]>
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> |
56e4ebf8 | 36 | #include <linux/vmalloc.h> |
04e4bd1c | 37 | #include <linux/kmemleak.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 *); | |
47 | static int nfs_readdir(struct file *, void *, filldir_t); | |
48 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
49 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
50 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
51 | static int nfs_rmdir(struct inode *, struct dentry *); | |
52 | static int nfs_unlink(struct inode *, struct dentry *); | |
53 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
54 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
55 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
56 | static int nfs_rename(struct inode *, struct dentry *, | |
57 | struct inode *, struct dentry *); | |
7ea80859 | 58 | static int nfs_fsync_dir(struct file *, int); |
f0dd2136 | 59 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
d1bacf9e | 60 | static int nfs_readdir_clear_array(struct page*, gfp_t); |
1da177e4 | 61 | |
4b6f5d20 | 62 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 63 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
64 | .read = generic_read_dir, |
65 | .readdir = nfs_readdir, | |
66 | .open = nfs_opendir, | |
67 | .release = nfs_release, | |
68 | .fsync = nfs_fsync_dir, | |
69 | }; | |
70 | ||
92e1d5be | 71 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
72 | .create = nfs_create, |
73 | .lookup = nfs_lookup, | |
74 | .link = nfs_link, | |
75 | .unlink = nfs_unlink, | |
76 | .symlink = nfs_symlink, | |
77 | .mkdir = nfs_mkdir, | |
78 | .rmdir = nfs_rmdir, | |
79 | .mknod = nfs_mknod, | |
80 | .rename = nfs_rename, | |
81 | .permission = nfs_permission, | |
82 | .getattr = nfs_getattr, | |
83 | .setattr = nfs_setattr, | |
84 | }; | |
85 | ||
d1bacf9e BS |
86 | const struct address_space_operations nfs_dir_addr_space_ops = { |
87 | .releasepage = nfs_readdir_clear_array, | |
88 | }; | |
89 | ||
b7fa0554 | 90 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 91 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
92 | .create = nfs_create, |
93 | .lookup = nfs_lookup, | |
94 | .link = nfs_link, | |
95 | .unlink = nfs_unlink, | |
96 | .symlink = nfs_symlink, | |
97 | .mkdir = nfs_mkdir, | |
98 | .rmdir = nfs_rmdir, | |
99 | .mknod = nfs_mknod, | |
100 | .rename = nfs_rename, | |
101 | .permission = nfs_permission, | |
102 | .getattr = nfs_getattr, | |
103 | .setattr = nfs_setattr, | |
104 | .listxattr = nfs3_listxattr, | |
105 | .getxattr = nfs3_getxattr, | |
106 | .setxattr = nfs3_setxattr, | |
107 | .removexattr = nfs3_removexattr, | |
108 | }; | |
109 | #endif /* CONFIG_NFS_V3 */ | |
110 | ||
1da177e4 LT |
111 | #ifdef CONFIG_NFS_V4 |
112 | ||
113 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
c0204fd2 | 114 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd); |
92e1d5be | 115 | const struct inode_operations nfs4_dir_inode_operations = { |
c0204fd2 | 116 | .create = nfs_open_create, |
1da177e4 LT |
117 | .lookup = nfs_atomic_lookup, |
118 | .link = nfs_link, | |
119 | .unlink = nfs_unlink, | |
120 | .symlink = nfs_symlink, | |
121 | .mkdir = nfs_mkdir, | |
122 | .rmdir = nfs_rmdir, | |
123 | .mknod = nfs_mknod, | |
124 | .rename = nfs_rename, | |
125 | .permission = nfs_permission, | |
126 | .getattr = nfs_getattr, | |
127 | .setattr = nfs_setattr, | |
6b3b5496 BF |
128 | .getxattr = nfs4_getxattr, |
129 | .setxattr = nfs4_setxattr, | |
130 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
131 | }; |
132 | ||
133 | #endif /* CONFIG_NFS_V4 */ | |
134 | ||
135 | /* | |
136 | * Open file | |
137 | */ | |
138 | static int | |
139 | nfs_opendir(struct inode *inode, struct file *filp) | |
140 | { | |
7451c4f0 | 141 | int res; |
1da177e4 | 142 | |
6da24bc9 | 143 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
cc0dd2d1 CL |
144 | filp->f_path.dentry->d_parent->d_name.name, |
145 | filp->f_path.dentry->d_name.name); | |
146 | ||
147 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 148 | |
1da177e4 | 149 | /* Call generic open code in order to cache credentials */ |
7451c4f0 | 150 | res = nfs_open(inode, filp); |
f5a73672 NB |
151 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
152 | /* This is a mountpoint, so d_revalidate will never | |
153 | * have been called, so we need to refresh the | |
154 | * inode (for close-open consistency) ourselves. | |
155 | */ | |
156 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
157 | } | |
1da177e4 LT |
158 | return res; |
159 | } | |
160 | ||
d1bacf9e BS |
161 | struct nfs_cache_array_entry { |
162 | u64 cookie; | |
163 | u64 ino; | |
164 | struct qstr string; | |
0b26a0bf | 165 | unsigned char d_type; |
d1bacf9e BS |
166 | }; |
167 | ||
168 | struct nfs_cache_array { | |
169 | unsigned int size; | |
170 | int eof_index; | |
171 | u64 last_cookie; | |
172 | struct nfs_cache_array_entry array[0]; | |
173 | }; | |
174 | ||
82f2e547 | 175 | typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int); |
1da177e4 LT |
176 | typedef struct { |
177 | struct file *file; | |
178 | struct page *page; | |
179 | unsigned long page_index; | |
f0dd2136 TM |
180 | u64 *dir_cookie; |
181 | loff_t current_index; | |
1da177e4 | 182 | decode_dirent_t decode; |
d1bacf9e | 183 | |
1f4eab7e | 184 | unsigned long timestamp; |
4704f0e2 | 185 | unsigned long gencount; |
d1bacf9e BS |
186 | unsigned int cache_entry_index; |
187 | unsigned int plus:1; | |
188 | unsigned int eof:1; | |
1da177e4 LT |
189 | } nfs_readdir_descriptor_t; |
190 | ||
d1bacf9e BS |
191 | /* |
192 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
1da177e4 LT |
193 | */ |
194 | static | |
d1bacf9e BS |
195 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
196 | { | |
8cd51a0c | 197 | void *ptr; |
d1bacf9e BS |
198 | if (page == NULL) |
199 | return ERR_PTR(-EIO); | |
8cd51a0c TM |
200 | ptr = kmap(page); |
201 | if (ptr == NULL) | |
202 | return ERR_PTR(-ENOMEM); | |
203 | return ptr; | |
d1bacf9e BS |
204 | } |
205 | ||
206 | static | |
207 | void nfs_readdir_release_array(struct page *page) | |
208 | { | |
209 | kunmap(page); | |
210 | } | |
211 | ||
212 | /* | |
213 | * we are freeing strings created by nfs_add_to_readdir_array() | |
214 | */ | |
215 | static | |
216 | int nfs_readdir_clear_array(struct page *page, gfp_t mask) | |
217 | { | |
218 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
219 | int i; | |
8cd51a0c TM |
220 | |
221 | if (IS_ERR(array)) | |
222 | return PTR_ERR(array); | |
d1bacf9e BS |
223 | for (i = 0; i < array->size; i++) |
224 | kfree(array->array[i].string.name); | |
225 | nfs_readdir_release_array(page); | |
226 | return 0; | |
227 | } | |
228 | ||
229 | /* | |
230 | * the caller is responsible for freeing qstr.name | |
231 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
232 | * nfs_clear_readdir_array() | |
233 | */ | |
234 | static | |
4a201d6e | 235 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
236 | { |
237 | string->len = len; | |
238 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
239 | if (string->name == NULL) |
240 | return -ENOMEM; | |
04e4bd1c CM |
241 | /* |
242 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
243 | * in a page cache page which kmemleak does not scan. | |
244 | */ | |
245 | kmemleak_not_leak(string->name); | |
4a201d6e TM |
246 | string->hash = full_name_hash(name, len); |
247 | return 0; | |
d1bacf9e BS |
248 | } |
249 | ||
250 | static | |
251 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
252 | { | |
253 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
254 | struct nfs_cache_array_entry *cache_entry; |
255 | int ret; | |
256 | ||
d1bacf9e BS |
257 | if (IS_ERR(array)) |
258 | return PTR_ERR(array); | |
3020093f TM |
259 | |
260 | cache_entry = &array->array[array->size]; | |
261 | ||
262 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 263 | ret = -ENOSPC; |
3020093f | 264 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 265 | goto out; |
d1bacf9e | 266 | |
4a201d6e TM |
267 | cache_entry->cookie = entry->prev_cookie; |
268 | cache_entry->ino = entry->ino; | |
0b26a0bf | 269 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
270 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
271 | if (ret) | |
272 | goto out; | |
d1bacf9e | 273 | array->last_cookie = entry->cookie; |
8cd51a0c | 274 | array->size++; |
d1bacf9e BS |
275 | if (entry->eof == 1) |
276 | array->eof_index = array->size; | |
4a201d6e | 277 | out: |
d1bacf9e | 278 | nfs_readdir_release_array(page); |
4a201d6e | 279 | return ret; |
d1bacf9e BS |
280 | } |
281 | ||
282 | static | |
283 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
284 | { | |
285 | loff_t diff = desc->file->f_pos - desc->current_index; | |
286 | unsigned int index; | |
287 | ||
288 | if (diff < 0) | |
289 | goto out_eof; | |
290 | if (diff >= array->size) { | |
8cd51a0c | 291 | if (array->eof_index >= 0) |
d1bacf9e BS |
292 | goto out_eof; |
293 | desc->current_index += array->size; | |
294 | return -EAGAIN; | |
295 | } | |
296 | ||
297 | index = (unsigned int)diff; | |
298 | *desc->dir_cookie = array->array[index].cookie; | |
299 | desc->cache_entry_index = index; | |
d1bacf9e BS |
300 | return 0; |
301 | out_eof: | |
302 | desc->eof = 1; | |
303 | return -EBADCOOKIE; | |
304 | } | |
305 | ||
306 | static | |
307 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
308 | { | |
309 | int i; | |
310 | int status = -EAGAIN; | |
311 | ||
312 | for (i = 0; i < array->size; i++) { | |
d1bacf9e BS |
313 | if (array->array[i].cookie == *desc->dir_cookie) { |
314 | desc->cache_entry_index = i; | |
315 | status = 0; | |
8cd51a0c | 316 | goto out; |
d1bacf9e BS |
317 | } |
318 | } | |
8cd51a0c TM |
319 | if (i == array->eof_index) { |
320 | desc->eof = 1; | |
321 | status = -EBADCOOKIE; | |
322 | } | |
323 | out: | |
d1bacf9e BS |
324 | return status; |
325 | } | |
326 | ||
327 | static | |
328 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
329 | { | |
330 | struct nfs_cache_array *array; | |
331 | int status = -EBADCOOKIE; | |
332 | ||
333 | if (desc->dir_cookie == NULL) | |
334 | goto out; | |
335 | ||
336 | array = nfs_readdir_get_array(desc->page); | |
337 | if (IS_ERR(array)) { | |
338 | status = PTR_ERR(array); | |
339 | goto out; | |
340 | } | |
341 | ||
342 | if (*desc->dir_cookie == 0) | |
343 | status = nfs_readdir_search_for_pos(array, desc); | |
344 | else | |
345 | status = nfs_readdir_search_for_cookie(array, desc); | |
346 | ||
347 | nfs_readdir_release_array(desc->page); | |
348 | out: | |
349 | return status; | |
350 | } | |
351 | ||
352 | /* Fill a page with xdr information before transferring to the cache page */ | |
353 | static | |
56e4ebf8 | 354 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 355 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 356 | { |
1da177e4 | 357 | struct rpc_cred *cred = nfs_file_cred(file); |
4704f0e2 | 358 | unsigned long timestamp, gencount; |
1da177e4 LT |
359 | int error; |
360 | ||
1da177e4 LT |
361 | again: |
362 | timestamp = jiffies; | |
4704f0e2 | 363 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 364 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
365 | NFS_SERVER(inode)->dtsize, desc->plus); |
366 | if (error < 0) { | |
367 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
368 | if (error == -ENOTSUPP && desc->plus) { | |
369 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 370 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
371 | desc->plus = 0; |
372 | goto again; | |
373 | } | |
374 | goto error; | |
375 | } | |
1f4eab7e | 376 | desc->timestamp = timestamp; |
4704f0e2 | 377 | desc->gencount = gencount; |
d1bacf9e BS |
378 | error: |
379 | return error; | |
1da177e4 LT |
380 | } |
381 | ||
d1bacf9e BS |
382 | /* Fill in an entry based on the xdr code stored in desc->page */ |
383 | static | |
babddc72 | 384 | int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream) |
1da177e4 | 385 | { |
82f2e547 | 386 | __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus); |
1da177e4 LT |
387 | if (IS_ERR(p)) |
388 | return PTR_ERR(p); | |
1da177e4 | 389 | |
d1bacf9e BS |
390 | entry->fattr->time_start = desc->timestamp; |
391 | entry->fattr->gencount = desc->gencount; | |
392 | return 0; | |
1da177e4 LT |
393 | } |
394 | ||
d39ab9de BS |
395 | static |
396 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
397 | { | |
398 | struct nfs_inode *node; | |
399 | if (dentry->d_inode == NULL) | |
400 | goto different; | |
401 | node = NFS_I(dentry->d_inode); | |
402 | if (node->fh.size != entry->fh->size) | |
403 | goto different; | |
404 | if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 0) | |
405 | goto different; | |
406 | return 1; | |
407 | different: | |
408 | return 0; | |
409 | } | |
410 | ||
411 | static | |
412 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
413 | { | |
4a201d6e TM |
414 | struct qstr filename = { |
415 | .len = entry->len, | |
416 | .name = entry->name, | |
417 | }; | |
418 | struct dentry *dentry; | |
419 | struct dentry *alias; | |
d39ab9de BS |
420 | struct inode *dir = parent->d_inode; |
421 | struct inode *inode; | |
422 | ||
4a201d6e TM |
423 | if (filename.name[0] == '.') { |
424 | if (filename.len == 1) | |
425 | return; | |
426 | if (filename.len == 2 && filename.name[1] == '.') | |
427 | return; | |
428 | } | |
429 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 430 | |
4a201d6e | 431 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
432 | if (dentry != NULL) { |
433 | if (nfs_same_file(dentry, entry)) { | |
434 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
435 | goto out; | |
436 | } else { | |
437 | d_drop(dentry); | |
438 | dput(dentry); | |
439 | } | |
440 | } | |
441 | ||
442 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
443 | if (dentry == NULL) |
444 | return; | |
445 | ||
d39ab9de BS |
446 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
447 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
448 | if (IS_ERR(inode)) | |
449 | goto out; | |
450 | ||
451 | alias = d_materialise_unique(dentry, inode); | |
452 | if (IS_ERR(alias)) | |
453 | goto out; | |
454 | else if (alias) { | |
455 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
456 | dput(alias); | |
457 | } else | |
458 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
459 | ||
460 | out: | |
461 | dput(dentry); | |
d39ab9de BS |
462 | } |
463 | ||
d1bacf9e BS |
464 | /* Perform conversion from xdr to cache array */ |
465 | static | |
8cd51a0c | 466 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
56e4ebf8 | 467 | void *xdr_page, struct page *page, unsigned int buflen) |
1da177e4 | 468 | { |
babddc72 BS |
469 | struct xdr_stream stream; |
470 | struct xdr_buf buf; | |
56e4ebf8 | 471 | __be32 *ptr = xdr_page; |
99424380 | 472 | struct nfs_cache_array *array; |
5c346854 TM |
473 | unsigned int count = 0; |
474 | int status; | |
babddc72 BS |
475 | |
476 | buf.head->iov_base = xdr_page; | |
477 | buf.head->iov_len = buflen; | |
478 | buf.tail->iov_len = 0; | |
479 | buf.page_base = 0; | |
480 | buf.page_len = 0; | |
481 | buf.buflen = buf.head->iov_len; | |
482 | buf.len = buf.head->iov_len; | |
483 | ||
484 | xdr_init_decode(&stream, &buf, ptr); | |
485 | ||
99424380 BS |
486 | |
487 | do { | |
488 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
489 | if (status != 0) { |
490 | if (status == -EAGAIN) | |
491 | status = 0; | |
99424380 | 492 | break; |
8cd51a0c | 493 | } |
99424380 | 494 | |
5c346854 TM |
495 | count++; |
496 | ||
d39ab9de BS |
497 | if (desc->plus == 1) |
498 | nfs_prime_dcache(desc->file->f_path.dentry, entry); | |
8cd51a0c TM |
499 | |
500 | status = nfs_readdir_add_to_array(entry, page); | |
501 | if (status != 0) | |
502 | break; | |
99424380 BS |
503 | } while (!entry->eof); |
504 | ||
5c346854 | 505 | if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) { |
99424380 | 506 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
507 | if (!IS_ERR(array)) { |
508 | array->eof_index = array->size; | |
509 | status = 0; | |
510 | nfs_readdir_release_array(page); | |
5c346854 TM |
511 | } else |
512 | status = PTR_ERR(array); | |
1da177e4 | 513 | } |
8cd51a0c | 514 | return status; |
56e4ebf8 BS |
515 | } |
516 | ||
517 | static | |
518 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
519 | { | |
520 | unsigned int i; | |
521 | for (i = 0; i < npages; i++) | |
522 | put_page(pages[i]); | |
523 | } | |
524 | ||
525 | static | |
526 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
527 | unsigned int npages) | |
528 | { | |
529 | vm_unmap_ram(ptr, npages); | |
530 | nfs_readdir_free_pagearray(pages, npages); | |
531 | } | |
532 | ||
533 | /* | |
534 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
535 | * to nfs_readdir_free_large_page | |
536 | */ | |
537 | static | |
538 | void *nfs_readdir_large_page(struct page **pages, unsigned int npages) | |
539 | { | |
540 | void *ptr; | |
541 | unsigned int i; | |
542 | ||
543 | for (i = 0; i < npages; i++) { | |
544 | struct page *page = alloc_page(GFP_KERNEL); | |
545 | if (page == NULL) | |
546 | goto out_freepages; | |
547 | pages[i] = page; | |
548 | } | |
549 | ||
4a201d6e | 550 | ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL); |
56e4ebf8 BS |
551 | if (!IS_ERR_OR_NULL(ptr)) |
552 | return ptr; | |
553 | out_freepages: | |
554 | nfs_readdir_free_pagearray(pages, i); | |
555 | return NULL; | |
1da177e4 LT |
556 | } |
557 | ||
d1bacf9e BS |
558 | static |
559 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 560 | { |
56e4ebf8 BS |
561 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
562 | void *pages_ptr = NULL; | |
d1bacf9e BS |
563 | struct nfs_entry entry; |
564 | struct file *file = desc->file; | |
565 | struct nfs_cache_array *array; | |
8cd51a0c | 566 | int status = -ENOMEM; |
56e4ebf8 | 567 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
568 | |
569 | entry.prev_cookie = 0; | |
570 | entry.cookie = *desc->dir_cookie; | |
571 | entry.eof = 0; | |
572 | entry.fh = nfs_alloc_fhandle(); | |
573 | entry.fattr = nfs_alloc_fattr(); | |
574 | if (entry.fh == NULL || entry.fattr == NULL) | |
575 | goto out; | |
00a92642 | 576 | |
d1bacf9e | 577 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
578 | if (IS_ERR(array)) { |
579 | status = PTR_ERR(array); | |
580 | goto out; | |
581 | } | |
d1bacf9e BS |
582 | memset(array, 0, sizeof(struct nfs_cache_array)); |
583 | array->eof_index = -1; | |
00a92642 | 584 | |
56e4ebf8 BS |
585 | pages_ptr = nfs_readdir_large_page(pages, array_size); |
586 | if (!pages_ptr) | |
d1bacf9e BS |
587 | goto out_release_array; |
588 | do { | |
ac396128 | 589 | unsigned int pglen; |
56e4ebf8 | 590 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 591 | |
d1bacf9e | 592 | if (status < 0) |
00a92642 | 593 | break; |
ac396128 TM |
594 | pglen = status; |
595 | status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen); | |
8cd51a0c TM |
596 | if (status < 0) { |
597 | if (status == -ENOSPC) | |
598 | status = 0; | |
599 | break; | |
600 | } | |
601 | } while (array->eof_index < 0); | |
d1bacf9e | 602 | |
56e4ebf8 | 603 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
604 | out_release_array: |
605 | nfs_readdir_release_array(page); | |
606 | out: | |
607 | nfs_free_fattr(entry.fattr); | |
608 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
609 | return status; |
610 | } | |
611 | ||
612 | /* | |
d1bacf9e BS |
613 | * Now we cache directories properly, by converting xdr information |
614 | * to an array that can be used for lookups later. This results in | |
615 | * fewer cache pages, since we can store more information on each page. | |
616 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 617 | */ |
d1bacf9e BS |
618 | static |
619 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 620 | { |
01cce933 | 621 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
8cd51a0c | 622 | int ret; |
1da177e4 | 623 | |
8cd51a0c TM |
624 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
625 | if (ret < 0) | |
d1bacf9e BS |
626 | goto error; |
627 | SetPageUptodate(page); | |
1da177e4 | 628 | |
d1bacf9e BS |
629 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
630 | /* Should never happen */ | |
631 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 632 | } |
d1bacf9e BS |
633 | unlock_page(page); |
634 | return 0; | |
635 | error: | |
636 | unlock_page(page); | |
8cd51a0c | 637 | return ret; |
d1bacf9e | 638 | } |
1da177e4 | 639 | |
d1bacf9e BS |
640 | static |
641 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
642 | { | |
643 | page_cache_release(desc->page); | |
644 | desc->page = NULL; | |
645 | } | |
646 | ||
647 | static | |
648 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
649 | { | |
8cd51a0c | 650 | return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, |
d1bacf9e | 651 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
652 | } |
653 | ||
654 | /* | |
d1bacf9e | 655 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 656 | */ |
d1bacf9e BS |
657 | static |
658 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
659 | { | |
660 | int res; | |
661 | ||
662 | desc->page = get_cache_page(desc); | |
663 | if (IS_ERR(desc->page)) | |
664 | return PTR_ERR(desc->page); | |
665 | ||
666 | res = nfs_readdir_search_array(desc); | |
667 | if (res == 0) | |
668 | return 0; | |
669 | cache_page_release(desc); | |
670 | return res; | |
671 | } | |
672 | ||
673 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
674 | static inline |
675 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
676 | { | |
8cd51a0c | 677 | int res; |
d1bacf9e | 678 | |
8cd51a0c TM |
679 | if (desc->page_index == 0) |
680 | desc->current_index = 0; | |
d1bacf9e BS |
681 | while (1) { |
682 | res = find_cache_page(desc); | |
1da177e4 LT |
683 | if (res != -EAGAIN) |
684 | break; | |
d1bacf9e | 685 | desc->page_index++; |
1da177e4 | 686 | } |
1da177e4 LT |
687 | return res; |
688 | } | |
689 | ||
690 | static inline unsigned int dt_type(struct inode *inode) | |
691 | { | |
692 | return (inode->i_mode >> 12) & 15; | |
693 | } | |
694 | ||
1da177e4 LT |
695 | /* |
696 | * Once we've found the start of the dirent within a page: fill 'er up... | |
697 | */ | |
698 | static | |
699 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
700 | filldir_t filldir) | |
701 | { | |
702 | struct file *file = desc->file; | |
d1bacf9e BS |
703 | int i = 0; |
704 | int res = 0; | |
705 | struct nfs_cache_array *array = NULL; | |
d1bacf9e BS |
706 | |
707 | array = nfs_readdir_get_array(desc->page); | |
e7c58e97 TM |
708 | if (IS_ERR(array)) { |
709 | res = PTR_ERR(array); | |
710 | goto out; | |
711 | } | |
d1bacf9e BS |
712 | |
713 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 714 | struct nfs_cache_array_entry *ent; |
1da177e4 | 715 | |
ece0b423 TM |
716 | ent = &array->array[i]; |
717 | if (filldir(dirent, ent->string.name, ent->string.len, | |
0b26a0bf TM |
718 | file->f_pos, nfs_compat_user_ino64(ent->ino), |
719 | ent->d_type) < 0) { | |
ece0b423 | 720 | desc->eof = 1; |
1da177e4 | 721 | break; |
ece0b423 | 722 | } |
00a92642 | 723 | file->f_pos++; |
d1bacf9e BS |
724 | desc->cache_entry_index = i; |
725 | if (i < (array->size-1)) | |
726 | *desc->dir_cookie = array->array[i+1].cookie; | |
727 | else | |
728 | *desc->dir_cookie = array->last_cookie; | |
1da177e4 | 729 | } |
8cd51a0c TM |
730 | if (i == array->eof_index) |
731 | desc->eof = 1; | |
d1bacf9e BS |
732 | |
733 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 734 | out: |
d1bacf9e | 735 | cache_page_release(desc); |
1e7cb3dc CL |
736 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
737 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
738 | return res; |
739 | } | |
740 | ||
741 | /* | |
742 | * If we cannot find a cookie in our cache, we suspect that this is | |
743 | * because it points to a deleted file, so we ask the server to return | |
744 | * whatever it thinks is the next entry. We then feed this to filldir. | |
745 | * If all goes well, we should then be able to find our way round the | |
746 | * cache on the next call to readdir_search_pagecache(); | |
747 | * | |
748 | * NOTE: we cannot add the anonymous page to the pagecache because | |
749 | * the data it contains might not be page aligned. Besides, | |
750 | * we should already have a complete representation of the | |
751 | * directory in the page cache by the time we get here. | |
752 | */ | |
753 | static inline | |
754 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
755 | filldir_t filldir) | |
756 | { | |
1da177e4 LT |
757 | struct page *page = NULL; |
758 | int status; | |
d1bacf9e | 759 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 760 | |
1e7cb3dc CL |
761 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
762 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
763 | |
764 | page = alloc_page(GFP_HIGHUSER); | |
765 | if (!page) { | |
766 | status = -ENOMEM; | |
767 | goto out; | |
768 | } | |
d1bacf9e | 769 | |
7a8e1dc3 TM |
770 | desc->page_index = 0; |
771 | desc->page = page; | |
772 | ||
85f8607e TM |
773 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
774 | if (status < 0) | |
1da177e4 LT |
775 | goto out_release; |
776 | ||
777 | status = nfs_do_filldir(desc, dirent, filldir); | |
778 | ||
1da177e4 | 779 | out: |
1e7cb3dc | 780 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 781 | __func__, status); |
1da177e4 LT |
782 | return status; |
783 | out_release: | |
d1bacf9e | 784 | cache_page_release(desc); |
1da177e4 LT |
785 | goto out; |
786 | } | |
787 | ||
00a92642 OG |
788 | /* The file offset position represents the dirent entry number. A |
789 | last cookie cache takes care of the common case of reading the | |
790 | whole directory. | |
1da177e4 LT |
791 | */ |
792 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
793 | { | |
01cce933 | 794 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
795 | struct inode *inode = dentry->d_inode; |
796 | nfs_readdir_descriptor_t my_desc, | |
797 | *desc = &my_desc; | |
aa49b4cf | 798 | int res = -ENOMEM; |
1da177e4 | 799 | |
6da24bc9 | 800 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
801 | dentry->d_parent->d_name.name, dentry->d_name.name, |
802 | (long long)filp->f_pos); | |
91d5b470 CL |
803 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
804 | ||
1da177e4 | 805 | /* |
00a92642 | 806 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 807 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
808 | * to either find the entry with the appropriate number or |
809 | * revalidate the cookie. | |
1da177e4 LT |
810 | */ |
811 | memset(desc, 0, sizeof(*desc)); | |
812 | ||
813 | desc->file = filp; | |
cd3758e3 | 814 | desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; |
1da177e4 LT |
815 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
816 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
817 | ||
565277f6 | 818 | nfs_block_sillyrename(dentry); |
1cda707d | 819 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
820 | if (res < 0) |
821 | goto out; | |
822 | ||
d1bacf9e | 823 | while (desc->eof != 1) { |
1da177e4 | 824 | res = readdir_search_pagecache(desc); |
00a92642 | 825 | |
1da177e4 | 826 | if (res == -EBADCOOKIE) { |
ece0b423 | 827 | res = 0; |
1da177e4 | 828 | /* This means either end of directory */ |
d1bacf9e | 829 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
830 | /* Or that the server has 'lost' a cookie */ |
831 | res = uncached_readdir(desc, dirent, filldir); | |
ece0b423 | 832 | if (res == 0) |
1da177e4 LT |
833 | continue; |
834 | } | |
1da177e4 LT |
835 | break; |
836 | } | |
837 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 838 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 839 | nfs_zap_caches(inode); |
baf57a09 | 840 | desc->page_index = 0; |
1da177e4 | 841 | desc->plus = 0; |
d1bacf9e | 842 | desc->eof = 0; |
1da177e4 LT |
843 | continue; |
844 | } | |
845 | if (res < 0) | |
846 | break; | |
847 | ||
848 | res = nfs_do_filldir(desc, dirent, filldir); | |
ece0b423 | 849 | if (res < 0) |
1da177e4 | 850 | break; |
1da177e4 | 851 | } |
fccca7fc | 852 | out: |
565277f6 | 853 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
854 | if (res > 0) |
855 | res = 0; | |
aa49b4cf | 856 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
857 | dentry->d_parent->d_name.name, dentry->d_name.name, |
858 | res); | |
859 | return res; | |
1da177e4 LT |
860 | } |
861 | ||
10afec90 | 862 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 863 | { |
b84e06c5 CL |
864 | struct dentry *dentry = filp->f_path.dentry; |
865 | struct inode *inode = dentry->d_inode; | |
866 | ||
6da24bc9 | 867 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
868 | dentry->d_parent->d_name.name, |
869 | dentry->d_name.name, | |
870 | offset, origin); | |
871 | ||
872 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
873 | switch (origin) { |
874 | case 1: | |
875 | offset += filp->f_pos; | |
876 | case 0: | |
877 | if (offset >= 0) | |
878 | break; | |
879 | default: | |
880 | offset = -EINVAL; | |
881 | goto out; | |
882 | } | |
883 | if (offset != filp->f_pos) { | |
884 | filp->f_pos = offset; | |
cd3758e3 | 885 | nfs_file_open_context(filp)->dir_cookie = 0; |
f0dd2136 TM |
886 | } |
887 | out: | |
b84e06c5 | 888 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
889 | return offset; |
890 | } | |
891 | ||
1da177e4 LT |
892 | /* |
893 | * All directory operations under NFS are synchronous, so fsync() | |
894 | * is a dummy operation. | |
895 | */ | |
7ea80859 | 896 | static int nfs_fsync_dir(struct file *filp, int datasync) |
1da177e4 | 897 | { |
7ea80859 CH |
898 | struct dentry *dentry = filp->f_path.dentry; |
899 | ||
6da24bc9 | 900 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
901 | dentry->d_parent->d_name.name, dentry->d_name.name, |
902 | datasync); | |
903 | ||
54917786 | 904 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
1da177e4 LT |
905 | return 0; |
906 | } | |
907 | ||
bfc69a45 TM |
908 | /** |
909 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
910 | * @dir - pointer to directory inode | |
911 | * | |
912 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
913 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
914 | * on the server that might have invalidated our dcache. | |
915 | * | |
916 | * The caller should be holding dir->i_lock | |
917 | */ | |
918 | void nfs_force_lookup_revalidate(struct inode *dir) | |
919 | { | |
011935a0 | 920 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
921 | } |
922 | ||
1da177e4 LT |
923 | /* |
924 | * A check for whether or not the parent directory has changed. | |
925 | * In the case it has, we assume that the dentries are untrustworthy | |
926 | * and may need to be looked up again. | |
927 | */ | |
c79ba787 | 928 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
929 | { |
930 | if (IS_ROOT(dentry)) | |
931 | return 1; | |
4eec952e TM |
932 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
933 | return 0; | |
f2c77f4e TM |
934 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
935 | return 0; | |
936 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
937 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
938 | return 0; | |
939 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
940 | return 0; | |
941 | return 1; | |
1da177e4 LT |
942 | } |
943 | ||
1d6757fb TM |
944 | /* |
945 | * Return the intent data that applies to this particular path component | |
946 | * | |
947 | * Note that the current set of intents only apply to the very last | |
948 | * component of the path. | |
949 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
950 | */ | |
951 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
952 | { | |
953 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
954 | return 0; | |
955 | return nd->flags & mask; | |
956 | } | |
957 | ||
a12802ca TM |
958 | /* |
959 | * Use intent information to check whether or not we're going to do | |
960 | * an O_EXCL create using this path component. | |
961 | */ | |
962 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
963 | { | |
964 | if (NFS_PROTO(dir)->version == 2) | |
965 | return 0; | |
3516586a | 966 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
967 | } |
968 | ||
1d6757fb TM |
969 | /* |
970 | * Inode and filehandle revalidation for lookups. | |
971 | * | |
972 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
973 | * or if the intent information indicates that we're about to open this | |
974 | * particular file and the "nocto" mount flag is not set. | |
975 | * | |
976 | */ | |
1da177e4 LT |
977 | static inline |
978 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
979 | { | |
980 | struct nfs_server *server = NFS_SERVER(inode); | |
981 | ||
4e99a1ff TM |
982 | if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) |
983 | return 0; | |
1da177e4 | 984 | if (nd != NULL) { |
1da177e4 | 985 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 986 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
987 | goto out_force; |
988 | /* This is an open(2) */ | |
1d6757fb | 989 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
990 | !(server->flags & NFS_MOUNT_NOCTO) && |
991 | (S_ISREG(inode->i_mode) || | |
992 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 993 | goto out_force; |
4f48af45 | 994 | return 0; |
1da177e4 LT |
995 | } |
996 | return nfs_revalidate_inode(server, inode); | |
997 | out_force: | |
998 | return __nfs_revalidate_inode(server, inode); | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * We judge how long we want to trust negative | |
1003 | * dentries by looking at the parent inode mtime. | |
1004 | * | |
1005 | * If parent mtime has changed, we revalidate, else we wait for a | |
1006 | * period corresponding to the parent's attribute cache timeout value. | |
1007 | */ | |
1008 | static inline | |
1009 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1010 | struct nameidata *nd) | |
1011 | { | |
1da177e4 | 1012 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 1013 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 1014 | return 0; |
4eec952e TM |
1015 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1016 | return 1; | |
1da177e4 LT |
1017 | return !nfs_check_verifier(dir, dentry); |
1018 | } | |
1019 | ||
1020 | /* | |
1021 | * This is called every time the dcache has a lookup hit, | |
1022 | * and we should check whether we can really trust that | |
1023 | * lookup. | |
1024 | * | |
1025 | * NOTE! The hit can be a negative hit too, don't assume | |
1026 | * we have an inode! | |
1027 | * | |
1028 | * If the parent directory is seen to have changed, we throw out the | |
1029 | * cached dentry and do a new lookup. | |
1030 | */ | |
1031 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
1032 | { | |
1033 | struct inode *dir; | |
1034 | struct inode *inode; | |
1035 | struct dentry *parent; | |
e1fb4d05 TM |
1036 | struct nfs_fh *fhandle = NULL; |
1037 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1038 | int error; |
1da177e4 LT |
1039 | |
1040 | parent = dget_parent(dentry); | |
1da177e4 | 1041 | dir = parent->d_inode; |
91d5b470 | 1042 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1043 | inode = dentry->d_inode; |
1044 | ||
1045 | if (!inode) { | |
1046 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1047 | goto out_bad; | |
1048 | goto out_valid; | |
1049 | } | |
1050 | ||
1051 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1052 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1053 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1054 | dentry->d_name.name); |
1da177e4 LT |
1055 | goto out_bad; |
1056 | } | |
1057 | ||
15860ab1 TM |
1058 | if (nfs_have_delegation(inode, FMODE_READ)) |
1059 | goto out_set_verifier; | |
1060 | ||
1da177e4 | 1061 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1062 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1063 | if (nfs_lookup_verify_inode(inode, nd)) |
1064 | goto out_zap_parent; | |
1065 | goto out_valid; | |
1066 | } | |
1067 | ||
1068 | if (NFS_STALE(inode)) | |
1069 | goto out_bad; | |
1070 | ||
e1fb4d05 TM |
1071 | error = -ENOMEM; |
1072 | fhandle = nfs_alloc_fhandle(); | |
1073 | fattr = nfs_alloc_fattr(); | |
1074 | if (fhandle == NULL || fattr == NULL) | |
1075 | goto out_error; | |
1076 | ||
1077 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1da177e4 LT |
1078 | if (error) |
1079 | goto out_bad; | |
e1fb4d05 | 1080 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1081 | goto out_bad; |
e1fb4d05 | 1082 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1083 | goto out_bad; |
1084 | ||
e1fb4d05 TM |
1085 | nfs_free_fattr(fattr); |
1086 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1087 | out_set_verifier: |
cf8ba45e | 1088 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1089 | out_valid: |
1da177e4 | 1090 | dput(parent); |
1e7cb3dc | 1091 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1092 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1093 | dentry->d_name.name); |
1da177e4 LT |
1094 | return 1; |
1095 | out_zap_parent: | |
1096 | nfs_zap_caches(dir); | |
1097 | out_bad: | |
a1643a92 | 1098 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1099 | if (inode && S_ISDIR(inode->i_mode)) { |
1100 | /* Purge readdir caches. */ | |
1101 | nfs_zap_caches(inode); | |
1102 | /* If we have submounts, don't unhash ! */ | |
1103 | if (have_submounts(dentry)) | |
1104 | goto out_valid; | |
d9e80b7d AV |
1105 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1106 | goto out_valid; | |
1da177e4 LT |
1107 | shrink_dcache_parent(dentry); |
1108 | } | |
1109 | d_drop(dentry); | |
e1fb4d05 TM |
1110 | nfs_free_fattr(fattr); |
1111 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1112 | dput(parent); |
1e7cb3dc | 1113 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1114 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1115 | dentry->d_name.name); |
1da177e4 | 1116 | return 0; |
e1fb4d05 TM |
1117 | out_error: |
1118 | nfs_free_fattr(fattr); | |
1119 | nfs_free_fhandle(fhandle); | |
1120 | dput(parent); | |
1121 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1122 | __func__, dentry->d_parent->d_name.name, | |
1123 | dentry->d_name.name, error); | |
1124 | return error; | |
1da177e4 LT |
1125 | } |
1126 | ||
1127 | /* | |
1128 | * This is called from dput() when d_count is going to 0. | |
1129 | */ | |
1130 | static int nfs_dentry_delete(struct dentry *dentry) | |
1131 | { | |
1132 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1133 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1134 | dentry->d_flags); | |
1135 | ||
77f11192 TM |
1136 | /* Unhash any dentry with a stale inode */ |
1137 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1138 | return 1; | |
1139 | ||
1da177e4 LT |
1140 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1141 | /* Unhash it, so that ->d_iput() would be called */ | |
1142 | return 1; | |
1143 | } | |
1144 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1145 | /* Unhash it, so that ancestors of killed async unlink | |
1146 | * files will be cleaned up during umount */ | |
1147 | return 1; | |
1148 | } | |
1149 | return 0; | |
1150 | ||
1151 | } | |
1152 | ||
1b83d707 TM |
1153 | static void nfs_drop_nlink(struct inode *inode) |
1154 | { | |
1155 | spin_lock(&inode->i_lock); | |
1156 | if (inode->i_nlink > 0) | |
1157 | drop_nlink(inode); | |
1158 | spin_unlock(&inode->i_lock); | |
1159 | } | |
1160 | ||
1da177e4 LT |
1161 | /* |
1162 | * Called when the dentry loses inode. | |
1163 | * We use it to clean up silly-renamed files. | |
1164 | */ | |
1165 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1166 | { | |
83672d39 NB |
1167 | if (S_ISDIR(inode->i_mode)) |
1168 | /* drop any readdir cache as it could easily be old */ | |
1169 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1170 | ||
1da177e4 | 1171 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1172 | drop_nlink(inode); |
e4eff1a6 | 1173 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1174 | } |
1da177e4 LT |
1175 | iput(inode); |
1176 | } | |
1177 | ||
f786aa90 | 1178 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1179 | .d_revalidate = nfs_lookup_revalidate, |
1180 | .d_delete = nfs_dentry_delete, | |
1181 | .d_iput = nfs_dentry_iput, | |
1182 | }; | |
1183 | ||
1da177e4 LT |
1184 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1185 | { | |
1186 | struct dentry *res; | |
565277f6 | 1187 | struct dentry *parent; |
1da177e4 | 1188 | struct inode *inode = NULL; |
e1fb4d05 TM |
1189 | struct nfs_fh *fhandle = NULL; |
1190 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1191 | int error; |
1da177e4 LT |
1192 | |
1193 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1194 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1195 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1196 | |
1197 | res = ERR_PTR(-ENAMETOOLONG); | |
1198 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1199 | goto out; | |
1200 | ||
1da177e4 LT |
1201 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
1202 | ||
fd684071 TM |
1203 | /* |
1204 | * If we're doing an exclusive create, optimize away the lookup | |
1205 | * but don't hash the dentry. | |
1206 | */ | |
1207 | if (nfs_is_exclusive_create(dir, nd)) { | |
1208 | d_instantiate(dentry, NULL); | |
1209 | res = NULL; | |
fc0f684c | 1210 | goto out; |
fd684071 | 1211 | } |
1da177e4 | 1212 | |
e1fb4d05 TM |
1213 | res = ERR_PTR(-ENOMEM); |
1214 | fhandle = nfs_alloc_fhandle(); | |
1215 | fattr = nfs_alloc_fattr(); | |
1216 | if (fhandle == NULL || fattr == NULL) | |
1217 | goto out; | |
1218 | ||
565277f6 TM |
1219 | parent = dentry->d_parent; |
1220 | /* Protect against concurrent sillydeletes */ | |
1221 | nfs_block_sillyrename(parent); | |
e1fb4d05 | 1222 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1223 | if (error == -ENOENT) |
1224 | goto no_entry; | |
1225 | if (error < 0) { | |
1226 | res = ERR_PTR(error); | |
565277f6 | 1227 | goto out_unblock_sillyrename; |
1da177e4 | 1228 | } |
e1fb4d05 | 1229 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1230 | res = (struct dentry *)inode; |
1231 | if (IS_ERR(res)) | |
565277f6 | 1232 | goto out_unblock_sillyrename; |
54ceac45 | 1233 | |
1da177e4 | 1234 | no_entry: |
54ceac45 | 1235 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1236 | if (res != NULL) { |
1237 | if (IS_ERR(res)) | |
565277f6 | 1238 | goto out_unblock_sillyrename; |
1da177e4 | 1239 | dentry = res; |
9eaef27b | 1240 | } |
1da177e4 | 1241 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1242 | out_unblock_sillyrename: |
1243 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1244 | out: |
e1fb4d05 TM |
1245 | nfs_free_fattr(fattr); |
1246 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1247 | return res; |
1248 | } | |
1249 | ||
1250 | #ifdef CONFIG_NFS_V4 | |
1251 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
1252 | ||
f786aa90 | 1253 | const struct dentry_operations nfs4_dentry_operations = { |
1da177e4 LT |
1254 | .d_revalidate = nfs_open_revalidate, |
1255 | .d_delete = nfs_dentry_delete, | |
1256 | .d_iput = nfs_dentry_iput, | |
1257 | }; | |
1258 | ||
1d6757fb TM |
1259 | /* |
1260 | * Use intent information to determine whether we need to substitute | |
1261 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1262 | */ | |
5584c306 | 1263 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1264 | { |
1d6757fb | 1265 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1266 | return 0; |
1267 | /* NFS does not (yet) have a stateful open for directories */ | |
1268 | if (nd->flags & LOOKUP_DIRECTORY) | |
1269 | return 0; | |
1270 | /* Are we trying to write to a read only partition? */ | |
2c463e95 DH |
1271 | if (__mnt_is_readonly(nd->path.mnt) && |
1272 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
1da177e4 LT |
1273 | return 0; |
1274 | return 1; | |
1275 | } | |
1276 | ||
cd9a1c0e TM |
1277 | static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd) |
1278 | { | |
1279 | struct path path = { | |
1280 | .mnt = nd->path.mnt, | |
1281 | .dentry = dentry, | |
1282 | }; | |
1283 | struct nfs_open_context *ctx; | |
1284 | struct rpc_cred *cred; | |
1285 | fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); | |
1286 | ||
1287 | cred = rpc_lookup_cred(); | |
1288 | if (IS_ERR(cred)) | |
1289 | return ERR_CAST(cred); | |
1290 | ctx = alloc_nfs_open_context(&path, cred, fmode); | |
1291 | put_rpccred(cred); | |
1292 | if (ctx == NULL) | |
1293 | return ERR_PTR(-ENOMEM); | |
1294 | return ctx; | |
1295 | } | |
1296 | ||
1297 | static int do_open(struct inode *inode, struct file *filp) | |
1298 | { | |
1299 | nfs_fscache_set_inode_cookie(inode, filp); | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx) | |
1304 | { | |
1305 | struct file *filp; | |
1306 | int ret = 0; | |
1307 | ||
1308 | /* If the open_intent is for execute, we have an extra check to make */ | |
1309 | if (ctx->mode & FMODE_EXEC) { | |
1310 | ret = nfs_may_open(ctx->path.dentry->d_inode, | |
1311 | ctx->cred, | |
1312 | nd->intent.open.flags); | |
1313 | if (ret < 0) | |
1314 | goto out; | |
1315 | } | |
1316 | filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open); | |
1317 | if (IS_ERR(filp)) | |
1318 | ret = PTR_ERR(filp); | |
1319 | else | |
1320 | nfs_file_set_open_context(filp, ctx); | |
1321 | out: | |
1322 | put_nfs_open_context(ctx); | |
1323 | return ret; | |
1324 | } | |
1325 | ||
1da177e4 LT |
1326 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
1327 | { | |
cd9a1c0e TM |
1328 | struct nfs_open_context *ctx; |
1329 | struct iattr attr; | |
1da177e4 | 1330 | struct dentry *res = NULL; |
f46e0bd3 | 1331 | struct inode *inode; |
cd9a1c0e | 1332 | int open_flags; |
898f635c | 1333 | int err; |
1da177e4 | 1334 | |
1e7cb3dc CL |
1335 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
1336 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1337 | ||
1da177e4 | 1338 | /* Check that we are indeed trying to open this file */ |
5584c306 | 1339 | if (!is_atomic_open(nd)) |
1da177e4 LT |
1340 | goto no_open; |
1341 | ||
1342 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1343 | res = ERR_PTR(-ENAMETOOLONG); | |
1344 | goto out; | |
1345 | } | |
1346 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1347 | ||
d4d9cdcb TM |
1348 | /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash |
1349 | * the dentry. */ | |
3516586a | 1350 | if (nd->flags & LOOKUP_EXCL) { |
d4d9cdcb | 1351 | d_instantiate(dentry, NULL); |
02a913a7 TM |
1352 | goto out; |
1353 | } | |
1da177e4 | 1354 | |
cd9a1c0e TM |
1355 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1356 | res = ERR_CAST(ctx); | |
1357 | if (IS_ERR(ctx)) | |
1358 | goto out; | |
1359 | ||
1360 | open_flags = nd->intent.open.flags; | |
1361 | if (nd->flags & LOOKUP_CREATE) { | |
1362 | attr.ia_mode = nd->intent.open.create_mode; | |
1363 | attr.ia_valid = ATTR_MODE; | |
1364 | if (!IS_POSIXACL(dir)) | |
1365 | attr.ia_mode &= ~current_umask(); | |
1366 | } else { | |
898f635c | 1367 | open_flags &= ~(O_EXCL | O_CREAT); |
cd9a1c0e | 1368 | attr.ia_valid = 0; |
cd9a1c0e TM |
1369 | } |
1370 | ||
1da177e4 | 1371 | /* Open the file on the server */ |
f46e0bd3 | 1372 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1373 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
f46e0bd3 TM |
1374 | if (IS_ERR(inode)) { |
1375 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1376 | put_nfs_open_context(ctx); |
f46e0bd3 | 1377 | switch (PTR_ERR(inode)) { |
1da177e4 LT |
1378 | /* Make a negative dentry */ |
1379 | case -ENOENT: | |
f46e0bd3 | 1380 | d_add(dentry, NULL); |
02a913a7 TM |
1381 | res = NULL; |
1382 | goto out; | |
1da177e4 | 1383 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1384 | case -ENOTDIR: |
1385 | goto no_open; | |
1da177e4 LT |
1386 | case -ELOOP: |
1387 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1388 | goto no_open; | |
23ebbd9a | 1389 | /* case -EISDIR: */ |
1da177e4 LT |
1390 | /* case -EINVAL: */ |
1391 | default: | |
f46e0bd3 | 1392 | res = ERR_CAST(inode); |
1da177e4 LT |
1393 | goto out; |
1394 | } | |
cd9a1c0e | 1395 | } |
f46e0bd3 | 1396 | res = d_add_unique(dentry, inode); |
898f635c | 1397 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 TM |
1398 | if (res != NULL) { |
1399 | dput(ctx->path.dentry); | |
1400 | ctx->path.dentry = dget(res); | |
1da177e4 | 1401 | dentry = res; |
f46e0bd3 | 1402 | } |
898f635c TM |
1403 | err = nfs_intent_set_file(nd, ctx); |
1404 | if (err < 0) { | |
1405 | if (res != NULL) | |
1406 | dput(res); | |
1407 | return ERR_PTR(err); | |
1408 | } | |
1da177e4 | 1409 | out: |
f46e0bd3 | 1410 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1411 | return res; |
1412 | no_open: | |
1413 | return nfs_lookup(dir, dentry, nd); | |
1414 | } | |
1415 | ||
1416 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1417 | { | |
1418 | struct dentry *parent = NULL; | |
1419 | struct inode *inode = dentry->d_inode; | |
1420 | struct inode *dir; | |
b8d4cadd | 1421 | struct nfs_open_context *ctx; |
1da177e4 LT |
1422 | int openflags, ret = 0; |
1423 | ||
1f063d2c | 1424 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1425 | goto no_open; |
2b484297 | 1426 | |
1da177e4 LT |
1427 | parent = dget_parent(dentry); |
1428 | dir = parent->d_inode; | |
2b484297 | 1429 | |
1da177e4 LT |
1430 | /* We can't create new files in nfs_open_revalidate(), so we |
1431 | * optimize away revalidation of negative dentries. | |
1432 | */ | |
216d5d06 TM |
1433 | if (inode == NULL) { |
1434 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1435 | ret = 1; | |
1da177e4 | 1436 | goto out; |
216d5d06 TM |
1437 | } |
1438 | ||
1da177e4 LT |
1439 | /* NFS only supports OPEN on regular files */ |
1440 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1441 | goto no_open_dput; |
1da177e4 LT |
1442 | openflags = nd->intent.open.flags; |
1443 | /* We cannot do exclusive creation on a positive dentry */ | |
1444 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1445 | goto no_open_dput; |
1da177e4 | 1446 | /* We can't create new files, or truncate existing ones here */ |
0a377cff | 1447 | openflags &= ~(O_CREAT|O_EXCL|O_TRUNC); |
1da177e4 | 1448 | |
b8d4cadd TM |
1449 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1450 | ret = PTR_ERR(ctx); | |
1451 | if (IS_ERR(ctx)) | |
1452 | goto out; | |
1da177e4 | 1453 | /* |
1b1dcc1b | 1454 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1455 | * operations that change the directory. We therefore save the |
1456 | * change attribute *before* we do the RPC call. | |
1457 | */ | |
2b484297 | 1458 | inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL); |
535918f1 TM |
1459 | if (IS_ERR(inode)) { |
1460 | ret = PTR_ERR(inode); | |
1461 | switch (ret) { | |
1462 | case -EPERM: | |
1463 | case -EACCES: | |
1464 | case -EDQUOT: | |
1465 | case -ENOSPC: | |
1466 | case -EROFS: | |
1467 | goto out_put_ctx; | |
1468 | default: | |
1469 | goto out_drop; | |
1470 | } | |
1471 | } | |
1472 | iput(inode); | |
898f635c | 1473 | if (inode != dentry->d_inode) |
535918f1 | 1474 | goto out_drop; |
898f635c TM |
1475 | |
1476 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1477 | ret = nfs_intent_set_file(nd, ctx); | |
1478 | if (ret >= 0) | |
1479 | ret = 1; | |
1da177e4 LT |
1480 | out: |
1481 | dput(parent); | |
1da177e4 | 1482 | return ret; |
535918f1 TM |
1483 | out_drop: |
1484 | d_drop(dentry); | |
1485 | ret = 0; | |
1486 | out_put_ctx: | |
1487 | put_nfs_open_context(ctx); | |
1488 | goto out; | |
1489 | ||
5584c306 | 1490 | no_open_dput: |
1da177e4 | 1491 | dput(parent); |
5584c306 | 1492 | no_open: |
1da177e4 LT |
1493 | return nfs_lookup_revalidate(dentry, nd); |
1494 | } | |
c0204fd2 TM |
1495 | |
1496 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, | |
1497 | struct nameidata *nd) | |
1498 | { | |
1499 | struct nfs_open_context *ctx = NULL; | |
1500 | struct iattr attr; | |
1501 | int error; | |
1502 | int open_flags = 0; | |
1503 | ||
1504 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1505 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1506 | ||
1507 | attr.ia_mode = mode; | |
1508 | attr.ia_valid = ATTR_MODE; | |
1509 | ||
1510 | if ((nd->flags & LOOKUP_CREATE) != 0) { | |
1511 | open_flags = nd->intent.open.flags; | |
1512 | ||
1513 | ctx = nameidata_to_nfs_open_context(dentry, nd); | |
1514 | error = PTR_ERR(ctx); | |
1515 | if (IS_ERR(ctx)) | |
898f635c | 1516 | goto out_err_drop; |
c0204fd2 TM |
1517 | } |
1518 | ||
1519 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1520 | if (error != 0) | |
1521 | goto out_put_ctx; | |
898f635c TM |
1522 | if (ctx != NULL) { |
1523 | error = nfs_intent_set_file(nd, ctx); | |
1524 | if (error < 0) | |
1525 | goto out_err; | |
1526 | } | |
c0204fd2 TM |
1527 | return 0; |
1528 | out_put_ctx: | |
1529 | if (ctx != NULL) | |
1530 | put_nfs_open_context(ctx); | |
898f635c | 1531 | out_err_drop: |
c0204fd2 | 1532 | d_drop(dentry); |
898f635c | 1533 | out_err: |
c0204fd2 TM |
1534 | return error; |
1535 | } | |
1536 | ||
1da177e4 LT |
1537 | #endif /* CONFIG_NFSV4 */ |
1538 | ||
1da177e4 LT |
1539 | /* |
1540 | * Code common to create, mkdir, and mknod. | |
1541 | */ | |
1542 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1543 | struct nfs_fattr *fattr) | |
1544 | { | |
fab728e1 TM |
1545 | struct dentry *parent = dget_parent(dentry); |
1546 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1547 | struct inode *inode; |
1548 | int error = -EACCES; | |
1549 | ||
fab728e1 TM |
1550 | d_drop(dentry); |
1551 | ||
1da177e4 LT |
1552 | /* We may have been initialized further down */ |
1553 | if (dentry->d_inode) | |
fab728e1 | 1554 | goto out; |
1da177e4 | 1555 | if (fhandle->size == 0) { |
1da177e4 LT |
1556 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1557 | if (error) | |
fab728e1 | 1558 | goto out_error; |
1da177e4 | 1559 | } |
5724ab37 | 1560 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1561 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1562 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1563 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1564 | if (error < 0) |
fab728e1 | 1565 | goto out_error; |
1da177e4 | 1566 | } |
1da177e4 | 1567 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1568 | error = PTR_ERR(inode); |
1569 | if (IS_ERR(inode)) | |
fab728e1 TM |
1570 | goto out_error; |
1571 | d_add(dentry, inode); | |
1572 | out: | |
1573 | dput(parent); | |
1da177e4 | 1574 | return 0; |
fab728e1 TM |
1575 | out_error: |
1576 | nfs_mark_for_revalidate(dir); | |
1577 | dput(parent); | |
1578 | return error; | |
1da177e4 LT |
1579 | } |
1580 | ||
1581 | /* | |
1582 | * Following a failed create operation, we drop the dentry rather | |
1583 | * than retain a negative dentry. This avoids a problem in the event | |
1584 | * that the operation succeeded on the server, but an error in the | |
1585 | * reply path made it appear to have failed. | |
1586 | */ | |
1587 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1588 | struct nameidata *nd) | |
1589 | { | |
1590 | struct iattr attr; | |
1591 | int error; | |
1da177e4 | 1592 | |
1e7cb3dc CL |
1593 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1594 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1595 | |
1596 | attr.ia_mode = mode; | |
1597 | attr.ia_valid = ATTR_MODE; | |
1598 | ||
c0204fd2 | 1599 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL); |
1da177e4 LT |
1600 | if (error != 0) |
1601 | goto out_err; | |
1da177e4 LT |
1602 | return 0; |
1603 | out_err: | |
1da177e4 LT |
1604 | d_drop(dentry); |
1605 | return error; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * See comments for nfs_proc_create regarding failed operations. | |
1610 | */ | |
1611 | static int | |
1612 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1613 | { | |
1614 | struct iattr attr; | |
1615 | int status; | |
1616 | ||
1e7cb3dc CL |
1617 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1618 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1619 | |
1620 | if (!new_valid_dev(rdev)) | |
1621 | return -EINVAL; | |
1622 | ||
1623 | attr.ia_mode = mode; | |
1624 | attr.ia_valid = ATTR_MODE; | |
1625 | ||
1da177e4 | 1626 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1627 | if (status != 0) |
1628 | goto out_err; | |
1da177e4 LT |
1629 | return 0; |
1630 | out_err: | |
1da177e4 LT |
1631 | d_drop(dentry); |
1632 | return status; | |
1633 | } | |
1634 | ||
1635 | /* | |
1636 | * See comments for nfs_proc_create regarding failed operations. | |
1637 | */ | |
1638 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1639 | { | |
1640 | struct iattr attr; | |
1641 | int error; | |
1642 | ||
1e7cb3dc CL |
1643 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1644 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1645 | |
1646 | attr.ia_valid = ATTR_MODE; | |
1647 | attr.ia_mode = mode | S_IFDIR; | |
1648 | ||
1da177e4 | 1649 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1650 | if (error != 0) |
1651 | goto out_err; | |
1da177e4 LT |
1652 | return 0; |
1653 | out_err: | |
1654 | d_drop(dentry); | |
1da177e4 LT |
1655 | return error; |
1656 | } | |
1657 | ||
d45b9d8b TM |
1658 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1659 | { | |
1660 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1661 | d_delete(dentry); | |
1662 | } | |
1663 | ||
1da177e4 LT |
1664 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1665 | { | |
1666 | int error; | |
1667 | ||
1e7cb3dc CL |
1668 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1669 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1670 | |
1da177e4 LT |
1671 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1672 | /* Ensure the VFS deletes this inode */ | |
1673 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1674 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1675 | else if (error == -ENOENT) |
1676 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1677 | |
1678 | return error; | |
1679 | } | |
1680 | ||
1da177e4 LT |
1681 | /* |
1682 | * Remove a file after making sure there are no pending writes, | |
1683 | * and after checking that the file has only one user. | |
1684 | * | |
1685 | * We invalidate the attribute cache and free the inode prior to the operation | |
1686 | * to avoid possible races if the server reuses the inode. | |
1687 | */ | |
1688 | static int nfs_safe_remove(struct dentry *dentry) | |
1689 | { | |
1690 | struct inode *dir = dentry->d_parent->d_inode; | |
1691 | struct inode *inode = dentry->d_inode; | |
1692 | int error = -EBUSY; | |
1693 | ||
1694 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1695 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1696 | ||
1697 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1698 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1699 | error = 0; | |
1700 | goto out; | |
1701 | } | |
1702 | ||
1da177e4 | 1703 | if (inode != NULL) { |
cae7a073 | 1704 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1705 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1706 | /* The VFS may want to delete this inode */ | |
1707 | if (error == 0) | |
1b83d707 | 1708 | nfs_drop_nlink(inode); |
5ba7cc48 | 1709 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1710 | } else |
1711 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1712 | if (error == -ENOENT) |
1713 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1714 | out: |
1715 | return error; | |
1716 | } | |
1717 | ||
1718 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1719 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1720 | * | |
1721 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1722 | */ | |
1723 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1724 | { | |
1725 | int error; | |
1726 | int need_rehash = 0; | |
1727 | ||
1728 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1729 | dir->i_ino, dentry->d_name.name); | |
1730 | ||
1da177e4 LT |
1731 | spin_lock(&dcache_lock); |
1732 | spin_lock(&dentry->d_lock); | |
1733 | if (atomic_read(&dentry->d_count) > 1) { | |
1734 | spin_unlock(&dentry->d_lock); | |
1735 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1736 | /* Start asynchronous writeout of the inode */ |
1737 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1738 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1739 | return error; |
1740 | } | |
1741 | if (!d_unhashed(dentry)) { | |
1742 | __d_drop(dentry); | |
1743 | need_rehash = 1; | |
1744 | } | |
1745 | spin_unlock(&dentry->d_lock); | |
1746 | spin_unlock(&dcache_lock); | |
1747 | error = nfs_safe_remove(dentry); | |
d45b9d8b | 1748 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1749 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1750 | } else if (need_rehash) | |
1751 | d_rehash(dentry); | |
1da177e4 LT |
1752 | return error; |
1753 | } | |
1754 | ||
873101b3 CL |
1755 | /* |
1756 | * To create a symbolic link, most file systems instantiate a new inode, | |
1757 | * add a page to it containing the path, then write it out to the disk | |
1758 | * using prepare_write/commit_write. | |
1759 | * | |
1760 | * Unfortunately the NFS client can't create the in-core inode first | |
1761 | * because it needs a file handle to create an in-core inode (see | |
1762 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1763 | * symlink request has completed on the server. | |
1764 | * | |
1765 | * So instead we allocate a raw page, copy the symname into it, then do | |
1766 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1767 | * now have a new file handle and can instantiate an in-core NFS inode | |
1768 | * and move the raw page into its mapping. | |
1769 | */ | |
1770 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1771 | { |
873101b3 CL |
1772 | struct pagevec lru_pvec; |
1773 | struct page *page; | |
1774 | char *kaddr; | |
1da177e4 | 1775 | struct iattr attr; |
873101b3 | 1776 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1777 | int error; |
1778 | ||
1779 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1780 | dir->i_ino, dentry->d_name.name, symname); | |
1781 | ||
873101b3 CL |
1782 | if (pathlen > PAGE_SIZE) |
1783 | return -ENAMETOOLONG; | |
1da177e4 | 1784 | |
873101b3 CL |
1785 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1786 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1787 | |
83d93f22 | 1788 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1789 | if (!page) |
873101b3 | 1790 | return -ENOMEM; |
873101b3 CL |
1791 | |
1792 | kaddr = kmap_atomic(page, KM_USER0); | |
1793 | memcpy(kaddr, symname, pathlen); | |
1794 | if (pathlen < PAGE_SIZE) | |
1795 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1796 | kunmap_atomic(kaddr, KM_USER0); | |
1797 | ||
94a6d753 | 1798 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1799 | if (error != 0) { |
1800 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1801 | dir->i_sb->s_id, dir->i_ino, | |
1802 | dentry->d_name.name, symname, error); | |
1da177e4 | 1803 | d_drop(dentry); |
873101b3 | 1804 | __free_page(page); |
873101b3 CL |
1805 | return error; |
1806 | } | |
1807 | ||
1808 | /* | |
1809 | * No big deal if we can't add this page to the page cache here. | |
1810 | * READLINK will get the missing page from the server if needed. | |
1811 | */ | |
1812 | pagevec_init(&lru_pvec, 0); | |
1813 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1814 | GFP_KERNEL)) { | |
39cf8a13 | 1815 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1816 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1817 | SetPageUptodate(page); |
1818 | unlock_page(page); | |
1819 | } else | |
1820 | __free_page(page); | |
1821 | ||
873101b3 | 1822 | return 0; |
1da177e4 LT |
1823 | } |
1824 | ||
1825 | static int | |
1826 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1827 | { | |
1828 | struct inode *inode = old_dentry->d_inode; | |
1829 | int error; | |
1830 | ||
1831 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1832 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1833 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1834 | ||
9a3936aa TM |
1835 | nfs_inode_return_delegation(inode); |
1836 | ||
9697d234 | 1837 | d_drop(dentry); |
1da177e4 | 1838 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1839 | if (error == 0) { |
7de9c6ee | 1840 | ihold(inode); |
9697d234 | 1841 | d_add(dentry, inode); |
cf809556 | 1842 | } |
1da177e4 LT |
1843 | return error; |
1844 | } | |
1845 | ||
1846 | /* | |
1847 | * RENAME | |
1848 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1849 | * different file handle for the same inode after a rename (e.g. when | |
1850 | * moving to a different directory). A fail-safe method to do so would | |
1851 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1852 | * rename the old file using the sillyrename stuff. This way, the original | |
1853 | * file in old_dir will go away when the last process iput()s the inode. | |
1854 | * | |
1855 | * FIXED. | |
1856 | * | |
1857 | * It actually works quite well. One needs to have the possibility for | |
1858 | * at least one ".nfs..." file in each directory the file ever gets | |
1859 | * moved or linked to which happens automagically with the new | |
1860 | * implementation that only depends on the dcache stuff instead of | |
1861 | * using the inode layer | |
1862 | * | |
1863 | * Unfortunately, things are a little more complicated than indicated | |
1864 | * above. For a cross-directory move, we want to make sure we can get | |
1865 | * rid of the old inode after the operation. This means there must be | |
1866 | * no pending writes (if it's a file), and the use count must be 1. | |
1867 | * If these conditions are met, we can drop the dentries before doing | |
1868 | * the rename. | |
1869 | */ | |
1870 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1871 | struct inode *new_dir, struct dentry *new_dentry) | |
1872 | { | |
1873 | struct inode *old_inode = old_dentry->d_inode; | |
1874 | struct inode *new_inode = new_dentry->d_inode; | |
1875 | struct dentry *dentry = NULL, *rehash = NULL; | |
1876 | int error = -EBUSY; | |
1877 | ||
1da177e4 LT |
1878 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1879 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1880 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1881 | atomic_read(&new_dentry->d_count)); | |
1882 | ||
1883 | /* | |
28f79a1a MS |
1884 | * For non-directories, check whether the target is busy and if so, |
1885 | * make a copy of the dentry and then do a silly-rename. If the | |
1886 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1887 | * the new target. | |
1da177e4 | 1888 | */ |
27226104 MS |
1889 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1890 | /* | |
1891 | * To prevent any new references to the target during the | |
1892 | * rename, we unhash the dentry in advance. | |
1893 | */ | |
1894 | if (!d_unhashed(new_dentry)) { | |
1895 | d_drop(new_dentry); | |
1896 | rehash = new_dentry; | |
1897 | } | |
1da177e4 | 1898 | |
27226104 MS |
1899 | if (atomic_read(&new_dentry->d_count) > 2) { |
1900 | int err; | |
1901 | ||
1902 | /* copy the target dentry's name */ | |
1903 | dentry = d_alloc(new_dentry->d_parent, | |
1904 | &new_dentry->d_name); | |
1905 | if (!dentry) | |
1906 | goto out; | |
1907 | ||
1908 | /* silly-rename the existing target ... */ | |
1909 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1910 | if (err) |
27226104 | 1911 | goto out; |
24e93025 MS |
1912 | |
1913 | new_dentry = dentry; | |
56335936 | 1914 | rehash = NULL; |
24e93025 | 1915 | new_inode = NULL; |
27226104 | 1916 | } |
b1e4adf4 | 1917 | } |
1da177e4 | 1918 | |
cae7a073 | 1919 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 1920 | if (new_inode != NULL) |
24174119 | 1921 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 1922 | |
1da177e4 LT |
1923 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1924 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1925 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1926 | out: |
1927 | if (rehash) | |
1928 | d_rehash(rehash); | |
1929 | if (!error) { | |
b1e4adf4 TM |
1930 | if (new_inode != NULL) |
1931 | nfs_drop_nlink(new_inode); | |
349457cc | 1932 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1933 | nfs_set_verifier(new_dentry, |
1934 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1935 | } else if (error == -ENOENT) |
1936 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1937 | |
1938 | /* new dentry created? */ | |
1939 | if (dentry) | |
1940 | dput(dentry); | |
1da177e4 LT |
1941 | return error; |
1942 | } | |
1943 | ||
cfcea3e8 TM |
1944 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1945 | static LIST_HEAD(nfs_access_lru_list); | |
1946 | static atomic_long_t nfs_access_nr_entries; | |
1947 | ||
1c3c07e9 TM |
1948 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1949 | { | |
1950 | put_rpccred(entry->cred); | |
1951 | kfree(entry); | |
cfcea3e8 TM |
1952 | smp_mb__before_atomic_dec(); |
1953 | atomic_long_dec(&nfs_access_nr_entries); | |
1954 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1955 | } |
1956 | ||
1a81bb8a TM |
1957 | static void nfs_access_free_list(struct list_head *head) |
1958 | { | |
1959 | struct nfs_access_entry *cache; | |
1960 | ||
1961 | while (!list_empty(head)) { | |
1962 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1963 | list_del(&cache->lru); | |
1964 | nfs_access_free_entry(cache); | |
1965 | } | |
1966 | } | |
1967 | ||
7f8275d0 | 1968 | int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
979df72e TM |
1969 | { |
1970 | LIST_HEAD(head); | |
aa510da5 | 1971 | struct nfs_inode *nfsi, *next; |
979df72e TM |
1972 | struct nfs_access_entry *cache; |
1973 | ||
61d5eb29 TM |
1974 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1975 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 1976 | |
a50f7951 | 1977 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 1978 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
1979 | struct inode *inode; |
1980 | ||
1981 | if (nr_to_scan-- == 0) | |
1982 | break; | |
9c7e7e23 | 1983 | inode = &nfsi->vfs_inode; |
979df72e TM |
1984 | spin_lock(&inode->i_lock); |
1985 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1986 | goto remove_lru_entry; | |
1987 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1988 | struct nfs_access_entry, lru); | |
1989 | list_move(&cache->lru, &head); | |
1990 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1991 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1992 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1993 | &nfs_access_lru_list); | |
1994 | else { | |
1995 | remove_lru_entry: | |
1996 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 1997 | smp_mb__before_clear_bit(); |
979df72e | 1998 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 1999 | smp_mb__after_clear_bit(); |
979df72e | 2000 | } |
59844a9b | 2001 | spin_unlock(&inode->i_lock); |
979df72e TM |
2002 | } |
2003 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2004 | nfs_access_free_list(&head); |
979df72e TM |
2005 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
2006 | } | |
2007 | ||
1a81bb8a | 2008 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2009 | { |
1c3c07e9 | 2010 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2011 | struct rb_node *n; |
1c3c07e9 TM |
2012 | struct nfs_access_entry *entry; |
2013 | ||
2014 | /* Unhook entries from the cache */ | |
2015 | while ((n = rb_first(root_node)) != NULL) { | |
2016 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2017 | rb_erase(n, root_node); | |
1a81bb8a | 2018 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2019 | } |
2020 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2021 | } |
2022 | ||
1c3c07e9 | 2023 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2024 | { |
1a81bb8a TM |
2025 | LIST_HEAD(head); |
2026 | ||
2027 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2028 | return; | |
cfcea3e8 | 2029 | /* Remove from global LRU init */ |
1a81bb8a TM |
2030 | spin_lock(&nfs_access_lru_lock); |
2031 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2032 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2033 | |
1c3c07e9 | 2034 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2035 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2036 | spin_unlock(&inode->i_lock); | |
2037 | spin_unlock(&nfs_access_lru_lock); | |
2038 | nfs_access_free_list(&head); | |
1c3c07e9 | 2039 | } |
1da177e4 | 2040 | |
1c3c07e9 TM |
2041 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2042 | { | |
2043 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2044 | struct nfs_access_entry *entry; | |
2045 | ||
2046 | while (n != NULL) { | |
2047 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2048 | ||
2049 | if (cred < entry->cred) | |
2050 | n = n->rb_left; | |
2051 | else if (cred > entry->cred) | |
2052 | n = n->rb_right; | |
2053 | else | |
2054 | return entry; | |
1da177e4 | 2055 | } |
1c3c07e9 TM |
2056 | return NULL; |
2057 | } | |
2058 | ||
af22f94a | 2059 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2060 | { |
2061 | struct nfs_inode *nfsi = NFS_I(inode); | |
2062 | struct nfs_access_entry *cache; | |
2063 | int err = -ENOENT; | |
2064 | ||
dc59250c | 2065 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2066 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2067 | goto out_zap; | |
2068 | cache = nfs_access_search_rbtree(inode, cred); | |
2069 | if (cache == NULL) | |
2070 | goto out; | |
b4d2314b | 2071 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2072 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2073 | goto out_stale; |
2074 | res->jiffies = cache->jiffies; | |
2075 | res->cred = cache->cred; | |
2076 | res->mask = cache->mask; | |
cfcea3e8 | 2077 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2078 | err = 0; |
2079 | out: | |
2080 | spin_unlock(&inode->i_lock); | |
2081 | return err; | |
2082 | out_stale: | |
2083 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2084 | list_del(&cache->lru); |
1c3c07e9 TM |
2085 | spin_unlock(&inode->i_lock); |
2086 | nfs_access_free_entry(cache); | |
2087 | return -ENOENT; | |
2088 | out_zap: | |
1a81bb8a TM |
2089 | spin_unlock(&inode->i_lock); |
2090 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2091 | return -ENOENT; |
2092 | } | |
2093 | ||
2094 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2095 | { | |
cfcea3e8 TM |
2096 | struct nfs_inode *nfsi = NFS_I(inode); |
2097 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2098 | struct rb_node **p = &root_node->rb_node; |
2099 | struct rb_node *parent = NULL; | |
2100 | struct nfs_access_entry *entry; | |
2101 | ||
2102 | spin_lock(&inode->i_lock); | |
2103 | while (*p != NULL) { | |
2104 | parent = *p; | |
2105 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2106 | ||
2107 | if (set->cred < entry->cred) | |
2108 | p = &parent->rb_left; | |
2109 | else if (set->cred > entry->cred) | |
2110 | p = &parent->rb_right; | |
2111 | else | |
2112 | goto found; | |
2113 | } | |
2114 | rb_link_node(&set->rb_node, parent, p); | |
2115 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2116 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2117 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2118 | return; |
2119 | found: | |
2120 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2121 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2122 | list_del(&entry->lru); | |
1c3c07e9 TM |
2123 | spin_unlock(&inode->i_lock); |
2124 | nfs_access_free_entry(entry); | |
2125 | } | |
2126 | ||
af22f94a | 2127 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2128 | { |
2129 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2130 | if (cache == NULL) | |
2131 | return; | |
2132 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2133 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2134 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2135 | cache->mask = set->mask; |
1c3c07e9 TM |
2136 | |
2137 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2138 | |
2139 | /* Update accounting */ | |
2140 | smp_mb__before_atomic_inc(); | |
2141 | atomic_long_inc(&nfs_access_nr_entries); | |
2142 | smp_mb__after_atomic_inc(); | |
2143 | ||
2144 | /* Add inode to global LRU list */ | |
1a81bb8a | 2145 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2146 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2147 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2148 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2149 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2150 | spin_unlock(&nfs_access_lru_lock); |
2151 | } | |
1da177e4 LT |
2152 | } |
2153 | ||
2154 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2155 | { | |
2156 | struct nfs_access_entry cache; | |
2157 | int status; | |
2158 | ||
2159 | status = nfs_access_get_cached(inode, cred, &cache); | |
2160 | if (status == 0) | |
2161 | goto out; | |
2162 | ||
2163 | /* Be clever: ask server to check for all possible rights */ | |
2164 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2165 | cache.cred = cred; | |
2166 | cache.jiffies = jiffies; | |
2167 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2168 | if (status != 0) { |
2169 | if (status == -ESTALE) { | |
2170 | nfs_zap_caches(inode); | |
2171 | if (!S_ISDIR(inode->i_mode)) | |
2172 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2173 | } | |
1da177e4 | 2174 | return status; |
a71ee337 | 2175 | } |
1da177e4 LT |
2176 | nfs_access_add_cache(inode, &cache); |
2177 | out: | |
e6305c43 | 2178 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2179 | return 0; |
2180 | return -EACCES; | |
2181 | } | |
2182 | ||
af22f94a TM |
2183 | static int nfs_open_permission_mask(int openflags) |
2184 | { | |
2185 | int mask = 0; | |
2186 | ||
2187 | if (openflags & FMODE_READ) | |
2188 | mask |= MAY_READ; | |
2189 | if (openflags & FMODE_WRITE) | |
2190 | mask |= MAY_WRITE; | |
2191 | if (openflags & FMODE_EXEC) | |
2192 | mask |= MAY_EXEC; | |
2193 | return mask; | |
2194 | } | |
2195 | ||
2196 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2197 | { | |
2198 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2199 | } | |
2200 | ||
e6305c43 | 2201 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2202 | { |
2203 | struct rpc_cred *cred; | |
2204 | int res = 0; | |
2205 | ||
91d5b470 CL |
2206 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2207 | ||
e6305c43 | 2208 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2209 | goto out; |
2210 | /* Is this sys_access() ? */ | |
9cfcac81 | 2211 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2212 | goto force_lookup; |
2213 | ||
2214 | switch (inode->i_mode & S_IFMT) { | |
2215 | case S_IFLNK: | |
2216 | goto out; | |
2217 | case S_IFREG: | |
2218 | /* NFSv4 has atomic_open... */ | |
2219 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2220 | && (mask & MAY_OPEN) |
2221 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2222 | goto out; |
2223 | break; | |
2224 | case S_IFDIR: | |
2225 | /* | |
2226 | * Optimize away all write operations, since the server | |
2227 | * will check permissions when we perform the op. | |
2228 | */ | |
2229 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2230 | goto out; | |
2231 | } | |
2232 | ||
2233 | force_lookup: | |
1da177e4 LT |
2234 | if (!NFS_PROTO(inode)->access) |
2235 | goto out_notsup; | |
2236 | ||
98a8e323 | 2237 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2238 | if (!IS_ERR(cred)) { |
2239 | res = nfs_do_access(inode, cred, mask); | |
2240 | put_rpccred(cred); | |
2241 | } else | |
2242 | res = PTR_ERR(cred); | |
1da177e4 | 2243 | out: |
f696a365 MS |
2244 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2245 | res = -EACCES; | |
2246 | ||
1e7cb3dc CL |
2247 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2248 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2249 | return res; |
2250 | out_notsup: | |
2251 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2252 | if (res == 0) | |
2253 | res = generic_permission(inode, mask, NULL); | |
1e7cb3dc | 2254 | goto out; |
1da177e4 LT |
2255 | } |
2256 | ||
2257 | /* | |
2258 | * Local variables: | |
2259 | * version-control: t | |
2260 | * kept-new-versions: 5 | |
2261 | * End: | |
2262 | */ |