blob: 3419c2da9ba9d25e40dcbc67846ba627477cdad2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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>
Chuck Lever873101b2006-08-22 20:06:23 -040033#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/namei.h>
David Howells54ceac42006-08-22 20:06:13 -040035#include <linux/mount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Trond Myklebust4ce79712005-06-22 17:16:21 +000037#include "nfs4_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "delegation.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050039#include "iostat.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define NFS_PARANOIA 1
42/* #define NFS_DEBUG_VERBOSE 1 */
43
44static int nfs_opendir(struct inode *, struct file *);
45static int nfs_readdir(struct file *, void *, filldir_t);
46static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
47static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
48static int nfs_mkdir(struct inode *, struct dentry *, int);
49static int nfs_rmdir(struct inode *, struct dentry *);
50static int nfs_unlink(struct inode *, struct dentry *);
51static int nfs_symlink(struct inode *, struct dentry *, const char *);
52static int nfs_link(struct dentry *, struct inode *, struct dentry *);
53static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
54static int nfs_rename(struct inode *, struct dentry *,
55 struct inode *, struct dentry *);
56static int nfs_fsync_dir(struct file *, struct dentry *, int);
Trond Myklebustf0dd2132005-06-22 17:16:29 +000057static loff_t nfs_llseek_dir(struct file *, loff_t, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080059const struct file_operations nfs_dir_operations = {
Trond Myklebustf0dd2132005-06-22 17:16:29 +000060 .llseek = nfs_llseek_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 .read = generic_read_dir,
62 .readdir = nfs_readdir,
63 .open = nfs_opendir,
64 .release = nfs_release,
65 .fsync = nfs_fsync_dir,
66};
67
68struct inode_operations nfs_dir_inode_operations = {
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
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000083#ifdef CONFIG_NFS_V3
84struct inode_operations nfs3_dir_inode_operations = {
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104#ifdef CONFIG_NFS_V4
105
106static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107struct inode_operations nfs4_dir_inode_operations = {
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,
J. Bruce Fields6b3b5492005-06-22 17:16:22 +0000120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123};
124
125#endif /* CONFIG_NFS_V4 */
126
127/*
128 * Open file
129 */
130static int
131nfs_opendir(struct inode *inode, struct file *filp)
132{
Carsten Otte7451c4f2006-04-19 13:06:37 -0400133 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 lock_kernel();
139 /* Call generic open code in order to cache credentials */
Carsten Otte7451c4f2006-04-19 13:06:37 -0400140 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 unlock_kernel();
142 return res;
143}
144
145typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
146typedef struct {
147 struct file *file;
148 struct page *page;
149 unsigned long page_index;
150 u32 *ptr;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000151 u64 *dir_cookie;
152 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 struct nfs_entry *entry;
154 decode_dirent_t decode;
155 int plus;
156 int error;
157} nfs_readdir_descriptor_t;
158
159/* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
161 *
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
165 * to be.
166 *
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
170 */
171static
172int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
173{
174 struct file *file = desc->file;
175 struct inode *inode = file->f_dentry->d_inode;
176 struct rpc_cred *cred = nfs_file_cred(file);
177 unsigned long timestamp;
178 int error;
179
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500180 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__, (long long)desc->entry->cookie,
182 page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 again:
185 timestamp = jiffies;
186 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
187 NFS_SERVER(inode)->dtsize, desc->plus);
188 if (error < 0) {
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error == -ENOTSUPP && desc->plus) {
191 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Chuck Lever412d5822005-08-18 11:24:11 -0700192 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 desc->plus = 0;
194 goto again;
195 }
196 goto error;
197 }
198 SetPageUptodate(page);
Chuck Leverdc592502005-08-18 11:24:12 -0700199 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700200 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700201 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800204 * through inode->i_mutex or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Trond Myklebusta656db92005-06-22 17:16:21 +0000206 if (page->index == 0)
207 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 unlock_page(page);
209 return 0;
210 error:
211 SetPageError(page);
212 unlock_page(page);
213 nfs_zap_caches(inode);
214 desc->error = error;
215 return -EIO;
216}
217
218static inline
219int dir_decode(nfs_readdir_descriptor_t *desc)
220{
221 u32 *p = desc->ptr;
222 p = desc->decode(p, desc->entry, desc->plus);
223 if (IS_ERR(p))
224 return PTR_ERR(p);
225 desc->ptr = p;
226 return 0;
227}
228
229static inline
230void dir_page_release(nfs_readdir_descriptor_t *desc)
231{
232 kunmap(desc->page);
233 page_cache_release(desc->page);
234 desc->page = NULL;
235 desc->ptr = NULL;
236}
237
238/*
239 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000240 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 *
242 * If the end of the buffer has been reached, return -EAGAIN, if not,
243 * return the offset within the buffer of the next entry to be
244 * read.
245 */
246static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000247int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
249 struct nfs_entry *entry = desc->entry;
250 int loop_count = 0,
251 status;
252
253 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500254 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
255 __FUNCTION__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000256 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 break;
258 if (loop_count++ > 200) {
259 loop_count = 0;
260 schedule();
261 }
262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 return status;
264}
265
266/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000267 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000268 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000269 *
270 * If the end of the buffer has been reached, return -EAGAIN, if not,
271 * return the offset within the buffer of the next entry to be
272 * read.
273 */
274static inline
275int find_dirent_index(nfs_readdir_descriptor_t *desc)
276{
277 struct nfs_entry *entry = desc->entry;
278 int loop_count = 0,
279 status;
280
281 for(;;) {
282 status = dir_decode(desc);
283 if (status)
284 break;
285
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500286 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
287 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000288
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000289 if (desc->file->f_pos == desc->current_index) {
290 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000291 break;
292 }
293 desc->current_index++;
294 if (loop_count++ > 200) {
295 loop_count = 0;
296 schedule();
297 }
298 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000299 return status;
300}
301
302/*
303 * Find the given page, and call find_dirent() or find_dirent_index in
304 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 */
306static inline
307int find_dirent_page(nfs_readdir_descriptor_t *desc)
308{
309 struct inode *inode = desc->file->f_dentry->d_inode;
310 struct page *page;
311 int status;
312
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500313 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
314 __FUNCTION__, desc->page_index,
315 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
317 page = read_cache_page(inode->i_mapping, desc->page_index,
318 (filler_t *)nfs_readdir_filler, desc);
319 if (IS_ERR(page)) {
320 status = PTR_ERR(page);
321 goto out;
322 }
323 if (!PageUptodate(page))
324 goto read_error;
325
326 /* NOTE: Someone else may have changed the READDIRPLUS flag */
327 desc->page = page;
328 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000329 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000330 status = find_dirent(desc);
331 else
332 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (status < 0)
334 dir_page_release(desc);
335 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500336 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return status;
338 read_error:
339 page_cache_release(page);
340 return -EIO;
341}
342
343/*
344 * Recurse through the page cache pages, and return a
345 * filled nfs_entry structure of the next directory entry if possible.
346 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000347 * The target for the search is '*desc->dir_cookie' if non-0,
348 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 */
350static inline
351int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
352{
353 int loop_count = 0;
354 int res;
355
Olivier Galibert00a92642005-06-22 17:16:29 +0000356 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000357 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500358 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
359 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000360 desc->page_index = 0;
361 desc->entry->cookie = desc->entry->prev_cookie = 0;
362 desc->entry->eof = 0;
363 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000364 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500365 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
366 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 for (;;) {
369 res = find_dirent_page(desc);
370 if (res != -EAGAIN)
371 break;
372 /* Align to beginning of next page */
373 desc->page_index ++;
374 if (loop_count++ > 200) {
375 loop_count = 0;
376 schedule();
377 }
378 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500379
380 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return res;
382}
383
384static inline unsigned int dt_type(struct inode *inode)
385{
386 return (inode->i_mode >> 12) & 15;
387}
388
389static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
390
391/*
392 * Once we've found the start of the dirent within a page: fill 'er up...
393 */
394static
395int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
396 filldir_t filldir)
397{
398 struct file *file = desc->file;
399 struct nfs_entry *entry = desc->entry;
400 struct dentry *dentry = NULL;
401 unsigned long fileid;
402 int loop_count = 0,
403 res;
404
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500405 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
406 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407
408 for(;;) {
409 unsigned d_type = DT_UNKNOWN;
410 /* Note: entry->prev_cookie contains the cookie for
411 * retrieving the current dirent on the server */
412 fileid = nfs_fileid_to_ino_t(entry->ino);
413
414 /* Get a dentry if we have one */
415 if (dentry != NULL)
416 dput(dentry);
417 dentry = nfs_readdir_lookup(desc);
418
419 /* Use readdirplus info */
420 if (dentry != NULL && dentry->d_inode != NULL) {
421 d_type = dt_type(dentry->d_inode);
422 fileid = dentry->d_inode->i_ino;
423 }
424
425 res = filldir(dirent, entry->name, entry->len,
Olivier Galibert00a92642005-06-22 17:16:29 +0000426 file->f_pos, fileid, d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (res < 0)
428 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000429 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000430 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 if (dir_decode(desc) != 0) {
432 desc->page_index ++;
433 break;
434 }
435 if (loop_count++ > 200) {
436 loop_count = 0;
437 schedule();
438 }
439 }
440 dir_page_release(desc);
441 if (dentry != NULL)
442 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500443 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
444 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445 return res;
446}
447
448/*
449 * If we cannot find a cookie in our cache, we suspect that this is
450 * because it points to a deleted file, so we ask the server to return
451 * whatever it thinks is the next entry. We then feed this to filldir.
452 * If all goes well, we should then be able to find our way round the
453 * cache on the next call to readdir_search_pagecache();
454 *
455 * NOTE: we cannot add the anonymous page to the pagecache because
456 * the data it contains might not be page aligned. Besides,
457 * we should already have a complete representation of the
458 * directory in the page cache by the time we get here.
459 */
460static inline
461int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
462 filldir_t filldir)
463{
464 struct file *file = desc->file;
465 struct inode *inode = file->f_dentry->d_inode;
466 struct rpc_cred *cred = nfs_file_cred(file);
467 struct page *page = NULL;
468 int status;
469
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500470 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
471 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
473 page = alloc_page(GFP_HIGHUSER);
474 if (!page) {
475 status = -ENOMEM;
476 goto out;
477 }
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000478 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 page,
480 NFS_SERVER(inode)->dtsize,
481 desc->plus);
Chuck Leverdc592502005-08-18 11:24:12 -0700482 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700483 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700484 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 desc->page = page;
486 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
487 if (desc->error >= 0) {
488 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000489 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 } else
491 status = -EIO;
492 if (status < 0)
493 goto out_release;
494
495 status = nfs_do_filldir(desc, dirent, filldir);
496
497 /* Reset read descriptor so it searches the page cache from
498 * the start upon the next call to readdir_search_pagecache() */
499 desc->page_index = 0;
500 desc->entry->cookie = desc->entry->prev_cookie = 0;
501 desc->entry->eof = 0;
502 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500503 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
504 __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505 return status;
506 out_release:
507 dir_page_release(desc);
508 goto out;
509}
510
Olivier Galibert00a92642005-06-22 17:16:29 +0000511/* The file offset position represents the dirent entry number. A
512 last cookie cache takes care of the common case of reading the
513 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 */
515static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
516{
517 struct dentry *dentry = filp->f_dentry;
518 struct inode *inode = dentry->d_inode;
519 nfs_readdir_descriptor_t my_desc,
520 *desc = &my_desc;
521 struct nfs_entry my_entry;
522 struct nfs_fh fh;
523 struct nfs_fattr fattr;
524 long res;
525
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500526 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
527 dentry->d_parent->d_name.name, dentry->d_name.name,
528 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500529 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
530
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 lock_kernel();
532
Trond Myklebust44b11872006-05-25 01:40:59 -0400533 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 if (res < 0) {
535 unlock_kernel();
536 return res;
537 }
538
539 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000540 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000541 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000542 * to either find the entry with the appropriate number or
543 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 */
545 memset(desc, 0, sizeof(*desc));
546
547 desc->file = filp;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000548 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 desc->decode = NFS_PROTO(inode)->decode_dirent;
550 desc->plus = NFS_USE_READDIRPLUS(inode);
551
552 my_entry.cookie = my_entry.prev_cookie = 0;
553 my_entry.eof = 0;
554 my_entry.fh = &fh;
555 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400556 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 desc->entry = &my_entry;
558
559 while(!desc->entry->eof) {
560 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 if (res == -EBADCOOKIE) {
563 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000564 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 /* Or that the server has 'lost' a cookie */
566 res = uncached_readdir(desc, dirent, filldir);
567 if (res >= 0)
568 continue;
569 }
570 res = 0;
571 break;
572 }
573 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700574 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 nfs_zap_caches(inode);
576 desc->plus = 0;
577 desc->entry->eof = 0;
578 continue;
579 }
580 if (res < 0)
581 break;
582
583 res = nfs_do_filldir(desc, dirent, filldir);
584 if (res < 0) {
585 res = 0;
586 break;
587 }
588 }
589 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500590 if (res > 0)
591 res = 0;
592 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
593 dentry->d_parent->d_name.name, dentry->d_name.name,
594 res);
595 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596}
597
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000598loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
599{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800600 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000601 switch (origin) {
602 case 1:
603 offset += filp->f_pos;
604 case 0:
605 if (offset >= 0)
606 break;
607 default:
608 offset = -EINVAL;
609 goto out;
610 }
611 if (offset != filp->f_pos) {
612 filp->f_pos = offset;
613 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
614 }
615out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800616 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000617 return offset;
618}
619
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620/*
621 * All directory operations under NFS are synchronous, so fsync()
622 * is a dummy operation.
623 */
624int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
625{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500626 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
627 dentry->d_parent->d_name.name, dentry->d_name.name,
628 datasync);
629
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 return 0;
631}
632
633/*
634 * A check for whether or not the parent directory has changed.
635 * In the case it has, we assume that the dentries are untrustworthy
636 * and may need to be looked up again.
637 */
638static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
639{
640 if (IS_ROOT(dentry))
641 return 1;
Chuck Lever55296802005-08-18 11:24:09 -0700642 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 || nfs_attribute_timeout(dir))
644 return 0;
645 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
646}
647
648static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
649{
650 dentry->d_fsdata = (void *)verf;
651}
652
653/*
654 * Whenever an NFS operation succeeds, we know that the dentry
655 * is valid, so we update the revalidation timestamp.
656 */
657static inline void nfs_renew_times(struct dentry * dentry)
658{
659 dentry->d_time = jiffies;
660}
661
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400662/*
663 * Return the intent data that applies to this particular path component
664 *
665 * Note that the current set of intents only apply to the very last
666 * component of the path.
667 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
668 */
669static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
670{
671 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
672 return 0;
673 return nd->flags & mask;
674}
675
676/*
677 * Inode and filehandle revalidation for lookups.
678 *
679 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
680 * or if the intent information indicates that we're about to open this
681 * particular file and the "nocto" mount flag is not set.
682 *
683 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684static inline
685int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
686{
687 struct nfs_server *server = NFS_SERVER(inode);
688
689 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400691 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 goto out_force;
693 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400694 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400695 !(server->flags & NFS_MOUNT_NOCTO) &&
696 (S_ISREG(inode->i_mode) ||
697 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 goto out_force;
699 }
700 return nfs_revalidate_inode(server, inode);
701out_force:
702 return __nfs_revalidate_inode(server, inode);
703}
704
705/*
706 * We judge how long we want to trust negative
707 * dentries by looking at the parent inode mtime.
708 *
709 * If parent mtime has changed, we revalidate, else we wait for a
710 * period corresponding to the parent's attribute cache timeout value.
711 */
712static inline
713int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
714 struct nameidata *nd)
715{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400717 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 return 0;
719 return !nfs_check_verifier(dir, dentry);
720}
721
722/*
723 * This is called every time the dcache has a lookup hit,
724 * and we should check whether we can really trust that
725 * lookup.
726 *
727 * NOTE! The hit can be a negative hit too, don't assume
728 * we have an inode!
729 *
730 * If the parent directory is seen to have changed, we throw out the
731 * cached dentry and do a new lookup.
732 */
733static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
734{
735 struct inode *dir;
736 struct inode *inode;
737 struct dentry *parent;
738 int error;
739 struct nfs_fh fhandle;
740 struct nfs_fattr fattr;
741 unsigned long verifier;
742
743 parent = dget_parent(dentry);
744 lock_kernel();
745 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500746 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 inode = dentry->d_inode;
748
749 if (!inode) {
750 if (nfs_neg_need_reval(dir, dentry, nd))
751 goto out_bad;
752 goto out_valid;
753 }
754
755 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500756 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
757 __FUNCTION__, dentry->d_parent->d_name.name,
758 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 goto out_bad;
760 }
761
762 /* Revalidate parent directory attribute cache */
763 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
764 goto out_zap_parent;
765
766 /* Force a full look up iff the parent directory has changed */
767 if (nfs_check_verifier(dir, dentry)) {
768 if (nfs_lookup_verify_inode(inode, nd))
769 goto out_zap_parent;
770 goto out_valid;
771 }
772
773 if (NFS_STALE(inode))
774 goto out_bad;
775
776 verifier = nfs_save_change_attribute(dir);
777 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
778 if (error)
779 goto out_bad;
780 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
781 goto out_bad;
782 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
783 goto out_bad;
784
785 nfs_renew_times(dentry);
786 nfs_set_verifier(dentry, verifier);
787 out_valid:
788 unlock_kernel();
789 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500790 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
791 __FUNCTION__, dentry->d_parent->d_name.name,
792 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 return 1;
794out_zap_parent:
795 nfs_zap_caches(dir);
796 out_bad:
797 NFS_CACHEINV(dir);
798 if (inode && S_ISDIR(inode->i_mode)) {
799 /* Purge readdir caches. */
800 nfs_zap_caches(inode);
801 /* If we have submounts, don't unhash ! */
802 if (have_submounts(dentry))
803 goto out_valid;
804 shrink_dcache_parent(dentry);
805 }
806 d_drop(dentry);
807 unlock_kernel();
808 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500809 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
810 __FUNCTION__, dentry->d_parent->d_name.name,
811 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 return 0;
813}
814
815/*
816 * This is called from dput() when d_count is going to 0.
817 */
818static int nfs_dentry_delete(struct dentry *dentry)
819{
820 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
821 dentry->d_parent->d_name.name, dentry->d_name.name,
822 dentry->d_flags);
823
824 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
825 /* Unhash it, so that ->d_iput() would be called */
826 return 1;
827 }
828 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
829 /* Unhash it, so that ancestors of killed async unlink
830 * files will be cleaned up during umount */
831 return 1;
832 }
833 return 0;
834
835}
836
837/*
838 * Called when the dentry loses inode.
839 * We use it to clean up silly-renamed files.
840 */
841static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
842{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700843 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
845 lock_kernel();
846 inode->i_nlink--;
847 nfs_complete_unlink(dentry);
848 unlock_kernel();
849 }
850 /* When creating a negative dentry, we want to renew d_time */
851 nfs_renew_times(dentry);
852 iput(inode);
853}
854
855struct dentry_operations nfs_dentry_operations = {
856 .d_revalidate = nfs_lookup_revalidate,
857 .d_delete = nfs_dentry_delete,
858 .d_iput = nfs_dentry_iput,
859};
860
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400861/*
862 * Use intent information to check whether or not we're going to do
863 * an O_EXCL create using this path component.
864 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865static inline
866int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
867{
868 if (NFS_PROTO(dir)->version == 2)
869 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400870 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return 0;
872 return (nd->intent.open.flags & O_EXCL) != 0;
873}
874
David Howells54ceac42006-08-22 20:06:13 -0400875static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
876 struct nfs_fh *fh, struct nfs_fattr *fattr)
Trond Myklebust55a97592006-06-09 09:34:19 -0400877{
878 struct nfs_server *server = NFS_SERVER(dir);
879
880 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
881 /* Revalidate fsid on root dir */
David Howells54ceac42006-08-22 20:06:13 -0400882 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
Trond Myklebust55a97592006-06-09 09:34:19 -0400883 return 0;
884}
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
887{
888 struct dentry *res;
889 struct inode *inode = NULL;
890 int error;
891 struct nfs_fh fhandle;
892 struct nfs_fattr fattr;
893
894 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
895 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500896 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897
898 res = ERR_PTR(-ENAMETOOLONG);
899 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
900 goto out;
901
902 res = ERR_PTR(-ENOMEM);
903 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
904
905 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906
Trond Myklebustfd684072006-09-05 12:27:44 -0400907 /*
908 * If we're doing an exclusive create, optimize away the lookup
909 * but don't hash the dentry.
910 */
911 if (nfs_is_exclusive_create(dir, nd)) {
912 d_instantiate(dentry, NULL);
913 res = NULL;
914 goto out_unlock;
915 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
917 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
918 if (error == -ENOENT)
919 goto no_entry;
920 if (error < 0) {
921 res = ERR_PTR(error);
922 goto out_unlock;
923 }
David Howells54ceac42006-08-22 20:06:13 -0400924 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
Trond Myklebust55a97592006-06-09 09:34:19 -0400925 if (error < 0) {
926 res = ERR_PTR(error);
927 goto out_unlock;
928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500930 res = (struct dentry *)inode;
931 if (IS_ERR(res))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 goto out_unlock;
David Howells54ceac42006-08-22 20:06:13 -0400933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400935 res = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 if (res != NULL)
937 dentry = res;
938 nfs_renew_times(dentry);
939 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
940out_unlock:
941 unlock_kernel();
942out:
943 return res;
944}
945
946#ifdef CONFIG_NFS_V4
947static int nfs_open_revalidate(struct dentry *, struct nameidata *);
948
949struct dentry_operations nfs4_dentry_operations = {
950 .d_revalidate = nfs_open_revalidate,
951 .d_delete = nfs_dentry_delete,
952 .d_iput = nfs_dentry_iput,
953};
954
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400955/*
956 * Use intent information to determine whether we need to substitute
957 * the NFSv4-style stateful OPEN for the LOOKUP call
958 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959static int is_atomic_open(struct inode *dir, struct nameidata *nd)
960{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400961 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 return 0;
963 /* NFS does not (yet) have a stateful open for directories */
964 if (nd->flags & LOOKUP_DIRECTORY)
965 return 0;
966 /* Are we trying to write to a read only partition? */
967 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
968 return 0;
969 return 1;
970}
971
972static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
973{
974 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 int error;
976
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500977 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
978 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 /* Check that we are indeed trying to open this file */
981 if (!is_atomic_open(dir, nd))
982 goto no_open;
983
984 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
985 res = ERR_PTR(-ENAMETOOLONG);
986 goto out;
987 }
988 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
989
990 /* Let vfs_create() deal with O_EXCL */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700991 if (nd->intent.open.flags & O_EXCL) {
992 d_add(dentry, NULL);
993 goto out;
994 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 /* Open the file on the server */
997 lock_kernel();
998 /* Revalidate parent directory attribute cache */
999 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1000 if (error < 0) {
1001 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -07001002 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 goto out;
1004 }
1005
1006 if (nd->intent.open.flags & O_CREAT) {
1007 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001008 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 nfs_end_data_update(dir);
1010 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -07001011 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001013 if (IS_ERR(res)) {
1014 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 switch (error) {
1016 /* Make a negative dentry */
1017 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001018 res = NULL;
1019 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001021 case -EISDIR:
1022 case -ENOTDIR:
1023 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 case -ELOOP:
1025 if (!(nd->intent.open.flags & O_NOFOLLOW))
1026 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 /* case -EINVAL: */
1028 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 goto out;
1030 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001031 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 dentry = res;
1033 nfs_renew_times(dentry);
1034 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1035out:
1036 return res;
1037no_open:
1038 return nfs_lookup(dir, dentry, nd);
1039}
1040
1041static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1042{
1043 struct dentry *parent = NULL;
1044 struct inode *inode = dentry->d_inode;
1045 struct inode *dir;
1046 unsigned long verifier;
1047 int openflags, ret = 0;
1048
1049 parent = dget_parent(dentry);
1050 dir = parent->d_inode;
1051 if (!is_atomic_open(dir, nd))
1052 goto no_open;
1053 /* We can't create new files in nfs_open_revalidate(), so we
1054 * optimize away revalidation of negative dentries.
1055 */
1056 if (inode == NULL)
1057 goto out;
1058 /* NFS only supports OPEN on regular files */
1059 if (!S_ISREG(inode->i_mode))
1060 goto no_open;
1061 openflags = nd->intent.open.flags;
1062 /* We cannot do exclusive creation on a positive dentry */
1063 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1064 goto no_open;
1065 /* We can't create new files, or truncate existing ones here */
1066 openflags &= ~(O_CREAT|O_TRUNC);
1067
1068 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001069 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 * operations that change the directory. We therefore save the
1071 * change attribute *before* we do the RPC call.
1072 */
1073 lock_kernel();
1074 verifier = nfs_save_change_attribute(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001075 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (!ret)
1077 nfs_set_verifier(dentry, verifier);
1078 unlock_kernel();
1079out:
1080 dput(parent);
1081 if (!ret)
1082 d_drop(dentry);
1083 return ret;
1084no_open:
1085 dput(parent);
1086 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1087 return 1;
1088 return nfs_lookup_revalidate(dentry, nd);
1089}
1090#endif /* CONFIG_NFSV4 */
1091
1092static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1093{
1094 struct dentry *parent = desc->file->f_dentry;
1095 struct inode *dir = parent->d_inode;
1096 struct nfs_entry *entry = desc->entry;
1097 struct dentry *dentry, *alias;
1098 struct qstr name = {
1099 .name = entry->name,
1100 .len = entry->len,
1101 };
1102 struct inode *inode;
1103
1104 switch (name.len) {
1105 case 2:
1106 if (name.name[0] == '.' && name.name[1] == '.')
1107 return dget_parent(parent);
1108 break;
1109 case 1:
1110 if (name.name[0] == '.')
1111 return dget(parent);
1112 }
1113 name.hash = full_name_hash(name.name, name.len);
1114 dentry = d_lookup(parent, &name);
1115 if (dentry != NULL)
1116 return dentry;
1117 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1118 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001119 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 dentry = d_alloc(parent, &name);
1121 if (dentry == NULL)
1122 return NULL;
1123 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1124 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001125 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126 dput(dentry);
1127 return NULL;
1128 }
David Howells54ceac42006-08-22 20:06:13 -04001129
1130 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 if (alias != NULL) {
1132 dput(dentry);
1133 dentry = alias;
1134 }
David Howells54ceac42006-08-22 20:06:13 -04001135
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136 nfs_renew_times(dentry);
1137 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1138 return dentry;
1139}
1140
1141/*
1142 * Code common to create, mkdir, and mknod.
1143 */
1144int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1145 struct nfs_fattr *fattr)
1146{
1147 struct inode *inode;
1148 int error = -EACCES;
1149
1150 /* We may have been initialized further down */
1151 if (dentry->d_inode)
1152 return 0;
1153 if (fhandle->size == 0) {
1154 struct inode *dir = dentry->d_parent->d_inode;
1155 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1156 if (error)
Chuck Lever4f390c12006-08-22 20:06:22 -04001157 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 }
1159 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1160 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001161 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 if (error < 0)
Chuck Lever4f390c12006-08-22 20:06:22 -04001163 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001166 error = PTR_ERR(inode);
1167 if (IS_ERR(inode))
Chuck Lever4f390c12006-08-22 20:06:22 -04001168 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 d_instantiate(dentry, inode);
Trond Myklebustfd684072006-09-05 12:27:44 -04001170 if (d_unhashed(dentry))
1171 d_rehash(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175/*
1176 * Following a failed create operation, we drop the dentry rather
1177 * than retain a negative dentry. This avoids a problem in the event
1178 * that the operation succeeded on the server, but an error in the
1179 * reply path made it appear to have failed.
1180 */
1181static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1182 struct nameidata *nd)
1183{
1184 struct iattr attr;
1185 int error;
1186 int open_flags = 0;
1187
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001188 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1189 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
1191 attr.ia_mode = mode;
1192 attr.ia_valid = ATTR_MODE;
1193
1194 if (nd && (nd->flags & LOOKUP_CREATE))
1195 open_flags = nd->intent.open.flags;
1196
1197 lock_kernel();
1198 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001199 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 nfs_end_data_update(dir);
1201 if (error != 0)
1202 goto out_err;
1203 nfs_renew_times(dentry);
1204 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1205 unlock_kernel();
1206 return 0;
1207out_err:
1208 unlock_kernel();
1209 d_drop(dentry);
1210 return error;
1211}
1212
1213/*
1214 * See comments for nfs_proc_create regarding failed operations.
1215 */
1216static int
1217nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1218{
1219 struct iattr attr;
1220 int status;
1221
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001222 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1223 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
1225 if (!new_valid_dev(rdev))
1226 return -EINVAL;
1227
1228 attr.ia_mode = mode;
1229 attr.ia_valid = ATTR_MODE;
1230
1231 lock_kernel();
1232 nfs_begin_data_update(dir);
1233 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1234 nfs_end_data_update(dir);
1235 if (status != 0)
1236 goto out_err;
1237 nfs_renew_times(dentry);
1238 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1239 unlock_kernel();
1240 return 0;
1241out_err:
1242 unlock_kernel();
1243 d_drop(dentry);
1244 return status;
1245}
1246
1247/*
1248 * See comments for nfs_proc_create regarding failed operations.
1249 */
1250static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1251{
1252 struct iattr attr;
1253 int error;
1254
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001255 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1256 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
1258 attr.ia_valid = ATTR_MODE;
1259 attr.ia_mode = mode | S_IFDIR;
1260
1261 lock_kernel();
1262 nfs_begin_data_update(dir);
1263 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1264 nfs_end_data_update(dir);
1265 if (error != 0)
1266 goto out_err;
1267 nfs_renew_times(dentry);
1268 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1269 unlock_kernel();
1270 return 0;
1271out_err:
1272 d_drop(dentry);
1273 unlock_kernel();
1274 return error;
1275}
1276
1277static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1278{
1279 int error;
1280
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001281 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1282 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283
1284 lock_kernel();
1285 nfs_begin_data_update(dir);
1286 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1287 /* Ensure the VFS deletes this inode */
1288 if (error == 0 && dentry->d_inode != NULL)
1289 dentry->d_inode->i_nlink = 0;
1290 nfs_end_data_update(dir);
1291 unlock_kernel();
1292
1293 return error;
1294}
1295
1296static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1297{
1298 static unsigned int sillycounter;
1299 const int i_inosize = sizeof(dir->i_ino)*2;
1300 const int countersize = sizeof(sillycounter)*2;
1301 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1302 char silly[slen+1];
1303 struct qstr qsilly;
1304 struct dentry *sdentry;
1305 int error = -EIO;
1306
1307 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1308 dentry->d_parent->d_name.name, dentry->d_name.name,
1309 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001310 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312#ifdef NFS_PARANOIA
1313if (!dentry->d_inode)
1314printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1315dentry->d_parent->d_name.name, dentry->d_name.name);
1316#endif
1317 /*
1318 * We don't allow a dentry to be silly-renamed twice.
1319 */
1320 error = -EBUSY;
1321 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1322 goto out;
1323
1324 sprintf(silly, ".nfs%*.*lx",
1325 i_inosize, i_inosize, dentry->d_inode->i_ino);
1326
Trond Myklebust34ea8182005-11-04 15:35:02 -05001327 /* Return delegation in anticipation of the rename */
1328 nfs_inode_return_delegation(dentry->d_inode);
1329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 sdentry = NULL;
1331 do {
1332 char *suffix = silly + slen - countersize;
1333
1334 dput(sdentry);
1335 sillycounter++;
1336 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1337
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001338 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1339 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
1341 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1342 /*
1343 * N.B. Better to return EBUSY here ... it could be
1344 * dangerous to delete the file while it's in use.
1345 */
1346 if (IS_ERR(sdentry))
1347 goto out;
1348 } while(sdentry->d_inode != NULL); /* need negative lookup */
1349
1350 qsilly.name = silly;
1351 qsilly.len = strlen(silly);
1352 nfs_begin_data_update(dir);
1353 if (dentry->d_inode) {
1354 nfs_begin_data_update(dentry->d_inode);
1355 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1356 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001357 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358 nfs_end_data_update(dentry->d_inode);
1359 } else
1360 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1361 dir, &qsilly);
1362 nfs_end_data_update(dir);
1363 if (!error) {
1364 nfs_renew_times(dentry);
1365 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1366 d_move(dentry, sdentry);
1367 error = nfs_async_unlink(dentry);
1368 /* If we return 0 we don't unlink */
1369 }
1370 dput(sdentry);
1371out:
1372 return error;
1373}
1374
1375/*
1376 * Remove a file after making sure there are no pending writes,
1377 * and after checking that the file has only one user.
1378 *
1379 * We invalidate the attribute cache and free the inode prior to the operation
1380 * to avoid possible races if the server reuses the inode.
1381 */
1382static int nfs_safe_remove(struct dentry *dentry)
1383{
1384 struct inode *dir = dentry->d_parent->d_inode;
1385 struct inode *inode = dentry->d_inode;
1386 int error = -EBUSY;
1387
1388 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1389 dentry->d_parent->d_name.name, dentry->d_name.name);
1390
1391 /* If the dentry was sillyrenamed, we simply call d_delete() */
1392 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1393 error = 0;
1394 goto out;
1395 }
1396
1397 nfs_begin_data_update(dir);
1398 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001399 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 nfs_begin_data_update(inode);
1401 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1402 /* The VFS may want to delete this inode */
1403 if (error == 0)
1404 inode->i_nlink--;
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001405 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 nfs_end_data_update(inode);
1407 } else
1408 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1409 nfs_end_data_update(dir);
1410out:
1411 return error;
1412}
1413
1414/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1415 * belongs to an active ".nfs..." file and we return -EBUSY.
1416 *
1417 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1418 */
1419static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1420{
1421 int error;
1422 int need_rehash = 0;
1423
1424 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1425 dir->i_ino, dentry->d_name.name);
1426
1427 lock_kernel();
1428 spin_lock(&dcache_lock);
1429 spin_lock(&dentry->d_lock);
1430 if (atomic_read(&dentry->d_count) > 1) {
1431 spin_unlock(&dentry->d_lock);
1432 spin_unlock(&dcache_lock);
1433 error = nfs_sillyrename(dir, dentry);
1434 unlock_kernel();
1435 return error;
1436 }
1437 if (!d_unhashed(dentry)) {
1438 __d_drop(dentry);
1439 need_rehash = 1;
1440 }
1441 spin_unlock(&dentry->d_lock);
1442 spin_unlock(&dcache_lock);
1443 error = nfs_safe_remove(dentry);
1444 if (!error) {
1445 nfs_renew_times(dentry);
1446 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1447 } else if (need_rehash)
1448 d_rehash(dentry);
1449 unlock_kernel();
1450 return error;
1451}
1452
Chuck Lever873101b2006-08-22 20:06:23 -04001453/*
1454 * To create a symbolic link, most file systems instantiate a new inode,
1455 * add a page to it containing the path, then write it out to the disk
1456 * using prepare_write/commit_write.
1457 *
1458 * Unfortunately the NFS client can't create the in-core inode first
1459 * because it needs a file handle to create an in-core inode (see
1460 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1461 * symlink request has completed on the server.
1462 *
1463 * So instead we allocate a raw page, copy the symname into it, then do
1464 * the SYMLINK request with the page as the buffer. If it succeeds, we
1465 * now have a new file handle and can instantiate an in-core NFS inode
1466 * and move the raw page into its mapping.
1467 */
1468static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469{
Chuck Lever873101b2006-08-22 20:06:23 -04001470 struct pagevec lru_pvec;
1471 struct page *page;
1472 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001474 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 int error;
1476
1477 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1478 dir->i_ino, dentry->d_name.name, symname);
1479
Chuck Lever873101b2006-08-22 20:06:23 -04001480 if (pathlen > PAGE_SIZE)
1481 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
Chuck Lever873101b2006-08-22 20:06:23 -04001483 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1484 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485
1486 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001487
1488 page = alloc_page(GFP_KERNEL);
1489 if (!page) {
1490 unlock_kernel();
1491 return -ENOMEM;
1492 }
1493
1494 kaddr = kmap_atomic(page, KM_USER0);
1495 memcpy(kaddr, symname, pathlen);
1496 if (pathlen < PAGE_SIZE)
1497 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1498 kunmap_atomic(kaddr, KM_USER0);
1499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 nfs_begin_data_update(dir);
Chuck Lever94a6d752006-08-22 20:06:23 -04001501 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 nfs_end_data_update(dir);
Chuck Lever873101b2006-08-22 20:06:23 -04001503 if (error != 0) {
1504 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1505 dir->i_sb->s_id, dir->i_ino,
1506 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001508 __free_page(page);
1509 unlock_kernel();
1510 return error;
1511 }
1512
1513 /*
1514 * No big deal if we can't add this page to the page cache here.
1515 * READLINK will get the missing page from the server if needed.
1516 */
1517 pagevec_init(&lru_pvec, 0);
1518 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1519 GFP_KERNEL)) {
1520 if (!pagevec_add(&lru_pvec, page))
1521 __pagevec_lru_add(&lru_pvec);
1522 SetPageUptodate(page);
1523 unlock_page(page);
1524 } else
1525 __free_page(page);
1526
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001528 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529}
1530
1531static int
1532nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1533{
1534 struct inode *inode = old_dentry->d_inode;
1535 int error;
1536
1537 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1538 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1539 dentry->d_parent->d_name.name, dentry->d_name.name);
1540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 nfs_begin_data_update(dir);
1543 nfs_begin_data_update(inode);
1544 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001545 if (error == 0) {
1546 atomic_inc(&inode->i_count);
1547 d_instantiate(dentry, inode);
1548 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 nfs_end_data_update(inode);
1550 nfs_end_data_update(dir);
1551 unlock_kernel();
1552 return error;
1553}
1554
1555/*
1556 * RENAME
1557 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1558 * different file handle for the same inode after a rename (e.g. when
1559 * moving to a different directory). A fail-safe method to do so would
1560 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1561 * rename the old file using the sillyrename stuff. This way, the original
1562 * file in old_dir will go away when the last process iput()s the inode.
1563 *
1564 * FIXED.
1565 *
1566 * It actually works quite well. One needs to have the possibility for
1567 * at least one ".nfs..." file in each directory the file ever gets
1568 * moved or linked to which happens automagically with the new
1569 * implementation that only depends on the dcache stuff instead of
1570 * using the inode layer
1571 *
1572 * Unfortunately, things are a little more complicated than indicated
1573 * above. For a cross-directory move, we want to make sure we can get
1574 * rid of the old inode after the operation. This means there must be
1575 * no pending writes (if it's a file), and the use count must be 1.
1576 * If these conditions are met, we can drop the dentries before doing
1577 * the rename.
1578 */
1579static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1580 struct inode *new_dir, struct dentry *new_dentry)
1581{
1582 struct inode *old_inode = old_dentry->d_inode;
1583 struct inode *new_inode = new_dentry->d_inode;
1584 struct dentry *dentry = NULL, *rehash = NULL;
1585 int error = -EBUSY;
1586
1587 /*
1588 * To prevent any new references to the target during the rename,
1589 * we unhash the dentry and free the inode in advance.
1590 */
1591 lock_kernel();
1592 if (!d_unhashed(new_dentry)) {
1593 d_drop(new_dentry);
1594 rehash = new_dentry;
1595 }
1596
1597 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1598 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1599 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1600 atomic_read(&new_dentry->d_count));
1601
1602 /*
1603 * First check whether the target is busy ... we can't
1604 * safely do _any_ rename if the target is in use.
1605 *
1606 * For files, make a copy of the dentry and then do a
1607 * silly-rename. If the silly-rename succeeds, the
1608 * copied dentry is hashed and becomes the new target.
1609 */
1610 if (!new_inode)
1611 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001612 if (S_ISDIR(new_inode->i_mode)) {
1613 error = -EISDIR;
1614 if (!S_ISDIR(old_inode->i_mode))
1615 goto out;
1616 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 int err;
1618 /* copy the target dentry's name */
1619 dentry = d_alloc(new_dentry->d_parent,
1620 &new_dentry->d_name);
1621 if (!dentry)
1622 goto out;
1623
1624 /* silly-rename the existing target ... */
1625 err = nfs_sillyrename(new_dir, new_dentry);
1626 if (!err) {
1627 new_dentry = rehash = dentry;
1628 new_inode = NULL;
1629 /* instantiate the replacement target */
1630 d_instantiate(new_dentry, NULL);
1631 } else if (atomic_read(&new_dentry->d_count) > 1) {
1632 /* dentry still busy? */
1633#ifdef NFS_PARANOIA
1634 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1635 new_dentry->d_parent->d_name.name,
1636 new_dentry->d_name.name,
1637 atomic_read(&new_dentry->d_count));
1638#endif
1639 goto out;
1640 }
Trond Myklebust20509f12005-08-25 16:25:34 -07001641 } else
1642 new_inode->i_nlink--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643
1644go_ahead:
1645 /*
1646 * ... prune child dentries and writebacks if needed.
1647 */
1648 if (atomic_read(&old_dentry->d_count) > 1) {
1649 nfs_wb_all(old_inode);
1650 shrink_dcache_parent(old_dentry);
1651 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001652 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653
Trond Myklebust24174112006-01-03 09:55:33 +01001654 if (new_inode != NULL) {
1655 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658
1659 nfs_begin_data_update(old_dir);
1660 nfs_begin_data_update(new_dir);
1661 nfs_begin_data_update(old_inode);
1662 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1663 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001664 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 nfs_end_data_update(old_inode);
1666 nfs_end_data_update(new_dir);
1667 nfs_end_data_update(old_dir);
1668out:
1669 if (rehash)
1670 d_rehash(rehash);
1671 if (!error) {
1672 if (!S_ISDIR(old_inode->i_mode))
1673 d_move(old_dentry, new_dentry);
1674 nfs_renew_times(new_dentry);
1675 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1676 }
1677
1678 /* new dentry created? */
1679 if (dentry)
1680 dput(dentry);
1681 unlock_kernel();
1682 return error;
1683}
1684
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001685static DEFINE_SPINLOCK(nfs_access_lru_lock);
1686static LIST_HEAD(nfs_access_lru_list);
1687static atomic_long_t nfs_access_nr_entries;
1688
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001689static void nfs_access_free_entry(struct nfs_access_entry *entry)
1690{
1691 put_rpccred(entry->cred);
1692 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001693 smp_mb__before_atomic_dec();
1694 atomic_long_dec(&nfs_access_nr_entries);
1695 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001696}
1697
Trond Myklebust979df722006-07-25 11:28:19 -04001698int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1699{
1700 LIST_HEAD(head);
1701 struct nfs_inode *nfsi;
1702 struct nfs_access_entry *cache;
1703
1704 spin_lock(&nfs_access_lru_lock);
1705restart:
1706 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1707 struct inode *inode;
1708
1709 if (nr_to_scan-- == 0)
1710 break;
1711 inode = igrab(&nfsi->vfs_inode);
1712 if (inode == NULL)
1713 continue;
1714 spin_lock(&inode->i_lock);
1715 if (list_empty(&nfsi->access_cache_entry_lru))
1716 goto remove_lru_entry;
1717 cache = list_entry(nfsi->access_cache_entry_lru.next,
1718 struct nfs_access_entry, lru);
1719 list_move(&cache->lru, &head);
1720 rb_erase(&cache->rb_node, &nfsi->access_cache);
1721 if (!list_empty(&nfsi->access_cache_entry_lru))
1722 list_move_tail(&nfsi->access_cache_inode_lru,
1723 &nfs_access_lru_list);
1724 else {
1725remove_lru_entry:
1726 list_del_init(&nfsi->access_cache_inode_lru);
1727 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1728 }
1729 spin_unlock(&inode->i_lock);
1730 iput(inode);
1731 goto restart;
1732 }
1733 spin_unlock(&nfs_access_lru_lock);
1734 while (!list_empty(&head)) {
1735 cache = list_entry(head.next, struct nfs_access_entry, lru);
1736 list_del(&cache->lru);
1737 nfs_access_free_entry(cache);
1738 }
1739 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1740}
1741
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001742static void __nfs_access_zap_cache(struct inode *inode)
1743{
1744 struct nfs_inode *nfsi = NFS_I(inode);
1745 struct rb_root *root_node = &nfsi->access_cache;
1746 struct rb_node *n, *dispose = NULL;
1747 struct nfs_access_entry *entry;
1748
1749 /* Unhook entries from the cache */
1750 while ((n = rb_first(root_node)) != NULL) {
1751 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1752 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001753 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001754 n->rb_left = dispose;
1755 dispose = n;
1756 }
1757 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1758 spin_unlock(&inode->i_lock);
1759
1760 /* Now kill them all! */
1761 while (dispose != NULL) {
1762 n = dispose;
1763 dispose = n->rb_left;
1764 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1765 }
1766}
1767
1768void nfs_access_zap_cache(struct inode *inode)
1769{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001770 /* Remove from global LRU init */
1771 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1772 spin_lock(&nfs_access_lru_lock);
1773 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1774 spin_unlock(&nfs_access_lru_lock);
1775 }
1776
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001777 spin_lock(&inode->i_lock);
1778 /* This will release the spinlock */
1779 __nfs_access_zap_cache(inode);
1780}
1781
1782static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1783{
1784 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1785 struct nfs_access_entry *entry;
1786
1787 while (n != NULL) {
1788 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1789
1790 if (cred < entry->cred)
1791 n = n->rb_left;
1792 else if (cred > entry->cred)
1793 n = n->rb_right;
1794 else
1795 return entry;
1796 }
1797 return NULL;
1798}
1799
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1801{
Chuck Lever55296802005-08-18 11:24:09 -07001802 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001803 struct nfs_access_entry *cache;
1804 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001806 spin_lock(&inode->i_lock);
1807 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1808 goto out_zap;
1809 cache = nfs_access_search_rbtree(inode, cred);
1810 if (cache == NULL)
1811 goto out;
1812 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1813 goto out_stale;
1814 res->jiffies = cache->jiffies;
1815 res->cred = cache->cred;
1816 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001817 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001818 err = 0;
1819out:
1820 spin_unlock(&inode->i_lock);
1821 return err;
1822out_stale:
1823 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001824 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001825 spin_unlock(&inode->i_lock);
1826 nfs_access_free_entry(cache);
1827 return -ENOENT;
1828out_zap:
1829 /* This will release the spinlock */
1830 __nfs_access_zap_cache(inode);
1831 return -ENOENT;
1832}
1833
1834static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1835{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001836 struct nfs_inode *nfsi = NFS_I(inode);
1837 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001838 struct rb_node **p = &root_node->rb_node;
1839 struct rb_node *parent = NULL;
1840 struct nfs_access_entry *entry;
1841
1842 spin_lock(&inode->i_lock);
1843 while (*p != NULL) {
1844 parent = *p;
1845 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1846
1847 if (set->cred < entry->cred)
1848 p = &parent->rb_left;
1849 else if (set->cred > entry->cred)
1850 p = &parent->rb_right;
1851 else
1852 goto found;
1853 }
1854 rb_link_node(&set->rb_node, parent, p);
1855 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001856 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001857 spin_unlock(&inode->i_lock);
1858 return;
1859found:
1860 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001861 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1862 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001863 spin_unlock(&inode->i_lock);
1864 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865}
1866
1867void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1868{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001869 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1870 if (cache == NULL)
1871 return;
1872 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001874 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001876
1877 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001878
1879 /* Update accounting */
1880 smp_mb__before_atomic_inc();
1881 atomic_long_inc(&nfs_access_nr_entries);
1882 smp_mb__after_atomic_inc();
1883
1884 /* Add inode to global LRU list */
1885 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1886 spin_lock(&nfs_access_lru_lock);
1887 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1888 spin_unlock(&nfs_access_lru_lock);
1889 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890}
1891
1892static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1893{
1894 struct nfs_access_entry cache;
1895 int status;
1896
1897 status = nfs_access_get_cached(inode, cred, &cache);
1898 if (status == 0)
1899 goto out;
1900
1901 /* Be clever: ask server to check for all possible rights */
1902 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1903 cache.cred = cred;
1904 cache.jiffies = jiffies;
1905 status = NFS_PROTO(inode)->access(inode, &cache);
1906 if (status != 0)
1907 return status;
1908 nfs_access_add_cache(inode, &cache);
1909out:
1910 if ((cache.mask & mask) == mask)
1911 return 0;
1912 return -EACCES;
1913}
1914
1915int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1916{
1917 struct rpc_cred *cred;
1918 int res = 0;
1919
Chuck Lever91d5b472006-03-20 13:44:14 -05001920 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1921
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 if (mask == 0)
1923 goto out;
1924 /* Is this sys_access() ? */
1925 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1926 goto force_lookup;
1927
1928 switch (inode->i_mode & S_IFMT) {
1929 case S_IFLNK:
1930 goto out;
1931 case S_IFREG:
1932 /* NFSv4 has atomic_open... */
1933 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1934 && nd != NULL
1935 && (nd->flags & LOOKUP_OPEN))
1936 goto out;
1937 break;
1938 case S_IFDIR:
1939 /*
1940 * Optimize away all write operations, since the server
1941 * will check permissions when we perform the op.
1942 */
1943 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1944 goto out;
1945 }
1946
1947force_lookup:
1948 lock_kernel();
1949
1950 if (!NFS_PROTO(inode)->access)
1951 goto out_notsup;
1952
1953 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1954 if (!IS_ERR(cred)) {
1955 res = nfs_do_access(inode, cred, mask);
1956 put_rpccred(cred);
1957 } else
1958 res = PTR_ERR(cred);
1959 unlock_kernel();
1960out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001961 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1962 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 return res;
1964out_notsup:
1965 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1966 if (res == 0)
1967 res = generic_permission(inode, mask, NULL);
1968 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001969 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970}
1971
1972/*
1973 * Local variables:
1974 * version-control: t
1975 * kept-new-versions: 5
1976 * End:
1977 */