blob: f697b5c74b7c38f5627ea71c9d2ec9ac95bc3eba [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>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040036#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Trond Myklebust4ce79712005-06-22 17:16:21 +000038#include "nfs4_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "delegation.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050040#include "iostat.h"
Adrian Bunk4c30d562007-11-21 15:04:31 -080041#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* #define NFS_DEBUG_VERBOSE 1 */
44
45static int nfs_opendir(struct inode *, struct file *);
46static int nfs_readdir(struct file *, void *, filldir_t);
47static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49static int nfs_mkdir(struct inode *, struct dentry *, int);
50static int nfs_rmdir(struct inode *, struct dentry *);
51static int nfs_unlink(struct inode *, struct dentry *);
52static int nfs_symlink(struct inode *, struct dentry *, const char *);
53static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57static int nfs_fsync_dir(struct file *, struct dentry *, int);
Trond Myklebustf0dd2132005-06-22 17:16:29 +000058static loff_t nfs_llseek_dir(struct file *, loff_t, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080060const struct file_operations nfs_dir_operations = {
Trond Myklebustf0dd2132005-06-22 17:16:29 +000061 .llseek = nfs_llseek_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 .read = generic_read_dir,
63 .readdir = nfs_readdir,
64 .open = nfs_opendir,
65 .release = nfs_release,
66 .fsync = nfs_fsync_dir,
67};
68
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080069const struct inode_operations nfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 .create = nfs_create,
71 .lookup = nfs_lookup,
72 .link = nfs_link,
73 .unlink = nfs_unlink,
74 .symlink = nfs_symlink,
75 .mkdir = nfs_mkdir,
76 .rmdir = nfs_rmdir,
77 .mknod = nfs_mknod,
78 .rename = nfs_rename,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
82};
83
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000084#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080085const struct inode_operations nfs3_dir_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000086 .create = nfs_create,
87 .lookup = nfs_lookup,
88 .link = nfs_link,
89 .unlink = nfs_unlink,
90 .symlink = nfs_symlink,
91 .mkdir = nfs_mkdir,
92 .rmdir = nfs_rmdir,
93 .mknod = nfs_mknod,
94 .rename = nfs_rename,
95 .permission = nfs_permission,
96 .getattr = nfs_getattr,
97 .setattr = nfs_setattr,
98 .listxattr = nfs3_listxattr,
99 .getxattr = nfs3_getxattr,
100 .setxattr = nfs3_setxattr,
101 .removexattr = nfs3_removexattr,
102};
103#endif /* CONFIG_NFS_V3 */
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105#ifdef CONFIG_NFS_V4
106
107static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800108const struct inode_operations nfs4_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 .create = nfs_create,
110 .lookup = nfs_atomic_lookup,
111 .link = nfs_link,
112 .unlink = nfs_unlink,
113 .symlink = nfs_symlink,
114 .mkdir = nfs_mkdir,
115 .rmdir = nfs_rmdir,
116 .mknod = nfs_mknod,
117 .rename = nfs_rename,
118 .permission = nfs_permission,
119 .getattr = nfs_getattr,
120 .setattr = nfs_setattr,
J. Bruce Fields6b3b5492005-06-22 17:16:22 +0000121 .getxattr = nfs4_getxattr,
122 .setxattr = nfs4_setxattr,
123 .listxattr = nfs4_listxattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124};
125
126#endif /* CONFIG_NFS_V4 */
127
128/*
129 * Open file
130 */
131static int
132nfs_opendir(struct inode *inode, struct file *filp)
133{
Carsten Otte7451c4f2006-04-19 13:06:37 -0400134 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500136 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
137 inode->i_sb->s_id, inode->i_ino);
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 lock_kernel();
140 /* Call generic open code in order to cache credentials */
Carsten Otte7451c4f2006-04-19 13:06:37 -0400141 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 unlock_kernel();
143 return res;
144}
145
Al Viro0dbb4c62006-10-19 23:28:49 -0700146typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147typedef struct {
148 struct file *file;
149 struct page *page;
150 unsigned long page_index;
Al Viro0dbb4c62006-10-19 23:28:49 -0700151 __be32 *ptr;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000152 u64 *dir_cookie;
153 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 struct nfs_entry *entry;
155 decode_dirent_t decode;
156 int plus;
157 int error;
Neil Brown1f4eab72007-04-16 09:35:27 +1000158 unsigned long timestamp;
159 int timestamp_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160} nfs_readdir_descriptor_t;
161
162/* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
164 *
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
168 * to be.
169 *
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
173 */
174static
175int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176{
177 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800178 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 struct rpc_cred *cred = nfs_file_cred(file);
180 unsigned long timestamp;
181 int error;
182
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
184 __FUNCTION__, (long long)desc->entry->cookie,
185 page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 again:
188 timestamp = jiffies;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800189 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 NFS_SERVER(inode)->dtsize, desc->plus);
191 if (error < 0) {
192 /* We requested READDIRPLUS, but the server doesn't grok it */
193 if (error == -ENOTSUPP && desc->plus) {
194 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Chuck Lever412d5822005-08-18 11:24:11 -0700195 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 desc->plus = 0;
197 goto again;
198 }
199 goto error;
200 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000201 desc->timestamp = timestamp;
202 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 /* Ensure consistent page alignment of the data.
205 * Note: assumes we have exclusive access to this mapping either
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800206 * through inode->i_mutex or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700208 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
209 /* Should never happen */
210 nfs_zap_mapping(inode, inode->i_mapping);
211 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 unlock_page(page);
213 return 0;
214 error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 desc->error = error;
217 return -EIO;
218}
219
220static inline
221int dir_decode(nfs_readdir_descriptor_t *desc)
222{
Al Viro0dbb4c62006-10-19 23:28:49 -0700223 __be32 *p = desc->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 p = desc->decode(p, desc->entry, desc->plus);
225 if (IS_ERR(p))
226 return PTR_ERR(p);
227 desc->ptr = p;
Neil Brown1f4eab72007-04-16 09:35:27 +1000228 if (desc->timestamp_valid)
229 desc->entry->fattr->time_start = desc->timestamp;
230 else
231 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 return 0;
233}
234
235static inline
236void dir_page_release(nfs_readdir_descriptor_t *desc)
237{
238 kunmap(desc->page);
239 page_cache_release(desc->page);
240 desc->page = NULL;
241 desc->ptr = NULL;
242}
243
244/*
245 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000246 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 *
248 * If the end of the buffer has been reached, return -EAGAIN, if not,
249 * return the offset within the buffer of the next entry to be
250 * read.
251 */
252static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000253int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254{
255 struct nfs_entry *entry = desc->entry;
256 int loop_count = 0,
257 status;
258
259 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500260 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
261 __FUNCTION__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000262 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 break;
264 if (loop_count++ > 200) {
265 loop_count = 0;
266 schedule();
267 }
268 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 return status;
270}
271
272/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000273 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000274 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000275 *
276 * If the end of the buffer has been reached, return -EAGAIN, if not,
277 * return the offset within the buffer of the next entry to be
278 * read.
279 */
280static inline
281int find_dirent_index(nfs_readdir_descriptor_t *desc)
282{
283 struct nfs_entry *entry = desc->entry;
284 int loop_count = 0,
285 status;
286
287 for(;;) {
288 status = dir_decode(desc);
289 if (status)
290 break;
291
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500292 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
293 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000294
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000295 if (desc->file->f_pos == desc->current_index) {
296 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000297 break;
298 }
299 desc->current_index++;
300 if (loop_count++ > 200) {
301 loop_count = 0;
302 schedule();
303 }
304 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000305 return status;
306}
307
308/*
309 * Find the given page, and call find_dirent() or find_dirent_index in
310 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 */
312static inline
313int find_dirent_page(nfs_readdir_descriptor_t *desc)
314{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800315 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 struct page *page;
317 int status;
318
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500319 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
320 __FUNCTION__, desc->page_index,
321 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Neil Brown1f4eab72007-04-16 09:35:27 +1000323 /* If we find the page in the page_cache, we cannot be sure
324 * how fresh the data is, so we will ignore readdir_plus attributes.
325 */
326 desc->timestamp_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 page = read_cache_page(inode->i_mapping, desc->page_index,
328 (filler_t *)nfs_readdir_filler, desc);
329 if (IS_ERR(page)) {
330 status = PTR_ERR(page);
331 goto out;
332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
334 /* NOTE: Someone else may have changed the READDIRPLUS flag */
335 desc->page = page;
336 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000337 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000338 status = find_dirent(desc);
339 else
340 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 if (status < 0)
342 dir_page_release(desc);
343 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500344 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346}
347
348/*
349 * Recurse through the page cache pages, and return a
350 * filled nfs_entry structure of the next directory entry if possible.
351 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000352 * The target for the search is '*desc->dir_cookie' if non-0,
353 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 */
355static inline
356int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
357{
358 int loop_count = 0;
359 int res;
360
Olivier Galibert00a92642005-06-22 17:16:29 +0000361 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000362 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500363 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
364 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000365 desc->page_index = 0;
366 desc->entry->cookie = desc->entry->prev_cookie = 0;
367 desc->entry->eof = 0;
368 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000369 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500370 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
371 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000372
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 for (;;) {
374 res = find_dirent_page(desc);
375 if (res != -EAGAIN)
376 break;
377 /* Align to beginning of next page */
378 desc->page_index ++;
379 if (loop_count++ > 200) {
380 loop_count = 0;
381 schedule();
382 }
383 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500384
385 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 return res;
387}
388
389static inline unsigned int dt_type(struct inode *inode)
390{
391 return (inode->i_mode >> 12) & 15;
392}
393
394static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
395
396/*
397 * Once we've found the start of the dirent within a page: fill 'er up...
398 */
399static
400int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
401 filldir_t filldir)
402{
403 struct file *file = desc->file;
404 struct nfs_entry *entry = desc->entry;
405 struct dentry *dentry = NULL;
Peter Staubach4e769b92007-08-03 15:07:10 -0400406 u64 fileid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 int loop_count = 0,
408 res;
409
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500410 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
411 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412
413 for(;;) {
414 unsigned d_type = DT_UNKNOWN;
415 /* Note: entry->prev_cookie contains the cookie for
416 * retrieving the current dirent on the server */
Peter Staubach4e769b92007-08-03 15:07:10 -0400417 fileid = entry->ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418
419 /* Get a dentry if we have one */
420 if (dentry != NULL)
421 dput(dentry);
422 dentry = nfs_readdir_lookup(desc);
423
424 /* Use readdirplus info */
425 if (dentry != NULL && dentry->d_inode != NULL) {
426 d_type = dt_type(dentry->d_inode);
Peter Staubach4e769b92007-08-03 15:07:10 -0400427 fileid = NFS_FILEID(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428 }
429
430 res = filldir(dirent, entry->name, entry->len,
Trond Myklebustf43bf0b2007-10-09 12:01:04 -0400431 file->f_pos, nfs_compat_user_ino64(fileid),
432 d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (res < 0)
434 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000435 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000436 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 if (dir_decode(desc) != 0) {
438 desc->page_index ++;
439 break;
440 }
441 if (loop_count++ > 200) {
442 loop_count = 0;
443 schedule();
444 }
445 }
446 dir_page_release(desc);
447 if (dentry != NULL)
448 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500449 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
450 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 return res;
452}
453
454/*
455 * If we cannot find a cookie in our cache, we suspect that this is
456 * because it points to a deleted file, so we ask the server to return
457 * whatever it thinks is the next entry. We then feed this to filldir.
458 * If all goes well, we should then be able to find our way round the
459 * cache on the next call to readdir_search_pagecache();
460 *
461 * NOTE: we cannot add the anonymous page to the pagecache because
462 * the data it contains might not be page aligned. Besides,
463 * we should already have a complete representation of the
464 * directory in the page cache by the time we get here.
465 */
466static inline
467int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
468 filldir_t filldir)
469{
470 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800471 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 struct rpc_cred *cred = nfs_file_cred(file);
473 struct page *page = NULL;
474 int status;
Neil Brown1f4eab72007-04-16 09:35:27 +1000475 unsigned long timestamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500477 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
478 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 page = alloc_page(GFP_HIGHUSER);
481 if (!page) {
482 status = -ENOMEM;
483 goto out;
484 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000485 timestamp = jiffies;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800486 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 page,
488 NFS_SERVER(inode)->dtsize,
489 desc->plus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 desc->page = page;
491 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
492 if (desc->error >= 0) {
Neil Brown1f4eab72007-04-16 09:35:27 +1000493 desc->timestamp = timestamp;
494 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000496 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 } else
498 status = -EIO;
499 if (status < 0)
500 goto out_release;
501
502 status = nfs_do_filldir(desc, dirent, filldir);
503
504 /* Reset read descriptor so it searches the page cache from
505 * the start upon the next call to readdir_search_pagecache() */
506 desc->page_index = 0;
507 desc->entry->cookie = desc->entry->prev_cookie = 0;
508 desc->entry->eof = 0;
509 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500510 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
511 __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 return status;
513 out_release:
514 dir_page_release(desc);
515 goto out;
516}
517
Olivier Galibert00a92642005-06-22 17:16:29 +0000518/* The file offset position represents the dirent entry number. A
519 last cookie cache takes care of the common case of reading the
520 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 */
522static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
523{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800524 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 struct inode *inode = dentry->d_inode;
526 nfs_readdir_descriptor_t my_desc,
527 *desc = &my_desc;
528 struct nfs_entry my_entry;
529 struct nfs_fh fh;
530 struct nfs_fattr fattr;
531 long res;
532
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500533 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
534 dentry->d_parent->d_name.name, dentry->d_name.name,
535 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500536 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 lock_kernel();
539
Trond Myklebust717d44e2007-01-24 11:54:55 -0800540 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (res < 0) {
542 unlock_kernel();
543 return res;
544 }
545
546 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000547 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000548 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000549 * to either find the entry with the appropriate number or
550 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 */
552 memset(desc, 0, sizeof(*desc));
553
554 desc->file = filp;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400555 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 desc->decode = NFS_PROTO(inode)->decode_dirent;
557 desc->plus = NFS_USE_READDIRPLUS(inode);
558
559 my_entry.cookie = my_entry.prev_cookie = 0;
560 my_entry.eof = 0;
561 my_entry.fh = &fh;
562 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400563 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 desc->entry = &my_entry;
565
Trond Myklebust565277f2007-10-15 18:17:53 -0400566 nfs_block_sillyrename(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 while(!desc->entry->eof) {
568 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (res == -EBADCOOKIE) {
571 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000572 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /* Or that the server has 'lost' a cookie */
574 res = uncached_readdir(desc, dirent, filldir);
575 if (res >= 0)
576 continue;
577 }
578 res = 0;
579 break;
580 }
581 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700582 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 nfs_zap_caches(inode);
584 desc->plus = 0;
585 desc->entry->eof = 0;
586 continue;
587 }
588 if (res < 0)
589 break;
590
591 res = nfs_do_filldir(desc, dirent, filldir);
592 if (res < 0) {
593 res = 0;
594 break;
595 }
596 }
Trond Myklebust565277f2007-10-15 18:17:53 -0400597 nfs_unblock_sillyrename(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500599 if (res > 0)
600 res = 0;
601 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
602 dentry->d_parent->d_name.name, dentry->d_name.name,
603 res);
604 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
Trond Myklebust10afec92007-05-14 17:16:04 -0400607static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000608{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800609 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000610 switch (origin) {
611 case 1:
612 offset += filp->f_pos;
613 case 0:
614 if (offset >= 0)
615 break;
616 default:
617 offset = -EINVAL;
618 goto out;
619 }
620 if (offset != filp->f_pos) {
621 filp->f_pos = offset;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400622 nfs_file_open_context(filp)->dir_cookie = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000623 }
624out:
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800625 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000626 return offset;
627}
628
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629/*
630 * All directory operations under NFS are synchronous, so fsync()
631 * is a dummy operation.
632 */
Trond Myklebust10afec92007-05-14 17:16:04 -0400633static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500635 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
636 dentry->d_parent->d_name.name, dentry->d_name.name,
637 datasync);
638
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 return 0;
640}
641
642/*
643 * A check for whether or not the parent directory has changed.
644 * In the case it has, we assume that the dentries are untrustworthy
645 * and may need to be looked up again.
646 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500647static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648{
649 if (IS_ROOT(dentry))
650 return 1;
Trond Myklebustf2c77f42007-10-02 12:54:39 -0400651 if (!nfs_verify_change_attribute(dir, dentry->d_time))
652 return 0;
653 /* Revalidate nfsi->cache_change_attribute before we declare a match */
654 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
655 return 0;
656 if (!nfs_verify_change_attribute(dir, dentry->d_time))
657 return 0;
658 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659}
660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400662 * 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 */
668static 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/*
Trond Myklebusta12802c2007-10-02 19:13:04 -0400676 * Use intent information to check whether or not we're going to do
677 * an O_EXCL create using this path component.
678 */
679static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
680{
681 if (NFS_PROTO(dir)->version == 2)
682 return 0;
683 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
684 return 0;
685 return (nd->intent.open.flags & O_EXCL) != 0;
686}
687
688/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400689 * Inode and filehandle revalidation for lookups.
690 *
691 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
692 * or if the intent information indicates that we're about to open this
693 * particular file and the "nocto" mount flag is not set.
694 *
695 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696static inline
697int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
698{
699 struct nfs_server *server = NFS_SERVER(inode);
700
701 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400703 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 goto out_force;
705 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400706 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400707 !(server->flags & NFS_MOUNT_NOCTO) &&
708 (S_ISREG(inode->i_mode) ||
709 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 goto out_force;
Trond Myklebust4f48af42007-10-02 23:13:32 -0400711 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 }
713 return nfs_revalidate_inode(server, inode);
714out_force:
715 return __nfs_revalidate_inode(server, inode);
716}
717
718/*
719 * We judge how long we want to trust negative
720 * dentries by looking at the parent inode mtime.
721 *
722 * If parent mtime has changed, we revalidate, else we wait for a
723 * period corresponding to the parent's attribute cache timeout value.
724 */
725static inline
726int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
727 struct nameidata *nd)
728{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400730 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 return 0;
732 return !nfs_check_verifier(dir, dentry);
733}
734
735/*
736 * This is called every time the dcache has a lookup hit,
737 * and we should check whether we can really trust that
738 * lookup.
739 *
740 * NOTE! The hit can be a negative hit too, don't assume
741 * we have an inode!
742 *
743 * If the parent directory is seen to have changed, we throw out the
744 * cached dentry and do a new lookup.
745 */
746static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
747{
748 struct inode *dir;
749 struct inode *inode;
750 struct dentry *parent;
751 int error;
752 struct nfs_fh fhandle;
753 struct nfs_fattr fattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
755 parent = dget_parent(dentry);
756 lock_kernel();
757 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500758 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 inode = dentry->d_inode;
760
761 if (!inode) {
762 if (nfs_neg_need_reval(dir, dentry, nd))
763 goto out_bad;
764 goto out_valid;
765 }
766
767 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500768 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
769 __FUNCTION__, dentry->d_parent->d_name.name,
770 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 goto out_bad;
772 }
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 /* Force a full look up iff the parent directory has changed */
Trond Myklebusta12802c2007-10-02 19:13:04 -0400775 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 if (nfs_lookup_verify_inode(inode, nd))
777 goto out_zap_parent;
778 goto out_valid;
779 }
780
781 if (NFS_STALE(inode))
782 goto out_bad;
783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
785 if (error)
786 goto out_bad;
787 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
788 goto out_bad;
789 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
790 goto out_bad;
791
Trond Myklebustcf8ba452007-10-01 13:46:53 -0400792 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 out_valid:
794 unlock_kernel();
795 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500796 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
797 __FUNCTION__, dentry->d_parent->d_name.name,
798 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return 1;
800out_zap_parent:
801 nfs_zap_caches(dir);
802 out_bad:
Trond Myklebusta1643a92007-09-29 17:25:43 -0400803 nfs_mark_for_revalidate(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 if (inode && S_ISDIR(inode->i_mode)) {
805 /* Purge readdir caches. */
806 nfs_zap_caches(inode);
807 /* If we have submounts, don't unhash ! */
808 if (have_submounts(dentry))
809 goto out_valid;
810 shrink_dcache_parent(dentry);
811 }
812 d_drop(dentry);
813 unlock_kernel();
814 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500815 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
816 __FUNCTION__, dentry->d_parent->d_name.name,
817 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 return 0;
819}
820
821/*
822 * This is called from dput() when d_count is going to 0.
823 */
824static int nfs_dentry_delete(struct dentry *dentry)
825{
826 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
827 dentry->d_parent->d_name.name, dentry->d_name.name,
828 dentry->d_flags);
829
830 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
831 /* Unhash it, so that ->d_iput() would be called */
832 return 1;
833 }
834 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
835 /* Unhash it, so that ancestors of killed async unlink
836 * files will be cleaned up during umount */
837 return 1;
838 }
839 return 0;
840
841}
842
843/*
844 * Called when the dentry loses inode.
845 * We use it to clean up silly-renamed files.
846 */
847static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
848{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700849 nfs_inode_return_delegation(inode);
Neil Brown83672d32007-02-26 12:48:25 +1100850 if (S_ISDIR(inode->i_mode))
851 /* drop any readdir cache as it could easily be old */
852 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
853
Linus Torvalds1da177e2005-04-16 15:20:36 -0700854 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
855 lock_kernel();
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700856 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -0400857 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858 unlock_kernel();
859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 iput(inode);
861}
862
863struct dentry_operations nfs_dentry_operations = {
864 .d_revalidate = nfs_lookup_revalidate,
865 .d_delete = nfs_dentry_delete,
866 .d_iput = nfs_dentry_iput,
867};
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
870{
871 struct dentry *res;
Trond Myklebust565277f2007-10-15 18:17:53 -0400872 struct dentry *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 struct inode *inode = NULL;
874 int error;
875 struct nfs_fh fhandle;
876 struct nfs_fattr fattr;
877
878 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
879 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500880 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881
882 res = ERR_PTR(-ENAMETOOLONG);
883 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
884 goto out;
885
886 res = ERR_PTR(-ENOMEM);
887 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
888
889 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890
Trond Myklebustfd684072006-09-05 12:27:44 -0400891 /*
892 * If we're doing an exclusive create, optimize away the lookup
893 * but don't hash the dentry.
894 */
895 if (nfs_is_exclusive_create(dir, nd)) {
896 d_instantiate(dentry, NULL);
897 res = NULL;
898 goto out_unlock;
899 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
Trond Myklebust565277f2007-10-15 18:17:53 -0400901 parent = dentry->d_parent;
902 /* Protect against concurrent sillydeletes */
903 nfs_block_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
905 if (error == -ENOENT)
906 goto no_entry;
907 if (error < 0) {
908 res = ERR_PTR(error);
Trond Myklebust565277f2007-10-15 18:17:53 -0400909 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500912 res = (struct dentry *)inode;
913 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400914 goto out_unblock_sillyrename;
David Howells54ceac42006-08-22 20:06:13 -0400915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400917 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -0700918 if (res != NULL) {
919 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400920 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Trond Myklebust565277f2007-10-15 18:17:53 -0400924out_unblock_sillyrename:
925 nfs_unblock_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926out_unlock:
927 unlock_kernel();
928out:
929 return res;
930}
931
932#ifdef CONFIG_NFS_V4
933static int nfs_open_revalidate(struct dentry *, struct nameidata *);
934
935struct dentry_operations nfs4_dentry_operations = {
936 .d_revalidate = nfs_open_revalidate,
937 .d_delete = nfs_dentry_delete,
938 .d_iput = nfs_dentry_iput,
939};
940
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400941/*
942 * Use intent information to determine whether we need to substitute
943 * the NFSv4-style stateful OPEN for the LOOKUP call
944 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945static int is_atomic_open(struct inode *dir, struct nameidata *nd)
946{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400947 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 return 0;
949 /* NFS does not (yet) have a stateful open for directories */
950 if (nd->flags & LOOKUP_DIRECTORY)
951 return 0;
952 /* Are we trying to write to a read only partition? */
953 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
954 return 0;
955 return 1;
956}
957
958static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
959{
960 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 int error;
962
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500963 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
964 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
965
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 /* Check that we are indeed trying to open this file */
967 if (!is_atomic_open(dir, nd))
968 goto no_open;
969
970 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
971 res = ERR_PTR(-ENAMETOOLONG);
972 goto out;
973 }
974 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
975
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400976 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
977 * the dentry. */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700978 if (nd->intent.open.flags & O_EXCL) {
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400979 d_instantiate(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -0700980 goto out;
981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
983 /* Open the file on the server */
984 lock_kernel();
Trond Myklebust60ccd4e2007-09-29 17:48:19 -0400985 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -0700987 if (IS_ERR(res)) {
988 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 switch (error) {
990 /* Make a negative dentry */
991 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -0700992 res = NULL;
993 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -0700995 case -EISDIR:
996 case -ENOTDIR:
997 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 case -ELOOP:
999 if (!(nd->intent.open.flags & O_NOFOLLOW))
1000 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 /* case -EINVAL: */
1002 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 goto out;
1004 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001005 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 dentry = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007out:
1008 return res;
1009no_open:
1010 return nfs_lookup(dir, dentry, nd);
1011}
1012
1013static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1014{
1015 struct dentry *parent = NULL;
1016 struct inode *inode = dentry->d_inode;
1017 struct inode *dir;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 int openflags, ret = 0;
1019
1020 parent = dget_parent(dentry);
1021 dir = parent->d_inode;
1022 if (!is_atomic_open(dir, nd))
1023 goto no_open;
1024 /* We can't create new files in nfs_open_revalidate(), so we
1025 * optimize away revalidation of negative dentries.
1026 */
Trond Myklebust216d5d062007-10-01 20:10:12 -04001027 if (inode == NULL) {
1028 if (!nfs_neg_need_reval(dir, dentry, nd))
1029 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 goto out;
Trond Myklebust216d5d062007-10-01 20:10:12 -04001031 }
1032
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 /* NFS only supports OPEN on regular files */
1034 if (!S_ISREG(inode->i_mode))
1035 goto no_open;
1036 openflags = nd->intent.open.flags;
1037 /* We cannot do exclusive creation on a positive dentry */
1038 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1039 goto no_open;
1040 /* We can't create new files, or truncate existing ones here */
1041 openflags &= ~(O_CREAT|O_TRUNC);
1042
1043 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001044 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 * operations that change the directory. We therefore save the
1046 * change attribute *before* we do the RPC call.
1047 */
1048 lock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001049 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 unlock_kernel();
1051out:
1052 dput(parent);
1053 if (!ret)
1054 d_drop(dentry);
1055 return ret;
1056no_open:
1057 dput(parent);
1058 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1059 return 1;
1060 return nfs_lookup_revalidate(dentry, nd);
1061}
1062#endif /* CONFIG_NFSV4 */
1063
1064static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1065{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -08001066 struct dentry *parent = desc->file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 struct inode *dir = parent->d_inode;
1068 struct nfs_entry *entry = desc->entry;
1069 struct dentry *dentry, *alias;
1070 struct qstr name = {
1071 .name = entry->name,
1072 .len = entry->len,
1073 };
1074 struct inode *inode;
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001075 unsigned long verf = nfs_save_change_attribute(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076
1077 switch (name.len) {
1078 case 2:
1079 if (name.name[0] == '.' && name.name[1] == '.')
1080 return dget_parent(parent);
1081 break;
1082 case 1:
1083 if (name.name[0] == '.')
1084 return dget(parent);
1085 }
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001086
1087 spin_lock(&dir->i_lock);
1088 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1089 spin_unlock(&dir->i_lock);
1090 return NULL;
1091 }
1092 spin_unlock(&dir->i_lock);
1093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 name.hash = full_name_hash(name.name, name.len);
1095 dentry = d_lookup(parent, &name);
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001096 if (dentry != NULL) {
Trond Myklebustef75c792007-01-16 10:09:44 -05001097 /* Is this a positive dentry that matches the readdir info? */
1098 if (dentry->d_inode != NULL &&
1099 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1100 d_mountpoint(dentry))) {
1101 if (!desc->plus || entry->fh->size == 0)
1102 return dentry;
1103 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1104 entry->fh) == 0)
1105 goto out_renew;
1106 }
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001107 /* No, so d_drop to allow one to be created */
1108 d_drop(dentry);
1109 dput(dentry);
1110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1112 return NULL;
Trond Myklebust54af3bb2007-09-28 12:27:41 -04001113 if (name.len > NFS_SERVER(dir)->namelen)
1114 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001115 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 dentry = d_alloc(parent, &name);
1117 if (dentry == NULL)
1118 return NULL;
1119 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1120 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001121 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 dput(dentry);
1123 return NULL;
1124 }
David Howells54ceac42006-08-22 20:06:13 -04001125
1126 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127 if (alias != NULL) {
1128 dput(dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001129 if (IS_ERR(alias))
1130 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 dentry = alias;
1132 }
David Howells54ceac42006-08-22 20:06:13 -04001133
Trond Myklebustc79ba782007-01-31 08:16:24 -05001134out_renew:
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001135 nfs_set_verifier(dentry, verf);
Trond Myklebustc79ba782007-01-31 08:16:24 -05001136 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137}
1138
1139/*
1140 * Code common to create, mkdir, and mknod.
1141 */
1142int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1143 struct nfs_fattr *fattr)
1144{
Trond Myklebustfab728e2007-09-29 17:41:33 -04001145 struct dentry *parent = dget_parent(dentry);
1146 struct inode *dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 struct inode *inode;
1148 int error = -EACCES;
1149
Trond Myklebustfab728e2007-09-29 17:41:33 -04001150 d_drop(dentry);
1151
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 /* We may have been initialized further down */
1153 if (dentry->d_inode)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001154 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (fhandle->size == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1157 if (error)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001158 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 }
Trond Myklebust5724ab32007-10-01 21:51:38 -04001160 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1162 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001163 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (error < 0)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001165 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001168 error = PTR_ERR(inode);
1169 if (IS_ERR(inode))
Trond Myklebustfab728e2007-09-29 17:41:33 -04001170 goto out_error;
1171 d_add(dentry, inode);
1172out:
1173 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 return 0;
Trond Myklebustfab728e2007-09-29 17:41:33 -04001175out_error:
1176 nfs_mark_for_revalidate(dir);
1177 dput(parent);
1178 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179}
1180
1181/*
1182 * Following a failed create operation, we drop the dentry rather
1183 * than retain a negative dentry. This avoids a problem in the event
1184 * that the operation succeeded on the server, but an error in the
1185 * reply path made it appear to have failed.
1186 */
1187static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1188 struct nameidata *nd)
1189{
1190 struct iattr attr;
1191 int error;
1192 int open_flags = 0;
1193
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001194 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1195 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
1197 attr.ia_mode = mode;
1198 attr.ia_valid = ATTR_MODE;
1199
Trond Myklebustad389da2007-06-05 12:30:00 -04001200 if ((nd->flags & LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 open_flags = nd->intent.open.flags;
1202
1203 lock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001204 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205 if (error != 0)
1206 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 unlock_kernel();
1208 return 0;
1209out_err:
1210 unlock_kernel();
1211 d_drop(dentry);
1212 return error;
1213}
1214
1215/*
1216 * See comments for nfs_proc_create regarding failed operations.
1217 */
1218static int
1219nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1220{
1221 struct iattr attr;
1222 int status;
1223
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001224 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1225 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 if (!new_valid_dev(rdev))
1228 return -EINVAL;
1229
1230 attr.ia_mode = mode;
1231 attr.ia_valid = ATTR_MODE;
1232
1233 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 if (status != 0)
1236 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 unlock_kernel();
1238 return 0;
1239out_err:
1240 unlock_kernel();
1241 d_drop(dentry);
1242 return status;
1243}
1244
1245/*
1246 * See comments for nfs_proc_create regarding failed operations.
1247 */
1248static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1249{
1250 struct iattr attr;
1251 int error;
1252
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001253 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1254 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255
1256 attr.ia_valid = ATTR_MODE;
1257 attr.ia_mode = mode | S_IFDIR;
1258
1259 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (error != 0)
1262 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 unlock_kernel();
1264 return 0;
1265out_err:
1266 d_drop(dentry);
1267 unlock_kernel();
1268 return error;
1269}
1270
1271static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1272{
1273 int error;
1274
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001275 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1276 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277
1278 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1280 /* Ensure the VFS deletes this inode */
1281 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001282 clear_nlink(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 unlock_kernel();
1284
1285 return error;
1286}
1287
1288static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1289{
1290 static unsigned int sillycounter;
Peter Staubach4e769b92007-08-03 15:07:10 -04001291 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 const int countersize = sizeof(sillycounter)*2;
Peter Staubach4e769b92007-08-03 15:07:10 -04001293 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 char silly[slen+1];
1295 struct qstr qsilly;
1296 struct dentry *sdentry;
1297 int error = -EIO;
1298
1299 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1300 dentry->d_parent->d_name.name, dentry->d_name.name,
1301 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001302 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304 /*
1305 * We don't allow a dentry to be silly-renamed twice.
1306 */
1307 error = -EBUSY;
1308 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1309 goto out;
1310
Peter Staubach4e769b92007-08-03 15:07:10 -04001311 sprintf(silly, ".nfs%*.*Lx",
1312 fileidsize, fileidsize,
1313 (unsigned long long)NFS_FILEID(dentry->d_inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314
Trond Myklebust34ea8182005-11-04 15:35:02 -05001315 /* Return delegation in anticipation of the rename */
1316 nfs_inode_return_delegation(dentry->d_inode);
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 sdentry = NULL;
1319 do {
1320 char *suffix = silly + slen - countersize;
1321
1322 dput(sdentry);
1323 sillycounter++;
1324 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1325
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001326 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1327 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1330 /*
1331 * N.B. Better to return EBUSY here ... it could be
1332 * dangerous to delete the file while it's in use.
1333 */
1334 if (IS_ERR(sdentry))
1335 goto out;
1336 } while(sdentry->d_inode != NULL); /* need negative lookup */
1337
1338 qsilly.name = silly;
1339 qsilly.len = strlen(silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 if (dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1342 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001343 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 } else
1345 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1346 dir, &qsilly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1349 d_move(dentry, sdentry);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001350 error = nfs_async_unlink(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 /* If we return 0 we don't unlink */
1352 }
1353 dput(sdentry);
1354out:
1355 return error;
1356}
1357
1358/*
1359 * Remove a file after making sure there are no pending writes,
1360 * and after checking that the file has only one user.
1361 *
1362 * We invalidate the attribute cache and free the inode prior to the operation
1363 * to avoid possible races if the server reuses the inode.
1364 */
1365static int nfs_safe_remove(struct dentry *dentry)
1366{
1367 struct inode *dir = dentry->d_parent->d_inode;
1368 struct inode *inode = dentry->d_inode;
1369 int error = -EBUSY;
1370
1371 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1372 dentry->d_parent->d_name.name, dentry->d_name.name);
1373
1374 /* If the dentry was sillyrenamed, we simply call d_delete() */
1375 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1376 error = 0;
1377 goto out;
1378 }
1379
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001381 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1383 /* The VFS may want to delete this inode */
1384 if (error == 0)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001385 drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001386 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 } else
1388 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389out:
1390 return error;
1391}
1392
1393/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1394 * belongs to an active ".nfs..." file and we return -EBUSY.
1395 *
1396 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1397 */
1398static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1399{
1400 int error;
1401 int need_rehash = 0;
1402
1403 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1404 dir->i_ino, dentry->d_name.name);
1405
1406 lock_kernel();
1407 spin_lock(&dcache_lock);
1408 spin_lock(&dentry->d_lock);
1409 if (atomic_read(&dentry->d_count) > 1) {
1410 spin_unlock(&dentry->d_lock);
1411 spin_unlock(&dcache_lock);
Trond Myklebustccfeb502007-01-13 02:28:12 -05001412 /* Start asynchronous writeout of the inode */
1413 write_inode_now(dentry->d_inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 error = nfs_sillyrename(dir, dentry);
1415 unlock_kernel();
1416 return error;
1417 }
1418 if (!d_unhashed(dentry)) {
1419 __d_drop(dentry);
1420 need_rehash = 1;
1421 }
1422 spin_unlock(&dentry->d_lock);
1423 spin_unlock(&dcache_lock);
1424 error = nfs_safe_remove(dentry);
1425 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1427 } else if (need_rehash)
1428 d_rehash(dentry);
1429 unlock_kernel();
1430 return error;
1431}
1432
Chuck Lever873101b2006-08-22 20:06:23 -04001433/*
1434 * To create a symbolic link, most file systems instantiate a new inode,
1435 * add a page to it containing the path, then write it out to the disk
1436 * using prepare_write/commit_write.
1437 *
1438 * Unfortunately the NFS client can't create the in-core inode first
1439 * because it needs a file handle to create an in-core inode (see
1440 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1441 * symlink request has completed on the server.
1442 *
1443 * So instead we allocate a raw page, copy the symname into it, then do
1444 * the SYMLINK request with the page as the buffer. If it succeeds, we
1445 * now have a new file handle and can instantiate an in-core NFS inode
1446 * and move the raw page into its mapping.
1447 */
1448static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
Chuck Lever873101b2006-08-22 20:06:23 -04001450 struct pagevec lru_pvec;
1451 struct page *page;
1452 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001454 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 int error;
1456
1457 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1458 dir->i_ino, dentry->d_name.name, symname);
1459
Chuck Lever873101b2006-08-22 20:06:23 -04001460 if (pathlen > PAGE_SIZE)
1461 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462
Chuck Lever873101b2006-08-22 20:06:23 -04001463 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1464 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465
1466 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001467
Jeff Layton83d93f22007-06-07 09:58:08 -04001468 page = alloc_page(GFP_HIGHUSER);
Chuck Lever873101b2006-08-22 20:06:23 -04001469 if (!page) {
1470 unlock_kernel();
1471 return -ENOMEM;
1472 }
1473
1474 kaddr = kmap_atomic(page, KM_USER0);
1475 memcpy(kaddr, symname, pathlen);
1476 if (pathlen < PAGE_SIZE)
1477 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1478 kunmap_atomic(kaddr, KM_USER0);
1479
Chuck Lever94a6d752006-08-22 20:06:23 -04001480 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Chuck Lever873101b2006-08-22 20:06:23 -04001481 if (error != 0) {
1482 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1483 dir->i_sb->s_id, dir->i_ino,
1484 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001486 __free_page(page);
1487 unlock_kernel();
1488 return error;
1489 }
1490
1491 /*
1492 * No big deal if we can't add this page to the page cache here.
1493 * READLINK will get the missing page from the server if needed.
1494 */
1495 pagevec_init(&lru_pvec, 0);
1496 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1497 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001498 pagevec_add(&lru_pvec, page);
1499 pagevec_lru_add(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001500 SetPageUptodate(page);
1501 unlock_page(page);
1502 } else
1503 __free_page(page);
1504
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001506 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
1509static int
1510nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1511{
1512 struct inode *inode = old_dentry->d_inode;
1513 int error;
1514
1515 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1516 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1517 dentry->d_parent->d_name.name, dentry->d_name.name);
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 lock_kernel();
Trond Myklebust9697d232007-10-02 21:58:05 -04001520 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001522 if (error == 0) {
1523 atomic_inc(&inode->i_count);
Trond Myklebust9697d232007-10-02 21:58:05 -04001524 d_add(dentry, inode);
Trond Myklebustcf809552005-10-27 22:12:42 -04001525 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 unlock_kernel();
1527 return error;
1528}
1529
1530/*
1531 * RENAME
1532 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1533 * different file handle for the same inode after a rename (e.g. when
1534 * moving to a different directory). A fail-safe method to do so would
1535 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1536 * rename the old file using the sillyrename stuff. This way, the original
1537 * file in old_dir will go away when the last process iput()s the inode.
1538 *
1539 * FIXED.
1540 *
1541 * It actually works quite well. One needs to have the possibility for
1542 * at least one ".nfs..." file in each directory the file ever gets
1543 * moved or linked to which happens automagically with the new
1544 * implementation that only depends on the dcache stuff instead of
1545 * using the inode layer
1546 *
1547 * Unfortunately, things are a little more complicated than indicated
1548 * above. For a cross-directory move, we want to make sure we can get
1549 * rid of the old inode after the operation. This means there must be
1550 * no pending writes (if it's a file), and the use count must be 1.
1551 * If these conditions are met, we can drop the dentries before doing
1552 * the rename.
1553 */
1554static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1555 struct inode *new_dir, struct dentry *new_dentry)
1556{
1557 struct inode *old_inode = old_dentry->d_inode;
1558 struct inode *new_inode = new_dentry->d_inode;
1559 struct dentry *dentry = NULL, *rehash = NULL;
1560 int error = -EBUSY;
1561
1562 /*
1563 * To prevent any new references to the target during the rename,
1564 * we unhash the dentry and free the inode in advance.
1565 */
1566 lock_kernel();
1567 if (!d_unhashed(new_dentry)) {
1568 d_drop(new_dentry);
1569 rehash = new_dentry;
1570 }
1571
1572 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1573 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1574 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1575 atomic_read(&new_dentry->d_count));
1576
1577 /*
1578 * First check whether the target is busy ... we can't
1579 * safely do _any_ rename if the target is in use.
1580 *
1581 * For files, make a copy of the dentry and then do a
1582 * silly-rename. If the silly-rename succeeds, the
1583 * copied dentry is hashed and becomes the new target.
1584 */
1585 if (!new_inode)
1586 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001587 if (S_ISDIR(new_inode->i_mode)) {
1588 error = -EISDIR;
1589 if (!S_ISDIR(old_inode->i_mode))
1590 goto out;
1591 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 int err;
1593 /* copy the target dentry's name */
1594 dentry = d_alloc(new_dentry->d_parent,
1595 &new_dentry->d_name);
1596 if (!dentry)
1597 goto out;
1598
1599 /* silly-rename the existing target ... */
1600 err = nfs_sillyrename(new_dir, new_dentry);
1601 if (!err) {
1602 new_dentry = rehash = dentry;
1603 new_inode = NULL;
1604 /* instantiate the replacement target */
1605 d_instantiate(new_dentry, NULL);
Jesper Juhl7a13e932007-04-26 00:29:02 -07001606 } else if (atomic_read(&new_dentry->d_count) > 1)
1607 /* dentry still busy? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608 goto out;
Trond Myklebust20509f12005-08-25 16:25:34 -07001609 } else
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001610 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
1612go_ahead:
1613 /*
1614 * ... prune child dentries and writebacks if needed.
1615 */
1616 if (atomic_read(&old_dentry->d_count) > 1) {
Trond Myklebuste1552e12007-04-14 19:07:28 -04001617 if (S_ISREG(old_inode->i_mode))
1618 nfs_wb_all(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 shrink_dcache_parent(old_dentry);
1620 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001621 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622
Trond Myklebust24174112006-01-03 09:55:33 +01001623 if (new_inode != NULL) {
1624 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1629 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001630 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631out:
1632 if (rehash)
1633 d_rehash(rehash);
1634 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07001635 d_move(old_dentry, new_dentry);
Chuck Lever8fb559f2007-09-24 15:40:16 -04001636 nfs_set_verifier(new_dentry,
1637 nfs_save_change_attribute(new_dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 }
1639
1640 /* new dentry created? */
1641 if (dentry)
1642 dput(dentry);
1643 unlock_kernel();
1644 return error;
1645}
1646
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001647static DEFINE_SPINLOCK(nfs_access_lru_lock);
1648static LIST_HEAD(nfs_access_lru_list);
1649static atomic_long_t nfs_access_nr_entries;
1650
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001651static void nfs_access_free_entry(struct nfs_access_entry *entry)
1652{
1653 put_rpccred(entry->cred);
1654 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001655 smp_mb__before_atomic_dec();
1656 atomic_long_dec(&nfs_access_nr_entries);
1657 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001658}
1659
Trond Myklebust979df722006-07-25 11:28:19 -04001660int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1661{
1662 LIST_HEAD(head);
1663 struct nfs_inode *nfsi;
1664 struct nfs_access_entry *cache;
1665
Trond Myklebust979df722006-07-25 11:28:19 -04001666restart:
Trond Myklebusta50f7952007-06-05 19:23:43 -04001667 spin_lock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001668 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1669 struct inode *inode;
1670
1671 if (nr_to_scan-- == 0)
1672 break;
1673 inode = igrab(&nfsi->vfs_inode);
1674 if (inode == NULL)
1675 continue;
1676 spin_lock(&inode->i_lock);
1677 if (list_empty(&nfsi->access_cache_entry_lru))
1678 goto remove_lru_entry;
1679 cache = list_entry(nfsi->access_cache_entry_lru.next,
1680 struct nfs_access_entry, lru);
1681 list_move(&cache->lru, &head);
1682 rb_erase(&cache->rb_node, &nfsi->access_cache);
1683 if (!list_empty(&nfsi->access_cache_entry_lru))
1684 list_move_tail(&nfsi->access_cache_inode_lru,
1685 &nfs_access_lru_list);
1686 else {
1687remove_lru_entry:
1688 list_del_init(&nfsi->access_cache_inode_lru);
1689 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1690 }
1691 spin_unlock(&inode->i_lock);
Trond Myklebusta50f7952007-06-05 19:23:43 -04001692 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001693 iput(inode);
1694 goto restart;
1695 }
1696 spin_unlock(&nfs_access_lru_lock);
1697 while (!list_empty(&head)) {
1698 cache = list_entry(head.next, struct nfs_access_entry, lru);
1699 list_del(&cache->lru);
1700 nfs_access_free_entry(cache);
1701 }
1702 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1703}
1704
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001705static void __nfs_access_zap_cache(struct inode *inode)
1706{
1707 struct nfs_inode *nfsi = NFS_I(inode);
1708 struct rb_root *root_node = &nfsi->access_cache;
1709 struct rb_node *n, *dispose = NULL;
1710 struct nfs_access_entry *entry;
1711
1712 /* Unhook entries from the cache */
1713 while ((n = rb_first(root_node)) != NULL) {
1714 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1715 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001716 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001717 n->rb_left = dispose;
1718 dispose = n;
1719 }
1720 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1721 spin_unlock(&inode->i_lock);
1722
1723 /* Now kill them all! */
1724 while (dispose != NULL) {
1725 n = dispose;
1726 dispose = n->rb_left;
1727 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1728 }
1729}
1730
1731void nfs_access_zap_cache(struct inode *inode)
1732{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001733 /* Remove from global LRU init */
1734 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1735 spin_lock(&nfs_access_lru_lock);
1736 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1737 spin_unlock(&nfs_access_lru_lock);
1738 }
1739
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001740 spin_lock(&inode->i_lock);
1741 /* This will release the spinlock */
1742 __nfs_access_zap_cache(inode);
1743}
1744
1745static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1746{
1747 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1748 struct nfs_access_entry *entry;
1749
1750 while (n != NULL) {
1751 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1752
1753 if (cred < entry->cred)
1754 n = n->rb_left;
1755 else if (cred > entry->cred)
1756 n = n->rb_right;
1757 else
1758 return entry;
1759 }
1760 return NULL;
1761}
1762
Trond Myklebustaf22f942007-08-10 17:45:10 -04001763static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764{
Chuck Lever55296802005-08-18 11:24:09 -07001765 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001766 struct nfs_access_entry *cache;
1767 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001769 spin_lock(&inode->i_lock);
1770 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1771 goto out_zap;
1772 cache = nfs_access_search_rbtree(inode, cred);
1773 if (cache == NULL)
1774 goto out;
Trond Myklebustc7c20972007-09-28 19:22:40 -04001775 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001776 goto out_stale;
1777 res->jiffies = cache->jiffies;
1778 res->cred = cache->cred;
1779 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001780 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001781 err = 0;
1782out:
1783 spin_unlock(&inode->i_lock);
1784 return err;
1785out_stale:
1786 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001787 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001788 spin_unlock(&inode->i_lock);
1789 nfs_access_free_entry(cache);
1790 return -ENOENT;
1791out_zap:
1792 /* This will release the spinlock */
1793 __nfs_access_zap_cache(inode);
1794 return -ENOENT;
1795}
1796
1797static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1798{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001799 struct nfs_inode *nfsi = NFS_I(inode);
1800 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001801 struct rb_node **p = &root_node->rb_node;
1802 struct rb_node *parent = NULL;
1803 struct nfs_access_entry *entry;
1804
1805 spin_lock(&inode->i_lock);
1806 while (*p != NULL) {
1807 parent = *p;
1808 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1809
1810 if (set->cred < entry->cred)
1811 p = &parent->rb_left;
1812 else if (set->cred > entry->cred)
1813 p = &parent->rb_right;
1814 else
1815 goto found;
1816 }
1817 rb_link_node(&set->rb_node, parent, p);
1818 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001819 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001820 spin_unlock(&inode->i_lock);
1821 return;
1822found:
1823 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001824 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1825 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001826 spin_unlock(&inode->i_lock);
1827 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828}
1829
Trond Myklebustaf22f942007-08-10 17:45:10 -04001830static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001832 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1833 if (cache == NULL)
1834 return;
1835 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001837 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001839
1840 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001841
1842 /* Update accounting */
1843 smp_mb__before_atomic_inc();
1844 atomic_long_inc(&nfs_access_nr_entries);
1845 smp_mb__after_atomic_inc();
1846
1847 /* Add inode to global LRU list */
1848 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1849 spin_lock(&nfs_access_lru_lock);
1850 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1851 spin_unlock(&nfs_access_lru_lock);
1852 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853}
1854
1855static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1856{
1857 struct nfs_access_entry cache;
1858 int status;
1859
1860 status = nfs_access_get_cached(inode, cred, &cache);
1861 if (status == 0)
1862 goto out;
1863
1864 /* Be clever: ask server to check for all possible rights */
1865 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1866 cache.cred = cred;
1867 cache.jiffies = jiffies;
1868 status = NFS_PROTO(inode)->access(inode, &cache);
1869 if (status != 0)
1870 return status;
1871 nfs_access_add_cache(inode, &cache);
1872out:
1873 if ((cache.mask & mask) == mask)
1874 return 0;
1875 return -EACCES;
1876}
1877
Trond Myklebustaf22f942007-08-10 17:45:10 -04001878static int nfs_open_permission_mask(int openflags)
1879{
1880 int mask = 0;
1881
1882 if (openflags & FMODE_READ)
1883 mask |= MAY_READ;
1884 if (openflags & FMODE_WRITE)
1885 mask |= MAY_WRITE;
1886 if (openflags & FMODE_EXEC)
1887 mask |= MAY_EXEC;
1888 return mask;
1889}
1890
1891int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1892{
1893 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1894}
1895
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1897{
1898 struct rpc_cred *cred;
1899 int res = 0;
1900
Chuck Lever91d5b472006-03-20 13:44:14 -05001901 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1902
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 if (mask == 0)
1904 goto out;
1905 /* Is this sys_access() ? */
1906 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1907 goto force_lookup;
1908
1909 switch (inode->i_mode & S_IFMT) {
1910 case S_IFLNK:
1911 goto out;
1912 case S_IFREG:
1913 /* NFSv4 has atomic_open... */
1914 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1915 && nd != NULL
1916 && (nd->flags & LOOKUP_OPEN))
1917 goto out;
1918 break;
1919 case S_IFDIR:
1920 /*
1921 * Optimize away all write operations, since the server
1922 * will check permissions when we perform the op.
1923 */
1924 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1925 goto out;
1926 }
1927
1928force_lookup:
1929 lock_kernel();
1930
1931 if (!NFS_PROTO(inode)->access)
1932 goto out_notsup;
1933
1934 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1935 if (!IS_ERR(cred)) {
1936 res = nfs_do_access(inode, cred, mask);
1937 put_rpccred(cred);
1938 } else
1939 res = PTR_ERR(cred);
1940 unlock_kernel();
1941out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001942 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1943 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 return res;
1945out_notsup:
1946 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1947 if (res == 0)
1948 res = generic_permission(inode, mask, NULL);
1949 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001950 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951}
1952
1953/*
1954 * Local variables:
1955 * version-control: t
1956 * kept-new-versions: 5
1957 * End:
1958 */