]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/dir.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * nfs directory handling functions | |
7 | * | |
8 | * 10 Apr 1996 Added silly rename for unlink --okir | |
9 | * 28 Sep 1996 Improved directory cache --okir | |
10 | * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de | |
11 | * Re-implemented silly rename for unlink, newly implemented | |
12 | * silly rename for nfs_rename() following the suggestions | |
13 | * of Olaf Kirch (okir) found in this file. | |
14 | * Following Linus comments on my original hack, this version | |
15 | * depends only on the dcache stuff and doesn't touch the inode | |
16 | * layer (iput() and friends). | |
17 | * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/stat.h> | |
23 | #include <linux/fcntl.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
29 | #include <linux/nfs_fs.h> | |
30 | #include <linux/nfs_mount.h> | |
31 | #include <linux/pagemap.h> | |
873101b3 | 32 | #include <linux/pagevec.h> |
1da177e4 | 33 | #include <linux/namei.h> |
54ceac45 | 34 | #include <linux/mount.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
04e4bd1c | 36 | #include <linux/kmemleak.h> |
64c2ce8b | 37 | #include <linux/xattr.h> |
1da177e4 LT |
38 | |
39 | #include "delegation.h" | |
91d5b470 | 40 | #include "iostat.h" |
4c30d56e | 41 | #include "internal.h" |
cd9a1c0e | 42 | #include "fscache.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | /* #define NFS_DEBUG_VERBOSE 1 */ |
45 | ||
46 | static int nfs_opendir(struct inode *, struct file *); | |
480c2006 | 47 | static int nfs_closedir(struct inode *, struct file *); |
1da177e4 LT |
48 | static int nfs_readdir(struct file *, void *, filldir_t); |
49 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
4acdaf27 | 50 | static int nfs_create(struct inode *, struct dentry *, umode_t, struct nameidata *); |
18bb1db3 | 51 | static int nfs_mkdir(struct inode *, struct dentry *, umode_t); |
1da177e4 LT |
52 | static int nfs_rmdir(struct inode *, struct dentry *); |
53 | static int nfs_unlink(struct inode *, struct dentry *); | |
54 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
55 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
1a67aafb | 56 | static int nfs_mknod(struct inode *, struct dentry *, umode_t, dev_t); |
1da177e4 LT |
57 | static int nfs_rename(struct inode *, struct dentry *, |
58 | struct inode *, struct dentry *); | |
02c24a82 | 59 | static int nfs_fsync_dir(struct file *, loff_t, loff_t, int); |
f0dd2136 | 60 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
11de3b11 | 61 | static void nfs_readdir_clear_array(struct page*); |
1da177e4 | 62 | |
4b6f5d20 | 63 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 64 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
65 | .read = generic_read_dir, |
66 | .readdir = nfs_readdir, | |
67 | .open = nfs_opendir, | |
480c2006 | 68 | .release = nfs_closedir, |
1da177e4 LT |
69 | .fsync = nfs_fsync_dir, |
70 | }; | |
71 | ||
92e1d5be | 72 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
73 | .create = nfs_create, |
74 | .lookup = nfs_lookup, | |
75 | .link = nfs_link, | |
76 | .unlink = nfs_unlink, | |
77 | .symlink = nfs_symlink, | |
78 | .mkdir = nfs_mkdir, | |
79 | .rmdir = nfs_rmdir, | |
80 | .mknod = nfs_mknod, | |
81 | .rename = nfs_rename, | |
82 | .permission = nfs_permission, | |
83 | .getattr = nfs_getattr, | |
84 | .setattr = nfs_setattr, | |
85 | }; | |
86 | ||
11de3b11 TM |
87 | const struct address_space_operations nfs_dir_aops = { |
88 | .freepage = nfs_readdir_clear_array, | |
d1bacf9e BS |
89 | }; |
90 | ||
b7fa0554 | 91 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 92 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
93 | .create = nfs_create, |
94 | .lookup = nfs_lookup, | |
95 | .link = nfs_link, | |
96 | .unlink = nfs_unlink, | |
97 | .symlink = nfs_symlink, | |
98 | .mkdir = nfs_mkdir, | |
99 | .rmdir = nfs_rmdir, | |
100 | .mknod = nfs_mknod, | |
101 | .rename = nfs_rename, | |
102 | .permission = nfs_permission, | |
103 | .getattr = nfs_getattr, | |
104 | .setattr = nfs_setattr, | |
105 | .listxattr = nfs3_listxattr, | |
106 | .getxattr = nfs3_getxattr, | |
107 | .setxattr = nfs3_setxattr, | |
108 | .removexattr = nfs3_removexattr, | |
109 | }; | |
110 | #endif /* CONFIG_NFS_V3 */ | |
111 | ||
1da177e4 LT |
112 | #ifdef CONFIG_NFS_V4 |
113 | ||
0dd2b474 MS |
114 | static struct file *nfs_atomic_open(struct inode *, struct dentry *, |
115 | struct opendata *, unsigned, umode_t, | |
116 | bool *); | |
117 | static int nfs4_create(struct inode *dir, struct dentry *dentry, | |
118 | umode_t mode, struct nameidata *nd); | |
92e1d5be | 119 | const struct inode_operations nfs4_dir_inode_operations = { |
0dd2b474 MS |
120 | .create = nfs4_create, |
121 | .lookup = nfs_lookup, | |
122 | .atomic_open = nfs_atomic_open, | |
1da177e4 LT |
123 | .link = nfs_link, |
124 | .unlink = nfs_unlink, | |
125 | .symlink = nfs_symlink, | |
126 | .mkdir = nfs_mkdir, | |
127 | .rmdir = nfs_rmdir, | |
128 | .mknod = nfs_mknod, | |
129 | .rename = nfs_rename, | |
130 | .permission = nfs_permission, | |
131 | .getattr = nfs_getattr, | |
132 | .setattr = nfs_setattr, | |
64c2ce8b AK |
133 | .getxattr = generic_getxattr, |
134 | .setxattr = generic_setxattr, | |
135 | .listxattr = generic_listxattr, | |
136 | .removexattr = generic_removexattr, | |
1da177e4 LT |
137 | }; |
138 | ||
139 | #endif /* CONFIG_NFS_V4 */ | |
140 | ||
0c030806 | 141 | static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred) |
480c2006 BS |
142 | { |
143 | struct nfs_open_dir_context *ctx; | |
144 | ctx = kmalloc(sizeof(*ctx), GFP_KERNEL); | |
145 | if (ctx != NULL) { | |
8ef2ce3e | 146 | ctx->duped = 0; |
0c030806 | 147 | ctx->attr_gencount = NFS_I(dir)->attr_gencount; |
480c2006 | 148 | ctx->dir_cookie = 0; |
8ef2ce3e | 149 | ctx->dup_cookie = 0; |
480c2006 | 150 | ctx->cred = get_rpccred(cred); |
0c030806 TM |
151 | return ctx; |
152 | } | |
153 | return ERR_PTR(-ENOMEM); | |
480c2006 BS |
154 | } |
155 | ||
156 | static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx) | |
157 | { | |
158 | put_rpccred(ctx->cred); | |
159 | kfree(ctx); | |
160 | } | |
161 | ||
1da177e4 LT |
162 | /* |
163 | * Open file | |
164 | */ | |
165 | static int | |
166 | nfs_opendir(struct inode *inode, struct file *filp) | |
167 | { | |
480c2006 BS |
168 | int res = 0; |
169 | struct nfs_open_dir_context *ctx; | |
170 | struct rpc_cred *cred; | |
1da177e4 | 171 | |
6da24bc9 | 172 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
cc0dd2d1 CL |
173 | filp->f_path.dentry->d_parent->d_name.name, |
174 | filp->f_path.dentry->d_name.name); | |
175 | ||
176 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 177 | |
480c2006 BS |
178 | cred = rpc_lookup_cred(); |
179 | if (IS_ERR(cred)) | |
180 | return PTR_ERR(cred); | |
0c030806 | 181 | ctx = alloc_nfs_open_dir_context(inode, cred); |
480c2006 BS |
182 | if (IS_ERR(ctx)) { |
183 | res = PTR_ERR(ctx); | |
184 | goto out; | |
185 | } | |
186 | filp->private_data = ctx; | |
f5a73672 NB |
187 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
188 | /* This is a mountpoint, so d_revalidate will never | |
189 | * have been called, so we need to refresh the | |
190 | * inode (for close-open consistency) ourselves. | |
191 | */ | |
192 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
193 | } | |
480c2006 BS |
194 | out: |
195 | put_rpccred(cred); | |
1da177e4 LT |
196 | return res; |
197 | } | |
198 | ||
480c2006 BS |
199 | static int |
200 | nfs_closedir(struct inode *inode, struct file *filp) | |
201 | { | |
202 | put_nfs_open_dir_context(filp->private_data); | |
203 | return 0; | |
204 | } | |
205 | ||
d1bacf9e BS |
206 | struct nfs_cache_array_entry { |
207 | u64 cookie; | |
208 | u64 ino; | |
209 | struct qstr string; | |
0b26a0bf | 210 | unsigned char d_type; |
d1bacf9e BS |
211 | }; |
212 | ||
213 | struct nfs_cache_array { | |
88b8e133 | 214 | int size; |
d1bacf9e BS |
215 | int eof_index; |
216 | u64 last_cookie; | |
217 | struct nfs_cache_array_entry array[0]; | |
218 | }; | |
219 | ||
573c4e1e | 220 | typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int); |
1da177e4 LT |
221 | typedef struct { |
222 | struct file *file; | |
223 | struct page *page; | |
224 | unsigned long page_index; | |
f0dd2136 | 225 | u64 *dir_cookie; |
0aded708 | 226 | u64 last_cookie; |
f0dd2136 | 227 | loff_t current_index; |
1da177e4 | 228 | decode_dirent_t decode; |
d1bacf9e | 229 | |
1f4eab7e | 230 | unsigned long timestamp; |
4704f0e2 | 231 | unsigned long gencount; |
d1bacf9e BS |
232 | unsigned int cache_entry_index; |
233 | unsigned int plus:1; | |
234 | unsigned int eof:1; | |
1da177e4 LT |
235 | } nfs_readdir_descriptor_t; |
236 | ||
d1bacf9e BS |
237 | /* |
238 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
1da177e4 LT |
239 | */ |
240 | static | |
d1bacf9e BS |
241 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
242 | { | |
8cd51a0c | 243 | void *ptr; |
d1bacf9e BS |
244 | if (page == NULL) |
245 | return ERR_PTR(-EIO); | |
8cd51a0c TM |
246 | ptr = kmap(page); |
247 | if (ptr == NULL) | |
248 | return ERR_PTR(-ENOMEM); | |
249 | return ptr; | |
d1bacf9e BS |
250 | } |
251 | ||
252 | static | |
253 | void nfs_readdir_release_array(struct page *page) | |
254 | { | |
255 | kunmap(page); | |
256 | } | |
257 | ||
258 | /* | |
259 | * we are freeing strings created by nfs_add_to_readdir_array() | |
260 | */ | |
261 | static | |
11de3b11 | 262 | void nfs_readdir_clear_array(struct page *page) |
d1bacf9e | 263 | { |
11de3b11 | 264 | struct nfs_cache_array *array; |
d1bacf9e | 265 | int i; |
8cd51a0c | 266 | |
2b86ce2d | 267 | array = kmap_atomic(page); |
d1bacf9e BS |
268 | for (i = 0; i < array->size; i++) |
269 | kfree(array->array[i].string.name); | |
2b86ce2d | 270 | kunmap_atomic(array); |
d1bacf9e BS |
271 | } |
272 | ||
273 | /* | |
274 | * the caller is responsible for freeing qstr.name | |
275 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
276 | * nfs_clear_readdir_array() | |
277 | */ | |
278 | static | |
4a201d6e | 279 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
280 | { |
281 | string->len = len; | |
282 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
283 | if (string->name == NULL) |
284 | return -ENOMEM; | |
04e4bd1c CM |
285 | /* |
286 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
287 | * in a page cache page which kmemleak does not scan. | |
288 | */ | |
289 | kmemleak_not_leak(string->name); | |
4a201d6e TM |
290 | string->hash = full_name_hash(name, len); |
291 | return 0; | |
d1bacf9e BS |
292 | } |
293 | ||
294 | static | |
295 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
296 | { | |
297 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
298 | struct nfs_cache_array_entry *cache_entry; |
299 | int ret; | |
300 | ||
d1bacf9e BS |
301 | if (IS_ERR(array)) |
302 | return PTR_ERR(array); | |
3020093f TM |
303 | |
304 | cache_entry = &array->array[array->size]; | |
305 | ||
306 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 307 | ret = -ENOSPC; |
3020093f | 308 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 309 | goto out; |
d1bacf9e | 310 | |
4a201d6e TM |
311 | cache_entry->cookie = entry->prev_cookie; |
312 | cache_entry->ino = entry->ino; | |
0b26a0bf | 313 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
314 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
315 | if (ret) | |
316 | goto out; | |
d1bacf9e | 317 | array->last_cookie = entry->cookie; |
8cd51a0c | 318 | array->size++; |
47c716cb | 319 | if (entry->eof != 0) |
d1bacf9e | 320 | array->eof_index = array->size; |
4a201d6e | 321 | out: |
d1bacf9e | 322 | nfs_readdir_release_array(page); |
4a201d6e | 323 | return ret; |
d1bacf9e BS |
324 | } |
325 | ||
326 | static | |
327 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
328 | { | |
329 | loff_t diff = desc->file->f_pos - desc->current_index; | |
330 | unsigned int index; | |
331 | ||
332 | if (diff < 0) | |
333 | goto out_eof; | |
334 | if (diff >= array->size) { | |
8cd51a0c | 335 | if (array->eof_index >= 0) |
d1bacf9e | 336 | goto out_eof; |
d1bacf9e BS |
337 | return -EAGAIN; |
338 | } | |
339 | ||
340 | index = (unsigned int)diff; | |
341 | *desc->dir_cookie = array->array[index].cookie; | |
342 | desc->cache_entry_index = index; | |
d1bacf9e BS |
343 | return 0; |
344 | out_eof: | |
345 | desc->eof = 1; | |
346 | return -EBADCOOKIE; | |
347 | } | |
348 | ||
349 | static | |
350 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
351 | { | |
352 | int i; | |
8ef2ce3e | 353 | loff_t new_pos; |
d1bacf9e BS |
354 | int status = -EAGAIN; |
355 | ||
356 | for (i = 0; i < array->size; i++) { | |
d1bacf9e | 357 | if (array->array[i].cookie == *desc->dir_cookie) { |
0c030806 TM |
358 | struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode); |
359 | struct nfs_open_dir_context *ctx = desc->file->private_data; | |
360 | ||
8ef2ce3e | 361 | new_pos = desc->current_index + i; |
0c030806 TM |
362 | if (ctx->attr_gencount != nfsi->attr_gencount |
363 | || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) { | |
364 | ctx->duped = 0; | |
365 | ctx->attr_gencount = nfsi->attr_gencount; | |
366 | } else if (new_pos < desc->file->f_pos) { | |
367 | if (ctx->duped > 0 | |
368 | && ctx->dup_cookie == *desc->dir_cookie) { | |
369 | if (printk_ratelimit()) { | |
370 | pr_notice("NFS: directory %s/%s contains a readdir loop." | |
371 | "Please contact your server vendor. " | |
374e4e3e | 372 | "The file: %s has duplicate cookie %llu\n", |
0c030806 TM |
373 | desc->file->f_dentry->d_parent->d_name.name, |
374 | desc->file->f_dentry->d_name.name, | |
374e4e3e | 375 | array->array[i].string.name, |
0c030806 TM |
376 | *desc->dir_cookie); |
377 | } | |
378 | status = -ELOOP; | |
379 | goto out; | |
380 | } | |
8ef2ce3e | 381 | ctx->dup_cookie = *desc->dir_cookie; |
0c030806 | 382 | ctx->duped = -1; |
8ef2ce3e BS |
383 | } |
384 | desc->file->f_pos = new_pos; | |
d1bacf9e | 385 | desc->cache_entry_index = i; |
47c716cb | 386 | return 0; |
d1bacf9e BS |
387 | } |
388 | } | |
47c716cb | 389 | if (array->eof_index >= 0) { |
8cd51a0c | 390 | status = -EBADCOOKIE; |
18fb5fe4 TM |
391 | if (*desc->dir_cookie == array->last_cookie) |
392 | desc->eof = 1; | |
8cd51a0c | 393 | } |
0c030806 | 394 | out: |
d1bacf9e BS |
395 | return status; |
396 | } | |
397 | ||
398 | static | |
399 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
400 | { | |
401 | struct nfs_cache_array *array; | |
47c716cb | 402 | int status; |
d1bacf9e BS |
403 | |
404 | array = nfs_readdir_get_array(desc->page); | |
405 | if (IS_ERR(array)) { | |
406 | status = PTR_ERR(array); | |
407 | goto out; | |
408 | } | |
409 | ||
410 | if (*desc->dir_cookie == 0) | |
411 | status = nfs_readdir_search_for_pos(array, desc); | |
412 | else | |
413 | status = nfs_readdir_search_for_cookie(array, desc); | |
414 | ||
47c716cb | 415 | if (status == -EAGAIN) { |
0aded708 | 416 | desc->last_cookie = array->last_cookie; |
e47c085a | 417 | desc->current_index += array->size; |
47c716cb TM |
418 | desc->page_index++; |
419 | } | |
d1bacf9e BS |
420 | nfs_readdir_release_array(desc->page); |
421 | out: | |
422 | return status; | |
423 | } | |
424 | ||
425 | /* Fill a page with xdr information before transferring to the cache page */ | |
426 | static | |
56e4ebf8 | 427 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 428 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 429 | { |
480c2006 BS |
430 | struct nfs_open_dir_context *ctx = file->private_data; |
431 | struct rpc_cred *cred = ctx->cred; | |
4704f0e2 | 432 | unsigned long timestamp, gencount; |
1da177e4 LT |
433 | int error; |
434 | ||
1da177e4 LT |
435 | again: |
436 | timestamp = jiffies; | |
4704f0e2 | 437 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 438 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
439 | NFS_SERVER(inode)->dtsize, desc->plus); |
440 | if (error < 0) { | |
441 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
442 | if (error == -ENOTSUPP && desc->plus) { | |
443 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 444 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
445 | desc->plus = 0; |
446 | goto again; | |
447 | } | |
448 | goto error; | |
449 | } | |
1f4eab7e | 450 | desc->timestamp = timestamp; |
4704f0e2 | 451 | desc->gencount = gencount; |
d1bacf9e BS |
452 | error: |
453 | return error; | |
1da177e4 LT |
454 | } |
455 | ||
573c4e1e CL |
456 | static int xdr_decode(nfs_readdir_descriptor_t *desc, |
457 | struct nfs_entry *entry, struct xdr_stream *xdr) | |
1da177e4 | 458 | { |
573c4e1e | 459 | int error; |
1da177e4 | 460 | |
573c4e1e CL |
461 | error = desc->decode(xdr, entry, desc->plus); |
462 | if (error) | |
463 | return error; | |
d1bacf9e BS |
464 | entry->fattr->time_start = desc->timestamp; |
465 | entry->fattr->gencount = desc->gencount; | |
466 | return 0; | |
1da177e4 LT |
467 | } |
468 | ||
d39ab9de BS |
469 | static |
470 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
471 | { | |
d39ab9de BS |
472 | if (dentry->d_inode == NULL) |
473 | goto different; | |
37a09f07 | 474 | if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) |
d39ab9de BS |
475 | goto different; |
476 | return 1; | |
477 | different: | |
478 | return 0; | |
479 | } | |
480 | ||
d69ee9b8 TM |
481 | static |
482 | bool nfs_use_readdirplus(struct inode *dir, struct file *filp) | |
483 | { | |
484 | if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS)) | |
485 | return false; | |
486 | if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags)) | |
487 | return true; | |
488 | if (filp->f_pos == 0) | |
489 | return true; | |
490 | return false; | |
491 | } | |
492 | ||
493 | /* | |
494 | * This function is called by the lookup code to request the use of | |
495 | * readdirplus to accelerate any future lookups in the same | |
496 | * directory. | |
497 | */ | |
498 | static | |
499 | void nfs_advise_use_readdirplus(struct inode *dir) | |
500 | { | |
501 | set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags); | |
502 | } | |
503 | ||
d39ab9de BS |
504 | static |
505 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
506 | { | |
26fe5750 | 507 | struct qstr filename = QSTR_INIT(entry->name, entry->len); |
4a201d6e TM |
508 | struct dentry *dentry; |
509 | struct dentry *alias; | |
d39ab9de BS |
510 | struct inode *dir = parent->d_inode; |
511 | struct inode *inode; | |
512 | ||
4a201d6e TM |
513 | if (filename.name[0] == '.') { |
514 | if (filename.len == 1) | |
515 | return; | |
516 | if (filename.len == 2 && filename.name[1] == '.') | |
517 | return; | |
518 | } | |
519 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 520 | |
4a201d6e | 521 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
522 | if (dentry != NULL) { |
523 | if (nfs_same_file(dentry, entry)) { | |
524 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
525 | goto out; | |
526 | } else { | |
527 | d_drop(dentry); | |
528 | dput(dentry); | |
529 | } | |
530 | } | |
531 | ||
532 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
533 | if (dentry == NULL) |
534 | return; | |
535 | ||
d39ab9de BS |
536 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); |
537 | if (IS_ERR(inode)) | |
538 | goto out; | |
539 | ||
540 | alias = d_materialise_unique(dentry, inode); | |
541 | if (IS_ERR(alias)) | |
542 | goto out; | |
543 | else if (alias) { | |
544 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
545 | dput(alias); | |
546 | } else | |
547 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
548 | ||
549 | out: | |
550 | dput(dentry); | |
d39ab9de BS |
551 | } |
552 | ||
d1bacf9e BS |
553 | /* Perform conversion from xdr to cache array */ |
554 | static | |
8cd51a0c | 555 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
6650239a | 556 | struct page **xdr_pages, struct page *page, unsigned int buflen) |
1da177e4 | 557 | { |
babddc72 | 558 | struct xdr_stream stream; |
f7da7a12 | 559 | struct xdr_buf buf; |
6650239a | 560 | struct page *scratch; |
99424380 | 561 | struct nfs_cache_array *array; |
5c346854 TM |
562 | unsigned int count = 0; |
563 | int status; | |
babddc72 | 564 | |
6650239a TM |
565 | scratch = alloc_page(GFP_KERNEL); |
566 | if (scratch == NULL) | |
567 | return -ENOMEM; | |
babddc72 | 568 | |
f7da7a12 | 569 | xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen); |
6650239a | 570 | xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE); |
99424380 BS |
571 | |
572 | do { | |
573 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
574 | if (status != 0) { |
575 | if (status == -EAGAIN) | |
576 | status = 0; | |
99424380 | 577 | break; |
8cd51a0c | 578 | } |
99424380 | 579 | |
5c346854 TM |
580 | count++; |
581 | ||
47c716cb | 582 | if (desc->plus != 0) |
d39ab9de | 583 | nfs_prime_dcache(desc->file->f_path.dentry, entry); |
8cd51a0c TM |
584 | |
585 | status = nfs_readdir_add_to_array(entry, page); | |
586 | if (status != 0) | |
587 | break; | |
99424380 BS |
588 | } while (!entry->eof); |
589 | ||
47c716cb | 590 | if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) { |
99424380 | 591 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
592 | if (!IS_ERR(array)) { |
593 | array->eof_index = array->size; | |
594 | status = 0; | |
595 | nfs_readdir_release_array(page); | |
5c346854 TM |
596 | } else |
597 | status = PTR_ERR(array); | |
1da177e4 | 598 | } |
6650239a TM |
599 | |
600 | put_page(scratch); | |
8cd51a0c | 601 | return status; |
56e4ebf8 BS |
602 | } |
603 | ||
604 | static | |
605 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
606 | { | |
607 | unsigned int i; | |
608 | for (i = 0; i < npages; i++) | |
609 | put_page(pages[i]); | |
610 | } | |
611 | ||
612 | static | |
613 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
614 | unsigned int npages) | |
615 | { | |
56e4ebf8 BS |
616 | nfs_readdir_free_pagearray(pages, npages); |
617 | } | |
618 | ||
619 | /* | |
620 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
621 | * to nfs_readdir_free_large_page | |
622 | */ | |
623 | static | |
6650239a | 624 | int nfs_readdir_large_page(struct page **pages, unsigned int npages) |
56e4ebf8 | 625 | { |
56e4ebf8 BS |
626 | unsigned int i; |
627 | ||
628 | for (i = 0; i < npages; i++) { | |
629 | struct page *page = alloc_page(GFP_KERNEL); | |
630 | if (page == NULL) | |
631 | goto out_freepages; | |
632 | pages[i] = page; | |
633 | } | |
6650239a | 634 | return 0; |
56e4ebf8 | 635 | |
56e4ebf8 BS |
636 | out_freepages: |
637 | nfs_readdir_free_pagearray(pages, i); | |
6650239a | 638 | return -ENOMEM; |
1da177e4 LT |
639 | } |
640 | ||
d1bacf9e BS |
641 | static |
642 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 643 | { |
56e4ebf8 BS |
644 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
645 | void *pages_ptr = NULL; | |
d1bacf9e BS |
646 | struct nfs_entry entry; |
647 | struct file *file = desc->file; | |
648 | struct nfs_cache_array *array; | |
8cd51a0c | 649 | int status = -ENOMEM; |
56e4ebf8 | 650 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
651 | |
652 | entry.prev_cookie = 0; | |
0aded708 | 653 | entry.cookie = desc->last_cookie; |
d1bacf9e BS |
654 | entry.eof = 0; |
655 | entry.fh = nfs_alloc_fhandle(); | |
656 | entry.fattr = nfs_alloc_fattr(); | |
573c4e1e | 657 | entry.server = NFS_SERVER(inode); |
d1bacf9e BS |
658 | if (entry.fh == NULL || entry.fattr == NULL) |
659 | goto out; | |
00a92642 | 660 | |
d1bacf9e | 661 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
662 | if (IS_ERR(array)) { |
663 | status = PTR_ERR(array); | |
664 | goto out; | |
665 | } | |
d1bacf9e BS |
666 | memset(array, 0, sizeof(struct nfs_cache_array)); |
667 | array->eof_index = -1; | |
00a92642 | 668 | |
6650239a TM |
669 | status = nfs_readdir_large_page(pages, array_size); |
670 | if (status < 0) | |
d1bacf9e BS |
671 | goto out_release_array; |
672 | do { | |
ac396128 | 673 | unsigned int pglen; |
56e4ebf8 | 674 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 675 | |
d1bacf9e | 676 | if (status < 0) |
00a92642 | 677 | break; |
ac396128 | 678 | pglen = status; |
6650239a | 679 | status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen); |
8cd51a0c TM |
680 | if (status < 0) { |
681 | if (status == -ENOSPC) | |
682 | status = 0; | |
683 | break; | |
684 | } | |
685 | } while (array->eof_index < 0); | |
d1bacf9e | 686 | |
56e4ebf8 | 687 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
688 | out_release_array: |
689 | nfs_readdir_release_array(page); | |
690 | out: | |
691 | nfs_free_fattr(entry.fattr); | |
692 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
693 | return status; |
694 | } | |
695 | ||
696 | /* | |
d1bacf9e BS |
697 | * Now we cache directories properly, by converting xdr information |
698 | * to an array that can be used for lookups later. This results in | |
699 | * fewer cache pages, since we can store more information on each page. | |
700 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 701 | */ |
d1bacf9e BS |
702 | static |
703 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 704 | { |
01cce933 | 705 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
8cd51a0c | 706 | int ret; |
1da177e4 | 707 | |
8cd51a0c TM |
708 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
709 | if (ret < 0) | |
d1bacf9e BS |
710 | goto error; |
711 | SetPageUptodate(page); | |
1da177e4 | 712 | |
d1bacf9e BS |
713 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
714 | /* Should never happen */ | |
715 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 716 | } |
d1bacf9e BS |
717 | unlock_page(page); |
718 | return 0; | |
719 | error: | |
720 | unlock_page(page); | |
8cd51a0c | 721 | return ret; |
d1bacf9e | 722 | } |
1da177e4 | 723 | |
d1bacf9e BS |
724 | static |
725 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
726 | { | |
11de3b11 TM |
727 | if (!desc->page->mapping) |
728 | nfs_readdir_clear_array(desc->page); | |
d1bacf9e BS |
729 | page_cache_release(desc->page); |
730 | desc->page = NULL; | |
731 | } | |
732 | ||
733 | static | |
734 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
735 | { | |
8cd51a0c | 736 | return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, |
d1bacf9e | 737 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
738 | } |
739 | ||
740 | /* | |
d1bacf9e | 741 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 742 | */ |
d1bacf9e BS |
743 | static |
744 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
745 | { | |
746 | int res; | |
747 | ||
748 | desc->page = get_cache_page(desc); | |
749 | if (IS_ERR(desc->page)) | |
750 | return PTR_ERR(desc->page); | |
751 | ||
752 | res = nfs_readdir_search_array(desc); | |
47c716cb TM |
753 | if (res != 0) |
754 | cache_page_release(desc); | |
d1bacf9e BS |
755 | return res; |
756 | } | |
757 | ||
758 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
759 | static inline |
760 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
761 | { | |
8cd51a0c | 762 | int res; |
d1bacf9e | 763 | |
0aded708 | 764 | if (desc->page_index == 0) { |
8cd51a0c | 765 | desc->current_index = 0; |
0aded708 TM |
766 | desc->last_cookie = 0; |
767 | } | |
47c716cb | 768 | do { |
d1bacf9e | 769 | res = find_cache_page(desc); |
47c716cb | 770 | } while (res == -EAGAIN); |
1da177e4 LT |
771 | return res; |
772 | } | |
773 | ||
1da177e4 LT |
774 | /* |
775 | * Once we've found the start of the dirent within a page: fill 'er up... | |
776 | */ | |
777 | static | |
778 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
779 | filldir_t filldir) | |
780 | { | |
781 | struct file *file = desc->file; | |
d1bacf9e BS |
782 | int i = 0; |
783 | int res = 0; | |
784 | struct nfs_cache_array *array = NULL; | |
8ef2ce3e BS |
785 | struct nfs_open_dir_context *ctx = file->private_data; |
786 | ||
d1bacf9e | 787 | array = nfs_readdir_get_array(desc->page); |
e7c58e97 TM |
788 | if (IS_ERR(array)) { |
789 | res = PTR_ERR(array); | |
790 | goto out; | |
791 | } | |
d1bacf9e BS |
792 | |
793 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 794 | struct nfs_cache_array_entry *ent; |
1da177e4 | 795 | |
ece0b423 TM |
796 | ent = &array->array[i]; |
797 | if (filldir(dirent, ent->string.name, ent->string.len, | |
0b26a0bf TM |
798 | file->f_pos, nfs_compat_user_ino64(ent->ino), |
799 | ent->d_type) < 0) { | |
ece0b423 | 800 | desc->eof = 1; |
1da177e4 | 801 | break; |
ece0b423 | 802 | } |
00a92642 | 803 | file->f_pos++; |
d1bacf9e BS |
804 | if (i < (array->size-1)) |
805 | *desc->dir_cookie = array->array[i+1].cookie; | |
806 | else | |
807 | *desc->dir_cookie = array->last_cookie; | |
0c030806 TM |
808 | if (ctx->duped != 0) |
809 | ctx->duped = 1; | |
1da177e4 | 810 | } |
47c716cb | 811 | if (array->eof_index >= 0) |
8cd51a0c | 812 | desc->eof = 1; |
d1bacf9e BS |
813 | |
814 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 815 | out: |
d1bacf9e | 816 | cache_page_release(desc); |
1e7cb3dc CL |
817 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
818 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
819 | return res; |
820 | } | |
821 | ||
822 | /* | |
823 | * If we cannot find a cookie in our cache, we suspect that this is | |
824 | * because it points to a deleted file, so we ask the server to return | |
825 | * whatever it thinks is the next entry. We then feed this to filldir. | |
826 | * If all goes well, we should then be able to find our way round the | |
827 | * cache on the next call to readdir_search_pagecache(); | |
828 | * | |
829 | * NOTE: we cannot add the anonymous page to the pagecache because | |
830 | * the data it contains might not be page aligned. Besides, | |
831 | * we should already have a complete representation of the | |
832 | * directory in the page cache by the time we get here. | |
833 | */ | |
834 | static inline | |
835 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
836 | filldir_t filldir) | |
837 | { | |
1da177e4 LT |
838 | struct page *page = NULL; |
839 | int status; | |
d1bacf9e | 840 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
0c030806 | 841 | struct nfs_open_dir_context *ctx = desc->file->private_data; |
1da177e4 | 842 | |
1e7cb3dc CL |
843 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
844 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
845 | |
846 | page = alloc_page(GFP_HIGHUSER); | |
847 | if (!page) { | |
848 | status = -ENOMEM; | |
849 | goto out; | |
850 | } | |
d1bacf9e | 851 | |
7a8e1dc3 | 852 | desc->page_index = 0; |
0aded708 | 853 | desc->last_cookie = *desc->dir_cookie; |
7a8e1dc3 | 854 | desc->page = page; |
0c030806 | 855 | ctx->duped = 0; |
7a8e1dc3 | 856 | |
85f8607e TM |
857 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
858 | if (status < 0) | |
1da177e4 LT |
859 | goto out_release; |
860 | ||
861 | status = nfs_do_filldir(desc, dirent, filldir); | |
862 | ||
1da177e4 | 863 | out: |
1e7cb3dc | 864 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 865 | __func__, status); |
1da177e4 LT |
866 | return status; |
867 | out_release: | |
d1bacf9e | 868 | cache_page_release(desc); |
1da177e4 LT |
869 | goto out; |
870 | } | |
871 | ||
00a92642 OG |
872 | /* The file offset position represents the dirent entry number. A |
873 | last cookie cache takes care of the common case of reading the | |
874 | whole directory. | |
1da177e4 LT |
875 | */ |
876 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
877 | { | |
01cce933 | 878 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
879 | struct inode *inode = dentry->d_inode; |
880 | nfs_readdir_descriptor_t my_desc, | |
881 | *desc = &my_desc; | |
480c2006 | 882 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
47c716cb | 883 | int res; |
1da177e4 | 884 | |
6da24bc9 | 885 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
886 | dentry->d_parent->d_name.name, dentry->d_name.name, |
887 | (long long)filp->f_pos); | |
91d5b470 CL |
888 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
889 | ||
1da177e4 | 890 | /* |
00a92642 | 891 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 892 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
893 | * to either find the entry with the appropriate number or |
894 | * revalidate the cookie. | |
1da177e4 LT |
895 | */ |
896 | memset(desc, 0, sizeof(*desc)); | |
897 | ||
898 | desc->file = filp; | |
480c2006 | 899 | desc->dir_cookie = &dir_ctx->dir_cookie; |
1da177e4 | 900 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
d69ee9b8 | 901 | desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0; |
1da177e4 | 902 | |
565277f6 | 903 | nfs_block_sillyrename(dentry); |
1cda707d | 904 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
905 | if (res < 0) |
906 | goto out; | |
907 | ||
47c716cb | 908 | do { |
1da177e4 | 909 | res = readdir_search_pagecache(desc); |
00a92642 | 910 | |
1da177e4 | 911 | if (res == -EBADCOOKIE) { |
ece0b423 | 912 | res = 0; |
1da177e4 | 913 | /* This means either end of directory */ |
d1bacf9e | 914 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
915 | /* Or that the server has 'lost' a cookie */ |
916 | res = uncached_readdir(desc, dirent, filldir); | |
ece0b423 | 917 | if (res == 0) |
1da177e4 LT |
918 | continue; |
919 | } | |
1da177e4 LT |
920 | break; |
921 | } | |
922 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 923 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 924 | nfs_zap_caches(inode); |
baf57a09 | 925 | desc->page_index = 0; |
1da177e4 | 926 | desc->plus = 0; |
d1bacf9e | 927 | desc->eof = 0; |
1da177e4 LT |
928 | continue; |
929 | } | |
930 | if (res < 0) | |
931 | break; | |
932 | ||
933 | res = nfs_do_filldir(desc, dirent, filldir); | |
ece0b423 | 934 | if (res < 0) |
1da177e4 | 935 | break; |
47c716cb | 936 | } while (!desc->eof); |
fccca7fc | 937 | out: |
565277f6 | 938 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
939 | if (res > 0) |
940 | res = 0; | |
aa49b4cf | 941 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
942 | dentry->d_parent->d_name.name, dentry->d_name.name, |
943 | res); | |
944 | return res; | |
1da177e4 LT |
945 | } |
946 | ||
10afec90 | 947 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 948 | { |
b84e06c5 CL |
949 | struct dentry *dentry = filp->f_path.dentry; |
950 | struct inode *inode = dentry->d_inode; | |
480c2006 | 951 | struct nfs_open_dir_context *dir_ctx = filp->private_data; |
b84e06c5 | 952 | |
6da24bc9 | 953 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
954 | dentry->d_parent->d_name.name, |
955 | dentry->d_name.name, | |
956 | offset, origin); | |
957 | ||
958 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
959 | switch (origin) { |
960 | case 1: | |
961 | offset += filp->f_pos; | |
962 | case 0: | |
963 | if (offset >= 0) | |
964 | break; | |
965 | default: | |
966 | offset = -EINVAL; | |
967 | goto out; | |
968 | } | |
969 | if (offset != filp->f_pos) { | |
970 | filp->f_pos = offset; | |
480c2006 | 971 | dir_ctx->dir_cookie = 0; |
8ef2ce3e | 972 | dir_ctx->duped = 0; |
f0dd2136 TM |
973 | } |
974 | out: | |
b84e06c5 | 975 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
976 | return offset; |
977 | } | |
978 | ||
1da177e4 LT |
979 | /* |
980 | * All directory operations under NFS are synchronous, so fsync() | |
981 | * is a dummy operation. | |
982 | */ | |
02c24a82 JB |
983 | static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end, |
984 | int datasync) | |
1da177e4 | 985 | { |
7ea80859 | 986 | struct dentry *dentry = filp->f_path.dentry; |
02c24a82 | 987 | struct inode *inode = dentry->d_inode; |
7ea80859 | 988 | |
6da24bc9 | 989 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
990 | dentry->d_parent->d_name.name, dentry->d_name.name, |
991 | datasync); | |
992 | ||
02c24a82 | 993 | mutex_lock(&inode->i_mutex); |
54917786 | 994 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
02c24a82 | 995 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
996 | return 0; |
997 | } | |
998 | ||
bfc69a45 TM |
999 | /** |
1000 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
1001 | * @dir - pointer to directory inode | |
1002 | * | |
1003 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
1004 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
1005 | * on the server that might have invalidated our dcache. | |
1006 | * | |
1007 | * The caller should be holding dir->i_lock | |
1008 | */ | |
1009 | void nfs_force_lookup_revalidate(struct inode *dir) | |
1010 | { | |
011935a0 | 1011 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
1012 | } |
1013 | ||
1da177e4 LT |
1014 | /* |
1015 | * A check for whether or not the parent directory has changed. | |
1016 | * In the case it has, we assume that the dentries are untrustworthy | |
1017 | * and may need to be looked up again. | |
1018 | */ | |
c79ba787 | 1019 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
1020 | { |
1021 | if (IS_ROOT(dentry)) | |
1022 | return 1; | |
4eec952e TM |
1023 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
1024 | return 0; | |
f2c77f4e TM |
1025 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
1026 | return 0; | |
1027 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
1028 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
1029 | return 0; | |
1030 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
1031 | return 0; | |
1032 | return 1; | |
1da177e4 LT |
1033 | } |
1034 | ||
1d6757fb TM |
1035 | /* |
1036 | * Return the intent data that applies to this particular path component | |
1037 | * | |
1038 | * Note that the current set of intents only apply to the very last | |
8aeb376c AV |
1039 | * component of the path and none of them is set before that last |
1040 | * component. | |
1d6757fb | 1041 | */ |
34286d66 NP |
1042 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, |
1043 | unsigned int mask) | |
1d6757fb | 1044 | { |
1d6757fb TM |
1045 | return nd->flags & mask; |
1046 | } | |
1047 | ||
a12802ca TM |
1048 | /* |
1049 | * Use intent information to check whether or not we're going to do | |
1050 | * an O_EXCL create using this path component. | |
1051 | */ | |
1052 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
1053 | { | |
1054 | if (NFS_PROTO(dir)->version == 2) | |
1055 | return 0; | |
3516586a | 1056 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
1057 | } |
1058 | ||
1d6757fb TM |
1059 | /* |
1060 | * Inode and filehandle revalidation for lookups. | |
1061 | * | |
1062 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
1063 | * or if the intent information indicates that we're about to open this | |
1064 | * particular file and the "nocto" mount flag is not set. | |
1065 | * | |
1066 | */ | |
1da177e4 LT |
1067 | static inline |
1068 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
1069 | { | |
1070 | struct nfs_server *server = NFS_SERVER(inode); | |
1071 | ||
36d43a43 | 1072 | if (IS_AUTOMOUNT(inode)) |
4e99a1ff | 1073 | return 0; |
1da177e4 | 1074 | if (nd != NULL) { |
1da177e4 | 1075 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 1076 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
1077 | goto out_force; |
1078 | /* This is an open(2) */ | |
1d6757fb | 1079 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
1080 | !(server->flags & NFS_MOUNT_NOCTO) && |
1081 | (S_ISREG(inode->i_mode) || | |
1082 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 1083 | goto out_force; |
4f48af45 | 1084 | return 0; |
1da177e4 LT |
1085 | } |
1086 | return nfs_revalidate_inode(server, inode); | |
1087 | out_force: | |
1088 | return __nfs_revalidate_inode(server, inode); | |
1089 | } | |
1090 | ||
1091 | /* | |
1092 | * We judge how long we want to trust negative | |
1093 | * dentries by looking at the parent inode mtime. | |
1094 | * | |
1095 | * If parent mtime has changed, we revalidate, else we wait for a | |
1096 | * period corresponding to the parent's attribute cache timeout value. | |
1097 | */ | |
1098 | static inline | |
1099 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1100 | struct nameidata *nd) | |
1101 | { | |
1da177e4 | 1102 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 1103 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 1104 | return 0; |
4eec952e TM |
1105 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1106 | return 1; | |
1da177e4 LT |
1107 | return !nfs_check_verifier(dir, dentry); |
1108 | } | |
1109 | ||
1110 | /* | |
1111 | * This is called every time the dcache has a lookup hit, | |
1112 | * and we should check whether we can really trust that | |
1113 | * lookup. | |
1114 | * | |
1115 | * NOTE! The hit can be a negative hit too, don't assume | |
1116 | * we have an inode! | |
1117 | * | |
1118 | * If the parent directory is seen to have changed, we throw out the | |
1119 | * cached dentry and do a new lookup. | |
1120 | */ | |
34286d66 | 1121 | static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1122 | { |
1123 | struct inode *dir; | |
1124 | struct inode *inode; | |
1125 | struct dentry *parent; | |
e1fb4d05 TM |
1126 | struct nfs_fh *fhandle = NULL; |
1127 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1128 | int error; |
1da177e4 | 1129 | |
34286d66 NP |
1130 | if (nd->flags & LOOKUP_RCU) |
1131 | return -ECHILD; | |
1132 | ||
1da177e4 | 1133 | parent = dget_parent(dentry); |
1da177e4 | 1134 | dir = parent->d_inode; |
91d5b470 | 1135 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1136 | inode = dentry->d_inode; |
1137 | ||
1138 | if (!inode) { | |
1139 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1140 | goto out_bad; | |
d69ee9b8 | 1141 | goto out_valid_noent; |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1145 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1146 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1147 | dentry->d_name.name); |
1da177e4 LT |
1148 | goto out_bad; |
1149 | } | |
1150 | ||
15860ab1 TM |
1151 | if (nfs_have_delegation(inode, FMODE_READ)) |
1152 | goto out_set_verifier; | |
1153 | ||
1da177e4 | 1154 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1155 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1156 | if (nfs_lookup_verify_inode(inode, nd)) |
1157 | goto out_zap_parent; | |
1158 | goto out_valid; | |
1159 | } | |
1160 | ||
1161 | if (NFS_STALE(inode)) | |
1162 | goto out_bad; | |
1163 | ||
e1fb4d05 TM |
1164 | error = -ENOMEM; |
1165 | fhandle = nfs_alloc_fhandle(); | |
1166 | fattr = nfs_alloc_fattr(); | |
1167 | if (fhandle == NULL || fattr == NULL) | |
1168 | goto out_error; | |
1169 | ||
80a16b21 | 1170 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1171 | if (error) |
1172 | goto out_bad; | |
e1fb4d05 | 1173 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1174 | goto out_bad; |
e1fb4d05 | 1175 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1176 | goto out_bad; |
1177 | ||
e1fb4d05 TM |
1178 | nfs_free_fattr(fattr); |
1179 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1180 | out_set_verifier: |
cf8ba45e | 1181 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1182 | out_valid: |
d69ee9b8 TM |
1183 | /* Success: notify readdir to use READDIRPLUS */ |
1184 | nfs_advise_use_readdirplus(dir); | |
1185 | out_valid_noent: | |
1da177e4 | 1186 | dput(parent); |
1e7cb3dc | 1187 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1188 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1189 | dentry->d_name.name); |
1da177e4 LT |
1190 | return 1; |
1191 | out_zap_parent: | |
1192 | nfs_zap_caches(dir); | |
1193 | out_bad: | |
a1643a92 | 1194 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1195 | if (inode && S_ISDIR(inode->i_mode)) { |
1196 | /* Purge readdir caches. */ | |
1197 | nfs_zap_caches(inode); | |
1198 | /* If we have submounts, don't unhash ! */ | |
1199 | if (have_submounts(dentry)) | |
1200 | goto out_valid; | |
d9e80b7d AV |
1201 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1202 | goto out_valid; | |
1da177e4 LT |
1203 | shrink_dcache_parent(dentry); |
1204 | } | |
1205 | d_drop(dentry); | |
e1fb4d05 TM |
1206 | nfs_free_fattr(fattr); |
1207 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1208 | dput(parent); |
1e7cb3dc | 1209 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1210 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1211 | dentry->d_name.name); |
1da177e4 | 1212 | return 0; |
e1fb4d05 TM |
1213 | out_error: |
1214 | nfs_free_fattr(fattr); | |
1215 | nfs_free_fhandle(fhandle); | |
1216 | dput(parent); | |
1217 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1218 | __func__, dentry->d_parent->d_name.name, | |
1219 | dentry->d_name.name, error); | |
1220 | return error; | |
1da177e4 LT |
1221 | } |
1222 | ||
1223 | /* | |
1224 | * This is called from dput() when d_count is going to 0. | |
1225 | */ | |
fe15ce44 | 1226 | static int nfs_dentry_delete(const struct dentry *dentry) |
1da177e4 LT |
1227 | { |
1228 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1229 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1230 | dentry->d_flags); | |
1231 | ||
77f11192 TM |
1232 | /* Unhash any dentry with a stale inode */ |
1233 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1234 | return 1; | |
1235 | ||
1da177e4 LT |
1236 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1237 | /* Unhash it, so that ->d_iput() would be called */ | |
1238 | return 1; | |
1239 | } | |
1240 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1241 | /* Unhash it, so that ancestors of killed async unlink | |
1242 | * files will be cleaned up during umount */ | |
1243 | return 1; | |
1244 | } | |
1245 | return 0; | |
1246 | ||
1247 | } | |
1248 | ||
1b83d707 TM |
1249 | static void nfs_drop_nlink(struct inode *inode) |
1250 | { | |
1251 | spin_lock(&inode->i_lock); | |
1252 | if (inode->i_nlink > 0) | |
1253 | drop_nlink(inode); | |
1254 | spin_unlock(&inode->i_lock); | |
1255 | } | |
1256 | ||
1da177e4 LT |
1257 | /* |
1258 | * Called when the dentry loses inode. | |
1259 | * We use it to clean up silly-renamed files. | |
1260 | */ | |
1261 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1262 | { | |
83672d39 NB |
1263 | if (S_ISDIR(inode->i_mode)) |
1264 | /* drop any readdir cache as it could easily be old */ | |
1265 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1266 | ||
1da177e4 | 1267 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1268 | drop_nlink(inode); |
e4eff1a6 | 1269 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1270 | } |
1da177e4 LT |
1271 | iput(inode); |
1272 | } | |
1273 | ||
b1942c5f AV |
1274 | static void nfs_d_release(struct dentry *dentry) |
1275 | { | |
1276 | /* free cached devname value, if it survived that far */ | |
1277 | if (unlikely(dentry->d_fsdata)) { | |
1278 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1279 | WARN_ON(1); | |
1280 | else | |
1281 | kfree(dentry->d_fsdata); | |
1282 | } | |
1283 | } | |
1284 | ||
f786aa90 | 1285 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1286 | .d_revalidate = nfs_lookup_revalidate, |
1287 | .d_delete = nfs_dentry_delete, | |
1288 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1289 | .d_automount = nfs_d_automount, |
b1942c5f | 1290 | .d_release = nfs_d_release, |
1da177e4 LT |
1291 | }; |
1292 | ||
1da177e4 LT |
1293 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1294 | { | |
1295 | struct dentry *res; | |
565277f6 | 1296 | struct dentry *parent; |
1da177e4 | 1297 | struct inode *inode = NULL; |
e1fb4d05 TM |
1298 | struct nfs_fh *fhandle = NULL; |
1299 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1300 | int error; |
1da177e4 LT |
1301 | |
1302 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1303 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1304 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1305 | |
1306 | res = ERR_PTR(-ENAMETOOLONG); | |
1307 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1308 | goto out; | |
1309 | ||
fd684071 TM |
1310 | /* |
1311 | * If we're doing an exclusive create, optimize away the lookup | |
1312 | * but don't hash the dentry. | |
1313 | */ | |
1314 | if (nfs_is_exclusive_create(dir, nd)) { | |
1315 | d_instantiate(dentry, NULL); | |
1316 | res = NULL; | |
fc0f684c | 1317 | goto out; |
fd684071 | 1318 | } |
1da177e4 | 1319 | |
e1fb4d05 TM |
1320 | res = ERR_PTR(-ENOMEM); |
1321 | fhandle = nfs_alloc_fhandle(); | |
1322 | fattr = nfs_alloc_fattr(); | |
1323 | if (fhandle == NULL || fattr == NULL) | |
1324 | goto out; | |
1325 | ||
565277f6 TM |
1326 | parent = dentry->d_parent; |
1327 | /* Protect against concurrent sillydeletes */ | |
1328 | nfs_block_sillyrename(parent); | |
80a16b21 | 1329 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1330 | if (error == -ENOENT) |
1331 | goto no_entry; | |
1332 | if (error < 0) { | |
1333 | res = ERR_PTR(error); | |
565277f6 | 1334 | goto out_unblock_sillyrename; |
1da177e4 | 1335 | } |
e1fb4d05 | 1336 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
bf0c84f1 | 1337 | res = ERR_CAST(inode); |
03f28e3a | 1338 | if (IS_ERR(res)) |
565277f6 | 1339 | goto out_unblock_sillyrename; |
54ceac45 | 1340 | |
d69ee9b8 TM |
1341 | /* Success: notify readdir to use READDIRPLUS */ |
1342 | nfs_advise_use_readdirplus(dir); | |
1343 | ||
1da177e4 | 1344 | no_entry: |
54ceac45 | 1345 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1346 | if (res != NULL) { |
1347 | if (IS_ERR(res)) | |
565277f6 | 1348 | goto out_unblock_sillyrename; |
1da177e4 | 1349 | dentry = res; |
9eaef27b | 1350 | } |
1da177e4 | 1351 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1352 | out_unblock_sillyrename: |
1353 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1354 | out: |
e1fb4d05 TM |
1355 | nfs_free_fattr(fattr); |
1356 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1357 | return res; |
1358 | } | |
1359 | ||
1360 | #ifdef CONFIG_NFS_V4 | |
0ef97dcf | 1361 | static int nfs4_lookup_revalidate(struct dentry *, struct nameidata *); |
1da177e4 | 1362 | |
f786aa90 | 1363 | const struct dentry_operations nfs4_dentry_operations = { |
0ef97dcf | 1364 | .d_revalidate = nfs4_lookup_revalidate, |
1da177e4 LT |
1365 | .d_delete = nfs_dentry_delete, |
1366 | .d_iput = nfs_dentry_iput, | |
36d43a43 | 1367 | .d_automount = nfs_d_automount, |
b1942c5f | 1368 | .d_release = nfs_d_release, |
1da177e4 LT |
1369 | }; |
1370 | ||
1d6757fb TM |
1371 | /* |
1372 | * Use intent information to determine whether we need to substitute | |
1373 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1374 | */ | |
5584c306 | 1375 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1376 | { |
1d6757fb | 1377 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1378 | return 0; |
1379 | /* NFS does not (yet) have a stateful open for directories */ | |
1380 | if (nd->flags & LOOKUP_DIRECTORY) | |
1381 | return 0; | |
1382 | /* Are we trying to write to a read only partition? */ | |
2c463e95 | 1383 | if (__mnt_is_readonly(nd->path.mnt) && |
8a5e929d | 1384 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE))) |
1da177e4 LT |
1385 | return 0; |
1386 | return 1; | |
1387 | } | |
1388 | ||
8a5e929d AV |
1389 | static fmode_t flags_to_mode(int flags) |
1390 | { | |
1391 | fmode_t res = (__force fmode_t)flags & FMODE_EXEC; | |
1392 | if ((flags & O_ACCMODE) != O_WRONLY) | |
1393 | res |= FMODE_READ; | |
1394 | if ((flags & O_ACCMODE) != O_RDONLY) | |
1395 | res |= FMODE_WRITE; | |
1396 | return res; | |
1397 | } | |
1398 | ||
51141598 | 1399 | static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags) |
cd9a1c0e | 1400 | { |
5ede7b1c | 1401 | return alloc_nfs_open_context(dentry, flags_to_mode(open_flags)); |
cd9a1c0e TM |
1402 | } |
1403 | ||
1404 | static int do_open(struct inode *inode, struct file *filp) | |
1405 | { | |
1406 | nfs_fscache_set_inode_cookie(inode, filp); | |
1407 | return 0; | |
1408 | } | |
1409 | ||
0dd2b474 MS |
1410 | static struct file *nfs_finish_open(struct nfs_open_context *ctx, |
1411 | struct dentry *dentry, | |
1412 | struct opendata *od, unsigned open_flags) | |
cd9a1c0e TM |
1413 | { |
1414 | struct file *filp; | |
0dd2b474 MS |
1415 | int err; |
1416 | ||
1417 | if (ctx->dentry != dentry) { | |
1418 | dput(ctx->dentry); | |
1419 | ctx->dentry = dget(dentry); | |
1420 | } | |
cd9a1c0e TM |
1421 | |
1422 | /* If the open_intent is for execute, we have an extra check to make */ | |
1423 | if (ctx->mode & FMODE_EXEC) { | |
0dd2b474 MS |
1424 | err = nfs_may_open(dentry->d_inode, ctx->cred, open_flags); |
1425 | if (err < 0) { | |
1426 | filp = ERR_PTR(err); | |
cd9a1c0e | 1427 | goto out; |
0dd2b474 | 1428 | } |
cd9a1c0e | 1429 | } |
0dd2b474 MS |
1430 | |
1431 | filp = finish_open(od, dentry, do_open); | |
1432 | if (!IS_ERR(filp)) | |
cd9a1c0e | 1433 | nfs_file_set_open_context(filp, ctx); |
0dd2b474 | 1434 | |
cd9a1c0e TM |
1435 | out: |
1436 | put_nfs_open_context(ctx); | |
0dd2b474 | 1437 | return filp; |
cd9a1c0e TM |
1438 | } |
1439 | ||
0dd2b474 MS |
1440 | static struct file *nfs_atomic_open(struct inode *dir, struct dentry *dentry, |
1441 | struct opendata *od, unsigned open_flags, | |
1442 | umode_t mode, bool *created) | |
1da177e4 | 1443 | { |
cd9a1c0e | 1444 | struct nfs_open_context *ctx; |
0dd2b474 MS |
1445 | struct dentry *res; |
1446 | struct iattr attr = { .ia_valid = ATTR_OPEN }; | |
f46e0bd3 | 1447 | struct inode *inode; |
0dd2b474 | 1448 | struct file *filp; |
898f635c | 1449 | int err; |
1da177e4 | 1450 | |
0dd2b474 MS |
1451 | /* Expect a negative dentry */ |
1452 | BUG_ON(dentry->d_inode); | |
1453 | ||
1454 | dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n", | |
1e7cb3dc CL |
1455 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); |
1456 | ||
0dd2b474 MS |
1457 | /* NFS only supports OPEN on regular files */ |
1458 | if ((open_flags & O_DIRECTORY)) { | |
1459 | err = -ENOENT; | |
1460 | if (!d_unhashed(dentry)) { | |
1461 | /* | |
1462 | * Hashed negative dentry with O_DIRECTORY: dentry was | |
1463 | * revalidated and is fine, no need to perform lookup | |
1464 | * again | |
1465 | */ | |
1466 | goto out_err; | |
1467 | } | |
1da177e4 | 1468 | goto no_open; |
02a913a7 | 1469 | } |
1da177e4 | 1470 | |
0dd2b474 MS |
1471 | err = -ENAMETOOLONG; |
1472 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1473 | goto out_err; | |
cd9a1c0e | 1474 | |
0dd2b474 | 1475 | if (open_flags & O_CREAT) { |
536e43d1 | 1476 | attr.ia_valid |= ATTR_MODE; |
0dd2b474 MS |
1477 | attr.ia_mode = mode & ~current_umask(); |
1478 | } | |
536e43d1 TM |
1479 | if (open_flags & O_TRUNC) { |
1480 | attr.ia_valid |= ATTR_SIZE; | |
1481 | attr.ia_size = 0; | |
cd9a1c0e TM |
1482 | } |
1483 | ||
0dd2b474 MS |
1484 | ctx = create_nfs_open_context(dentry, open_flags); |
1485 | err = PTR_ERR(ctx); | |
1486 | if (IS_ERR(ctx)) | |
1487 | goto out_err; | |
1488 | ||
f46e0bd3 | 1489 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1490 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
0dd2b474 | 1491 | d_drop(dentry); |
f46e0bd3 TM |
1492 | if (IS_ERR(inode)) { |
1493 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1494 | put_nfs_open_context(ctx); |
0dd2b474 MS |
1495 | err = PTR_ERR(inode); |
1496 | switch (err) { | |
1497 | case -ENOENT: | |
1498 | d_add(dentry, NULL); | |
1499 | break; | |
1500 | case -EISDIR: | |
1501 | case -ENOTDIR: | |
1502 | goto no_open; | |
1503 | case -ELOOP: | |
1504 | if (!(open_flags & O_NOFOLLOW)) | |
6f926b5b | 1505 | goto no_open; |
0dd2b474 | 1506 | break; |
1da177e4 | 1507 | /* case -EINVAL: */ |
0dd2b474 MS |
1508 | default: |
1509 | break; | |
1da177e4 | 1510 | } |
0dd2b474 | 1511 | goto out_err; |
cd9a1c0e | 1512 | } |
f46e0bd3 | 1513 | res = d_add_unique(dentry, inode); |
0dd2b474 | 1514 | if (res != NULL) |
1da177e4 | 1515 | dentry = res; |
0dd2b474 MS |
1516 | |
1517 | nfs_unblock_sillyrename(dentry->d_parent); | |
f46e0bd3 | 1518 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
0dd2b474 MS |
1519 | |
1520 | filp = nfs_finish_open(ctx, dentry, od, open_flags); | |
1521 | ||
1522 | dput(res); | |
1523 | return filp; | |
1524 | ||
1525 | out_err: | |
1526 | return ERR_PTR(err); | |
1527 | ||
1da177e4 | 1528 | no_open: |
0dd2b474 MS |
1529 | res = nfs_lookup(dir, dentry, NULL); |
1530 | err = PTR_ERR(res); | |
1531 | if (IS_ERR(res)) | |
1532 | goto out_err; | |
1533 | ||
1534 | finish_no_open(od, res); | |
1535 | return NULL; | |
1da177e4 LT |
1536 | } |
1537 | ||
0ef97dcf | 1538 | static int nfs4_lookup_revalidate(struct dentry *dentry, struct nameidata *nd) |
1da177e4 LT |
1539 | { |
1540 | struct dentry *parent = NULL; | |
657e94b6 | 1541 | struct inode *inode; |
1da177e4 | 1542 | struct inode *dir; |
1da177e4 LT |
1543 | int openflags, ret = 0; |
1544 | ||
657e94b6 NP |
1545 | if (nd->flags & LOOKUP_RCU) |
1546 | return -ECHILD; | |
1547 | ||
1548 | inode = dentry->d_inode; | |
1f063d2c | 1549 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1550 | goto no_open; |
2b484297 | 1551 | |
1da177e4 LT |
1552 | parent = dget_parent(dentry); |
1553 | dir = parent->d_inode; | |
2b484297 | 1554 | |
1da177e4 LT |
1555 | /* We can't create new files in nfs_open_revalidate(), so we |
1556 | * optimize away revalidation of negative dentries. | |
1557 | */ | |
216d5d06 TM |
1558 | if (inode == NULL) { |
1559 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1560 | ret = 1; | |
1da177e4 | 1561 | goto out; |
216d5d06 TM |
1562 | } |
1563 | ||
1da177e4 LT |
1564 | /* NFS only supports OPEN on regular files */ |
1565 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1566 | goto no_open_dput; |
1da177e4 LT |
1567 | openflags = nd->intent.open.flags; |
1568 | /* We cannot do exclusive creation on a positive dentry */ | |
1569 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1570 | goto no_open_dput; |
1da177e4 | 1571 | |
0ef97dcf MS |
1572 | /* Let f_op->open() actually open (and revalidate) the file */ |
1573 | ret = 1; | |
536e43d1 | 1574 | |
1da177e4 LT |
1575 | out: |
1576 | dput(parent); | |
1da177e4 | 1577 | return ret; |
535918f1 | 1578 | |
5584c306 | 1579 | no_open_dput: |
1da177e4 | 1580 | dput(parent); |
5584c306 | 1581 | no_open: |
1da177e4 LT |
1582 | return nfs_lookup_revalidate(dentry, nd); |
1583 | } | |
c0204fd2 | 1584 | |
0dd2b474 MS |
1585 | static int nfs4_create(struct inode *dir, struct dentry *dentry, |
1586 | umode_t mode, struct nameidata *nd) | |
c0204fd2 TM |
1587 | { |
1588 | struct nfs_open_context *ctx = NULL; | |
1589 | struct iattr attr; | |
1590 | int error; | |
8a5e929d | 1591 | int open_flags = O_CREAT|O_EXCL; |
c0204fd2 TM |
1592 | |
1593 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1594 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1595 | ||
1596 | attr.ia_mode = mode; | |
1597 | attr.ia_valid = ATTR_MODE; | |
1598 | ||
dd7dd556 | 1599 | if (nd) |
c0204fd2 TM |
1600 | open_flags = nd->intent.open.flags; |
1601 | ||
f7c85868 AV |
1602 | ctx = create_nfs_open_context(dentry, open_flags); |
1603 | error = PTR_ERR(ctx); | |
1604 | if (IS_ERR(ctx)) | |
1605 | goto out_err_drop; | |
c0204fd2 TM |
1606 | |
1607 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1608 | if (error != 0) | |
1609 | goto out_put_ctx; | |
0dd2b474 MS |
1610 | |
1611 | put_nfs_open_context(ctx); | |
1612 | ||
c0204fd2 TM |
1613 | return 0; |
1614 | out_put_ctx: | |
f7c85868 | 1615 | put_nfs_open_context(ctx); |
898f635c | 1616 | out_err_drop: |
c0204fd2 TM |
1617 | d_drop(dentry); |
1618 | return error; | |
1619 | } | |
1620 | ||
1da177e4 LT |
1621 | #endif /* CONFIG_NFSV4 */ |
1622 | ||
1da177e4 LT |
1623 | /* |
1624 | * Code common to create, mkdir, and mknod. | |
1625 | */ | |
1626 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1627 | struct nfs_fattr *fattr) | |
1628 | { | |
fab728e1 TM |
1629 | struct dentry *parent = dget_parent(dentry); |
1630 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1631 | struct inode *inode; |
1632 | int error = -EACCES; | |
1633 | ||
fab728e1 TM |
1634 | d_drop(dentry); |
1635 | ||
1da177e4 LT |
1636 | /* We may have been initialized further down */ |
1637 | if (dentry->d_inode) | |
fab728e1 | 1638 | goto out; |
1da177e4 | 1639 | if (fhandle->size == 0) { |
80a16b21 | 1640 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 | 1641 | if (error) |
fab728e1 | 1642 | goto out_error; |
1da177e4 | 1643 | } |
5724ab37 | 1644 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1645 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1646 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1647 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1648 | if (error < 0) |
fab728e1 | 1649 | goto out_error; |
1da177e4 | 1650 | } |
1da177e4 | 1651 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1652 | error = PTR_ERR(inode); |
1653 | if (IS_ERR(inode)) | |
fab728e1 TM |
1654 | goto out_error; |
1655 | d_add(dentry, inode); | |
1656 | out: | |
1657 | dput(parent); | |
1da177e4 | 1658 | return 0; |
fab728e1 TM |
1659 | out_error: |
1660 | nfs_mark_for_revalidate(dir); | |
1661 | dput(parent); | |
1662 | return error; | |
1da177e4 LT |
1663 | } |
1664 | ||
1665 | /* | |
1666 | * Following a failed create operation, we drop the dentry rather | |
1667 | * than retain a negative dentry. This avoids a problem in the event | |
1668 | * that the operation succeeded on the server, but an error in the | |
1669 | * reply path made it appear to have failed. | |
1670 | */ | |
4acdaf27 AV |
1671 | static int nfs_create(struct inode *dir, struct dentry *dentry, |
1672 | umode_t mode, struct nameidata *nd) | |
1da177e4 LT |
1673 | { |
1674 | struct iattr attr; | |
1675 | int error; | |
8a5e929d | 1676 | int open_flags = O_CREAT|O_EXCL; |
1da177e4 | 1677 | |
1e7cb3dc CL |
1678 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1679 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1680 | |
1681 | attr.ia_mode = mode; | |
1682 | attr.ia_valid = ATTR_MODE; | |
1683 | ||
dd7dd556 | 1684 | if (nd) |
8a0eebf6 TM |
1685 | open_flags = nd->intent.open.flags; |
1686 | ||
1687 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL); | |
1da177e4 LT |
1688 | if (error != 0) |
1689 | goto out_err; | |
1da177e4 LT |
1690 | return 0; |
1691 | out_err: | |
1da177e4 LT |
1692 | d_drop(dentry); |
1693 | return error; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * See comments for nfs_proc_create regarding failed operations. | |
1698 | */ | |
1699 | static int | |
1a67aafb | 1700 | nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev) |
1da177e4 LT |
1701 | { |
1702 | struct iattr attr; | |
1703 | int status; | |
1704 | ||
1e7cb3dc CL |
1705 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1706 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1707 | |
1708 | if (!new_valid_dev(rdev)) | |
1709 | return -EINVAL; | |
1710 | ||
1711 | attr.ia_mode = mode; | |
1712 | attr.ia_valid = ATTR_MODE; | |
1713 | ||
1da177e4 | 1714 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1715 | if (status != 0) |
1716 | goto out_err; | |
1da177e4 LT |
1717 | return 0; |
1718 | out_err: | |
1da177e4 LT |
1719 | d_drop(dentry); |
1720 | return status; | |
1721 | } | |
1722 | ||
1723 | /* | |
1724 | * See comments for nfs_proc_create regarding failed operations. | |
1725 | */ | |
18bb1db3 | 1726 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
1727 | { |
1728 | struct iattr attr; | |
1729 | int error; | |
1730 | ||
1e7cb3dc CL |
1731 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1732 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1733 | |
1734 | attr.ia_valid = ATTR_MODE; | |
1735 | attr.ia_mode = mode | S_IFDIR; | |
1736 | ||
1da177e4 | 1737 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1738 | if (error != 0) |
1739 | goto out_err; | |
1da177e4 LT |
1740 | return 0; |
1741 | out_err: | |
1742 | d_drop(dentry); | |
1da177e4 LT |
1743 | return error; |
1744 | } | |
1745 | ||
d45b9d8b TM |
1746 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1747 | { | |
1748 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1749 | d_delete(dentry); | |
1750 | } | |
1751 | ||
1da177e4 LT |
1752 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1753 | { | |
1754 | int error; | |
1755 | ||
1e7cb3dc CL |
1756 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1757 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1758 | |
1da177e4 LT |
1759 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1760 | /* Ensure the VFS deletes this inode */ | |
1761 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1762 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1763 | else if (error == -ENOENT) |
1764 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1765 | |
1766 | return error; | |
1767 | } | |
1768 | ||
1da177e4 LT |
1769 | /* |
1770 | * Remove a file after making sure there are no pending writes, | |
1771 | * and after checking that the file has only one user. | |
1772 | * | |
1773 | * We invalidate the attribute cache and free the inode prior to the operation | |
1774 | * to avoid possible races if the server reuses the inode. | |
1775 | */ | |
1776 | static int nfs_safe_remove(struct dentry *dentry) | |
1777 | { | |
1778 | struct inode *dir = dentry->d_parent->d_inode; | |
1779 | struct inode *inode = dentry->d_inode; | |
1780 | int error = -EBUSY; | |
1781 | ||
1782 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1783 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1784 | ||
1785 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1786 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1787 | error = 0; | |
1788 | goto out; | |
1789 | } | |
1790 | ||
1da177e4 | 1791 | if (inode != NULL) { |
cae7a073 | 1792 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1793 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1794 | /* The VFS may want to delete this inode */ | |
1795 | if (error == 0) | |
1b83d707 | 1796 | nfs_drop_nlink(inode); |
5ba7cc48 | 1797 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1798 | } else |
1799 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1800 | if (error == -ENOENT) |
1801 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1802 | out: |
1803 | return error; | |
1804 | } | |
1805 | ||
1806 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1807 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1808 | * | |
1809 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1810 | */ | |
1811 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1812 | { | |
1813 | int error; | |
1814 | int need_rehash = 0; | |
1815 | ||
1816 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1817 | dir->i_ino, dentry->d_name.name); | |
1818 | ||
1da177e4 | 1819 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 1820 | if (dentry->d_count > 1) { |
1da177e4 | 1821 | spin_unlock(&dentry->d_lock); |
ccfeb506 TM |
1822 | /* Start asynchronous writeout of the inode */ |
1823 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1824 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1825 | return error; |
1826 | } | |
1827 | if (!d_unhashed(dentry)) { | |
1828 | __d_drop(dentry); | |
1829 | need_rehash = 1; | |
1830 | } | |
1831 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1832 | error = nfs_safe_remove(dentry); |
d45b9d8b | 1833 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1834 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1835 | } else if (need_rehash) | |
1836 | d_rehash(dentry); | |
1da177e4 LT |
1837 | return error; |
1838 | } | |
1839 | ||
873101b3 CL |
1840 | /* |
1841 | * To create a symbolic link, most file systems instantiate a new inode, | |
1842 | * add a page to it containing the path, then write it out to the disk | |
1843 | * using prepare_write/commit_write. | |
1844 | * | |
1845 | * Unfortunately the NFS client can't create the in-core inode first | |
1846 | * because it needs a file handle to create an in-core inode (see | |
1847 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1848 | * symlink request has completed on the server. | |
1849 | * | |
1850 | * So instead we allocate a raw page, copy the symname into it, then do | |
1851 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1852 | * now have a new file handle and can instantiate an in-core NFS inode | |
1853 | * and move the raw page into its mapping. | |
1854 | */ | |
1855 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1856 | { |
873101b3 CL |
1857 | struct pagevec lru_pvec; |
1858 | struct page *page; | |
1859 | char *kaddr; | |
1da177e4 | 1860 | struct iattr attr; |
873101b3 | 1861 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1862 | int error; |
1863 | ||
1864 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1865 | dir->i_ino, dentry->d_name.name, symname); | |
1866 | ||
873101b3 CL |
1867 | if (pathlen > PAGE_SIZE) |
1868 | return -ENAMETOOLONG; | |
1da177e4 | 1869 | |
873101b3 CL |
1870 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1871 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1872 | |
83d93f22 | 1873 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1874 | if (!page) |
873101b3 | 1875 | return -ENOMEM; |
873101b3 | 1876 | |
2b86ce2d | 1877 | kaddr = kmap_atomic(page); |
873101b3 CL |
1878 | memcpy(kaddr, symname, pathlen); |
1879 | if (pathlen < PAGE_SIZE) | |
1880 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
2b86ce2d | 1881 | kunmap_atomic(kaddr); |
873101b3 | 1882 | |
94a6d753 | 1883 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1884 | if (error != 0) { |
1885 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1886 | dir->i_sb->s_id, dir->i_ino, | |
1887 | dentry->d_name.name, symname, error); | |
1da177e4 | 1888 | d_drop(dentry); |
873101b3 | 1889 | __free_page(page); |
873101b3 CL |
1890 | return error; |
1891 | } | |
1892 | ||
1893 | /* | |
1894 | * No big deal if we can't add this page to the page cache here. | |
1895 | * READLINK will get the missing page from the server if needed. | |
1896 | */ | |
1897 | pagevec_init(&lru_pvec, 0); | |
1898 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1899 | GFP_KERNEL)) { | |
39cf8a13 | 1900 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1901 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1902 | SetPageUptodate(page); |
1903 | unlock_page(page); | |
1904 | } else | |
1905 | __free_page(page); | |
1906 | ||
873101b3 | 1907 | return 0; |
1da177e4 LT |
1908 | } |
1909 | ||
1910 | static int | |
1911 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1912 | { | |
1913 | struct inode *inode = old_dentry->d_inode; | |
1914 | int error; | |
1915 | ||
1916 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1917 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1918 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1919 | ||
9a3936aa TM |
1920 | nfs_inode_return_delegation(inode); |
1921 | ||
9697d234 | 1922 | d_drop(dentry); |
1da177e4 | 1923 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1924 | if (error == 0) { |
7de9c6ee | 1925 | ihold(inode); |
9697d234 | 1926 | d_add(dentry, inode); |
cf809556 | 1927 | } |
1da177e4 LT |
1928 | return error; |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * RENAME | |
1933 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1934 | * different file handle for the same inode after a rename (e.g. when | |
1935 | * moving to a different directory). A fail-safe method to do so would | |
1936 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1937 | * rename the old file using the sillyrename stuff. This way, the original | |
1938 | * file in old_dir will go away when the last process iput()s the inode. | |
1939 | * | |
1940 | * FIXED. | |
1941 | * | |
1942 | * It actually works quite well. One needs to have the possibility for | |
1943 | * at least one ".nfs..." file in each directory the file ever gets | |
1944 | * moved or linked to which happens automagically with the new | |
1945 | * implementation that only depends on the dcache stuff instead of | |
1946 | * using the inode layer | |
1947 | * | |
1948 | * Unfortunately, things are a little more complicated than indicated | |
1949 | * above. For a cross-directory move, we want to make sure we can get | |
1950 | * rid of the old inode after the operation. This means there must be | |
1951 | * no pending writes (if it's a file), and the use count must be 1. | |
1952 | * If these conditions are met, we can drop the dentries before doing | |
1953 | * the rename. | |
1954 | */ | |
1955 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1956 | struct inode *new_dir, struct dentry *new_dentry) | |
1957 | { | |
1958 | struct inode *old_inode = old_dentry->d_inode; | |
1959 | struct inode *new_inode = new_dentry->d_inode; | |
1960 | struct dentry *dentry = NULL, *rehash = NULL; | |
1961 | int error = -EBUSY; | |
1962 | ||
1da177e4 LT |
1963 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1964 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1965 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
b7ab39f6 | 1966 | new_dentry->d_count); |
1da177e4 LT |
1967 | |
1968 | /* | |
28f79a1a MS |
1969 | * For non-directories, check whether the target is busy and if so, |
1970 | * make a copy of the dentry and then do a silly-rename. If the | |
1971 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1972 | * the new target. | |
1da177e4 | 1973 | */ |
27226104 MS |
1974 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1975 | /* | |
1976 | * To prevent any new references to the target during the | |
1977 | * rename, we unhash the dentry in advance. | |
1978 | */ | |
1979 | if (!d_unhashed(new_dentry)) { | |
1980 | d_drop(new_dentry); | |
1981 | rehash = new_dentry; | |
1982 | } | |
1da177e4 | 1983 | |
b7ab39f6 | 1984 | if (new_dentry->d_count > 2) { |
27226104 MS |
1985 | int err; |
1986 | ||
1987 | /* copy the target dentry's name */ | |
1988 | dentry = d_alloc(new_dentry->d_parent, | |
1989 | &new_dentry->d_name); | |
1990 | if (!dentry) | |
1991 | goto out; | |
1992 | ||
1993 | /* silly-rename the existing target ... */ | |
1994 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1995 | if (err) |
27226104 | 1996 | goto out; |
24e93025 MS |
1997 | |
1998 | new_dentry = dentry; | |
56335936 | 1999 | rehash = NULL; |
24e93025 | 2000 | new_inode = NULL; |
27226104 | 2001 | } |
b1e4adf4 | 2002 | } |
1da177e4 | 2003 | |
cae7a073 | 2004 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 2005 | if (new_inode != NULL) |
24174119 | 2006 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 2007 | |
1da177e4 LT |
2008 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
2009 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 2010 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
2011 | out: |
2012 | if (rehash) | |
2013 | d_rehash(rehash); | |
2014 | if (!error) { | |
b1e4adf4 TM |
2015 | if (new_inode != NULL) |
2016 | nfs_drop_nlink(new_inode); | |
349457cc | 2017 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
2018 | nfs_set_verifier(new_dentry, |
2019 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
2020 | } else if (error == -ENOENT) |
2021 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
2022 | |
2023 | /* new dentry created? */ | |
2024 | if (dentry) | |
2025 | dput(dentry); | |
1da177e4 LT |
2026 | return error; |
2027 | } | |
2028 | ||
cfcea3e8 TM |
2029 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
2030 | static LIST_HEAD(nfs_access_lru_list); | |
2031 | static atomic_long_t nfs_access_nr_entries; | |
2032 | ||
1c3c07e9 TM |
2033 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
2034 | { | |
2035 | put_rpccred(entry->cred); | |
2036 | kfree(entry); | |
cfcea3e8 TM |
2037 | smp_mb__before_atomic_dec(); |
2038 | atomic_long_dec(&nfs_access_nr_entries); | |
2039 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
2040 | } |
2041 | ||
1a81bb8a TM |
2042 | static void nfs_access_free_list(struct list_head *head) |
2043 | { | |
2044 | struct nfs_access_entry *cache; | |
2045 | ||
2046 | while (!list_empty(head)) { | |
2047 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
2048 | list_del(&cache->lru); | |
2049 | nfs_access_free_entry(cache); | |
2050 | } | |
2051 | } | |
2052 | ||
1495f230 YH |
2053 | int nfs_access_cache_shrinker(struct shrinker *shrink, |
2054 | struct shrink_control *sc) | |
979df72e TM |
2055 | { |
2056 | LIST_HEAD(head); | |
aa510da5 | 2057 | struct nfs_inode *nfsi, *next; |
979df72e | 2058 | struct nfs_access_entry *cache; |
1495f230 YH |
2059 | int nr_to_scan = sc->nr_to_scan; |
2060 | gfp_t gfp_mask = sc->gfp_mask; | |
979df72e | 2061 | |
61d5eb29 TM |
2062 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
2063 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 2064 | |
a50f7951 | 2065 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 2066 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
2067 | struct inode *inode; |
2068 | ||
2069 | if (nr_to_scan-- == 0) | |
2070 | break; | |
9c7e7e23 | 2071 | inode = &nfsi->vfs_inode; |
979df72e TM |
2072 | spin_lock(&inode->i_lock); |
2073 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
2074 | goto remove_lru_entry; | |
2075 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
2076 | struct nfs_access_entry, lru); | |
2077 | list_move(&cache->lru, &head); | |
2078 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
2079 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
2080 | list_move_tail(&nfsi->access_cache_inode_lru, | |
2081 | &nfs_access_lru_list); | |
2082 | else { | |
2083 | remove_lru_entry: | |
2084 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 2085 | smp_mb__before_clear_bit(); |
979df72e | 2086 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 2087 | smp_mb__after_clear_bit(); |
979df72e | 2088 | } |
59844a9b | 2089 | spin_unlock(&inode->i_lock); |
979df72e TM |
2090 | } |
2091 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2092 | nfs_access_free_list(&head); |
979df72e TM |
2093 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
2094 | } | |
2095 | ||
1a81bb8a | 2096 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2097 | { |
1c3c07e9 | 2098 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2099 | struct rb_node *n; |
1c3c07e9 TM |
2100 | struct nfs_access_entry *entry; |
2101 | ||
2102 | /* Unhook entries from the cache */ | |
2103 | while ((n = rb_first(root_node)) != NULL) { | |
2104 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2105 | rb_erase(n, root_node); | |
1a81bb8a | 2106 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2107 | } |
2108 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2109 | } |
2110 | ||
1c3c07e9 | 2111 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2112 | { |
1a81bb8a TM |
2113 | LIST_HEAD(head); |
2114 | ||
2115 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2116 | return; | |
cfcea3e8 | 2117 | /* Remove from global LRU init */ |
1a81bb8a TM |
2118 | spin_lock(&nfs_access_lru_lock); |
2119 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2120 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2121 | |
1c3c07e9 | 2122 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2123 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2124 | spin_unlock(&inode->i_lock); | |
2125 | spin_unlock(&nfs_access_lru_lock); | |
2126 | nfs_access_free_list(&head); | |
1c3c07e9 | 2127 | } |
1da177e4 | 2128 | |
1c3c07e9 TM |
2129 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2130 | { | |
2131 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2132 | struct nfs_access_entry *entry; | |
2133 | ||
2134 | while (n != NULL) { | |
2135 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2136 | ||
2137 | if (cred < entry->cred) | |
2138 | n = n->rb_left; | |
2139 | else if (cred > entry->cred) | |
2140 | n = n->rb_right; | |
2141 | else | |
2142 | return entry; | |
1da177e4 | 2143 | } |
1c3c07e9 TM |
2144 | return NULL; |
2145 | } | |
2146 | ||
af22f94a | 2147 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2148 | { |
2149 | struct nfs_inode *nfsi = NFS_I(inode); | |
2150 | struct nfs_access_entry *cache; | |
2151 | int err = -ENOENT; | |
2152 | ||
dc59250c | 2153 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2154 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2155 | goto out_zap; | |
2156 | cache = nfs_access_search_rbtree(inode, cred); | |
2157 | if (cache == NULL) | |
2158 | goto out; | |
b4d2314b | 2159 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2160 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2161 | goto out_stale; |
2162 | res->jiffies = cache->jiffies; | |
2163 | res->cred = cache->cred; | |
2164 | res->mask = cache->mask; | |
cfcea3e8 | 2165 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2166 | err = 0; |
2167 | out: | |
2168 | spin_unlock(&inode->i_lock); | |
2169 | return err; | |
2170 | out_stale: | |
2171 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2172 | list_del(&cache->lru); |
1c3c07e9 TM |
2173 | spin_unlock(&inode->i_lock); |
2174 | nfs_access_free_entry(cache); | |
2175 | return -ENOENT; | |
2176 | out_zap: | |
1a81bb8a TM |
2177 | spin_unlock(&inode->i_lock); |
2178 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2179 | return -ENOENT; |
2180 | } | |
2181 | ||
2182 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2183 | { | |
cfcea3e8 TM |
2184 | struct nfs_inode *nfsi = NFS_I(inode); |
2185 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2186 | struct rb_node **p = &root_node->rb_node; |
2187 | struct rb_node *parent = NULL; | |
2188 | struct nfs_access_entry *entry; | |
2189 | ||
2190 | spin_lock(&inode->i_lock); | |
2191 | while (*p != NULL) { | |
2192 | parent = *p; | |
2193 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2194 | ||
2195 | if (set->cred < entry->cred) | |
2196 | p = &parent->rb_left; | |
2197 | else if (set->cred > entry->cred) | |
2198 | p = &parent->rb_right; | |
2199 | else | |
2200 | goto found; | |
2201 | } | |
2202 | rb_link_node(&set->rb_node, parent, p); | |
2203 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2204 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2205 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2206 | return; |
2207 | found: | |
2208 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2209 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2210 | list_del(&entry->lru); | |
1c3c07e9 TM |
2211 | spin_unlock(&inode->i_lock); |
2212 | nfs_access_free_entry(entry); | |
2213 | } | |
2214 | ||
af22f94a | 2215 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2216 | { |
2217 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2218 | if (cache == NULL) | |
2219 | return; | |
2220 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2221 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2222 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2223 | cache->mask = set->mask; |
1c3c07e9 TM |
2224 | |
2225 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2226 | |
2227 | /* Update accounting */ | |
2228 | smp_mb__before_atomic_inc(); | |
2229 | atomic_long_inc(&nfs_access_nr_entries); | |
2230 | smp_mb__after_atomic_inc(); | |
2231 | ||
2232 | /* Add inode to global LRU list */ | |
1a81bb8a | 2233 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2234 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2235 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2236 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2237 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2238 | spin_unlock(&nfs_access_lru_lock); |
2239 | } | |
1da177e4 LT |
2240 | } |
2241 | ||
2242 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2243 | { | |
2244 | struct nfs_access_entry cache; | |
2245 | int status; | |
2246 | ||
2247 | status = nfs_access_get_cached(inode, cred, &cache); | |
2248 | if (status == 0) | |
2249 | goto out; | |
2250 | ||
2251 | /* Be clever: ask server to check for all possible rights */ | |
2252 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2253 | cache.cred = cred; | |
2254 | cache.jiffies = jiffies; | |
2255 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2256 | if (status != 0) { |
2257 | if (status == -ESTALE) { | |
2258 | nfs_zap_caches(inode); | |
2259 | if (!S_ISDIR(inode->i_mode)) | |
2260 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2261 | } | |
1da177e4 | 2262 | return status; |
a71ee337 | 2263 | } |
1da177e4 LT |
2264 | nfs_access_add_cache(inode, &cache); |
2265 | out: | |
e6305c43 | 2266 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2267 | return 0; |
2268 | return -EACCES; | |
2269 | } | |
2270 | ||
af22f94a TM |
2271 | static int nfs_open_permission_mask(int openflags) |
2272 | { | |
2273 | int mask = 0; | |
2274 | ||
8a5e929d | 2275 | if ((openflags & O_ACCMODE) != O_WRONLY) |
af22f94a | 2276 | mask |= MAY_READ; |
8a5e929d | 2277 | if ((openflags & O_ACCMODE) != O_RDONLY) |
af22f94a | 2278 | mask |= MAY_WRITE; |
8a5e929d | 2279 | if (openflags & __FMODE_EXEC) |
af22f94a TM |
2280 | mask |= MAY_EXEC; |
2281 | return mask; | |
2282 | } | |
2283 | ||
2284 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2285 | { | |
2286 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2287 | } | |
2288 | ||
10556cb2 | 2289 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2290 | { |
2291 | struct rpc_cred *cred; | |
2292 | int res = 0; | |
2293 | ||
10556cb2 | 2294 | if (mask & MAY_NOT_BLOCK) |
b74c79e9 NP |
2295 | return -ECHILD; |
2296 | ||
91d5b470 CL |
2297 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2298 | ||
e6305c43 | 2299 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2300 | goto out; |
2301 | /* Is this sys_access() ? */ | |
9cfcac81 | 2302 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2303 | goto force_lookup; |
2304 | ||
2305 | switch (inode->i_mode & S_IFMT) { | |
2306 | case S_IFLNK: | |
2307 | goto out; | |
2308 | case S_IFREG: | |
2309 | /* NFSv4 has atomic_open... */ | |
2310 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2311 | && (mask & MAY_OPEN) |
2312 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2313 | goto out; |
2314 | break; | |
2315 | case S_IFDIR: | |
2316 | /* | |
2317 | * Optimize away all write operations, since the server | |
2318 | * will check permissions when we perform the op. | |
2319 | */ | |
2320 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2321 | goto out; | |
2322 | } | |
2323 | ||
2324 | force_lookup: | |
1da177e4 LT |
2325 | if (!NFS_PROTO(inode)->access) |
2326 | goto out_notsup; | |
2327 | ||
98a8e323 | 2328 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2329 | if (!IS_ERR(cred)) { |
2330 | res = nfs_do_access(inode, cred, mask); | |
2331 | put_rpccred(cred); | |
2332 | } else | |
2333 | res = PTR_ERR(cred); | |
1da177e4 | 2334 | out: |
f696a365 MS |
2335 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2336 | res = -EACCES; | |
2337 | ||
1e7cb3dc CL |
2338 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2339 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2340 | return res; |
2341 | out_notsup: | |
2342 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2343 | if (res == 0) | |
2830ba7f | 2344 | res = generic_permission(inode, mask); |
1e7cb3dc | 2345 | goto out; |
1da177e4 LT |
2346 | } |
2347 | ||
2348 | /* | |
2349 | * Local variables: | |
2350 | * version-control: t | |
2351 | * kept-new-versions: 5 | |
2352 | * End: | |
2353 | */ |