]>
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> | |
32 | #include <linux/smp_lock.h> | |
873101b3 | 33 | #include <linux/pagevec.h> |
1da177e4 | 34 | #include <linux/namei.h> |
54ceac45 | 35 | #include <linux/mount.h> |
e8edc6e0 | 36 | #include <linux/sched.h> |
1da177e4 | 37 | |
4ce79717 | 38 | #include "nfs4_fs.h" |
1da177e4 | 39 | #include "delegation.h" |
91d5b470 | 40 | #include "iostat.h" |
1da177e4 | 41 | |
1da177e4 LT |
42 | /* #define NFS_DEBUG_VERBOSE 1 */ |
43 | ||
44 | static int nfs_opendir(struct inode *, struct file *); | |
45 | static int nfs_readdir(struct file *, void *, filldir_t); | |
46 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
47 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
48 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
49 | static int nfs_rmdir(struct inode *, struct dentry *); | |
50 | static int nfs_unlink(struct inode *, struct dentry *); | |
51 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
52 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
53 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
54 | static int nfs_rename(struct inode *, struct dentry *, | |
55 | struct inode *, struct dentry *); | |
56 | static int nfs_fsync_dir(struct file *, struct dentry *, int); | |
f0dd2136 | 57 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
1da177e4 | 58 | |
4b6f5d20 | 59 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 60 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
61 | .read = generic_read_dir, |
62 | .readdir = nfs_readdir, | |
63 | .open = nfs_opendir, | |
64 | .release = nfs_release, | |
65 | .fsync = nfs_fsync_dir, | |
66 | }; | |
67 | ||
92e1d5be | 68 | const struct inode_operations nfs_dir_inode_operations = { |
1da177e4 LT |
69 | .create = nfs_create, |
70 | .lookup = nfs_lookup, | |
71 | .link = nfs_link, | |
72 | .unlink = nfs_unlink, | |
73 | .symlink = nfs_symlink, | |
74 | .mkdir = nfs_mkdir, | |
75 | .rmdir = nfs_rmdir, | |
76 | .mknod = nfs_mknod, | |
77 | .rename = nfs_rename, | |
78 | .permission = nfs_permission, | |
79 | .getattr = nfs_getattr, | |
80 | .setattr = nfs_setattr, | |
81 | }; | |
82 | ||
b7fa0554 | 83 | #ifdef CONFIG_NFS_V3 |
92e1d5be | 84 | const struct inode_operations nfs3_dir_inode_operations = { |
b7fa0554 AG |
85 | .create = nfs_create, |
86 | .lookup = nfs_lookup, | |
87 | .link = nfs_link, | |
88 | .unlink = nfs_unlink, | |
89 | .symlink = nfs_symlink, | |
90 | .mkdir = nfs_mkdir, | |
91 | .rmdir = nfs_rmdir, | |
92 | .mknod = nfs_mknod, | |
93 | .rename = nfs_rename, | |
94 | .permission = nfs_permission, | |
95 | .getattr = nfs_getattr, | |
96 | .setattr = nfs_setattr, | |
97 | .listxattr = nfs3_listxattr, | |
98 | .getxattr = nfs3_getxattr, | |
99 | .setxattr = nfs3_setxattr, | |
100 | .removexattr = nfs3_removexattr, | |
101 | }; | |
102 | #endif /* CONFIG_NFS_V3 */ | |
103 | ||
1da177e4 LT |
104 | #ifdef CONFIG_NFS_V4 |
105 | ||
106 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
92e1d5be | 107 | const struct inode_operations nfs4_dir_inode_operations = { |
1da177e4 LT |
108 | .create = nfs_create, |
109 | .lookup = nfs_atomic_lookup, | |
110 | .link = nfs_link, | |
111 | .unlink = nfs_unlink, | |
112 | .symlink = nfs_symlink, | |
113 | .mkdir = nfs_mkdir, | |
114 | .rmdir = nfs_rmdir, | |
115 | .mknod = nfs_mknod, | |
116 | .rename = nfs_rename, | |
117 | .permission = nfs_permission, | |
118 | .getattr = nfs_getattr, | |
119 | .setattr = nfs_setattr, | |
6b3b5496 BF |
120 | .getxattr = nfs4_getxattr, |
121 | .setxattr = nfs4_setxattr, | |
122 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
123 | }; |
124 | ||
125 | #endif /* CONFIG_NFS_V4 */ | |
126 | ||
127 | /* | |
128 | * Open file | |
129 | */ | |
130 | static int | |
131 | nfs_opendir(struct inode *inode, struct file *filp) | |
132 | { | |
7451c4f0 | 133 | int res; |
1da177e4 | 134 | |
1e7cb3dc CL |
135 | dfprintk(VFS, "NFS: opendir(%s/%ld)\n", |
136 | inode->i_sb->s_id, inode->i_ino); | |
137 | ||
1da177e4 LT |
138 | lock_kernel(); |
139 | /* Call generic open code in order to cache credentials */ | |
7451c4f0 | 140 | res = nfs_open(inode, filp); |
1da177e4 LT |
141 | unlock_kernel(); |
142 | return res; | |
143 | } | |
144 | ||
0dbb4c67 | 145 | typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int); |
1da177e4 LT |
146 | typedef struct { |
147 | struct file *file; | |
148 | struct page *page; | |
149 | unsigned long page_index; | |
0dbb4c67 | 150 | __be32 *ptr; |
f0dd2136 TM |
151 | u64 *dir_cookie; |
152 | loff_t current_index; | |
1da177e4 LT |
153 | struct nfs_entry *entry; |
154 | decode_dirent_t decode; | |
155 | int plus; | |
156 | int error; | |
1f4eab7e NB |
157 | unsigned long timestamp; |
158 | int timestamp_valid; | |
1da177e4 LT |
159 | } nfs_readdir_descriptor_t; |
160 | ||
161 | /* Now we cache directories properly, by stuffing the dirent | |
162 | * data directly in the page cache. | |
163 | * | |
164 | * Inode invalidation due to refresh etc. takes care of | |
165 | * _everything_, no sloppy entry flushing logic, no extraneous | |
166 | * copying, network direct to page cache, the way it was meant | |
167 | * to be. | |
168 | * | |
169 | * NOTE: Dirent information verification is done always by the | |
170 | * page-in of the RPC reply, nowhere else, this simplies | |
171 | * things substantially. | |
172 | */ | |
173 | static | |
174 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) | |
175 | { | |
176 | struct file *file = desc->file; | |
01cce933 | 177 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
178 | struct rpc_cred *cred = nfs_file_cred(file); |
179 | unsigned long timestamp; | |
180 | int error; | |
181 | ||
1e7cb3dc CL |
182 | dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", |
183 | __FUNCTION__, (long long)desc->entry->cookie, | |
184 | page->index); | |
1da177e4 LT |
185 | |
186 | again: | |
187 | timestamp = jiffies; | |
01cce933 | 188 | error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page, |
1da177e4 LT |
189 | NFS_SERVER(inode)->dtsize, desc->plus); |
190 | if (error < 0) { | |
191 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
192 | if (error == -ENOTSUPP && desc->plus) { | |
193 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
412d582e | 194 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); |
1da177e4 LT |
195 | desc->plus = 0; |
196 | goto again; | |
197 | } | |
198 | goto error; | |
199 | } | |
1f4eab7e NB |
200 | desc->timestamp = timestamp; |
201 | desc->timestamp_valid = 1; | |
1da177e4 | 202 | SetPageUptodate(page); |
dc59250c | 203 | spin_lock(&inode->i_lock); |
55296809 | 204 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 205 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
206 | /* Ensure consistent page alignment of the data. |
207 | * Note: assumes we have exclusive access to this mapping either | |
1b1dcc1b | 208 | * through inode->i_mutex or some other mechanism. |
1da177e4 | 209 | */ |
cd9ae2b6 TM |
210 | if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) { |
211 | /* Should never happen */ | |
212 | nfs_zap_mapping(inode, inode->i_mapping); | |
213 | } | |
1da177e4 LT |
214 | unlock_page(page); |
215 | return 0; | |
216 | error: | |
217 | SetPageError(page); | |
218 | unlock_page(page); | |
219 | nfs_zap_caches(inode); | |
220 | desc->error = error; | |
221 | return -EIO; | |
222 | } | |
223 | ||
224 | static inline | |
225 | int dir_decode(nfs_readdir_descriptor_t *desc) | |
226 | { | |
0dbb4c67 | 227 | __be32 *p = desc->ptr; |
1da177e4 LT |
228 | p = desc->decode(p, desc->entry, desc->plus); |
229 | if (IS_ERR(p)) | |
230 | return PTR_ERR(p); | |
231 | desc->ptr = p; | |
1f4eab7e NB |
232 | if (desc->timestamp_valid) |
233 | desc->entry->fattr->time_start = desc->timestamp; | |
234 | else | |
235 | desc->entry->fattr->valid &= ~NFS_ATTR_FATTR; | |
1da177e4 LT |
236 | return 0; |
237 | } | |
238 | ||
239 | static inline | |
240 | void dir_page_release(nfs_readdir_descriptor_t *desc) | |
241 | { | |
242 | kunmap(desc->page); | |
243 | page_cache_release(desc->page); | |
244 | desc->page = NULL; | |
245 | desc->ptr = NULL; | |
246 | } | |
247 | ||
248 | /* | |
249 | * Given a pointer to a buffer that has already been filled by a call | |
f0dd2136 | 250 | * to readdir, find the next entry with cookie '*desc->dir_cookie'. |
1da177e4 LT |
251 | * |
252 | * If the end of the buffer has been reached, return -EAGAIN, if not, | |
253 | * return the offset within the buffer of the next entry to be | |
254 | * read. | |
255 | */ | |
256 | static inline | |
00a92642 | 257 | int find_dirent(nfs_readdir_descriptor_t *desc) |
1da177e4 LT |
258 | { |
259 | struct nfs_entry *entry = desc->entry; | |
260 | int loop_count = 0, | |
261 | status; | |
262 | ||
263 | while((status = dir_decode(desc)) == 0) { | |
1e7cb3dc CL |
264 | dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", |
265 | __FUNCTION__, (unsigned long long)entry->cookie); | |
f0dd2136 | 266 | if (entry->prev_cookie == *desc->dir_cookie) |
1da177e4 LT |
267 | break; |
268 | if (loop_count++ > 200) { | |
269 | loop_count = 0; | |
270 | schedule(); | |
271 | } | |
272 | } | |
1da177e4 LT |
273 | return status; |
274 | } | |
275 | ||
276 | /* | |
00a92642 | 277 | * Given a pointer to a buffer that has already been filled by a call |
f0dd2136 | 278 | * to readdir, find the entry at offset 'desc->file->f_pos'. |
00a92642 OG |
279 | * |
280 | * If the end of the buffer has been reached, return -EAGAIN, if not, | |
281 | * return the offset within the buffer of the next entry to be | |
282 | * read. | |
283 | */ | |
284 | static inline | |
285 | int find_dirent_index(nfs_readdir_descriptor_t *desc) | |
286 | { | |
287 | struct nfs_entry *entry = desc->entry; | |
288 | int loop_count = 0, | |
289 | status; | |
290 | ||
291 | for(;;) { | |
292 | status = dir_decode(desc); | |
293 | if (status) | |
294 | break; | |
295 | ||
1e7cb3dc CL |
296 | dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", |
297 | (unsigned long long)entry->cookie, desc->current_index); | |
00a92642 | 298 | |
f0dd2136 TM |
299 | if (desc->file->f_pos == desc->current_index) { |
300 | *desc->dir_cookie = entry->cookie; | |
00a92642 OG |
301 | break; |
302 | } | |
303 | desc->current_index++; | |
304 | if (loop_count++ > 200) { | |
305 | loop_count = 0; | |
306 | schedule(); | |
307 | } | |
308 | } | |
00a92642 OG |
309 | return status; |
310 | } | |
311 | ||
312 | /* | |
313 | * Find the given page, and call find_dirent() or find_dirent_index in | |
314 | * order to try to return the next entry. | |
1da177e4 LT |
315 | */ |
316 | static inline | |
317 | int find_dirent_page(nfs_readdir_descriptor_t *desc) | |
318 | { | |
01cce933 | 319 | struct inode *inode = desc->file->f_path.dentry->d_inode; |
1da177e4 LT |
320 | struct page *page; |
321 | int status; | |
322 | ||
1e7cb3dc CL |
323 | dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", |
324 | __FUNCTION__, desc->page_index, | |
325 | (long long) *desc->dir_cookie); | |
1da177e4 | 326 | |
1f4eab7e NB |
327 | /* If we find the page in the page_cache, we cannot be sure |
328 | * how fresh the data is, so we will ignore readdir_plus attributes. | |
329 | */ | |
330 | desc->timestamp_valid = 0; | |
1da177e4 LT |
331 | page = read_cache_page(inode->i_mapping, desc->page_index, |
332 | (filler_t *)nfs_readdir_filler, desc); | |
333 | if (IS_ERR(page)) { | |
334 | status = PTR_ERR(page); | |
335 | goto out; | |
336 | } | |
1da177e4 LT |
337 | |
338 | /* NOTE: Someone else may have changed the READDIRPLUS flag */ | |
339 | desc->page = page; | |
340 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ | |
f0dd2136 | 341 | if (*desc->dir_cookie != 0) |
00a92642 OG |
342 | status = find_dirent(desc); |
343 | else | |
344 | status = find_dirent_index(desc); | |
1da177e4 LT |
345 | if (status < 0) |
346 | dir_page_release(desc); | |
347 | out: | |
1e7cb3dc | 348 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); |
1da177e4 | 349 | return status; |
1da177e4 LT |
350 | } |
351 | ||
352 | /* | |
353 | * Recurse through the page cache pages, and return a | |
354 | * filled nfs_entry structure of the next directory entry if possible. | |
355 | * | |
f0dd2136 TM |
356 | * The target for the search is '*desc->dir_cookie' if non-0, |
357 | * 'desc->file->f_pos' otherwise | |
1da177e4 LT |
358 | */ |
359 | static inline | |
360 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
361 | { | |
362 | int loop_count = 0; | |
363 | int res; | |
364 | ||
00a92642 | 365 | /* Always search-by-index from the beginning of the cache */ |
f0dd2136 | 366 | if (*desc->dir_cookie == 0) { |
1e7cb3dc CL |
367 | dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", |
368 | (long long)desc->file->f_pos); | |
00a92642 OG |
369 | desc->page_index = 0; |
370 | desc->entry->cookie = desc->entry->prev_cookie = 0; | |
371 | desc->entry->eof = 0; | |
372 | desc->current_index = 0; | |
f0dd2136 | 373 | } else |
1e7cb3dc CL |
374 | dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", |
375 | (unsigned long long)*desc->dir_cookie); | |
00a92642 | 376 | |
1da177e4 LT |
377 | for (;;) { |
378 | res = find_dirent_page(desc); | |
379 | if (res != -EAGAIN) | |
380 | break; | |
381 | /* Align to beginning of next page */ | |
382 | desc->page_index ++; | |
383 | if (loop_count++ > 200) { | |
384 | loop_count = 0; | |
385 | schedule(); | |
386 | } | |
387 | } | |
1e7cb3dc CL |
388 | |
389 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); | |
1da177e4 LT |
390 | return res; |
391 | } | |
392 | ||
393 | static inline unsigned int dt_type(struct inode *inode) | |
394 | { | |
395 | return (inode->i_mode >> 12) & 15; | |
396 | } | |
397 | ||
398 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); | |
399 | ||
400 | /* | |
401 | * Once we've found the start of the dirent within a page: fill 'er up... | |
402 | */ | |
403 | static | |
404 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
405 | filldir_t filldir) | |
406 | { | |
407 | struct file *file = desc->file; | |
408 | struct nfs_entry *entry = desc->entry; | |
409 | struct dentry *dentry = NULL; | |
4e769b93 | 410 | u64 fileid; |
1da177e4 LT |
411 | int loop_count = 0, |
412 | res; | |
413 | ||
1e7cb3dc CL |
414 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", |
415 | (unsigned long long)entry->cookie); | |
1da177e4 LT |
416 | |
417 | for(;;) { | |
418 | unsigned d_type = DT_UNKNOWN; | |
419 | /* Note: entry->prev_cookie contains the cookie for | |
420 | * retrieving the current dirent on the server */ | |
4e769b93 | 421 | fileid = entry->ino; |
1da177e4 LT |
422 | |
423 | /* Get a dentry if we have one */ | |
424 | if (dentry != NULL) | |
425 | dput(dentry); | |
426 | dentry = nfs_readdir_lookup(desc); | |
427 | ||
428 | /* Use readdirplus info */ | |
429 | if (dentry != NULL && dentry->d_inode != NULL) { | |
430 | d_type = dt_type(dentry->d_inode); | |
4e769b93 | 431 | fileid = NFS_FILEID(dentry->d_inode); |
1da177e4 LT |
432 | } |
433 | ||
434 | res = filldir(dirent, entry->name, entry->len, | |
00a92642 | 435 | file->f_pos, fileid, d_type); |
1da177e4 LT |
436 | if (res < 0) |
437 | break; | |
00a92642 | 438 | file->f_pos++; |
f0dd2136 | 439 | *desc->dir_cookie = entry->cookie; |
1da177e4 LT |
440 | if (dir_decode(desc) != 0) { |
441 | desc->page_index ++; | |
442 | break; | |
443 | } | |
444 | if (loop_count++ > 200) { | |
445 | loop_count = 0; | |
446 | schedule(); | |
447 | } | |
448 | } | |
449 | dir_page_release(desc); | |
450 | if (dentry != NULL) | |
451 | dput(dentry); | |
1e7cb3dc CL |
452 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
453 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
454 | return res; |
455 | } | |
456 | ||
457 | /* | |
458 | * If we cannot find a cookie in our cache, we suspect that this is | |
459 | * because it points to a deleted file, so we ask the server to return | |
460 | * whatever it thinks is the next entry. We then feed this to filldir. | |
461 | * If all goes well, we should then be able to find our way round the | |
462 | * cache on the next call to readdir_search_pagecache(); | |
463 | * | |
464 | * NOTE: we cannot add the anonymous page to the pagecache because | |
465 | * the data it contains might not be page aligned. Besides, | |
466 | * we should already have a complete representation of the | |
467 | * directory in the page cache by the time we get here. | |
468 | */ | |
469 | static inline | |
470 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
471 | filldir_t filldir) | |
472 | { | |
473 | struct file *file = desc->file; | |
01cce933 | 474 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
475 | struct rpc_cred *cred = nfs_file_cred(file); |
476 | struct page *page = NULL; | |
477 | int status; | |
1f4eab7e | 478 | unsigned long timestamp; |
1da177e4 | 479 | |
1e7cb3dc CL |
480 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
481 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
482 | |
483 | page = alloc_page(GFP_HIGHUSER); | |
484 | if (!page) { | |
485 | status = -ENOMEM; | |
486 | goto out; | |
487 | } | |
1f4eab7e | 488 | timestamp = jiffies; |
01cce933 | 489 | desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie, |
1da177e4 LT |
490 | page, |
491 | NFS_SERVER(inode)->dtsize, | |
492 | desc->plus); | |
dc59250c | 493 | spin_lock(&inode->i_lock); |
55296809 | 494 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 495 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
496 | desc->page = page; |
497 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ | |
498 | if (desc->error >= 0) { | |
1f4eab7e NB |
499 | desc->timestamp = timestamp; |
500 | desc->timestamp_valid = 1; | |
1da177e4 | 501 | if ((status = dir_decode(desc)) == 0) |
f0dd2136 | 502 | desc->entry->prev_cookie = *desc->dir_cookie; |
1da177e4 LT |
503 | } else |
504 | status = -EIO; | |
505 | if (status < 0) | |
506 | goto out_release; | |
507 | ||
508 | status = nfs_do_filldir(desc, dirent, filldir); | |
509 | ||
510 | /* Reset read descriptor so it searches the page cache from | |
511 | * the start upon the next call to readdir_search_pagecache() */ | |
512 | desc->page_index = 0; | |
513 | desc->entry->cookie = desc->entry->prev_cookie = 0; | |
514 | desc->entry->eof = 0; | |
515 | out: | |
1e7cb3dc CL |
516 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
517 | __FUNCTION__, status); | |
1da177e4 LT |
518 | return status; |
519 | out_release: | |
520 | dir_page_release(desc); | |
521 | goto out; | |
522 | } | |
523 | ||
00a92642 OG |
524 | /* The file offset position represents the dirent entry number. A |
525 | last cookie cache takes care of the common case of reading the | |
526 | whole directory. | |
1da177e4 LT |
527 | */ |
528 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
529 | { | |
01cce933 | 530 | struct dentry *dentry = filp->f_path.dentry; |
1da177e4 LT |
531 | struct inode *inode = dentry->d_inode; |
532 | nfs_readdir_descriptor_t my_desc, | |
533 | *desc = &my_desc; | |
534 | struct nfs_entry my_entry; | |
535 | struct nfs_fh fh; | |
536 | struct nfs_fattr fattr; | |
537 | long res; | |
538 | ||
1e7cb3dc CL |
539 | dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", |
540 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
541 | (long long)filp->f_pos); | |
91d5b470 CL |
542 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
543 | ||
1da177e4 LT |
544 | lock_kernel(); |
545 | ||
717d44e8 | 546 | res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping); |
1da177e4 LT |
547 | if (res < 0) { |
548 | unlock_kernel(); | |
549 | return res; | |
550 | } | |
551 | ||
552 | /* | |
00a92642 | 553 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 554 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
555 | * to either find the entry with the appropriate number or |
556 | * revalidate the cookie. | |
1da177e4 LT |
557 | */ |
558 | memset(desc, 0, sizeof(*desc)); | |
559 | ||
560 | desc->file = filp; | |
f0dd2136 | 561 | desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie; |
1da177e4 LT |
562 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
563 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
564 | ||
565 | my_entry.cookie = my_entry.prev_cookie = 0; | |
566 | my_entry.eof = 0; | |
567 | my_entry.fh = &fh; | |
568 | my_entry.fattr = &fattr; | |
0e574af1 | 569 | nfs_fattr_init(&fattr); |
1da177e4 LT |
570 | desc->entry = &my_entry; |
571 | ||
572 | while(!desc->entry->eof) { | |
573 | res = readdir_search_pagecache(desc); | |
00a92642 | 574 | |
1da177e4 LT |
575 | if (res == -EBADCOOKIE) { |
576 | /* This means either end of directory */ | |
f0dd2136 | 577 | if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { |
1da177e4 LT |
578 | /* Or that the server has 'lost' a cookie */ |
579 | res = uncached_readdir(desc, dirent, filldir); | |
580 | if (res >= 0) | |
581 | continue; | |
582 | } | |
583 | res = 0; | |
584 | break; | |
585 | } | |
586 | if (res == -ETOOSMALL && desc->plus) { | |
412d582e | 587 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); |
1da177e4 LT |
588 | nfs_zap_caches(inode); |
589 | desc->plus = 0; | |
590 | desc->entry->eof = 0; | |
591 | continue; | |
592 | } | |
593 | if (res < 0) | |
594 | break; | |
595 | ||
596 | res = nfs_do_filldir(desc, dirent, filldir); | |
597 | if (res < 0) { | |
598 | res = 0; | |
599 | break; | |
600 | } | |
601 | } | |
602 | unlock_kernel(); | |
1e7cb3dc CL |
603 | if (res > 0) |
604 | res = 0; | |
605 | dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", | |
606 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
607 | res); | |
608 | return res; | |
1da177e4 LT |
609 | } |
610 | ||
10afec90 | 611 | static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
f0dd2136 | 612 | { |
01cce933 | 613 | mutex_lock(&filp->f_path.dentry->d_inode->i_mutex); |
f0dd2136 TM |
614 | switch (origin) { |
615 | case 1: | |
616 | offset += filp->f_pos; | |
617 | case 0: | |
618 | if (offset >= 0) | |
619 | break; | |
620 | default: | |
621 | offset = -EINVAL; | |
622 | goto out; | |
623 | } | |
624 | if (offset != filp->f_pos) { | |
625 | filp->f_pos = offset; | |
626 | ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0; | |
627 | } | |
628 | out: | |
01cce933 | 629 | mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex); |
f0dd2136 TM |
630 | return offset; |
631 | } | |
632 | ||
1da177e4 LT |
633 | /* |
634 | * All directory operations under NFS are synchronous, so fsync() | |
635 | * is a dummy operation. | |
636 | */ | |
10afec90 | 637 | static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) |
1da177e4 | 638 | { |
1e7cb3dc CL |
639 | dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", |
640 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
641 | datasync); | |
642 | ||
1da177e4 LT |
643 | return 0; |
644 | } | |
645 | ||
646 | /* | |
647 | * A check for whether or not the parent directory has changed. | |
648 | * In the case it has, we assume that the dentries are untrustworthy | |
649 | * and may need to be looked up again. | |
650 | */ | |
c79ba787 | 651 | static int nfs_check_verifier(struct inode *dir, struct dentry *dentry) |
1da177e4 | 652 | { |
e62c2bba TM |
653 | unsigned long verf; |
654 | ||
1da177e4 LT |
655 | if (IS_ROOT(dentry)) |
656 | return 1; | |
3062c532 | 657 | verf = dentry->d_time; |
e70c4908 | 658 | if (nfs_caches_unstable(dir) |
e62c2bba | 659 | || verf != NFS_I(dir)->cache_change_attribute) |
1da177e4 | 660 | return 0; |
e62c2bba | 661 | return 1; |
1da177e4 LT |
662 | } |
663 | ||
664 | static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf) | |
665 | { | |
3062c532 | 666 | dentry->d_time = verf; |
1da177e4 LT |
667 | } |
668 | ||
1d6757fb TM |
669 | /* |
670 | * Return the intent data that applies to this particular path component | |
671 | * | |
672 | * Note that the current set of intents only apply to the very last | |
673 | * component of the path. | |
674 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
675 | */ | |
676 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
677 | { | |
678 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
679 | return 0; | |
680 | return nd->flags & mask; | |
681 | } | |
682 | ||
683 | /* | |
684 | * Inode and filehandle revalidation for lookups. | |
685 | * | |
686 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
687 | * or if the intent information indicates that we're about to open this | |
688 | * particular file and the "nocto" mount flag is not set. | |
689 | * | |
690 | */ | |
1da177e4 LT |
691 | static inline |
692 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
693 | { | |
694 | struct nfs_server *server = NFS_SERVER(inode); | |
695 | ||
696 | if (nd != NULL) { | |
1da177e4 | 697 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 698 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
699 | goto out_force; |
700 | /* This is an open(2) */ | |
1d6757fb | 701 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
702 | !(server->flags & NFS_MOUNT_NOCTO) && |
703 | (S_ISREG(inode->i_mode) || | |
704 | S_ISDIR(inode->i_mode))) | |
1da177e4 LT |
705 | goto out_force; |
706 | } | |
707 | return nfs_revalidate_inode(server, inode); | |
708 | out_force: | |
709 | return __nfs_revalidate_inode(server, inode); | |
710 | } | |
711 | ||
712 | /* | |
713 | * We judge how long we want to trust negative | |
714 | * dentries by looking at the parent inode mtime. | |
715 | * | |
716 | * If parent mtime has changed, we revalidate, else we wait for a | |
717 | * period corresponding to the parent's attribute cache timeout value. | |
718 | */ | |
719 | static inline | |
720 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
721 | struct nameidata *nd) | |
722 | { | |
1da177e4 | 723 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 724 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 LT |
725 | return 0; |
726 | return !nfs_check_verifier(dir, dentry); | |
727 | } | |
728 | ||
729 | /* | |
730 | * This is called every time the dcache has a lookup hit, | |
731 | * and we should check whether we can really trust that | |
732 | * lookup. | |
733 | * | |
734 | * NOTE! The hit can be a negative hit too, don't assume | |
735 | * we have an inode! | |
736 | * | |
737 | * If the parent directory is seen to have changed, we throw out the | |
738 | * cached dentry and do a new lookup. | |
739 | */ | |
740 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
741 | { | |
742 | struct inode *dir; | |
743 | struct inode *inode; | |
744 | struct dentry *parent; | |
745 | int error; | |
746 | struct nfs_fh fhandle; | |
747 | struct nfs_fattr fattr; | |
748 | unsigned long verifier; | |
749 | ||
750 | parent = dget_parent(dentry); | |
751 | lock_kernel(); | |
752 | dir = parent->d_inode; | |
91d5b470 | 753 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
754 | inode = dentry->d_inode; |
755 | ||
e70c4908 TM |
756 | /* Revalidate parent directory attribute cache */ |
757 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
758 | goto out_zap_parent; | |
759 | ||
1da177e4 LT |
760 | if (!inode) { |
761 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
762 | goto out_bad; | |
763 | goto out_valid; | |
764 | } | |
765 | ||
766 | if (is_bad_inode(inode)) { | |
1e7cb3dc CL |
767 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
768 | __FUNCTION__, dentry->d_parent->d_name.name, | |
769 | dentry->d_name.name); | |
1da177e4 LT |
770 | goto out_bad; |
771 | } | |
772 | ||
1da177e4 LT |
773 | /* Force a full look up iff the parent directory has changed */ |
774 | if (nfs_check_verifier(dir, dentry)) { | |
775 | if (nfs_lookup_verify_inode(inode, nd)) | |
776 | goto out_zap_parent; | |
777 | goto out_valid; | |
778 | } | |
779 | ||
780 | if (NFS_STALE(inode)) | |
781 | goto out_bad; | |
782 | ||
783 | verifier = nfs_save_change_attribute(dir); | |
784 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); | |
785 | if (error) | |
786 | goto out_bad; | |
787 | if (nfs_compare_fh(NFS_FH(inode), &fhandle)) | |
788 | goto out_bad; | |
789 | if ((error = nfs_refresh_inode(inode, &fattr)) != 0) | |
790 | goto out_bad; | |
791 | ||
8fb559f8 | 792 | nfs_set_verifier(dentry, verifier); |
1da177e4 LT |
793 | out_valid: |
794 | unlock_kernel(); | |
795 | dput(parent); | |
1e7cb3dc CL |
796 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
797 | __FUNCTION__, dentry->d_parent->d_name.name, | |
798 | dentry->d_name.name); | |
1da177e4 LT |
799 | return 1; |
800 | out_zap_parent: | |
801 | nfs_zap_caches(dir); | |
802 | out_bad: | |
803 | NFS_CACHEINV(dir); | |
804 | if (inode && S_ISDIR(inode->i_mode)) { | |
805 | /* Purge readdir caches. */ | |
806 | nfs_zap_caches(inode); | |
807 | /* If we have submounts, don't unhash ! */ | |
808 | if (have_submounts(dentry)) | |
809 | goto out_valid; | |
810 | shrink_dcache_parent(dentry); | |
811 | } | |
812 | d_drop(dentry); | |
813 | unlock_kernel(); | |
814 | dput(parent); | |
1e7cb3dc CL |
815 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
816 | __FUNCTION__, dentry->d_parent->d_name.name, | |
817 | dentry->d_name.name); | |
1da177e4 LT |
818 | return 0; |
819 | } | |
820 | ||
821 | /* | |
822 | * This is called from dput() when d_count is going to 0. | |
823 | */ | |
824 | static int nfs_dentry_delete(struct dentry *dentry) | |
825 | { | |
826 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
827 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
828 | dentry->d_flags); | |
829 | ||
830 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
831 | /* Unhash it, so that ->d_iput() would be called */ | |
832 | return 1; | |
833 | } | |
834 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
835 | /* Unhash it, so that ancestors of killed async unlink | |
836 | * files will be cleaned up during umount */ | |
837 | return 1; | |
838 | } | |
839 | return 0; | |
840 | ||
841 | } | |
842 | ||
843 | /* | |
844 | * Called when the dentry loses inode. | |
845 | * We use it to clean up silly-renamed files. | |
846 | */ | |
847 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
848 | { | |
cae7a073 | 849 | nfs_inode_return_delegation(inode); |
83672d39 NB |
850 | if (S_ISDIR(inode->i_mode)) |
851 | /* drop any readdir cache as it could easily be old */ | |
852 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA; | |
853 | ||
1da177e4 LT |
854 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
855 | lock_kernel(); | |
9a53c3a7 | 856 | drop_nlink(inode); |
e4eff1a6 | 857 | nfs_complete_unlink(dentry, inode); |
1da177e4 LT |
858 | unlock_kernel(); |
859 | } | |
1da177e4 LT |
860 | iput(inode); |
861 | } | |
862 | ||
863 | struct dentry_operations nfs_dentry_operations = { | |
864 | .d_revalidate = nfs_lookup_revalidate, | |
865 | .d_delete = nfs_dentry_delete, | |
866 | .d_iput = nfs_dentry_iput, | |
867 | }; | |
868 | ||
1d6757fb TM |
869 | /* |
870 | * Use intent information to check whether or not we're going to do | |
871 | * an O_EXCL create using this path component. | |
872 | */ | |
1da177e4 LT |
873 | static inline |
874 | int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
875 | { | |
876 | if (NFS_PROTO(dir)->version == 2) | |
877 | return 0; | |
1d6757fb | 878 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) |
1da177e4 LT |
879 | return 0; |
880 | return (nd->intent.open.flags & O_EXCL) != 0; | |
881 | } | |
882 | ||
a0356862 | 883 | static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr) |
55a97593 TM |
884 | { |
885 | struct nfs_server *server = NFS_SERVER(dir); | |
886 | ||
887 | if (!nfs_fsid_equal(&server->fsid, &fattr->fsid)) | |
a0356862 TM |
888 | /* Revalidate fsid using the parent directory */ |
889 | return __nfs_revalidate_inode(server, dir); | |
55a97593 TM |
890 | return 0; |
891 | } | |
892 | ||
1da177e4 LT |
893 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
894 | { | |
895 | struct dentry *res; | |
896 | struct inode *inode = NULL; | |
897 | int error; | |
898 | struct nfs_fh fhandle; | |
899 | struct nfs_fattr fattr; | |
900 | ||
901 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
902 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 903 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
904 | |
905 | res = ERR_PTR(-ENAMETOOLONG); | |
906 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
907 | goto out; | |
908 | ||
909 | res = ERR_PTR(-ENOMEM); | |
910 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
911 | ||
912 | lock_kernel(); | |
1da177e4 | 913 | |
fd684071 TM |
914 | /* |
915 | * If we're doing an exclusive create, optimize away the lookup | |
916 | * but don't hash the dentry. | |
917 | */ | |
918 | if (nfs_is_exclusive_create(dir, nd)) { | |
919 | d_instantiate(dentry, NULL); | |
920 | res = NULL; | |
921 | goto out_unlock; | |
922 | } | |
1da177e4 LT |
923 | |
924 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); | |
925 | if (error == -ENOENT) | |
926 | goto no_entry; | |
927 | if (error < 0) { | |
928 | res = ERR_PTR(error); | |
929 | goto out_unlock; | |
930 | } | |
a0356862 | 931 | error = nfs_reval_fsid(dir, &fattr); |
55a97593 TM |
932 | if (error < 0) { |
933 | res = ERR_PTR(error); | |
934 | goto out_unlock; | |
935 | } | |
1da177e4 | 936 | inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); |
03f28e3a TM |
937 | res = (struct dentry *)inode; |
938 | if (IS_ERR(res)) | |
1da177e4 | 939 | goto out_unlock; |
54ceac45 | 940 | |
1da177e4 | 941 | no_entry: |
54ceac45 | 942 | res = d_materialise_unique(dentry, inode); |
9eaef27b | 943 | if (res != NULL) { |
fc22617e | 944 | struct dentry *parent; |
9eaef27b TM |
945 | if (IS_ERR(res)) |
946 | goto out_unlock; | |
fc22617e TM |
947 | /* Was a directory renamed! */ |
948 | parent = dget_parent(res); | |
949 | if (!IS_ROOT(parent)) | |
950 | nfs_mark_for_revalidate(parent->d_inode); | |
951 | dput(parent); | |
1da177e4 | 952 | dentry = res; |
9eaef27b | 953 | } |
1da177e4 LT |
954 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
955 | out_unlock: | |
956 | unlock_kernel(); | |
957 | out: | |
958 | return res; | |
959 | } | |
960 | ||
961 | #ifdef CONFIG_NFS_V4 | |
962 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
963 | ||
964 | struct dentry_operations nfs4_dentry_operations = { | |
965 | .d_revalidate = nfs_open_revalidate, | |
966 | .d_delete = nfs_dentry_delete, | |
967 | .d_iput = nfs_dentry_iput, | |
968 | }; | |
969 | ||
1d6757fb TM |
970 | /* |
971 | * Use intent information to determine whether we need to substitute | |
972 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
973 | */ | |
1da177e4 LT |
974 | static int is_atomic_open(struct inode *dir, struct nameidata *nd) |
975 | { | |
1d6757fb | 976 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
977 | return 0; |
978 | /* NFS does not (yet) have a stateful open for directories */ | |
979 | if (nd->flags & LOOKUP_DIRECTORY) | |
980 | return 0; | |
981 | /* Are we trying to write to a read only partition? */ | |
982 | if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
983 | return 0; | |
984 | return 1; | |
985 | } | |
986 | ||
987 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) | |
988 | { | |
989 | struct dentry *res = NULL; | |
1da177e4 LT |
990 | int error; |
991 | ||
1e7cb3dc CL |
992 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
993 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
994 | ||
1da177e4 LT |
995 | /* Check that we are indeed trying to open this file */ |
996 | if (!is_atomic_open(dir, nd)) | |
997 | goto no_open; | |
998 | ||
999 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
1000 | res = ERR_PTR(-ENAMETOOLONG); | |
1001 | goto out; | |
1002 | } | |
1003 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1004 | ||
1005 | /* Let vfs_create() deal with O_EXCL */ | |
02a913a7 TM |
1006 | if (nd->intent.open.flags & O_EXCL) { |
1007 | d_add(dentry, NULL); | |
1008 | goto out; | |
1009 | } | |
1da177e4 LT |
1010 | |
1011 | /* Open the file on the server */ | |
1012 | lock_kernel(); | |
1013 | /* Revalidate parent directory attribute cache */ | |
1014 | error = nfs_revalidate_inode(NFS_SERVER(dir), dir); | |
1015 | if (error < 0) { | |
1016 | res = ERR_PTR(error); | |
01c314a0 | 1017 | unlock_kernel(); |
1da177e4 LT |
1018 | goto out; |
1019 | } | |
1020 | ||
1021 | if (nd->intent.open.flags & O_CREAT) { | |
1022 | nfs_begin_data_update(dir); | |
02a913a7 | 1023 | res = nfs4_atomic_open(dir, dentry, nd); |
1da177e4 LT |
1024 | nfs_end_data_update(dir); |
1025 | } else | |
02a913a7 | 1026 | res = nfs4_atomic_open(dir, dentry, nd); |
1da177e4 | 1027 | unlock_kernel(); |
02a913a7 TM |
1028 | if (IS_ERR(res)) { |
1029 | error = PTR_ERR(res); | |
1da177e4 LT |
1030 | switch (error) { |
1031 | /* Make a negative dentry */ | |
1032 | case -ENOENT: | |
02a913a7 TM |
1033 | res = NULL; |
1034 | goto out; | |
1da177e4 | 1035 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1036 | case -EISDIR: |
1037 | case -ENOTDIR: | |
1038 | goto no_open; | |
1da177e4 LT |
1039 | case -ELOOP: |
1040 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1041 | goto no_open; | |
1da177e4 LT |
1042 | /* case -EINVAL: */ |
1043 | default: | |
1da177e4 LT |
1044 | goto out; |
1045 | } | |
02a913a7 | 1046 | } else if (res != NULL) |
1da177e4 | 1047 | dentry = res; |
1da177e4 LT |
1048 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1049 | out: | |
1050 | return res; | |
1051 | no_open: | |
1052 | return nfs_lookup(dir, dentry, nd); | |
1053 | } | |
1054 | ||
1055 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1056 | { | |
1057 | struct dentry *parent = NULL; | |
1058 | struct inode *inode = dentry->d_inode; | |
1059 | struct inode *dir; | |
1060 | unsigned long verifier; | |
1061 | int openflags, ret = 0; | |
1062 | ||
1063 | parent = dget_parent(dentry); | |
1064 | dir = parent->d_inode; | |
1065 | if (!is_atomic_open(dir, nd)) | |
1066 | goto no_open; | |
1067 | /* We can't create new files in nfs_open_revalidate(), so we | |
1068 | * optimize away revalidation of negative dentries. | |
1069 | */ | |
1070 | if (inode == NULL) | |
1071 | goto out; | |
1072 | /* NFS only supports OPEN on regular files */ | |
1073 | if (!S_ISREG(inode->i_mode)) | |
1074 | goto no_open; | |
1075 | openflags = nd->intent.open.flags; | |
1076 | /* We cannot do exclusive creation on a positive dentry */ | |
1077 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
1078 | goto no_open; | |
1079 | /* We can't create new files, or truncate existing ones here */ | |
1080 | openflags &= ~(O_CREAT|O_TRUNC); | |
1081 | ||
1082 | /* | |
1b1dcc1b | 1083 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1084 | * operations that change the directory. We therefore save the |
1085 | * change attribute *before* we do the RPC call. | |
1086 | */ | |
1087 | lock_kernel(); | |
1088 | verifier = nfs_save_change_attribute(dir); | |
02a913a7 | 1089 | ret = nfs4_open_revalidate(dir, dentry, openflags, nd); |
1da177e4 | 1090 | if (!ret) |
8fb559f8 | 1091 | nfs_set_verifier(dentry, verifier); |
1da177e4 LT |
1092 | unlock_kernel(); |
1093 | out: | |
1094 | dput(parent); | |
1095 | if (!ret) | |
1096 | d_drop(dentry); | |
1097 | return ret; | |
1098 | no_open: | |
1099 | dput(parent); | |
1100 | if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) | |
1101 | return 1; | |
1102 | return nfs_lookup_revalidate(dentry, nd); | |
1103 | } | |
1104 | #endif /* CONFIG_NFSV4 */ | |
1105 | ||
1106 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) | |
1107 | { | |
01cce933 | 1108 | struct dentry *parent = desc->file->f_path.dentry; |
1da177e4 LT |
1109 | struct inode *dir = parent->d_inode; |
1110 | struct nfs_entry *entry = desc->entry; | |
1111 | struct dentry *dentry, *alias; | |
1112 | struct qstr name = { | |
1113 | .name = entry->name, | |
1114 | .len = entry->len, | |
1115 | }; | |
1116 | struct inode *inode; | |
1117 | ||
1118 | switch (name.len) { | |
1119 | case 2: | |
1120 | if (name.name[0] == '.' && name.name[1] == '.') | |
1121 | return dget_parent(parent); | |
1122 | break; | |
1123 | case 1: | |
1124 | if (name.name[0] == '.') | |
1125 | return dget(parent); | |
1126 | } | |
1127 | name.hash = full_name_hash(name.name, name.len); | |
1128 | dentry = d_lookup(parent, &name); | |
df1d5d23 | 1129 | if (dentry != NULL) { |
ef75c797 TM |
1130 | /* Is this a positive dentry that matches the readdir info? */ |
1131 | if (dentry->d_inode != NULL && | |
1132 | (NFS_FILEID(dentry->d_inode) == entry->ino || | |
1133 | d_mountpoint(dentry))) { | |
1134 | if (!desc->plus || entry->fh->size == 0) | |
1135 | return dentry; | |
1136 | if (nfs_compare_fh(NFS_FH(dentry->d_inode), | |
1137 | entry->fh) == 0) | |
1138 | goto out_renew; | |
1139 | } | |
df1d5d23 TM |
1140 | /* No, so d_drop to allow one to be created */ |
1141 | d_drop(dentry); | |
1142 | dput(dentry); | |
1143 | } | |
1da177e4 LT |
1144 | if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) |
1145 | return NULL; | |
54af3bb5 TM |
1146 | if (name.len > NFS_SERVER(dir)->namelen) |
1147 | return NULL; | |
1b1dcc1b | 1148 | /* Note: caller is already holding the dir->i_mutex! */ |
1da177e4 LT |
1149 | dentry = d_alloc(parent, &name); |
1150 | if (dentry == NULL) | |
1151 | return NULL; | |
1152 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1153 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
03f28e3a | 1154 | if (IS_ERR(inode)) { |
1da177e4 LT |
1155 | dput(dentry); |
1156 | return NULL; | |
1157 | } | |
54ceac45 DH |
1158 | |
1159 | alias = d_materialise_unique(dentry, inode); | |
1da177e4 LT |
1160 | if (alias != NULL) { |
1161 | dput(dentry); | |
9eaef27b TM |
1162 | if (IS_ERR(alias)) |
1163 | return NULL; | |
1da177e4 LT |
1164 | dentry = alias; |
1165 | } | |
54ceac45 | 1166 | |
c79ba787 | 1167 | out_renew: |
8fb559f8 | 1168 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
c79ba787 | 1169 | return dentry; |
1da177e4 LT |
1170 | } |
1171 | ||
1172 | /* | |
1173 | * Code common to create, mkdir, and mknod. | |
1174 | */ | |
1175 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1176 | struct nfs_fattr *fattr) | |
1177 | { | |
1178 | struct inode *inode; | |
1179 | int error = -EACCES; | |
1180 | ||
1181 | /* We may have been initialized further down */ | |
1182 | if (dentry->d_inode) | |
1183 | return 0; | |
1184 | if (fhandle->size == 0) { | |
1185 | struct inode *dir = dentry->d_parent->d_inode; | |
1186 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1187 | if (error) | |
4f390c15 | 1188 | return error; |
1da177e4 LT |
1189 | } |
1190 | if (!(fattr->valid & NFS_ATTR_FATTR)) { | |
1191 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1192 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1193 | if (error < 0) |
4f390c15 | 1194 | return error; |
1da177e4 | 1195 | } |
1da177e4 | 1196 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1197 | error = PTR_ERR(inode); |
1198 | if (IS_ERR(inode)) | |
4f390c15 | 1199 | return error; |
1da177e4 | 1200 | d_instantiate(dentry, inode); |
fd684071 TM |
1201 | if (d_unhashed(dentry)) |
1202 | d_rehash(dentry); | |
1da177e4 | 1203 | return 0; |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /* | |
1207 | * Following a failed create operation, we drop the dentry rather | |
1208 | * than retain a negative dentry. This avoids a problem in the event | |
1209 | * that the operation succeeded on the server, but an error in the | |
1210 | * reply path made it appear to have failed. | |
1211 | */ | |
1212 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1213 | struct nameidata *nd) | |
1214 | { | |
1215 | struct iattr attr; | |
1216 | int error; | |
1217 | int open_flags = 0; | |
1218 | ||
1e7cb3dc CL |
1219 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1220 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1221 | |
1222 | attr.ia_mode = mode; | |
1223 | attr.ia_valid = ATTR_MODE; | |
1224 | ||
ad389da7 | 1225 | if ((nd->flags & LOOKUP_CREATE) != 0) |
1da177e4 LT |
1226 | open_flags = nd->intent.open.flags; |
1227 | ||
1228 | lock_kernel(); | |
1229 | nfs_begin_data_update(dir); | |
02a913a7 | 1230 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); |
1da177e4 LT |
1231 | nfs_end_data_update(dir); |
1232 | if (error != 0) | |
1233 | goto out_err; | |
1da177e4 LT |
1234 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1235 | unlock_kernel(); | |
1236 | return 0; | |
1237 | out_err: | |
1238 | unlock_kernel(); | |
1239 | d_drop(dentry); | |
1240 | return error; | |
1241 | } | |
1242 | ||
1243 | /* | |
1244 | * See comments for nfs_proc_create regarding failed operations. | |
1245 | */ | |
1246 | static int | |
1247 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1248 | { | |
1249 | struct iattr attr; | |
1250 | int status; | |
1251 | ||
1e7cb3dc CL |
1252 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1253 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1254 | |
1255 | if (!new_valid_dev(rdev)) | |
1256 | return -EINVAL; | |
1257 | ||
1258 | attr.ia_mode = mode; | |
1259 | attr.ia_valid = ATTR_MODE; | |
1260 | ||
1261 | lock_kernel(); | |
1262 | nfs_begin_data_update(dir); | |
1263 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); | |
1264 | nfs_end_data_update(dir); | |
1265 | if (status != 0) | |
1266 | goto out_err; | |
1da177e4 LT |
1267 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1268 | unlock_kernel(); | |
1269 | return 0; | |
1270 | out_err: | |
1271 | unlock_kernel(); | |
1272 | d_drop(dentry); | |
1273 | return status; | |
1274 | } | |
1275 | ||
1276 | /* | |
1277 | * See comments for nfs_proc_create regarding failed operations. | |
1278 | */ | |
1279 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1280 | { | |
1281 | struct iattr attr; | |
1282 | int error; | |
1283 | ||
1e7cb3dc CL |
1284 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1285 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1286 | |
1287 | attr.ia_valid = ATTR_MODE; | |
1288 | attr.ia_mode = mode | S_IFDIR; | |
1289 | ||
1290 | lock_kernel(); | |
1291 | nfs_begin_data_update(dir); | |
1292 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); | |
1293 | nfs_end_data_update(dir); | |
1294 | if (error != 0) | |
1295 | goto out_err; | |
1da177e4 LT |
1296 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1297 | unlock_kernel(); | |
1298 | return 0; | |
1299 | out_err: | |
1300 | d_drop(dentry); | |
1301 | unlock_kernel(); | |
1302 | return error; | |
1303 | } | |
1304 | ||
1305 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) | |
1306 | { | |
1307 | int error; | |
1308 | ||
1e7cb3dc CL |
1309 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1310 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1311 | |
1312 | lock_kernel(); | |
1313 | nfs_begin_data_update(dir); | |
1314 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1315 | /* Ensure the VFS deletes this inode */ | |
1316 | if (error == 0 && dentry->d_inode != NULL) | |
ce71ec36 | 1317 | clear_nlink(dentry->d_inode); |
1da177e4 LT |
1318 | nfs_end_data_update(dir); |
1319 | unlock_kernel(); | |
1320 | ||
1321 | return error; | |
1322 | } | |
1323 | ||
1324 | static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) | |
1325 | { | |
1326 | static unsigned int sillycounter; | |
4e769b93 | 1327 | const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2; |
1da177e4 | 1328 | const int countersize = sizeof(sillycounter)*2; |
4e769b93 | 1329 | const int slen = sizeof(".nfs")+fileidsize+countersize-1; |
1da177e4 LT |
1330 | char silly[slen+1]; |
1331 | struct qstr qsilly; | |
1332 | struct dentry *sdentry; | |
1333 | int error = -EIO; | |
1334 | ||
1335 | dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", | |
1336 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1337 | atomic_read(&dentry->d_count)); | |
91d5b470 | 1338 | nfs_inc_stats(dir, NFSIOS_SILLYRENAME); |
1da177e4 | 1339 | |
1da177e4 LT |
1340 | /* |
1341 | * We don't allow a dentry to be silly-renamed twice. | |
1342 | */ | |
1343 | error = -EBUSY; | |
1344 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1345 | goto out; | |
1346 | ||
4e769b93 PS |
1347 | sprintf(silly, ".nfs%*.*Lx", |
1348 | fileidsize, fileidsize, | |
1349 | (unsigned long long)NFS_FILEID(dentry->d_inode)); | |
1da177e4 | 1350 | |
34ea8188 TM |
1351 | /* Return delegation in anticipation of the rename */ |
1352 | nfs_inode_return_delegation(dentry->d_inode); | |
1353 | ||
1da177e4 LT |
1354 | sdentry = NULL; |
1355 | do { | |
1356 | char *suffix = silly + slen - countersize; | |
1357 | ||
1358 | dput(sdentry); | |
1359 | sillycounter++; | |
1360 | sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); | |
1361 | ||
1e7cb3dc CL |
1362 | dfprintk(VFS, "NFS: trying to rename %s to %s\n", |
1363 | dentry->d_name.name, silly); | |
1da177e4 LT |
1364 | |
1365 | sdentry = lookup_one_len(silly, dentry->d_parent, slen); | |
1366 | /* | |
1367 | * N.B. Better to return EBUSY here ... it could be | |
1368 | * dangerous to delete the file while it's in use. | |
1369 | */ | |
1370 | if (IS_ERR(sdentry)) | |
1371 | goto out; | |
1372 | } while(sdentry->d_inode != NULL); /* need negative lookup */ | |
1373 | ||
1374 | qsilly.name = silly; | |
1375 | qsilly.len = strlen(silly); | |
1376 | nfs_begin_data_update(dir); | |
1377 | if (dentry->d_inode) { | |
1378 | nfs_begin_data_update(dentry->d_inode); | |
1379 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, | |
1380 | dir, &qsilly); | |
5ba7cc48 | 1381 | nfs_mark_for_revalidate(dentry->d_inode); |
1da177e4 LT |
1382 | nfs_end_data_update(dentry->d_inode); |
1383 | } else | |
1384 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, | |
1385 | dir, &qsilly); | |
1386 | nfs_end_data_update(dir); | |
1387 | if (!error) { | |
1da177e4 LT |
1388 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1389 | d_move(dentry, sdentry); | |
e4eff1a6 | 1390 | error = nfs_async_unlink(dir, dentry); |
1da177e4 LT |
1391 | /* If we return 0 we don't unlink */ |
1392 | } | |
1393 | dput(sdentry); | |
1394 | out: | |
1395 | return error; | |
1396 | } | |
1397 | ||
1398 | /* | |
1399 | * Remove a file after making sure there are no pending writes, | |
1400 | * and after checking that the file has only one user. | |
1401 | * | |
1402 | * We invalidate the attribute cache and free the inode prior to the operation | |
1403 | * to avoid possible races if the server reuses the inode. | |
1404 | */ | |
1405 | static int nfs_safe_remove(struct dentry *dentry) | |
1406 | { | |
1407 | struct inode *dir = dentry->d_parent->d_inode; | |
1408 | struct inode *inode = dentry->d_inode; | |
1409 | int error = -EBUSY; | |
1410 | ||
1411 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1412 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1413 | ||
1414 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1415 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1416 | error = 0; | |
1417 | goto out; | |
1418 | } | |
1419 | ||
1420 | nfs_begin_data_update(dir); | |
1421 | if (inode != NULL) { | |
cae7a073 | 1422 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1423 | nfs_begin_data_update(inode); |
1424 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1425 | /* The VFS may want to delete this inode */ | |
1426 | if (error == 0) | |
9a53c3a7 | 1427 | drop_nlink(inode); |
5ba7cc48 | 1428 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1429 | nfs_end_data_update(inode); |
1430 | } else | |
1431 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1432 | nfs_end_data_update(dir); | |
1433 | out: | |
1434 | return error; | |
1435 | } | |
1436 | ||
1437 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1438 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1439 | * | |
1440 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1441 | */ | |
1442 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1443 | { | |
1444 | int error; | |
1445 | int need_rehash = 0; | |
1446 | ||
1447 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1448 | dir->i_ino, dentry->d_name.name); | |
1449 | ||
1450 | lock_kernel(); | |
1451 | spin_lock(&dcache_lock); | |
1452 | spin_lock(&dentry->d_lock); | |
1453 | if (atomic_read(&dentry->d_count) > 1) { | |
1454 | spin_unlock(&dentry->d_lock); | |
1455 | spin_unlock(&dcache_lock); | |
ccfeb506 TM |
1456 | /* Start asynchronous writeout of the inode */ |
1457 | write_inode_now(dentry->d_inode, 0); | |
1da177e4 LT |
1458 | error = nfs_sillyrename(dir, dentry); |
1459 | unlock_kernel(); | |
1460 | return error; | |
1461 | } | |
1462 | if (!d_unhashed(dentry)) { | |
1463 | __d_drop(dentry); | |
1464 | need_rehash = 1; | |
1465 | } | |
1466 | spin_unlock(&dentry->d_lock); | |
1467 | spin_unlock(&dcache_lock); | |
1468 | error = nfs_safe_remove(dentry); | |
1469 | if (!error) { | |
1da177e4 LT |
1470 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); |
1471 | } else if (need_rehash) | |
1472 | d_rehash(dentry); | |
1473 | unlock_kernel(); | |
1474 | return error; | |
1475 | } | |
1476 | ||
873101b3 CL |
1477 | /* |
1478 | * To create a symbolic link, most file systems instantiate a new inode, | |
1479 | * add a page to it containing the path, then write it out to the disk | |
1480 | * using prepare_write/commit_write. | |
1481 | * | |
1482 | * Unfortunately the NFS client can't create the in-core inode first | |
1483 | * because it needs a file handle to create an in-core inode (see | |
1484 | * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the | |
1485 | * symlink request has completed on the server. | |
1486 | * | |
1487 | * So instead we allocate a raw page, copy the symname into it, then do | |
1488 | * the SYMLINK request with the page as the buffer. If it succeeds, we | |
1489 | * now have a new file handle and can instantiate an in-core NFS inode | |
1490 | * and move the raw page into its mapping. | |
1491 | */ | |
1492 | static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1da177e4 | 1493 | { |
873101b3 CL |
1494 | struct pagevec lru_pvec; |
1495 | struct page *page; | |
1496 | char *kaddr; | |
1da177e4 | 1497 | struct iattr attr; |
873101b3 | 1498 | unsigned int pathlen = strlen(symname); |
1da177e4 LT |
1499 | int error; |
1500 | ||
1501 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1502 | dir->i_ino, dentry->d_name.name, symname); | |
1503 | ||
873101b3 CL |
1504 | if (pathlen > PAGE_SIZE) |
1505 | return -ENAMETOOLONG; | |
1da177e4 | 1506 | |
873101b3 CL |
1507 | attr.ia_mode = S_IFLNK | S_IRWXUGO; |
1508 | attr.ia_valid = ATTR_MODE; | |
1da177e4 LT |
1509 | |
1510 | lock_kernel(); | |
873101b3 | 1511 | |
83d93f22 | 1512 | page = alloc_page(GFP_HIGHUSER); |
873101b3 CL |
1513 | if (!page) { |
1514 | unlock_kernel(); | |
1515 | return -ENOMEM; | |
1516 | } | |
1517 | ||
1518 | kaddr = kmap_atomic(page, KM_USER0); | |
1519 | memcpy(kaddr, symname, pathlen); | |
1520 | if (pathlen < PAGE_SIZE) | |
1521 | memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen); | |
1522 | kunmap_atomic(kaddr, KM_USER0); | |
1523 | ||
1da177e4 | 1524 | nfs_begin_data_update(dir); |
94a6d753 | 1525 | error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr); |
1da177e4 | 1526 | nfs_end_data_update(dir); |
873101b3 CL |
1527 | if (error != 0) { |
1528 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n", | |
1529 | dir->i_sb->s_id, dir->i_ino, | |
1530 | dentry->d_name.name, symname, error); | |
1da177e4 | 1531 | d_drop(dentry); |
873101b3 CL |
1532 | __free_page(page); |
1533 | unlock_kernel(); | |
1534 | return error; | |
1535 | } | |
1536 | ||
1537 | /* | |
1538 | * No big deal if we can't add this page to the page cache here. | |
1539 | * READLINK will get the missing page from the server if needed. | |
1540 | */ | |
1541 | pagevec_init(&lru_pvec, 0); | |
1542 | if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0, | |
1543 | GFP_KERNEL)) { | |
39cf8a13 CL |
1544 | pagevec_add(&lru_pvec, page); |
1545 | pagevec_lru_add(&lru_pvec); | |
873101b3 CL |
1546 | SetPageUptodate(page); |
1547 | unlock_page(page); | |
1548 | } else | |
1549 | __free_page(page); | |
1550 | ||
1da177e4 | 1551 | unlock_kernel(); |
873101b3 | 1552 | return 0; |
1da177e4 LT |
1553 | } |
1554 | ||
1555 | static int | |
1556 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1557 | { | |
1558 | struct inode *inode = old_dentry->d_inode; | |
1559 | int error; | |
1560 | ||
1561 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1562 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1563 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1564 | ||
1da177e4 | 1565 | lock_kernel(); |
1da177e4 LT |
1566 | nfs_begin_data_update(dir); |
1567 | nfs_begin_data_update(inode); | |
1568 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); | |
cf809556 TM |
1569 | if (error == 0) { |
1570 | atomic_inc(&inode->i_count); | |
1571 | d_instantiate(dentry, inode); | |
1572 | } | |
1da177e4 LT |
1573 | nfs_end_data_update(inode); |
1574 | nfs_end_data_update(dir); | |
1575 | unlock_kernel(); | |
1576 | return error; | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * RENAME | |
1581 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1582 | * different file handle for the same inode after a rename (e.g. when | |
1583 | * moving to a different directory). A fail-safe method to do so would | |
1584 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1585 | * rename the old file using the sillyrename stuff. This way, the original | |
1586 | * file in old_dir will go away when the last process iput()s the inode. | |
1587 | * | |
1588 | * FIXED. | |
1589 | * | |
1590 | * It actually works quite well. One needs to have the possibility for | |
1591 | * at least one ".nfs..." file in each directory the file ever gets | |
1592 | * moved or linked to which happens automagically with the new | |
1593 | * implementation that only depends on the dcache stuff instead of | |
1594 | * using the inode layer | |
1595 | * | |
1596 | * Unfortunately, things are a little more complicated than indicated | |
1597 | * above. For a cross-directory move, we want to make sure we can get | |
1598 | * rid of the old inode after the operation. This means there must be | |
1599 | * no pending writes (if it's a file), and the use count must be 1. | |
1600 | * If these conditions are met, we can drop the dentries before doing | |
1601 | * the rename. | |
1602 | */ | |
1603 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1604 | struct inode *new_dir, struct dentry *new_dentry) | |
1605 | { | |
1606 | struct inode *old_inode = old_dentry->d_inode; | |
1607 | struct inode *new_inode = new_dentry->d_inode; | |
1608 | struct dentry *dentry = NULL, *rehash = NULL; | |
1609 | int error = -EBUSY; | |
1610 | ||
1611 | /* | |
1612 | * To prevent any new references to the target during the rename, | |
1613 | * we unhash the dentry and free the inode in advance. | |
1614 | */ | |
1615 | lock_kernel(); | |
1616 | if (!d_unhashed(new_dentry)) { | |
1617 | d_drop(new_dentry); | |
1618 | rehash = new_dentry; | |
1619 | } | |
1620 | ||
1621 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", | |
1622 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1623 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1624 | atomic_read(&new_dentry->d_count)); | |
1625 | ||
1626 | /* | |
1627 | * First check whether the target is busy ... we can't | |
1628 | * safely do _any_ rename if the target is in use. | |
1629 | * | |
1630 | * For files, make a copy of the dentry and then do a | |
1631 | * silly-rename. If the silly-rename succeeds, the | |
1632 | * copied dentry is hashed and becomes the new target. | |
1633 | */ | |
1634 | if (!new_inode) | |
1635 | goto go_ahead; | |
6fe43f9e TM |
1636 | if (S_ISDIR(new_inode->i_mode)) { |
1637 | error = -EISDIR; | |
1638 | if (!S_ISDIR(old_inode->i_mode)) | |
1639 | goto out; | |
1640 | } else if (atomic_read(&new_dentry->d_count) > 2) { | |
1da177e4 LT |
1641 | int err; |
1642 | /* copy the target dentry's name */ | |
1643 | dentry = d_alloc(new_dentry->d_parent, | |
1644 | &new_dentry->d_name); | |
1645 | if (!dentry) | |
1646 | goto out; | |
1647 | ||
1648 | /* silly-rename the existing target ... */ | |
1649 | err = nfs_sillyrename(new_dir, new_dentry); | |
1650 | if (!err) { | |
1651 | new_dentry = rehash = dentry; | |
1652 | new_inode = NULL; | |
1653 | /* instantiate the replacement target */ | |
1654 | d_instantiate(new_dentry, NULL); | |
7a13e932 JJ |
1655 | } else if (atomic_read(&new_dentry->d_count) > 1) |
1656 | /* dentry still busy? */ | |
1da177e4 | 1657 | goto out; |
20509f1b | 1658 | } else |
9a53c3a7 | 1659 | drop_nlink(new_inode); |
1da177e4 LT |
1660 | |
1661 | go_ahead: | |
1662 | /* | |
1663 | * ... prune child dentries and writebacks if needed. | |
1664 | */ | |
1665 | if (atomic_read(&old_dentry->d_count) > 1) { | |
e1552e19 TM |
1666 | if (S_ISREG(old_inode->i_mode)) |
1667 | nfs_wb_all(old_inode); | |
1da177e4 LT |
1668 | shrink_dcache_parent(old_dentry); |
1669 | } | |
cae7a073 | 1670 | nfs_inode_return_delegation(old_inode); |
1da177e4 | 1671 | |
24174119 TM |
1672 | if (new_inode != NULL) { |
1673 | nfs_inode_return_delegation(new_inode); | |
1da177e4 | 1674 | d_delete(new_dentry); |
24174119 | 1675 | } |
1da177e4 LT |
1676 | |
1677 | nfs_begin_data_update(old_dir); | |
1678 | nfs_begin_data_update(new_dir); | |
1679 | nfs_begin_data_update(old_inode); | |
1680 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, | |
1681 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1682 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1683 | nfs_end_data_update(old_inode); |
1684 | nfs_end_data_update(new_dir); | |
1685 | nfs_end_data_update(old_dir); | |
1686 | out: | |
1687 | if (rehash) | |
1688 | d_rehash(rehash); | |
1689 | if (!error) { | |
349457cc | 1690 | d_move(old_dentry, new_dentry); |
8fb559f8 CL |
1691 | nfs_set_verifier(new_dentry, |
1692 | nfs_save_change_attribute(new_dir)); | |
1da177e4 LT |
1693 | } |
1694 | ||
1695 | /* new dentry created? */ | |
1696 | if (dentry) | |
1697 | dput(dentry); | |
1698 | unlock_kernel(); | |
1699 | return error; | |
1700 | } | |
1701 | ||
cfcea3e8 TM |
1702 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1703 | static LIST_HEAD(nfs_access_lru_list); | |
1704 | static atomic_long_t nfs_access_nr_entries; | |
1705 | ||
1c3c07e9 TM |
1706 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1707 | { | |
1708 | put_rpccred(entry->cred); | |
1709 | kfree(entry); | |
cfcea3e8 TM |
1710 | smp_mb__before_atomic_dec(); |
1711 | atomic_long_dec(&nfs_access_nr_entries); | |
1712 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1713 | } |
1714 | ||
979df72e TM |
1715 | int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) |
1716 | { | |
1717 | LIST_HEAD(head); | |
1718 | struct nfs_inode *nfsi; | |
1719 | struct nfs_access_entry *cache; | |
1720 | ||
979df72e | 1721 | restart: |
a50f7951 | 1722 | spin_lock(&nfs_access_lru_lock); |
979df72e TM |
1723 | list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { |
1724 | struct inode *inode; | |
1725 | ||
1726 | if (nr_to_scan-- == 0) | |
1727 | break; | |
1728 | inode = igrab(&nfsi->vfs_inode); | |
1729 | if (inode == NULL) | |
1730 | continue; | |
1731 | spin_lock(&inode->i_lock); | |
1732 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1733 | goto remove_lru_entry; | |
1734 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1735 | struct nfs_access_entry, lru); | |
1736 | list_move(&cache->lru, &head); | |
1737 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1738 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1739 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1740 | &nfs_access_lru_list); | |
1741 | else { | |
1742 | remove_lru_entry: | |
1743 | list_del_init(&nfsi->access_cache_inode_lru); | |
1744 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); | |
1745 | } | |
1746 | spin_unlock(&inode->i_lock); | |
a50f7951 | 1747 | spin_unlock(&nfs_access_lru_lock); |
979df72e TM |
1748 | iput(inode); |
1749 | goto restart; | |
1750 | } | |
1751 | spin_unlock(&nfs_access_lru_lock); | |
1752 | while (!list_empty(&head)) { | |
1753 | cache = list_entry(head.next, struct nfs_access_entry, lru); | |
1754 | list_del(&cache->lru); | |
1755 | nfs_access_free_entry(cache); | |
1756 | } | |
1757 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; | |
1758 | } | |
1759 | ||
1c3c07e9 | 1760 | static void __nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 1761 | { |
55296809 | 1762 | struct nfs_inode *nfsi = NFS_I(inode); |
1c3c07e9 TM |
1763 | struct rb_root *root_node = &nfsi->access_cache; |
1764 | struct rb_node *n, *dispose = NULL; | |
1765 | struct nfs_access_entry *entry; | |
1766 | ||
1767 | /* Unhook entries from the cache */ | |
1768 | while ((n = rb_first(root_node)) != NULL) { | |
1769 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
1770 | rb_erase(n, root_node); | |
cfcea3e8 | 1771 | list_del(&entry->lru); |
1c3c07e9 TM |
1772 | n->rb_left = dispose; |
1773 | dispose = n; | |
1774 | } | |
1775 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1776 | spin_unlock(&inode->i_lock); | |
1da177e4 | 1777 | |
1c3c07e9 TM |
1778 | /* Now kill them all! */ |
1779 | while (dispose != NULL) { | |
1780 | n = dispose; | |
1781 | dispose = n->rb_left; | |
1782 | nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); | |
1783 | } | |
1da177e4 LT |
1784 | } |
1785 | ||
1c3c07e9 | 1786 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 1787 | { |
cfcea3e8 TM |
1788 | /* Remove from global LRU init */ |
1789 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { | |
1790 | spin_lock(&nfs_access_lru_lock); | |
1791 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); | |
1792 | spin_unlock(&nfs_access_lru_lock); | |
1793 | } | |
1794 | ||
1c3c07e9 TM |
1795 | spin_lock(&inode->i_lock); |
1796 | /* This will release the spinlock */ | |
1797 | __nfs_access_zap_cache(inode); | |
1798 | } | |
1da177e4 | 1799 | |
1c3c07e9 TM |
1800 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
1801 | { | |
1802 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
1803 | struct nfs_access_entry *entry; | |
1804 | ||
1805 | while (n != NULL) { | |
1806 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
1807 | ||
1808 | if (cred < entry->cred) | |
1809 | n = n->rb_left; | |
1810 | else if (cred > entry->cred) | |
1811 | n = n->rb_right; | |
1812 | else | |
1813 | return entry; | |
1da177e4 | 1814 | } |
1c3c07e9 TM |
1815 | return NULL; |
1816 | } | |
1817 | ||
1818 | int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) | |
1819 | { | |
1820 | struct nfs_inode *nfsi = NFS_I(inode); | |
1821 | struct nfs_access_entry *cache; | |
1822 | int err = -ENOENT; | |
1823 | ||
dc59250c | 1824 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
1825 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
1826 | goto out_zap; | |
1827 | cache = nfs_access_search_rbtree(inode, cred); | |
1828 | if (cache == NULL) | |
1829 | goto out; | |
c7e15961 | 1830 | if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) |
1c3c07e9 TM |
1831 | goto out_stale; |
1832 | res->jiffies = cache->jiffies; | |
1833 | res->cred = cache->cred; | |
1834 | res->mask = cache->mask; | |
cfcea3e8 | 1835 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
1836 | err = 0; |
1837 | out: | |
1838 | spin_unlock(&inode->i_lock); | |
1839 | return err; | |
1840 | out_stale: | |
1841 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 1842 | list_del(&cache->lru); |
1c3c07e9 TM |
1843 | spin_unlock(&inode->i_lock); |
1844 | nfs_access_free_entry(cache); | |
1845 | return -ENOENT; | |
1846 | out_zap: | |
1847 | /* This will release the spinlock */ | |
1848 | __nfs_access_zap_cache(inode); | |
1849 | return -ENOENT; | |
1850 | } | |
1851 | ||
1852 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
1853 | { | |
cfcea3e8 TM |
1854 | struct nfs_inode *nfsi = NFS_I(inode); |
1855 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
1856 | struct rb_node **p = &root_node->rb_node; |
1857 | struct rb_node *parent = NULL; | |
1858 | struct nfs_access_entry *entry; | |
1859 | ||
1860 | spin_lock(&inode->i_lock); | |
1861 | while (*p != NULL) { | |
1862 | parent = *p; | |
1863 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
1864 | ||
1865 | if (set->cred < entry->cred) | |
1866 | p = &parent->rb_left; | |
1867 | else if (set->cred > entry->cred) | |
1868 | p = &parent->rb_right; | |
1869 | else | |
1870 | goto found; | |
1871 | } | |
1872 | rb_link_node(&set->rb_node, parent, p); | |
1873 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 1874 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 1875 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
1876 | return; |
1877 | found: | |
1878 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
1879 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
1880 | list_del(&entry->lru); | |
1c3c07e9 TM |
1881 | spin_unlock(&inode->i_lock); |
1882 | nfs_access_free_entry(entry); | |
1883 | } | |
1884 | ||
1885 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) | |
1886 | { | |
1887 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
1888 | if (cache == NULL) | |
1889 | return; | |
1890 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 1891 | cache->jiffies = set->jiffies; |
1c3c07e9 | 1892 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 1893 | cache->mask = set->mask; |
1c3c07e9 TM |
1894 | |
1895 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
1896 | |
1897 | /* Update accounting */ | |
1898 | smp_mb__before_atomic_inc(); | |
1899 | atomic_long_inc(&nfs_access_nr_entries); | |
1900 | smp_mb__after_atomic_inc(); | |
1901 | ||
1902 | /* Add inode to global LRU list */ | |
1903 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { | |
1904 | spin_lock(&nfs_access_lru_lock); | |
1905 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); | |
1906 | spin_unlock(&nfs_access_lru_lock); | |
1907 | } | |
1da177e4 LT |
1908 | } |
1909 | ||
1910 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
1911 | { | |
1912 | struct nfs_access_entry cache; | |
1913 | int status; | |
1914 | ||
1915 | status = nfs_access_get_cached(inode, cred, &cache); | |
1916 | if (status == 0) | |
1917 | goto out; | |
1918 | ||
1919 | /* Be clever: ask server to check for all possible rights */ | |
1920 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
1921 | cache.cred = cred; | |
1922 | cache.jiffies = jiffies; | |
1923 | status = NFS_PROTO(inode)->access(inode, &cache); | |
1924 | if (status != 0) | |
1925 | return status; | |
1926 | nfs_access_add_cache(inode, &cache); | |
1927 | out: | |
1928 | if ((cache.mask & mask) == mask) | |
1929 | return 0; | |
1930 | return -EACCES; | |
1931 | } | |
1932 | ||
1933 | int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) | |
1934 | { | |
1935 | struct rpc_cred *cred; | |
1936 | int res = 0; | |
1937 | ||
91d5b470 CL |
1938 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
1939 | ||
1da177e4 LT |
1940 | if (mask == 0) |
1941 | goto out; | |
1942 | /* Is this sys_access() ? */ | |
1943 | if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) | |
1944 | goto force_lookup; | |
1945 | ||
1946 | switch (inode->i_mode & S_IFMT) { | |
1947 | case S_IFLNK: | |
1948 | goto out; | |
1949 | case S_IFREG: | |
1950 | /* NFSv4 has atomic_open... */ | |
1951 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
1952 | && nd != NULL | |
1953 | && (nd->flags & LOOKUP_OPEN)) | |
1954 | goto out; | |
1955 | break; | |
1956 | case S_IFDIR: | |
1957 | /* | |
1958 | * Optimize away all write operations, since the server | |
1959 | * will check permissions when we perform the op. | |
1960 | */ | |
1961 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
1962 | goto out; | |
1963 | } | |
1964 | ||
1965 | force_lookup: | |
1966 | lock_kernel(); | |
1967 | ||
1968 | if (!NFS_PROTO(inode)->access) | |
1969 | goto out_notsup; | |
1970 | ||
1971 | cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); | |
1972 | if (!IS_ERR(cred)) { | |
1973 | res = nfs_do_access(inode, cred, mask); | |
1974 | put_rpccred(cred); | |
1975 | } else | |
1976 | res = PTR_ERR(cred); | |
1977 | unlock_kernel(); | |
1978 | out: | |
1e7cb3dc CL |
1979 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
1980 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
1981 | return res; |
1982 | out_notsup: | |
1983 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
1984 | if (res == 0) | |
1985 | res = generic_permission(inode, mask, NULL); | |
1986 | unlock_kernel(); | |
1e7cb3dc | 1987 | goto out; |
1da177e4 LT |
1988 | } |
1989 | ||
1990 | /* | |
1991 | * Local variables: | |
1992 | * version-control: t | |
1993 | * kept-new-versions: 5 | |
1994 | * End: | |
1995 | */ |