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