]>
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 | { | |
d39ab9de BS |
398 | if (dentry->d_inode == NULL) |
399 | goto different; | |
37a09f07 | 400 | if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) |
d39ab9de BS |
401 | goto different; |
402 | return 1; | |
403 | different: | |
404 | return 0; | |
405 | } | |
406 | ||
407 | static | |
408 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
409 | { | |
4a201d6e TM |
410 | struct qstr filename = { |
411 | .len = entry->len, | |
412 | .name = entry->name, | |
413 | }; | |
414 | struct dentry *dentry; | |
415 | struct dentry *alias; | |
d39ab9de BS |
416 | struct inode *dir = parent->d_inode; |
417 | struct inode *inode; | |
418 | ||
4a201d6e TM |
419 | if (filename.name[0] == '.') { |
420 | if (filename.len == 1) | |
421 | return; | |
422 | if (filename.len == 2 && filename.name[1] == '.') | |
423 | return; | |
424 | } | |
425 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 426 | |
4a201d6e | 427 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
428 | if (dentry != NULL) { |
429 | if (nfs_same_file(dentry, entry)) { | |
430 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
431 | goto out; | |
432 | } else { | |
433 | d_drop(dentry); | |
434 | dput(dentry); | |
435 | } | |
436 | } | |
437 | ||
438 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
439 | if (dentry == NULL) |
440 | return; | |
441 | ||
d39ab9de BS |
442 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
443 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
444 | if (IS_ERR(inode)) | |
445 | goto out; | |
446 | ||
447 | alias = d_materialise_unique(dentry, inode); | |
448 | if (IS_ERR(alias)) | |
449 | goto out; | |
450 | else if (alias) { | |
451 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
452 | dput(alias); | |
453 | } else | |
454 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
455 | ||
456 | out: | |
457 | dput(dentry); | |
d39ab9de BS |
458 | } |
459 | ||
d1bacf9e BS |
460 | /* Perform conversion from xdr to cache array */ |
461 | static | |
8cd51a0c | 462 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
56e4ebf8 | 463 | void *xdr_page, struct page *page, unsigned int buflen) |
1da177e4 | 464 | { |
babddc72 BS |
465 | struct xdr_stream stream; |
466 | struct xdr_buf buf; | |
56e4ebf8 | 467 | __be32 *ptr = xdr_page; |
99424380 | 468 | struct nfs_cache_array *array; |
5c346854 TM |
469 | unsigned int count = 0; |
470 | int status; | |
babddc72 BS |
471 | |
472 | buf.head->iov_base = xdr_page; | |
473 | buf.head->iov_len = buflen; | |
474 | buf.tail->iov_len = 0; | |
475 | buf.page_base = 0; | |
476 | buf.page_len = 0; | |
477 | buf.buflen = buf.head->iov_len; | |
478 | buf.len = buf.head->iov_len; | |
479 | ||
480 | xdr_init_decode(&stream, &buf, ptr); | |
481 | ||
99424380 BS |
482 | |
483 | do { | |
484 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
485 | if (status != 0) { |
486 | if (status == -EAGAIN) | |
487 | status = 0; | |
99424380 | 488 | break; |
8cd51a0c | 489 | } |
99424380 | 490 | |
5c346854 TM |
491 | count++; |
492 | ||
d39ab9de BS |
493 | if (desc->plus == 1) |
494 | nfs_prime_dcache(desc->file->f_path.dentry, entry); | |
8cd51a0c TM |
495 | |
496 | status = nfs_readdir_add_to_array(entry, page); | |
497 | if (status != 0) | |
498 | break; | |
99424380 BS |
499 | } while (!entry->eof); |
500 | ||
5c346854 | 501 | if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) { |
99424380 | 502 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
503 | if (!IS_ERR(array)) { |
504 | array->eof_index = array->size; | |
505 | status = 0; | |
506 | nfs_readdir_release_array(page); | |
5c346854 TM |
507 | } else |
508 | status = PTR_ERR(array); | |
1da177e4 | 509 | } |
8cd51a0c | 510 | return status; |
56e4ebf8 BS |
511 | } |
512 | ||
513 | static | |
514 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
515 | { | |
516 | unsigned int i; | |
517 | for (i = 0; i < npages; i++) | |
518 | put_page(pages[i]); | |
519 | } | |
520 | ||
521 | static | |
522 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
523 | unsigned int npages) | |
524 | { | |
525 | vm_unmap_ram(ptr, npages); | |
526 | nfs_readdir_free_pagearray(pages, npages); | |
527 | } | |
528 | ||
529 | /* | |
530 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
531 | * to nfs_readdir_free_large_page | |
532 | */ | |
533 | static | |
534 | void *nfs_readdir_large_page(struct page **pages, unsigned int npages) | |
535 | { | |
536 | void *ptr; | |
537 | unsigned int i; | |
538 | ||
539 | for (i = 0; i < npages; i++) { | |
540 | struct page *page = alloc_page(GFP_KERNEL); | |
541 | if (page == NULL) | |
542 | goto out_freepages; | |
543 | pages[i] = page; | |
544 | } | |
545 | ||
4a201d6e | 546 | ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL); |
56e4ebf8 BS |
547 | if (!IS_ERR_OR_NULL(ptr)) |
548 | return ptr; | |
549 | out_freepages: | |
550 | nfs_readdir_free_pagearray(pages, i); | |
551 | return NULL; | |
1da177e4 LT |
552 | } |
553 | ||
d1bacf9e BS |
554 | static |
555 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 556 | { |
56e4ebf8 BS |
557 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
558 | void *pages_ptr = NULL; | |
d1bacf9e BS |
559 | struct nfs_entry entry; |
560 | struct file *file = desc->file; | |
561 | struct nfs_cache_array *array; | |
8cd51a0c | 562 | int status = -ENOMEM; |
56e4ebf8 | 563 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
564 | |
565 | entry.prev_cookie = 0; | |
566 | entry.cookie = *desc->dir_cookie; | |
567 | entry.eof = 0; | |
568 | entry.fh = nfs_alloc_fhandle(); | |
569 | entry.fattr = nfs_alloc_fattr(); | |
570 | if (entry.fh == NULL || entry.fattr == NULL) | |
571 | goto out; | |
00a92642 | 572 | |
d1bacf9e | 573 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
574 | if (IS_ERR(array)) { |
575 | status = PTR_ERR(array); | |
576 | goto out; | |
577 | } | |
d1bacf9e BS |
578 | memset(array, 0, sizeof(struct nfs_cache_array)); |
579 | array->eof_index = -1; | |
00a92642 | 580 | |
56e4ebf8 BS |
581 | pages_ptr = nfs_readdir_large_page(pages, array_size); |
582 | if (!pages_ptr) | |
d1bacf9e BS |
583 | goto out_release_array; |
584 | do { | |
ac396128 | 585 | unsigned int pglen; |
56e4ebf8 | 586 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 587 | |
d1bacf9e | 588 | if (status < 0) |
00a92642 | 589 | break; |
ac396128 TM |
590 | pglen = status; |
591 | status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen); | |
8cd51a0c TM |
592 | if (status < 0) { |
593 | if (status == -ENOSPC) | |
594 | status = 0; | |
595 | break; | |
596 | } | |
597 | } while (array->eof_index < 0); | |
d1bacf9e | 598 | |
56e4ebf8 | 599 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
600 | out_release_array: |
601 | nfs_readdir_release_array(page); | |
602 | out: | |
603 | nfs_free_fattr(entry.fattr); | |
604 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
605 | return status; |
606 | } | |
607 | ||
608 | /* | |
d1bacf9e BS |
609 | * Now we cache directories properly, by converting xdr information |
610 | * to an array that can be used for lookups later. This results in | |
611 | * fewer cache pages, since we can store more information on each page. | |
612 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 613 | */ |
d1bacf9e BS |
614 | static |
615 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 616 | { |
01cce933 | 617 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
8cd51a0c | 618 | int ret; |
1da177e4 | 619 | |
8cd51a0c TM |
620 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
621 | if (ret < 0) | |
d1bacf9e BS |
622 | goto error; |
623 | SetPageUptodate(page); | |
1da177e4 | 624 | |
d1bacf9e BS |
625 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
626 | /* Should never happen */ | |
627 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 628 | } |
d1bacf9e BS |
629 | unlock_page(page); |
630 | return 0; | |
631 | error: | |
632 | unlock_page(page); | |
8cd51a0c | 633 | return ret; |
d1bacf9e | 634 | } |
1da177e4 | 635 | |
d1bacf9e BS |
636 | static |
637 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
638 | { | |
639 | page_cache_release(desc->page); | |
640 | desc->page = NULL; | |
641 | } | |
642 | ||
643 | static | |
644 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
645 | { | |
8cd51a0c | 646 | return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, |
d1bacf9e | 647 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
648 | } |
649 | ||
650 | /* | |
d1bacf9e | 651 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 652 | */ |
d1bacf9e BS |
653 | static |
654 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
655 | { | |
656 | int res; | |
657 | ||
658 | desc->page = get_cache_page(desc); | |
659 | if (IS_ERR(desc->page)) | |
660 | return PTR_ERR(desc->page); | |
661 | ||
662 | res = nfs_readdir_search_array(desc); | |
663 | if (res == 0) | |
664 | return 0; | |
665 | cache_page_release(desc); | |
666 | return res; | |
667 | } | |
668 | ||
669 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
670 | static inline |
671 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
672 | { | |
8cd51a0c | 673 | int res; |
d1bacf9e | 674 | |
8cd51a0c TM |
675 | if (desc->page_index == 0) |
676 | desc->current_index = 0; | |
d1bacf9e BS |
677 | while (1) { |
678 | res = find_cache_page(desc); | |
1da177e4 LT |
679 | if (res != -EAGAIN) |
680 | break; | |
d1bacf9e | 681 | desc->page_index++; |
1da177e4 | 682 | } |
1da177e4 LT |
683 | return res; |
684 | } | |
685 | ||
686 | static inline unsigned int dt_type(struct inode *inode) | |
687 | { | |
688 | return (inode->i_mode >> 12) & 15; | |
689 | } | |
690 | ||
1da177e4 LT |
691 | /* |
692 | * Once we've found the start of the dirent within a page: fill 'er up... | |
693 | */ | |
694 | static | |
695 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
696 | filldir_t filldir) | |
697 | { | |
698 | struct file *file = desc->file; | |
d1bacf9e BS |
699 | int i = 0; |
700 | int res = 0; | |
701 | struct nfs_cache_array *array = NULL; | |
d1bacf9e BS |
702 | |
703 | array = nfs_readdir_get_array(desc->page); | |
e7c58e97 TM |
704 | if (IS_ERR(array)) { |
705 | res = PTR_ERR(array); | |
706 | goto out; | |
707 | } | |
d1bacf9e BS |
708 | |
709 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 710 | struct nfs_cache_array_entry *ent; |
1da177e4 | 711 | |
ece0b423 TM |
712 | ent = &array->array[i]; |
713 | if (filldir(dirent, ent->string.name, ent->string.len, | |
0b26a0bf TM |
714 | file->f_pos, nfs_compat_user_ino64(ent->ino), |
715 | ent->d_type) < 0) { | |
ece0b423 | 716 | desc->eof = 1; |
1da177e4 | 717 | break; |
ece0b423 | 718 | } |
00a92642 | 719 | file->f_pos++; |
d1bacf9e BS |
720 | desc->cache_entry_index = i; |
721 | if (i < (array->size-1)) | |
722 | *desc->dir_cookie = array->array[i+1].cookie; | |
723 | else | |
724 | *desc->dir_cookie = array->last_cookie; | |
1da177e4 | 725 | } |
8cd51a0c TM |
726 | if (i == array->eof_index) |
727 | desc->eof = 1; | |
d1bacf9e BS |
728 | |
729 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 730 | out: |
d1bacf9e | 731 | cache_page_release(desc); |
1e7cb3dc CL |
732 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
733 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
734 | return res; |
735 | } | |
736 | ||
737 | /* | |
738 | * If we cannot find a cookie in our cache, we suspect that this is | |
739 | * because it points to a deleted file, so we ask the server to return | |
740 | * whatever it thinks is the next entry. We then feed this to filldir. | |
741 | * If all goes well, we should then be able to find our way round the | |
742 | * cache on the next call to readdir_search_pagecache(); | |
743 | * | |
744 | * NOTE: we cannot add the anonymous page to the pagecache because | |
745 | * the data it contains might not be page aligned. Besides, | |
746 | * we should already have a complete representation of the | |
747 | * directory in the page cache by the time we get here. | |
748 | */ | |
749 | static inline | |
750 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
751 | filldir_t filldir) | |
752 | { | |
1da177e4 LT |
753 | struct page *page = NULL; |
754 | int status; | |
d1bacf9e | 755 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 756 | |
1e7cb3dc CL |
757 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
758 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
759 | |
760 | page = alloc_page(GFP_HIGHUSER); | |
761 | if (!page) { | |
762 | status = -ENOMEM; | |
763 | goto out; | |
764 | } | |
d1bacf9e | 765 | |
7a8e1dc3 TM |
766 | desc->page_index = 0; |
767 | desc->page = page; | |
768 | ||
85f8607e TM |
769 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
770 | if (status < 0) | |
1da177e4 LT |
771 | goto out_release; |
772 | ||
773 | status = nfs_do_filldir(desc, dirent, filldir); | |
774 | ||
1da177e4 | 775 | out: |
1e7cb3dc | 776 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 777 | __func__, status); |
1da177e4 LT |
778 | return status; |
779 | out_release: | |
d1bacf9e | 780 | cache_page_release(desc); |
1da177e4 LT |
781 | goto out; |
782 | } | |
783 | ||
00a92642 OG |
784 | /* The file offset position represents the dirent entry number. A |
785 | last cookie cache takes care of the common case of reading the | |
786 | whole directory. | |
1da177e4 LT |
787 | */ |
788 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
789 | { | |
01cce933 | 790 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
791 | struct inode *inode = dentry->d_inode; |
792 | nfs_readdir_descriptor_t my_desc, | |
793 | *desc = &my_desc; | |
aa49b4cf | 794 | int res = -ENOMEM; |
1da177e4 | 795 | |
6da24bc9 | 796 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
797 | dentry->d_parent->d_name.name, dentry->d_name.name, |
798 | (long long)filp->f_pos); | |
91d5b470 CL |
799 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
800 | ||
1da177e4 | 801 | /* |
00a92642 | 802 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 803 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
804 | * to either find the entry with the appropriate number or |
805 | * revalidate the cookie. | |
1da177e4 LT |
806 | */ |
807 | memset(desc, 0, sizeof(*desc)); | |
808 | ||
809 | desc->file = filp; | |
cd3758e3 | 810 | desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; |
1da177e4 LT |
811 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
812 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
813 | ||
565277f6 | 814 | nfs_block_sillyrename(dentry); |
1cda707d | 815 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
816 | if (res < 0) |
817 | goto out; | |
818 | ||
d1bacf9e | 819 | while (desc->eof != 1) { |
1da177e4 | 820 | res = readdir_search_pagecache(desc); |
00a92642 | 821 | |
1da177e4 | 822 | if (res == -EBADCOOKIE) { |
ece0b423 | 823 | res = 0; |
1da177e4 | 824 | /* This means either end of directory */ |
d1bacf9e | 825 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
826 | /* Or that the server has 'lost' a cookie */ |
827 | res = uncached_readdir(desc, dirent, filldir); | |
ece0b423 | 828 | if (res == 0) |
1da177e4 LT |
829 | continue; |
830 | } | |
1da177e4 LT |
831 | break; |
832 | } | |
833 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 834 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 835 | nfs_zap_caches(inode); |
baf57a09 | 836 | desc->page_index = 0; |
1da177e4 | 837 | desc->plus = 0; |
d1bacf9e | 838 | desc->eof = 0; |
1da177e4 LT |
839 | continue; |
840 | } | |
841 | if (res < 0) | |
842 | break; | |
843 | ||
844 | res = nfs_do_filldir(desc, dirent, filldir); | |
ece0b423 | 845 | if (res < 0) |
1da177e4 | 846 | break; |
1da177e4 | 847 | } |
fccca7fc | 848 | out: |
565277f6 | 849 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
850 | if (res > 0) |
851 | res = 0; | |
aa49b4cf | 852 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
853 | dentry->d_parent->d_name.name, dentry->d_name.name, |
854 | res); | |
855 | return res; | |
1da177e4 LT |
856 | } |
857 | ||
10afec90 | 858 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 859 | { |
b84e06c5 CL |
860 | struct dentry *dentry = filp->f_path.dentry; |
861 | struct inode *inode = dentry->d_inode; | |
862 | ||
6da24bc9 | 863 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
864 | dentry->d_parent->d_name.name, |
865 | dentry->d_name.name, | |
866 | offset, origin); | |
867 | ||
868 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
869 | switch (origin) { |
870 | case 1: | |
871 | offset += filp->f_pos; | |
872 | case 0: | |
873 | if (offset >= 0) | |
874 | break; | |
875 | default: | |
876 | offset = -EINVAL; | |
877 | goto out; | |
878 | } | |
879 | if (offset != filp->f_pos) { | |
880 | filp->f_pos = offset; | |
cd3758e3 | 881 | nfs_file_open_context(filp)->dir_cookie = 0; |
f0dd2136 TM |
882 | } |
883 | out: | |
b84e06c5 | 884 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
885 | return offset; |
886 | } | |
887 | ||
1da177e4 LT |
888 | /* |
889 | * All directory operations under NFS are synchronous, so fsync() | |
890 | * is a dummy operation. | |
891 | */ | |
7ea80859 | 892 | static int nfs_fsync_dir(struct file *filp, int datasync) |
1da177e4 | 893 | { |
7ea80859 CH |
894 | struct dentry *dentry = filp->f_path.dentry; |
895 | ||
6da24bc9 | 896 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
897 | dentry->d_parent->d_name.name, dentry->d_name.name, |
898 | datasync); | |
899 | ||
54917786 | 900 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
1da177e4 LT |
901 | return 0; |
902 | } | |
903 | ||
bfc69a45 TM |
904 | /** |
905 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
906 | * @dir - pointer to directory inode | |
907 | * | |
908 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
909 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
910 | * on the server that might have invalidated our dcache. | |
911 | * | |
912 | * The caller should be holding dir->i_lock | |
913 | */ | |
914 | void nfs_force_lookup_revalidate(struct inode *dir) | |
915 | { | |
011935a0 | 916 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
917 | } |
918 | ||
1da177e4 LT |
919 | /* |
920 | * A check for whether or not the parent directory has changed. | |
921 | * In the case it has, we assume that the dentries are untrustworthy | |
922 | * and may need to be looked up again. | |
923 | */ | |
c79ba787 | 924 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
925 | { |
926 | if (IS_ROOT(dentry)) | |
927 | return 1; | |
4eec952e TM |
928 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
929 | return 0; | |
f2c77f4e TM |
930 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
931 | return 0; | |
932 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
933 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
934 | return 0; | |
935 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
936 | return 0; | |
937 | return 1; | |
1da177e4 LT |
938 | } |
939 | ||
1d6757fb TM |
940 | /* |
941 | * Return the intent data that applies to this particular path component | |
942 | * | |
943 | * Note that the current set of intents only apply to the very last | |
944 | * component of the path. | |
945 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
946 | */ | |
947 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
948 | { | |
949 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
950 | return 0; | |
951 | return nd->flags & mask; | |
952 | } | |
953 | ||
a12802ca TM |
954 | /* |
955 | * Use intent information to check whether or not we're going to do | |
956 | * an O_EXCL create using this path component. | |
957 | */ | |
958 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
959 | { | |
960 | if (NFS_PROTO(dir)->version == 2) | |
961 | return 0; | |
3516586a | 962 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
963 | } |
964 | ||
1d6757fb TM |
965 | /* |
966 | * Inode and filehandle revalidation for lookups. | |
967 | * | |
968 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
969 | * or if the intent information indicates that we're about to open this | |
970 | * particular file and the "nocto" mount flag is not set. | |
971 | * | |
972 | */ | |
1da177e4 LT |
973 | static inline |
974 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
975 | { | |
976 | struct nfs_server *server = NFS_SERVER(inode); | |
977 | ||
4e99a1ff TM |
978 | if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) |
979 | return 0; | |
1da177e4 | 980 | if (nd != NULL) { |
1da177e4 | 981 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 982 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
983 | goto out_force; |
984 | /* This is an open(2) */ | |
1d6757fb | 985 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
986 | !(server->flags & NFS_MOUNT_NOCTO) && |
987 | (S_ISREG(inode->i_mode) || | |
988 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 989 | goto out_force; |
4f48af45 | 990 | return 0; |
1da177e4 LT |
991 | } |
992 | return nfs_revalidate_inode(server, inode); | |
993 | out_force: | |
994 | return __nfs_revalidate_inode(server, inode); | |
995 | } | |
996 | ||
997 | /* | |
998 | * We judge how long we want to trust negative | |
999 | * dentries by looking at the parent inode mtime. | |
1000 | * | |
1001 | * If parent mtime has changed, we revalidate, else we wait for a | |
1002 | * period corresponding to the parent's attribute cache timeout value. | |
1003 | */ | |
1004 | static inline | |
1005 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1006 | struct nameidata *nd) | |
1007 | { | |
1da177e4 | 1008 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 1009 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 1010 | return 0; |
4eec952e TM |
1011 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1012 | return 1; | |
1da177e4 LT |
1013 | return !nfs_check_verifier(dir, dentry); |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * This is called every time the dcache has a lookup hit, | |
1018 | * and we should check whether we can really trust that | |
1019 | * lookup. | |
1020 | * | |
1021 | * NOTE! The hit can be a negative hit too, don't assume | |
1022 | * we have an inode! | |
1023 | * | |
1024 | * If the parent directory is seen to have changed, we throw out the | |
1025 | * cached dentry and do a new lookup. | |
1026 | */ | |
1027 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
1028 | { | |
1029 | struct inode *dir; | |
1030 | struct inode *inode; | |
1031 | struct dentry *parent; | |
e1fb4d05 TM |
1032 | struct nfs_fh *fhandle = NULL; |
1033 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1034 | int error; |
1da177e4 LT |
1035 | |
1036 | parent = dget_parent(dentry); | |
1da177e4 | 1037 | dir = parent->d_inode; |
91d5b470 | 1038 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1039 | inode = dentry->d_inode; |
1040 | ||
1041 | if (!inode) { | |
1042 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1043 | goto out_bad; | |
1044 | goto out_valid; | |
1045 | } | |
1046 | ||
1047 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1048 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1049 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1050 | dentry->d_name.name); |
1da177e4 LT |
1051 | goto out_bad; |
1052 | } | |
1053 | ||
15860ab1 TM |
1054 | if (nfs_have_delegation(inode, FMODE_READ)) |
1055 | goto out_set_verifier; | |
1056 | ||
1da177e4 | 1057 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1058 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1059 | if (nfs_lookup_verify_inode(inode, nd)) |
1060 | goto out_zap_parent; | |
1061 | goto out_valid; | |
1062 | } | |
1063 | ||
1064 | if (NFS_STALE(inode)) | |
1065 | goto out_bad; | |
1066 | ||
e1fb4d05 TM |
1067 | error = -ENOMEM; |
1068 | fhandle = nfs_alloc_fhandle(); | |
1069 | fattr = nfs_alloc_fattr(); | |
1070 | if (fhandle == NULL || fattr == NULL) | |
1071 | goto out_error; | |
1072 | ||
1073 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1da177e4 LT |
1074 | if (error) |
1075 | goto out_bad; | |
e1fb4d05 | 1076 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1077 | goto out_bad; |
e1fb4d05 | 1078 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1079 | goto out_bad; |
1080 | ||
e1fb4d05 TM |
1081 | nfs_free_fattr(fattr); |
1082 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1083 | out_set_verifier: |
cf8ba45e | 1084 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1085 | out_valid: |
1da177e4 | 1086 | dput(parent); |
1e7cb3dc | 1087 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1088 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1089 | dentry->d_name.name); |
1da177e4 LT |
1090 | return 1; |
1091 | out_zap_parent: | |
1092 | nfs_zap_caches(dir); | |
1093 | out_bad: | |
a1643a92 | 1094 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1095 | if (inode && S_ISDIR(inode->i_mode)) { |
1096 | /* Purge readdir caches. */ | |
1097 | nfs_zap_caches(inode); | |
1098 | /* If we have submounts, don't unhash ! */ | |
1099 | if (have_submounts(dentry)) | |
1100 | goto out_valid; | |
d9e80b7d AV |
1101 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1102 | goto out_valid; | |
1da177e4 LT |
1103 | shrink_dcache_parent(dentry); |
1104 | } | |
1105 | d_drop(dentry); | |
e1fb4d05 TM |
1106 | nfs_free_fattr(fattr); |
1107 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1108 | dput(parent); |
1e7cb3dc | 1109 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1110 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1111 | dentry->d_name.name); |
1da177e4 | 1112 | return 0; |
e1fb4d05 TM |
1113 | out_error: |
1114 | nfs_free_fattr(fattr); | |
1115 | nfs_free_fhandle(fhandle); | |
1116 | dput(parent); | |
1117 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1118 | __func__, dentry->d_parent->d_name.name, | |
1119 | dentry->d_name.name, error); | |
1120 | return error; | |
1da177e4 LT |
1121 | } |
1122 | ||
1123 | /* | |
1124 | * This is called from dput() when d_count is going to 0. | |
1125 | */ | |
1126 | static int nfs_dentry_delete(struct dentry *dentry) | |
1127 | { | |
1128 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1129 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1130 | dentry->d_flags); | |
1131 | ||
77f11192 TM |
1132 | /* Unhash any dentry with a stale inode */ |
1133 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1134 | return 1; | |
1135 | ||
1da177e4 LT |
1136 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1137 | /* Unhash it, so that ->d_iput() would be called */ | |
1138 | return 1; | |
1139 | } | |
1140 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1141 | /* Unhash it, so that ancestors of killed async unlink | |
1142 | * files will be cleaned up during umount */ | |
1143 | return 1; | |
1144 | } | |
1145 | return 0; | |
1146 | ||
1147 | } | |
1148 | ||
1b83d707 TM |
1149 | static void nfs_drop_nlink(struct inode *inode) |
1150 | { | |
1151 | spin_lock(&inode->i_lock); | |
1152 | if (inode->i_nlink > 0) | |
1153 | drop_nlink(inode); | |
1154 | spin_unlock(&inode->i_lock); | |
1155 | } | |
1156 | ||
1da177e4 LT |
1157 | /* |
1158 | * Called when the dentry loses inode. | |
1159 | * We use it to clean up silly-renamed files. | |
1160 | */ | |
1161 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1162 | { | |
83672d39 NB |
1163 | if (S_ISDIR(inode->i_mode)) |
1164 | /* drop any readdir cache as it could easily be old */ | |
1165 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1166 | ||
1da177e4 | 1167 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1168 | drop_nlink(inode); |
e4eff1a6 | 1169 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1170 | } |
1da177e4 LT |
1171 | iput(inode); |
1172 | } | |
1173 | ||
f786aa90 | 1174 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1175 | .d_revalidate = nfs_lookup_revalidate, |
1176 | .d_delete = nfs_dentry_delete, | |
1177 | .d_iput = nfs_dentry_iput, | |
1178 | }; | |
1179 | ||
1da177e4 LT |
1180 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1181 | { | |
1182 | struct dentry *res; | |
565277f6 | 1183 | struct dentry *parent; |
1da177e4 | 1184 | struct inode *inode = NULL; |
e1fb4d05 TM |
1185 | struct nfs_fh *fhandle = NULL; |
1186 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1187 | int error; |
1da177e4 LT |
1188 | |
1189 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1190 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1191 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1192 | |
1193 | res = ERR_PTR(-ENAMETOOLONG); | |
1194 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1195 | goto out; | |
1196 | ||
1da177e4 LT |
1197 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
1198 | ||
fd684071 TM |
1199 | /* |
1200 | * If we're doing an exclusive create, optimize away the lookup | |
1201 | * but don't hash the dentry. | |
1202 | */ | |
1203 | if (nfs_is_exclusive_create(dir, nd)) { | |
1204 | d_instantiate(dentry, NULL); | |
1205 | res = NULL; | |
fc0f684c | 1206 | goto out; |
fd684071 | 1207 | } |
1da177e4 | 1208 | |
e1fb4d05 TM |
1209 | res = ERR_PTR(-ENOMEM); |
1210 | fhandle = nfs_alloc_fhandle(); | |
1211 | fattr = nfs_alloc_fattr(); | |
1212 | if (fhandle == NULL || fattr == NULL) | |
1213 | goto out; | |
1214 | ||
565277f6 TM |
1215 | parent = dentry->d_parent; |
1216 | /* Protect against concurrent sillydeletes */ | |
1217 | nfs_block_sillyrename(parent); | |
e1fb4d05 | 1218 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1219 | if (error == -ENOENT) |
1220 | goto no_entry; | |
1221 | if (error < 0) { | |
1222 | res = ERR_PTR(error); | |
565277f6 | 1223 | goto out_unblock_sillyrename; |
1da177e4 | 1224 | } |
e1fb4d05 | 1225 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1226 | res = (struct dentry *)inode; |
1227 | if (IS_ERR(res)) | |
565277f6 | 1228 | goto out_unblock_sillyrename; |
54ceac45 | 1229 | |
1da177e4 | 1230 | no_entry: |
54ceac45 | 1231 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1232 | if (res != NULL) { |
1233 | if (IS_ERR(res)) | |
565277f6 | 1234 | goto out_unblock_sillyrename; |
1da177e4 | 1235 | dentry = res; |
9eaef27b | 1236 | } |
1da177e4 | 1237 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1238 | out_unblock_sillyrename: |
1239 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1240 | out: |
e1fb4d05 TM |
1241 | nfs_free_fattr(fattr); |
1242 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1243 | return res; |
1244 | } | |
1245 | ||
1246 | #ifdef CONFIG_NFS_V4 | |
1247 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
1248 | ||
f786aa90 | 1249 | const struct dentry_operations nfs4_dentry_operations = { |
1da177e4 LT |
1250 | .d_revalidate = nfs_open_revalidate, |
1251 | .d_delete = nfs_dentry_delete, | |
1252 | .d_iput = nfs_dentry_iput, | |
1253 | }; | |
1254 | ||
1d6757fb TM |
1255 | /* |
1256 | * Use intent information to determine whether we need to substitute | |
1257 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1258 | */ | |
5584c306 | 1259 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1260 | { |
1d6757fb | 1261 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1262 | return 0; |
1263 | /* NFS does not (yet) have a stateful open for directories */ | |
1264 | if (nd->flags & LOOKUP_DIRECTORY) | |
1265 | return 0; | |
1266 | /* Are we trying to write to a read only partition? */ | |
2c463e95 DH |
1267 | if (__mnt_is_readonly(nd->path.mnt) && |
1268 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
1da177e4 LT |
1269 | return 0; |
1270 | return 1; | |
1271 | } | |
1272 | ||
cd9a1c0e TM |
1273 | static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd) |
1274 | { | |
1275 | struct path path = { | |
1276 | .mnt = nd->path.mnt, | |
1277 | .dentry = dentry, | |
1278 | }; | |
1279 | struct nfs_open_context *ctx; | |
1280 | struct rpc_cred *cred; | |
1281 | fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); | |
1282 | ||
1283 | cred = rpc_lookup_cred(); | |
1284 | if (IS_ERR(cred)) | |
1285 | return ERR_CAST(cred); | |
1286 | ctx = alloc_nfs_open_context(&path, cred, fmode); | |
1287 | put_rpccred(cred); | |
1288 | if (ctx == NULL) | |
1289 | return ERR_PTR(-ENOMEM); | |
1290 | return ctx; | |
1291 | } | |
1292 | ||
1293 | static int do_open(struct inode *inode, struct file *filp) | |
1294 | { | |
1295 | nfs_fscache_set_inode_cookie(inode, filp); | |
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx) | |
1300 | { | |
1301 | struct file *filp; | |
1302 | int ret = 0; | |
1303 | ||
1304 | /* If the open_intent is for execute, we have an extra check to make */ | |
1305 | if (ctx->mode & FMODE_EXEC) { | |
1306 | ret = nfs_may_open(ctx->path.dentry->d_inode, | |
1307 | ctx->cred, | |
1308 | nd->intent.open.flags); | |
1309 | if (ret < 0) | |
1310 | goto out; | |
1311 | } | |
1312 | filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open); | |
1313 | if (IS_ERR(filp)) | |
1314 | ret = PTR_ERR(filp); | |
1315 | else | |
1316 | nfs_file_set_open_context(filp, ctx); | |
1317 | out: | |
1318 | put_nfs_open_context(ctx); | |
1319 | return ret; | |
1320 | } | |
1321 | ||
1da177e4 LT |
1322 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
1323 | { | |
cd9a1c0e TM |
1324 | struct nfs_open_context *ctx; |
1325 | struct iattr attr; | |
1da177e4 | 1326 | struct dentry *res = NULL; |
f46e0bd3 | 1327 | struct inode *inode; |
cd9a1c0e | 1328 | int open_flags; |
898f635c | 1329 | int err; |
1da177e4 | 1330 | |
1e7cb3dc CL |
1331 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
1332 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1333 | ||
1da177e4 | 1334 | /* Check that we are indeed trying to open this file */ |
5584c306 | 1335 | if (!is_atomic_open(nd)) |
1da177e4 LT |
1336 | goto no_open; |
1337 | ||
1338 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1339 | res = ERR_PTR(-ENAMETOOLONG); | |
1340 | goto out; | |
1341 | } | |
1342 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1343 | ||
d4d9cdcb TM |
1344 | /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash |
1345 | * the dentry. */ | |
3516586a | 1346 | if (nd->flags & LOOKUP_EXCL) { |
d4d9cdcb | 1347 | d_instantiate(dentry, NULL); |
02a913a7 TM |
1348 | goto out; |
1349 | } | |
1da177e4 | 1350 | |
cd9a1c0e TM |
1351 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1352 | res = ERR_CAST(ctx); | |
1353 | if (IS_ERR(ctx)) | |
1354 | goto out; | |
1355 | ||
1356 | open_flags = nd->intent.open.flags; | |
1357 | if (nd->flags & LOOKUP_CREATE) { | |
1358 | attr.ia_mode = nd->intent.open.create_mode; | |
1359 | attr.ia_valid = ATTR_MODE; | |
1360 | if (!IS_POSIXACL(dir)) | |
1361 | attr.ia_mode &= ~current_umask(); | |
1362 | } else { | |
898f635c | 1363 | open_flags &= ~(O_EXCL | O_CREAT); |
cd9a1c0e | 1364 | attr.ia_valid = 0; |
cd9a1c0e TM |
1365 | } |
1366 | ||
1da177e4 | 1367 | /* Open the file on the server */ |
f46e0bd3 | 1368 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1369 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
f46e0bd3 TM |
1370 | if (IS_ERR(inode)) { |
1371 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1372 | put_nfs_open_context(ctx); |
f46e0bd3 | 1373 | switch (PTR_ERR(inode)) { |
1da177e4 LT |
1374 | /* Make a negative dentry */ |
1375 | case -ENOENT: | |
f46e0bd3 | 1376 | d_add(dentry, NULL); |
02a913a7 TM |
1377 | res = NULL; |
1378 | goto out; | |
1da177e4 | 1379 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1380 | case -ENOTDIR: |
1381 | goto no_open; | |
1da177e4 LT |
1382 | case -ELOOP: |
1383 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1384 | goto no_open; | |
23ebbd9a | 1385 | /* case -EISDIR: */ |
1da177e4 LT |
1386 | /* case -EINVAL: */ |
1387 | default: | |
f46e0bd3 | 1388 | res = ERR_CAST(inode); |
1da177e4 LT |
1389 | goto out; |
1390 | } | |
cd9a1c0e | 1391 | } |
f46e0bd3 | 1392 | res = d_add_unique(dentry, inode); |
898f635c | 1393 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 TM |
1394 | if (res != NULL) { |
1395 | dput(ctx->path.dentry); | |
1396 | ctx->path.dentry = dget(res); | |
1da177e4 | 1397 | dentry = res; |
f46e0bd3 | 1398 | } |
898f635c TM |
1399 | err = nfs_intent_set_file(nd, ctx); |
1400 | if (err < 0) { | |
1401 | if (res != NULL) | |
1402 | dput(res); | |
1403 | return ERR_PTR(err); | |
1404 | } | |
1da177e4 | 1405 | out: |
f46e0bd3 | 1406 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1407 | return res; |
1408 | no_open: | |
1409 | return nfs_lookup(dir, dentry, nd); | |
1410 | } | |
1411 | ||
1412 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1413 | { | |
1414 | struct dentry *parent = NULL; | |
1415 | struct inode *inode = dentry->d_inode; | |
1416 | struct inode *dir; | |
b8d4cadd | 1417 | struct nfs_open_context *ctx; |
1da177e4 LT |
1418 | int openflags, ret = 0; |
1419 | ||
1f063d2c | 1420 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1421 | goto no_open; |
2b484297 | 1422 | |
1da177e4 LT |
1423 | parent = dget_parent(dentry); |
1424 | dir = parent->d_inode; | |
2b484297 | 1425 | |
1da177e4 LT |
1426 | /* We can't create new files in nfs_open_revalidate(), so we |
1427 | * optimize away revalidation of negative dentries. | |
1428 | */ | |
216d5d06 TM |
1429 | if (inode == NULL) { |
1430 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1431 | ret = 1; | |
1da177e4 | 1432 | goto out; |
216d5d06 TM |
1433 | } |
1434 | ||
1da177e4 LT |
1435 | /* NFS only supports OPEN on regular files */ |
1436 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1437 | goto no_open_dput; |
1da177e4 LT |
1438 | openflags = nd->intent.open.flags; |
1439 | /* We cannot do exclusive creation on a positive dentry */ | |
1440 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1441 | goto no_open_dput; |
1da177e4 | 1442 | /* We can't create new files, or truncate existing ones here */ |
0a377cff | 1443 | openflags &= ~(O_CREAT|O_EXCL|O_TRUNC); |
1da177e4 | 1444 | |
b8d4cadd TM |
1445 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1446 | ret = PTR_ERR(ctx); | |
1447 | if (IS_ERR(ctx)) | |
1448 | goto out; | |
1da177e4 | 1449 | /* |
1b1dcc1b | 1450 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1451 | * operations that change the directory. We therefore save the |
1452 | * change attribute *before* we do the RPC call. | |
1453 | */ | |
2b484297 | 1454 | inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL); |
535918f1 TM |
1455 | if (IS_ERR(inode)) { |
1456 | ret = PTR_ERR(inode); | |
1457 | switch (ret) { | |
1458 | case -EPERM: | |
1459 | case -EACCES: | |
1460 | case -EDQUOT: | |
1461 | case -ENOSPC: | |
1462 | case -EROFS: | |
1463 | goto out_put_ctx; | |
1464 | default: | |
1465 | goto out_drop; | |
1466 | } | |
1467 | } | |
1468 | iput(inode); | |
898f635c | 1469 | if (inode != dentry->d_inode) |
535918f1 | 1470 | goto out_drop; |
898f635c TM |
1471 | |
1472 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1473 | ret = nfs_intent_set_file(nd, ctx); | |
1474 | if (ret >= 0) | |
1475 | ret = 1; | |
1da177e4 LT |
1476 | out: |
1477 | dput(parent); | |
1da177e4 | 1478 | return ret; |
535918f1 TM |
1479 | out_drop: |
1480 | d_drop(dentry); | |
1481 | ret = 0; | |
1482 | out_put_ctx: | |
1483 | put_nfs_open_context(ctx); | |
1484 | goto out; | |
1485 | ||
5584c306 | 1486 | no_open_dput: |
1da177e4 | 1487 | dput(parent); |
5584c306 | 1488 | no_open: |
1da177e4 LT |
1489 | return nfs_lookup_revalidate(dentry, nd); |
1490 | } | |
c0204fd2 TM |
1491 | |
1492 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, | |
1493 | struct nameidata *nd) | |
1494 | { | |
1495 | struct nfs_open_context *ctx = NULL; | |
1496 | struct iattr attr; | |
1497 | int error; | |
1498 | int open_flags = 0; | |
1499 | ||
1500 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1501 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1502 | ||
1503 | attr.ia_mode = mode; | |
1504 | attr.ia_valid = ATTR_MODE; | |
1505 | ||
1506 | if ((nd->flags & LOOKUP_CREATE) != 0) { | |
1507 | open_flags = nd->intent.open.flags; | |
1508 | ||
1509 | ctx = nameidata_to_nfs_open_context(dentry, nd); | |
1510 | error = PTR_ERR(ctx); | |
1511 | if (IS_ERR(ctx)) | |
898f635c | 1512 | goto out_err_drop; |
c0204fd2 TM |
1513 | } |
1514 | ||
1515 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1516 | if (error != 0) | |
1517 | goto out_put_ctx; | |
898f635c TM |
1518 | if (ctx != NULL) { |
1519 | error = nfs_intent_set_file(nd, ctx); | |
1520 | if (error < 0) | |
1521 | goto out_err; | |
1522 | } | |
c0204fd2 TM |
1523 | return 0; |
1524 | out_put_ctx: | |
1525 | if (ctx != NULL) | |
1526 | put_nfs_open_context(ctx); | |
898f635c | 1527 | out_err_drop: |
c0204fd2 | 1528 | d_drop(dentry); |
898f635c | 1529 | out_err: |
c0204fd2 TM |
1530 | return error; |
1531 | } | |
1532 | ||
1da177e4 LT |
1533 | #endif /* CONFIG_NFSV4 */ |
1534 | ||
1da177e4 LT |
1535 | /* |
1536 | * Code common to create, mkdir, and mknod. | |
1537 | */ | |
1538 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1539 | struct nfs_fattr *fattr) | |
1540 | { | |
fab728e1 TM |
1541 | struct dentry *parent = dget_parent(dentry); |
1542 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1543 | struct inode *inode; |
1544 | int error = -EACCES; | |
1545 | ||
fab728e1 TM |
1546 | d_drop(dentry); |
1547 | ||
1da177e4 LT |
1548 | /* We may have been initialized further down */ |
1549 | if (dentry->d_inode) | |
fab728e1 | 1550 | goto out; |
1da177e4 | 1551 | if (fhandle->size == 0) { |
1da177e4 LT |
1552 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1553 | if (error) | |
fab728e1 | 1554 | goto out_error; |
1da177e4 | 1555 | } |
5724ab37 | 1556 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1557 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1558 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1559 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1560 | if (error < 0) |
fab728e1 | 1561 | goto out_error; |
1da177e4 | 1562 | } |
1da177e4 | 1563 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1564 | error = PTR_ERR(inode); |
1565 | if (IS_ERR(inode)) | |
fab728e1 TM |
1566 | goto out_error; |
1567 | d_add(dentry, inode); | |
1568 | out: | |
1569 | dput(parent); | |
1da177e4 | 1570 | return 0; |
fab728e1 TM |
1571 | out_error: |
1572 | nfs_mark_for_revalidate(dir); | |
1573 | dput(parent); | |
1574 | return error; | |
1da177e4 LT |
1575 | } |
1576 | ||
1577 | /* | |
1578 | * Following a failed create operation, we drop the dentry rather | |
1579 | * than retain a negative dentry. This avoids a problem in the event | |
1580 | * that the operation succeeded on the server, but an error in the | |
1581 | * reply path made it appear to have failed. | |
1582 | */ | |
1583 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1584 | struct nameidata *nd) | |
1585 | { | |
1586 | struct iattr attr; | |
1587 | int error; | |
1da177e4 | 1588 | |
1e7cb3dc CL |
1589 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1590 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1591 | |
1592 | attr.ia_mode = mode; | |
1593 | attr.ia_valid = ATTR_MODE; | |
1594 | ||
c0204fd2 | 1595 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL); |
1da177e4 LT |
1596 | if (error != 0) |
1597 | goto out_err; | |
1da177e4 LT |
1598 | return 0; |
1599 | out_err: | |
1da177e4 LT |
1600 | d_drop(dentry); |
1601 | return error; | |
1602 | } | |
1603 | ||
1604 | /* | |
1605 | * See comments for nfs_proc_create regarding failed operations. | |
1606 | */ | |
1607 | static int | |
1608 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1609 | { | |
1610 | struct iattr attr; | |
1611 | int status; | |
1612 | ||
1e7cb3dc CL |
1613 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1614 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1615 | |
1616 | if (!new_valid_dev(rdev)) | |
1617 | return -EINVAL; | |
1618 | ||
1619 | attr.ia_mode = mode; | |
1620 | attr.ia_valid = ATTR_MODE; | |
1621 | ||
1da177e4 | 1622 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1623 | if (status != 0) |
1624 | goto out_err; | |
1da177e4 LT |
1625 | return 0; |
1626 | out_err: | |
1da177e4 LT |
1627 | d_drop(dentry); |
1628 | return status; | |
1629 | } | |
1630 | ||
1631 | /* | |
1632 | * See comments for nfs_proc_create regarding failed operations. | |
1633 | */ | |
1634 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1635 | { | |
1636 | struct iattr attr; | |
1637 | int error; | |
1638 | ||
1e7cb3dc CL |
1639 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1640 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1641 | |
1642 | attr.ia_valid = ATTR_MODE; | |
1643 | attr.ia_mode = mode | S_IFDIR; | |
1644 | ||
1da177e4 | 1645 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1646 | if (error != 0) |
1647 | goto out_err; | |
1da177e4 LT |
1648 | return 0; |
1649 | out_err: | |
1650 | d_drop(dentry); | |
1da177e4 LT |
1651 | return error; |
1652 | } | |
1653 | ||
d45b9d8b TM |
1654 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1655 | { | |
1656 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1657 | d_delete(dentry); | |
1658 | } | |
1659 | ||
1da177e4 LT |
1660 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1661 | { | |
1662 | int error; | |
1663 | ||
1e7cb3dc CL |
1664 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1665 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1666 | |
1da177e4 LT |
1667 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1668 | /* Ensure the VFS deletes this inode */ | |
1669 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1670 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1671 | else if (error == -ENOENT) |
1672 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1673 | |
1674 | return error; | |
1675 | } | |
1676 | ||
1da177e4 LT |
1677 | /* |
1678 | * Remove a file after making sure there are no pending writes, | |
1679 | * and after checking that the file has only one user. | |
1680 | * | |
1681 | * We invalidate the attribute cache and free the inode prior to the operation | |
1682 | * to avoid possible races if the server reuses the inode. | |
1683 | */ | |
1684 | static int nfs_safe_remove(struct dentry *dentry) | |
1685 | { | |
1686 | struct inode *dir = dentry->d_parent->d_inode; | |
1687 | struct inode *inode = dentry->d_inode; | |
1688 | int error = -EBUSY; | |
1689 | ||
1690 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1691 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1692 | ||
1693 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1694 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1695 | error = 0; | |
1696 | goto out; | |
1697 | } | |
1698 | ||
1da177e4 | 1699 | if (inode != NULL) { |
cae7a073 | 1700 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1701 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1702 | /* The VFS may want to delete this inode */ | |
1703 | if (error == 0) | |
1b83d707 | 1704 | nfs_drop_nlink(inode); |
5ba7cc48 | 1705 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1706 | } else |
1707 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1708 | if (error == -ENOENT) |
1709 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1710 | out: |
1711 | return error; | |
1712 | } | |
1713 | ||
1714 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1715 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1716 | * | |
1717 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1718 | */ | |
1719 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1720 | { | |
1721 | int error; | |
1722 | int need_rehash = 0; | |
1723 | ||
1724 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1725 | dir->i_ino, dentry->d_name.name); | |
1726 | ||
1da177e4 LT |
1727 | spin_lock(&dcache_lock); |
1728 | spin_lock(&dentry->d_lock); | |
1729 | if (atomic_read(&dentry->d_count) > 1) { | |
1730 | spin_unlock(&dentry->d_lock); | |
1731 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1732 | /* Start asynchronous writeout of the inode */ |
1733 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1734 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1735 | return error; |
1736 | } | |
1737 | if (!d_unhashed(dentry)) { | |
1738 | __d_drop(dentry); | |
1739 | need_rehash = 1; | |
1740 | } | |
1741 | spin_unlock(&dentry->d_lock); | |
1742 | spin_unlock(&dcache_lock); | |
1743 | error = nfs_safe_remove(dentry); | |
d45b9d8b | 1744 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1745 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1746 | } else if (need_rehash) | |
1747 | d_rehash(dentry); | |
1da177e4 LT |
1748 | return error; |
1749 | } | |
1750 | ||
873101b3 CL |
1751 | /* |
1752 | * To create a symbolic link, most file systems instantiate a new inode, | |
1753 | * add a page to it containing the path, then write it out to the disk | |
1754 | * using prepare_write/commit_write. | |
1755 | * | |
1756 | * Unfortunately the NFS client can't create the in-core inode first | |
1757 | * because it needs a file handle to create an in-core inode (see | |
1758 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1759 | * symlink request has completed on the server. | |
1760 | * | |
1761 | * So instead we allocate a raw page, copy the symname into it, then do | |
1762 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1763 | * now have a new file handle and can instantiate an in-core NFS inode | |
1764 | * and move the raw page into its mapping. | |
1765 | */ | |
1766 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1767 | { |
873101b3 CL |
1768 | struct pagevec lru_pvec; |
1769 | struct page *page; | |
1770 | char *kaddr; | |
1da177e4 | 1771 | struct iattr attr; |
873101b3 | 1772 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1773 | int error; |
1774 | ||
1775 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1776 | dir->i_ino, dentry->d_name.name, symname); | |
1777 | ||
873101b3 CL |
1778 | if (pathlen > PAGE_SIZE) |
1779 | return -ENAMETOOLONG; | |
1da177e4 | 1780 | |
873101b3 CL |
1781 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1782 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1783 | |
83d93f22 | 1784 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1785 | if (!page) |
873101b3 | 1786 | return -ENOMEM; |
873101b3 CL |
1787 | |
1788 | kaddr = kmap_atomic(page, KM_USER0); | |
1789 | memcpy(kaddr, symname, pathlen); | |
1790 | if (pathlen < PAGE_SIZE) | |
1791 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1792 | kunmap_atomic(kaddr, KM_USER0); | |
1793 | ||
94a6d753 | 1794 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1795 | if (error != 0) { |
1796 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1797 | dir->i_sb->s_id, dir->i_ino, | |
1798 | dentry->d_name.name, symname, error); | |
1da177e4 | 1799 | d_drop(dentry); |
873101b3 | 1800 | __free_page(page); |
873101b3 CL |
1801 | return error; |
1802 | } | |
1803 | ||
1804 | /* | |
1805 | * No big deal if we can't add this page to the page cache here. | |
1806 | * READLINK will get the missing page from the server if needed. | |
1807 | */ | |
1808 | pagevec_init(&lru_pvec, 0); | |
1809 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1810 | GFP_KERNEL)) { | |
39cf8a13 | 1811 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1812 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1813 | SetPageUptodate(page); |
1814 | unlock_page(page); | |
1815 | } else | |
1816 | __free_page(page); | |
1817 | ||
873101b3 | 1818 | return 0; |
1da177e4 LT |
1819 | } |
1820 | ||
1821 | static int | |
1822 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1823 | { | |
1824 | struct inode *inode = old_dentry->d_inode; | |
1825 | int error; | |
1826 | ||
1827 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1828 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1829 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1830 | ||
9a3936aa TM |
1831 | nfs_inode_return_delegation(inode); |
1832 | ||
9697d234 | 1833 | d_drop(dentry); |
1da177e4 | 1834 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1835 | if (error == 0) { |
7de9c6ee | 1836 | ihold(inode); |
9697d234 | 1837 | d_add(dentry, inode); |
cf809556 | 1838 | } |
1da177e4 LT |
1839 | return error; |
1840 | } | |
1841 | ||
1842 | /* | |
1843 | * RENAME | |
1844 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1845 | * different file handle for the same inode after a rename (e.g. when | |
1846 | * moving to a different directory). A fail-safe method to do so would | |
1847 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1848 | * rename the old file using the sillyrename stuff. This way, the original | |
1849 | * file in old_dir will go away when the last process iput()s the inode. | |
1850 | * | |
1851 | * FIXED. | |
1852 | * | |
1853 | * It actually works quite well. One needs to have the possibility for | |
1854 | * at least one ".nfs..." file in each directory the file ever gets | |
1855 | * moved or linked to which happens automagically with the new | |
1856 | * implementation that only depends on the dcache stuff instead of | |
1857 | * using the inode layer | |
1858 | * | |
1859 | * Unfortunately, things are a little more complicated than indicated | |
1860 | * above. For a cross-directory move, we want to make sure we can get | |
1861 | * rid of the old inode after the operation. This means there must be | |
1862 | * no pending writes (if it's a file), and the use count must be 1. | |
1863 | * If these conditions are met, we can drop the dentries before doing | |
1864 | * the rename. | |
1865 | */ | |
1866 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1867 | struct inode *new_dir, struct dentry *new_dentry) | |
1868 | { | |
1869 | struct inode *old_inode = old_dentry->d_inode; | |
1870 | struct inode *new_inode = new_dentry->d_inode; | |
1871 | struct dentry *dentry = NULL, *rehash = NULL; | |
1872 | int error = -EBUSY; | |
1873 | ||
1da177e4 LT |
1874 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1875 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1876 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1877 | atomic_read(&new_dentry->d_count)); | |
1878 | ||
1879 | /* | |
28f79a1a MS |
1880 | * For non-directories, check whether the target is busy and if so, |
1881 | * make a copy of the dentry and then do a silly-rename. If the | |
1882 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1883 | * the new target. | |
1da177e4 | 1884 | */ |
27226104 MS |
1885 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1886 | /* | |
1887 | * To prevent any new references to the target during the | |
1888 | * rename, we unhash the dentry in advance. | |
1889 | */ | |
1890 | if (!d_unhashed(new_dentry)) { | |
1891 | d_drop(new_dentry); | |
1892 | rehash = new_dentry; | |
1893 | } | |
1da177e4 | 1894 | |
27226104 MS |
1895 | if (atomic_read(&new_dentry->d_count) > 2) { |
1896 | int err; | |
1897 | ||
1898 | /* copy the target dentry's name */ | |
1899 | dentry = d_alloc(new_dentry->d_parent, | |
1900 | &new_dentry->d_name); | |
1901 | if (!dentry) | |
1902 | goto out; | |
1903 | ||
1904 | /* silly-rename the existing target ... */ | |
1905 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1906 | if (err) |
27226104 | 1907 | goto out; |
24e93025 MS |
1908 | |
1909 | new_dentry = dentry; | |
56335936 | 1910 | rehash = NULL; |
24e93025 | 1911 | new_inode = NULL; |
27226104 | 1912 | } |
b1e4adf4 | 1913 | } |
1da177e4 | 1914 | |
cae7a073 | 1915 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 1916 | if (new_inode != NULL) |
24174119 | 1917 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 1918 | |
1da177e4 LT |
1919 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1920 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1921 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1922 | out: |
1923 | if (rehash) | |
1924 | d_rehash(rehash); | |
1925 | if (!error) { | |
b1e4adf4 TM |
1926 | if (new_inode != NULL) |
1927 | nfs_drop_nlink(new_inode); | |
349457cc | 1928 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1929 | nfs_set_verifier(new_dentry, |
1930 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1931 | } else if (error == -ENOENT) |
1932 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1933 | |
1934 | /* new dentry created? */ | |
1935 | if (dentry) | |
1936 | dput(dentry); | |
1da177e4 LT |
1937 | return error; |
1938 | } | |
1939 | ||
cfcea3e8 TM |
1940 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1941 | static LIST_HEAD(nfs_access_lru_list); | |
1942 | static atomic_long_t nfs_access_nr_entries; | |
1943 | ||
1c3c07e9 TM |
1944 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1945 | { | |
1946 | put_rpccred(entry->cred); | |
1947 | kfree(entry); | |
cfcea3e8 TM |
1948 | smp_mb__before_atomic_dec(); |
1949 | atomic_long_dec(&nfs_access_nr_entries); | |
1950 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1951 | } |
1952 | ||
1a81bb8a TM |
1953 | static void nfs_access_free_list(struct list_head *head) |
1954 | { | |
1955 | struct nfs_access_entry *cache; | |
1956 | ||
1957 | while (!list_empty(head)) { | |
1958 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1959 | list_del(&cache->lru); | |
1960 | nfs_access_free_entry(cache); | |
1961 | } | |
1962 | } | |
1963 | ||
7f8275d0 | 1964 | int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
979df72e TM |
1965 | { |
1966 | LIST_HEAD(head); | |
aa510da5 | 1967 | struct nfs_inode *nfsi, *next; |
979df72e TM |
1968 | struct nfs_access_entry *cache; |
1969 | ||
61d5eb29 TM |
1970 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1971 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 1972 | |
a50f7951 | 1973 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 1974 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
1975 | struct inode *inode; |
1976 | ||
1977 | if (nr_to_scan-- == 0) | |
1978 | break; | |
9c7e7e23 | 1979 | inode = &nfsi->vfs_inode; |
979df72e TM |
1980 | spin_lock(&inode->i_lock); |
1981 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1982 | goto remove_lru_entry; | |
1983 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1984 | struct nfs_access_entry, lru); | |
1985 | list_move(&cache->lru, &head); | |
1986 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1987 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1988 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1989 | &nfs_access_lru_list); | |
1990 | else { | |
1991 | remove_lru_entry: | |
1992 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 1993 | smp_mb__before_clear_bit(); |
979df72e | 1994 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 1995 | smp_mb__after_clear_bit(); |
979df72e | 1996 | } |
59844a9b | 1997 | spin_unlock(&inode->i_lock); |
979df72e TM |
1998 | } |
1999 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2000 | nfs_access_free_list(&head); |
979df72e TM |
2001 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
2002 | } | |
2003 | ||
1a81bb8a | 2004 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2005 | { |
1c3c07e9 | 2006 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2007 | struct rb_node *n; |
1c3c07e9 TM |
2008 | struct nfs_access_entry *entry; |
2009 | ||
2010 | /* Unhook entries from the cache */ | |
2011 | while ((n = rb_first(root_node)) != NULL) { | |
2012 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2013 | rb_erase(n, root_node); | |
1a81bb8a | 2014 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2015 | } |
2016 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2017 | } |
2018 | ||
1c3c07e9 | 2019 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2020 | { |
1a81bb8a TM |
2021 | LIST_HEAD(head); |
2022 | ||
2023 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2024 | return; | |
cfcea3e8 | 2025 | /* Remove from global LRU init */ |
1a81bb8a TM |
2026 | spin_lock(&nfs_access_lru_lock); |
2027 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2028 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2029 | |
1c3c07e9 | 2030 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2031 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2032 | spin_unlock(&inode->i_lock); | |
2033 | spin_unlock(&nfs_access_lru_lock); | |
2034 | nfs_access_free_list(&head); | |
1c3c07e9 | 2035 | } |
1da177e4 | 2036 | |
1c3c07e9 TM |
2037 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2038 | { | |
2039 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2040 | struct nfs_access_entry *entry; | |
2041 | ||
2042 | while (n != NULL) { | |
2043 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2044 | ||
2045 | if (cred < entry->cred) | |
2046 | n = n->rb_left; | |
2047 | else if (cred > entry->cred) | |
2048 | n = n->rb_right; | |
2049 | else | |
2050 | return entry; | |
1da177e4 | 2051 | } |
1c3c07e9 TM |
2052 | return NULL; |
2053 | } | |
2054 | ||
af22f94a | 2055 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2056 | { |
2057 | struct nfs_inode *nfsi = NFS_I(inode); | |
2058 | struct nfs_access_entry *cache; | |
2059 | int err = -ENOENT; | |
2060 | ||
dc59250c | 2061 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2062 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2063 | goto out_zap; | |
2064 | cache = nfs_access_search_rbtree(inode, cred); | |
2065 | if (cache == NULL) | |
2066 | goto out; | |
b4d2314b | 2067 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2068 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2069 | goto out_stale; |
2070 | res->jiffies = cache->jiffies; | |
2071 | res->cred = cache->cred; | |
2072 | res->mask = cache->mask; | |
cfcea3e8 | 2073 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2074 | err = 0; |
2075 | out: | |
2076 | spin_unlock(&inode->i_lock); | |
2077 | return err; | |
2078 | out_stale: | |
2079 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2080 | list_del(&cache->lru); |
1c3c07e9 TM |
2081 | spin_unlock(&inode->i_lock); |
2082 | nfs_access_free_entry(cache); | |
2083 | return -ENOENT; | |
2084 | out_zap: | |
1a81bb8a TM |
2085 | spin_unlock(&inode->i_lock); |
2086 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2087 | return -ENOENT; |
2088 | } | |
2089 | ||
2090 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2091 | { | |
cfcea3e8 TM |
2092 | struct nfs_inode *nfsi = NFS_I(inode); |
2093 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2094 | struct rb_node **p = &root_node->rb_node; |
2095 | struct rb_node *parent = NULL; | |
2096 | struct nfs_access_entry *entry; | |
2097 | ||
2098 | spin_lock(&inode->i_lock); | |
2099 | while (*p != NULL) { | |
2100 | parent = *p; | |
2101 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2102 | ||
2103 | if (set->cred < entry->cred) | |
2104 | p = &parent->rb_left; | |
2105 | else if (set->cred > entry->cred) | |
2106 | p = &parent->rb_right; | |
2107 | else | |
2108 | goto found; | |
2109 | } | |
2110 | rb_link_node(&set->rb_node, parent, p); | |
2111 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2112 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2113 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2114 | return; |
2115 | found: | |
2116 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2117 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2118 | list_del(&entry->lru); | |
1c3c07e9 TM |
2119 | spin_unlock(&inode->i_lock); |
2120 | nfs_access_free_entry(entry); | |
2121 | } | |
2122 | ||
af22f94a | 2123 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2124 | { |
2125 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2126 | if (cache == NULL) | |
2127 | return; | |
2128 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2129 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2130 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2131 | cache->mask = set->mask; |
1c3c07e9 TM |
2132 | |
2133 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2134 | |
2135 | /* Update accounting */ | |
2136 | smp_mb__before_atomic_inc(); | |
2137 | atomic_long_inc(&nfs_access_nr_entries); | |
2138 | smp_mb__after_atomic_inc(); | |
2139 | ||
2140 | /* Add inode to global LRU list */ | |
1a81bb8a | 2141 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2142 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2143 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2144 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2145 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2146 | spin_unlock(&nfs_access_lru_lock); |
2147 | } | |
1da177e4 LT |
2148 | } |
2149 | ||
2150 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2151 | { | |
2152 | struct nfs_access_entry cache; | |
2153 | int status; | |
2154 | ||
2155 | status = nfs_access_get_cached(inode, cred, &cache); | |
2156 | if (status == 0) | |
2157 | goto out; | |
2158 | ||
2159 | /* Be clever: ask server to check for all possible rights */ | |
2160 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2161 | cache.cred = cred; | |
2162 | cache.jiffies = jiffies; | |
2163 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2164 | if (status != 0) { |
2165 | if (status == -ESTALE) { | |
2166 | nfs_zap_caches(inode); | |
2167 | if (!S_ISDIR(inode->i_mode)) | |
2168 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2169 | } | |
1da177e4 | 2170 | return status; |
a71ee337 | 2171 | } |
1da177e4 LT |
2172 | nfs_access_add_cache(inode, &cache); |
2173 | out: | |
e6305c43 | 2174 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2175 | return 0; |
2176 | return -EACCES; | |
2177 | } | |
2178 | ||
af22f94a TM |
2179 | static int nfs_open_permission_mask(int openflags) |
2180 | { | |
2181 | int mask = 0; | |
2182 | ||
2183 | if (openflags & FMODE_READ) | |
2184 | mask |= MAY_READ; | |
2185 | if (openflags & FMODE_WRITE) | |
2186 | mask |= MAY_WRITE; | |
2187 | if (openflags & FMODE_EXEC) | |
2188 | mask |= MAY_EXEC; | |
2189 | return mask; | |
2190 | } | |
2191 | ||
2192 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2193 | { | |
2194 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2195 | } | |
2196 | ||
e6305c43 | 2197 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2198 | { |
2199 | struct rpc_cred *cred; | |
2200 | int res = 0; | |
2201 | ||
91d5b470 CL |
2202 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2203 | ||
e6305c43 | 2204 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2205 | goto out; |
2206 | /* Is this sys_access() ? */ | |
9cfcac81 | 2207 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2208 | goto force_lookup; |
2209 | ||
2210 | switch (inode->i_mode & S_IFMT) { | |
2211 | case S_IFLNK: | |
2212 | goto out; | |
2213 | case S_IFREG: | |
2214 | /* NFSv4 has atomic_open... */ | |
2215 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2216 | && (mask & MAY_OPEN) |
2217 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2218 | goto out; |
2219 | break; | |
2220 | case S_IFDIR: | |
2221 | /* | |
2222 | * Optimize away all write operations, since the server | |
2223 | * will check permissions when we perform the op. | |
2224 | */ | |
2225 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2226 | goto out; | |
2227 | } | |
2228 | ||
2229 | force_lookup: | |
1da177e4 LT |
2230 | if (!NFS_PROTO(inode)->access) |
2231 | goto out_notsup; | |
2232 | ||
98a8e323 | 2233 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2234 | if (!IS_ERR(cred)) { |
2235 | res = nfs_do_access(inode, cred, mask); | |
2236 | put_rpccred(cred); | |
2237 | } else | |
2238 | res = PTR_ERR(cred); | |
1da177e4 | 2239 | out: |
f696a365 MS |
2240 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2241 | res = -EACCES; | |
2242 | ||
1e7cb3dc CL |
2243 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2244 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2245 | return res; |
2246 | out_notsup: | |
2247 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2248 | if (res == 0) | |
2249 | res = generic_permission(inode, mask, NULL); | |
1e7cb3dc | 2250 | goto out; |
1da177e4 LT |
2251 | } |
2252 | ||
2253 | /* | |
2254 | * Local variables: | |
2255 | * version-control: t | |
2256 | * kept-new-versions: 5 | |
2257 | * End: | |
2258 | */ |