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> |
1da177e4 LT |
37 | |
38 | #include "delegation.h" | |
91d5b470 | 39 | #include "iostat.h" |
4c30d56e | 40 | #include "internal.h" |
cd9a1c0e | 41 | #include "fscache.h" |
1da177e4 | 42 | |
1da177e4 LT |
43 | /* #define NFS_DEBUG_VERBOSE 1 */ |
44 | ||
45 | static int nfs_opendir(struct inode *, struct file *); | |
46 | static int nfs_readdir(struct file *, void *, filldir_t); | |
47 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
48 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
49 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
50 | static int nfs_rmdir(struct inode *, struct dentry *); | |
51 | static int nfs_unlink(struct inode *, struct dentry *); | |
52 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
53 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
54 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
55 | static int nfs_rename(struct inode *, struct dentry *, | |
56 | struct inode *, struct dentry *); | |
7ea80859 | 57 | static int nfs_fsync_dir(struct file *, int); |
f0dd2136 | 58 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
d1bacf9e | 59 | static int nfs_readdir_clear_array(struct page*, gfp_t); |
1da177e4 | 60 | |
4b6f5d20 | 61 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 62 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
63 | .read = generic_read_dir, |
64 | .readdir = nfs_readdir, | |
65 | .open = nfs_opendir, | |
66 | .release = nfs_release, | |
67 | .fsync = nfs_fsync_dir, | |
68 | }; | |
69 | ||
92e1d5be | 70 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
71 | .create = nfs_create, |
72 | .lookup = nfs_lookup, | |
73 | .link = nfs_link, | |
74 | .unlink = nfs_unlink, | |
75 | .symlink = nfs_symlink, | |
76 | .mkdir = nfs_mkdir, | |
77 | .rmdir = nfs_rmdir, | |
78 | .mknod = nfs_mknod, | |
79 | .rename = nfs_rename, | |
80 | .permission = nfs_permission, | |
81 | .getattr = nfs_getattr, | |
82 | .setattr = nfs_setattr, | |
83 | }; | |
84 | ||
d1bacf9e BS |
85 | const struct address_space_operations nfs_dir_addr_space_ops = { |
86 | .releasepage = nfs_readdir_clear_array, | |
87 | }; | |
88 | ||
b7fa0554 | 89 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 90 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
91 | .create = nfs_create, |
92 | .lookup = nfs_lookup, | |
93 | .link = nfs_link, | |
94 | .unlink = nfs_unlink, | |
95 | .symlink = nfs_symlink, | |
96 | .mkdir = nfs_mkdir, | |
97 | .rmdir = nfs_rmdir, | |
98 | .mknod = nfs_mknod, | |
99 | .rename = nfs_rename, | |
100 | .permission = nfs_permission, | |
101 | .getattr = nfs_getattr, | |
102 | .setattr = nfs_setattr, | |
103 | .listxattr = nfs3_listxattr, | |
104 | .getxattr = nfs3_getxattr, | |
105 | .setxattr = nfs3_setxattr, | |
106 | .removexattr = nfs3_removexattr, | |
107 | }; | |
108 | #endif /* CONFIG_NFS_V3 */ | |
109 | ||
1da177e4 LT |
110 | #ifdef CONFIG_NFS_V4 |
111 | ||
112 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
c0204fd2 | 113 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd); |
92e1d5be | 114 | const struct inode_operations nfs4_dir_inode_operations = { |
c0204fd2 | 115 | .create = nfs_open_create, |
1da177e4 LT |
116 | .lookup = nfs_atomic_lookup, |
117 | .link = nfs_link, | |
118 | .unlink = nfs_unlink, | |
119 | .symlink = nfs_symlink, | |
120 | .mkdir = nfs_mkdir, | |
121 | .rmdir = nfs_rmdir, | |
122 | .mknod = nfs_mknod, | |
123 | .rename = nfs_rename, | |
124 | .permission = nfs_permission, | |
125 | .getattr = nfs_getattr, | |
126 | .setattr = nfs_setattr, | |
6b3b5496 BF |
127 | .getxattr = nfs4_getxattr, |
128 | .setxattr = nfs4_setxattr, | |
129 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
130 | }; |
131 | ||
132 | #endif /* CONFIG_NFS_V4 */ | |
133 | ||
134 | /* | |
135 | * Open file | |
136 | */ | |
137 | static int | |
138 | nfs_opendir(struct inode *inode, struct file *filp) | |
139 | { | |
7451c4f0 | 140 | int res; |
1da177e4 | 141 | |
6da24bc9 | 142 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
cc0dd2d1 CL |
143 | filp->f_path.dentry->d_parent->d_name.name, |
144 | filp->f_path.dentry->d_name.name); | |
145 | ||
146 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 147 | |
1da177e4 | 148 | /* Call generic open code in order to cache credentials */ |
7451c4f0 | 149 | res = nfs_open(inode, filp); |
f5a73672 NB |
150 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
151 | /* This is a mountpoint, so d_revalidate will never | |
152 | * have been called, so we need to refresh the | |
153 | * inode (for close-open consistency) ourselves. | |
154 | */ | |
155 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
156 | } | |
1da177e4 LT |
157 | return res; |
158 | } | |
159 | ||
d1bacf9e BS |
160 | struct nfs_cache_array_entry { |
161 | u64 cookie; | |
162 | u64 ino; | |
163 | struct qstr string; | |
164 | }; | |
165 | ||
166 | struct nfs_cache_array { | |
167 | unsigned int size; | |
168 | int eof_index; | |
169 | u64 last_cookie; | |
170 | struct nfs_cache_array_entry array[0]; | |
171 | }; | |
172 | ||
173 | #define MAX_READDIR_ARRAY ((PAGE_SIZE - sizeof(struct nfs_cache_array)) / sizeof(struct nfs_cache_array_entry)) | |
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 | { | |
197 | if (page == NULL) | |
198 | return ERR_PTR(-EIO); | |
199 | return (struct nfs_cache_array *)kmap(page); | |
200 | } | |
201 | ||
202 | static | |
203 | void nfs_readdir_release_array(struct page *page) | |
204 | { | |
205 | kunmap(page); | |
206 | } | |
207 | ||
208 | /* | |
209 | * we are freeing strings created by nfs_add_to_readdir_array() | |
210 | */ | |
211 | static | |
212 | int nfs_readdir_clear_array(struct page *page, gfp_t mask) | |
213 | { | |
214 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
215 | int i; | |
216 | for (i = 0; i < array->size; i++) | |
217 | kfree(array->array[i].string.name); | |
218 | nfs_readdir_release_array(page); | |
219 | return 0; | |
220 | } | |
221 | ||
222 | /* | |
223 | * the caller is responsible for freeing qstr.name | |
224 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
225 | * nfs_clear_readdir_array() | |
226 | */ | |
227 | static | |
4a201d6e | 228 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
229 | { |
230 | string->len = len; | |
231 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
232 | if (string->name == NULL) |
233 | return -ENOMEM; | |
234 | string->hash = full_name_hash(name, len); | |
235 | return 0; | |
d1bacf9e BS |
236 | } |
237 | ||
238 | static | |
239 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
240 | { | |
241 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
242 | struct nfs_cache_array_entry *cache_entry; |
243 | int ret; | |
244 | ||
d1bacf9e BS |
245 | if (IS_ERR(array)) |
246 | return PTR_ERR(array); | |
4a201d6e TM |
247 | ret = -EIO; |
248 | if (array->size >= MAX_READDIR_ARRAY) | |
249 | goto out; | |
d1bacf9e | 250 | |
4a201d6e TM |
251 | cache_entry = &array->array[array->size]; |
252 | cache_entry->cookie = entry->prev_cookie; | |
253 | cache_entry->ino = entry->ino; | |
254 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); | |
255 | if (ret) | |
256 | goto out; | |
d1bacf9e | 257 | array->last_cookie = entry->cookie; |
d1bacf9e BS |
258 | if (entry->eof == 1) |
259 | array->eof_index = array->size; | |
260 | array->size++; | |
4a201d6e | 261 | out: |
d1bacf9e | 262 | nfs_readdir_release_array(page); |
4a201d6e | 263 | return ret; |
d1bacf9e BS |
264 | } |
265 | ||
266 | static | |
267 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
268 | { | |
269 | loff_t diff = desc->file->f_pos - desc->current_index; | |
270 | unsigned int index; | |
271 | ||
272 | if (diff < 0) | |
273 | goto out_eof; | |
274 | if (diff >= array->size) { | |
275 | if (array->eof_index > 0) | |
276 | goto out_eof; | |
277 | desc->current_index += array->size; | |
278 | return -EAGAIN; | |
279 | } | |
280 | ||
281 | index = (unsigned int)diff; | |
282 | *desc->dir_cookie = array->array[index].cookie; | |
283 | desc->cache_entry_index = index; | |
284 | if (index == array->eof_index) | |
285 | desc->eof = 1; | |
286 | return 0; | |
287 | out_eof: | |
288 | desc->eof = 1; | |
289 | return -EBADCOOKIE; | |
290 | } | |
291 | ||
292 | static | |
293 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
294 | { | |
295 | int i; | |
296 | int status = -EAGAIN; | |
297 | ||
298 | for (i = 0; i < array->size; i++) { | |
299 | if (i == array->eof_index) { | |
300 | desc->eof = 1; | |
301 | status = -EBADCOOKIE; | |
302 | } | |
303 | if (array->array[i].cookie == *desc->dir_cookie) { | |
304 | desc->cache_entry_index = i; | |
305 | status = 0; | |
306 | break; | |
307 | } | |
308 | } | |
309 | ||
310 | return status; | |
311 | } | |
312 | ||
313 | static | |
314 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
315 | { | |
316 | struct nfs_cache_array *array; | |
317 | int status = -EBADCOOKIE; | |
318 | ||
319 | if (desc->dir_cookie == NULL) | |
320 | goto out; | |
321 | ||
322 | array = nfs_readdir_get_array(desc->page); | |
323 | if (IS_ERR(array)) { | |
324 | status = PTR_ERR(array); | |
325 | goto out; | |
326 | } | |
327 | ||
328 | if (*desc->dir_cookie == 0) | |
329 | status = nfs_readdir_search_for_pos(array, desc); | |
330 | else | |
331 | status = nfs_readdir_search_for_cookie(array, desc); | |
332 | ||
333 | nfs_readdir_release_array(desc->page); | |
334 | out: | |
335 | return status; | |
336 | } | |
337 | ||
338 | /* Fill a page with xdr information before transferring to the cache page */ | |
339 | static | |
56e4ebf8 | 340 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 341 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 342 | { |
1da177e4 | 343 | struct rpc_cred *cred = nfs_file_cred(file); |
4704f0e2 | 344 | unsigned long timestamp, gencount; |
1da177e4 LT |
345 | int error; |
346 | ||
1da177e4 LT |
347 | again: |
348 | timestamp = jiffies; | |
4704f0e2 | 349 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 350 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
351 | NFS_SERVER(inode)->dtsize, desc->plus); |
352 | if (error < 0) { | |
353 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
354 | if (error == -ENOTSUPP && desc->plus) { | |
355 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 356 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
357 | desc->plus = 0; |
358 | goto again; | |
359 | } | |
360 | goto error; | |
361 | } | |
1f4eab7e | 362 | desc->timestamp = timestamp; |
4704f0e2 | 363 | desc->gencount = gencount; |
d1bacf9e BS |
364 | error: |
365 | return error; | |
1da177e4 LT |
366 | } |
367 | ||
d1bacf9e BS |
368 | /* Fill in an entry based on the xdr code stored in desc->page */ |
369 | static | |
babddc72 | 370 | int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream) |
1da177e4 | 371 | { |
82f2e547 | 372 | __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus); |
1da177e4 LT |
373 | if (IS_ERR(p)) |
374 | return PTR_ERR(p); | |
1da177e4 | 375 | |
d1bacf9e BS |
376 | entry->fattr->time_start = desc->timestamp; |
377 | entry->fattr->gencount = desc->gencount; | |
378 | return 0; | |
1da177e4 LT |
379 | } |
380 | ||
d39ab9de BS |
381 | static |
382 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
383 | { | |
384 | struct nfs_inode *node; | |
385 | if (dentry->d_inode == NULL) | |
386 | goto different; | |
387 | node = NFS_I(dentry->d_inode); | |
388 | if (node->fh.size != entry->fh->size) | |
389 | goto different; | |
390 | if (strncmp(node->fh.data, entry->fh->data, node->fh.size) != 0) | |
391 | goto different; | |
392 | return 1; | |
393 | different: | |
394 | return 0; | |
395 | } | |
396 | ||
397 | static | |
398 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
399 | { | |
4a201d6e TM |
400 | struct qstr filename = { |
401 | .len = entry->len, | |
402 | .name = entry->name, | |
403 | }; | |
404 | struct dentry *dentry; | |
405 | struct dentry *alias; | |
d39ab9de BS |
406 | struct inode *dir = parent->d_inode; |
407 | struct inode *inode; | |
408 | ||
4a201d6e TM |
409 | if (filename.name[0] == '.') { |
410 | if (filename.len == 1) | |
411 | return; | |
412 | if (filename.len == 2 && filename.name[1] == '.') | |
413 | return; | |
414 | } | |
415 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 416 | |
4a201d6e | 417 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
418 | if (dentry != NULL) { |
419 | if (nfs_same_file(dentry, entry)) { | |
420 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
421 | goto out; | |
422 | } else { | |
423 | d_drop(dentry); | |
424 | dput(dentry); | |
425 | } | |
426 | } | |
427 | ||
428 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
429 | if (dentry == NULL) |
430 | return; | |
431 | ||
d39ab9de BS |
432 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
433 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
434 | if (IS_ERR(inode)) | |
435 | goto out; | |
436 | ||
437 | alias = d_materialise_unique(dentry, inode); | |
438 | if (IS_ERR(alias)) | |
439 | goto out; | |
440 | else if (alias) { | |
441 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
442 | dput(alias); | |
443 | } else | |
444 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
445 | ||
446 | out: | |
447 | dput(dentry); | |
d39ab9de BS |
448 | } |
449 | ||
d1bacf9e BS |
450 | /* Perform conversion from xdr to cache array */ |
451 | static | |
452 | void nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, | |
56e4ebf8 | 453 | void *xdr_page, struct page *page, unsigned int buflen) |
1da177e4 | 454 | { |
babddc72 BS |
455 | struct xdr_stream stream; |
456 | struct xdr_buf buf; | |
56e4ebf8 | 457 | __be32 *ptr = xdr_page; |
99424380 BS |
458 | int status; |
459 | struct nfs_cache_array *array; | |
babddc72 BS |
460 | |
461 | buf.head->iov_base = xdr_page; | |
462 | buf.head->iov_len = buflen; | |
463 | buf.tail->iov_len = 0; | |
464 | buf.page_base = 0; | |
465 | buf.page_len = 0; | |
466 | buf.buflen = buf.head->iov_len; | |
467 | buf.len = buf.head->iov_len; | |
468 | ||
469 | xdr_init_decode(&stream, &buf, ptr); | |
470 | ||
99424380 BS |
471 | |
472 | do { | |
473 | status = xdr_decode(desc, entry, &stream); | |
474 | if (status != 0) | |
475 | break; | |
476 | ||
d1bacf9e | 477 | if (nfs_readdir_add_to_array(entry, page) == -1) |
1da177e4 | 478 | break; |
d39ab9de BS |
479 | if (desc->plus == 1) |
480 | nfs_prime_dcache(desc->file->f_path.dentry, entry); | |
99424380 BS |
481 | } while (!entry->eof); |
482 | ||
483 | if (status == -EBADCOOKIE && entry->eof) { | |
484 | array = nfs_readdir_get_array(page); | |
485 | array->eof_index = array->size - 1; | |
486 | status = 0; | |
487 | nfs_readdir_release_array(page); | |
1da177e4 | 488 | } |
56e4ebf8 BS |
489 | } |
490 | ||
491 | static | |
492 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
493 | { | |
494 | unsigned int i; | |
495 | for (i = 0; i < npages; i++) | |
496 | put_page(pages[i]); | |
497 | } | |
498 | ||
499 | static | |
500 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
501 | unsigned int npages) | |
502 | { | |
503 | vm_unmap_ram(ptr, npages); | |
504 | nfs_readdir_free_pagearray(pages, npages); | |
505 | } | |
506 | ||
507 | /* | |
508 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
509 | * to nfs_readdir_free_large_page | |
510 | */ | |
511 | static | |
512 | void *nfs_readdir_large_page(struct page **pages, unsigned int npages) | |
513 | { | |
514 | void *ptr; | |
515 | unsigned int i; | |
516 | ||
517 | for (i = 0; i < npages; i++) { | |
518 | struct page *page = alloc_page(GFP_KERNEL); | |
519 | if (page == NULL) | |
520 | goto out_freepages; | |
521 | pages[i] = page; | |
522 | } | |
523 | ||
4a201d6e | 524 | ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL); |
56e4ebf8 BS |
525 | if (!IS_ERR_OR_NULL(ptr)) |
526 | return ptr; | |
527 | out_freepages: | |
528 | nfs_readdir_free_pagearray(pages, i); | |
529 | return NULL; | |
1da177e4 LT |
530 | } |
531 | ||
d1bacf9e BS |
532 | static |
533 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 534 | { |
56e4ebf8 BS |
535 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
536 | void *pages_ptr = NULL; | |
d1bacf9e BS |
537 | struct nfs_entry entry; |
538 | struct file *file = desc->file; | |
539 | struct nfs_cache_array *array; | |
540 | int status = 0; | |
56e4ebf8 | 541 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
542 | |
543 | entry.prev_cookie = 0; | |
544 | entry.cookie = *desc->dir_cookie; | |
545 | entry.eof = 0; | |
546 | entry.fh = nfs_alloc_fhandle(); | |
547 | entry.fattr = nfs_alloc_fattr(); | |
548 | if (entry.fh == NULL || entry.fattr == NULL) | |
549 | goto out; | |
00a92642 | 550 | |
d1bacf9e BS |
551 | array = nfs_readdir_get_array(page); |
552 | memset(array, 0, sizeof(struct nfs_cache_array)); | |
553 | array->eof_index = -1; | |
00a92642 | 554 | |
56e4ebf8 BS |
555 | pages_ptr = nfs_readdir_large_page(pages, array_size); |
556 | if (!pages_ptr) | |
d1bacf9e BS |
557 | goto out_release_array; |
558 | do { | |
56e4ebf8 | 559 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 560 | |
d1bacf9e | 561 | if (status < 0) |
00a92642 | 562 | break; |
56e4ebf8 | 563 | nfs_readdir_page_filler(desc, &entry, pages_ptr, page, array_size * PAGE_SIZE); |
d1bacf9e BS |
564 | } while (array->eof_index < 0 && array->size < MAX_READDIR_ARRAY); |
565 | ||
56e4ebf8 | 566 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
567 | out_release_array: |
568 | nfs_readdir_release_array(page); | |
569 | out: | |
570 | nfs_free_fattr(entry.fattr); | |
571 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
572 | return status; |
573 | } | |
574 | ||
575 | /* | |
d1bacf9e BS |
576 | * Now we cache directories properly, by converting xdr information |
577 | * to an array that can be used for lookups later. This results in | |
578 | * fewer cache pages, since we can store more information on each page. | |
579 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 580 | */ |
d1bacf9e BS |
581 | static |
582 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 583 | { |
01cce933 | 584 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 585 | |
3c8a1aee | 586 | if (nfs_readdir_xdr_to_array(desc, page, inode) < 0) |
d1bacf9e BS |
587 | goto error; |
588 | SetPageUptodate(page); | |
1da177e4 | 589 | |
d1bacf9e BS |
590 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
591 | /* Should never happen */ | |
592 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 593 | } |
d1bacf9e BS |
594 | unlock_page(page); |
595 | return 0; | |
596 | error: | |
597 | unlock_page(page); | |
598 | return -EIO; | |
599 | } | |
1da177e4 | 600 | |
d1bacf9e BS |
601 | static |
602 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
603 | { | |
604 | page_cache_release(desc->page); | |
605 | desc->page = NULL; | |
606 | } | |
607 | ||
608 | static | |
609 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
610 | { | |
611 | struct page *page; | |
612 | page = read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, | |
613 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); | |
614 | if (IS_ERR(page)) | |
615 | desc->eof = 1; | |
616 | return page; | |
1da177e4 LT |
617 | } |
618 | ||
619 | /* | |
d1bacf9e | 620 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 621 | */ |
d1bacf9e BS |
622 | static |
623 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
624 | { | |
625 | int res; | |
626 | ||
627 | desc->page = get_cache_page(desc); | |
628 | if (IS_ERR(desc->page)) | |
629 | return PTR_ERR(desc->page); | |
630 | ||
631 | res = nfs_readdir_search_array(desc); | |
632 | if (res == 0) | |
633 | return 0; | |
634 | cache_page_release(desc); | |
635 | return res; | |
636 | } | |
637 | ||
638 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
639 | static inline |
640 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
641 | { | |
d1bacf9e | 642 | int res = -EAGAIN; |
d1bacf9e BS |
643 | |
644 | while (1) { | |
645 | res = find_cache_page(desc); | |
1da177e4 LT |
646 | if (res != -EAGAIN) |
647 | break; | |
d1bacf9e | 648 | desc->page_index++; |
1da177e4 | 649 | } |
1da177e4 LT |
650 | return res; |
651 | } | |
652 | ||
653 | static inline unsigned int dt_type(struct inode *inode) | |
654 | { | |
655 | return (inode->i_mode >> 12) & 15; | |
656 | } | |
657 | ||
1da177e4 LT |
658 | /* |
659 | * Once we've found the start of the dirent within a page: fill 'er up... | |
660 | */ | |
661 | static | |
662 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
663 | filldir_t filldir) | |
664 | { | |
665 | struct file *file = desc->file; | |
d1bacf9e BS |
666 | int i = 0; |
667 | int res = 0; | |
668 | struct nfs_cache_array *array = NULL; | |
669 | unsigned int d_type = DT_UNKNOWN; | |
670 | struct dentry *dentry = NULL; | |
671 | ||
672 | array = nfs_readdir_get_array(desc->page); | |
673 | ||
674 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
675 | d_type = DT_UNKNOWN; | |
1da177e4 | 676 | |
d1bacf9e BS |
677 | res = filldir(dirent, array->array[i].string.name, |
678 | array->array[i].string.len, file->f_pos, | |
679 | nfs_compat_user_ino64(array->array[i].ino), d_type); | |
1da177e4 LT |
680 | if (res < 0) |
681 | break; | |
00a92642 | 682 | file->f_pos++; |
d1bacf9e BS |
683 | desc->cache_entry_index = i; |
684 | if (i < (array->size-1)) | |
685 | *desc->dir_cookie = array->array[i+1].cookie; | |
686 | else | |
687 | *desc->dir_cookie = array->last_cookie; | |
688 | if (i == array->eof_index) { | |
689 | desc->eof = 1; | |
1da177e4 LT |
690 | break; |
691 | } | |
1da177e4 | 692 | } |
d1bacf9e BS |
693 | |
694 | nfs_readdir_release_array(desc->page); | |
695 | cache_page_release(desc); | |
1da177e4 LT |
696 | if (dentry != NULL) |
697 | dput(dentry); | |
1e7cb3dc CL |
698 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
699 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
700 | return res; |
701 | } | |
702 | ||
703 | /* | |
704 | * If we cannot find a cookie in our cache, we suspect that this is | |
705 | * because it points to a deleted file, so we ask the server to return | |
706 | * whatever it thinks is the next entry. We then feed this to filldir. | |
707 | * If all goes well, we should then be able to find our way round the | |
708 | * cache on the next call to readdir_search_pagecache(); | |
709 | * | |
710 | * NOTE: we cannot add the anonymous page to the pagecache because | |
711 | * the data it contains might not be page aligned. Besides, | |
712 | * we should already have a complete representation of the | |
713 | * directory in the page cache by the time we get here. | |
714 | */ | |
715 | static inline | |
716 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
717 | filldir_t filldir) | |
718 | { | |
1da177e4 LT |
719 | struct page *page = NULL; |
720 | int status; | |
d1bacf9e | 721 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 722 | |
1e7cb3dc CL |
723 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
724 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
725 | |
726 | page = alloc_page(GFP_HIGHUSER); | |
727 | if (!page) { | |
728 | status = -ENOMEM; | |
729 | goto out; | |
730 | } | |
d1bacf9e BS |
731 | |
732 | if (nfs_readdir_xdr_to_array(desc, page, inode) == -1) { | |
1da177e4 | 733 | status = -EIO; |
1da177e4 | 734 | goto out_release; |
d1bacf9e | 735 | } |
1da177e4 | 736 | |
baf57a09 | 737 | desc->page_index = 0; |
d1bacf9e | 738 | desc->page = page; |
1da177e4 LT |
739 | status = nfs_do_filldir(desc, dirent, filldir); |
740 | ||
1da177e4 | 741 | out: |
1e7cb3dc | 742 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 743 | __func__, status); |
1da177e4 LT |
744 | return status; |
745 | out_release: | |
d1bacf9e | 746 | cache_page_release(desc); |
1da177e4 LT |
747 | goto out; |
748 | } | |
749 | ||
00a92642 OG |
750 | /* The file offset position represents the dirent entry number. A |
751 | last cookie cache takes care of the common case of reading the | |
752 | whole directory. | |
1da177e4 LT |
753 | */ |
754 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
755 | { | |
01cce933 | 756 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
757 | struct inode *inode = dentry->d_inode; |
758 | nfs_readdir_descriptor_t my_desc, | |
759 | *desc = &my_desc; | |
aa49b4cf | 760 | int res = -ENOMEM; |
1da177e4 | 761 | |
6da24bc9 | 762 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
763 | dentry->d_parent->d_name.name, dentry->d_name.name, |
764 | (long long)filp->f_pos); | |
91d5b470 CL |
765 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
766 | ||
1da177e4 | 767 | /* |
00a92642 | 768 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 769 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
770 | * to either find the entry with the appropriate number or |
771 | * revalidate the cookie. | |
1da177e4 LT |
772 | */ |
773 | memset(desc, 0, sizeof(*desc)); | |
774 | ||
775 | desc->file = filp; | |
cd3758e3 | 776 | desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; |
1da177e4 LT |
777 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
778 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
779 | ||
565277f6 | 780 | nfs_block_sillyrename(dentry); |
1cda707d | 781 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
782 | if (res < 0) |
783 | goto out; | |
784 | ||
d1bacf9e | 785 | while (desc->eof != 1) { |
1da177e4 | 786 | res = readdir_search_pagecache(desc); |
00a92642 | 787 | |
1da177e4 LT |
788 | if (res == -EBADCOOKIE) { |
789 | /* This means either end of directory */ | |
d1bacf9e | 790 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
791 | /* Or that the server has 'lost' a cookie */ |
792 | res = uncached_readdir(desc, dirent, filldir); | |
793 | if (res >= 0) | |
794 | continue; | |
795 | } | |
796 | res = 0; | |
797 | break; | |
798 | } | |
799 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 800 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 801 | nfs_zap_caches(inode); |
baf57a09 | 802 | desc->page_index = 0; |
1da177e4 | 803 | desc->plus = 0; |
d1bacf9e | 804 | desc->eof = 0; |
1da177e4 LT |
805 | continue; |
806 | } | |
807 | if (res < 0) | |
808 | break; | |
809 | ||
810 | res = nfs_do_filldir(desc, dirent, filldir); | |
811 | if (res < 0) { | |
812 | res = 0; | |
813 | break; | |
814 | } | |
815 | } | |
fccca7fc | 816 | out: |
565277f6 | 817 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
818 | if (res > 0) |
819 | res = 0; | |
aa49b4cf | 820 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
821 | dentry->d_parent->d_name.name, dentry->d_name.name, |
822 | res); | |
823 | return res; | |
1da177e4 LT |
824 | } |
825 | ||
10afec90 | 826 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 827 | { |
b84e06c5 CL |
828 | struct dentry *dentry = filp->f_path.dentry; |
829 | struct inode *inode = dentry->d_inode; | |
830 | ||
6da24bc9 | 831 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
832 | dentry->d_parent->d_name.name, |
833 | dentry->d_name.name, | |
834 | offset, origin); | |
835 | ||
836 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
837 | switch (origin) { |
838 | case 1: | |
839 | offset += filp->f_pos; | |
840 | case 0: | |
841 | if (offset >= 0) | |
842 | break; | |
843 | default: | |
844 | offset = -EINVAL; | |
845 | goto out; | |
846 | } | |
847 | if (offset != filp->f_pos) { | |
848 | filp->f_pos = offset; | |
cd3758e3 | 849 | nfs_file_open_context(filp)->dir_cookie = 0; |
f0dd2136 TM |
850 | } |
851 | out: | |
b84e06c5 | 852 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
853 | return offset; |
854 | } | |
855 | ||
1da177e4 LT |
856 | /* |
857 | * All directory operations under NFS are synchronous, so fsync() | |
858 | * is a dummy operation. | |
859 | */ | |
7ea80859 | 860 | static int nfs_fsync_dir(struct file *filp, int datasync) |
1da177e4 | 861 | { |
7ea80859 CH |
862 | struct dentry *dentry = filp->f_path.dentry; |
863 | ||
6da24bc9 | 864 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
865 | dentry->d_parent->d_name.name, dentry->d_name.name, |
866 | datasync); | |
867 | ||
54917786 | 868 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
1da177e4 LT |
869 | return 0; |
870 | } | |
871 | ||
bfc69a45 TM |
872 | /** |
873 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
874 | * @dir - pointer to directory inode | |
875 | * | |
876 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
877 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
878 | * on the server that might have invalidated our dcache. | |
879 | * | |
880 | * The caller should be holding dir->i_lock | |
881 | */ | |
882 | void nfs_force_lookup_revalidate(struct inode *dir) | |
883 | { | |
011935a0 | 884 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
885 | } |
886 | ||
1da177e4 LT |
887 | /* |
888 | * A check for whether or not the parent directory has changed. | |
889 | * In the case it has, we assume that the dentries are untrustworthy | |
890 | * and may need to be looked up again. | |
891 | */ | |
c79ba787 | 892 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
893 | { |
894 | if (IS_ROOT(dentry)) | |
895 | return 1; | |
4eec952e TM |
896 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
897 | return 0; | |
f2c77f4e TM |
898 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
899 | return 0; | |
900 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
901 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
902 | return 0; | |
903 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
904 | return 0; | |
905 | return 1; | |
1da177e4 LT |
906 | } |
907 | ||
1d6757fb TM |
908 | /* |
909 | * Return the intent data that applies to this particular path component | |
910 | * | |
911 | * Note that the current set of intents only apply to the very last | |
912 | * component of the path. | |
913 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
914 | */ | |
915 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
916 | { | |
917 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
918 | return 0; | |
919 | return nd->flags & mask; | |
920 | } | |
921 | ||
a12802ca TM |
922 | /* |
923 | * Use intent information to check whether or not we're going to do | |
924 | * an O_EXCL create using this path component. | |
925 | */ | |
926 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
927 | { | |
928 | if (NFS_PROTO(dir)->version == 2) | |
929 | return 0; | |
3516586a | 930 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
931 | } |
932 | ||
1d6757fb TM |
933 | /* |
934 | * Inode and filehandle revalidation for lookups. | |
935 | * | |
936 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
937 | * or if the intent information indicates that we're about to open this | |
938 | * particular file and the "nocto" mount flag is not set. | |
939 | * | |
940 | */ | |
1da177e4 LT |
941 | static inline |
942 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
943 | { | |
944 | struct nfs_server *server = NFS_SERVER(inode); | |
945 | ||
4e99a1ff TM |
946 | if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) |
947 | return 0; | |
1da177e4 | 948 | if (nd != NULL) { |
1da177e4 | 949 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 950 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
951 | goto out_force; |
952 | /* This is an open(2) */ | |
1d6757fb | 953 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
954 | !(server->flags & NFS_MOUNT_NOCTO) && |
955 | (S_ISREG(inode->i_mode) || | |
956 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 957 | goto out_force; |
4f48af45 | 958 | return 0; |
1da177e4 LT |
959 | } |
960 | return nfs_revalidate_inode(server, inode); | |
961 | out_force: | |
962 | return __nfs_revalidate_inode(server, inode); | |
963 | } | |
964 | ||
965 | /* | |
966 | * We judge how long we want to trust negative | |
967 | * dentries by looking at the parent inode mtime. | |
968 | * | |
969 | * If parent mtime has changed, we revalidate, else we wait for a | |
970 | * period corresponding to the parent's attribute cache timeout value. | |
971 | */ | |
972 | static inline | |
973 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
974 | struct nameidata *nd) | |
975 | { | |
1da177e4 | 976 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 977 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 978 | return 0; |
4eec952e TM |
979 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
980 | return 1; | |
1da177e4 LT |
981 | return !nfs_check_verifier(dir, dentry); |
982 | } | |
983 | ||
984 | /* | |
985 | * This is called every time the dcache has a lookup hit, | |
986 | * and we should check whether we can really trust that | |
987 | * lookup. | |
988 | * | |
989 | * NOTE! The hit can be a negative hit too, don't assume | |
990 | * we have an inode! | |
991 | * | |
992 | * If the parent directory is seen to have changed, we throw out the | |
993 | * cached dentry and do a new lookup. | |
994 | */ | |
995 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
996 | { | |
997 | struct inode *dir; | |
998 | struct inode *inode; | |
999 | struct dentry *parent; | |
e1fb4d05 TM |
1000 | struct nfs_fh *fhandle = NULL; |
1001 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1002 | int error; |
1da177e4 LT |
1003 | |
1004 | parent = dget_parent(dentry); | |
1da177e4 | 1005 | dir = parent->d_inode; |
91d5b470 | 1006 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1007 | inode = dentry->d_inode; |
1008 | ||
1009 | if (!inode) { | |
1010 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1011 | goto out_bad; | |
1012 | goto out_valid; | |
1013 | } | |
1014 | ||
1015 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1016 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1017 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1018 | dentry->d_name.name); |
1da177e4 LT |
1019 | goto out_bad; |
1020 | } | |
1021 | ||
15860ab1 TM |
1022 | if (nfs_have_delegation(inode, FMODE_READ)) |
1023 | goto out_set_verifier; | |
1024 | ||
1da177e4 | 1025 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1026 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1027 | if (nfs_lookup_verify_inode(inode, nd)) |
1028 | goto out_zap_parent; | |
1029 | goto out_valid; | |
1030 | } | |
1031 | ||
1032 | if (NFS_STALE(inode)) | |
1033 | goto out_bad; | |
1034 | ||
e1fb4d05 TM |
1035 | error = -ENOMEM; |
1036 | fhandle = nfs_alloc_fhandle(); | |
1037 | fattr = nfs_alloc_fattr(); | |
1038 | if (fhandle == NULL || fattr == NULL) | |
1039 | goto out_error; | |
1040 | ||
1041 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1da177e4 LT |
1042 | if (error) |
1043 | goto out_bad; | |
e1fb4d05 | 1044 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1045 | goto out_bad; |
e1fb4d05 | 1046 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1047 | goto out_bad; |
1048 | ||
e1fb4d05 TM |
1049 | nfs_free_fattr(fattr); |
1050 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1051 | out_set_verifier: |
cf8ba45e | 1052 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1053 | out_valid: |
1da177e4 | 1054 | dput(parent); |
1e7cb3dc | 1055 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1056 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1057 | dentry->d_name.name); |
1da177e4 LT |
1058 | return 1; |
1059 | out_zap_parent: | |
1060 | nfs_zap_caches(dir); | |
1061 | out_bad: | |
a1643a92 | 1062 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1063 | if (inode && S_ISDIR(inode->i_mode)) { |
1064 | /* Purge readdir caches. */ | |
1065 | nfs_zap_caches(inode); | |
1066 | /* If we have submounts, don't unhash ! */ | |
1067 | if (have_submounts(dentry)) | |
1068 | goto out_valid; | |
d9e80b7d AV |
1069 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1070 | goto out_valid; | |
1da177e4 LT |
1071 | shrink_dcache_parent(dentry); |
1072 | } | |
1073 | d_drop(dentry); | |
e1fb4d05 TM |
1074 | nfs_free_fattr(fattr); |
1075 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1076 | dput(parent); |
1e7cb3dc | 1077 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1078 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1079 | dentry->d_name.name); |
1da177e4 | 1080 | return 0; |
e1fb4d05 TM |
1081 | out_error: |
1082 | nfs_free_fattr(fattr); | |
1083 | nfs_free_fhandle(fhandle); | |
1084 | dput(parent); | |
1085 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1086 | __func__, dentry->d_parent->d_name.name, | |
1087 | dentry->d_name.name, error); | |
1088 | return error; | |
1da177e4 LT |
1089 | } |
1090 | ||
1091 | /* | |
1092 | * This is called from dput() when d_count is going to 0. | |
1093 | */ | |
1094 | static int nfs_dentry_delete(struct dentry *dentry) | |
1095 | { | |
1096 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1097 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1098 | dentry->d_flags); | |
1099 | ||
77f11192 TM |
1100 | /* Unhash any dentry with a stale inode */ |
1101 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1102 | return 1; | |
1103 | ||
1da177e4 LT |
1104 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1105 | /* Unhash it, so that ->d_iput() would be called */ | |
1106 | return 1; | |
1107 | } | |
1108 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1109 | /* Unhash it, so that ancestors of killed async unlink | |
1110 | * files will be cleaned up during umount */ | |
1111 | return 1; | |
1112 | } | |
1113 | return 0; | |
1114 | ||
1115 | } | |
1116 | ||
1b83d707 TM |
1117 | static void nfs_drop_nlink(struct inode *inode) |
1118 | { | |
1119 | spin_lock(&inode->i_lock); | |
1120 | if (inode->i_nlink > 0) | |
1121 | drop_nlink(inode); | |
1122 | spin_unlock(&inode->i_lock); | |
1123 | } | |
1124 | ||
1da177e4 LT |
1125 | /* |
1126 | * Called when the dentry loses inode. | |
1127 | * We use it to clean up silly-renamed files. | |
1128 | */ | |
1129 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1130 | { | |
83672d39 NB |
1131 | if (S_ISDIR(inode->i_mode)) |
1132 | /* drop any readdir cache as it could easily be old */ | |
1133 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1134 | ||
1da177e4 | 1135 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1136 | drop_nlink(inode); |
e4eff1a6 | 1137 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1138 | } |
1da177e4 LT |
1139 | iput(inode); |
1140 | } | |
1141 | ||
f786aa90 | 1142 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1143 | .d_revalidate = nfs_lookup_revalidate, |
1144 | .d_delete = nfs_dentry_delete, | |
1145 | .d_iput = nfs_dentry_iput, | |
1146 | }; | |
1147 | ||
1da177e4 LT |
1148 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1149 | { | |
1150 | struct dentry *res; | |
565277f6 | 1151 | struct dentry *parent; |
1da177e4 | 1152 | struct inode *inode = NULL; |
e1fb4d05 TM |
1153 | struct nfs_fh *fhandle = NULL; |
1154 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1155 | int error; |
1da177e4 LT |
1156 | |
1157 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1158 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1159 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1160 | |
1161 | res = ERR_PTR(-ENAMETOOLONG); | |
1162 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1163 | goto out; | |
1164 | ||
1da177e4 LT |
1165 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
1166 | ||
fd684071 TM |
1167 | /* |
1168 | * If we're doing an exclusive create, optimize away the lookup | |
1169 | * but don't hash the dentry. | |
1170 | */ | |
1171 | if (nfs_is_exclusive_create(dir, nd)) { | |
1172 | d_instantiate(dentry, NULL); | |
1173 | res = NULL; | |
fc0f684c | 1174 | goto out; |
fd684071 | 1175 | } |
1da177e4 | 1176 | |
e1fb4d05 TM |
1177 | res = ERR_PTR(-ENOMEM); |
1178 | fhandle = nfs_alloc_fhandle(); | |
1179 | fattr = nfs_alloc_fattr(); | |
1180 | if (fhandle == NULL || fattr == NULL) | |
1181 | goto out; | |
1182 | ||
565277f6 TM |
1183 | parent = dentry->d_parent; |
1184 | /* Protect against concurrent sillydeletes */ | |
1185 | nfs_block_sillyrename(parent); | |
e1fb4d05 | 1186 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1187 | if (error == -ENOENT) |
1188 | goto no_entry; | |
1189 | if (error < 0) { | |
1190 | res = ERR_PTR(error); | |
565277f6 | 1191 | goto out_unblock_sillyrename; |
1da177e4 | 1192 | } |
e1fb4d05 | 1193 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1194 | res = (struct dentry *)inode; |
1195 | if (IS_ERR(res)) | |
565277f6 | 1196 | goto out_unblock_sillyrename; |
54ceac45 | 1197 | |
1da177e4 | 1198 | no_entry: |
54ceac45 | 1199 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1200 | if (res != NULL) { |
1201 | if (IS_ERR(res)) | |
565277f6 | 1202 | goto out_unblock_sillyrename; |
1da177e4 | 1203 | dentry = res; |
9eaef27b | 1204 | } |
1da177e4 | 1205 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1206 | out_unblock_sillyrename: |
1207 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1208 | out: |
e1fb4d05 TM |
1209 | nfs_free_fattr(fattr); |
1210 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1211 | return res; |
1212 | } | |
1213 | ||
1214 | #ifdef CONFIG_NFS_V4 | |
1215 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
1216 | ||
f786aa90 | 1217 | const struct dentry_operations nfs4_dentry_operations = { |
1da177e4 LT |
1218 | .d_revalidate = nfs_open_revalidate, |
1219 | .d_delete = nfs_dentry_delete, | |
1220 | .d_iput = nfs_dentry_iput, | |
1221 | }; | |
1222 | ||
1d6757fb TM |
1223 | /* |
1224 | * Use intent information to determine whether we need to substitute | |
1225 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1226 | */ | |
5584c306 | 1227 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1228 | { |
1d6757fb | 1229 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1230 | return 0; |
1231 | /* NFS does not (yet) have a stateful open for directories */ | |
1232 | if (nd->flags & LOOKUP_DIRECTORY) | |
1233 | return 0; | |
1234 | /* Are we trying to write to a read only partition? */ | |
2c463e95 DH |
1235 | if (__mnt_is_readonly(nd->path.mnt) && |
1236 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
1da177e4 LT |
1237 | return 0; |
1238 | return 1; | |
1239 | } | |
1240 | ||
cd9a1c0e TM |
1241 | static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd) |
1242 | { | |
1243 | struct path path = { | |
1244 | .mnt = nd->path.mnt, | |
1245 | .dentry = dentry, | |
1246 | }; | |
1247 | struct nfs_open_context *ctx; | |
1248 | struct rpc_cred *cred; | |
1249 | fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); | |
1250 | ||
1251 | cred = rpc_lookup_cred(); | |
1252 | if (IS_ERR(cred)) | |
1253 | return ERR_CAST(cred); | |
1254 | ctx = alloc_nfs_open_context(&path, cred, fmode); | |
1255 | put_rpccred(cred); | |
1256 | if (ctx == NULL) | |
1257 | return ERR_PTR(-ENOMEM); | |
1258 | return ctx; | |
1259 | } | |
1260 | ||
1261 | static int do_open(struct inode *inode, struct file *filp) | |
1262 | { | |
1263 | nfs_fscache_set_inode_cookie(inode, filp); | |
1264 | return 0; | |
1265 | } | |
1266 | ||
1267 | static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx) | |
1268 | { | |
1269 | struct file *filp; | |
1270 | int ret = 0; | |
1271 | ||
1272 | /* If the open_intent is for execute, we have an extra check to make */ | |
1273 | if (ctx->mode & FMODE_EXEC) { | |
1274 | ret = nfs_may_open(ctx->path.dentry->d_inode, | |
1275 | ctx->cred, | |
1276 | nd->intent.open.flags); | |
1277 | if (ret < 0) | |
1278 | goto out; | |
1279 | } | |
1280 | filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open); | |
1281 | if (IS_ERR(filp)) | |
1282 | ret = PTR_ERR(filp); | |
1283 | else | |
1284 | nfs_file_set_open_context(filp, ctx); | |
1285 | out: | |
1286 | put_nfs_open_context(ctx); | |
1287 | return ret; | |
1288 | } | |
1289 | ||
1da177e4 LT |
1290 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
1291 | { | |
cd9a1c0e TM |
1292 | struct nfs_open_context *ctx; |
1293 | struct iattr attr; | |
1da177e4 | 1294 | struct dentry *res = NULL; |
f46e0bd3 | 1295 | struct inode *inode; |
cd9a1c0e | 1296 | int open_flags; |
898f635c | 1297 | int err; |
1da177e4 | 1298 | |
1e7cb3dc CL |
1299 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
1300 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1301 | ||
1da177e4 | 1302 | /* Check that we are indeed trying to open this file */ |
5584c306 | 1303 | if (!is_atomic_open(nd)) |
1da177e4 LT |
1304 | goto no_open; |
1305 | ||
1306 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1307 | res = ERR_PTR(-ENAMETOOLONG); | |
1308 | goto out; | |
1309 | } | |
1310 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1311 | ||
d4d9cdcb TM |
1312 | /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash |
1313 | * the dentry. */ | |
3516586a | 1314 | if (nd->flags & LOOKUP_EXCL) { |
d4d9cdcb | 1315 | d_instantiate(dentry, NULL); |
02a913a7 TM |
1316 | goto out; |
1317 | } | |
1da177e4 | 1318 | |
cd9a1c0e TM |
1319 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1320 | res = ERR_CAST(ctx); | |
1321 | if (IS_ERR(ctx)) | |
1322 | goto out; | |
1323 | ||
1324 | open_flags = nd->intent.open.flags; | |
1325 | if (nd->flags & LOOKUP_CREATE) { | |
1326 | attr.ia_mode = nd->intent.open.create_mode; | |
1327 | attr.ia_valid = ATTR_MODE; | |
1328 | if (!IS_POSIXACL(dir)) | |
1329 | attr.ia_mode &= ~current_umask(); | |
1330 | } else { | |
898f635c | 1331 | open_flags &= ~(O_EXCL | O_CREAT); |
cd9a1c0e | 1332 | attr.ia_valid = 0; |
cd9a1c0e TM |
1333 | } |
1334 | ||
1da177e4 | 1335 | /* Open the file on the server */ |
f46e0bd3 | 1336 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1337 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
f46e0bd3 TM |
1338 | if (IS_ERR(inode)) { |
1339 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1340 | put_nfs_open_context(ctx); |
f46e0bd3 | 1341 | switch (PTR_ERR(inode)) { |
1da177e4 LT |
1342 | /* Make a negative dentry */ |
1343 | case -ENOENT: | |
f46e0bd3 | 1344 | d_add(dentry, NULL); |
02a913a7 TM |
1345 | res = NULL; |
1346 | goto out; | |
1da177e4 | 1347 | /* This turned out not to be a regular file */ |
80e60639 | 1348 | case -EISDIR: |
6f926b5b TM |
1349 | case -ENOTDIR: |
1350 | goto no_open; | |
1da177e4 LT |
1351 | case -ELOOP: |
1352 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1353 | goto no_open; | |
1da177e4 LT |
1354 | /* case -EINVAL: */ |
1355 | default: | |
f46e0bd3 | 1356 | res = ERR_CAST(inode); |
1da177e4 LT |
1357 | goto out; |
1358 | } | |
cd9a1c0e | 1359 | } |
f46e0bd3 | 1360 | res = d_add_unique(dentry, inode); |
898f635c | 1361 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 TM |
1362 | if (res != NULL) { |
1363 | dput(ctx->path.dentry); | |
1364 | ctx->path.dentry = dget(res); | |
1da177e4 | 1365 | dentry = res; |
f46e0bd3 | 1366 | } |
898f635c TM |
1367 | err = nfs_intent_set_file(nd, ctx); |
1368 | if (err < 0) { | |
1369 | if (res != NULL) | |
1370 | dput(res); | |
1371 | return ERR_PTR(err); | |
1372 | } | |
1da177e4 | 1373 | out: |
f46e0bd3 | 1374 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1375 | return res; |
1376 | no_open: | |
1377 | return nfs_lookup(dir, dentry, nd); | |
1378 | } | |
1379 | ||
1380 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1381 | { | |
1382 | struct dentry *parent = NULL; | |
1383 | struct inode *inode = dentry->d_inode; | |
1384 | struct inode *dir; | |
b8d4cadd | 1385 | struct nfs_open_context *ctx; |
1da177e4 LT |
1386 | int openflags, ret = 0; |
1387 | ||
1f063d2c | 1388 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1389 | goto no_open; |
2b484297 | 1390 | |
1da177e4 LT |
1391 | parent = dget_parent(dentry); |
1392 | dir = parent->d_inode; | |
2b484297 | 1393 | |
1da177e4 LT |
1394 | /* We can't create new files in nfs_open_revalidate(), so we |
1395 | * optimize away revalidation of negative dentries. | |
1396 | */ | |
216d5d06 TM |
1397 | if (inode == NULL) { |
1398 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1399 | ret = 1; | |
1da177e4 | 1400 | goto out; |
216d5d06 TM |
1401 | } |
1402 | ||
1da177e4 LT |
1403 | /* NFS only supports OPEN on regular files */ |
1404 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1405 | goto no_open_dput; |
1da177e4 LT |
1406 | openflags = nd->intent.open.flags; |
1407 | /* We cannot do exclusive creation on a positive dentry */ | |
1408 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1409 | goto no_open_dput; |
1da177e4 | 1410 | /* We can't create new files, or truncate existing ones here */ |
0a377cff | 1411 | openflags &= ~(O_CREAT|O_EXCL|O_TRUNC); |
1da177e4 | 1412 | |
b8d4cadd TM |
1413 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1414 | ret = PTR_ERR(ctx); | |
1415 | if (IS_ERR(ctx)) | |
1416 | goto out; | |
1da177e4 | 1417 | /* |
1b1dcc1b | 1418 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1419 | * operations that change the directory. We therefore save the |
1420 | * change attribute *before* we do the RPC call. | |
1421 | */ | |
2b484297 | 1422 | inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL); |
535918f1 TM |
1423 | if (IS_ERR(inode)) { |
1424 | ret = PTR_ERR(inode); | |
1425 | switch (ret) { | |
1426 | case -EPERM: | |
1427 | case -EACCES: | |
1428 | case -EDQUOT: | |
1429 | case -ENOSPC: | |
1430 | case -EROFS: | |
1431 | goto out_put_ctx; | |
1432 | default: | |
1433 | goto out_drop; | |
1434 | } | |
1435 | } | |
1436 | iput(inode); | |
898f635c | 1437 | if (inode != dentry->d_inode) |
535918f1 | 1438 | goto out_drop; |
898f635c TM |
1439 | |
1440 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1441 | ret = nfs_intent_set_file(nd, ctx); | |
1442 | if (ret >= 0) | |
1443 | ret = 1; | |
1da177e4 LT |
1444 | out: |
1445 | dput(parent); | |
1da177e4 | 1446 | return ret; |
535918f1 TM |
1447 | out_drop: |
1448 | d_drop(dentry); | |
1449 | ret = 0; | |
1450 | out_put_ctx: | |
1451 | put_nfs_open_context(ctx); | |
1452 | goto out; | |
1453 | ||
5584c306 | 1454 | no_open_dput: |
1da177e4 | 1455 | dput(parent); |
5584c306 | 1456 | no_open: |
1da177e4 LT |
1457 | return nfs_lookup_revalidate(dentry, nd); |
1458 | } | |
c0204fd2 TM |
1459 | |
1460 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, | |
1461 | struct nameidata *nd) | |
1462 | { | |
1463 | struct nfs_open_context *ctx = NULL; | |
1464 | struct iattr attr; | |
1465 | int error; | |
1466 | int open_flags = 0; | |
1467 | ||
1468 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1469 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1470 | ||
1471 | attr.ia_mode = mode; | |
1472 | attr.ia_valid = ATTR_MODE; | |
1473 | ||
1474 | if ((nd->flags & LOOKUP_CREATE) != 0) { | |
1475 | open_flags = nd->intent.open.flags; | |
1476 | ||
1477 | ctx = nameidata_to_nfs_open_context(dentry, nd); | |
1478 | error = PTR_ERR(ctx); | |
1479 | if (IS_ERR(ctx)) | |
898f635c | 1480 | goto out_err_drop; |
c0204fd2 TM |
1481 | } |
1482 | ||
1483 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1484 | if (error != 0) | |
1485 | goto out_put_ctx; | |
898f635c TM |
1486 | if (ctx != NULL) { |
1487 | error = nfs_intent_set_file(nd, ctx); | |
1488 | if (error < 0) | |
1489 | goto out_err; | |
1490 | } | |
c0204fd2 TM |
1491 | return 0; |
1492 | out_put_ctx: | |
1493 | if (ctx != NULL) | |
1494 | put_nfs_open_context(ctx); | |
898f635c | 1495 | out_err_drop: |
c0204fd2 | 1496 | d_drop(dentry); |
898f635c | 1497 | out_err: |
c0204fd2 TM |
1498 | return error; |
1499 | } | |
1500 | ||
1da177e4 LT |
1501 | #endif /* CONFIG_NFSV4 */ |
1502 | ||
1da177e4 LT |
1503 | /* |
1504 | * Code common to create, mkdir, and mknod. | |
1505 | */ | |
1506 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1507 | struct nfs_fattr *fattr) | |
1508 | { | |
fab728e1 TM |
1509 | struct dentry *parent = dget_parent(dentry); |
1510 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1511 | struct inode *inode; |
1512 | int error = -EACCES; | |
1513 | ||
fab728e1 TM |
1514 | d_drop(dentry); |
1515 | ||
1da177e4 LT |
1516 | /* We may have been initialized further down */ |
1517 | if (dentry->d_inode) | |
fab728e1 | 1518 | goto out; |
1da177e4 | 1519 | if (fhandle->size == 0) { |
1da177e4 LT |
1520 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1521 | if (error) | |
fab728e1 | 1522 | goto out_error; |
1da177e4 | 1523 | } |
5724ab37 | 1524 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1525 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1526 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1527 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1528 | if (error < 0) |
fab728e1 | 1529 | goto out_error; |
1da177e4 | 1530 | } |
1da177e4 | 1531 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1532 | error = PTR_ERR(inode); |
1533 | if (IS_ERR(inode)) | |
fab728e1 TM |
1534 | goto out_error; |
1535 | d_add(dentry, inode); | |
1536 | out: | |
1537 | dput(parent); | |
1da177e4 | 1538 | return 0; |
fab728e1 TM |
1539 | out_error: |
1540 | nfs_mark_for_revalidate(dir); | |
1541 | dput(parent); | |
1542 | return error; | |
1da177e4 LT |
1543 | } |
1544 | ||
1545 | /* | |
1546 | * Following a failed create operation, we drop the dentry rather | |
1547 | * than retain a negative dentry. This avoids a problem in the event | |
1548 | * that the operation succeeded on the server, but an error in the | |
1549 | * reply path made it appear to have failed. | |
1550 | */ | |
1551 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1552 | struct nameidata *nd) | |
1553 | { | |
1554 | struct iattr attr; | |
1555 | int error; | |
1da177e4 | 1556 | |
1e7cb3dc CL |
1557 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1558 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1559 | |
1560 | attr.ia_mode = mode; | |
1561 | attr.ia_valid = ATTR_MODE; | |
1562 | ||
c0204fd2 | 1563 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL); |
1da177e4 LT |
1564 | if (error != 0) |
1565 | goto out_err; | |
1da177e4 LT |
1566 | return 0; |
1567 | out_err: | |
1da177e4 LT |
1568 | d_drop(dentry); |
1569 | return error; | |
1570 | } | |
1571 | ||
1572 | /* | |
1573 | * See comments for nfs_proc_create regarding failed operations. | |
1574 | */ | |
1575 | static int | |
1576 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1577 | { | |
1578 | struct iattr attr; | |
1579 | int status; | |
1580 | ||
1e7cb3dc CL |
1581 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1582 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1583 | |
1584 | if (!new_valid_dev(rdev)) | |
1585 | return -EINVAL; | |
1586 | ||
1587 | attr.ia_mode = mode; | |
1588 | attr.ia_valid = ATTR_MODE; | |
1589 | ||
1da177e4 | 1590 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1591 | if (status != 0) |
1592 | goto out_err; | |
1da177e4 LT |
1593 | return 0; |
1594 | out_err: | |
1da177e4 LT |
1595 | d_drop(dentry); |
1596 | return status; | |
1597 | } | |
1598 | ||
1599 | /* | |
1600 | * See comments for nfs_proc_create regarding failed operations. | |
1601 | */ | |
1602 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1603 | { | |
1604 | struct iattr attr; | |
1605 | int error; | |
1606 | ||
1e7cb3dc CL |
1607 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1608 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1609 | |
1610 | attr.ia_valid = ATTR_MODE; | |
1611 | attr.ia_mode = mode | S_IFDIR; | |
1612 | ||
1da177e4 | 1613 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1614 | if (error != 0) |
1615 | goto out_err; | |
1da177e4 LT |
1616 | return 0; |
1617 | out_err: | |
1618 | d_drop(dentry); | |
1da177e4 LT |
1619 | return error; |
1620 | } | |
1621 | ||
d45b9d8b TM |
1622 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1623 | { | |
1624 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1625 | d_delete(dentry); | |
1626 | } | |
1627 | ||
1da177e4 LT |
1628 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1629 | { | |
1630 | int error; | |
1631 | ||
1e7cb3dc CL |
1632 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1633 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1634 | |
1da177e4 LT |
1635 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1636 | /* Ensure the VFS deletes this inode */ | |
1637 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1638 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1639 | else if (error == -ENOENT) |
1640 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1641 | |
1642 | return error; | |
1643 | } | |
1644 | ||
1da177e4 LT |
1645 | /* |
1646 | * Remove a file after making sure there are no pending writes, | |
1647 | * and after checking that the file has only one user. | |
1648 | * | |
1649 | * We invalidate the attribute cache and free the inode prior to the operation | |
1650 | * to avoid possible races if the server reuses the inode. | |
1651 | */ | |
1652 | static int nfs_safe_remove(struct dentry *dentry) | |
1653 | { | |
1654 | struct inode *dir = dentry->d_parent->d_inode; | |
1655 | struct inode *inode = dentry->d_inode; | |
1656 | int error = -EBUSY; | |
1657 | ||
1658 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1659 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1660 | ||
1661 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1662 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1663 | error = 0; | |
1664 | goto out; | |
1665 | } | |
1666 | ||
1da177e4 | 1667 | if (inode != NULL) { |
cae7a073 | 1668 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1669 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1670 | /* The VFS may want to delete this inode */ | |
1671 | if (error == 0) | |
1b83d707 | 1672 | nfs_drop_nlink(inode); |
5ba7cc48 | 1673 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1674 | } else |
1675 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1676 | if (error == -ENOENT) |
1677 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1678 | out: |
1679 | return error; | |
1680 | } | |
1681 | ||
1682 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1683 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1684 | * | |
1685 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1686 | */ | |
1687 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1688 | { | |
1689 | int error; | |
1690 | int need_rehash = 0; | |
1691 | ||
1692 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1693 | dir->i_ino, dentry->d_name.name); | |
1694 | ||
1da177e4 LT |
1695 | spin_lock(&dcache_lock); |
1696 | spin_lock(&dentry->d_lock); | |
1697 | if (atomic_read(&dentry->d_count) > 1) { | |
1698 | spin_unlock(&dentry->d_lock); | |
1699 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1700 | /* Start asynchronous writeout of the inode */ |
1701 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1702 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1703 | return error; |
1704 | } | |
1705 | if (!d_unhashed(dentry)) { | |
1706 | __d_drop(dentry); | |
1707 | need_rehash = 1; | |
1708 | } | |
1709 | spin_unlock(&dentry->d_lock); | |
1710 | spin_unlock(&dcache_lock); | |
1711 | error = nfs_safe_remove(dentry); | |
d45b9d8b | 1712 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1713 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1714 | } else if (need_rehash) | |
1715 | d_rehash(dentry); | |
1da177e4 LT |
1716 | return error; |
1717 | } | |
1718 | ||
873101b3 CL |
1719 | /* |
1720 | * To create a symbolic link, most file systems instantiate a new inode, | |
1721 | * add a page to it containing the path, then write it out to the disk | |
1722 | * using prepare_write/commit_write. | |
1723 | * | |
1724 | * Unfortunately the NFS client can't create the in-core inode first | |
1725 | * because it needs a file handle to create an in-core inode (see | |
1726 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1727 | * symlink request has completed on the server. | |
1728 | * | |
1729 | * So instead we allocate a raw page, copy the symname into it, then do | |
1730 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1731 | * now have a new file handle and can instantiate an in-core NFS inode | |
1732 | * and move the raw page into its mapping. | |
1733 | */ | |
1734 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1735 | { |
873101b3 CL |
1736 | struct pagevec lru_pvec; |
1737 | struct page *page; | |
1738 | char *kaddr; | |
1da177e4 | 1739 | struct iattr attr; |
873101b3 | 1740 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1741 | int error; |
1742 | ||
1743 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1744 | dir->i_ino, dentry->d_name.name, symname); | |
1745 | ||
873101b3 CL |
1746 | if (pathlen > PAGE_SIZE) |
1747 | return -ENAMETOOLONG; | |
1da177e4 | 1748 | |
873101b3 CL |
1749 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1750 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1751 | |
83d93f22 | 1752 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1753 | if (!page) |
873101b3 | 1754 | return -ENOMEM; |
873101b3 CL |
1755 | |
1756 | kaddr = kmap_atomic(page, KM_USER0); | |
1757 | memcpy(kaddr, symname, pathlen); | |
1758 | if (pathlen < PAGE_SIZE) | |
1759 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1760 | kunmap_atomic(kaddr, KM_USER0); | |
1761 | ||
94a6d753 | 1762 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1763 | if (error != 0) { |
1764 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1765 | dir->i_sb->s_id, dir->i_ino, | |
1766 | dentry->d_name.name, symname, error); | |
1da177e4 | 1767 | d_drop(dentry); |
873101b3 | 1768 | __free_page(page); |
873101b3 CL |
1769 | return error; |
1770 | } | |
1771 | ||
1772 | /* | |
1773 | * No big deal if we can't add this page to the page cache here. | |
1774 | * READLINK will get the missing page from the server if needed. | |
1775 | */ | |
1776 | pagevec_init(&lru_pvec, 0); | |
1777 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1778 | GFP_KERNEL)) { | |
39cf8a13 | 1779 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1780 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1781 | SetPageUptodate(page); |
1782 | unlock_page(page); | |
1783 | } else | |
1784 | __free_page(page); | |
1785 | ||
873101b3 | 1786 | return 0; |
1da177e4 LT |
1787 | } |
1788 | ||
1789 | static int | |
1790 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1791 | { | |
1792 | struct inode *inode = old_dentry->d_inode; | |
1793 | int error; | |
1794 | ||
1795 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1796 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1797 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1798 | ||
9a3936aa TM |
1799 | nfs_inode_return_delegation(inode); |
1800 | ||
9697d234 | 1801 | d_drop(dentry); |
1da177e4 | 1802 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1803 | if (error == 0) { |
7de9c6ee | 1804 | ihold(inode); |
9697d234 | 1805 | d_add(dentry, inode); |
cf809556 | 1806 | } |
1da177e4 LT |
1807 | return error; |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * RENAME | |
1812 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1813 | * different file handle for the same inode after a rename (e.g. when | |
1814 | * moving to a different directory). A fail-safe method to do so would | |
1815 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1816 | * rename the old file using the sillyrename stuff. This way, the original | |
1817 | * file in old_dir will go away when the last process iput()s the inode. | |
1818 | * | |
1819 | * FIXED. | |
1820 | * | |
1821 | * It actually works quite well. One needs to have the possibility for | |
1822 | * at least one ".nfs..." file in each directory the file ever gets | |
1823 | * moved or linked to which happens automagically with the new | |
1824 | * implementation that only depends on the dcache stuff instead of | |
1825 | * using the inode layer | |
1826 | * | |
1827 | * Unfortunately, things are a little more complicated than indicated | |
1828 | * above. For a cross-directory move, we want to make sure we can get | |
1829 | * rid of the old inode after the operation. This means there must be | |
1830 | * no pending writes (if it's a file), and the use count must be 1. | |
1831 | * If these conditions are met, we can drop the dentries before doing | |
1832 | * the rename. | |
1833 | */ | |
1834 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1835 | struct inode *new_dir, struct dentry *new_dentry) | |
1836 | { | |
1837 | struct inode *old_inode = old_dentry->d_inode; | |
1838 | struct inode *new_inode = new_dentry->d_inode; | |
1839 | struct dentry *dentry = NULL, *rehash = NULL; | |
1840 | int error = -EBUSY; | |
1841 | ||
1da177e4 LT |
1842 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1843 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1844 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1845 | atomic_read(&new_dentry->d_count)); | |
1846 | ||
1847 | /* | |
28f79a1a MS |
1848 | * For non-directories, check whether the target is busy and if so, |
1849 | * make a copy of the dentry and then do a silly-rename. If the | |
1850 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1851 | * the new target. | |
1da177e4 | 1852 | */ |
27226104 MS |
1853 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1854 | /* | |
1855 | * To prevent any new references to the target during the | |
1856 | * rename, we unhash the dentry in advance. | |
1857 | */ | |
1858 | if (!d_unhashed(new_dentry)) { | |
1859 | d_drop(new_dentry); | |
1860 | rehash = new_dentry; | |
1861 | } | |
1da177e4 | 1862 | |
27226104 MS |
1863 | if (atomic_read(&new_dentry->d_count) > 2) { |
1864 | int err; | |
1865 | ||
1866 | /* copy the target dentry's name */ | |
1867 | dentry = d_alloc(new_dentry->d_parent, | |
1868 | &new_dentry->d_name); | |
1869 | if (!dentry) | |
1870 | goto out; | |
1871 | ||
1872 | /* silly-rename the existing target ... */ | |
1873 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1874 | if (err) |
27226104 | 1875 | goto out; |
24e93025 MS |
1876 | |
1877 | new_dentry = dentry; | |
56335936 | 1878 | rehash = NULL; |
24e93025 | 1879 | new_inode = NULL; |
27226104 | 1880 | } |
b1e4adf4 | 1881 | } |
1da177e4 | 1882 | |
cae7a073 | 1883 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 1884 | if (new_inode != NULL) |
24174119 | 1885 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 1886 | |
1da177e4 LT |
1887 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1888 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1889 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1890 | out: |
1891 | if (rehash) | |
1892 | d_rehash(rehash); | |
1893 | if (!error) { | |
b1e4adf4 TM |
1894 | if (new_inode != NULL) |
1895 | nfs_drop_nlink(new_inode); | |
349457cc | 1896 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1897 | nfs_set_verifier(new_dentry, |
1898 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1899 | } else if (error == -ENOENT) |
1900 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1901 | |
1902 | /* new dentry created? */ | |
1903 | if (dentry) | |
1904 | dput(dentry); | |
1da177e4 LT |
1905 | return error; |
1906 | } | |
1907 | ||
cfcea3e8 TM |
1908 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1909 | static LIST_HEAD(nfs_access_lru_list); | |
1910 | static atomic_long_t nfs_access_nr_entries; | |
1911 | ||
1c3c07e9 TM |
1912 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1913 | { | |
1914 | put_rpccred(entry->cred); | |
1915 | kfree(entry); | |
cfcea3e8 TM |
1916 | smp_mb__before_atomic_dec(); |
1917 | atomic_long_dec(&nfs_access_nr_entries); | |
1918 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1919 | } |
1920 | ||
1a81bb8a TM |
1921 | static void nfs_access_free_list(struct list_head *head) |
1922 | { | |
1923 | struct nfs_access_entry *cache; | |
1924 | ||
1925 | while (!list_empty(head)) { | |
1926 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1927 | list_del(&cache->lru); | |
1928 | nfs_access_free_entry(cache); | |
1929 | } | |
1930 | } | |
1931 | ||
7f8275d0 | 1932 | int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
979df72e TM |
1933 | { |
1934 | LIST_HEAD(head); | |
aa510da5 | 1935 | struct nfs_inode *nfsi, *next; |
979df72e TM |
1936 | struct nfs_access_entry *cache; |
1937 | ||
61d5eb29 TM |
1938 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1939 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 1940 | |
a50f7951 | 1941 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 1942 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
1943 | struct inode *inode; |
1944 | ||
1945 | if (nr_to_scan-- == 0) | |
1946 | break; | |
9c7e7e23 | 1947 | inode = &nfsi->vfs_inode; |
979df72e TM |
1948 | spin_lock(&inode->i_lock); |
1949 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1950 | goto remove_lru_entry; | |
1951 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1952 | struct nfs_access_entry, lru); | |
1953 | list_move(&cache->lru, &head); | |
1954 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1955 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1956 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1957 | &nfs_access_lru_list); | |
1958 | else { | |
1959 | remove_lru_entry: | |
1960 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 1961 | smp_mb__before_clear_bit(); |
979df72e | 1962 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 1963 | smp_mb__after_clear_bit(); |
979df72e | 1964 | } |
59844a9b | 1965 | spin_unlock(&inode->i_lock); |
979df72e TM |
1966 | } |
1967 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 1968 | nfs_access_free_list(&head); |
979df72e TM |
1969 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
1970 | } | |
1971 | ||
1a81bb8a | 1972 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 1973 | { |
1c3c07e9 | 1974 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 1975 | struct rb_node *n; |
1c3c07e9 TM |
1976 | struct nfs_access_entry *entry; |
1977 | ||
1978 | /* Unhook entries from the cache */ | |
1979 | while ((n = rb_first(root_node)) != NULL) { | |
1980 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
1981 | rb_erase(n, root_node); | |
1a81bb8a | 1982 | list_move(&entry->lru, head); |
1c3c07e9 TM |
1983 | } |
1984 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
1985 | } |
1986 | ||
1c3c07e9 | 1987 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 1988 | { |
1a81bb8a TM |
1989 | LIST_HEAD(head); |
1990 | ||
1991 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
1992 | return; | |
cfcea3e8 | 1993 | /* Remove from global LRU init */ |
1a81bb8a TM |
1994 | spin_lock(&nfs_access_lru_lock); |
1995 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 1996 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 1997 | |
1c3c07e9 | 1998 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
1999 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2000 | spin_unlock(&inode->i_lock); | |
2001 | spin_unlock(&nfs_access_lru_lock); | |
2002 | nfs_access_free_list(&head); | |
1c3c07e9 | 2003 | } |
1da177e4 | 2004 | |
1c3c07e9 TM |
2005 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2006 | { | |
2007 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2008 | struct nfs_access_entry *entry; | |
2009 | ||
2010 | while (n != NULL) { | |
2011 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2012 | ||
2013 | if (cred < entry->cred) | |
2014 | n = n->rb_left; | |
2015 | else if (cred > entry->cred) | |
2016 | n = n->rb_right; | |
2017 | else | |
2018 | return entry; | |
1da177e4 | 2019 | } |
1c3c07e9 TM |
2020 | return NULL; |
2021 | } | |
2022 | ||
af22f94a | 2023 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2024 | { |
2025 | struct nfs_inode *nfsi = NFS_I(inode); | |
2026 | struct nfs_access_entry *cache; | |
2027 | int err = -ENOENT; | |
2028 | ||
dc59250c | 2029 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2030 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2031 | goto out_zap; | |
2032 | cache = nfs_access_search_rbtree(inode, cred); | |
2033 | if (cache == NULL) | |
2034 | goto out; | |
b4d2314b | 2035 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2036 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2037 | goto out_stale; |
2038 | res->jiffies = cache->jiffies; | |
2039 | res->cred = cache->cred; | |
2040 | res->mask = cache->mask; | |
cfcea3e8 | 2041 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2042 | err = 0; |
2043 | out: | |
2044 | spin_unlock(&inode->i_lock); | |
2045 | return err; | |
2046 | out_stale: | |
2047 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2048 | list_del(&cache->lru); |
1c3c07e9 TM |
2049 | spin_unlock(&inode->i_lock); |
2050 | nfs_access_free_entry(cache); | |
2051 | return -ENOENT; | |
2052 | out_zap: | |
1a81bb8a TM |
2053 | spin_unlock(&inode->i_lock); |
2054 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2055 | return -ENOENT; |
2056 | } | |
2057 | ||
2058 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2059 | { | |
cfcea3e8 TM |
2060 | struct nfs_inode *nfsi = NFS_I(inode); |
2061 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2062 | struct rb_node **p = &root_node->rb_node; |
2063 | struct rb_node *parent = NULL; | |
2064 | struct nfs_access_entry *entry; | |
2065 | ||
2066 | spin_lock(&inode->i_lock); | |
2067 | while (*p != NULL) { | |
2068 | parent = *p; | |
2069 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2070 | ||
2071 | if (set->cred < entry->cred) | |
2072 | p = &parent->rb_left; | |
2073 | else if (set->cred > entry->cred) | |
2074 | p = &parent->rb_right; | |
2075 | else | |
2076 | goto found; | |
2077 | } | |
2078 | rb_link_node(&set->rb_node, parent, p); | |
2079 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2080 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2081 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2082 | return; |
2083 | found: | |
2084 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2085 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2086 | list_del(&entry->lru); | |
1c3c07e9 TM |
2087 | spin_unlock(&inode->i_lock); |
2088 | nfs_access_free_entry(entry); | |
2089 | } | |
2090 | ||
af22f94a | 2091 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2092 | { |
2093 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2094 | if (cache == NULL) | |
2095 | return; | |
2096 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2097 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2098 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2099 | cache->mask = set->mask; |
1c3c07e9 TM |
2100 | |
2101 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2102 | |
2103 | /* Update accounting */ | |
2104 | smp_mb__before_atomic_inc(); | |
2105 | atomic_long_inc(&nfs_access_nr_entries); | |
2106 | smp_mb__after_atomic_inc(); | |
2107 | ||
2108 | /* Add inode to global LRU list */ | |
1a81bb8a | 2109 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2110 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2111 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2112 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2113 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2114 | spin_unlock(&nfs_access_lru_lock); |
2115 | } | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2119 | { | |
2120 | struct nfs_access_entry cache; | |
2121 | int status; | |
2122 | ||
2123 | status = nfs_access_get_cached(inode, cred, &cache); | |
2124 | if (status == 0) | |
2125 | goto out; | |
2126 | ||
2127 | /* Be clever: ask server to check for all possible rights */ | |
2128 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2129 | cache.cred = cred; | |
2130 | cache.jiffies = jiffies; | |
2131 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2132 | if (status != 0) { |
2133 | if (status == -ESTALE) { | |
2134 | nfs_zap_caches(inode); | |
2135 | if (!S_ISDIR(inode->i_mode)) | |
2136 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2137 | } | |
1da177e4 | 2138 | return status; |
a71ee337 | 2139 | } |
1da177e4 LT |
2140 | nfs_access_add_cache(inode, &cache); |
2141 | out: | |
e6305c43 | 2142 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2143 | return 0; |
2144 | return -EACCES; | |
2145 | } | |
2146 | ||
af22f94a TM |
2147 | static int nfs_open_permission_mask(int openflags) |
2148 | { | |
2149 | int mask = 0; | |
2150 | ||
2151 | if (openflags & FMODE_READ) | |
2152 | mask |= MAY_READ; | |
2153 | if (openflags & FMODE_WRITE) | |
2154 | mask |= MAY_WRITE; | |
2155 | if (openflags & FMODE_EXEC) | |
2156 | mask |= MAY_EXEC; | |
2157 | return mask; | |
2158 | } | |
2159 | ||
2160 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2161 | { | |
2162 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2163 | } | |
2164 | ||
e6305c43 | 2165 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2166 | { |
2167 | struct rpc_cred *cred; | |
2168 | int res = 0; | |
2169 | ||
91d5b470 CL |
2170 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2171 | ||
e6305c43 | 2172 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2173 | goto out; |
2174 | /* Is this sys_access() ? */ | |
9cfcac81 | 2175 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2176 | goto force_lookup; |
2177 | ||
2178 | switch (inode->i_mode & S_IFMT) { | |
2179 | case S_IFLNK: | |
2180 | goto out; | |
2181 | case S_IFREG: | |
2182 | /* NFSv4 has atomic_open... */ | |
2183 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2184 | && (mask & MAY_OPEN) |
2185 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2186 | goto out; |
2187 | break; | |
2188 | case S_IFDIR: | |
2189 | /* | |
2190 | * Optimize away all write operations, since the server | |
2191 | * will check permissions when we perform the op. | |
2192 | */ | |
2193 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2194 | goto out; | |
2195 | } | |
2196 | ||
2197 | force_lookup: | |
1da177e4 LT |
2198 | if (!NFS_PROTO(inode)->access) |
2199 | goto out_notsup; | |
2200 | ||
98a8e323 | 2201 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2202 | if (!IS_ERR(cred)) { |
2203 | res = nfs_do_access(inode, cred, mask); | |
2204 | put_rpccred(cred); | |
2205 | } else | |
2206 | res = PTR_ERR(cred); | |
1da177e4 | 2207 | out: |
f696a365 MS |
2208 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2209 | res = -EACCES; | |
2210 | ||
1e7cb3dc CL |
2211 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2212 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2213 | return res; |
2214 | out_notsup: | |
2215 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2216 | if (res == 0) | |
2217 | res = generic_permission(inode, mask, NULL); | |
1e7cb3dc | 2218 | goto out; |
1da177e4 LT |
2219 | } |
2220 | ||
2221 | /* | |
2222 | * Local variables: | |
2223 | * version-control: t | |
2224 | * kept-new-versions: 5 | |
2225 | * End: | |
2226 | */ |