Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/nfs/dir.c | |
3 | * | |
4 | * Copyright (C) 1992 Rick Sladkey | |
5 | * | |
6 | * nfs directory handling functions | |
7 | * | |
8 | * 10 Apr 1996 Added silly rename for unlink --okir | |
9 | * 28 Sep 1996 Improved directory cache --okir | |
10 | * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de | |
11 | * Re-implemented silly rename for unlink, newly implemented | |
12 | * silly rename for nfs_rename() following the suggestions | |
13 | * of Olaf Kirch (okir) found in this file. | |
14 | * Following Linus comments on my original hack, this version | |
15 | * depends only on the dcache stuff and doesn't touch the inode | |
16 | * layer (iput() and friends). | |
17 | * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/errno.h> | |
22 | #include <linux/stat.h> | |
23 | #include <linux/fcntl.h> | |
24 | #include <linux/string.h> | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/slab.h> | |
27 | #include <linux/mm.h> | |
28 | #include <linux/sunrpc/clnt.h> | |
29 | #include <linux/nfs_fs.h> | |
30 | #include <linux/nfs_mount.h> | |
31 | #include <linux/pagemap.h> | |
873101b3 | 32 | #include <linux/pagevec.h> |
1da177e4 | 33 | #include <linux/namei.h> |
54ceac45 | 34 | #include <linux/mount.h> |
e8edc6e0 | 35 | #include <linux/sched.h> |
56e4ebf8 | 36 | #include <linux/vmalloc.h> |
04e4bd1c | 37 | #include <linux/kmemleak.h> |
1da177e4 LT |
38 | |
39 | #include "delegation.h" | |
91d5b470 | 40 | #include "iostat.h" |
4c30d56e | 41 | #include "internal.h" |
cd9a1c0e | 42 | #include "fscache.h" |
1da177e4 | 43 | |
1da177e4 LT |
44 | /* #define NFS_DEBUG_VERBOSE 1 */ |
45 | ||
46 | static int nfs_opendir(struct inode *, struct file *); | |
47 | static int nfs_readdir(struct file *, void *, filldir_t); | |
48 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
49 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
50 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
51 | static int nfs_rmdir(struct inode *, struct dentry *); | |
52 | static int nfs_unlink(struct inode *, struct dentry *); | |
53 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
54 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
55 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
56 | static int nfs_rename(struct inode *, struct dentry *, | |
57 | struct inode *, struct dentry *); | |
7ea80859 | 58 | static int nfs_fsync_dir(struct file *, int); |
f0dd2136 | 59 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
d1bacf9e | 60 | static int nfs_readdir_clear_array(struct page*, gfp_t); |
1da177e4 | 61 | |
4b6f5d20 | 62 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 63 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
64 | .read = generic_read_dir, |
65 | .readdir = nfs_readdir, | |
66 | .open = nfs_opendir, | |
67 | .release = nfs_release, | |
68 | .fsync = nfs_fsync_dir, | |
69 | }; | |
70 | ||
92e1d5be | 71 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
72 | .create = nfs_create, |
73 | .lookup = nfs_lookup, | |
74 | .link = nfs_link, | |
75 | .unlink = nfs_unlink, | |
76 | .symlink = nfs_symlink, | |
77 | .mkdir = nfs_mkdir, | |
78 | .rmdir = nfs_rmdir, | |
79 | .mknod = nfs_mknod, | |
80 | .rename = nfs_rename, | |
81 | .permission = nfs_permission, | |
82 | .getattr = nfs_getattr, | |
83 | .setattr = nfs_setattr, | |
84 | }; | |
85 | ||
d1bacf9e BS |
86 | const struct address_space_operations nfs_dir_addr_space_ops = { |
87 | .releasepage = nfs_readdir_clear_array, | |
88 | }; | |
89 | ||
b7fa0554 | 90 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 91 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
92 | .create = nfs_create, |
93 | .lookup = nfs_lookup, | |
94 | .link = nfs_link, | |
95 | .unlink = nfs_unlink, | |
96 | .symlink = nfs_symlink, | |
97 | .mkdir = nfs_mkdir, | |
98 | .rmdir = nfs_rmdir, | |
99 | .mknod = nfs_mknod, | |
100 | .rename = nfs_rename, | |
101 | .permission = nfs_permission, | |
102 | .getattr = nfs_getattr, | |
103 | .setattr = nfs_setattr, | |
104 | .listxattr = nfs3_listxattr, | |
105 | .getxattr = nfs3_getxattr, | |
106 | .setxattr = nfs3_setxattr, | |
107 | .removexattr = nfs3_removexattr, | |
108 | }; | |
109 | #endif /* CONFIG_NFS_V3 */ | |
110 | ||
1da177e4 LT |
111 | #ifdef CONFIG_NFS_V4 |
112 | ||
113 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
c0204fd2 | 114 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd); |
92e1d5be | 115 | const struct inode_operations nfs4_dir_inode_operations = { |
c0204fd2 | 116 | .create = nfs_open_create, |
1da177e4 LT |
117 | .lookup = nfs_atomic_lookup, |
118 | .link = nfs_link, | |
119 | .unlink = nfs_unlink, | |
120 | .symlink = nfs_symlink, | |
121 | .mkdir = nfs_mkdir, | |
122 | .rmdir = nfs_rmdir, | |
123 | .mknod = nfs_mknod, | |
124 | .rename = nfs_rename, | |
125 | .permission = nfs_permission, | |
126 | .getattr = nfs_getattr, | |
127 | .setattr = nfs_setattr, | |
6b3b5496 BF |
128 | .getxattr = nfs4_getxattr, |
129 | .setxattr = nfs4_setxattr, | |
130 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
131 | }; |
132 | ||
133 | #endif /* CONFIG_NFS_V4 */ | |
134 | ||
135 | /* | |
136 | * Open file | |
137 | */ | |
138 | static int | |
139 | nfs_opendir(struct inode *inode, struct file *filp) | |
140 | { | |
7451c4f0 | 141 | int res; |
1da177e4 | 142 | |
6da24bc9 | 143 | dfprintk(FILE, "NFS: open dir(%s/%s)\n", |
cc0dd2d1 CL |
144 | filp->f_path.dentry->d_parent->d_name.name, |
145 | filp->f_path.dentry->d_name.name); | |
146 | ||
147 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); | |
1e7cb3dc | 148 | |
1da177e4 | 149 | /* Call generic open code in order to cache credentials */ |
7451c4f0 | 150 | res = nfs_open(inode, filp); |
f5a73672 NB |
151 | if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) { |
152 | /* This is a mountpoint, so d_revalidate will never | |
153 | * have been called, so we need to refresh the | |
154 | * inode (for close-open consistency) ourselves. | |
155 | */ | |
156 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
157 | } | |
1da177e4 LT |
158 | return res; |
159 | } | |
160 | ||
d1bacf9e BS |
161 | struct nfs_cache_array_entry { |
162 | u64 cookie; | |
163 | u64 ino; | |
164 | struct qstr string; | |
0b26a0bf | 165 | unsigned char d_type; |
d1bacf9e BS |
166 | }; |
167 | ||
168 | struct nfs_cache_array { | |
169 | unsigned int size; | |
170 | int eof_index; | |
171 | u64 last_cookie; | |
172 | struct nfs_cache_array_entry array[0]; | |
173 | }; | |
174 | ||
82f2e547 | 175 | typedef __be32 * (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, struct nfs_server *, int); |
1da177e4 LT |
176 | typedef struct { |
177 | struct file *file; | |
178 | struct page *page; | |
179 | unsigned long page_index; | |
f0dd2136 | 180 | u64 *dir_cookie; |
0aded708 | 181 | u64 last_cookie; |
f0dd2136 | 182 | loff_t current_index; |
1da177e4 | 183 | decode_dirent_t decode; |
d1bacf9e | 184 | |
1f4eab7e | 185 | unsigned long timestamp; |
4704f0e2 | 186 | unsigned long gencount; |
d1bacf9e BS |
187 | unsigned int cache_entry_index; |
188 | unsigned int plus:1; | |
189 | unsigned int eof:1; | |
1da177e4 LT |
190 | } nfs_readdir_descriptor_t; |
191 | ||
d1bacf9e BS |
192 | /* |
193 | * The caller is responsible for calling nfs_readdir_release_array(page) | |
1da177e4 LT |
194 | */ |
195 | static | |
d1bacf9e BS |
196 | struct nfs_cache_array *nfs_readdir_get_array(struct page *page) |
197 | { | |
8cd51a0c | 198 | void *ptr; |
d1bacf9e BS |
199 | if (page == NULL) |
200 | return ERR_PTR(-EIO); | |
8cd51a0c TM |
201 | ptr = kmap(page); |
202 | if (ptr == NULL) | |
203 | return ERR_PTR(-ENOMEM); | |
204 | return ptr; | |
d1bacf9e BS |
205 | } |
206 | ||
207 | static | |
208 | void nfs_readdir_release_array(struct page *page) | |
209 | { | |
210 | kunmap(page); | |
211 | } | |
212 | ||
213 | /* | |
214 | * we are freeing strings created by nfs_add_to_readdir_array() | |
215 | */ | |
216 | static | |
217 | int nfs_readdir_clear_array(struct page *page, gfp_t mask) | |
218 | { | |
219 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
220 | int i; | |
8cd51a0c TM |
221 | |
222 | if (IS_ERR(array)) | |
223 | return PTR_ERR(array); | |
d1bacf9e BS |
224 | for (i = 0; i < array->size; i++) |
225 | kfree(array->array[i].string.name); | |
226 | nfs_readdir_release_array(page); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | /* | |
231 | * the caller is responsible for freeing qstr.name | |
232 | * when called by nfs_readdir_add_to_array, the strings will be freed in | |
233 | * nfs_clear_readdir_array() | |
234 | */ | |
235 | static | |
4a201d6e | 236 | int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len) |
d1bacf9e BS |
237 | { |
238 | string->len = len; | |
239 | string->name = kmemdup(name, len, GFP_KERNEL); | |
4a201d6e TM |
240 | if (string->name == NULL) |
241 | return -ENOMEM; | |
04e4bd1c CM |
242 | /* |
243 | * Avoid a kmemleak false positive. The pointer to the name is stored | |
244 | * in a page cache page which kmemleak does not scan. | |
245 | */ | |
246 | kmemleak_not_leak(string->name); | |
4a201d6e TM |
247 | string->hash = full_name_hash(name, len); |
248 | return 0; | |
d1bacf9e BS |
249 | } |
250 | ||
251 | static | |
252 | int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page) | |
253 | { | |
254 | struct nfs_cache_array *array = nfs_readdir_get_array(page); | |
4a201d6e TM |
255 | struct nfs_cache_array_entry *cache_entry; |
256 | int ret; | |
257 | ||
d1bacf9e BS |
258 | if (IS_ERR(array)) |
259 | return PTR_ERR(array); | |
3020093f TM |
260 | |
261 | cache_entry = &array->array[array->size]; | |
262 | ||
263 | /* Check that this entry lies within the page bounds */ | |
8cd51a0c | 264 | ret = -ENOSPC; |
3020093f | 265 | if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE) |
4a201d6e | 266 | goto out; |
d1bacf9e | 267 | |
4a201d6e TM |
268 | cache_entry->cookie = entry->prev_cookie; |
269 | cache_entry->ino = entry->ino; | |
0b26a0bf | 270 | cache_entry->d_type = entry->d_type; |
4a201d6e TM |
271 | ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len); |
272 | if (ret) | |
273 | goto out; | |
d1bacf9e | 274 | array->last_cookie = entry->cookie; |
8cd51a0c | 275 | array->size++; |
d1bacf9e BS |
276 | if (entry->eof == 1) |
277 | array->eof_index = array->size; | |
4a201d6e | 278 | out: |
d1bacf9e | 279 | nfs_readdir_release_array(page); |
4a201d6e | 280 | return ret; |
d1bacf9e BS |
281 | } |
282 | ||
283 | static | |
284 | int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
285 | { | |
286 | loff_t diff = desc->file->f_pos - desc->current_index; | |
287 | unsigned int index; | |
288 | ||
289 | if (diff < 0) | |
290 | goto out_eof; | |
291 | if (diff >= array->size) { | |
8cd51a0c | 292 | if (array->eof_index >= 0) |
d1bacf9e BS |
293 | goto out_eof; |
294 | desc->current_index += array->size; | |
295 | return -EAGAIN; | |
296 | } | |
297 | ||
298 | index = (unsigned int)diff; | |
299 | *desc->dir_cookie = array->array[index].cookie; | |
300 | desc->cache_entry_index = index; | |
d1bacf9e BS |
301 | return 0; |
302 | out_eof: | |
303 | desc->eof = 1; | |
304 | return -EBADCOOKIE; | |
305 | } | |
306 | ||
307 | static | |
308 | int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc) | |
309 | { | |
310 | int i; | |
311 | int status = -EAGAIN; | |
312 | ||
313 | for (i = 0; i < array->size; i++) { | |
d1bacf9e BS |
314 | if (array->array[i].cookie == *desc->dir_cookie) { |
315 | desc->cache_entry_index = i; | |
316 | status = 0; | |
8cd51a0c | 317 | goto out; |
d1bacf9e BS |
318 | } |
319 | } | |
8cd51a0c TM |
320 | if (i == array->eof_index) { |
321 | desc->eof = 1; | |
322 | status = -EBADCOOKIE; | |
323 | } | |
324 | out: | |
d1bacf9e BS |
325 | return status; |
326 | } | |
327 | ||
328 | static | |
329 | int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc) | |
330 | { | |
331 | struct nfs_cache_array *array; | |
332 | int status = -EBADCOOKIE; | |
333 | ||
334 | if (desc->dir_cookie == NULL) | |
335 | goto out; | |
336 | ||
337 | array = nfs_readdir_get_array(desc->page); | |
338 | if (IS_ERR(array)) { | |
339 | status = PTR_ERR(array); | |
340 | goto out; | |
341 | } | |
342 | ||
343 | if (*desc->dir_cookie == 0) | |
344 | status = nfs_readdir_search_for_pos(array, desc); | |
345 | else | |
346 | status = nfs_readdir_search_for_cookie(array, desc); | |
347 | ||
0aded708 TM |
348 | if (status == -EAGAIN) |
349 | desc->last_cookie = array->last_cookie; | |
d1bacf9e BS |
350 | nfs_readdir_release_array(desc->page); |
351 | out: | |
352 | return status; | |
353 | } | |
354 | ||
355 | /* Fill a page with xdr information before transferring to the cache page */ | |
356 | static | |
56e4ebf8 | 357 | int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc, |
d1bacf9e | 358 | struct nfs_entry *entry, struct file *file, struct inode *inode) |
1da177e4 | 359 | { |
1da177e4 | 360 | struct rpc_cred *cred = nfs_file_cred(file); |
4704f0e2 | 361 | unsigned long timestamp, gencount; |
1da177e4 LT |
362 | int error; |
363 | ||
1da177e4 LT |
364 | again: |
365 | timestamp = jiffies; | |
4704f0e2 | 366 | gencount = nfs_inc_attr_generation_counter(); |
56e4ebf8 | 367 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages, |
1da177e4 LT |
368 | NFS_SERVER(inode)->dtsize, desc->plus); |
369 | if (error < 0) { | |
370 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
371 | if (error == -ENOTSUPP && desc->plus) { | |
372 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
3a10c30a | 373 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 LT |
374 | desc->plus = 0; |
375 | goto again; | |
376 | } | |
377 | goto error; | |
378 | } | |
1f4eab7e | 379 | desc->timestamp = timestamp; |
4704f0e2 | 380 | desc->gencount = gencount; |
d1bacf9e BS |
381 | error: |
382 | return error; | |
1da177e4 LT |
383 | } |
384 | ||
d1bacf9e BS |
385 | /* Fill in an entry based on the xdr code stored in desc->page */ |
386 | static | |
babddc72 | 387 | int xdr_decode(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, struct xdr_stream *stream) |
1da177e4 | 388 | { |
82f2e547 | 389 | __be32 *p = desc->decode(stream, entry, NFS_SERVER(desc->file->f_path.dentry->d_inode), desc->plus); |
1da177e4 LT |
390 | if (IS_ERR(p)) |
391 | return PTR_ERR(p); | |
1da177e4 | 392 | |
d1bacf9e BS |
393 | entry->fattr->time_start = desc->timestamp; |
394 | entry->fattr->gencount = desc->gencount; | |
395 | return 0; | |
1da177e4 LT |
396 | } |
397 | ||
d39ab9de BS |
398 | static |
399 | int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry) | |
400 | { | |
d39ab9de BS |
401 | if (dentry->d_inode == NULL) |
402 | goto different; | |
37a09f07 | 403 | if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0) |
d39ab9de BS |
404 | goto different; |
405 | return 1; | |
406 | different: | |
407 | return 0; | |
408 | } | |
409 | ||
410 | static | |
411 | void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry) | |
412 | { | |
4a201d6e TM |
413 | struct qstr filename = { |
414 | .len = entry->len, | |
415 | .name = entry->name, | |
416 | }; | |
417 | struct dentry *dentry; | |
418 | struct dentry *alias; | |
d39ab9de BS |
419 | struct inode *dir = parent->d_inode; |
420 | struct inode *inode; | |
421 | ||
4a201d6e TM |
422 | if (filename.name[0] == '.') { |
423 | if (filename.len == 1) | |
424 | return; | |
425 | if (filename.len == 2 && filename.name[1] == '.') | |
426 | return; | |
427 | } | |
428 | filename.hash = full_name_hash(filename.name, filename.len); | |
d39ab9de | 429 | |
4a201d6e | 430 | dentry = d_lookup(parent, &filename); |
d39ab9de BS |
431 | if (dentry != NULL) { |
432 | if (nfs_same_file(dentry, entry)) { | |
433 | nfs_refresh_inode(dentry->d_inode, entry->fattr); | |
434 | goto out; | |
435 | } else { | |
436 | d_drop(dentry); | |
437 | dput(dentry); | |
438 | } | |
439 | } | |
440 | ||
441 | dentry = d_alloc(parent, &filename); | |
4a201d6e TM |
442 | if (dentry == NULL) |
443 | return; | |
444 | ||
d39ab9de BS |
445 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
446 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
447 | if (IS_ERR(inode)) | |
448 | goto out; | |
449 | ||
450 | alias = d_materialise_unique(dentry, inode); | |
451 | if (IS_ERR(alias)) | |
452 | goto out; | |
453 | else if (alias) { | |
454 | nfs_set_verifier(alias, nfs_save_change_attribute(dir)); | |
455 | dput(alias); | |
456 | } else | |
457 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
458 | ||
459 | out: | |
460 | dput(dentry); | |
d39ab9de BS |
461 | } |
462 | ||
d1bacf9e BS |
463 | /* Perform conversion from xdr to cache array */ |
464 | static | |
8cd51a0c | 465 | int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry, |
56e4ebf8 | 466 | void *xdr_page, struct page *page, unsigned int buflen) |
1da177e4 | 467 | { |
babddc72 BS |
468 | struct xdr_stream stream; |
469 | struct xdr_buf buf; | |
56e4ebf8 | 470 | __be32 *ptr = xdr_page; |
99424380 | 471 | struct nfs_cache_array *array; |
5c346854 TM |
472 | unsigned int count = 0; |
473 | int status; | |
babddc72 BS |
474 | |
475 | buf.head->iov_base = xdr_page; | |
476 | buf.head->iov_len = buflen; | |
477 | buf.tail->iov_len = 0; | |
478 | buf.page_base = 0; | |
479 | buf.page_len = 0; | |
480 | buf.buflen = buf.head->iov_len; | |
481 | buf.len = buf.head->iov_len; | |
482 | ||
483 | xdr_init_decode(&stream, &buf, ptr); | |
484 | ||
99424380 BS |
485 | |
486 | do { | |
487 | status = xdr_decode(desc, entry, &stream); | |
8cd51a0c TM |
488 | if (status != 0) { |
489 | if (status == -EAGAIN) | |
490 | status = 0; | |
99424380 | 491 | break; |
8cd51a0c | 492 | } |
99424380 | 493 | |
5c346854 TM |
494 | count++; |
495 | ||
d39ab9de BS |
496 | if (desc->plus == 1) |
497 | nfs_prime_dcache(desc->file->f_path.dentry, entry); | |
8cd51a0c TM |
498 | |
499 | status = nfs_readdir_add_to_array(entry, page); | |
500 | if (status != 0) | |
501 | break; | |
99424380 BS |
502 | } while (!entry->eof); |
503 | ||
5c346854 | 504 | if (count == 0 || (status == -EBADCOOKIE && entry->eof == 1)) { |
99424380 | 505 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
506 | if (!IS_ERR(array)) { |
507 | array->eof_index = array->size; | |
508 | status = 0; | |
509 | nfs_readdir_release_array(page); | |
5c346854 TM |
510 | } else |
511 | status = PTR_ERR(array); | |
1da177e4 | 512 | } |
8cd51a0c | 513 | return status; |
56e4ebf8 BS |
514 | } |
515 | ||
516 | static | |
517 | void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages) | |
518 | { | |
519 | unsigned int i; | |
520 | for (i = 0; i < npages; i++) | |
521 | put_page(pages[i]); | |
522 | } | |
523 | ||
524 | static | |
525 | void nfs_readdir_free_large_page(void *ptr, struct page **pages, | |
526 | unsigned int npages) | |
527 | { | |
528 | vm_unmap_ram(ptr, npages); | |
529 | nfs_readdir_free_pagearray(pages, npages); | |
530 | } | |
531 | ||
532 | /* | |
533 | * nfs_readdir_large_page will allocate pages that must be freed with a call | |
534 | * to nfs_readdir_free_large_page | |
535 | */ | |
536 | static | |
537 | void *nfs_readdir_large_page(struct page **pages, unsigned int npages) | |
538 | { | |
539 | void *ptr; | |
540 | unsigned int i; | |
541 | ||
542 | for (i = 0; i < npages; i++) { | |
543 | struct page *page = alloc_page(GFP_KERNEL); | |
544 | if (page == NULL) | |
545 | goto out_freepages; | |
546 | pages[i] = page; | |
547 | } | |
548 | ||
4a201d6e | 549 | ptr = vm_map_ram(pages, npages, 0, PAGE_KERNEL); |
56e4ebf8 BS |
550 | if (!IS_ERR_OR_NULL(ptr)) |
551 | return ptr; | |
552 | out_freepages: | |
553 | nfs_readdir_free_pagearray(pages, i); | |
554 | return NULL; | |
1da177e4 LT |
555 | } |
556 | ||
d1bacf9e BS |
557 | static |
558 | int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode) | |
00a92642 | 559 | { |
56e4ebf8 BS |
560 | struct page *pages[NFS_MAX_READDIR_PAGES]; |
561 | void *pages_ptr = NULL; | |
d1bacf9e BS |
562 | struct nfs_entry entry; |
563 | struct file *file = desc->file; | |
564 | struct nfs_cache_array *array; | |
8cd51a0c | 565 | int status = -ENOMEM; |
56e4ebf8 | 566 | unsigned int array_size = ARRAY_SIZE(pages); |
d1bacf9e BS |
567 | |
568 | entry.prev_cookie = 0; | |
0aded708 | 569 | entry.cookie = desc->last_cookie; |
d1bacf9e BS |
570 | entry.eof = 0; |
571 | entry.fh = nfs_alloc_fhandle(); | |
572 | entry.fattr = nfs_alloc_fattr(); | |
573 | if (entry.fh == NULL || entry.fattr == NULL) | |
574 | goto out; | |
00a92642 | 575 | |
d1bacf9e | 576 | array = nfs_readdir_get_array(page); |
8cd51a0c TM |
577 | if (IS_ERR(array)) { |
578 | status = PTR_ERR(array); | |
579 | goto out; | |
580 | } | |
d1bacf9e BS |
581 | memset(array, 0, sizeof(struct nfs_cache_array)); |
582 | array->eof_index = -1; | |
00a92642 | 583 | |
56e4ebf8 BS |
584 | pages_ptr = nfs_readdir_large_page(pages, array_size); |
585 | if (!pages_ptr) | |
d1bacf9e BS |
586 | goto out_release_array; |
587 | do { | |
ac396128 | 588 | unsigned int pglen; |
56e4ebf8 | 589 | status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode); |
babddc72 | 590 | |
d1bacf9e | 591 | if (status < 0) |
00a92642 | 592 | break; |
ac396128 TM |
593 | pglen = status; |
594 | status = nfs_readdir_page_filler(desc, &entry, pages_ptr, page, pglen); | |
8cd51a0c TM |
595 | if (status < 0) { |
596 | if (status == -ENOSPC) | |
597 | status = 0; | |
598 | break; | |
599 | } | |
600 | } while (array->eof_index < 0); | |
d1bacf9e | 601 | |
56e4ebf8 | 602 | nfs_readdir_free_large_page(pages_ptr, pages, array_size); |
d1bacf9e BS |
603 | out_release_array: |
604 | nfs_readdir_release_array(page); | |
605 | out: | |
606 | nfs_free_fattr(entry.fattr); | |
607 | nfs_free_fhandle(entry.fh); | |
00a92642 OG |
608 | return status; |
609 | } | |
610 | ||
611 | /* | |
d1bacf9e BS |
612 | * Now we cache directories properly, by converting xdr information |
613 | * to an array that can be used for lookups later. This results in | |
614 | * fewer cache pages, since we can store more information on each page. | |
615 | * We only need to convert from xdr once so future lookups are much simpler | |
1da177e4 | 616 | */ |
d1bacf9e BS |
617 | static |
618 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page) | |
1da177e4 | 619 | { |
01cce933 | 620 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
8cd51a0c | 621 | int ret; |
1da177e4 | 622 | |
8cd51a0c TM |
623 | ret = nfs_readdir_xdr_to_array(desc, page, inode); |
624 | if (ret < 0) | |
d1bacf9e BS |
625 | goto error; |
626 | SetPageUptodate(page); | |
1da177e4 | 627 | |
d1bacf9e BS |
628 | if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) { |
629 | /* Should never happen */ | |
630 | nfs_zap_mapping(inode, inode->i_mapping); | |
1da177e4 | 631 | } |
d1bacf9e BS |
632 | unlock_page(page); |
633 | return 0; | |
634 | error: | |
635 | unlock_page(page); | |
8cd51a0c | 636 | return ret; |
d1bacf9e | 637 | } |
1da177e4 | 638 | |
d1bacf9e BS |
639 | static |
640 | void cache_page_release(nfs_readdir_descriptor_t *desc) | |
641 | { | |
642 | page_cache_release(desc->page); | |
643 | desc->page = NULL; | |
644 | } | |
645 | ||
646 | static | |
647 | struct page *get_cache_page(nfs_readdir_descriptor_t *desc) | |
648 | { | |
8cd51a0c | 649 | return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping, |
d1bacf9e | 650 | desc->page_index, (filler_t *)nfs_readdir_filler, desc); |
1da177e4 LT |
651 | } |
652 | ||
653 | /* | |
d1bacf9e | 654 | * Returns 0 if desc->dir_cookie was found on page desc->page_index |
1da177e4 | 655 | */ |
d1bacf9e BS |
656 | static |
657 | int find_cache_page(nfs_readdir_descriptor_t *desc) | |
658 | { | |
659 | int res; | |
660 | ||
661 | desc->page = get_cache_page(desc); | |
662 | if (IS_ERR(desc->page)) | |
663 | return PTR_ERR(desc->page); | |
664 | ||
665 | res = nfs_readdir_search_array(desc); | |
666 | if (res == 0) | |
667 | return 0; | |
668 | cache_page_release(desc); | |
669 | return res; | |
670 | } | |
671 | ||
672 | /* Search for desc->dir_cookie from the beginning of the page cache */ | |
1da177e4 LT |
673 | static inline |
674 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
675 | { | |
8cd51a0c | 676 | int res; |
d1bacf9e | 677 | |
0aded708 | 678 | if (desc->page_index == 0) { |
8cd51a0c | 679 | desc->current_index = 0; |
0aded708 TM |
680 | desc->last_cookie = 0; |
681 | } | |
d1bacf9e BS |
682 | while (1) { |
683 | res = find_cache_page(desc); | |
1da177e4 LT |
684 | if (res != -EAGAIN) |
685 | break; | |
d1bacf9e | 686 | desc->page_index++; |
1da177e4 | 687 | } |
1da177e4 LT |
688 | return res; |
689 | } | |
690 | ||
691 | static inline unsigned int dt_type(struct inode *inode) | |
692 | { | |
693 | return (inode->i_mode >> 12) & 15; | |
694 | } | |
695 | ||
1da177e4 LT |
696 | /* |
697 | * Once we've found the start of the dirent within a page: fill 'er up... | |
698 | */ | |
699 | static | |
700 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
701 | filldir_t filldir) | |
702 | { | |
703 | struct file *file = desc->file; | |
d1bacf9e BS |
704 | int i = 0; |
705 | int res = 0; | |
706 | struct nfs_cache_array *array = NULL; | |
d1bacf9e BS |
707 | |
708 | array = nfs_readdir_get_array(desc->page); | |
e7c58e97 TM |
709 | if (IS_ERR(array)) { |
710 | res = PTR_ERR(array); | |
711 | goto out; | |
712 | } | |
d1bacf9e BS |
713 | |
714 | for (i = desc->cache_entry_index; i < array->size; i++) { | |
ece0b423 | 715 | struct nfs_cache_array_entry *ent; |
1da177e4 | 716 | |
ece0b423 TM |
717 | ent = &array->array[i]; |
718 | if (filldir(dirent, ent->string.name, ent->string.len, | |
0b26a0bf TM |
719 | file->f_pos, nfs_compat_user_ino64(ent->ino), |
720 | ent->d_type) < 0) { | |
ece0b423 | 721 | desc->eof = 1; |
1da177e4 | 722 | break; |
ece0b423 | 723 | } |
00a92642 | 724 | file->f_pos++; |
d1bacf9e BS |
725 | desc->cache_entry_index = i; |
726 | if (i < (array->size-1)) | |
727 | *desc->dir_cookie = array->array[i+1].cookie; | |
728 | else | |
729 | *desc->dir_cookie = array->last_cookie; | |
1da177e4 | 730 | } |
8cd51a0c TM |
731 | if (i == array->eof_index) |
732 | desc->eof = 1; | |
d1bacf9e BS |
733 | |
734 | nfs_readdir_release_array(desc->page); | |
e7c58e97 | 735 | out: |
d1bacf9e | 736 | cache_page_release(desc); |
1e7cb3dc CL |
737 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
738 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
739 | return res; |
740 | } | |
741 | ||
742 | /* | |
743 | * If we cannot find a cookie in our cache, we suspect that this is | |
744 | * because it points to a deleted file, so we ask the server to return | |
745 | * whatever it thinks is the next entry. We then feed this to filldir. | |
746 | * If all goes well, we should then be able to find our way round the | |
747 | * cache on the next call to readdir_search_pagecache(); | |
748 | * | |
749 | * NOTE: we cannot add the anonymous page to the pagecache because | |
750 | * the data it contains might not be page aligned. Besides, | |
751 | * we should already have a complete representation of the | |
752 | * directory in the page cache by the time we get here. | |
753 | */ | |
754 | static inline | |
755 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
756 | filldir_t filldir) | |
757 | { | |
1da177e4 LT |
758 | struct page *page = NULL; |
759 | int status; | |
d1bacf9e | 760 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 | 761 | |
1e7cb3dc CL |
762 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
763 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
764 | |
765 | page = alloc_page(GFP_HIGHUSER); | |
766 | if (!page) { | |
767 | status = -ENOMEM; | |
768 | goto out; | |
769 | } | |
d1bacf9e | 770 | |
7a8e1dc3 | 771 | desc->page_index = 0; |
0aded708 | 772 | desc->last_cookie = *desc->dir_cookie; |
7a8e1dc3 TM |
773 | desc->page = page; |
774 | ||
85f8607e TM |
775 | status = nfs_readdir_xdr_to_array(desc, page, inode); |
776 | if (status < 0) | |
1da177e4 LT |
777 | goto out_release; |
778 | ||
779 | status = nfs_do_filldir(desc, dirent, filldir); | |
780 | ||
1da177e4 | 781 | out: |
1e7cb3dc | 782 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
3110ff80 | 783 | __func__, status); |
1da177e4 LT |
784 | return status; |
785 | out_release: | |
d1bacf9e | 786 | cache_page_release(desc); |
1da177e4 LT |
787 | goto out; |
788 | } | |
789 | ||
00a92642 OG |
790 | /* The file offset position represents the dirent entry number. A |
791 | last cookie cache takes care of the common case of reading the | |
792 | whole directory. | |
1da177e4 LT |
793 | */ |
794 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
795 | { | |
01cce933 | 796 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
797 | struct inode *inode = dentry->d_inode; |
798 | nfs_readdir_descriptor_t my_desc, | |
799 | *desc = &my_desc; | |
aa49b4cf | 800 | int res = -ENOMEM; |
1da177e4 | 801 | |
6da24bc9 | 802 | dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n", |
1e7cb3dc CL |
803 | dentry->d_parent->d_name.name, dentry->d_name.name, |
804 | (long long)filp->f_pos); | |
91d5b470 CL |
805 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
806 | ||
1da177e4 | 807 | /* |
00a92642 | 808 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 809 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
810 | * to either find the entry with the appropriate number or |
811 | * revalidate the cookie. | |
1da177e4 LT |
812 | */ |
813 | memset(desc, 0, sizeof(*desc)); | |
814 | ||
815 | desc->file = filp; | |
cd3758e3 | 816 | desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie; |
1da177e4 LT |
817 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
818 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
819 | ||
565277f6 | 820 | nfs_block_sillyrename(dentry); |
1cda707d | 821 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
fccca7fc TM |
822 | if (res < 0) |
823 | goto out; | |
824 | ||
d1bacf9e | 825 | while (desc->eof != 1) { |
1da177e4 | 826 | res = readdir_search_pagecache(desc); |
00a92642 | 827 | |
1da177e4 | 828 | if (res == -EBADCOOKIE) { |
ece0b423 | 829 | res = 0; |
1da177e4 | 830 | /* This means either end of directory */ |
d1bacf9e | 831 | if (*desc->dir_cookie && desc->eof == 0) { |
1da177e4 LT |
832 | /* Or that the server has 'lost' a cookie */ |
833 | res = uncached_readdir(desc, dirent, filldir); | |
ece0b423 | 834 | if (res == 0) |
1da177e4 LT |
835 | continue; |
836 | } | |
1da177e4 LT |
837 | break; |
838 | } | |
839 | if (res == -ETOOSMALL && desc->plus) { | |
3a10c30a | 840 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags); |
1da177e4 | 841 | nfs_zap_caches(inode); |
baf57a09 | 842 | desc->page_index = 0; |
1da177e4 | 843 | desc->plus = 0; |
d1bacf9e | 844 | desc->eof = 0; |
1da177e4 LT |
845 | continue; |
846 | } | |
847 | if (res < 0) | |
848 | break; | |
849 | ||
850 | res = nfs_do_filldir(desc, dirent, filldir); | |
ece0b423 | 851 | if (res < 0) |
1da177e4 | 852 | break; |
1da177e4 | 853 | } |
fccca7fc | 854 | out: |
565277f6 | 855 | nfs_unblock_sillyrename(dentry); |
1e7cb3dc CL |
856 | if (res > 0) |
857 | res = 0; | |
aa49b4cf | 858 | dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n", |
1e7cb3dc CL |
859 | dentry->d_parent->d_name.name, dentry->d_name.name, |
860 | res); | |
861 | return res; | |
1da177e4 LT |
862 | } |
863 | ||
10afec90 | 864 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 865 | { |
b84e06c5 CL |
866 | struct dentry *dentry = filp->f_path.dentry; |
867 | struct inode *inode = dentry->d_inode; | |
868 | ||
6da24bc9 | 869 | dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n", |
b84e06c5 CL |
870 | dentry->d_parent->d_name.name, |
871 | dentry->d_name.name, | |
872 | offset, origin); | |
873 | ||
874 | mutex_lock(&inode->i_mutex); | |
f0dd2136 TM |
875 | switch (origin) { |
876 | case 1: | |
877 | offset += filp->f_pos; | |
878 | case 0: | |
879 | if (offset >= 0) | |
880 | break; | |
881 | default: | |
882 | offset = -EINVAL; | |
883 | goto out; | |
884 | } | |
885 | if (offset != filp->f_pos) { | |
886 | filp->f_pos = offset; | |
cd3758e3 | 887 | nfs_file_open_context(filp)->dir_cookie = 0; |
f0dd2136 TM |
888 | } |
889 | out: | |
b84e06c5 | 890 | mutex_unlock(&inode->i_mutex); |
f0dd2136 TM |
891 | return offset; |
892 | } | |
893 | ||
1da177e4 LT |
894 | /* |
895 | * All directory operations under NFS are synchronous, so fsync() | |
896 | * is a dummy operation. | |
897 | */ | |
7ea80859 | 898 | static int nfs_fsync_dir(struct file *filp, int datasync) |
1da177e4 | 899 | { |
7ea80859 CH |
900 | struct dentry *dentry = filp->f_path.dentry; |
901 | ||
6da24bc9 | 902 | dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n", |
1e7cb3dc CL |
903 | dentry->d_parent->d_name.name, dentry->d_name.name, |
904 | datasync); | |
905 | ||
54917786 | 906 | nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC); |
1da177e4 LT |
907 | return 0; |
908 | } | |
909 | ||
bfc69a45 TM |
910 | /** |
911 | * nfs_force_lookup_revalidate - Mark the directory as having changed | |
912 | * @dir - pointer to directory inode | |
913 | * | |
914 | * This forces the revalidation code in nfs_lookup_revalidate() to do a | |
915 | * full lookup on all child dentries of 'dir' whenever a change occurs | |
916 | * on the server that might have invalidated our dcache. | |
917 | * | |
918 | * The caller should be holding dir->i_lock | |
919 | */ | |
920 | void nfs_force_lookup_revalidate(struct inode *dir) | |
921 | { | |
011935a0 | 922 | NFS_I(dir)->cache_change_attribute++; |
bfc69a45 TM |
923 | } |
924 | ||
1da177e4 LT |
925 | /* |
926 | * A check for whether or not the parent directory has changed. | |
927 | * In the case it has, we assume that the dentries are untrustworthy | |
928 | * and may need to be looked up again. | |
929 | */ | |
c79ba787 | 930 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 LT |
931 | { |
932 | if (IS_ROOT(dentry)) | |
933 | return 1; | |
4eec952e TM |
934 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE) |
935 | return 0; | |
f2c77f4e TM |
936 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) |
937 | return 0; | |
938 | /* Revalidate nfsi->cache_change_attribute before we declare a match */ | |
939 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
940 | return 0; | |
941 | if (!nfs_verify_change_attribute(dir, dentry->d_time)) | |
942 | return 0; | |
943 | return 1; | |
1da177e4 LT |
944 | } |
945 | ||
1d6757fb TM |
946 | /* |
947 | * Return the intent data that applies to this particular path component | |
948 | * | |
949 | * Note that the current set of intents only apply to the very last | |
950 | * component of the path. | |
951 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
952 | */ | |
953 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
954 | { | |
955 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
956 | return 0; | |
957 | return nd->flags & mask; | |
958 | } | |
959 | ||
a12802ca TM |
960 | /* |
961 | * Use intent information to check whether or not we're going to do | |
962 | * an O_EXCL create using this path component. | |
963 | */ | |
964 | static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
965 | { | |
966 | if (NFS_PROTO(dir)->version == 2) | |
967 | return 0; | |
3516586a | 968 | return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL); |
a12802ca TM |
969 | } |
970 | ||
1d6757fb TM |
971 | /* |
972 | * Inode and filehandle revalidation for lookups. | |
973 | * | |
974 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
975 | * or if the intent information indicates that we're about to open this | |
976 | * particular file and the "nocto" mount flag is not set. | |
977 | * | |
978 | */ | |
1da177e4 LT |
979 | static inline |
980 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
981 | { | |
982 | struct nfs_server *server = NFS_SERVER(inode); | |
983 | ||
4e99a1ff TM |
984 | if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags)) |
985 | return 0; | |
1da177e4 | 986 | if (nd != NULL) { |
1da177e4 | 987 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 988 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
989 | goto out_force; |
990 | /* This is an open(2) */ | |
1d6757fb | 991 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
992 | !(server->flags & NFS_MOUNT_NOCTO) && |
993 | (S_ISREG(inode->i_mode) || | |
994 | S_ISDIR(inode->i_mode))) | |
1da177e4 | 995 | goto out_force; |
4f48af45 | 996 | return 0; |
1da177e4 LT |
997 | } |
998 | return nfs_revalidate_inode(server, inode); | |
999 | out_force: | |
1000 | return __nfs_revalidate_inode(server, inode); | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * We judge how long we want to trust negative | |
1005 | * dentries by looking at the parent inode mtime. | |
1006 | * | |
1007 | * If parent mtime has changed, we revalidate, else we wait for a | |
1008 | * period corresponding to the parent's attribute cache timeout value. | |
1009 | */ | |
1010 | static inline | |
1011 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
1012 | struct nameidata *nd) | |
1013 | { | |
1da177e4 | 1014 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 1015 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 | 1016 | return 0; |
4eec952e TM |
1017 | if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG) |
1018 | return 1; | |
1da177e4 LT |
1019 | return !nfs_check_verifier(dir, dentry); |
1020 | } | |
1021 | ||
1022 | /* | |
1023 | * This is called every time the dcache has a lookup hit, | |
1024 | * and we should check whether we can really trust that | |
1025 | * lookup. | |
1026 | * | |
1027 | * NOTE! The hit can be a negative hit too, don't assume | |
1028 | * we have an inode! | |
1029 | * | |
1030 | * If the parent directory is seen to have changed, we throw out the | |
1031 | * cached dentry and do a new lookup. | |
1032 | */ | |
1033 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
1034 | { | |
1035 | struct inode *dir; | |
1036 | struct inode *inode; | |
1037 | struct dentry *parent; | |
e1fb4d05 TM |
1038 | struct nfs_fh *fhandle = NULL; |
1039 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1040 | int error; |
1da177e4 LT |
1041 | |
1042 | parent = dget_parent(dentry); | |
1da177e4 | 1043 | dir = parent->d_inode; |
91d5b470 | 1044 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
1045 | inode = dentry->d_inode; |
1046 | ||
1047 | if (!inode) { | |
1048 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
1049 | goto out_bad; | |
1050 | goto out_valid; | |
1051 | } | |
1052 | ||
1053 | if (is_bad_inode(inode)) { | |
1e7cb3dc | 1054 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
3110ff80 | 1055 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1056 | dentry->d_name.name); |
1da177e4 LT |
1057 | goto out_bad; |
1058 | } | |
1059 | ||
15860ab1 TM |
1060 | if (nfs_have_delegation(inode, FMODE_READ)) |
1061 | goto out_set_verifier; | |
1062 | ||
1da177e4 | 1063 | /* Force a full look up iff the parent directory has changed */ |
a12802ca | 1064 | if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) { |
1da177e4 LT |
1065 | if (nfs_lookup_verify_inode(inode, nd)) |
1066 | goto out_zap_parent; | |
1067 | goto out_valid; | |
1068 | } | |
1069 | ||
1070 | if (NFS_STALE(inode)) | |
1071 | goto out_bad; | |
1072 | ||
e1fb4d05 TM |
1073 | error = -ENOMEM; |
1074 | fhandle = nfs_alloc_fhandle(); | |
1075 | fattr = nfs_alloc_fattr(); | |
1076 | if (fhandle == NULL || fattr == NULL) | |
1077 | goto out_error; | |
1078 | ||
1079 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1da177e4 LT |
1080 | if (error) |
1081 | goto out_bad; | |
e1fb4d05 | 1082 | if (nfs_compare_fh(NFS_FH(inode), fhandle)) |
1da177e4 | 1083 | goto out_bad; |
e1fb4d05 | 1084 | if ((error = nfs_refresh_inode(inode, fattr)) != 0) |
1da177e4 LT |
1085 | goto out_bad; |
1086 | ||
e1fb4d05 TM |
1087 | nfs_free_fattr(fattr); |
1088 | nfs_free_fhandle(fhandle); | |
15860ab1 | 1089 | out_set_verifier: |
cf8ba45e | 1090 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 | 1091 | out_valid: |
1da177e4 | 1092 | dput(parent); |
1e7cb3dc | 1093 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
3110ff80 | 1094 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1095 | dentry->d_name.name); |
1da177e4 LT |
1096 | return 1; |
1097 | out_zap_parent: | |
1098 | nfs_zap_caches(dir); | |
1099 | out_bad: | |
a1643a92 | 1100 | nfs_mark_for_revalidate(dir); |
1da177e4 LT |
1101 | if (inode && S_ISDIR(inode->i_mode)) { |
1102 | /* Purge readdir caches. */ | |
1103 | nfs_zap_caches(inode); | |
1104 | /* If we have submounts, don't unhash ! */ | |
1105 | if (have_submounts(dentry)) | |
1106 | goto out_valid; | |
d9e80b7d AV |
1107 | if (dentry->d_flags & DCACHE_DISCONNECTED) |
1108 | goto out_valid; | |
1da177e4 LT |
1109 | shrink_dcache_parent(dentry); |
1110 | } | |
1111 | d_drop(dentry); | |
e1fb4d05 TM |
1112 | nfs_free_fattr(fattr); |
1113 | nfs_free_fhandle(fhandle); | |
1da177e4 | 1114 | dput(parent); |
1e7cb3dc | 1115 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
3110ff80 | 1116 | __func__, dentry->d_parent->d_name.name, |
1e7cb3dc | 1117 | dentry->d_name.name); |
1da177e4 | 1118 | return 0; |
e1fb4d05 TM |
1119 | out_error: |
1120 | nfs_free_fattr(fattr); | |
1121 | nfs_free_fhandle(fhandle); | |
1122 | dput(parent); | |
1123 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n", | |
1124 | __func__, dentry->d_parent->d_name.name, | |
1125 | dentry->d_name.name, error); | |
1126 | return error; | |
1da177e4 LT |
1127 | } |
1128 | ||
1129 | /* | |
1130 | * This is called from dput() when d_count is going to 0. | |
1131 | */ | |
1132 | static int nfs_dentry_delete(struct dentry *dentry) | |
1133 | { | |
1134 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
1135 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1136 | dentry->d_flags); | |
1137 | ||
77f11192 TM |
1138 | /* Unhash any dentry with a stale inode */ |
1139 | if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode)) | |
1140 | return 1; | |
1141 | ||
1da177e4 LT |
1142 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
1143 | /* Unhash it, so that ->d_iput() would be called */ | |
1144 | return 1; | |
1145 | } | |
1146 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
1147 | /* Unhash it, so that ancestors of killed async unlink | |
1148 | * files will be cleaned up during umount */ | |
1149 | return 1; | |
1150 | } | |
1151 | return 0; | |
1152 | ||
1153 | } | |
1154 | ||
1b83d707 TM |
1155 | static void nfs_drop_nlink(struct inode *inode) |
1156 | { | |
1157 | spin_lock(&inode->i_lock); | |
1158 | if (inode->i_nlink > 0) | |
1159 | drop_nlink(inode); | |
1160 | spin_unlock(&inode->i_lock); | |
1161 | } | |
1162 | ||
1da177e4 LT |
1163 | /* |
1164 | * Called when the dentry loses inode. | |
1165 | * We use it to clean up silly-renamed files. | |
1166 | */ | |
1167 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
1168 | { | |
83672d39 NB |
1169 | if (S_ISDIR(inode->i_mode)) |
1170 | /* drop any readdir cache as it could easily be old */ | |
1171 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
1172 | ||
1da177e4 | 1173 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
9a53c3a7 | 1174 | drop_nlink(inode); |
e4eff1a6 | 1175 | nfs_complete_unlink(dentry, inode); |
1da177e4 | 1176 | } |
1da177e4 LT |
1177 | iput(inode); |
1178 | } | |
1179 | ||
f786aa90 | 1180 | const struct dentry_operations nfs_dentry_operations = { |
1da177e4 LT |
1181 | .d_revalidate = nfs_lookup_revalidate, |
1182 | .d_delete = nfs_dentry_delete, | |
1183 | .d_iput = nfs_dentry_iput, | |
1184 | }; | |
1185 | ||
1da177e4 LT |
1186 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
1187 | { | |
1188 | struct dentry *res; | |
565277f6 | 1189 | struct dentry *parent; |
1da177e4 | 1190 | struct inode *inode = NULL; |
e1fb4d05 TM |
1191 | struct nfs_fh *fhandle = NULL; |
1192 | struct nfs_fattr *fattr = NULL; | |
1da177e4 | 1193 | int error; |
1da177e4 LT |
1194 | |
1195 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
1196 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 1197 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
1198 | |
1199 | res = ERR_PTR(-ENAMETOOLONG); | |
1200 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
1201 | goto out; | |
1202 | ||
1da177e4 LT |
1203 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; |
1204 | ||
fd684071 TM |
1205 | /* |
1206 | * If we're doing an exclusive create, optimize away the lookup | |
1207 | * but don't hash the dentry. | |
1208 | */ | |
1209 | if (nfs_is_exclusive_create(dir, nd)) { | |
1210 | d_instantiate(dentry, NULL); | |
1211 | res = NULL; | |
fc0f684c | 1212 | goto out; |
fd684071 | 1213 | } |
1da177e4 | 1214 | |
e1fb4d05 TM |
1215 | res = ERR_PTR(-ENOMEM); |
1216 | fhandle = nfs_alloc_fhandle(); | |
1217 | fattr = nfs_alloc_fattr(); | |
1218 | if (fhandle == NULL || fattr == NULL) | |
1219 | goto out; | |
1220 | ||
565277f6 TM |
1221 | parent = dentry->d_parent; |
1222 | /* Protect against concurrent sillydeletes */ | |
1223 | nfs_block_sillyrename(parent); | |
e1fb4d05 | 1224 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1da177e4 LT |
1225 | if (error == -ENOENT) |
1226 | goto no_entry; | |
1227 | if (error < 0) { | |
1228 | res = ERR_PTR(error); | |
565277f6 | 1229 | goto out_unblock_sillyrename; |
1da177e4 | 1230 | } |
e1fb4d05 | 1231 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1232 | res = (struct dentry *)inode; |
1233 | if (IS_ERR(res)) | |
565277f6 | 1234 | goto out_unblock_sillyrename; |
54ceac45 | 1235 | |
1da177e4 | 1236 | no_entry: |
54ceac45 | 1237 | res = d_materialise_unique(dentry, inode); |
9eaef27b TM |
1238 | if (res != NULL) { |
1239 | if (IS_ERR(res)) | |
565277f6 | 1240 | goto out_unblock_sillyrename; |
1da177e4 | 1241 | dentry = res; |
9eaef27b | 1242 | } |
1da177e4 | 1243 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
565277f6 TM |
1244 | out_unblock_sillyrename: |
1245 | nfs_unblock_sillyrename(parent); | |
1da177e4 | 1246 | out: |
e1fb4d05 TM |
1247 | nfs_free_fattr(fattr); |
1248 | nfs_free_fhandle(fhandle); | |
1da177e4 LT |
1249 | return res; |
1250 | } | |
1251 | ||
1252 | #ifdef CONFIG_NFS_V4 | |
1253 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
1254 | ||
f786aa90 | 1255 | const struct dentry_operations nfs4_dentry_operations = { |
1da177e4 LT |
1256 | .d_revalidate = nfs_open_revalidate, |
1257 | .d_delete = nfs_dentry_delete, | |
1258 | .d_iput = nfs_dentry_iput, | |
1259 | }; | |
1260 | ||
1d6757fb TM |
1261 | /* |
1262 | * Use intent information to determine whether we need to substitute | |
1263 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
1264 | */ | |
5584c306 | 1265 | static int is_atomic_open(struct nameidata *nd) |
1da177e4 | 1266 | { |
1d6757fb | 1267 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
1268 | return 0; |
1269 | /* NFS does not (yet) have a stateful open for directories */ | |
1270 | if (nd->flags & LOOKUP_DIRECTORY) | |
1271 | return 0; | |
1272 | /* Are we trying to write to a read only partition? */ | |
2c463e95 DH |
1273 | if (__mnt_is_readonly(nd->path.mnt) && |
1274 | (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
1da177e4 LT |
1275 | return 0; |
1276 | return 1; | |
1277 | } | |
1278 | ||
cd9a1c0e TM |
1279 | static struct nfs_open_context *nameidata_to_nfs_open_context(struct dentry *dentry, struct nameidata *nd) |
1280 | { | |
1281 | struct path path = { | |
1282 | .mnt = nd->path.mnt, | |
1283 | .dentry = dentry, | |
1284 | }; | |
1285 | struct nfs_open_context *ctx; | |
1286 | struct rpc_cred *cred; | |
1287 | fmode_t fmode = nd->intent.open.flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC); | |
1288 | ||
1289 | cred = rpc_lookup_cred(); | |
1290 | if (IS_ERR(cred)) | |
1291 | return ERR_CAST(cred); | |
1292 | ctx = alloc_nfs_open_context(&path, cred, fmode); | |
1293 | put_rpccred(cred); | |
1294 | if (ctx == NULL) | |
1295 | return ERR_PTR(-ENOMEM); | |
1296 | return ctx; | |
1297 | } | |
1298 | ||
1299 | static int do_open(struct inode *inode, struct file *filp) | |
1300 | { | |
1301 | nfs_fscache_set_inode_cookie(inode, filp); | |
1302 | return 0; | |
1303 | } | |
1304 | ||
1305 | static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx) | |
1306 | { | |
1307 | struct file *filp; | |
1308 | int ret = 0; | |
1309 | ||
1310 | /* If the open_intent is for execute, we have an extra check to make */ | |
1311 | if (ctx->mode & FMODE_EXEC) { | |
1312 | ret = nfs_may_open(ctx->path.dentry->d_inode, | |
1313 | ctx->cred, | |
1314 | nd->intent.open.flags); | |
1315 | if (ret < 0) | |
1316 | goto out; | |
1317 | } | |
1318 | filp = lookup_instantiate_filp(nd, ctx->path.dentry, do_open); | |
1319 | if (IS_ERR(filp)) | |
1320 | ret = PTR_ERR(filp); | |
1321 | else | |
1322 | nfs_file_set_open_context(filp, ctx); | |
1323 | out: | |
1324 | put_nfs_open_context(ctx); | |
1325 | return ret; | |
1326 | } | |
1327 | ||
1da177e4 LT |
1328 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) |
1329 | { | |
cd9a1c0e TM |
1330 | struct nfs_open_context *ctx; |
1331 | struct iattr attr; | |
1da177e4 | 1332 | struct dentry *res = NULL; |
f46e0bd3 | 1333 | struct inode *inode; |
cd9a1c0e | 1334 | int open_flags; |
898f635c | 1335 | int err; |
1da177e4 | 1336 | |
1e7cb3dc CL |
1337 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
1338 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1339 | ||
1da177e4 | 1340 | /* Check that we are indeed trying to open this file */ |
5584c306 | 1341 | if (!is_atomic_open(nd)) |
1da177e4 LT |
1342 | goto no_open; |
1343 | ||
1344 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1345 | res = ERR_PTR(-ENAMETOOLONG); | |
1346 | goto out; | |
1347 | } | |
1348 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1349 | ||
d4d9cdcb TM |
1350 | /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash |
1351 | * the dentry. */ | |
3516586a | 1352 | if (nd->flags & LOOKUP_EXCL) { |
d4d9cdcb | 1353 | d_instantiate(dentry, NULL); |
02a913a7 TM |
1354 | goto out; |
1355 | } | |
1da177e4 | 1356 | |
cd9a1c0e TM |
1357 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1358 | res = ERR_CAST(ctx); | |
1359 | if (IS_ERR(ctx)) | |
1360 | goto out; | |
1361 | ||
1362 | open_flags = nd->intent.open.flags; | |
1363 | if (nd->flags & LOOKUP_CREATE) { | |
1364 | attr.ia_mode = nd->intent.open.create_mode; | |
1365 | attr.ia_valid = ATTR_MODE; | |
1366 | if (!IS_POSIXACL(dir)) | |
1367 | attr.ia_mode &= ~current_umask(); | |
1368 | } else { | |
898f635c | 1369 | open_flags &= ~(O_EXCL | O_CREAT); |
cd9a1c0e | 1370 | attr.ia_valid = 0; |
cd9a1c0e TM |
1371 | } |
1372 | ||
1da177e4 | 1373 | /* Open the file on the server */ |
f46e0bd3 | 1374 | nfs_block_sillyrename(dentry->d_parent); |
2b484297 | 1375 | inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr); |
f46e0bd3 TM |
1376 | if (IS_ERR(inode)) { |
1377 | nfs_unblock_sillyrename(dentry->d_parent); | |
cd9a1c0e | 1378 | put_nfs_open_context(ctx); |
f46e0bd3 | 1379 | switch (PTR_ERR(inode)) { |
1da177e4 LT |
1380 | /* Make a negative dentry */ |
1381 | case -ENOENT: | |
f46e0bd3 | 1382 | d_add(dentry, NULL); |
02a913a7 TM |
1383 | res = NULL; |
1384 | goto out; | |
1da177e4 | 1385 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1386 | case -ENOTDIR: |
1387 | goto no_open; | |
1da177e4 LT |
1388 | case -ELOOP: |
1389 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1390 | goto no_open; | |
23ebbd9a | 1391 | /* case -EISDIR: */ |
1da177e4 LT |
1392 | /* case -EINVAL: */ |
1393 | default: | |
f46e0bd3 | 1394 | res = ERR_CAST(inode); |
1da177e4 LT |
1395 | goto out; |
1396 | } | |
cd9a1c0e | 1397 | } |
f46e0bd3 | 1398 | res = d_add_unique(dentry, inode); |
898f635c | 1399 | nfs_unblock_sillyrename(dentry->d_parent); |
f46e0bd3 TM |
1400 | if (res != NULL) { |
1401 | dput(ctx->path.dentry); | |
1402 | ctx->path.dentry = dget(res); | |
1da177e4 | 1403 | dentry = res; |
f46e0bd3 | 1404 | } |
898f635c TM |
1405 | err = nfs_intent_set_file(nd, ctx); |
1406 | if (err < 0) { | |
1407 | if (res != NULL) | |
1408 | dput(res); | |
1409 | return ERR_PTR(err); | |
1410 | } | |
1da177e4 | 1411 | out: |
f46e0bd3 | 1412 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1413 | return res; |
1414 | no_open: | |
1415 | return nfs_lookup(dir, dentry, nd); | |
1416 | } | |
1417 | ||
1418 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1419 | { | |
1420 | struct dentry *parent = NULL; | |
1421 | struct inode *inode = dentry->d_inode; | |
1422 | struct inode *dir; | |
b8d4cadd | 1423 | struct nfs_open_context *ctx; |
1da177e4 LT |
1424 | int openflags, ret = 0; |
1425 | ||
1f063d2c | 1426 | if (!is_atomic_open(nd) || d_mountpoint(dentry)) |
5584c306 | 1427 | goto no_open; |
2b484297 | 1428 | |
1da177e4 LT |
1429 | parent = dget_parent(dentry); |
1430 | dir = parent->d_inode; | |
2b484297 | 1431 | |
1da177e4 LT |
1432 | /* We can't create new files in nfs_open_revalidate(), so we |
1433 | * optimize away revalidation of negative dentries. | |
1434 | */ | |
216d5d06 TM |
1435 | if (inode == NULL) { |
1436 | if (!nfs_neg_need_reval(dir, dentry, nd)) | |
1437 | ret = 1; | |
1da177e4 | 1438 | goto out; |
216d5d06 TM |
1439 | } |
1440 | ||
1da177e4 LT |
1441 | /* NFS only supports OPEN on regular files */ |
1442 | if (!S_ISREG(inode->i_mode)) | |
5584c306 | 1443 | goto no_open_dput; |
1da177e4 LT |
1444 | openflags = nd->intent.open.flags; |
1445 | /* We cannot do exclusive creation on a positive dentry */ | |
1446 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
5584c306 | 1447 | goto no_open_dput; |
1da177e4 | 1448 | /* We can't create new files, or truncate existing ones here */ |
0a377cff | 1449 | openflags &= ~(O_CREAT|O_EXCL|O_TRUNC); |
1da177e4 | 1450 | |
b8d4cadd TM |
1451 | ctx = nameidata_to_nfs_open_context(dentry, nd); |
1452 | ret = PTR_ERR(ctx); | |
1453 | if (IS_ERR(ctx)) | |
1454 | goto out; | |
1da177e4 | 1455 | /* |
1b1dcc1b | 1456 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1457 | * operations that change the directory. We therefore save the |
1458 | * change attribute *before* we do the RPC call. | |
1459 | */ | |
2b484297 | 1460 | inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL); |
535918f1 TM |
1461 | if (IS_ERR(inode)) { |
1462 | ret = PTR_ERR(inode); | |
1463 | switch (ret) { | |
1464 | case -EPERM: | |
1465 | case -EACCES: | |
1466 | case -EDQUOT: | |
1467 | case -ENOSPC: | |
1468 | case -EROFS: | |
1469 | goto out_put_ctx; | |
1470 | default: | |
1471 | goto out_drop; | |
1472 | } | |
1473 | } | |
1474 | iput(inode); | |
898f635c | 1475 | if (inode != dentry->d_inode) |
535918f1 | 1476 | goto out_drop; |
898f635c TM |
1477 | |
1478 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1479 | ret = nfs_intent_set_file(nd, ctx); | |
1480 | if (ret >= 0) | |
1481 | ret = 1; | |
1da177e4 LT |
1482 | out: |
1483 | dput(parent); | |
1da177e4 | 1484 | return ret; |
535918f1 TM |
1485 | out_drop: |
1486 | d_drop(dentry); | |
1487 | ret = 0; | |
1488 | out_put_ctx: | |
1489 | put_nfs_open_context(ctx); | |
1490 | goto out; | |
1491 | ||
5584c306 | 1492 | no_open_dput: |
1da177e4 | 1493 | dput(parent); |
5584c306 | 1494 | no_open: |
1da177e4 LT |
1495 | return nfs_lookup_revalidate(dentry, nd); |
1496 | } | |
c0204fd2 TM |
1497 | |
1498 | static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, | |
1499 | struct nameidata *nd) | |
1500 | { | |
1501 | struct nfs_open_context *ctx = NULL; | |
1502 | struct iattr attr; | |
1503 | int error; | |
1504 | int open_flags = 0; | |
1505 | ||
1506 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", | |
1507 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1508 | ||
1509 | attr.ia_mode = mode; | |
1510 | attr.ia_valid = ATTR_MODE; | |
1511 | ||
1512 | if ((nd->flags & LOOKUP_CREATE) != 0) { | |
1513 | open_flags = nd->intent.open.flags; | |
1514 | ||
1515 | ctx = nameidata_to_nfs_open_context(dentry, nd); | |
1516 | error = PTR_ERR(ctx); | |
1517 | if (IS_ERR(ctx)) | |
898f635c | 1518 | goto out_err_drop; |
c0204fd2 TM |
1519 | } |
1520 | ||
1521 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx); | |
1522 | if (error != 0) | |
1523 | goto out_put_ctx; | |
898f635c TM |
1524 | if (ctx != NULL) { |
1525 | error = nfs_intent_set_file(nd, ctx); | |
1526 | if (error < 0) | |
1527 | goto out_err; | |
1528 | } | |
c0204fd2 TM |
1529 | return 0; |
1530 | out_put_ctx: | |
1531 | if (ctx != NULL) | |
1532 | put_nfs_open_context(ctx); | |
898f635c | 1533 | out_err_drop: |
c0204fd2 | 1534 | d_drop(dentry); |
898f635c | 1535 | out_err: |
c0204fd2 TM |
1536 | return error; |
1537 | } | |
1538 | ||
1da177e4 LT |
1539 | #endif /* CONFIG_NFSV4 */ |
1540 | ||
1da177e4 LT |
1541 | /* |
1542 | * Code common to create, mkdir, and mknod. | |
1543 | */ | |
1544 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1545 | struct nfs_fattr *fattr) | |
1546 | { | |
fab728e1 TM |
1547 | struct dentry *parent = dget_parent(dentry); |
1548 | struct inode *dir = parent->d_inode; | |
1da177e4 LT |
1549 | struct inode *inode; |
1550 | int error = -EACCES; | |
1551 | ||
fab728e1 TM |
1552 | d_drop(dentry); |
1553 | ||
1da177e4 LT |
1554 | /* We may have been initialized further down */ |
1555 | if (dentry->d_inode) | |
fab728e1 | 1556 | goto out; |
1da177e4 | 1557 | if (fhandle->size == 0) { |
1da177e4 LT |
1558 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); |
1559 | if (error) | |
fab728e1 | 1560 | goto out_error; |
1da177e4 | 1561 | } |
5724ab37 | 1562 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1da177e4 LT |
1563 | if (!(fattr->valid & NFS_ATTR_FATTR)) { |
1564 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1565 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1566 | if (error < 0) |
fab728e1 | 1567 | goto out_error; |
1da177e4 | 1568 | } |
1da177e4 | 1569 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1570 | error = PTR_ERR(inode); |
1571 | if (IS_ERR(inode)) | |
fab728e1 TM |
1572 | goto out_error; |
1573 | d_add(dentry, inode); | |
1574 | out: | |
1575 | dput(parent); | |
1da177e4 | 1576 | return 0; |
fab728e1 TM |
1577 | out_error: |
1578 | nfs_mark_for_revalidate(dir); | |
1579 | dput(parent); | |
1580 | return error; | |
1da177e4 LT |
1581 | } |
1582 | ||
1583 | /* | |
1584 | * Following a failed create operation, we drop the dentry rather | |
1585 | * than retain a negative dentry. This avoids a problem in the event | |
1586 | * that the operation succeeded on the server, but an error in the | |
1587 | * reply path made it appear to have failed. | |
1588 | */ | |
1589 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1590 | struct nameidata *nd) | |
1591 | { | |
1592 | struct iattr attr; | |
1593 | int error; | |
1da177e4 | 1594 | |
1e7cb3dc CL |
1595 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1596 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1597 | |
1598 | attr.ia_mode = mode; | |
1599 | attr.ia_valid = ATTR_MODE; | |
1600 | ||
c0204fd2 | 1601 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, 0, NULL); |
1da177e4 LT |
1602 | if (error != 0) |
1603 | goto out_err; | |
1da177e4 LT |
1604 | return 0; |
1605 | out_err: | |
1da177e4 LT |
1606 | d_drop(dentry); |
1607 | return error; | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * See comments for nfs_proc_create regarding failed operations. | |
1612 | */ | |
1613 | static int | |
1614 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1615 | { | |
1616 | struct iattr attr; | |
1617 | int status; | |
1618 | ||
1e7cb3dc CL |
1619 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1620 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1621 | |
1622 | if (!new_valid_dev(rdev)) | |
1623 | return -EINVAL; | |
1624 | ||
1625 | attr.ia_mode = mode; | |
1626 | attr.ia_valid = ATTR_MODE; | |
1627 | ||
1da177e4 | 1628 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); |
1da177e4 LT |
1629 | if (status != 0) |
1630 | goto out_err; | |
1da177e4 LT |
1631 | return 0; |
1632 | out_err: | |
1da177e4 LT |
1633 | d_drop(dentry); |
1634 | return status; | |
1635 | } | |
1636 | ||
1637 | /* | |
1638 | * See comments for nfs_proc_create regarding failed operations. | |
1639 | */ | |
1640 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1641 | { | |
1642 | struct iattr attr; | |
1643 | int error; | |
1644 | ||
1e7cb3dc CL |
1645 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1646 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1647 | |
1648 | attr.ia_valid = ATTR_MODE; | |
1649 | attr.ia_mode = mode | S_IFDIR; | |
1650 | ||
1da177e4 | 1651 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); |
1da177e4 LT |
1652 | if (error != 0) |
1653 | goto out_err; | |
1da177e4 LT |
1654 | return 0; |
1655 | out_err: | |
1656 | d_drop(dentry); | |
1da177e4 LT |
1657 | return error; |
1658 | } | |
1659 | ||
d45b9d8b TM |
1660 | static void nfs_dentry_handle_enoent(struct dentry *dentry) |
1661 | { | |
1662 | if (dentry->d_inode != NULL && !d_unhashed(dentry)) | |
1663 | d_delete(dentry); | |
1664 | } | |
1665 | ||
1da177e4 LT |
1666 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) |
1667 | { | |
1668 | int error; | |
1669 | ||
1e7cb3dc CL |
1670 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1671 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 | 1672 | |
1da177e4 LT |
1673 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); |
1674 | /* Ensure the VFS deletes this inode */ | |
1675 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1676 | clear_nlink(dentry->d_inode); |
d45b9d8b TM |
1677 | else if (error == -ENOENT) |
1678 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1679 | |
1680 | return error; | |
1681 | } | |
1682 | ||
1da177e4 LT |
1683 | /* |
1684 | * Remove a file after making sure there are no pending writes, | |
1685 | * and after checking that the file has only one user. | |
1686 | * | |
1687 | * We invalidate the attribute cache and free the inode prior to the operation | |
1688 | * to avoid possible races if the server reuses the inode. | |
1689 | */ | |
1690 | static int nfs_safe_remove(struct dentry *dentry) | |
1691 | { | |
1692 | struct inode *dir = dentry->d_parent->d_inode; | |
1693 | struct inode *inode = dentry->d_inode; | |
1694 | int error = -EBUSY; | |
1695 | ||
1696 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1697 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1698 | ||
1699 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1700 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1701 | error = 0; | |
1702 | goto out; | |
1703 | } | |
1704 | ||
1da177e4 | 1705 | if (inode != NULL) { |
cae7a073 | 1706 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1707 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); |
1708 | /* The VFS may want to delete this inode */ | |
1709 | if (error == 0) | |
1b83d707 | 1710 | nfs_drop_nlink(inode); |
5ba7cc48 | 1711 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1712 | } else |
1713 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
d45b9d8b TM |
1714 | if (error == -ENOENT) |
1715 | nfs_dentry_handle_enoent(dentry); | |
1da177e4 LT |
1716 | out: |
1717 | return error; | |
1718 | } | |
1719 | ||
1720 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1721 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1722 | * | |
1723 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1724 | */ | |
1725 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1726 | { | |
1727 | int error; | |
1728 | int need_rehash = 0; | |
1729 | ||
1730 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1731 | dir->i_ino, dentry->d_name.name); | |
1732 | ||
1da177e4 LT |
1733 | spin_lock(&dcache_lock); |
1734 | spin_lock(&dentry->d_lock); | |
1735 | if (atomic_read(&dentry->d_count) > 1) { | |
1736 | spin_unlock(&dentry->d_lock); | |
1737 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1738 | /* Start asynchronous writeout of the inode */ |
1739 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 | 1740 | error = nfs_sillyrename(dir, dentry); |
1da177e4 LT |
1741 | return error; |
1742 | } | |
1743 | if (!d_unhashed(dentry)) { | |
1744 | __d_drop(dentry); | |
1745 | need_rehash = 1; | |
1746 | } | |
1747 | spin_unlock(&dentry->d_lock); | |
1748 | spin_unlock(&dcache_lock); | |
1749 | error = nfs_safe_remove(dentry); | |
d45b9d8b | 1750 | if (!error || error == -ENOENT) { |
1da177e4 LT |
1751 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1752 | } else if (need_rehash) | |
1753 | d_rehash(dentry); | |
1da177e4 LT |
1754 | return error; |
1755 | } | |
1756 | ||
873101b3 CL |
1757 | /* |
1758 | * To create a symbolic link, most file systems instantiate a new inode, | |
1759 | * add a page to it containing the path, then write it out to the disk | |
1760 | * using prepare_write/commit_write. | |
1761 | * | |
1762 | * Unfortunately the NFS client can't create the in-core inode first | |
1763 | * because it needs a file handle to create an in-core inode (see | |
1764 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1765 | * symlink request has completed on the server. | |
1766 | * | |
1767 | * So instead we allocate a raw page, copy the symname into it, then do | |
1768 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1769 | * now have a new file handle and can instantiate an in-core NFS inode | |
1770 | * and move the raw page into its mapping. | |
1771 | */ | |
1772 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1773 | { |
873101b3 CL |
1774 | struct pagevec lru_pvec; |
1775 | struct page *page; | |
1776 | char *kaddr; | |
1da177e4 | 1777 | struct iattr attr; |
873101b3 | 1778 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1779 | int error; |
1780 | ||
1781 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1782 | dir->i_ino, dentry->d_name.name, symname); | |
1783 | ||
873101b3 CL |
1784 | if (pathlen > PAGE_SIZE) |
1785 | return -ENAMETOOLONG; | |
1da177e4 | 1786 | |
873101b3 CL |
1787 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1788 | attr.ia_valid = ATTR_MODE; | |
1da177e4 | 1789 | |
83d93f22 | 1790 | page = alloc_page(GFP_HIGHUSER); |
76566991 | 1791 | if (!page) |
873101b3 | 1792 | return -ENOMEM; |
873101b3 CL |
1793 | |
1794 | kaddr = kmap_atomic(page, KM_USER0); | |
1795 | memcpy(kaddr, symname, pathlen); | |
1796 | if (pathlen < PAGE_SIZE) | |
1797 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1798 | kunmap_atomic(kaddr, KM_USER0); | |
1799 | ||
94a6d753 | 1800 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
873101b3 CL |
1801 | if (error != 0) { |
1802 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1803 | dir->i_sb->s_id, dir->i_ino, | |
1804 | dentry->d_name.name, symname, error); | |
1da177e4 | 1805 | d_drop(dentry); |
873101b3 | 1806 | __free_page(page); |
873101b3 CL |
1807 | return error; |
1808 | } | |
1809 | ||
1810 | /* | |
1811 | * No big deal if we can't add this page to the page cache here. | |
1812 | * READLINK will get the missing page from the server if needed. | |
1813 | */ | |
1814 | pagevec_init(&lru_pvec, 0); | |
1815 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1816 | GFP_KERNEL)) { | |
39cf8a13 | 1817 | pagevec_add(&lru_pvec, page); |
4f98a2fe | 1818 | pagevec_lru_add_file(&lru_pvec); |
873101b3 CL |
1819 | SetPageUptodate(page); |
1820 | unlock_page(page); | |
1821 | } else | |
1822 | __free_page(page); | |
1823 | ||
873101b3 | 1824 | return 0; |
1da177e4 LT |
1825 | } |
1826 | ||
1827 | static int | |
1828 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1829 | { | |
1830 | struct inode *inode = old_dentry->d_inode; | |
1831 | int error; | |
1832 | ||
1833 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1834 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1835 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1836 | ||
9a3936aa TM |
1837 | nfs_inode_return_delegation(inode); |
1838 | ||
9697d234 | 1839 | d_drop(dentry); |
1da177e4 | 1840 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); |
cf809556 | 1841 | if (error == 0) { |
7de9c6ee | 1842 | ihold(inode); |
9697d234 | 1843 | d_add(dentry, inode); |
cf809556 | 1844 | } |
1da177e4 LT |
1845 | return error; |
1846 | } | |
1847 | ||
1848 | /* | |
1849 | * RENAME | |
1850 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1851 | * different file handle for the same inode after a rename (e.g. when | |
1852 | * moving to a different directory). A fail-safe method to do so would | |
1853 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1854 | * rename the old file using the sillyrename stuff. This way, the original | |
1855 | * file in old_dir will go away when the last process iput()s the inode. | |
1856 | * | |
1857 | * FIXED. | |
1858 | * | |
1859 | * It actually works quite well. One needs to have the possibility for | |
1860 | * at least one ".nfs..." file in each directory the file ever gets | |
1861 | * moved or linked to which happens automagically with the new | |
1862 | * implementation that only depends on the dcache stuff instead of | |
1863 | * using the inode layer | |
1864 | * | |
1865 | * Unfortunately, things are a little more complicated than indicated | |
1866 | * above. For a cross-directory move, we want to make sure we can get | |
1867 | * rid of the old inode after the operation. This means there must be | |
1868 | * no pending writes (if it's a file), and the use count must be 1. | |
1869 | * If these conditions are met, we can drop the dentries before doing | |
1870 | * the rename. | |
1871 | */ | |
1872 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1873 | struct inode *new_dir, struct dentry *new_dentry) | |
1874 | { | |
1875 | struct inode *old_inode = old_dentry->d_inode; | |
1876 | struct inode *new_inode = new_dentry->d_inode; | |
1877 | struct dentry *dentry = NULL, *rehash = NULL; | |
1878 | int error = -EBUSY; | |
1879 | ||
1da177e4 LT |
1880 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", |
1881 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1882 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1883 | atomic_read(&new_dentry->d_count)); | |
1884 | ||
1885 | /* | |
28f79a1a MS |
1886 | * For non-directories, check whether the target is busy and if so, |
1887 | * make a copy of the dentry and then do a silly-rename. If the | |
1888 | * silly-rename succeeds, the copied dentry is hashed and becomes | |
1889 | * the new target. | |
1da177e4 | 1890 | */ |
27226104 MS |
1891 | if (new_inode && !S_ISDIR(new_inode->i_mode)) { |
1892 | /* | |
1893 | * To prevent any new references to the target during the | |
1894 | * rename, we unhash the dentry in advance. | |
1895 | */ | |
1896 | if (!d_unhashed(new_dentry)) { | |
1897 | d_drop(new_dentry); | |
1898 | rehash = new_dentry; | |
1899 | } | |
1da177e4 | 1900 | |
27226104 MS |
1901 | if (atomic_read(&new_dentry->d_count) > 2) { |
1902 | int err; | |
1903 | ||
1904 | /* copy the target dentry's name */ | |
1905 | dentry = d_alloc(new_dentry->d_parent, | |
1906 | &new_dentry->d_name); | |
1907 | if (!dentry) | |
1908 | goto out; | |
1909 | ||
1910 | /* silly-rename the existing target ... */ | |
1911 | err = nfs_sillyrename(new_dir, new_dentry); | |
24e93025 | 1912 | if (err) |
27226104 | 1913 | goto out; |
24e93025 MS |
1914 | |
1915 | new_dentry = dentry; | |
56335936 | 1916 | rehash = NULL; |
24e93025 | 1917 | new_inode = NULL; |
27226104 | 1918 | } |
b1e4adf4 | 1919 | } |
1da177e4 | 1920 | |
cae7a073 | 1921 | nfs_inode_return_delegation(old_inode); |
b1e4adf4 | 1922 | if (new_inode != NULL) |
24174119 | 1923 | nfs_inode_return_delegation(new_inode); |
1da177e4 | 1924 | |
1da177e4 LT |
1925 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, |
1926 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1927 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1928 | out: |
1929 | if (rehash) | |
1930 | d_rehash(rehash); | |
1931 | if (!error) { | |
b1e4adf4 TM |
1932 | if (new_inode != NULL) |
1933 | nfs_drop_nlink(new_inode); | |
349457cc | 1934 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1935 | nfs_set_verifier(new_dentry, |
1936 | nfs_save_change_attribute(new_dir)); | |
d45b9d8b TM |
1937 | } else if (error == -ENOENT) |
1938 | nfs_dentry_handle_enoent(old_dentry); | |
1da177e4 LT |
1939 | |
1940 | /* new dentry created? */ | |
1941 | if (dentry) | |
1942 | dput(dentry); | |
1da177e4 LT |
1943 | return error; |
1944 | } | |
1945 | ||
cfcea3e8 TM |
1946 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1947 | static LIST_HEAD(nfs_access_lru_list); | |
1948 | static atomic_long_t nfs_access_nr_entries; | |
1949 | ||
1c3c07e9 TM |
1950 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1951 | { | |
1952 | put_rpccred(entry->cred); | |
1953 | kfree(entry); | |
cfcea3e8 TM |
1954 | smp_mb__before_atomic_dec(); |
1955 | atomic_long_dec(&nfs_access_nr_entries); | |
1956 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1957 | } |
1958 | ||
1a81bb8a TM |
1959 | static void nfs_access_free_list(struct list_head *head) |
1960 | { | |
1961 | struct nfs_access_entry *cache; | |
1962 | ||
1963 | while (!list_empty(head)) { | |
1964 | cache = list_entry(head->next, struct nfs_access_entry, lru); | |
1965 | list_del(&cache->lru); | |
1966 | nfs_access_free_entry(cache); | |
1967 | } | |
1968 | } | |
1969 | ||
7f8275d0 | 1970 | int nfs_access_cache_shrinker(struct shrinker *shrink, int nr_to_scan, gfp_t gfp_mask) |
979df72e TM |
1971 | { |
1972 | LIST_HEAD(head); | |
aa510da5 | 1973 | struct nfs_inode *nfsi, *next; |
979df72e TM |
1974 | struct nfs_access_entry *cache; |
1975 | ||
61d5eb29 TM |
1976 | if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL) |
1977 | return (nr_to_scan == 0) ? 0 : -1; | |
9c7e7e23 | 1978 | |
a50f7951 | 1979 | spin_lock(&nfs_access_lru_lock); |
aa510da5 | 1980 | list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) { |
979df72e TM |
1981 | struct inode *inode; |
1982 | ||
1983 | if (nr_to_scan-- == 0) | |
1984 | break; | |
9c7e7e23 | 1985 | inode = &nfsi->vfs_inode; |
979df72e TM |
1986 | spin_lock(&inode->i_lock); |
1987 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1988 | goto remove_lru_entry; | |
1989 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1990 | struct nfs_access_entry, lru); | |
1991 | list_move(&cache->lru, &head); | |
1992 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1993 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1994 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1995 | &nfs_access_lru_list); | |
1996 | else { | |
1997 | remove_lru_entry: | |
1998 | list_del_init(&nfsi->access_cache_inode_lru); | |
9c7e7e23 | 1999 | smp_mb__before_clear_bit(); |
979df72e | 2000 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); |
9c7e7e23 | 2001 | smp_mb__after_clear_bit(); |
979df72e | 2002 | } |
59844a9b | 2003 | spin_unlock(&inode->i_lock); |
979df72e TM |
2004 | } |
2005 | spin_unlock(&nfs_access_lru_lock); | |
1a81bb8a | 2006 | nfs_access_free_list(&head); |
979df72e TM |
2007 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; |
2008 | } | |
2009 | ||
1a81bb8a | 2010 | static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head) |
1da177e4 | 2011 | { |
1c3c07e9 | 2012 | struct rb_root *root_node = &nfsi->access_cache; |
1a81bb8a | 2013 | struct rb_node *n; |
1c3c07e9 TM |
2014 | struct nfs_access_entry *entry; |
2015 | ||
2016 | /* Unhook entries from the cache */ | |
2017 | while ((n = rb_first(root_node)) != NULL) { | |
2018 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2019 | rb_erase(n, root_node); | |
1a81bb8a | 2020 | list_move(&entry->lru, head); |
1c3c07e9 TM |
2021 | } |
2022 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1da177e4 LT |
2023 | } |
2024 | ||
1c3c07e9 | 2025 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 2026 | { |
1a81bb8a TM |
2027 | LIST_HEAD(head); |
2028 | ||
2029 | if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0) | |
2030 | return; | |
cfcea3e8 | 2031 | /* Remove from global LRU init */ |
1a81bb8a TM |
2032 | spin_lock(&nfs_access_lru_lock); |
2033 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) | |
cfcea3e8 | 2034 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); |
cfcea3e8 | 2035 | |
1c3c07e9 | 2036 | spin_lock(&inode->i_lock); |
1a81bb8a TM |
2037 | __nfs_access_zap_cache(NFS_I(inode), &head); |
2038 | spin_unlock(&inode->i_lock); | |
2039 | spin_unlock(&nfs_access_lru_lock); | |
2040 | nfs_access_free_list(&head); | |
1c3c07e9 | 2041 | } |
1da177e4 | 2042 | |
1c3c07e9 TM |
2043 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
2044 | { | |
2045 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
2046 | struct nfs_access_entry *entry; | |
2047 | ||
2048 | while (n != NULL) { | |
2049 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
2050 | ||
2051 | if (cred < entry->cred) | |
2052 | n = n->rb_left; | |
2053 | else if (cred > entry->cred) | |
2054 | n = n->rb_right; | |
2055 | else | |
2056 | return entry; | |
1da177e4 | 2057 | } |
1c3c07e9 TM |
2058 | return NULL; |
2059 | } | |
2060 | ||
af22f94a | 2061 | static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) |
1c3c07e9 TM |
2062 | { |
2063 | struct nfs_inode *nfsi = NFS_I(inode); | |
2064 | struct nfs_access_entry *cache; | |
2065 | int err = -ENOENT; | |
2066 | ||
dc59250c | 2067 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
2068 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
2069 | goto out_zap; | |
2070 | cache = nfs_access_search_rbtree(inode, cred); | |
2071 | if (cache == NULL) | |
2072 | goto out; | |
b4d2314b | 2073 | if (!nfs_have_delegated_attributes(inode) && |
64672d55 | 2074 | !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo)) |
1c3c07e9 TM |
2075 | goto out_stale; |
2076 | res->jiffies = cache->jiffies; | |
2077 | res->cred = cache->cred; | |
2078 | res->mask = cache->mask; | |
cfcea3e8 | 2079 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
2080 | err = 0; |
2081 | out: | |
2082 | spin_unlock(&inode->i_lock); | |
2083 | return err; | |
2084 | out_stale: | |
2085 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 2086 | list_del(&cache->lru); |
1c3c07e9 TM |
2087 | spin_unlock(&inode->i_lock); |
2088 | nfs_access_free_entry(cache); | |
2089 | return -ENOENT; | |
2090 | out_zap: | |
1a81bb8a TM |
2091 | spin_unlock(&inode->i_lock); |
2092 | nfs_access_zap_cache(inode); | |
1c3c07e9 TM |
2093 | return -ENOENT; |
2094 | } | |
2095 | ||
2096 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
2097 | { | |
cfcea3e8 TM |
2098 | struct nfs_inode *nfsi = NFS_I(inode); |
2099 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
2100 | struct rb_node **p = &root_node->rb_node; |
2101 | struct rb_node *parent = NULL; | |
2102 | struct nfs_access_entry *entry; | |
2103 | ||
2104 | spin_lock(&inode->i_lock); | |
2105 | while (*p != NULL) { | |
2106 | parent = *p; | |
2107 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
2108 | ||
2109 | if (set->cred < entry->cred) | |
2110 | p = &parent->rb_left; | |
2111 | else if (set->cred > entry->cred) | |
2112 | p = &parent->rb_right; | |
2113 | else | |
2114 | goto found; | |
2115 | } | |
2116 | rb_link_node(&set->rb_node, parent, p); | |
2117 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 2118 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 2119 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
2120 | return; |
2121 | found: | |
2122 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
2123 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
2124 | list_del(&entry->lru); | |
1c3c07e9 TM |
2125 | spin_unlock(&inode->i_lock); |
2126 | nfs_access_free_entry(entry); | |
2127 | } | |
2128 | ||
af22f94a | 2129 | static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) |
1c3c07e9 TM |
2130 | { |
2131 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
2132 | if (cache == NULL) | |
2133 | return; | |
2134 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 2135 | cache->jiffies = set->jiffies; |
1c3c07e9 | 2136 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 2137 | cache->mask = set->mask; |
1c3c07e9 TM |
2138 | |
2139 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
2140 | |
2141 | /* Update accounting */ | |
2142 | smp_mb__before_atomic_inc(); | |
2143 | atomic_long_inc(&nfs_access_nr_entries); | |
2144 | smp_mb__after_atomic_inc(); | |
2145 | ||
2146 | /* Add inode to global LRU list */ | |
1a81bb8a | 2147 | if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) { |
cfcea3e8 | 2148 | spin_lock(&nfs_access_lru_lock); |
1a81bb8a TM |
2149 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) |
2150 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, | |
2151 | &nfs_access_lru_list); | |
cfcea3e8 TM |
2152 | spin_unlock(&nfs_access_lru_lock); |
2153 | } | |
1da177e4 LT |
2154 | } |
2155 | ||
2156 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
2157 | { | |
2158 | struct nfs_access_entry cache; | |
2159 | int status; | |
2160 | ||
2161 | status = nfs_access_get_cached(inode, cred, &cache); | |
2162 | if (status == 0) | |
2163 | goto out; | |
2164 | ||
2165 | /* Be clever: ask server to check for all possible rights */ | |
2166 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
2167 | cache.cred = cred; | |
2168 | cache.jiffies = jiffies; | |
2169 | status = NFS_PROTO(inode)->access(inode, &cache); | |
a71ee337 SJ |
2170 | if (status != 0) { |
2171 | if (status == -ESTALE) { | |
2172 | nfs_zap_caches(inode); | |
2173 | if (!S_ISDIR(inode->i_mode)) | |
2174 | set_bit(NFS_INO_STALE, &NFS_I(inode)->flags); | |
2175 | } | |
1da177e4 | 2176 | return status; |
a71ee337 | 2177 | } |
1da177e4 LT |
2178 | nfs_access_add_cache(inode, &cache); |
2179 | out: | |
e6305c43 | 2180 | if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2181 | return 0; |
2182 | return -EACCES; | |
2183 | } | |
2184 | ||
af22f94a TM |
2185 | static int nfs_open_permission_mask(int openflags) |
2186 | { | |
2187 | int mask = 0; | |
2188 | ||
2189 | if (openflags & FMODE_READ) | |
2190 | mask |= MAY_READ; | |
2191 | if (openflags & FMODE_WRITE) | |
2192 | mask |= MAY_WRITE; | |
2193 | if (openflags & FMODE_EXEC) | |
2194 | mask |= MAY_EXEC; | |
2195 | return mask; | |
2196 | } | |
2197 | ||
2198 | int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags) | |
2199 | { | |
2200 | return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags)); | |
2201 | } | |
2202 | ||
e6305c43 | 2203 | int nfs_permission(struct inode *inode, int mask) |
1da177e4 LT |
2204 | { |
2205 | struct rpc_cred *cred; | |
2206 | int res = 0; | |
2207 | ||
91d5b470 CL |
2208 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
2209 | ||
e6305c43 | 2210 | if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0) |
1da177e4 LT |
2211 | goto out; |
2212 | /* Is this sys_access() ? */ | |
9cfcac81 | 2213 | if (mask & (MAY_ACCESS | MAY_CHDIR)) |
1da177e4 LT |
2214 | goto force_lookup; |
2215 | ||
2216 | switch (inode->i_mode & S_IFMT) { | |
2217 | case S_IFLNK: | |
2218 | goto out; | |
2219 | case S_IFREG: | |
2220 | /* NFSv4 has atomic_open... */ | |
2221 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
7ee2cb7f FF |
2222 | && (mask & MAY_OPEN) |
2223 | && !(mask & MAY_EXEC)) | |
1da177e4 LT |
2224 | goto out; |
2225 | break; | |
2226 | case S_IFDIR: | |
2227 | /* | |
2228 | * Optimize away all write operations, since the server | |
2229 | * will check permissions when we perform the op. | |
2230 | */ | |
2231 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
2232 | goto out; | |
2233 | } | |
2234 | ||
2235 | force_lookup: | |
1da177e4 LT |
2236 | if (!NFS_PROTO(inode)->access) |
2237 | goto out_notsup; | |
2238 | ||
98a8e323 | 2239 | cred = rpc_lookup_cred(); |
1da177e4 LT |
2240 | if (!IS_ERR(cred)) { |
2241 | res = nfs_do_access(inode, cred, mask); | |
2242 | put_rpccred(cred); | |
2243 | } else | |
2244 | res = PTR_ERR(cred); | |
1da177e4 | 2245 | out: |
f696a365 MS |
2246 | if (!res && (mask & MAY_EXEC) && !execute_ok(inode)) |
2247 | res = -EACCES; | |
2248 | ||
1e7cb3dc CL |
2249 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
2250 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
2251 | return res; |
2252 | out_notsup: | |
2253 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
2254 | if (res == 0) | |
2255 | res = generic_permission(inode, mask, NULL); | |
1e7cb3dc | 2256 | goto out; |
1da177e4 LT |
2257 | } |
2258 | ||
2259 | /* | |
2260 | * Local variables: | |
2261 | * version-control: t | |
2262 | * kept-new-versions: 5 | |
2263 | * End: | |
2264 | */ |