blob: a03ed2f85047e70cdf92b8e0dd9a0229b9257ca2 [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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Linus Torvalds1da177e2005-04-16 15:20:36 -070042/* #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
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080068const struct inode_operations nfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 .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
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080084const struct inode_operations nfs3_dir_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000085 .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 *);
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800107const struct inode_operations nfs4_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 .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
Al Viro0dbb4c62006-10-19 23:28:49 -0700145typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146typedef struct {
147 struct file *file;
148 struct page *page;
149 unsigned long page_index;
Al Viro0dbb4c62006-10-19 23:28:49 -0700150 __be32 *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;
Neil Brown1f4eab72007-04-16 09:35:27 +1000157 unsigned long timestamp;
158 int timestamp_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159} nfs_readdir_descriptor_t;
160
161/* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
163 *
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
167 * to be.
168 *
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
172 */
173static
174int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
175{
176 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800177 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 struct rpc_cred *cred = nfs_file_cred(file);
179 unsigned long timestamp;
180 int error;
181
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__, (long long)desc->entry->cookie,
184 page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185
186 again:
187 timestamp = jiffies;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 NFS_SERVER(inode)->dtsize, desc->plus);
190 if (error < 0) {
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error == -ENOTSUPP && desc->plus) {
193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Chuck Lever412d5822005-08-18 11:24:11 -0700194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 desc->plus = 0;
196 goto again;
197 }
198 goto error;
199 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000200 desc->timestamp = timestamp;
201 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 /* Ensure consistent page alignment of the data.
204 * Note: assumes we have exclusive access to this mapping either
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800205 * through inode->i_mutex or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 */
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700207 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
208 /* Should never happen */
209 nfs_zap_mapping(inode, inode->i_mapping);
210 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 unlock_page(page);
212 return 0;
213 error:
214 SetPageError(page);
215 unlock_page(page);
216 nfs_zap_caches(inode);
217 desc->error = error;
218 return -EIO;
219}
220
221static inline
222int dir_decode(nfs_readdir_descriptor_t *desc)
223{
Al Viro0dbb4c62006-10-19 23:28:49 -0700224 __be32 *p = desc->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 p = desc->decode(p, desc->entry, desc->plus);
226 if (IS_ERR(p))
227 return PTR_ERR(p);
228 desc->ptr = p;
Neil Brown1f4eab72007-04-16 09:35:27 +1000229 if (desc->timestamp_valid)
230 desc->entry->fattr->time_start = desc->timestamp;
231 else
232 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return 0;
234}
235
236static inline
237void dir_page_release(nfs_readdir_descriptor_t *desc)
238{
239 kunmap(desc->page);
240 page_cache_release(desc->page);
241 desc->page = NULL;
242 desc->ptr = NULL;
243}
244
245/*
246 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000247 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 *
249 * If the end of the buffer has been reached, return -EAGAIN, if not,
250 * return the offset within the buffer of the next entry to be
251 * read.
252 */
253static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000254int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255{
256 struct nfs_entry *entry = desc->entry;
257 int loop_count = 0,
258 status;
259
260 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500261 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
262 __FUNCTION__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000263 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 break;
265 if (loop_count++ > 200) {
266 loop_count = 0;
267 schedule();
268 }
269 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 return status;
271}
272
273/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000274 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000275 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000276 *
277 * If the end of the buffer has been reached, return -EAGAIN, if not,
278 * return the offset within the buffer of the next entry to be
279 * read.
280 */
281static inline
282int find_dirent_index(nfs_readdir_descriptor_t *desc)
283{
284 struct nfs_entry *entry = desc->entry;
285 int loop_count = 0,
286 status;
287
288 for(;;) {
289 status = dir_decode(desc);
290 if (status)
291 break;
292
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500293 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
294 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000295
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000296 if (desc->file->f_pos == desc->current_index) {
297 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000298 break;
299 }
300 desc->current_index++;
301 if (loop_count++ > 200) {
302 loop_count = 0;
303 schedule();
304 }
305 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000306 return status;
307}
308
309/*
310 * Find the given page, and call find_dirent() or find_dirent_index in
311 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
313static inline
314int find_dirent_page(nfs_readdir_descriptor_t *desc)
315{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800316 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 struct page *page;
318 int status;
319
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500320 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
321 __FUNCTION__, desc->page_index,
322 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
Neil Brown1f4eab72007-04-16 09:35:27 +1000324 /* If we find the page in the page_cache, we cannot be sure
325 * how fresh the data is, so we will ignore readdir_plus attributes.
326 */
327 desc->timestamp_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 page = read_cache_page(inode->i_mapping, desc->page_index,
329 (filler_t *)nfs_readdir_filler, desc);
330 if (IS_ERR(page)) {
331 status = PTR_ERR(page);
332 goto out;
333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 /* NOTE: Someone else may have changed the READDIRPLUS flag */
336 desc->page = page;
337 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000338 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000339 status = find_dirent(desc);
340 else
341 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 if (status < 0)
343 dir_page_release(desc);
344 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500345 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
349/*
350 * Recurse through the page cache pages, and return a
351 * filled nfs_entry structure of the next directory entry if possible.
352 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000353 * The target for the search is '*desc->dir_cookie' if non-0,
354 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 */
356static inline
357int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
358{
359 int loop_count = 0;
360 int res;
361
Olivier Galibert00a92642005-06-22 17:16:29 +0000362 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000363 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500364 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
365 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000366 desc->page_index = 0;
367 desc->entry->cookie = desc->entry->prev_cookie = 0;
368 desc->entry->eof = 0;
369 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000370 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500371 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
372 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 for (;;) {
375 res = find_dirent_page(desc);
376 if (res != -EAGAIN)
377 break;
378 /* Align to beginning of next page */
379 desc->page_index ++;
380 if (loop_count++ > 200) {
381 loop_count = 0;
382 schedule();
383 }
384 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500385
386 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 return res;
388}
389
390static inline unsigned int dt_type(struct inode *inode)
391{
392 return (inode->i_mode >> 12) & 15;
393}
394
395static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
396
397/*
398 * Once we've found the start of the dirent within a page: fill 'er up...
399 */
400static
401int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
402 filldir_t filldir)
403{
404 struct file *file = desc->file;
405 struct nfs_entry *entry = desc->entry;
406 struct dentry *dentry = NULL;
Peter Staubach4e769b92007-08-03 15:07:10 -0400407 u64 fileid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 int loop_count = 0,
409 res;
410
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500411 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
412 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
414 for(;;) {
415 unsigned d_type = DT_UNKNOWN;
416 /* Note: entry->prev_cookie contains the cookie for
417 * retrieving the current dirent on the server */
Peter Staubach4e769b92007-08-03 15:07:10 -0400418 fileid = entry->ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
420 /* Get a dentry if we have one */
421 if (dentry != NULL)
422 dput(dentry);
423 dentry = nfs_readdir_lookup(desc);
424
425 /* Use readdirplus info */
426 if (dentry != NULL && dentry->d_inode != NULL) {
427 d_type = dt_type(dentry->d_inode);
Peter Staubach4e769b92007-08-03 15:07:10 -0400428 fileid = NFS_FILEID(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 }
430
431 res = filldir(dirent, entry->name, entry->len,
Olivier Galibert00a92642005-06-22 17:16:29 +0000432 file->f_pos, fileid, 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
566 while(!desc->entry->eof) {
567 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000568
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 if (res == -EBADCOOKIE) {
570 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000571 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 /* Or that the server has 'lost' a cookie */
573 res = uncached_readdir(desc, dirent, filldir);
574 if (res >= 0)
575 continue;
576 }
577 res = 0;
578 break;
579 }
580 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700581 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 nfs_zap_caches(inode);
583 desc->plus = 0;
584 desc->entry->eof = 0;
585 continue;
586 }
587 if (res < 0)
588 break;
589
590 res = nfs_do_filldir(desc, dirent, filldir);
591 if (res < 0) {
592 res = 0;
593 break;
594 }
595 }
596 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500597 if (res > 0)
598 res = 0;
599 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
600 dentry->d_parent->d_name.name, dentry->d_name.name,
601 res);
602 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Trond Myklebust10afec92007-05-14 17:16:04 -0400605static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000606{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800607 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000608 switch (origin) {
609 case 1:
610 offset += filp->f_pos;
611 case 0:
612 if (offset >= 0)
613 break;
614 default:
615 offset = -EINVAL;
616 goto out;
617 }
618 if (offset != filp->f_pos) {
619 filp->f_pos = offset;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400620 nfs_file_open_context(filp)->dir_cookie = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000621 }
622out:
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800623 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000624 return offset;
625}
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627/*
628 * All directory operations under NFS are synchronous, so fsync()
629 * is a dummy operation.
630 */
Trond Myklebust10afec92007-05-14 17:16:04 -0400631static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500633 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
634 dentry->d_parent->d_name.name, dentry->d_name.name,
635 datasync);
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 return 0;
638}
639
640/*
641 * A check for whether or not the parent directory has changed.
642 * In the case it has, we assume that the dentries are untrustworthy
643 * and may need to be looked up again.
644 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500645static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
647 if (IS_ROOT(dentry))
648 return 1;
Trond Myklebust4b841732007-09-29 17:15:01 -0400649 if (nfs_verify_change_attribute(dir, dentry->d_time))
Trond Myklebust6ecc5e82007-09-28 14:20:33 -0400650 return 1;
651 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652}
653
654static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
655{
Trond Myklebust3062c532007-07-14 17:36:45 -0400656 dentry->d_time = verf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657}
658
659/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400660 * Return the intent data that applies to this particular path component
661 *
662 * Note that the current set of intents only apply to the very last
663 * component of the path.
664 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
665 */
666static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
667{
668 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
669 return 0;
670 return nd->flags & mask;
671}
672
673/*
674 * Inode and filehandle revalidation for lookups.
675 *
676 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
677 * or if the intent information indicates that we're about to open this
678 * particular file and the "nocto" mount flag is not set.
679 *
680 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681static inline
682int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
683{
684 struct nfs_server *server = NFS_SERVER(inode);
685
686 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400688 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 goto out_force;
690 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400691 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400692 !(server->flags & NFS_MOUNT_NOCTO) &&
693 (S_ISREG(inode->i_mode) ||
694 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 goto out_force;
696 }
697 return nfs_revalidate_inode(server, inode);
698out_force:
699 return __nfs_revalidate_inode(server, inode);
700}
701
702/*
703 * We judge how long we want to trust negative
704 * dentries by looking at the parent inode mtime.
705 *
706 * If parent mtime has changed, we revalidate, else we wait for a
707 * period corresponding to the parent's attribute cache timeout value.
708 */
709static inline
710int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
711 struct nameidata *nd)
712{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400714 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 return 0;
716 return !nfs_check_verifier(dir, dentry);
717}
718
719/*
720 * This is called every time the dcache has a lookup hit,
721 * and we should check whether we can really trust that
722 * lookup.
723 *
724 * NOTE! The hit can be a negative hit too, don't assume
725 * we have an inode!
726 *
727 * If the parent directory is seen to have changed, we throw out the
728 * cached dentry and do a new lookup.
729 */
730static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
731{
732 struct inode *dir;
733 struct inode *inode;
734 struct dentry *parent;
735 int error;
736 struct nfs_fh fhandle;
737 struct nfs_fattr fattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 parent = dget_parent(dentry);
740 lock_kernel();
741 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500742 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 inode = dentry->d_inode;
744
Trond Myklebuste70c4902007-05-09 09:00:18 -0400745 /* Revalidate parent directory attribute cache */
746 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
747 goto out_zap_parent;
748
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 /* Force a full look up iff the parent directory has changed */
763 if (nfs_check_verifier(dir, dentry)) {
764 if (nfs_lookup_verify_inode(inode, nd))
765 goto out_zap_parent;
766 goto out_valid;
767 }
768
769 if (NFS_STALE(inode))
770 goto out_bad;
771
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
773 if (error)
774 goto out_bad;
775 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
776 goto out_bad;
777 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
778 goto out_bad;
779
Trond Myklebustcf8ba452007-10-01 13:46:53 -0400780 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 out_valid:
782 unlock_kernel();
783 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500784 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
785 __FUNCTION__, dentry->d_parent->d_name.name,
786 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 return 1;
788out_zap_parent:
789 nfs_zap_caches(dir);
790 out_bad:
Trond Myklebusta1643a92007-09-29 17:25:43 -0400791 nfs_mark_for_revalidate(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 if (inode && S_ISDIR(inode->i_mode)) {
793 /* Purge readdir caches. */
794 nfs_zap_caches(inode);
795 /* If we have submounts, don't unhash ! */
796 if (have_submounts(dentry))
797 goto out_valid;
798 shrink_dcache_parent(dentry);
799 }
800 d_drop(dentry);
801 unlock_kernel();
802 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500803 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
804 __FUNCTION__, dentry->d_parent->d_name.name,
805 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 return 0;
807}
808
809/*
810 * This is called from dput() when d_count is going to 0.
811 */
812static int nfs_dentry_delete(struct dentry *dentry)
813{
814 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
815 dentry->d_parent->d_name.name, dentry->d_name.name,
816 dentry->d_flags);
817
818 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
819 /* Unhash it, so that ->d_iput() would be called */
820 return 1;
821 }
822 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
823 /* Unhash it, so that ancestors of killed async unlink
824 * files will be cleaned up during umount */
825 return 1;
826 }
827 return 0;
828
829}
830
831/*
832 * Called when the dentry loses inode.
833 * We use it to clean up silly-renamed files.
834 */
835static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
836{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700837 nfs_inode_return_delegation(inode);
Neil Brown83672d32007-02-26 12:48:25 +1100838 if (S_ISDIR(inode->i_mode))
839 /* drop any readdir cache as it could easily be old */
840 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
841
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
843 lock_kernel();
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700844 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -0400845 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 unlock_kernel();
847 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 iput(inode);
849}
850
851struct dentry_operations nfs_dentry_operations = {
852 .d_revalidate = nfs_lookup_revalidate,
853 .d_delete = nfs_dentry_delete,
854 .d_iput = nfs_dentry_iput,
855};
856
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400857/*
858 * Use intent information to check whether or not we're going to do
859 * an O_EXCL create using this path component.
860 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861static inline
862int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
863{
864 if (NFS_PROTO(dir)->version == 2)
865 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400866 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 return 0;
868 return (nd->intent.open.flags & O_EXCL) != 0;
869}
870
Trond Myklebusta0356862007-06-05 13:26:15 -0400871static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
Trond Myklebust55a97592006-06-09 09:34:19 -0400872{
873 struct nfs_server *server = NFS_SERVER(dir);
874
875 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
Trond Myklebusta0356862007-06-05 13:26:15 -0400876 /* Revalidate fsid using the parent directory */
877 return __nfs_revalidate_inode(server, dir);
Trond Myklebust55a97592006-06-09 09:34:19 -0400878 return 0;
879}
880
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
882{
883 struct dentry *res;
884 struct inode *inode = NULL;
885 int error;
886 struct nfs_fh fhandle;
887 struct nfs_fattr fattr;
888
889 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
890 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500891 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892
893 res = ERR_PTR(-ENAMETOOLONG);
894 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
895 goto out;
896
897 res = ERR_PTR(-ENOMEM);
898 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
899
900 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Trond Myklebustfd684072006-09-05 12:27:44 -0400902 /*
903 * If we're doing an exclusive create, optimize away the lookup
904 * but don't hash the dentry.
905 */
906 if (nfs_is_exclusive_create(dir, nd)) {
907 d_instantiate(dentry, NULL);
908 res = NULL;
909 goto out_unlock;
910 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911
912 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
913 if (error == -ENOENT)
914 goto no_entry;
915 if (error < 0) {
916 res = ERR_PTR(error);
917 goto out_unlock;
918 }
Trond Myklebusta0356862007-06-05 13:26:15 -0400919 error = nfs_reval_fsid(dir, &fattr);
Trond Myklebust55a97592006-06-09 09:34:19 -0400920 if (error < 0) {
921 res = ERR_PTR(error);
922 goto out_unlock;
923 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500925 res = (struct dentry *)inode;
926 if (IS_ERR(res))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 goto out_unlock;
David Howells54ceac42006-08-22 20:06:13 -0400928
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400930 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -0700931 if (res != NULL) {
932 if (IS_ERR(res))
933 goto out_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
937out_unlock:
938 unlock_kernel();
939out:
940 return res;
941}
942
943#ifdef CONFIG_NFS_V4
944static int nfs_open_revalidate(struct dentry *, struct nameidata *);
945
946struct dentry_operations nfs4_dentry_operations = {
947 .d_revalidate = nfs_open_revalidate,
948 .d_delete = nfs_dentry_delete,
949 .d_iput = nfs_dentry_iput,
950};
951
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400952/*
953 * Use intent information to determine whether we need to substitute
954 * the NFSv4-style stateful OPEN for the LOOKUP call
955 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956static int is_atomic_open(struct inode *dir, struct nameidata *nd)
957{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400958 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 return 0;
960 /* NFS does not (yet) have a stateful open for directories */
961 if (nd->flags & LOOKUP_DIRECTORY)
962 return 0;
963 /* Are we trying to write to a read only partition? */
964 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
965 return 0;
966 return 1;
967}
968
969static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
970{
971 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 int error;
973
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500974 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
975 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
976
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 /* Check that we are indeed trying to open this file */
978 if (!is_atomic_open(dir, nd))
979 goto no_open;
980
981 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
982 res = ERR_PTR(-ENAMETOOLONG);
983 goto out;
984 }
985 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
986
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400987 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
988 * the dentry. */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700989 if (nd->intent.open.flags & O_EXCL) {
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400990 d_instantiate(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -0700991 goto out;
992 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
994 /* Open the file on the server */
995 lock_kernel();
996 /* Revalidate parent directory attribute cache */
997 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
998 if (error < 0) {
999 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -07001000 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 goto out;
1002 }
1003
1004 if (nd->intent.open.flags & O_CREAT) {
1005 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001006 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 nfs_end_data_update(dir);
1008 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -07001009 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001011 if (IS_ERR(res)) {
1012 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 switch (error) {
1014 /* Make a negative dentry */
1015 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001016 res = NULL;
1017 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001019 case -EISDIR:
1020 case -ENOTDIR:
1021 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 case -ELOOP:
1023 if (!(nd->intent.open.flags & O_NOFOLLOW))
1024 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 /* case -EINVAL: */
1026 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 goto out;
1028 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001029 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 dentry = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1032out:
1033 return res;
1034no_open:
1035 return nfs_lookup(dir, dentry, nd);
1036}
1037
1038static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1039{
1040 struct dentry *parent = NULL;
1041 struct inode *inode = dentry->d_inode;
1042 struct inode *dir;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 int openflags, ret = 0;
1044
1045 parent = dget_parent(dentry);
1046 dir = parent->d_inode;
1047 if (!is_atomic_open(dir, nd))
1048 goto no_open;
1049 /* We can't create new files in nfs_open_revalidate(), so we
1050 * optimize away revalidation of negative dentries.
1051 */
1052 if (inode == NULL)
1053 goto out;
1054 /* NFS only supports OPEN on regular files */
1055 if (!S_ISREG(inode->i_mode))
1056 goto no_open;
1057 openflags = nd->intent.open.flags;
1058 /* We cannot do exclusive creation on a positive dentry */
1059 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1060 goto no_open;
1061 /* We can't create new files, or truncate existing ones here */
1062 openflags &= ~(O_CREAT|O_TRUNC);
1063
1064 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001065 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 * operations that change the directory. We therefore save the
1067 * change attribute *before* we do the RPC call.
1068 */
1069 lock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001070 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Trond Myklebust446e5342007-10-03 15:58:38 -04001071 if (ret == 1)
Trond Myklebustcf8ba452007-10-01 13:46:53 -04001072 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 unlock_kernel();
1074out:
1075 dput(parent);
1076 if (!ret)
1077 d_drop(dentry);
1078 return ret;
1079no_open:
1080 dput(parent);
1081 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1082 return 1;
1083 return nfs_lookup_revalidate(dentry, nd);
1084}
1085#endif /* CONFIG_NFSV4 */
1086
1087static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1088{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -08001089 struct dentry *parent = desc->file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 struct inode *dir = parent->d_inode;
1091 struct nfs_entry *entry = desc->entry;
1092 struct dentry *dentry, *alias;
1093 struct qstr name = {
1094 .name = entry->name,
1095 .len = entry->len,
1096 };
1097 struct inode *inode;
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001098 unsigned long verf = nfs_save_change_attribute(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 switch (name.len) {
1101 case 2:
1102 if (name.name[0] == '.' && name.name[1] == '.')
1103 return dget_parent(parent);
1104 break;
1105 case 1:
1106 if (name.name[0] == '.')
1107 return dget(parent);
1108 }
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001109
1110 spin_lock(&dir->i_lock);
1111 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1112 spin_unlock(&dir->i_lock);
1113 return NULL;
1114 }
1115 spin_unlock(&dir->i_lock);
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 name.hash = full_name_hash(name.name, name.len);
1118 dentry = d_lookup(parent, &name);
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001119 if (dentry != NULL) {
Trond Myklebustef75c792007-01-16 10:09:44 -05001120 /* Is this a positive dentry that matches the readdir info? */
1121 if (dentry->d_inode != NULL &&
1122 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1123 d_mountpoint(dentry))) {
1124 if (!desc->plus || entry->fh->size == 0)
1125 return dentry;
1126 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1127 entry->fh) == 0)
1128 goto out_renew;
1129 }
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001130 /* No, so d_drop to allow one to be created */
1131 d_drop(dentry);
1132 dput(dentry);
1133 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1135 return NULL;
Trond Myklebust54af3bb2007-09-28 12:27:41 -04001136 if (name.len > NFS_SERVER(dir)->namelen)
1137 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001138 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 dentry = d_alloc(parent, &name);
1140 if (dentry == NULL)
1141 return NULL;
1142 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1143 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001144 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 dput(dentry);
1146 return NULL;
1147 }
David Howells54ceac42006-08-22 20:06:13 -04001148
1149 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 if (alias != NULL) {
1151 dput(dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001152 if (IS_ERR(alias))
1153 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 dentry = alias;
1155 }
David Howells54ceac42006-08-22 20:06:13 -04001156
Trond Myklebustc79ba782007-01-31 08:16:24 -05001157out_renew:
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001158 nfs_set_verifier(dentry, verf);
Trond Myklebustc79ba782007-01-31 08:16:24 -05001159 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160}
1161
1162/*
1163 * Code common to create, mkdir, and mknod.
1164 */
1165int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1166 struct nfs_fattr *fattr)
1167{
Trond Myklebustfab728e2007-09-29 17:41:33 -04001168 struct dentry *parent = dget_parent(dentry);
1169 struct inode *dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 struct inode *inode;
1171 int error = -EACCES;
1172
Trond Myklebustfab728e2007-09-29 17:41:33 -04001173 d_drop(dentry);
1174
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 /* We may have been initialized further down */
1176 if (dentry->d_inode)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001177 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 if (fhandle->size == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1180 if (error)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001181 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 }
Trond Myklebust5724ab32007-10-01 21:51:38 -04001183 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1185 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001186 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187 if (error < 0)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001188 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001191 error = PTR_ERR(inode);
1192 if (IS_ERR(inode))
Trond Myklebustfab728e2007-09-29 17:41:33 -04001193 goto out_error;
1194 d_add(dentry, inode);
1195out:
1196 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 return 0;
Trond Myklebustfab728e2007-09-29 17:41:33 -04001198out_error:
1199 nfs_mark_for_revalidate(dir);
1200 dput(parent);
1201 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202}
1203
1204/*
1205 * Following a failed create operation, we drop the dentry rather
1206 * than retain a negative dentry. This avoids a problem in the event
1207 * that the operation succeeded on the server, but an error in the
1208 * reply path made it appear to have failed.
1209 */
1210static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1211 struct nameidata *nd)
1212{
1213 struct iattr attr;
1214 int error;
1215 int open_flags = 0;
1216
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001217 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1218 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220 attr.ia_mode = mode;
1221 attr.ia_valid = ATTR_MODE;
1222
Trond Myklebustad389da2007-06-05 12:30:00 -04001223 if ((nd->flags & LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 open_flags = nd->intent.open.flags;
1225
1226 lock_kernel();
1227 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001228 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 nfs_end_data_update(dir);
1230 if (error != 0)
1231 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 unlock_kernel();
1233 return 0;
1234out_err:
1235 unlock_kernel();
1236 d_drop(dentry);
1237 return error;
1238}
1239
1240/*
1241 * See comments for nfs_proc_create regarding failed operations.
1242 */
1243static int
1244nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1245{
1246 struct iattr attr;
1247 int status;
1248
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001249 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1250 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251
1252 if (!new_valid_dev(rdev))
1253 return -EINVAL;
1254
1255 attr.ia_mode = mode;
1256 attr.ia_valid = ATTR_MODE;
1257
1258 lock_kernel();
1259 nfs_begin_data_update(dir);
1260 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1261 nfs_end_data_update(dir);
1262 if (status != 0)
1263 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 unlock_kernel();
1265 return 0;
1266out_err:
1267 unlock_kernel();
1268 d_drop(dentry);
1269 return status;
1270}
1271
1272/*
1273 * See comments for nfs_proc_create regarding failed operations.
1274 */
1275static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1276{
1277 struct iattr attr;
1278 int error;
1279
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001280 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1281 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282
1283 attr.ia_valid = ATTR_MODE;
1284 attr.ia_mode = mode | S_IFDIR;
1285
1286 lock_kernel();
1287 nfs_begin_data_update(dir);
1288 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1289 nfs_end_data_update(dir);
1290 if (error != 0)
1291 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 unlock_kernel();
1293 return 0;
1294out_err:
1295 d_drop(dentry);
1296 unlock_kernel();
1297 return error;
1298}
1299
1300static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1301{
1302 int error;
1303
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001304 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1305 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306
1307 lock_kernel();
1308 nfs_begin_data_update(dir);
1309 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1310 /* Ensure the VFS deletes this inode */
1311 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001312 clear_nlink(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 nfs_end_data_update(dir);
1314 unlock_kernel();
1315
1316 return error;
1317}
1318
1319static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1320{
1321 static unsigned int sillycounter;
Peter Staubach4e769b92007-08-03 15:07:10 -04001322 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 const int countersize = sizeof(sillycounter)*2;
Peter Staubach4e769b92007-08-03 15:07:10 -04001324 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 char silly[slen+1];
1326 struct qstr qsilly;
1327 struct dentry *sdentry;
1328 int error = -EIO;
1329
1330 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1331 dentry->d_parent->d_name.name, dentry->d_name.name,
1332 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001333 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 /*
1336 * We don't allow a dentry to be silly-renamed twice.
1337 */
1338 error = -EBUSY;
1339 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1340 goto out;
1341
Peter Staubach4e769b92007-08-03 15:07:10 -04001342 sprintf(silly, ".nfs%*.*Lx",
1343 fileidsize, fileidsize,
1344 (unsigned long long)NFS_FILEID(dentry->d_inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Trond Myklebust34ea8182005-11-04 15:35:02 -05001346 /* Return delegation in anticipation of the rename */
1347 nfs_inode_return_delegation(dentry->d_inode);
1348
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 sdentry = NULL;
1350 do {
1351 char *suffix = silly + slen - countersize;
1352
1353 dput(sdentry);
1354 sillycounter++;
1355 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1356
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001357 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1358 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
1360 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1361 /*
1362 * N.B. Better to return EBUSY here ... it could be
1363 * dangerous to delete the file while it's in use.
1364 */
1365 if (IS_ERR(sdentry))
1366 goto out;
1367 } while(sdentry->d_inode != NULL); /* need negative lookup */
1368
1369 qsilly.name = silly;
1370 qsilly.len = strlen(silly);
1371 nfs_begin_data_update(dir);
1372 if (dentry->d_inode) {
1373 nfs_begin_data_update(dentry->d_inode);
1374 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1375 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001376 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 nfs_end_data_update(dentry->d_inode);
1378 } else
1379 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1380 dir, &qsilly);
1381 nfs_end_data_update(dir);
1382 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1384 d_move(dentry, sdentry);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001385 error = nfs_async_unlink(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 /* If we return 0 we don't unlink */
1387 }
1388 dput(sdentry);
1389out:
1390 return error;
1391}
1392
1393/*
1394 * Remove a file after making sure there are no pending writes,
1395 * and after checking that the file has only one user.
1396 *
1397 * We invalidate the attribute cache and free the inode prior to the operation
1398 * to avoid possible races if the server reuses the inode.
1399 */
1400static int nfs_safe_remove(struct dentry *dentry)
1401{
1402 struct inode *dir = dentry->d_parent->d_inode;
1403 struct inode *inode = dentry->d_inode;
1404 int error = -EBUSY;
1405
1406 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1407 dentry->d_parent->d_name.name, dentry->d_name.name);
1408
1409 /* If the dentry was sillyrenamed, we simply call d_delete() */
1410 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1411 error = 0;
1412 goto out;
1413 }
1414
1415 nfs_begin_data_update(dir);
1416 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001417 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 nfs_begin_data_update(inode);
1419 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1420 /* The VFS may want to delete this inode */
1421 if (error == 0)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001422 drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001423 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 nfs_end_data_update(inode);
1425 } else
1426 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1427 nfs_end_data_update(dir);
1428out:
1429 return error;
1430}
1431
1432/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1433 * belongs to an active ".nfs..." file and we return -EBUSY.
1434 *
1435 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1436 */
1437static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1438{
1439 int error;
1440 int need_rehash = 0;
1441
1442 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1443 dir->i_ino, dentry->d_name.name);
1444
1445 lock_kernel();
1446 spin_lock(&dcache_lock);
1447 spin_lock(&dentry->d_lock);
1448 if (atomic_read(&dentry->d_count) > 1) {
1449 spin_unlock(&dentry->d_lock);
1450 spin_unlock(&dcache_lock);
Trond Myklebustccfeb502007-01-13 02:28:12 -05001451 /* Start asynchronous writeout of the inode */
1452 write_inode_now(dentry->d_inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 error = nfs_sillyrename(dir, dentry);
1454 unlock_kernel();
1455 return error;
1456 }
1457 if (!d_unhashed(dentry)) {
1458 __d_drop(dentry);
1459 need_rehash = 1;
1460 }
1461 spin_unlock(&dentry->d_lock);
1462 spin_unlock(&dcache_lock);
1463 error = nfs_safe_remove(dentry);
1464 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1466 } else if (need_rehash)
1467 d_rehash(dentry);
1468 unlock_kernel();
1469 return error;
1470}
1471
Chuck Lever873101b2006-08-22 20:06:23 -04001472/*
1473 * To create a symbolic link, most file systems instantiate a new inode,
1474 * add a page to it containing the path, then write it out to the disk
1475 * using prepare_write/commit_write.
1476 *
1477 * Unfortunately the NFS client can't create the in-core inode first
1478 * because it needs a file handle to create an in-core inode (see
1479 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1480 * symlink request has completed on the server.
1481 *
1482 * So instead we allocate a raw page, copy the symname into it, then do
1483 * the SYMLINK request with the page as the buffer. If it succeeds, we
1484 * now have a new file handle and can instantiate an in-core NFS inode
1485 * and move the raw page into its mapping.
1486 */
1487static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488{
Chuck Lever873101b2006-08-22 20:06:23 -04001489 struct pagevec lru_pvec;
1490 struct page *page;
1491 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001493 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 int error;
1495
1496 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1497 dir->i_ino, dentry->d_name.name, symname);
1498
Chuck Lever873101b2006-08-22 20:06:23 -04001499 if (pathlen > PAGE_SIZE)
1500 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501
Chuck Lever873101b2006-08-22 20:06:23 -04001502 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1503 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
1505 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001506
Jeff Layton83d93f22007-06-07 09:58:08 -04001507 page = alloc_page(GFP_HIGHUSER);
Chuck Lever873101b2006-08-22 20:06:23 -04001508 if (!page) {
1509 unlock_kernel();
1510 return -ENOMEM;
1511 }
1512
1513 kaddr = kmap_atomic(page, KM_USER0);
1514 memcpy(kaddr, symname, pathlen);
1515 if (pathlen < PAGE_SIZE)
1516 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1517 kunmap_atomic(kaddr, KM_USER0);
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 nfs_begin_data_update(dir);
Chuck Lever94a6d752006-08-22 20:06:23 -04001520 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 nfs_end_data_update(dir);
Chuck Lever873101b2006-08-22 20:06:23 -04001522 if (error != 0) {
1523 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1524 dir->i_sb->s_id, dir->i_ino,
1525 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001527 __free_page(page);
1528 unlock_kernel();
1529 return error;
1530 }
1531
1532 /*
1533 * No big deal if we can't add this page to the page cache here.
1534 * READLINK will get the missing page from the server if needed.
1535 */
1536 pagevec_init(&lru_pvec, 0);
1537 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1538 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001539 pagevec_add(&lru_pvec, page);
1540 pagevec_lru_add(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001541 SetPageUptodate(page);
1542 unlock_page(page);
1543 } else
1544 __free_page(page);
1545
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001547 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548}
1549
1550static int
1551nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1552{
1553 struct inode *inode = old_dentry->d_inode;
1554 int error;
1555
1556 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1557 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1558 dentry->d_parent->d_name.name, dentry->d_name.name);
1559
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 nfs_begin_data_update(dir);
1562 nfs_begin_data_update(inode);
1563 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001564 if (error == 0) {
1565 atomic_inc(&inode->i_count);
1566 d_instantiate(dentry, inode);
1567 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568 nfs_end_data_update(inode);
1569 nfs_end_data_update(dir);
1570 unlock_kernel();
1571 return error;
1572}
1573
1574/*
1575 * RENAME
1576 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1577 * different file handle for the same inode after a rename (e.g. when
1578 * moving to a different directory). A fail-safe method to do so would
1579 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1580 * rename the old file using the sillyrename stuff. This way, the original
1581 * file in old_dir will go away when the last process iput()s the inode.
1582 *
1583 * FIXED.
1584 *
1585 * It actually works quite well. One needs to have the possibility for
1586 * at least one ".nfs..." file in each directory the file ever gets
1587 * moved or linked to which happens automagically with the new
1588 * implementation that only depends on the dcache stuff instead of
1589 * using the inode layer
1590 *
1591 * Unfortunately, things are a little more complicated than indicated
1592 * above. For a cross-directory move, we want to make sure we can get
1593 * rid of the old inode after the operation. This means there must be
1594 * no pending writes (if it's a file), and the use count must be 1.
1595 * If these conditions are met, we can drop the dentries before doing
1596 * the rename.
1597 */
1598static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1599 struct inode *new_dir, struct dentry *new_dentry)
1600{
1601 struct inode *old_inode = old_dentry->d_inode;
1602 struct inode *new_inode = new_dentry->d_inode;
1603 struct dentry *dentry = NULL, *rehash = NULL;
1604 int error = -EBUSY;
1605
1606 /*
1607 * To prevent any new references to the target during the rename,
1608 * we unhash the dentry and free the inode in advance.
1609 */
1610 lock_kernel();
1611 if (!d_unhashed(new_dentry)) {
1612 d_drop(new_dentry);
1613 rehash = new_dentry;
1614 }
1615
1616 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1617 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1618 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1619 atomic_read(&new_dentry->d_count));
1620
1621 /*
1622 * First check whether the target is busy ... we can't
1623 * safely do _any_ rename if the target is in use.
1624 *
1625 * For files, make a copy of the dentry and then do a
1626 * silly-rename. If the silly-rename succeeds, the
1627 * copied dentry is hashed and becomes the new target.
1628 */
1629 if (!new_inode)
1630 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001631 if (S_ISDIR(new_inode->i_mode)) {
1632 error = -EISDIR;
1633 if (!S_ISDIR(old_inode->i_mode))
1634 goto out;
1635 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 int err;
1637 /* copy the target dentry's name */
1638 dentry = d_alloc(new_dentry->d_parent,
1639 &new_dentry->d_name);
1640 if (!dentry)
1641 goto out;
1642
1643 /* silly-rename the existing target ... */
1644 err = nfs_sillyrename(new_dir, new_dentry);
1645 if (!err) {
1646 new_dentry = rehash = dentry;
1647 new_inode = NULL;
1648 /* instantiate the replacement target */
1649 d_instantiate(new_dentry, NULL);
Jesper Juhl7a13e932007-04-26 00:29:02 -07001650 } else if (atomic_read(&new_dentry->d_count) > 1)
1651 /* dentry still busy? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 goto out;
Trond Myklebust20509f12005-08-25 16:25:34 -07001653 } else
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001654 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656go_ahead:
1657 /*
1658 * ... prune child dentries and writebacks if needed.
1659 */
1660 if (atomic_read(&old_dentry->d_count) > 1) {
Trond Myklebuste1552e12007-04-14 19:07:28 -04001661 if (S_ISREG(old_inode->i_mode))
1662 nfs_wb_all(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 shrink_dcache_parent(old_dentry);
1664 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001665 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666
Trond Myklebust24174112006-01-03 09:55:33 +01001667 if (new_inode != NULL) {
1668 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001670 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
1672 nfs_begin_data_update(old_dir);
1673 nfs_begin_data_update(new_dir);
1674 nfs_begin_data_update(old_inode);
1675 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1676 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001677 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 nfs_end_data_update(old_inode);
1679 nfs_end_data_update(new_dir);
1680 nfs_end_data_update(old_dir);
1681out:
1682 if (rehash)
1683 d_rehash(rehash);
1684 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07001685 d_move(old_dentry, new_dentry);
Chuck Lever8fb559f2007-09-24 15:40:16 -04001686 nfs_set_verifier(new_dentry,
1687 nfs_save_change_attribute(new_dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 }
1689
1690 /* new dentry created? */
1691 if (dentry)
1692 dput(dentry);
1693 unlock_kernel();
1694 return error;
1695}
1696
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001697static DEFINE_SPINLOCK(nfs_access_lru_lock);
1698static LIST_HEAD(nfs_access_lru_list);
1699static atomic_long_t nfs_access_nr_entries;
1700
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001701static void nfs_access_free_entry(struct nfs_access_entry *entry)
1702{
1703 put_rpccred(entry->cred);
1704 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001705 smp_mb__before_atomic_dec();
1706 atomic_long_dec(&nfs_access_nr_entries);
1707 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001708}
1709
Trond Myklebust979df722006-07-25 11:28:19 -04001710int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1711{
1712 LIST_HEAD(head);
1713 struct nfs_inode *nfsi;
1714 struct nfs_access_entry *cache;
1715
Trond Myklebust979df722006-07-25 11:28:19 -04001716restart:
Trond Myklebusta50f7952007-06-05 19:23:43 -04001717 spin_lock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001718 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1719 struct inode *inode;
1720
1721 if (nr_to_scan-- == 0)
1722 break;
1723 inode = igrab(&nfsi->vfs_inode);
1724 if (inode == NULL)
1725 continue;
1726 spin_lock(&inode->i_lock);
1727 if (list_empty(&nfsi->access_cache_entry_lru))
1728 goto remove_lru_entry;
1729 cache = list_entry(nfsi->access_cache_entry_lru.next,
1730 struct nfs_access_entry, lru);
1731 list_move(&cache->lru, &head);
1732 rb_erase(&cache->rb_node, &nfsi->access_cache);
1733 if (!list_empty(&nfsi->access_cache_entry_lru))
1734 list_move_tail(&nfsi->access_cache_inode_lru,
1735 &nfs_access_lru_list);
1736 else {
1737remove_lru_entry:
1738 list_del_init(&nfsi->access_cache_inode_lru);
1739 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1740 }
1741 spin_unlock(&inode->i_lock);
Trond Myklebusta50f7952007-06-05 19:23:43 -04001742 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001743 iput(inode);
1744 goto restart;
1745 }
1746 spin_unlock(&nfs_access_lru_lock);
1747 while (!list_empty(&head)) {
1748 cache = list_entry(head.next, struct nfs_access_entry, lru);
1749 list_del(&cache->lru);
1750 nfs_access_free_entry(cache);
1751 }
1752 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1753}
1754
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001755static void __nfs_access_zap_cache(struct inode *inode)
1756{
1757 struct nfs_inode *nfsi = NFS_I(inode);
1758 struct rb_root *root_node = &nfsi->access_cache;
1759 struct rb_node *n, *dispose = NULL;
1760 struct nfs_access_entry *entry;
1761
1762 /* Unhook entries from the cache */
1763 while ((n = rb_first(root_node)) != NULL) {
1764 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1765 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001766 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001767 n->rb_left = dispose;
1768 dispose = n;
1769 }
1770 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1771 spin_unlock(&inode->i_lock);
1772
1773 /* Now kill them all! */
1774 while (dispose != NULL) {
1775 n = dispose;
1776 dispose = n->rb_left;
1777 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1778 }
1779}
1780
1781void nfs_access_zap_cache(struct inode *inode)
1782{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001783 /* Remove from global LRU init */
1784 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1785 spin_lock(&nfs_access_lru_lock);
1786 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1787 spin_unlock(&nfs_access_lru_lock);
1788 }
1789
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001790 spin_lock(&inode->i_lock);
1791 /* This will release the spinlock */
1792 __nfs_access_zap_cache(inode);
1793}
1794
1795static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1796{
1797 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1798 struct nfs_access_entry *entry;
1799
1800 while (n != NULL) {
1801 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1802
1803 if (cred < entry->cred)
1804 n = n->rb_left;
1805 else if (cred > entry->cred)
1806 n = n->rb_right;
1807 else
1808 return entry;
1809 }
1810 return NULL;
1811}
1812
Trond Myklebustaf22f942007-08-10 17:45:10 -04001813static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814{
Chuck Lever55296802005-08-18 11:24:09 -07001815 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001816 struct nfs_access_entry *cache;
1817 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001819 spin_lock(&inode->i_lock);
1820 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1821 goto out_zap;
1822 cache = nfs_access_search_rbtree(inode, cred);
1823 if (cache == NULL)
1824 goto out;
Fabio Olive Leitec7e15962007-07-26 22:59:00 -03001825 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001826 goto out_stale;
1827 res->jiffies = cache->jiffies;
1828 res->cred = cache->cred;
1829 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001830 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001831 err = 0;
1832out:
1833 spin_unlock(&inode->i_lock);
1834 return err;
1835out_stale:
1836 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001837 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001838 spin_unlock(&inode->i_lock);
1839 nfs_access_free_entry(cache);
1840 return -ENOENT;
1841out_zap:
1842 /* This will release the spinlock */
1843 __nfs_access_zap_cache(inode);
1844 return -ENOENT;
1845}
1846
1847static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1848{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001849 struct nfs_inode *nfsi = NFS_I(inode);
1850 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001851 struct rb_node **p = &root_node->rb_node;
1852 struct rb_node *parent = NULL;
1853 struct nfs_access_entry *entry;
1854
1855 spin_lock(&inode->i_lock);
1856 while (*p != NULL) {
1857 parent = *p;
1858 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1859
1860 if (set->cred < entry->cred)
1861 p = &parent->rb_left;
1862 else if (set->cred > entry->cred)
1863 p = &parent->rb_right;
1864 else
1865 goto found;
1866 }
1867 rb_link_node(&set->rb_node, parent, p);
1868 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001869 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001870 spin_unlock(&inode->i_lock);
1871 return;
1872found:
1873 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001874 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1875 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001876 spin_unlock(&inode->i_lock);
1877 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878}
1879
Trond Myklebustaf22f942007-08-10 17:45:10 -04001880static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001882 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1883 if (cache == NULL)
1884 return;
1885 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001887 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001889
1890 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001891
1892 /* Update accounting */
1893 smp_mb__before_atomic_inc();
1894 atomic_long_inc(&nfs_access_nr_entries);
1895 smp_mb__after_atomic_inc();
1896
1897 /* Add inode to global LRU list */
1898 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1899 spin_lock(&nfs_access_lru_lock);
1900 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1901 spin_unlock(&nfs_access_lru_lock);
1902 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903}
1904
1905static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1906{
1907 struct nfs_access_entry cache;
1908 int status;
1909
1910 status = nfs_access_get_cached(inode, cred, &cache);
1911 if (status == 0)
1912 goto out;
1913
1914 /* Be clever: ask server to check for all possible rights */
1915 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1916 cache.cred = cred;
1917 cache.jiffies = jiffies;
1918 status = NFS_PROTO(inode)->access(inode, &cache);
1919 if (status != 0)
1920 return status;
1921 nfs_access_add_cache(inode, &cache);
1922out:
1923 if ((cache.mask & mask) == mask)
1924 return 0;
1925 return -EACCES;
1926}
1927
Trond Myklebustaf22f942007-08-10 17:45:10 -04001928static int nfs_open_permission_mask(int openflags)
1929{
1930 int mask = 0;
1931
1932 if (openflags & FMODE_READ)
1933 mask |= MAY_READ;
1934 if (openflags & FMODE_WRITE)
1935 mask |= MAY_WRITE;
1936 if (openflags & FMODE_EXEC)
1937 mask |= MAY_EXEC;
1938 return mask;
1939}
1940
1941int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1942{
1943 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1944}
1945
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1947{
1948 struct rpc_cred *cred;
1949 int res = 0;
1950
Chuck Lever91d5b472006-03-20 13:44:14 -05001951 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 if (mask == 0)
1954 goto out;
1955 /* Is this sys_access() ? */
1956 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1957 goto force_lookup;
1958
1959 switch (inode->i_mode & S_IFMT) {
1960 case S_IFLNK:
1961 goto out;
1962 case S_IFREG:
1963 /* NFSv4 has atomic_open... */
1964 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1965 && nd != NULL
1966 && (nd->flags & LOOKUP_OPEN))
1967 goto out;
1968 break;
1969 case S_IFDIR:
1970 /*
1971 * Optimize away all write operations, since the server
1972 * will check permissions when we perform the op.
1973 */
1974 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1975 goto out;
1976 }
1977
1978force_lookup:
1979 lock_kernel();
1980
1981 if (!NFS_PROTO(inode)->access)
1982 goto out_notsup;
1983
1984 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1985 if (!IS_ERR(cred)) {
1986 res = nfs_do_access(inode, cred, mask);
1987 put_rpccred(cred);
1988 } else
1989 res = PTR_ERR(cred);
1990 unlock_kernel();
1991out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001992 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1993 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 return res;
1995out_notsup:
1996 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1997 if (res == 0)
1998 res = generic_permission(inode, mask, NULL);
1999 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05002000 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001}
2002
2003/*
2004 * Local variables:
2005 * version-control: t
2006 * kept-new-versions: 5
2007 * End:
2008 */