nfs_destroy_nfspagecache(); | |
1331 | #ifdef CONFIG_PROC_FS | |
1332 | rpc_proc_unregister("nfs"); | |
1333 | #endif | |
f7b422b1 | 1334 | unregister_nfs_fs(); |
6aaca566 | 1335 | nfs_fs_proc_exit(); |
5746006f | 1336 | nfsiod_stop(); |
1da177e4 LT |
1337 | } |
1338 | ||
1339 | /* Not quite true; I just maintain it */ | |
1340 | MODULE_AUTHOR("Olaf Kirch <okir@monad.swb.de>"); | |
1341 | MODULE_LICENSE("GPL"); | |
f43bf0be | 1342 | module_param(enable_ino64, bool, 0644); |
1da177e4 LT |
1343 | |
1344 | module_init(init_nfs_fs) | |
1345 | module_exit(exit_nfs_fs) |