]>
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> | |
33 | #include <linux/namei.h> | |
54ceac45 | 34 | #include <linux/mount.h> |
1da177e4 | 35 | |
4ce79717 | 36 | #include "nfs4_fs.h" |
1da177e4 | 37 | #include "delegation.h" |
91d5b470 | 38 | #include "iostat.h" |
1da177e4 LT |
39 | |
40 | #define NFS_PARANOIA 1 | |
41 | /* #define NFS_DEBUG_VERBOSE 1 */ | |
42 | ||
43 | static int nfs_opendir(struct inode *, struct file *); | |
44 | static int nfs_readdir(struct file *, void *, filldir_t); | |
45 | static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *); | |
46 | static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *); | |
47 | static int nfs_mkdir(struct inode *, struct dentry *, int); | |
48 | static int nfs_rmdir(struct inode *, struct dentry *); | |
49 | static int nfs_unlink(struct inode *, struct dentry *); | |
50 | static int nfs_symlink(struct inode *, struct dentry *, const char *); | |
51 | static int nfs_link(struct dentry *, struct inode *, struct dentry *); | |
52 | static int nfs_mknod(struct inode *, struct dentry *, int, dev_t); | |
53 | static int nfs_rename(struct inode *, struct dentry *, | |
54 | struct inode *, struct dentry *); | |
55 | static int nfs_fsync_dir(struct file *, struct dentry *, int); | |
f0dd2136 | 56 | static loff_t nfs_llseek_dir(struct file *, loff_t, int); |
1da177e4 | 57 | |
4b6f5d20 | 58 | const struct file_operations nfs_dir_operations = { |
f0dd2136 | 59 | .llseek = nfs_llseek_dir, |
1da177e4 LT |
60 | .read = generic_read_dir, |
61 | .readdir = nfs_readdir, | |
62 | .open = nfs_opendir, | |
63 | .release = nfs_release, | |
64 | .fsync = nfs_fsync_dir, | |
65 | }; | |
66 | ||
67 | struct inode_operations nfs_dir_inode_operations = { | |
68 | .create = nfs_create, | |
69 | .lookup = nfs_lookup, | |
70 | .link = nfs_link, | |
71 | .unlink = nfs_unlink, | |
72 | .symlink = nfs_symlink, | |
73 | .mkdir = nfs_mkdir, | |
74 | .rmdir = nfs_rmdir, | |
75 | .mknod = nfs_mknod, | |
76 | .rename = nfs_rename, | |
77 | .permission = nfs_permission, | |
78 | .getattr = nfs_getattr, | |
79 | .setattr = nfs_setattr, | |
80 | }; | |
81 | ||
b7fa0554 AG |
82 | #ifdef CONFIG_NFS_V3 |
83 | struct inode_operations nfs3_dir_inode_operations = { | |
84 | .create = nfs_create, | |
85 | .lookup = nfs_lookup, | |
86 | .link = nfs_link, | |
87 | .unlink = nfs_unlink, | |
88 | .symlink = nfs_symlink, | |
89 | .mkdir = nfs_mkdir, | |
90 | .rmdir = nfs_rmdir, | |
91 | .mknod = nfs_mknod, | |
92 | .rename = nfs_rename, | |
93 | .permission = nfs_permission, | |
94 | .getattr = nfs_getattr, | |
95 | .setattr = nfs_setattr, | |
96 | .listxattr = nfs3_listxattr, | |
97 | .getxattr = nfs3_getxattr, | |
98 | .setxattr = nfs3_setxattr, | |
99 | .removexattr = nfs3_removexattr, | |
100 | }; | |
101 | #endif /* CONFIG_NFS_V3 */ | |
102 | ||
1da177e4 LT |
103 | #ifdef CONFIG_NFS_V4 |
104 | ||
105 | static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *); | |
106 | struct inode_operations nfs4_dir_inode_operations = { | |
107 | .create = nfs_create, | |
108 | .lookup = nfs_atomic_lookup, | |
109 | .link = nfs_link, | |
110 | .unlink = nfs_unlink, | |
111 | .symlink = nfs_symlink, | |
112 | .mkdir = nfs_mkdir, | |
113 | .rmdir = nfs_rmdir, | |
114 | .mknod = nfs_mknod, | |
115 | .rename = nfs_rename, | |
116 | .permission = nfs_permission, | |
117 | .getattr = nfs_getattr, | |
118 | .setattr = nfs_setattr, | |
6b3b5496 BF |
119 | .getxattr = nfs4_getxattr, |
120 | .setxattr = nfs4_setxattr, | |
121 | .listxattr = nfs4_listxattr, | |
1da177e4 LT |
122 | }; |
123 | ||
124 | #endif /* CONFIG_NFS_V4 */ | |
125 | ||
126 | /* | |
127 | * Open file | |
128 | */ | |
129 | static int | |
130 | nfs_opendir(struct inode *inode, struct file *filp) | |
131 | { | |
7451c4f0 | 132 | int res; |
1da177e4 | 133 | |
1e7cb3dc CL |
134 | dfprintk(VFS, "NFS: opendir(%s/%ld)\n", |
135 | inode->i_sb->s_id, inode->i_ino); | |
136 | ||
1da177e4 LT |
137 | lock_kernel(); |
138 | /* Call generic open code in order to cache credentials */ | |
7451c4f0 | 139 | res = nfs_open(inode, filp); |
1da177e4 LT |
140 | unlock_kernel(); |
141 | return res; | |
142 | } | |
143 | ||
144 | typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int); | |
145 | typedef struct { | |
146 | struct file *file; | |
147 | struct page *page; | |
148 | unsigned long page_index; | |
149 | u32 *ptr; | |
f0dd2136 TM |
150 | u64 *dir_cookie; |
151 | loff_t current_index; | |
1da177e4 LT |
152 | struct nfs_entry *entry; |
153 | decode_dirent_t decode; | |
154 | int plus; | |
155 | int error; | |
156 | } nfs_readdir_descriptor_t; | |
157 | ||
158 | /* Now we cache directories properly, by stuffing the dirent | |
159 | * data directly in the page cache. | |
160 | * | |
161 | * Inode invalidation due to refresh etc. takes care of | |
162 | * _everything_, no sloppy entry flushing logic, no extraneous | |
163 | * copying, network direct to page cache, the way it was meant | |
164 | * to be. | |
165 | * | |
166 | * NOTE: Dirent information verification is done always by the | |
167 | * page-in of the RPC reply, nowhere else, this simplies | |
168 | * things substantially. | |
169 | */ | |
170 | static | |
171 | int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page) | |
172 | { | |
173 | struct file *file = desc->file; | |
174 | struct inode *inode = file->f_dentry->d_inode; | |
175 | struct rpc_cred *cred = nfs_file_cred(file); | |
176 | unsigned long timestamp; | |
177 | int error; | |
178 | ||
1e7cb3dc CL |
179 | dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n", |
180 | __FUNCTION__, (long long)desc->entry->cookie, | |
181 | page->index); | |
1da177e4 LT |
182 | |
183 | again: | |
184 | timestamp = jiffies; | |
185 | error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page, | |
186 | NFS_SERVER(inode)->dtsize, desc->plus); | |
187 | if (error < 0) { | |
188 | /* We requested READDIRPLUS, but the server doesn't grok it */ | |
189 | if (error == -ENOTSUPP && desc->plus) { | |
190 | NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS; | |
412d582e | 191 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); |
1da177e4 LT |
192 | desc->plus = 0; |
193 | goto again; | |
194 | } | |
195 | goto error; | |
196 | } | |
197 | SetPageUptodate(page); | |
dc59250c | 198 | spin_lock(&inode->i_lock); |
55296809 | 199 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 200 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
201 | /* Ensure consistent page alignment of the data. |
202 | * Note: assumes we have exclusive access to this mapping either | |
1b1dcc1b | 203 | * through inode->i_mutex or some other mechanism. |
1da177e4 | 204 | */ |
a656db99 TM |
205 | if (page->index == 0) |
206 | invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1); | |
1da177e4 LT |
207 | unlock_page(page); |
208 | return 0; | |
209 | error: | |
210 | SetPageError(page); | |
211 | unlock_page(page); | |
212 | nfs_zap_caches(inode); | |
213 | desc->error = error; | |
214 | return -EIO; | |
215 | } | |
216 | ||
217 | static inline | |
218 | int dir_decode(nfs_readdir_descriptor_t *desc) | |
219 | { | |
220 | u32 *p = desc->ptr; | |
221 | p = desc->decode(p, desc->entry, desc->plus); | |
222 | if (IS_ERR(p)) | |
223 | return PTR_ERR(p); | |
224 | desc->ptr = p; | |
225 | return 0; | |
226 | } | |
227 | ||
228 | static inline | |
229 | void dir_page_release(nfs_readdir_descriptor_t *desc) | |
230 | { | |
231 | kunmap(desc->page); | |
232 | page_cache_release(desc->page); | |
233 | desc->page = NULL; | |
234 | desc->ptr = NULL; | |
235 | } | |
236 | ||
237 | /* | |
238 | * Given a pointer to a buffer that has already been filled by a call | |
f0dd2136 | 239 | * to readdir, find the next entry with cookie '*desc->dir_cookie'. |
1da177e4 LT |
240 | * |
241 | * If the end of the buffer has been reached, return -EAGAIN, if not, | |
242 | * return the offset within the buffer of the next entry to be | |
243 | * read. | |
244 | */ | |
245 | static inline | |
00a92642 | 246 | int find_dirent(nfs_readdir_descriptor_t *desc) |
1da177e4 LT |
247 | { |
248 | struct nfs_entry *entry = desc->entry; | |
249 | int loop_count = 0, | |
250 | status; | |
251 | ||
252 | while((status = dir_decode(desc)) == 0) { | |
1e7cb3dc CL |
253 | dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n", |
254 | __FUNCTION__, (unsigned long long)entry->cookie); | |
f0dd2136 | 255 | if (entry->prev_cookie == *desc->dir_cookie) |
1da177e4 LT |
256 | break; |
257 | if (loop_count++ > 200) { | |
258 | loop_count = 0; | |
259 | schedule(); | |
260 | } | |
261 | } | |
1da177e4 LT |
262 | return status; |
263 | } | |
264 | ||
265 | /* | |
00a92642 | 266 | * Given a pointer to a buffer that has already been filled by a call |
f0dd2136 | 267 | * to readdir, find the entry at offset 'desc->file->f_pos'. |
00a92642 OG |
268 | * |
269 | * If the end of the buffer has been reached, return -EAGAIN, if not, | |
270 | * return the offset within the buffer of the next entry to be | |
271 | * read. | |
272 | */ | |
273 | static inline | |
274 | int find_dirent_index(nfs_readdir_descriptor_t *desc) | |
275 | { | |
276 | struct nfs_entry *entry = desc->entry; | |
277 | int loop_count = 0, | |
278 | status; | |
279 | ||
280 | for(;;) { | |
281 | status = dir_decode(desc); | |
282 | if (status) | |
283 | break; | |
284 | ||
1e7cb3dc CL |
285 | dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n", |
286 | (unsigned long long)entry->cookie, desc->current_index); | |
00a92642 | 287 | |
f0dd2136 TM |
288 | if (desc->file->f_pos == desc->current_index) { |
289 | *desc->dir_cookie = entry->cookie; | |
00a92642 OG |
290 | break; |
291 | } | |
292 | desc->current_index++; | |
293 | if (loop_count++ > 200) { | |
294 | loop_count = 0; | |
295 | schedule(); | |
296 | } | |
297 | } | |
00a92642 OG |
298 | return status; |
299 | } | |
300 | ||
301 | /* | |
302 | * Find the given page, and call find_dirent() or find_dirent_index in | |
303 | * order to try to return the next entry. | |
1da177e4 LT |
304 | */ |
305 | static inline | |
306 | int find_dirent_page(nfs_readdir_descriptor_t *desc) | |
307 | { | |
308 | struct inode *inode = desc->file->f_dentry->d_inode; | |
309 | struct page *page; | |
310 | int status; | |
311 | ||
1e7cb3dc CL |
312 | dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n", |
313 | __FUNCTION__, desc->page_index, | |
314 | (long long) *desc->dir_cookie); | |
1da177e4 LT |
315 | |
316 | page = read_cache_page(inode->i_mapping, desc->page_index, | |
317 | (filler_t *)nfs_readdir_filler, desc); | |
318 | if (IS_ERR(page)) { | |
319 | status = PTR_ERR(page); | |
320 | goto out; | |
321 | } | |
322 | if (!PageUptodate(page)) | |
323 | goto read_error; | |
324 | ||
325 | /* NOTE: Someone else may have changed the READDIRPLUS flag */ | |
326 | desc->page = page; | |
327 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ | |
f0dd2136 | 328 | if (*desc->dir_cookie != 0) |
00a92642 OG |
329 | status = find_dirent(desc); |
330 | else | |
331 | status = find_dirent_index(desc); | |
1da177e4 LT |
332 | if (status < 0) |
333 | dir_page_release(desc); | |
334 | out: | |
1e7cb3dc | 335 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status); |
1da177e4 LT |
336 | return status; |
337 | read_error: | |
338 | page_cache_release(page); | |
339 | return -EIO; | |
340 | } | |
341 | ||
342 | /* | |
343 | * Recurse through the page cache pages, and return a | |
344 | * filled nfs_entry structure of the next directory entry if possible. | |
345 | * | |
f0dd2136 TM |
346 | * The target for the search is '*desc->dir_cookie' if non-0, |
347 | * 'desc->file->f_pos' otherwise | |
1da177e4 LT |
348 | */ |
349 | static inline | |
350 | int readdir_search_pagecache(nfs_readdir_descriptor_t *desc) | |
351 | { | |
352 | int loop_count = 0; | |
353 | int res; | |
354 | ||
00a92642 | 355 | /* Always search-by-index from the beginning of the cache */ |
f0dd2136 | 356 | if (*desc->dir_cookie == 0) { |
1e7cb3dc CL |
357 | dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n", |
358 | (long long)desc->file->f_pos); | |
00a92642 OG |
359 | desc->page_index = 0; |
360 | desc->entry->cookie = desc->entry->prev_cookie = 0; | |
361 | desc->entry->eof = 0; | |
362 | desc->current_index = 0; | |
f0dd2136 | 363 | } else |
1e7cb3dc CL |
364 | dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", |
365 | (unsigned long long)*desc->dir_cookie); | |
00a92642 | 366 | |
1da177e4 LT |
367 | for (;;) { |
368 | res = find_dirent_page(desc); | |
369 | if (res != -EAGAIN) | |
370 | break; | |
371 | /* Align to beginning of next page */ | |
372 | desc->page_index ++; | |
373 | if (loop_count++ > 200) { | |
374 | loop_count = 0; | |
375 | schedule(); | |
376 | } | |
377 | } | |
1e7cb3dc CL |
378 | |
379 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res); | |
1da177e4 LT |
380 | return res; |
381 | } | |
382 | ||
383 | static inline unsigned int dt_type(struct inode *inode) | |
384 | { | |
385 | return (inode->i_mode >> 12) & 15; | |
386 | } | |
387 | ||
388 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc); | |
389 | ||
390 | /* | |
391 | * Once we've found the start of the dirent within a page: fill 'er up... | |
392 | */ | |
393 | static | |
394 | int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent, | |
395 | filldir_t filldir) | |
396 | { | |
397 | struct file *file = desc->file; | |
398 | struct nfs_entry *entry = desc->entry; | |
399 | struct dentry *dentry = NULL; | |
400 | unsigned long fileid; | |
401 | int loop_count = 0, | |
402 | res; | |
403 | ||
1e7cb3dc CL |
404 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", |
405 | (unsigned long long)entry->cookie); | |
1da177e4 LT |
406 | |
407 | for(;;) { | |
408 | unsigned d_type = DT_UNKNOWN; | |
409 | /* Note: entry->prev_cookie contains the cookie for | |
410 | * retrieving the current dirent on the server */ | |
411 | fileid = nfs_fileid_to_ino_t(entry->ino); | |
412 | ||
413 | /* Get a dentry if we have one */ | |
414 | if (dentry != NULL) | |
415 | dput(dentry); | |
416 | dentry = nfs_readdir_lookup(desc); | |
417 | ||
418 | /* Use readdirplus info */ | |
419 | if (dentry != NULL && dentry->d_inode != NULL) { | |
420 | d_type = dt_type(dentry->d_inode); | |
421 | fileid = dentry->d_inode->i_ino; | |
422 | } | |
423 | ||
424 | res = filldir(dirent, entry->name, entry->len, | |
00a92642 | 425 | file->f_pos, fileid, d_type); |
1da177e4 LT |
426 | if (res < 0) |
427 | break; | |
00a92642 | 428 | file->f_pos++; |
f0dd2136 | 429 | *desc->dir_cookie = entry->cookie; |
1da177e4 LT |
430 | if (dir_decode(desc) != 0) { |
431 | desc->page_index ++; | |
432 | break; | |
433 | } | |
434 | if (loop_count++ > 200) { | |
435 | loop_count = 0; | |
436 | schedule(); | |
437 | } | |
438 | } | |
439 | dir_page_release(desc); | |
440 | if (dentry != NULL) | |
441 | dput(dentry); | |
1e7cb3dc CL |
442 | dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", |
443 | (unsigned long long)*desc->dir_cookie, res); | |
1da177e4 LT |
444 | return res; |
445 | } | |
446 | ||
447 | /* | |
448 | * If we cannot find a cookie in our cache, we suspect that this is | |
449 | * because it points to a deleted file, so we ask the server to return | |
450 | * whatever it thinks is the next entry. We then feed this to filldir. | |
451 | * If all goes well, we should then be able to find our way round the | |
452 | * cache on the next call to readdir_search_pagecache(); | |
453 | * | |
454 | * NOTE: we cannot add the anonymous page to the pagecache because | |
455 | * the data it contains might not be page aligned. Besides, | |
456 | * we should already have a complete representation of the | |
457 | * directory in the page cache by the time we get here. | |
458 | */ | |
459 | static inline | |
460 | int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent, | |
461 | filldir_t filldir) | |
462 | { | |
463 | struct file *file = desc->file; | |
464 | struct inode *inode = file->f_dentry->d_inode; | |
465 | struct rpc_cred *cred = nfs_file_cred(file); | |
466 | struct page *page = NULL; | |
467 | int status; | |
468 | ||
1e7cb3dc CL |
469 | dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n", |
470 | (unsigned long long)*desc->dir_cookie); | |
1da177e4 LT |
471 | |
472 | page = alloc_page(GFP_HIGHUSER); | |
473 | if (!page) { | |
474 | status = -ENOMEM; | |
475 | goto out; | |
476 | } | |
f0dd2136 | 477 | desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie, |
1da177e4 LT |
478 | page, |
479 | NFS_SERVER(inode)->dtsize, | |
480 | desc->plus); | |
dc59250c | 481 | spin_lock(&inode->i_lock); |
55296809 | 482 | NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME; |
dc59250c | 483 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
484 | desc->page = page; |
485 | desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */ | |
486 | if (desc->error >= 0) { | |
487 | if ((status = dir_decode(desc)) == 0) | |
f0dd2136 | 488 | desc->entry->prev_cookie = *desc->dir_cookie; |
1da177e4 LT |
489 | } else |
490 | status = -EIO; | |
491 | if (status < 0) | |
492 | goto out_release; | |
493 | ||
494 | status = nfs_do_filldir(desc, dirent, filldir); | |
495 | ||
496 | /* Reset read descriptor so it searches the page cache from | |
497 | * the start upon the next call to readdir_search_pagecache() */ | |
498 | desc->page_index = 0; | |
499 | desc->entry->cookie = desc->entry->prev_cookie = 0; | |
500 | desc->entry->eof = 0; | |
501 | out: | |
1e7cb3dc CL |
502 | dfprintk(DIRCACHE, "NFS: %s: returns %d\n", |
503 | __FUNCTION__, status); | |
1da177e4 LT |
504 | return status; |
505 | out_release: | |
506 | dir_page_release(desc); | |
507 | goto out; | |
508 | } | |
509 | ||
00a92642 OG |
510 | /* The file offset position represents the dirent entry number. A |
511 | last cookie cache takes care of the common case of reading the | |
512 | whole directory. | |
1da177e4 LT |
513 | */ |
514 | static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir) | |
515 | { | |
516 | struct dentry *dentry = filp->f_dentry; | |
517 | struct inode *inode = dentry->d_inode; | |
518 | nfs_readdir_descriptor_t my_desc, | |
519 | *desc = &my_desc; | |
520 | struct nfs_entry my_entry; | |
521 | struct nfs_fh fh; | |
522 | struct nfs_fattr fattr; | |
523 | long res; | |
524 | ||
1e7cb3dc CL |
525 | dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n", |
526 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
527 | (long long)filp->f_pos); | |
91d5b470 CL |
528 | nfs_inc_stats(inode, NFSIOS_VFSGETDENTS); |
529 | ||
1da177e4 LT |
530 | lock_kernel(); |
531 | ||
44b11874 | 532 | res = nfs_revalidate_mapping(inode, filp->f_mapping); |
1da177e4 LT |
533 | if (res < 0) { |
534 | unlock_kernel(); | |
535 | return res; | |
536 | } | |
537 | ||
538 | /* | |
00a92642 | 539 | * filp->f_pos points to the dirent entry number. |
f0dd2136 | 540 | * *desc->dir_cookie has the cookie for the next entry. We have |
00a92642 OG |
541 | * to either find the entry with the appropriate number or |
542 | * revalidate the cookie. | |
1da177e4 LT |
543 | */ |
544 | memset(desc, 0, sizeof(*desc)); | |
545 | ||
546 | desc->file = filp; | |
f0dd2136 | 547 | desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie; |
1da177e4 LT |
548 | desc->decode = NFS_PROTO(inode)->decode_dirent; |
549 | desc->plus = NFS_USE_READDIRPLUS(inode); | |
550 | ||
551 | my_entry.cookie = my_entry.prev_cookie = 0; | |
552 | my_entry.eof = 0; | |
553 | my_entry.fh = &fh; | |
554 | my_entry.fattr = &fattr; | |
0e574af1 | 555 | nfs_fattr_init(&fattr); |
1da177e4 LT |
556 | desc->entry = &my_entry; |
557 | ||
558 | while(!desc->entry->eof) { | |
559 | res = readdir_search_pagecache(desc); | |
00a92642 | 560 | |
1da177e4 LT |
561 | if (res == -EBADCOOKIE) { |
562 | /* This means either end of directory */ | |
f0dd2136 | 563 | if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) { |
1da177e4 LT |
564 | /* Or that the server has 'lost' a cookie */ |
565 | res = uncached_readdir(desc, dirent, filldir); | |
566 | if (res >= 0) | |
567 | continue; | |
568 | } | |
569 | res = 0; | |
570 | break; | |
571 | } | |
572 | if (res == -ETOOSMALL && desc->plus) { | |
412d582e | 573 | clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode)); |
1da177e4 LT |
574 | nfs_zap_caches(inode); |
575 | desc->plus = 0; | |
576 | desc->entry->eof = 0; | |
577 | continue; | |
578 | } | |
579 | if (res < 0) | |
580 | break; | |
581 | ||
582 | res = nfs_do_filldir(desc, dirent, filldir); | |
583 | if (res < 0) { | |
584 | res = 0; | |
585 | break; | |
586 | } | |
587 | } | |
588 | unlock_kernel(); | |
1e7cb3dc CL |
589 | if (res > 0) |
590 | res = 0; | |
591 | dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n", | |
592 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
593 | res); | |
594 | return res; | |
1da177e4 LT |
595 | } |
596 | ||
f0dd2136 TM |
597 | loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin) |
598 | { | |
1b1dcc1b | 599 | mutex_lock(&filp->f_dentry->d_inode->i_mutex); |
f0dd2136 TM |
600 | switch (origin) { |
601 | case 1: | |
602 | offset += filp->f_pos; | |
603 | case 0: | |
604 | if (offset >= 0) | |
605 | break; | |
606 | default: | |
607 | offset = -EINVAL; | |
608 | goto out; | |
609 | } | |
610 | if (offset != filp->f_pos) { | |
611 | filp->f_pos = offset; | |
612 | ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0; | |
613 | } | |
614 | out: | |
1b1dcc1b | 615 | mutex_unlock(&filp->f_dentry->d_inode->i_mutex); |
f0dd2136 TM |
616 | return offset; |
617 | } | |
618 | ||
1da177e4 LT |
619 | /* |
620 | * All directory operations under NFS are synchronous, so fsync() | |
621 | * is a dummy operation. | |
622 | */ | |
623 | int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync) | |
624 | { | |
1e7cb3dc CL |
625 | dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n", |
626 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
627 | datasync); | |
628 | ||
1da177e4 LT |
629 | return 0; |
630 | } | |
631 | ||
632 | /* | |
633 | * A check for whether or not the parent directory has changed. | |
634 | * In the case it has, we assume that the dentries are untrustworthy | |
635 | * and may need to be looked up again. | |
636 | */ | |
637 | static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry) | |
638 | { | |
639 | if (IS_ROOT(dentry)) | |
640 | return 1; | |
55296809 | 641 | if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0 |
1da177e4 LT |
642 | || nfs_attribute_timeout(dir)) |
643 | return 0; | |
644 | return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata); | |
645 | } | |
646 | ||
647 | static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf) | |
648 | { | |
649 | dentry->d_fsdata = (void *)verf; | |
650 | } | |
651 | ||
652 | /* | |
653 | * Whenever an NFS operation succeeds, we know that the dentry | |
654 | * is valid, so we update the revalidation timestamp. | |
655 | */ | |
656 | static inline void nfs_renew_times(struct dentry * dentry) | |
657 | { | |
658 | dentry->d_time = jiffies; | |
659 | } | |
660 | ||
1d6757fb TM |
661 | /* |
662 | * Return the intent data that applies to this particular path component | |
663 | * | |
664 | * Note that the current set of intents only apply to the very last | |
665 | * component of the path. | |
666 | * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT. | |
667 | */ | |
668 | static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask) | |
669 | { | |
670 | if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT)) | |
671 | return 0; | |
672 | return nd->flags & mask; | |
673 | } | |
674 | ||
675 | /* | |
676 | * Inode and filehandle revalidation for lookups. | |
677 | * | |
678 | * We force revalidation in the cases where the VFS sets LOOKUP_REVAL, | |
679 | * or if the intent information indicates that we're about to open this | |
680 | * particular file and the "nocto" mount flag is not set. | |
681 | * | |
682 | */ | |
1da177e4 LT |
683 | static inline |
684 | int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd) | |
685 | { | |
686 | struct nfs_server *server = NFS_SERVER(inode); | |
687 | ||
688 | if (nd != NULL) { | |
1da177e4 | 689 | /* VFS wants an on-the-wire revalidation */ |
1d6757fb | 690 | if (nd->flags & LOOKUP_REVAL) |
1da177e4 LT |
691 | goto out_force; |
692 | /* This is an open(2) */ | |
1d6757fb | 693 | if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 && |
4e0641a7 TM |
694 | !(server->flags & NFS_MOUNT_NOCTO) && |
695 | (S_ISREG(inode->i_mode) || | |
696 | S_ISDIR(inode->i_mode))) | |
1da177e4 LT |
697 | goto out_force; |
698 | } | |
699 | return nfs_revalidate_inode(server, inode); | |
700 | out_force: | |
701 | return __nfs_revalidate_inode(server, inode); | |
702 | } | |
703 | ||
704 | /* | |
705 | * We judge how long we want to trust negative | |
706 | * dentries by looking at the parent inode mtime. | |
707 | * | |
708 | * If parent mtime has changed, we revalidate, else we wait for a | |
709 | * period corresponding to the parent's attribute cache timeout value. | |
710 | */ | |
711 | static inline | |
712 | int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry, | |
713 | struct nameidata *nd) | |
714 | { | |
1da177e4 | 715 | /* Don't revalidate a negative dentry if we're creating a new file */ |
1d6757fb | 716 | if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0) |
1da177e4 LT |
717 | return 0; |
718 | return !nfs_check_verifier(dir, dentry); | |
719 | } | |
720 | ||
721 | /* | |
722 | * This is called every time the dcache has a lookup hit, | |
723 | * and we should check whether we can really trust that | |
724 | * lookup. | |
725 | * | |
726 | * NOTE! The hit can be a negative hit too, don't assume | |
727 | * we have an inode! | |
728 | * | |
729 | * If the parent directory is seen to have changed, we throw out the | |
730 | * cached dentry and do a new lookup. | |
731 | */ | |
732 | static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd) | |
733 | { | |
734 | struct inode *dir; | |
735 | struct inode *inode; | |
736 | struct dentry *parent; | |
737 | int error; | |
738 | struct nfs_fh fhandle; | |
739 | struct nfs_fattr fattr; | |
740 | unsigned long verifier; | |
741 | ||
742 | parent = dget_parent(dentry); | |
743 | lock_kernel(); | |
744 | dir = parent->d_inode; | |
91d5b470 | 745 | nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE); |
1da177e4 LT |
746 | inode = dentry->d_inode; |
747 | ||
748 | if (!inode) { | |
749 | if (nfs_neg_need_reval(dir, dentry, nd)) | |
750 | goto out_bad; | |
751 | goto out_valid; | |
752 | } | |
753 | ||
754 | if (is_bad_inode(inode)) { | |
1e7cb3dc CL |
755 | dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n", |
756 | __FUNCTION__, dentry->d_parent->d_name.name, | |
757 | dentry->d_name.name); | |
1da177e4 LT |
758 | goto out_bad; |
759 | } | |
760 | ||
761 | /* Revalidate parent directory attribute cache */ | |
762 | if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0) | |
763 | goto out_zap_parent; | |
764 | ||
765 | /* Force a full look up iff the parent directory has changed */ | |
766 | if (nfs_check_verifier(dir, dentry)) { | |
767 | if (nfs_lookup_verify_inode(inode, nd)) | |
768 | goto out_zap_parent; | |
769 | goto out_valid; | |
770 | } | |
771 | ||
772 | if (NFS_STALE(inode)) | |
773 | goto out_bad; | |
774 | ||
775 | verifier = nfs_save_change_attribute(dir); | |
776 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); | |
777 | if (error) | |
778 | goto out_bad; | |
779 | if (nfs_compare_fh(NFS_FH(inode), &fhandle)) | |
780 | goto out_bad; | |
781 | if ((error = nfs_refresh_inode(inode, &fattr)) != 0) | |
782 | goto out_bad; | |
783 | ||
784 | nfs_renew_times(dentry); | |
785 | nfs_set_verifier(dentry, verifier); | |
786 | out_valid: | |
787 | unlock_kernel(); | |
788 | dput(parent); | |
1e7cb3dc CL |
789 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n", |
790 | __FUNCTION__, dentry->d_parent->d_name.name, | |
791 | dentry->d_name.name); | |
1da177e4 LT |
792 | return 1; |
793 | out_zap_parent: | |
794 | nfs_zap_caches(dir); | |
795 | out_bad: | |
796 | NFS_CACHEINV(dir); | |
797 | if (inode && S_ISDIR(inode->i_mode)) { | |
798 | /* Purge readdir caches. */ | |
799 | nfs_zap_caches(inode); | |
800 | /* If we have submounts, don't unhash ! */ | |
801 | if (have_submounts(dentry)) | |
802 | goto out_valid; | |
803 | shrink_dcache_parent(dentry); | |
804 | } | |
805 | d_drop(dentry); | |
806 | unlock_kernel(); | |
807 | dput(parent); | |
1e7cb3dc CL |
808 | dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n", |
809 | __FUNCTION__, dentry->d_parent->d_name.name, | |
810 | dentry->d_name.name); | |
1da177e4 LT |
811 | return 0; |
812 | } | |
813 | ||
814 | /* | |
815 | * This is called from dput() when d_count is going to 0. | |
816 | */ | |
817 | static int nfs_dentry_delete(struct dentry *dentry) | |
818 | { | |
819 | dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n", | |
820 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
821 | dentry->d_flags); | |
822 | ||
823 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
824 | /* Unhash it, so that ->d_iput() would be called */ | |
825 | return 1; | |
826 | } | |
827 | if (!(dentry->d_sb->s_flags & MS_ACTIVE)) { | |
828 | /* Unhash it, so that ancestors of killed async unlink | |
829 | * files will be cleaned up during umount */ | |
830 | return 1; | |
831 | } | |
832 | return 0; | |
833 | ||
834 | } | |
835 | ||
836 | /* | |
837 | * Called when the dentry loses inode. | |
838 | * We use it to clean up silly-renamed files. | |
839 | */ | |
840 | static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode) | |
841 | { | |
cae7a073 | 842 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
843 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { |
844 | lock_kernel(); | |
845 | inode->i_nlink--; | |
846 | nfs_complete_unlink(dentry); | |
847 | unlock_kernel(); | |
848 | } | |
849 | /* When creating a negative dentry, we want to renew d_time */ | |
850 | nfs_renew_times(dentry); | |
851 | iput(inode); | |
852 | } | |
853 | ||
854 | struct dentry_operations nfs_dentry_operations = { | |
855 | .d_revalidate = nfs_lookup_revalidate, | |
856 | .d_delete = nfs_dentry_delete, | |
857 | .d_iput = nfs_dentry_iput, | |
858 | }; | |
859 | ||
1d6757fb TM |
860 | /* |
861 | * Use intent information to check whether or not we're going to do | |
862 | * an O_EXCL create using this path component. | |
863 | */ | |
1da177e4 LT |
864 | static inline |
865 | int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd) | |
866 | { | |
867 | if (NFS_PROTO(dir)->version == 2) | |
868 | return 0; | |
1d6757fb | 869 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0) |
1da177e4 LT |
870 | return 0; |
871 | return (nd->intent.open.flags & O_EXCL) != 0; | |
872 | } | |
873 | ||
54ceac45 DH |
874 | static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir, |
875 | struct nfs_fh *fh, struct nfs_fattr *fattr) | |
55a97593 TM |
876 | { |
877 | struct nfs_server *server = NFS_SERVER(dir); | |
878 | ||
879 | if (!nfs_fsid_equal(&server->fsid, &fattr->fsid)) | |
880 | /* Revalidate fsid on root dir */ | |
54ceac45 | 881 | return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode); |
55a97593 TM |
882 | return 0; |
883 | } | |
884 | ||
1da177e4 LT |
885 | static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd) |
886 | { | |
887 | struct dentry *res; | |
888 | struct inode *inode = NULL; | |
889 | int error; | |
890 | struct nfs_fh fhandle; | |
891 | struct nfs_fattr fattr; | |
892 | ||
893 | dfprintk(VFS, "NFS: lookup(%s/%s)\n", | |
894 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
91d5b470 | 895 | nfs_inc_stats(dir, NFSIOS_VFSLOOKUP); |
1da177e4 LT |
896 | |
897 | res = ERR_PTR(-ENAMETOOLONG); | |
898 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) | |
899 | goto out; | |
900 | ||
901 | res = ERR_PTR(-ENOMEM); | |
902 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
903 | ||
904 | lock_kernel(); | |
1da177e4 LT |
905 | |
906 | /* If we're doing an exclusive create, optimize away the lookup */ | |
907 | if (nfs_is_exclusive_create(dir, nd)) | |
908 | goto no_entry; | |
909 | ||
910 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr); | |
911 | if (error == -ENOENT) | |
912 | goto no_entry; | |
913 | if (error < 0) { | |
914 | res = ERR_PTR(error); | |
915 | goto out_unlock; | |
916 | } | |
54ceac45 | 917 | error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr); |
55a97593 TM |
918 | if (error < 0) { |
919 | res = ERR_PTR(error); | |
920 | goto out_unlock; | |
921 | } | |
1da177e4 | 922 | inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr); |
03f28e3a TM |
923 | res = (struct dentry *)inode; |
924 | if (IS_ERR(res)) | |
1da177e4 | 925 | goto out_unlock; |
54ceac45 | 926 | |
1da177e4 | 927 | no_entry: |
54ceac45 | 928 | res = d_materialise_unique(dentry, inode); |
1da177e4 LT |
929 | if (res != NULL) |
930 | dentry = res; | |
931 | nfs_renew_times(dentry); | |
932 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
933 | out_unlock: | |
934 | unlock_kernel(); | |
935 | out: | |
936 | return res; | |
937 | } | |
938 | ||
939 | #ifdef CONFIG_NFS_V4 | |
940 | static int nfs_open_revalidate(struct dentry *, struct nameidata *); | |
941 | ||
942 | struct dentry_operations nfs4_dentry_operations = { | |
943 | .d_revalidate = nfs_open_revalidate, | |
944 | .d_delete = nfs_dentry_delete, | |
945 | .d_iput = nfs_dentry_iput, | |
946 | }; | |
947 | ||
1d6757fb TM |
948 | /* |
949 | * Use intent information to determine whether we need to substitute | |
950 | * the NFSv4-style stateful OPEN for the LOOKUP call | |
951 | */ | |
1da177e4 LT |
952 | static int is_atomic_open(struct inode *dir, struct nameidata *nd) |
953 | { | |
1d6757fb | 954 | if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0) |
1da177e4 LT |
955 | return 0; |
956 | /* NFS does not (yet) have a stateful open for directories */ | |
957 | if (nd->flags & LOOKUP_DIRECTORY) | |
958 | return 0; | |
959 | /* Are we trying to write to a read only partition? */ | |
960 | if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE))) | |
961 | return 0; | |
962 | return 1; | |
963 | } | |
964 | ||
965 | static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd) | |
966 | { | |
967 | struct dentry *res = NULL; | |
1da177e4 LT |
968 | int error; |
969 | ||
1e7cb3dc CL |
970 | dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n", |
971 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
972 | ||
1da177e4 LT |
973 | /* Check that we are indeed trying to open this file */ |
974 | if (!is_atomic_open(dir, nd)) | |
975 | goto no_open; | |
976 | ||
977 | if (dentry->d_name.len > NFS_SERVER(dir)->namelen) { | |
978 | res = ERR_PTR(-ENAMETOOLONG); | |
979 | goto out; | |
980 | } | |
981 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
982 | ||
983 | /* Let vfs_create() deal with O_EXCL */ | |
02a913a7 TM |
984 | if (nd->intent.open.flags & O_EXCL) { |
985 | d_add(dentry, NULL); | |
986 | goto out; | |
987 | } | |
1da177e4 LT |
988 | |
989 | /* Open the file on the server */ | |
990 | lock_kernel(); | |
991 | /* Revalidate parent directory attribute cache */ | |
992 | error = nfs_revalidate_inode(NFS_SERVER(dir), dir); | |
993 | if (error < 0) { | |
994 | res = ERR_PTR(error); | |
01c314a0 | 995 | unlock_kernel(); |
1da177e4 LT |
996 | goto out; |
997 | } | |
998 | ||
999 | if (nd->intent.open.flags & O_CREAT) { | |
1000 | nfs_begin_data_update(dir); | |
02a913a7 | 1001 | res = nfs4_atomic_open(dir, dentry, nd); |
1da177e4 LT |
1002 | nfs_end_data_update(dir); |
1003 | } else | |
02a913a7 | 1004 | res = nfs4_atomic_open(dir, dentry, nd); |
1da177e4 | 1005 | unlock_kernel(); |
02a913a7 TM |
1006 | if (IS_ERR(res)) { |
1007 | error = PTR_ERR(res); | |
1da177e4 LT |
1008 | switch (error) { |
1009 | /* Make a negative dentry */ | |
1010 | case -ENOENT: | |
02a913a7 TM |
1011 | res = NULL; |
1012 | goto out; | |
1da177e4 | 1013 | /* This turned out not to be a regular file */ |
6f926b5b TM |
1014 | case -EISDIR: |
1015 | case -ENOTDIR: | |
1016 | goto no_open; | |
1da177e4 LT |
1017 | case -ELOOP: |
1018 | if (!(nd->intent.open.flags & O_NOFOLLOW)) | |
1019 | goto no_open; | |
1da177e4 LT |
1020 | /* case -EINVAL: */ |
1021 | default: | |
1da177e4 LT |
1022 | goto out; |
1023 | } | |
02a913a7 | 1024 | } else if (res != NULL) |
1da177e4 LT |
1025 | dentry = res; |
1026 | nfs_renew_times(dentry); | |
1027 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1028 | out: | |
1029 | return res; | |
1030 | no_open: | |
1031 | return nfs_lookup(dir, dentry, nd); | |
1032 | } | |
1033 | ||
1034 | static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd) | |
1035 | { | |
1036 | struct dentry *parent = NULL; | |
1037 | struct inode *inode = dentry->d_inode; | |
1038 | struct inode *dir; | |
1039 | unsigned long verifier; | |
1040 | int openflags, ret = 0; | |
1041 | ||
1042 | parent = dget_parent(dentry); | |
1043 | dir = parent->d_inode; | |
1044 | if (!is_atomic_open(dir, nd)) | |
1045 | goto no_open; | |
1046 | /* We can't create new files in nfs_open_revalidate(), so we | |
1047 | * optimize away revalidation of negative dentries. | |
1048 | */ | |
1049 | if (inode == NULL) | |
1050 | goto out; | |
1051 | /* NFS only supports OPEN on regular files */ | |
1052 | if (!S_ISREG(inode->i_mode)) | |
1053 | goto no_open; | |
1054 | openflags = nd->intent.open.flags; | |
1055 | /* We cannot do exclusive creation on a positive dentry */ | |
1056 | if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL)) | |
1057 | goto no_open; | |
1058 | /* We can't create new files, or truncate existing ones here */ | |
1059 | openflags &= ~(O_CREAT|O_TRUNC); | |
1060 | ||
1061 | /* | |
1b1dcc1b | 1062 | * Note: we're not holding inode->i_mutex and so may be racing with |
1da177e4 LT |
1063 | * operations that change the directory. We therefore save the |
1064 | * change attribute *before* we do the RPC call. | |
1065 | */ | |
1066 | lock_kernel(); | |
1067 | verifier = nfs_save_change_attribute(dir); | |
02a913a7 | 1068 | ret = nfs4_open_revalidate(dir, dentry, openflags, nd); |
1da177e4 LT |
1069 | if (!ret) |
1070 | nfs_set_verifier(dentry, verifier); | |
1071 | unlock_kernel(); | |
1072 | out: | |
1073 | dput(parent); | |
1074 | if (!ret) | |
1075 | d_drop(dentry); | |
1076 | return ret; | |
1077 | no_open: | |
1078 | dput(parent); | |
1079 | if (inode != NULL && nfs_have_delegation(inode, FMODE_READ)) | |
1080 | return 1; | |
1081 | return nfs_lookup_revalidate(dentry, nd); | |
1082 | } | |
1083 | #endif /* CONFIG_NFSV4 */ | |
1084 | ||
1085 | static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc) | |
1086 | { | |
1087 | struct dentry *parent = desc->file->f_dentry; | |
1088 | struct inode *dir = parent->d_inode; | |
1089 | struct nfs_entry *entry = desc->entry; | |
1090 | struct dentry *dentry, *alias; | |
1091 | struct qstr name = { | |
1092 | .name = entry->name, | |
1093 | .len = entry->len, | |
1094 | }; | |
1095 | struct inode *inode; | |
1096 | ||
1097 | switch (name.len) { | |
1098 | case 2: | |
1099 | if (name.name[0] == '.' && name.name[1] == '.') | |
1100 | return dget_parent(parent); | |
1101 | break; | |
1102 | case 1: | |
1103 | if (name.name[0] == '.') | |
1104 | return dget(parent); | |
1105 | } | |
1106 | name.hash = full_name_hash(name.name, name.len); | |
1107 | dentry = d_lookup(parent, &name); | |
1108 | if (dentry != NULL) | |
1109 | return dentry; | |
1110 | if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR)) | |
1111 | return NULL; | |
1b1dcc1b | 1112 | /* Note: caller is already holding the dir->i_mutex! */ |
1da177e4 LT |
1113 | dentry = d_alloc(parent, &name); |
1114 | if (dentry == NULL) | |
1115 | return NULL; | |
1116 | dentry->d_op = NFS_PROTO(dir)->dentry_ops; | |
1117 | inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr); | |
03f28e3a | 1118 | if (IS_ERR(inode)) { |
1da177e4 LT |
1119 | dput(dentry); |
1120 | return NULL; | |
1121 | } | |
54ceac45 DH |
1122 | |
1123 | alias = d_materialise_unique(dentry, inode); | |
1da177e4 LT |
1124 | if (alias != NULL) { |
1125 | dput(dentry); | |
1126 | dentry = alias; | |
1127 | } | |
54ceac45 | 1128 | |
1da177e4 LT |
1129 | nfs_renew_times(dentry); |
1130 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1131 | return dentry; | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Code common to create, mkdir, and mknod. | |
1136 | */ | |
1137 | int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle, | |
1138 | struct nfs_fattr *fattr) | |
1139 | { | |
1140 | struct inode *inode; | |
1141 | int error = -EACCES; | |
1142 | ||
1143 | /* We may have been initialized further down */ | |
1144 | if (dentry->d_inode) | |
1145 | return 0; | |
1146 | if (fhandle->size == 0) { | |
1147 | struct inode *dir = dentry->d_parent->d_inode; | |
1148 | error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr); | |
1149 | if (error) | |
4f390c15 | 1150 | return error; |
1da177e4 LT |
1151 | } |
1152 | if (!(fattr->valid & NFS_ATTR_FATTR)) { | |
1153 | struct nfs_server *server = NFS_SB(dentry->d_sb); | |
8fa5c000 | 1154 | error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr); |
1da177e4 | 1155 | if (error < 0) |
4f390c15 | 1156 | return error; |
1da177e4 | 1157 | } |
1da177e4 | 1158 | inode = nfs_fhget(dentry->d_sb, fhandle, fattr); |
03f28e3a TM |
1159 | error = PTR_ERR(inode); |
1160 | if (IS_ERR(inode)) | |
4f390c15 | 1161 | return error; |
1da177e4 LT |
1162 | d_instantiate(dentry, inode); |
1163 | return 0; | |
1da177e4 LT |
1164 | } |
1165 | ||
1166 | /* | |
1167 | * Following a failed create operation, we drop the dentry rather | |
1168 | * than retain a negative dentry. This avoids a problem in the event | |
1169 | * that the operation succeeded on the server, but an error in the | |
1170 | * reply path made it appear to have failed. | |
1171 | */ | |
1172 | static int nfs_create(struct inode *dir, struct dentry *dentry, int mode, | |
1173 | struct nameidata *nd) | |
1174 | { | |
1175 | struct iattr attr; | |
1176 | int error; | |
1177 | int open_flags = 0; | |
1178 | ||
1e7cb3dc CL |
1179 | dfprintk(VFS, "NFS: create(%s/%ld), %s\n", |
1180 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1181 | |
1182 | attr.ia_mode = mode; | |
1183 | attr.ia_valid = ATTR_MODE; | |
1184 | ||
1185 | if (nd && (nd->flags & LOOKUP_CREATE)) | |
1186 | open_flags = nd->intent.open.flags; | |
1187 | ||
1188 | lock_kernel(); | |
1189 | nfs_begin_data_update(dir); | |
02a913a7 | 1190 | error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd); |
1da177e4 LT |
1191 | nfs_end_data_update(dir); |
1192 | if (error != 0) | |
1193 | goto out_err; | |
1194 | nfs_renew_times(dentry); | |
1195 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1196 | unlock_kernel(); | |
1197 | return 0; | |
1198 | out_err: | |
1199 | unlock_kernel(); | |
1200 | d_drop(dentry); | |
1201 | return error; | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * See comments for nfs_proc_create regarding failed operations. | |
1206 | */ | |
1207 | static int | |
1208 | nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev) | |
1209 | { | |
1210 | struct iattr attr; | |
1211 | int status; | |
1212 | ||
1e7cb3dc CL |
1213 | dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n", |
1214 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1215 | |
1216 | if (!new_valid_dev(rdev)) | |
1217 | return -EINVAL; | |
1218 | ||
1219 | attr.ia_mode = mode; | |
1220 | attr.ia_valid = ATTR_MODE; | |
1221 | ||
1222 | lock_kernel(); | |
1223 | nfs_begin_data_update(dir); | |
1224 | status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev); | |
1225 | nfs_end_data_update(dir); | |
1226 | if (status != 0) | |
1227 | goto out_err; | |
1228 | nfs_renew_times(dentry); | |
1229 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1230 | unlock_kernel(); | |
1231 | return 0; | |
1232 | out_err: | |
1233 | unlock_kernel(); | |
1234 | d_drop(dentry); | |
1235 | return status; | |
1236 | } | |
1237 | ||
1238 | /* | |
1239 | * See comments for nfs_proc_create regarding failed operations. | |
1240 | */ | |
1241 | static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode) | |
1242 | { | |
1243 | struct iattr attr; | |
1244 | int error; | |
1245 | ||
1e7cb3dc CL |
1246 | dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n", |
1247 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1248 | |
1249 | attr.ia_valid = ATTR_MODE; | |
1250 | attr.ia_mode = mode | S_IFDIR; | |
1251 | ||
1252 | lock_kernel(); | |
1253 | nfs_begin_data_update(dir); | |
1254 | error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr); | |
1255 | nfs_end_data_update(dir); | |
1256 | if (error != 0) | |
1257 | goto out_err; | |
1258 | nfs_renew_times(dentry); | |
1259 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1260 | unlock_kernel(); | |
1261 | return 0; | |
1262 | out_err: | |
1263 | d_drop(dentry); | |
1264 | unlock_kernel(); | |
1265 | return error; | |
1266 | } | |
1267 | ||
1268 | static int nfs_rmdir(struct inode *dir, struct dentry *dentry) | |
1269 | { | |
1270 | int error; | |
1271 | ||
1e7cb3dc CL |
1272 | dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n", |
1273 | dir->i_sb->s_id, dir->i_ino, dentry->d_name.name); | |
1da177e4 LT |
1274 | |
1275 | lock_kernel(); | |
1276 | nfs_begin_data_update(dir); | |
1277 | error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name); | |
1278 | /* Ensure the VFS deletes this inode */ | |
1279 | if (error == 0 && dentry->d_inode != NULL) | |
1280 | dentry->d_inode->i_nlink = 0; | |
1281 | nfs_end_data_update(dir); | |
1282 | unlock_kernel(); | |
1283 | ||
1284 | return error; | |
1285 | } | |
1286 | ||
1287 | static int nfs_sillyrename(struct inode *dir, struct dentry *dentry) | |
1288 | { | |
1289 | static unsigned int sillycounter; | |
1290 | const int i_inosize = sizeof(dir->i_ino)*2; | |
1291 | const int countersize = sizeof(sillycounter)*2; | |
1292 | const int slen = sizeof(".nfs") + i_inosize + countersize - 1; | |
1293 | char silly[slen+1]; | |
1294 | struct qstr qsilly; | |
1295 | struct dentry *sdentry; | |
1296 | int error = -EIO; | |
1297 | ||
1298 | dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n", | |
1299 | dentry->d_parent->d_name.name, dentry->d_name.name, | |
1300 | atomic_read(&dentry->d_count)); | |
91d5b470 | 1301 | nfs_inc_stats(dir, NFSIOS_SILLYRENAME); |
1da177e4 LT |
1302 | |
1303 | #ifdef NFS_PARANOIA | |
1304 | if (!dentry->d_inode) | |
1305 | printk("NFS: silly-renaming %s/%s, negative dentry??\n", | |
1306 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1307 | #endif | |
1308 | /* | |
1309 | * We don't allow a dentry to be silly-renamed twice. | |
1310 | */ | |
1311 | error = -EBUSY; | |
1312 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) | |
1313 | goto out; | |
1314 | ||
1315 | sprintf(silly, ".nfs%*.*lx", | |
1316 | i_inosize, i_inosize, dentry->d_inode->i_ino); | |
1317 | ||
34ea8188 TM |
1318 | /* Return delegation in anticipation of the rename */ |
1319 | nfs_inode_return_delegation(dentry->d_inode); | |
1320 | ||
1da177e4 LT |
1321 | sdentry = NULL; |
1322 | do { | |
1323 | char *suffix = silly + slen - countersize; | |
1324 | ||
1325 | dput(sdentry); | |
1326 | sillycounter++; | |
1327 | sprintf(suffix, "%*.*x", countersize, countersize, sillycounter); | |
1328 | ||
1e7cb3dc CL |
1329 | dfprintk(VFS, "NFS: trying to rename %s to %s\n", |
1330 | dentry->d_name.name, silly); | |
1da177e4 LT |
1331 | |
1332 | sdentry = lookup_one_len(silly, dentry->d_parent, slen); | |
1333 | /* | |
1334 | * N.B. Better to return EBUSY here ... it could be | |
1335 | * dangerous to delete the file while it's in use. | |
1336 | */ | |
1337 | if (IS_ERR(sdentry)) | |
1338 | goto out; | |
1339 | } while(sdentry->d_inode != NULL); /* need negative lookup */ | |
1340 | ||
1341 | qsilly.name = silly; | |
1342 | qsilly.len = strlen(silly); | |
1343 | nfs_begin_data_update(dir); | |
1344 | if (dentry->d_inode) { | |
1345 | nfs_begin_data_update(dentry->d_inode); | |
1346 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, | |
1347 | dir, &qsilly); | |
5ba7cc48 | 1348 | nfs_mark_for_revalidate(dentry->d_inode); |
1da177e4 LT |
1349 | nfs_end_data_update(dentry->d_inode); |
1350 | } else | |
1351 | error = NFS_PROTO(dir)->rename(dir, &dentry->d_name, | |
1352 | dir, &qsilly); | |
1353 | nfs_end_data_update(dir); | |
1354 | if (!error) { | |
1355 | nfs_renew_times(dentry); | |
1356 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1357 | d_move(dentry, sdentry); | |
1358 | error = nfs_async_unlink(dentry); | |
1359 | /* If we return 0 we don't unlink */ | |
1360 | } | |
1361 | dput(sdentry); | |
1362 | out: | |
1363 | return error; | |
1364 | } | |
1365 | ||
1366 | /* | |
1367 | * Remove a file after making sure there are no pending writes, | |
1368 | * and after checking that the file has only one user. | |
1369 | * | |
1370 | * We invalidate the attribute cache and free the inode prior to the operation | |
1371 | * to avoid possible races if the server reuses the inode. | |
1372 | */ | |
1373 | static int nfs_safe_remove(struct dentry *dentry) | |
1374 | { | |
1375 | struct inode *dir = dentry->d_parent->d_inode; | |
1376 | struct inode *inode = dentry->d_inode; | |
1377 | int error = -EBUSY; | |
1378 | ||
1379 | dfprintk(VFS, "NFS: safe_remove(%s/%s)\n", | |
1380 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1381 | ||
1382 | /* If the dentry was sillyrenamed, we simply call d_delete() */ | |
1383 | if (dentry->d_flags & DCACHE_NFSFS_RENAMED) { | |
1384 | error = 0; | |
1385 | goto out; | |
1386 | } | |
1387 | ||
1388 | nfs_begin_data_update(dir); | |
1389 | if (inode != NULL) { | |
cae7a073 | 1390 | nfs_inode_return_delegation(inode); |
1da177e4 LT |
1391 | nfs_begin_data_update(inode); |
1392 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1393 | /* The VFS may want to delete this inode */ | |
1394 | if (error == 0) | |
1395 | inode->i_nlink--; | |
5ba7cc48 | 1396 | nfs_mark_for_revalidate(inode); |
1da177e4 LT |
1397 | nfs_end_data_update(inode); |
1398 | } else | |
1399 | error = NFS_PROTO(dir)->remove(dir, &dentry->d_name); | |
1400 | nfs_end_data_update(dir); | |
1401 | out: | |
1402 | return error; | |
1403 | } | |
1404 | ||
1405 | /* We do silly rename. In case sillyrename() returns -EBUSY, the inode | |
1406 | * belongs to an active ".nfs..." file and we return -EBUSY. | |
1407 | * | |
1408 | * If sillyrename() returns 0, we do nothing, otherwise we unlink. | |
1409 | */ | |
1410 | static int nfs_unlink(struct inode *dir, struct dentry *dentry) | |
1411 | { | |
1412 | int error; | |
1413 | int need_rehash = 0; | |
1414 | ||
1415 | dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id, | |
1416 | dir->i_ino, dentry->d_name.name); | |
1417 | ||
1418 | lock_kernel(); | |
1419 | spin_lock(&dcache_lock); | |
1420 | spin_lock(&dentry->d_lock); | |
1421 | if (atomic_read(&dentry->d_count) > 1) { | |
1422 | spin_unlock(&dentry->d_lock); | |
1423 | spin_unlock(&dcache_lock); | |
1424 | error = nfs_sillyrename(dir, dentry); | |
1425 | unlock_kernel(); | |
1426 | return error; | |
1427 | } | |
1428 | if (!d_unhashed(dentry)) { | |
1429 | __d_drop(dentry); | |
1430 | need_rehash = 1; | |
1431 | } | |
1432 | spin_unlock(&dentry->d_lock); | |
1433 | spin_unlock(&dcache_lock); | |
1434 | error = nfs_safe_remove(dentry); | |
1435 | if (!error) { | |
1436 | nfs_renew_times(dentry); | |
1437 | nfs_set_verifier(dentry, nfs_save_change_attribute(dir)); | |
1438 | } else if (need_rehash) | |
1439 | d_rehash(dentry); | |
1440 | unlock_kernel(); | |
1441 | return error; | |
1442 | } | |
1443 | ||
1444 | static int | |
1445 | nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname) | |
1446 | { | |
1447 | struct iattr attr; | |
1da177e4 LT |
1448 | struct qstr qsymname; |
1449 | int error; | |
1450 | ||
1451 | dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id, | |
1452 | dir->i_ino, dentry->d_name.name, symname); | |
1453 | ||
1454 | #ifdef NFS_PARANOIA | |
1455 | if (dentry->d_inode) | |
1456 | printk("nfs_proc_symlink: %s/%s not negative!\n", | |
1457 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1458 | #endif | |
1459 | /* | |
1460 | * Fill in the sattr for the call. | |
1461 | * Note: SunOS 4.1.2 crashes if the mode isn't initialized! | |
1462 | */ | |
1463 | attr.ia_valid = ATTR_MODE; | |
1464 | attr.ia_mode = S_IFLNK | S_IRWXUGO; | |
1465 | ||
1466 | qsymname.name = symname; | |
1467 | qsymname.len = strlen(symname); | |
1468 | ||
1469 | lock_kernel(); | |
1470 | nfs_begin_data_update(dir); | |
4f390c15 | 1471 | error = NFS_PROTO(dir)->symlink(dir, dentry, &qsymname, &attr); |
1da177e4 | 1472 | nfs_end_data_update(dir); |
d3db90e2 | 1473 | if (!error) |
1da177e4 | 1474 | d_drop(dentry); |
1da177e4 LT |
1475 | unlock_kernel(); |
1476 | return error; | |
1477 | } | |
1478 | ||
1479 | static int | |
1480 | nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) | |
1481 | { | |
1482 | struct inode *inode = old_dentry->d_inode; | |
1483 | int error; | |
1484 | ||
1485 | dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n", | |
1486 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1487 | dentry->d_parent->d_name.name, dentry->d_name.name); | |
1488 | ||
1da177e4 | 1489 | lock_kernel(); |
1da177e4 LT |
1490 | nfs_begin_data_update(dir); |
1491 | nfs_begin_data_update(inode); | |
1492 | error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name); | |
cf809556 TM |
1493 | if (error == 0) { |
1494 | atomic_inc(&inode->i_count); | |
1495 | d_instantiate(dentry, inode); | |
1496 | } | |
1da177e4 LT |
1497 | nfs_end_data_update(inode); |
1498 | nfs_end_data_update(dir); | |
1499 | unlock_kernel(); | |
1500 | return error; | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * RENAME | |
1505 | * FIXME: Some nfsds, like the Linux user space nfsd, may generate a | |
1506 | * different file handle for the same inode after a rename (e.g. when | |
1507 | * moving to a different directory). A fail-safe method to do so would | |
1508 | * be to look up old_dir/old_name, create a link to new_dir/new_name and | |
1509 | * rename the old file using the sillyrename stuff. This way, the original | |
1510 | * file in old_dir will go away when the last process iput()s the inode. | |
1511 | * | |
1512 | * FIXED. | |
1513 | * | |
1514 | * It actually works quite well. One needs to have the possibility for | |
1515 | * at least one ".nfs..." file in each directory the file ever gets | |
1516 | * moved or linked to which happens automagically with the new | |
1517 | * implementation that only depends on the dcache stuff instead of | |
1518 | * using the inode layer | |
1519 | * | |
1520 | * Unfortunately, things are a little more complicated than indicated | |
1521 | * above. For a cross-directory move, we want to make sure we can get | |
1522 | * rid of the old inode after the operation. This means there must be | |
1523 | * no pending writes (if it's a file), and the use count must be 1. | |
1524 | * If these conditions are met, we can drop the dentries before doing | |
1525 | * the rename. | |
1526 | */ | |
1527 | static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1528 | struct inode *new_dir, struct dentry *new_dentry) | |
1529 | { | |
1530 | struct inode *old_inode = old_dentry->d_inode; | |
1531 | struct inode *new_inode = new_dentry->d_inode; | |
1532 | struct dentry *dentry = NULL, *rehash = NULL; | |
1533 | int error = -EBUSY; | |
1534 | ||
1535 | /* | |
1536 | * To prevent any new references to the target during the rename, | |
1537 | * we unhash the dentry and free the inode in advance. | |
1538 | */ | |
1539 | lock_kernel(); | |
1540 | if (!d_unhashed(new_dentry)) { | |
1541 | d_drop(new_dentry); | |
1542 | rehash = new_dentry; | |
1543 | } | |
1544 | ||
1545 | dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n", | |
1546 | old_dentry->d_parent->d_name.name, old_dentry->d_name.name, | |
1547 | new_dentry->d_parent->d_name.name, new_dentry->d_name.name, | |
1548 | atomic_read(&new_dentry->d_count)); | |
1549 | ||
1550 | /* | |
1551 | * First check whether the target is busy ... we can't | |
1552 | * safely do _any_ rename if the target is in use. | |
1553 | * | |
1554 | * For files, make a copy of the dentry and then do a | |
1555 | * silly-rename. If the silly-rename succeeds, the | |
1556 | * copied dentry is hashed and becomes the new target. | |
1557 | */ | |
1558 | if (!new_inode) | |
1559 | goto go_ahead; | |
6fe43f9e TM |
1560 | if (S_ISDIR(new_inode->i_mode)) { |
1561 | error = -EISDIR; | |
1562 | if (!S_ISDIR(old_inode->i_mode)) | |
1563 | goto out; | |
1564 | } else if (atomic_read(&new_dentry->d_count) > 2) { | |
1da177e4 LT |
1565 | int err; |
1566 | /* copy the target dentry's name */ | |
1567 | dentry = d_alloc(new_dentry->d_parent, | |
1568 | &new_dentry->d_name); | |
1569 | if (!dentry) | |
1570 | goto out; | |
1571 | ||
1572 | /* silly-rename the existing target ... */ | |
1573 | err = nfs_sillyrename(new_dir, new_dentry); | |
1574 | if (!err) { | |
1575 | new_dentry = rehash = dentry; | |
1576 | new_inode = NULL; | |
1577 | /* instantiate the replacement target */ | |
1578 | d_instantiate(new_dentry, NULL); | |
1579 | } else if (atomic_read(&new_dentry->d_count) > 1) { | |
1580 | /* dentry still busy? */ | |
1581 | #ifdef NFS_PARANOIA | |
1582 | printk("nfs_rename: target %s/%s busy, d_count=%d\n", | |
1583 | new_dentry->d_parent->d_name.name, | |
1584 | new_dentry->d_name.name, | |
1585 | atomic_read(&new_dentry->d_count)); | |
1586 | #endif | |
1587 | goto out; | |
1588 | } | |
20509f1b TM |
1589 | } else |
1590 | new_inode->i_nlink--; | |
1da177e4 LT |
1591 | |
1592 | go_ahead: | |
1593 | /* | |
1594 | * ... prune child dentries and writebacks if needed. | |
1595 | */ | |
1596 | if (atomic_read(&old_dentry->d_count) > 1) { | |
1597 | nfs_wb_all(old_inode); | |
1598 | shrink_dcache_parent(old_dentry); | |
1599 | } | |
cae7a073 | 1600 | nfs_inode_return_delegation(old_inode); |
1da177e4 | 1601 | |
24174119 TM |
1602 | if (new_inode != NULL) { |
1603 | nfs_inode_return_delegation(new_inode); | |
1da177e4 | 1604 | d_delete(new_dentry); |
24174119 | 1605 | } |
1da177e4 LT |
1606 | |
1607 | nfs_begin_data_update(old_dir); | |
1608 | nfs_begin_data_update(new_dir); | |
1609 | nfs_begin_data_update(old_inode); | |
1610 | error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name, | |
1611 | new_dir, &new_dentry->d_name); | |
5ba7cc48 | 1612 | nfs_mark_for_revalidate(old_inode); |
1da177e4 LT |
1613 | nfs_end_data_update(old_inode); |
1614 | nfs_end_data_update(new_dir); | |
1615 | nfs_end_data_update(old_dir); | |
1616 | out: | |
1617 | if (rehash) | |
1618 | d_rehash(rehash); | |
1619 | if (!error) { | |
1620 | if (!S_ISDIR(old_inode->i_mode)) | |
1621 | d_move(old_dentry, new_dentry); | |
1622 | nfs_renew_times(new_dentry); | |
1623 | nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir)); | |
1624 | } | |
1625 | ||
1626 | /* new dentry created? */ | |
1627 | if (dentry) | |
1628 | dput(dentry); | |
1629 | unlock_kernel(); | |
1630 | return error; | |
1631 | } | |
1632 | ||
cfcea3e8 TM |
1633 | static DEFINE_SPINLOCK(nfs_access_lru_lock); |
1634 | static LIST_HEAD(nfs_access_lru_list); | |
1635 | static atomic_long_t nfs_access_nr_entries; | |
1636 | ||
1c3c07e9 TM |
1637 | static void nfs_access_free_entry(struct nfs_access_entry *entry) |
1638 | { | |
1639 | put_rpccred(entry->cred); | |
1640 | kfree(entry); | |
cfcea3e8 TM |
1641 | smp_mb__before_atomic_dec(); |
1642 | atomic_long_dec(&nfs_access_nr_entries); | |
1643 | smp_mb__after_atomic_dec(); | |
1c3c07e9 TM |
1644 | } |
1645 | ||
979df72e TM |
1646 | int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask) |
1647 | { | |
1648 | LIST_HEAD(head); | |
1649 | struct nfs_inode *nfsi; | |
1650 | struct nfs_access_entry *cache; | |
1651 | ||
1652 | spin_lock(&nfs_access_lru_lock); | |
1653 | restart: | |
1654 | list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) { | |
1655 | struct inode *inode; | |
1656 | ||
1657 | if (nr_to_scan-- == 0) | |
1658 | break; | |
1659 | inode = igrab(&nfsi->vfs_inode); | |
1660 | if (inode == NULL) | |
1661 | continue; | |
1662 | spin_lock(&inode->i_lock); | |
1663 | if (list_empty(&nfsi->access_cache_entry_lru)) | |
1664 | goto remove_lru_entry; | |
1665 | cache = list_entry(nfsi->access_cache_entry_lru.next, | |
1666 | struct nfs_access_entry, lru); | |
1667 | list_move(&cache->lru, &head); | |
1668 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
1669 | if (!list_empty(&nfsi->access_cache_entry_lru)) | |
1670 | list_move_tail(&nfsi->access_cache_inode_lru, | |
1671 | &nfs_access_lru_list); | |
1672 | else { | |
1673 | remove_lru_entry: | |
1674 | list_del_init(&nfsi->access_cache_inode_lru); | |
1675 | clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags); | |
1676 | } | |
1677 | spin_unlock(&inode->i_lock); | |
1678 | iput(inode); | |
1679 | goto restart; | |
1680 | } | |
1681 | spin_unlock(&nfs_access_lru_lock); | |
1682 | while (!list_empty(&head)) { | |
1683 | cache = list_entry(head.next, struct nfs_access_entry, lru); | |
1684 | list_del(&cache->lru); | |
1685 | nfs_access_free_entry(cache); | |
1686 | } | |
1687 | return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure; | |
1688 | } | |
1689 | ||
1c3c07e9 | 1690 | static void __nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 1691 | { |
55296809 | 1692 | struct nfs_inode *nfsi = NFS_I(inode); |
1c3c07e9 TM |
1693 | struct rb_root *root_node = &nfsi->access_cache; |
1694 | struct rb_node *n, *dispose = NULL; | |
1695 | struct nfs_access_entry *entry; | |
1696 | ||
1697 | /* Unhook entries from the cache */ | |
1698 | while ((n = rb_first(root_node)) != NULL) { | |
1699 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
1700 | rb_erase(n, root_node); | |
cfcea3e8 | 1701 | list_del(&entry->lru); |
1c3c07e9 TM |
1702 | n->rb_left = dispose; |
1703 | dispose = n; | |
1704 | } | |
1705 | nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS; | |
1706 | spin_unlock(&inode->i_lock); | |
1da177e4 | 1707 | |
1c3c07e9 TM |
1708 | /* Now kill them all! */ |
1709 | while (dispose != NULL) { | |
1710 | n = dispose; | |
1711 | dispose = n->rb_left; | |
1712 | nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node)); | |
1713 | } | |
1da177e4 LT |
1714 | } |
1715 | ||
1c3c07e9 | 1716 | void nfs_access_zap_cache(struct inode *inode) |
1da177e4 | 1717 | { |
cfcea3e8 TM |
1718 | /* Remove from global LRU init */ |
1719 | if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { | |
1720 | spin_lock(&nfs_access_lru_lock); | |
1721 | list_del_init(&NFS_I(inode)->access_cache_inode_lru); | |
1722 | spin_unlock(&nfs_access_lru_lock); | |
1723 | } | |
1724 | ||
1c3c07e9 TM |
1725 | spin_lock(&inode->i_lock); |
1726 | /* This will release the spinlock */ | |
1727 | __nfs_access_zap_cache(inode); | |
1728 | } | |
1da177e4 | 1729 | |
1c3c07e9 TM |
1730 | static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred) |
1731 | { | |
1732 | struct rb_node *n = NFS_I(inode)->access_cache.rb_node; | |
1733 | struct nfs_access_entry *entry; | |
1734 | ||
1735 | while (n != NULL) { | |
1736 | entry = rb_entry(n, struct nfs_access_entry, rb_node); | |
1737 | ||
1738 | if (cred < entry->cred) | |
1739 | n = n->rb_left; | |
1740 | else if (cred > entry->cred) | |
1741 | n = n->rb_right; | |
1742 | else | |
1743 | return entry; | |
1da177e4 | 1744 | } |
1c3c07e9 TM |
1745 | return NULL; |
1746 | } | |
1747 | ||
1748 | int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res) | |
1749 | { | |
1750 | struct nfs_inode *nfsi = NFS_I(inode); | |
1751 | struct nfs_access_entry *cache; | |
1752 | int err = -ENOENT; | |
1753 | ||
dc59250c | 1754 | spin_lock(&inode->i_lock); |
1c3c07e9 TM |
1755 | if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS) |
1756 | goto out_zap; | |
1757 | cache = nfs_access_search_rbtree(inode, cred); | |
1758 | if (cache == NULL) | |
1759 | goto out; | |
1760 | if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))) | |
1761 | goto out_stale; | |
1762 | res->jiffies = cache->jiffies; | |
1763 | res->cred = cache->cred; | |
1764 | res->mask = cache->mask; | |
cfcea3e8 | 1765 | list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru); |
1c3c07e9 TM |
1766 | err = 0; |
1767 | out: | |
1768 | spin_unlock(&inode->i_lock); | |
1769 | return err; | |
1770 | out_stale: | |
1771 | rb_erase(&cache->rb_node, &nfsi->access_cache); | |
cfcea3e8 | 1772 | list_del(&cache->lru); |
1c3c07e9 TM |
1773 | spin_unlock(&inode->i_lock); |
1774 | nfs_access_free_entry(cache); | |
1775 | return -ENOENT; | |
1776 | out_zap: | |
1777 | /* This will release the spinlock */ | |
1778 | __nfs_access_zap_cache(inode); | |
1779 | return -ENOENT; | |
1780 | } | |
1781 | ||
1782 | static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set) | |
1783 | { | |
cfcea3e8 TM |
1784 | struct nfs_inode *nfsi = NFS_I(inode); |
1785 | struct rb_root *root_node = &nfsi->access_cache; | |
1c3c07e9 TM |
1786 | struct rb_node **p = &root_node->rb_node; |
1787 | struct rb_node *parent = NULL; | |
1788 | struct nfs_access_entry *entry; | |
1789 | ||
1790 | spin_lock(&inode->i_lock); | |
1791 | while (*p != NULL) { | |
1792 | parent = *p; | |
1793 | entry = rb_entry(parent, struct nfs_access_entry, rb_node); | |
1794 | ||
1795 | if (set->cred < entry->cred) | |
1796 | p = &parent->rb_left; | |
1797 | else if (set->cred > entry->cred) | |
1798 | p = &parent->rb_right; | |
1799 | else | |
1800 | goto found; | |
1801 | } | |
1802 | rb_link_node(&set->rb_node, parent, p); | |
1803 | rb_insert_color(&set->rb_node, root_node); | |
cfcea3e8 | 1804 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
dc59250c | 1805 | spin_unlock(&inode->i_lock); |
1c3c07e9 TM |
1806 | return; |
1807 | found: | |
1808 | rb_replace_node(parent, &set->rb_node, root_node); | |
cfcea3e8 TM |
1809 | list_add_tail(&set->lru, &nfsi->access_cache_entry_lru); |
1810 | list_del(&entry->lru); | |
1c3c07e9 TM |
1811 | spin_unlock(&inode->i_lock); |
1812 | nfs_access_free_entry(entry); | |
1813 | } | |
1814 | ||
1815 | void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set) | |
1816 | { | |
1817 | struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL); | |
1818 | if (cache == NULL) | |
1819 | return; | |
1820 | RB_CLEAR_NODE(&cache->rb_node); | |
1da177e4 | 1821 | cache->jiffies = set->jiffies; |
1c3c07e9 | 1822 | cache->cred = get_rpccred(set->cred); |
1da177e4 | 1823 | cache->mask = set->mask; |
1c3c07e9 TM |
1824 | |
1825 | nfs_access_add_rbtree(inode, cache); | |
cfcea3e8 TM |
1826 | |
1827 | /* Update accounting */ | |
1828 | smp_mb__before_atomic_inc(); | |
1829 | atomic_long_inc(&nfs_access_nr_entries); | |
1830 | smp_mb__after_atomic_inc(); | |
1831 | ||
1832 | /* Add inode to global LRU list */ | |
1833 | if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) { | |
1834 | spin_lock(&nfs_access_lru_lock); | |
1835 | list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list); | |
1836 | spin_unlock(&nfs_access_lru_lock); | |
1837 | } | |
1da177e4 LT |
1838 | } |
1839 | ||
1840 | static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask) | |
1841 | { | |
1842 | struct nfs_access_entry cache; | |
1843 | int status; | |
1844 | ||
1845 | status = nfs_access_get_cached(inode, cred, &cache); | |
1846 | if (status == 0) | |
1847 | goto out; | |
1848 | ||
1849 | /* Be clever: ask server to check for all possible rights */ | |
1850 | cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ; | |
1851 | cache.cred = cred; | |
1852 | cache.jiffies = jiffies; | |
1853 | status = NFS_PROTO(inode)->access(inode, &cache); | |
1854 | if (status != 0) | |
1855 | return status; | |
1856 | nfs_access_add_cache(inode, &cache); | |
1857 | out: | |
1858 | if ((cache.mask & mask) == mask) | |
1859 | return 0; | |
1860 | return -EACCES; | |
1861 | } | |
1862 | ||
1863 | int nfs_permission(struct inode *inode, int mask, struct nameidata *nd) | |
1864 | { | |
1865 | struct rpc_cred *cred; | |
1866 | int res = 0; | |
1867 | ||
91d5b470 CL |
1868 | nfs_inc_stats(inode, NFSIOS_VFSACCESS); |
1869 | ||
1da177e4 LT |
1870 | if (mask == 0) |
1871 | goto out; | |
1872 | /* Is this sys_access() ? */ | |
1873 | if (nd != NULL && (nd->flags & LOOKUP_ACCESS)) | |
1874 | goto force_lookup; | |
1875 | ||
1876 | switch (inode->i_mode & S_IFMT) { | |
1877 | case S_IFLNK: | |
1878 | goto out; | |
1879 | case S_IFREG: | |
1880 | /* NFSv4 has atomic_open... */ | |
1881 | if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN) | |
1882 | && nd != NULL | |
1883 | && (nd->flags & LOOKUP_OPEN)) | |
1884 | goto out; | |
1885 | break; | |
1886 | case S_IFDIR: | |
1887 | /* | |
1888 | * Optimize away all write operations, since the server | |
1889 | * will check permissions when we perform the op. | |
1890 | */ | |
1891 | if ((mask & MAY_WRITE) && !(mask & MAY_READ)) | |
1892 | goto out; | |
1893 | } | |
1894 | ||
1895 | force_lookup: | |
1896 | lock_kernel(); | |
1897 | ||
1898 | if (!NFS_PROTO(inode)->access) | |
1899 | goto out_notsup; | |
1900 | ||
1901 | cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0); | |
1902 | if (!IS_ERR(cred)) { | |
1903 | res = nfs_do_access(inode, cred, mask); | |
1904 | put_rpccred(cred); | |
1905 | } else | |
1906 | res = PTR_ERR(cred); | |
1907 | unlock_kernel(); | |
1908 | out: | |
1e7cb3dc CL |
1909 | dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n", |
1910 | inode->i_sb->s_id, inode->i_ino, mask, res); | |
1da177e4 LT |
1911 | return res; |
1912 | out_notsup: | |
1913 | res = nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
1914 | if (res == 0) | |
1915 | res = generic_permission(inode, mask, NULL); | |
1916 | unlock_kernel(); | |
1e7cb3dc | 1917 | goto out; |
1da177e4 LT |
1918 | } |
1919 | ||
1920 | /* | |
1921 | * Local variables: | |
1922 | * version-control: t | |
1923 | * kept-new-versions: 5 | |
1924 | * End: | |
1925 | */ |