blob: e4a04d16b8b06fd4faf22b7ba238f404403a8c7f [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);
Chuck Leverdc592502005-08-18 11:24:12 -0700203 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700205 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800208 * through inode->i_mutex or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 */
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700210 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
213 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 unlock_page(page);
215 return 0;
216 error:
217 SetPageError(page);
218 unlock_page(page);
219 nfs_zap_caches(inode);
220 desc->error = error;
221 return -EIO;
222}
223
224static inline
225int dir_decode(nfs_readdir_descriptor_t *desc)
226{
Al Viro0dbb4c62006-10-19 23:28:49 -0700227 __be32 *p = desc->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 p = desc->decode(p, desc->entry, desc->plus);
229 if (IS_ERR(p))
230 return PTR_ERR(p);
231 desc->ptr = p;
Neil Brown1f4eab72007-04-16 09:35:27 +1000232 if (desc->timestamp_valid)
233 desc->entry->fattr->time_start = desc->timestamp;
234 else
235 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 return 0;
237}
238
239static inline
240void dir_page_release(nfs_readdir_descriptor_t *desc)
241{
242 kunmap(desc->page);
243 page_cache_release(desc->page);
244 desc->page = NULL;
245 desc->ptr = NULL;
246}
247
248/*
249 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000250 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 *
252 * If the end of the buffer has been reached, return -EAGAIN, if not,
253 * return the offset within the buffer of the next entry to be
254 * read.
255 */
256static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000257int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258{
259 struct nfs_entry *entry = desc->entry;
260 int loop_count = 0,
261 status;
262
263 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500264 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
265 __FUNCTION__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000266 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 break;
268 if (loop_count++ > 200) {
269 loop_count = 0;
270 schedule();
271 }
272 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 return status;
274}
275
276/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000277 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000278 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000279 *
280 * If the end of the buffer has been reached, return -EAGAIN, if not,
281 * return the offset within the buffer of the next entry to be
282 * read.
283 */
284static inline
285int find_dirent_index(nfs_readdir_descriptor_t *desc)
286{
287 struct nfs_entry *entry = desc->entry;
288 int loop_count = 0,
289 status;
290
291 for(;;) {
292 status = dir_decode(desc);
293 if (status)
294 break;
295
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500296 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
297 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000298
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000299 if (desc->file->f_pos == desc->current_index) {
300 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000301 break;
302 }
303 desc->current_index++;
304 if (loop_count++ > 200) {
305 loop_count = 0;
306 schedule();
307 }
308 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000309 return status;
310}
311
312/*
313 * Find the given page, and call find_dirent() or find_dirent_index in
314 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 */
316static inline
317int find_dirent_page(nfs_readdir_descriptor_t *desc)
318{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800319 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 struct page *page;
321 int status;
322
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500323 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
324 __FUNCTION__, desc->page_index,
325 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326
Neil Brown1f4eab72007-04-16 09:35:27 +1000327 /* If we find the page in the page_cache, we cannot be sure
328 * how fresh the data is, so we will ignore readdir_plus attributes.
329 */
330 desc->timestamp_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 page = read_cache_page(inode->i_mapping, desc->page_index,
332 (filler_t *)nfs_readdir_filler, desc);
333 if (IS_ERR(page)) {
334 status = PTR_ERR(page);
335 goto out;
336 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 /* NOTE: Someone else may have changed the READDIRPLUS flag */
339 desc->page = page;
340 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000341 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000342 status = find_dirent(desc);
343 else
344 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 if (status < 0)
346 dir_page_release(desc);
347 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500348 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350}
351
352/*
353 * Recurse through the page cache pages, and return a
354 * filled nfs_entry structure of the next directory entry if possible.
355 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000356 * The target for the search is '*desc->dir_cookie' if non-0,
357 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 */
359static inline
360int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
361{
362 int loop_count = 0;
363 int res;
364
Olivier Galibert00a92642005-06-22 17:16:29 +0000365 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000366 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
368 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000369 desc->page_index = 0;
370 desc->entry->cookie = desc->entry->prev_cookie = 0;
371 desc->entry->eof = 0;
372 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000373 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500374 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
375 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000376
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 for (;;) {
378 res = find_dirent_page(desc);
379 if (res != -EAGAIN)
380 break;
381 /* Align to beginning of next page */
382 desc->page_index ++;
383 if (loop_count++ > 200) {
384 loop_count = 0;
385 schedule();
386 }
387 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500388
389 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 return res;
391}
392
393static inline unsigned int dt_type(struct inode *inode)
394{
395 return (inode->i_mode >> 12) & 15;
396}
397
398static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
399
400/*
401 * Once we've found the start of the dirent within a page: fill 'er up...
402 */
403static
404int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
405 filldir_t filldir)
406{
407 struct file *file = desc->file;
408 struct nfs_entry *entry = desc->entry;
409 struct dentry *dentry = NULL;
410 unsigned long fileid;
411 int loop_count = 0,
412 res;
413
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500414 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
415 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 for(;;) {
418 unsigned d_type = DT_UNKNOWN;
419 /* Note: entry->prev_cookie contains the cookie for
420 * retrieving the current dirent on the server */
421 fileid = nfs_fileid_to_ino_t(entry->ino);
422
423 /* Get a dentry if we have one */
424 if (dentry != NULL)
425 dput(dentry);
426 dentry = nfs_readdir_lookup(desc);
427
428 /* Use readdirplus info */
429 if (dentry != NULL && dentry->d_inode != NULL) {
430 d_type = dt_type(dentry->d_inode);
431 fileid = dentry->d_inode->i_ino;
432 }
433
434 res = filldir(dirent, entry->name, entry->len,
Olivier Galibert00a92642005-06-22 17:16:29 +0000435 file->f_pos, fileid, d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 if (res < 0)
437 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000438 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000439 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 if (dir_decode(desc) != 0) {
441 desc->page_index ++;
442 break;
443 }
444 if (loop_count++ > 200) {
445 loop_count = 0;
446 schedule();
447 }
448 }
449 dir_page_release(desc);
450 if (dentry != NULL)
451 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500452 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
453 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 return res;
455}
456
457/*
458 * If we cannot find a cookie in our cache, we suspect that this is
459 * because it points to a deleted file, so we ask the server to return
460 * whatever it thinks is the next entry. We then feed this to filldir.
461 * If all goes well, we should then be able to find our way round the
462 * cache on the next call to readdir_search_pagecache();
463 *
464 * NOTE: we cannot add the anonymous page to the pagecache because
465 * the data it contains might not be page aligned. Besides,
466 * we should already have a complete representation of the
467 * directory in the page cache by the time we get here.
468 */
469static inline
470int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
471 filldir_t filldir)
472{
473 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800474 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 struct rpc_cred *cred = nfs_file_cred(file);
476 struct page *page = NULL;
477 int status;
Neil Brown1f4eab72007-04-16 09:35:27 +1000478 unsigned long timestamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500480 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
481 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482
483 page = alloc_page(GFP_HIGHUSER);
484 if (!page) {
485 status = -ENOMEM;
486 goto out;
487 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000488 timestamp = jiffies;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800489 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 page,
491 NFS_SERVER(inode)->dtsize,
492 desc->plus);
Chuck Leverdc592502005-08-18 11:24:12 -0700493 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700494 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700495 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 desc->page = page;
497 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
498 if (desc->error >= 0) {
Neil Brown1f4eab72007-04-16 09:35:27 +1000499 desc->timestamp = timestamp;
500 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000502 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 } else
504 status = -EIO;
505 if (status < 0)
506 goto out_release;
507
508 status = nfs_do_filldir(desc, dirent, filldir);
509
510 /* Reset read descriptor so it searches the page cache from
511 * the start upon the next call to readdir_search_pagecache() */
512 desc->page_index = 0;
513 desc->entry->cookie = desc->entry->prev_cookie = 0;
514 desc->entry->eof = 0;
515 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500516 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
517 __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 return status;
519 out_release:
520 dir_page_release(desc);
521 goto out;
522}
523
Olivier Galibert00a92642005-06-22 17:16:29 +0000524/* The file offset position represents the dirent entry number. A
525 last cookie cache takes care of the common case of reading the
526 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527 */
528static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
529{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800530 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531 struct inode *inode = dentry->d_inode;
532 nfs_readdir_descriptor_t my_desc,
533 *desc = &my_desc;
534 struct nfs_entry my_entry;
535 struct nfs_fh fh;
536 struct nfs_fattr fattr;
537 long res;
538
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500539 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
540 dentry->d_parent->d_name.name, dentry->d_name.name,
541 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500542 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 lock_kernel();
545
Trond Myklebust717d44e2007-01-24 11:54:55 -0800546 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 if (res < 0) {
548 unlock_kernel();
549 return res;
550 }
551
552 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000553 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000554 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000555 * to either find the entry with the appropriate number or
556 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557 */
558 memset(desc, 0, sizeof(*desc));
559
560 desc->file = filp;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000561 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 desc->decode = NFS_PROTO(inode)->decode_dirent;
563 desc->plus = NFS_USE_READDIRPLUS(inode);
564
565 my_entry.cookie = my_entry.prev_cookie = 0;
566 my_entry.eof = 0;
567 my_entry.fh = &fh;
568 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400569 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 desc->entry = &my_entry;
571
572 while(!desc->entry->eof) {
573 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (res == -EBADCOOKIE) {
576 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000577 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 /* Or that the server has 'lost' a cookie */
579 res = uncached_readdir(desc, dirent, filldir);
580 if (res >= 0)
581 continue;
582 }
583 res = 0;
584 break;
585 }
586 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700587 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 nfs_zap_caches(inode);
589 desc->plus = 0;
590 desc->entry->eof = 0;
591 continue;
592 }
593 if (res < 0)
594 break;
595
596 res = nfs_do_filldir(desc, dirent, filldir);
597 if (res < 0) {
598 res = 0;
599 break;
600 }
601 }
602 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500603 if (res > 0)
604 res = 0;
605 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
606 dentry->d_parent->d_name.name, dentry->d_name.name,
607 res);
608 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
Trond Myklebust10afec92007-05-14 17:16:04 -0400611static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000612{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800613 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000614 switch (origin) {
615 case 1:
616 offset += filp->f_pos;
617 case 0:
618 if (offset >= 0)
619 break;
620 default:
621 offset = -EINVAL;
622 goto out;
623 }
624 if (offset != filp->f_pos) {
625 filp->f_pos = offset;
626 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
627 }
628out:
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800629 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000630 return offset;
631}
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633/*
634 * All directory operations under NFS are synchronous, so fsync()
635 * is a dummy operation.
636 */
Trond Myklebust10afec92007-05-14 17:16:04 -0400637static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500639 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
640 dentry->d_parent->d_name.name, dentry->d_name.name,
641 datasync);
642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 return 0;
644}
645
646/*
647 * A check for whether or not the parent directory has changed.
648 * In the case it has, we assume that the dentries are untrustworthy
649 * and may need to be looked up again.
650 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500651static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
Trond Myklebuste62c2bb2007-05-09 09:00:17 -0400653 unsigned long verf;
654
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 if (IS_ROOT(dentry))
656 return 1;
Trond Myklebust3062c532007-07-14 17:36:45 -0400657 verf = dentry->d_time;
Trond Myklebuste70c4902007-05-09 09:00:18 -0400658 if (nfs_caches_unstable(dir)
Trond Myklebuste62c2bb2007-05-09 09:00:17 -0400659 || verf != NFS_I(dir)->cache_change_attribute)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 return 0;
Trond Myklebuste62c2bb2007-05-09 09:00:17 -0400661 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662}
663
664static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
665{
Trond Myklebust3062c532007-07-14 17:36:45 -0400666 dentry->d_time = verf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667}
668
Trond Myklebustc79ba782007-01-31 08:16:24 -0500669static void nfs_refresh_verifier(struct dentry * dentry, unsigned long verf)
670{
Trond Myklebuste62c2bb2007-05-09 09:00:17 -0400671 nfs_set_verifier(dentry, verf);
Trond Myklebustc79ba782007-01-31 08:16:24 -0500672}
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674/*
675 * Whenever an NFS operation succeeds, we know that the dentry
676 * is valid, so we update the revalidation timestamp.
677 */
678static inline void nfs_renew_times(struct dentry * dentry)
679{
680 dentry->d_time = jiffies;
681}
682
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400683/*
684 * Return the intent data that applies to this particular path component
685 *
686 * Note that the current set of intents only apply to the very last
687 * component of the path.
688 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
689 */
690static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
691{
692 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
693 return 0;
694 return nd->flags & mask;
695}
696
697/*
698 * Inode and filehandle revalidation for lookups.
699 *
700 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
701 * or if the intent information indicates that we're about to open this
702 * particular file and the "nocto" mount flag is not set.
703 *
704 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705static inline
706int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
707{
708 struct nfs_server *server = NFS_SERVER(inode);
709
710 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400712 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 goto out_force;
714 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400715 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400716 !(server->flags & NFS_MOUNT_NOCTO) &&
717 (S_ISREG(inode->i_mode) ||
718 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 goto out_force;
720 }
721 return nfs_revalidate_inode(server, inode);
722out_force:
723 return __nfs_revalidate_inode(server, inode);
724}
725
726/*
727 * We judge how long we want to trust negative
728 * dentries by looking at the parent inode mtime.
729 *
730 * If parent mtime has changed, we revalidate, else we wait for a
731 * period corresponding to the parent's attribute cache timeout value.
732 */
733static inline
734int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
735 struct nameidata *nd)
736{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400738 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 return 0;
740 return !nfs_check_verifier(dir, dentry);
741}
742
743/*
744 * This is called every time the dcache has a lookup hit,
745 * and we should check whether we can really trust that
746 * lookup.
747 *
748 * NOTE! The hit can be a negative hit too, don't assume
749 * we have an inode!
750 *
751 * If the parent directory is seen to have changed, we throw out the
752 * cached dentry and do a new lookup.
753 */
754static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
755{
756 struct inode *dir;
757 struct inode *inode;
758 struct dentry *parent;
759 int error;
760 struct nfs_fh fhandle;
761 struct nfs_fattr fattr;
762 unsigned long verifier;
763
764 parent = dget_parent(dentry);
765 lock_kernel();
766 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500767 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 inode = dentry->d_inode;
769
Trond Myklebuste70c4902007-05-09 09:00:18 -0400770 /* Revalidate parent directory attribute cache */
771 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
772 goto out_zap_parent;
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 if (!inode) {
775 if (nfs_neg_need_reval(dir, dentry, nd))
776 goto out_bad;
777 goto out_valid;
778 }
779
780 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500781 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
782 __FUNCTION__, dentry->d_parent->d_name.name,
783 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 goto out_bad;
785 }
786
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 /* Force a full look up iff the parent directory has changed */
788 if (nfs_check_verifier(dir, dentry)) {
789 if (nfs_lookup_verify_inode(inode, nd))
790 goto out_zap_parent;
791 goto out_valid;
792 }
793
794 if (NFS_STALE(inode))
795 goto out_bad;
796
797 verifier = nfs_save_change_attribute(dir);
798 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
799 if (error)
800 goto out_bad;
801 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
802 goto out_bad;
803 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
804 goto out_bad;
805
806 nfs_renew_times(dentry);
Trond Myklebustc79ba782007-01-31 08:16:24 -0500807 nfs_refresh_verifier(dentry, verifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 out_valid:
809 unlock_kernel();
810 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
812 __FUNCTION__, dentry->d_parent->d_name.name,
813 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return 1;
815out_zap_parent:
816 nfs_zap_caches(dir);
817 out_bad:
818 NFS_CACHEINV(dir);
819 if (inode && S_ISDIR(inode->i_mode)) {
820 /* Purge readdir caches. */
821 nfs_zap_caches(inode);
822 /* If we have submounts, don't unhash ! */
823 if (have_submounts(dentry))
824 goto out_valid;
825 shrink_dcache_parent(dentry);
826 }
827 d_drop(dentry);
828 unlock_kernel();
829 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500830 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
831 __FUNCTION__, dentry->d_parent->d_name.name,
832 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return 0;
834}
835
836/*
837 * This is called from dput() when d_count is going to 0.
838 */
839static int nfs_dentry_delete(struct dentry *dentry)
840{
841 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
842 dentry->d_parent->d_name.name, dentry->d_name.name,
843 dentry->d_flags);
844
845 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
846 /* Unhash it, so that ->d_iput() would be called */
847 return 1;
848 }
849 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
850 /* Unhash it, so that ancestors of killed async unlink
851 * files will be cleaned up during umount */
852 return 1;
853 }
854 return 0;
855
856}
857
858/*
859 * Called when the dentry loses inode.
860 * We use it to clean up silly-renamed files.
861 */
862static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
863{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700864 nfs_inode_return_delegation(inode);
Neil Brown83672d32007-02-26 12:48:25 +1100865 if (S_ISDIR(inode->i_mode))
866 /* drop any readdir cache as it could easily be old */
867 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
868
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
870 lock_kernel();
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700871 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -0400872 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 unlock_kernel();
874 }
875 /* When creating a negative dentry, we want to renew d_time */
876 nfs_renew_times(dentry);
877 iput(inode);
878}
879
880struct dentry_operations nfs_dentry_operations = {
881 .d_revalidate = nfs_lookup_revalidate,
882 .d_delete = nfs_dentry_delete,
883 .d_iput = nfs_dentry_iput,
884};
885
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400886/*
887 * Use intent information to check whether or not we're going to do
888 * an O_EXCL create using this path component.
889 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890static inline
891int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
892{
893 if (NFS_PROTO(dir)->version == 2)
894 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400895 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 return 0;
897 return (nd->intent.open.flags & O_EXCL) != 0;
898}
899
Trond Myklebusta0356862007-06-05 13:26:15 -0400900static inline int nfs_reval_fsid(struct inode *dir, const struct nfs_fattr *fattr)
Trond Myklebust55a97592006-06-09 09:34:19 -0400901{
902 struct nfs_server *server = NFS_SERVER(dir);
903
904 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
Trond Myklebusta0356862007-06-05 13:26:15 -0400905 /* Revalidate fsid using the parent directory */
906 return __nfs_revalidate_inode(server, dir);
Trond Myklebust55a97592006-06-09 09:34:19 -0400907 return 0;
908}
909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
911{
912 struct dentry *res;
913 struct inode *inode = NULL;
914 int error;
915 struct nfs_fh fhandle;
916 struct nfs_fattr fattr;
917
918 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
919 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500920 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
922 res = ERR_PTR(-ENAMETOOLONG);
923 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
924 goto out;
925
926 res = ERR_PTR(-ENOMEM);
927 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
928
929 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
Trond Myklebustfd684072006-09-05 12:27:44 -0400931 /*
932 * If we're doing an exclusive create, optimize away the lookup
933 * but don't hash the dentry.
934 */
935 if (nfs_is_exclusive_create(dir, nd)) {
936 d_instantiate(dentry, NULL);
937 res = NULL;
938 goto out_unlock;
939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
941 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
942 if (error == -ENOENT)
943 goto no_entry;
944 if (error < 0) {
945 res = ERR_PTR(error);
946 goto out_unlock;
947 }
Trond Myklebusta0356862007-06-05 13:26:15 -0400948 error = nfs_reval_fsid(dir, &fattr);
Trond Myklebust55a97592006-06-09 09:34:19 -0400949 if (error < 0) {
950 res = ERR_PTR(error);
951 goto out_unlock;
952 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500954 res = (struct dentry *)inode;
955 if (IS_ERR(res))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 goto out_unlock;
David Howells54ceac42006-08-22 20:06:13 -0400957
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400959 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -0700960 if (res != NULL) {
Trond Myklebustfc226172006-10-21 10:24:24 -0700961 struct dentry *parent;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700962 if (IS_ERR(res))
963 goto out_unlock;
Trond Myklebustfc226172006-10-21 10:24:24 -0700964 /* Was a directory renamed! */
965 parent = dget_parent(res);
966 if (!IS_ROOT(parent))
967 nfs_mark_for_revalidate(parent->d_inode);
968 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700970 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 nfs_renew_times(dentry);
972 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
973out_unlock:
974 unlock_kernel();
975out:
976 return res;
977}
978
979#ifdef CONFIG_NFS_V4
980static int nfs_open_revalidate(struct dentry *, struct nameidata *);
981
982struct dentry_operations nfs4_dentry_operations = {
983 .d_revalidate = nfs_open_revalidate,
984 .d_delete = nfs_dentry_delete,
985 .d_iput = nfs_dentry_iput,
986};
987
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400988/*
989 * Use intent information to determine whether we need to substitute
990 * the NFSv4-style stateful OPEN for the LOOKUP call
991 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992static int is_atomic_open(struct inode *dir, struct nameidata *nd)
993{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400994 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 return 0;
996 /* NFS does not (yet) have a stateful open for directories */
997 if (nd->flags & LOOKUP_DIRECTORY)
998 return 0;
999 /* Are we trying to write to a read only partition? */
1000 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1001 return 0;
1002 return 1;
1003}
1004
1005static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1006{
1007 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 int error;
1009
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001010 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1011 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 /* Check that we are indeed trying to open this file */
1014 if (!is_atomic_open(dir, nd))
1015 goto no_open;
1016
1017 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1018 res = ERR_PTR(-ENAMETOOLONG);
1019 goto out;
1020 }
1021 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1022
1023 /* Let vfs_create() deal with O_EXCL */
Trond Myklebust02a913a2005-10-18 14:20:17 -07001024 if (nd->intent.open.flags & O_EXCL) {
1025 d_add(dentry, NULL);
1026 goto out;
1027 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
1029 /* Open the file on the server */
1030 lock_kernel();
1031 /* Revalidate parent directory attribute cache */
1032 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1033 if (error < 0) {
1034 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -07001035 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 goto out;
1037 }
1038
1039 if (nd->intent.open.flags & O_CREAT) {
1040 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001041 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 nfs_end_data_update(dir);
1043 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -07001044 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001046 if (IS_ERR(res)) {
1047 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 switch (error) {
1049 /* Make a negative dentry */
1050 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001051 res = NULL;
1052 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001054 case -EISDIR:
1055 case -ENOTDIR:
1056 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 case -ELOOP:
1058 if (!(nd->intent.open.flags & O_NOFOLLOW))
1059 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 /* case -EINVAL: */
1061 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 goto out;
1063 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001064 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 dentry = res;
1066 nfs_renew_times(dentry);
1067 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1068out:
1069 return res;
1070no_open:
1071 return nfs_lookup(dir, dentry, nd);
1072}
1073
1074static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1075{
1076 struct dentry *parent = NULL;
1077 struct inode *inode = dentry->d_inode;
1078 struct inode *dir;
1079 unsigned long verifier;
1080 int openflags, ret = 0;
1081
1082 parent = dget_parent(dentry);
1083 dir = parent->d_inode;
1084 if (!is_atomic_open(dir, nd))
1085 goto no_open;
1086 /* We can't create new files in nfs_open_revalidate(), so we
1087 * optimize away revalidation of negative dentries.
1088 */
1089 if (inode == NULL)
1090 goto out;
1091 /* NFS only supports OPEN on regular files */
1092 if (!S_ISREG(inode->i_mode))
1093 goto no_open;
1094 openflags = nd->intent.open.flags;
1095 /* We cannot do exclusive creation on a positive dentry */
1096 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1097 goto no_open;
1098 /* We can't create new files, or truncate existing ones here */
1099 openflags &= ~(O_CREAT|O_TRUNC);
1100
1101 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001102 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 * operations that change the directory. We therefore save the
1104 * change attribute *before* we do the RPC call.
1105 */
1106 lock_kernel();
1107 verifier = nfs_save_change_attribute(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001108 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 if (!ret)
Trond Myklebustc79ba782007-01-31 08:16:24 -05001110 nfs_refresh_verifier(dentry, verifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 unlock_kernel();
1112out:
1113 dput(parent);
1114 if (!ret)
1115 d_drop(dentry);
1116 return ret;
1117no_open:
1118 dput(parent);
1119 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1120 return 1;
1121 return nfs_lookup_revalidate(dentry, nd);
1122}
1123#endif /* CONFIG_NFSV4 */
1124
1125static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1126{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -08001127 struct dentry *parent = desc->file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 struct inode *dir = parent->d_inode;
1129 struct nfs_entry *entry = desc->entry;
1130 struct dentry *dentry, *alias;
1131 struct qstr name = {
1132 .name = entry->name,
1133 .len = entry->len,
1134 };
1135 struct inode *inode;
1136
1137 switch (name.len) {
1138 case 2:
1139 if (name.name[0] == '.' && name.name[1] == '.')
1140 return dget_parent(parent);
1141 break;
1142 case 1:
1143 if (name.name[0] == '.')
1144 return dget(parent);
1145 }
1146 name.hash = full_name_hash(name.name, name.len);
1147 dentry = d_lookup(parent, &name);
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001148 if (dentry != NULL) {
Trond Myklebustef75c792007-01-16 10:09:44 -05001149 /* Is this a positive dentry that matches the readdir info? */
1150 if (dentry->d_inode != NULL &&
1151 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1152 d_mountpoint(dentry))) {
1153 if (!desc->plus || entry->fh->size == 0)
1154 return dentry;
1155 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1156 entry->fh) == 0)
1157 goto out_renew;
1158 }
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001159 /* No, so d_drop to allow one to be created */
1160 d_drop(dentry);
1161 dput(dentry);
1162 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1164 return NULL;
Trond Myklebust54af3bb2007-09-28 12:27:41 -04001165 if (name.len > NFS_SERVER(dir)->namelen)
1166 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001167 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 dentry = d_alloc(parent, &name);
1169 if (dentry == NULL)
1170 return NULL;
1171 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1172 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001173 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 dput(dentry);
1175 return NULL;
1176 }
David Howells54ceac42006-08-22 20:06:13 -04001177
1178 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (alias != NULL) {
1180 dput(dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001181 if (IS_ERR(alias))
1182 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 dentry = alias;
1184 }
David Howells54ceac42006-08-22 20:06:13 -04001185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 nfs_renew_times(dentry);
1187 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1188 return dentry;
Trond Myklebustc79ba782007-01-31 08:16:24 -05001189out_renew:
1190 nfs_renew_times(dentry);
1191 nfs_refresh_verifier(dentry, nfs_save_change_attribute(dir));
1192 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193}
1194
1195/*
1196 * Code common to create, mkdir, and mknod.
1197 */
1198int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1199 struct nfs_fattr *fattr)
1200{
1201 struct inode *inode;
1202 int error = -EACCES;
1203
1204 /* We may have been initialized further down */
1205 if (dentry->d_inode)
1206 return 0;
1207 if (fhandle->size == 0) {
1208 struct inode *dir = dentry->d_parent->d_inode;
1209 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1210 if (error)
Chuck Lever4f390c12006-08-22 20:06:22 -04001211 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212 }
1213 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1214 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001215 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 if (error < 0)
Chuck Lever4f390c12006-08-22 20:06:22 -04001217 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001220 error = PTR_ERR(inode);
1221 if (IS_ERR(inode))
Chuck Lever4f390c12006-08-22 20:06:22 -04001222 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 d_instantiate(dentry, inode);
Trond Myklebustfd684072006-09-05 12:27:44 -04001224 if (d_unhashed(dentry))
1225 d_rehash(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227}
1228
1229/*
1230 * Following a failed create operation, we drop the dentry rather
1231 * than retain a negative dentry. This avoids a problem in the event
1232 * that the operation succeeded on the server, but an error in the
1233 * reply path made it appear to have failed.
1234 */
1235static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1236 struct nameidata *nd)
1237{
1238 struct iattr attr;
1239 int error;
1240 int open_flags = 0;
1241
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001242 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1243 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
1245 attr.ia_mode = mode;
1246 attr.ia_valid = ATTR_MODE;
1247
Trond Myklebustad389da2007-06-05 12:30:00 -04001248 if ((nd->flags & LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 open_flags = nd->intent.open.flags;
1250
1251 lock_kernel();
1252 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001253 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 nfs_end_data_update(dir);
1255 if (error != 0)
1256 goto out_err;
1257 nfs_renew_times(dentry);
1258 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1259 unlock_kernel();
1260 return 0;
1261out_err:
1262 unlock_kernel();
1263 d_drop(dentry);
1264 return error;
1265}
1266
1267/*
1268 * See comments for nfs_proc_create regarding failed operations.
1269 */
1270static int
1271nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1272{
1273 struct iattr attr;
1274 int status;
1275
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001276 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1277 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 if (!new_valid_dev(rdev))
1280 return -EINVAL;
1281
1282 attr.ia_mode = mode;
1283 attr.ia_valid = ATTR_MODE;
1284
1285 lock_kernel();
1286 nfs_begin_data_update(dir);
1287 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1288 nfs_end_data_update(dir);
1289 if (status != 0)
1290 goto out_err;
1291 nfs_renew_times(dentry);
1292 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1293 unlock_kernel();
1294 return 0;
1295out_err:
1296 unlock_kernel();
1297 d_drop(dentry);
1298 return status;
1299}
1300
1301/*
1302 * See comments for nfs_proc_create regarding failed operations.
1303 */
1304static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1305{
1306 struct iattr attr;
1307 int error;
1308
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001309 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1310 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 attr.ia_valid = ATTR_MODE;
1313 attr.ia_mode = mode | S_IFDIR;
1314
1315 lock_kernel();
1316 nfs_begin_data_update(dir);
1317 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1318 nfs_end_data_update(dir);
1319 if (error != 0)
1320 goto out_err;
1321 nfs_renew_times(dentry);
1322 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1323 unlock_kernel();
1324 return 0;
1325out_err:
1326 d_drop(dentry);
1327 unlock_kernel();
1328 return error;
1329}
1330
1331static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1332{
1333 int error;
1334
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001335 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1336 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338 lock_kernel();
1339 nfs_begin_data_update(dir);
1340 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1341 /* Ensure the VFS deletes this inode */
1342 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001343 clear_nlink(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 nfs_end_data_update(dir);
1345 unlock_kernel();
1346
1347 return error;
1348}
1349
1350static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1351{
1352 static unsigned int sillycounter;
1353 const int i_inosize = sizeof(dir->i_ino)*2;
1354 const int countersize = sizeof(sillycounter)*2;
1355 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1356 char silly[slen+1];
1357 struct qstr qsilly;
1358 struct dentry *sdentry;
1359 int error = -EIO;
1360
1361 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1362 dentry->d_parent->d_name.name, dentry->d_name.name,
1363 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001364 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 /*
1367 * We don't allow a dentry to be silly-renamed twice.
1368 */
1369 error = -EBUSY;
1370 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1371 goto out;
1372
1373 sprintf(silly, ".nfs%*.*lx",
1374 i_inosize, i_inosize, dentry->d_inode->i_ino);
1375
Trond Myklebust34ea8182005-11-04 15:35:02 -05001376 /* Return delegation in anticipation of the rename */
1377 nfs_inode_return_delegation(dentry->d_inode);
1378
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 sdentry = NULL;
1380 do {
1381 char *suffix = silly + slen - countersize;
1382
1383 dput(sdentry);
1384 sillycounter++;
1385 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1386
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001387 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1388 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1391 /*
1392 * N.B. Better to return EBUSY here ... it could be
1393 * dangerous to delete the file while it's in use.
1394 */
1395 if (IS_ERR(sdentry))
1396 goto out;
1397 } while(sdentry->d_inode != NULL); /* need negative lookup */
1398
1399 qsilly.name = silly;
1400 qsilly.len = strlen(silly);
1401 nfs_begin_data_update(dir);
1402 if (dentry->d_inode) {
1403 nfs_begin_data_update(dentry->d_inode);
1404 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1405 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001406 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 nfs_end_data_update(dentry->d_inode);
1408 } else
1409 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1410 dir, &qsilly);
1411 nfs_end_data_update(dir);
1412 if (!error) {
1413 nfs_renew_times(dentry);
1414 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1415 d_move(dentry, sdentry);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001416 error = nfs_async_unlink(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 /* If we return 0 we don't unlink */
1418 }
1419 dput(sdentry);
1420out:
1421 return error;
1422}
1423
1424/*
1425 * Remove a file after making sure there are no pending writes,
1426 * and after checking that the file has only one user.
1427 *
1428 * We invalidate the attribute cache and free the inode prior to the operation
1429 * to avoid possible races if the server reuses the inode.
1430 */
1431static int nfs_safe_remove(struct dentry *dentry)
1432{
1433 struct inode *dir = dentry->d_parent->d_inode;
1434 struct inode *inode = dentry->d_inode;
1435 int error = -EBUSY;
1436
1437 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1438 dentry->d_parent->d_name.name, dentry->d_name.name);
1439
1440 /* If the dentry was sillyrenamed, we simply call d_delete() */
1441 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1442 error = 0;
1443 goto out;
1444 }
1445
1446 nfs_begin_data_update(dir);
1447 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001448 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 nfs_begin_data_update(inode);
1450 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1451 /* The VFS may want to delete this inode */
1452 if (error == 0)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001453 drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001454 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455 nfs_end_data_update(inode);
1456 } else
1457 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1458 nfs_end_data_update(dir);
1459out:
1460 return error;
1461}
1462
1463/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1464 * belongs to an active ".nfs..." file and we return -EBUSY.
1465 *
1466 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1467 */
1468static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1469{
1470 int error;
1471 int need_rehash = 0;
1472
1473 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1474 dir->i_ino, dentry->d_name.name);
1475
1476 lock_kernel();
1477 spin_lock(&dcache_lock);
1478 spin_lock(&dentry->d_lock);
1479 if (atomic_read(&dentry->d_count) > 1) {
1480 spin_unlock(&dentry->d_lock);
1481 spin_unlock(&dcache_lock);
Trond Myklebustccfeb502007-01-13 02:28:12 -05001482 /* Start asynchronous writeout of the inode */
1483 write_inode_now(dentry->d_inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 error = nfs_sillyrename(dir, dentry);
1485 unlock_kernel();
1486 return error;
1487 }
1488 if (!d_unhashed(dentry)) {
1489 __d_drop(dentry);
1490 need_rehash = 1;
1491 }
1492 spin_unlock(&dentry->d_lock);
1493 spin_unlock(&dcache_lock);
1494 error = nfs_safe_remove(dentry);
1495 if (!error) {
1496 nfs_renew_times(dentry);
1497 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1498 } else if (need_rehash)
1499 d_rehash(dentry);
1500 unlock_kernel();
1501 return error;
1502}
1503
Chuck Lever873101b2006-08-22 20:06:23 -04001504/*
1505 * To create a symbolic link, most file systems instantiate a new inode,
1506 * add a page to it containing the path, then write it out to the disk
1507 * using prepare_write/commit_write.
1508 *
1509 * Unfortunately the NFS client can't create the in-core inode first
1510 * because it needs a file handle to create an in-core inode (see
1511 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1512 * symlink request has completed on the server.
1513 *
1514 * So instead we allocate a raw page, copy the symname into it, then do
1515 * the SYMLINK request with the page as the buffer. If it succeeds, we
1516 * now have a new file handle and can instantiate an in-core NFS inode
1517 * and move the raw page into its mapping.
1518 */
1519static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520{
Chuck Lever873101b2006-08-22 20:06:23 -04001521 struct pagevec lru_pvec;
1522 struct page *page;
1523 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001525 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 int error;
1527
1528 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1529 dir->i_ino, dentry->d_name.name, symname);
1530
Chuck Lever873101b2006-08-22 20:06:23 -04001531 if (pathlen > PAGE_SIZE)
1532 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533
Chuck Lever873101b2006-08-22 20:06:23 -04001534 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1535 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536
1537 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001538
Jeff Layton83d93f22007-06-07 09:58:08 -04001539 page = alloc_page(GFP_HIGHUSER);
Chuck Lever873101b2006-08-22 20:06:23 -04001540 if (!page) {
1541 unlock_kernel();
1542 return -ENOMEM;
1543 }
1544
1545 kaddr = kmap_atomic(page, KM_USER0);
1546 memcpy(kaddr, symname, pathlen);
1547 if (pathlen < PAGE_SIZE)
1548 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1549 kunmap_atomic(kaddr, KM_USER0);
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 nfs_begin_data_update(dir);
Chuck Lever94a6d752006-08-22 20:06:23 -04001552 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 nfs_end_data_update(dir);
Chuck Lever873101b2006-08-22 20:06:23 -04001554 if (error != 0) {
1555 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1556 dir->i_sb->s_id, dir->i_ino,
1557 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001559 __free_page(page);
1560 unlock_kernel();
1561 return error;
1562 }
1563
1564 /*
1565 * No big deal if we can't add this page to the page cache here.
1566 * READLINK will get the missing page from the server if needed.
1567 */
1568 pagevec_init(&lru_pvec, 0);
1569 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1570 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001571 pagevec_add(&lru_pvec, page);
1572 pagevec_lru_add(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001573 SetPageUptodate(page);
1574 unlock_page(page);
1575 } else
1576 __free_page(page);
1577
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001579 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580}
1581
1582static int
1583nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1584{
1585 struct inode *inode = old_dentry->d_inode;
1586 int error;
1587
1588 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1589 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1590 dentry->d_parent->d_name.name, dentry->d_name.name);
1591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 nfs_begin_data_update(dir);
1594 nfs_begin_data_update(inode);
1595 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001596 if (error == 0) {
1597 atomic_inc(&inode->i_count);
1598 d_instantiate(dentry, inode);
1599 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 nfs_end_data_update(inode);
1601 nfs_end_data_update(dir);
1602 unlock_kernel();
1603 return error;
1604}
1605
1606/*
1607 * RENAME
1608 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1609 * different file handle for the same inode after a rename (e.g. when
1610 * moving to a different directory). A fail-safe method to do so would
1611 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1612 * rename the old file using the sillyrename stuff. This way, the original
1613 * file in old_dir will go away when the last process iput()s the inode.
1614 *
1615 * FIXED.
1616 *
1617 * It actually works quite well. One needs to have the possibility for
1618 * at least one ".nfs..." file in each directory the file ever gets
1619 * moved or linked to which happens automagically with the new
1620 * implementation that only depends on the dcache stuff instead of
1621 * using the inode layer
1622 *
1623 * Unfortunately, things are a little more complicated than indicated
1624 * above. For a cross-directory move, we want to make sure we can get
1625 * rid of the old inode after the operation. This means there must be
1626 * no pending writes (if it's a file), and the use count must be 1.
1627 * If these conditions are met, we can drop the dentries before doing
1628 * the rename.
1629 */
1630static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1631 struct inode *new_dir, struct dentry *new_dentry)
1632{
1633 struct inode *old_inode = old_dentry->d_inode;
1634 struct inode *new_inode = new_dentry->d_inode;
1635 struct dentry *dentry = NULL, *rehash = NULL;
1636 int error = -EBUSY;
1637
1638 /*
1639 * To prevent any new references to the target during the rename,
1640 * we unhash the dentry and free the inode in advance.
1641 */
1642 lock_kernel();
1643 if (!d_unhashed(new_dentry)) {
1644 d_drop(new_dentry);
1645 rehash = new_dentry;
1646 }
1647
1648 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1649 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1650 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1651 atomic_read(&new_dentry->d_count));
1652
1653 /*
1654 * First check whether the target is busy ... we can't
1655 * safely do _any_ rename if the target is in use.
1656 *
1657 * For files, make a copy of the dentry and then do a
1658 * silly-rename. If the silly-rename succeeds, the
1659 * copied dentry is hashed and becomes the new target.
1660 */
1661 if (!new_inode)
1662 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001663 if (S_ISDIR(new_inode->i_mode)) {
1664 error = -EISDIR;
1665 if (!S_ISDIR(old_inode->i_mode))
1666 goto out;
1667 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 int err;
1669 /* copy the target dentry's name */
1670 dentry = d_alloc(new_dentry->d_parent,
1671 &new_dentry->d_name);
1672 if (!dentry)
1673 goto out;
1674
1675 /* silly-rename the existing target ... */
1676 err = nfs_sillyrename(new_dir, new_dentry);
1677 if (!err) {
1678 new_dentry = rehash = dentry;
1679 new_inode = NULL;
1680 /* instantiate the replacement target */
1681 d_instantiate(new_dentry, NULL);
Jesper Juhl7a13e932007-04-26 00:29:02 -07001682 } else if (atomic_read(&new_dentry->d_count) > 1)
1683 /* dentry still busy? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 goto out;
Trond Myklebust20509f12005-08-25 16:25:34 -07001685 } else
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001686 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687
1688go_ahead:
1689 /*
1690 * ... prune child dentries and writebacks if needed.
1691 */
1692 if (atomic_read(&old_dentry->d_count) > 1) {
Trond Myklebuste1552e12007-04-14 19:07:28 -04001693 if (S_ISREG(old_inode->i_mode))
1694 nfs_wb_all(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 shrink_dcache_parent(old_dentry);
1696 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001697 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698
Trond Myklebust24174112006-01-03 09:55:33 +01001699 if (new_inode != NULL) {
1700 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001702 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703
1704 nfs_begin_data_update(old_dir);
1705 nfs_begin_data_update(new_dir);
1706 nfs_begin_data_update(old_inode);
1707 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1708 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001709 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 nfs_end_data_update(old_inode);
1711 nfs_end_data_update(new_dir);
1712 nfs_end_data_update(old_dir);
1713out:
1714 if (rehash)
1715 d_rehash(rehash);
1716 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07001717 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 nfs_renew_times(new_dentry);
Trond Myklebustc79ba782007-01-31 08:16:24 -05001719 nfs_refresh_verifier(new_dentry, nfs_save_change_attribute(new_dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 }
1721
1722 /* new dentry created? */
1723 if (dentry)
1724 dput(dentry);
1725 unlock_kernel();
1726 return error;
1727}
1728
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001729static DEFINE_SPINLOCK(nfs_access_lru_lock);
1730static LIST_HEAD(nfs_access_lru_list);
1731static atomic_long_t nfs_access_nr_entries;
1732
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001733static void nfs_access_free_entry(struct nfs_access_entry *entry)
1734{
1735 put_rpccred(entry->cred);
1736 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001737 smp_mb__before_atomic_dec();
1738 atomic_long_dec(&nfs_access_nr_entries);
1739 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001740}
1741
Trond Myklebust979df722006-07-25 11:28:19 -04001742int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1743{
1744 LIST_HEAD(head);
1745 struct nfs_inode *nfsi;
1746 struct nfs_access_entry *cache;
1747
Trond Myklebust979df722006-07-25 11:28:19 -04001748restart:
Trond Myklebusta50f7952007-06-05 19:23:43 -04001749 spin_lock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001750 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1751 struct inode *inode;
1752
1753 if (nr_to_scan-- == 0)
1754 break;
1755 inode = igrab(&nfsi->vfs_inode);
1756 if (inode == NULL)
1757 continue;
1758 spin_lock(&inode->i_lock);
1759 if (list_empty(&nfsi->access_cache_entry_lru))
1760 goto remove_lru_entry;
1761 cache = list_entry(nfsi->access_cache_entry_lru.next,
1762 struct nfs_access_entry, lru);
1763 list_move(&cache->lru, &head);
1764 rb_erase(&cache->rb_node, &nfsi->access_cache);
1765 if (!list_empty(&nfsi->access_cache_entry_lru))
1766 list_move_tail(&nfsi->access_cache_inode_lru,
1767 &nfs_access_lru_list);
1768 else {
1769remove_lru_entry:
1770 list_del_init(&nfsi->access_cache_inode_lru);
1771 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1772 }
1773 spin_unlock(&inode->i_lock);
Trond Myklebusta50f7952007-06-05 19:23:43 -04001774 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001775 iput(inode);
1776 goto restart;
1777 }
1778 spin_unlock(&nfs_access_lru_lock);
1779 while (!list_empty(&head)) {
1780 cache = list_entry(head.next, struct nfs_access_entry, lru);
1781 list_del(&cache->lru);
1782 nfs_access_free_entry(cache);
1783 }
1784 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1785}
1786
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001787static void __nfs_access_zap_cache(struct inode *inode)
1788{
1789 struct nfs_inode *nfsi = NFS_I(inode);
1790 struct rb_root *root_node = &nfsi->access_cache;
1791 struct rb_node *n, *dispose = NULL;
1792 struct nfs_access_entry *entry;
1793
1794 /* Unhook entries from the cache */
1795 while ((n = rb_first(root_node)) != NULL) {
1796 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1797 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001798 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001799 n->rb_left = dispose;
1800 dispose = n;
1801 }
1802 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1803 spin_unlock(&inode->i_lock);
1804
1805 /* Now kill them all! */
1806 while (dispose != NULL) {
1807 n = dispose;
1808 dispose = n->rb_left;
1809 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1810 }
1811}
1812
1813void nfs_access_zap_cache(struct inode *inode)
1814{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001815 /* Remove from global LRU init */
1816 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1817 spin_lock(&nfs_access_lru_lock);
1818 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1819 spin_unlock(&nfs_access_lru_lock);
1820 }
1821
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001822 spin_lock(&inode->i_lock);
1823 /* This will release the spinlock */
1824 __nfs_access_zap_cache(inode);
1825}
1826
1827static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1828{
1829 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1830 struct nfs_access_entry *entry;
1831
1832 while (n != NULL) {
1833 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1834
1835 if (cred < entry->cred)
1836 n = n->rb_left;
1837 else if (cred > entry->cred)
1838 n = n->rb_right;
1839 else
1840 return entry;
1841 }
1842 return NULL;
1843}
1844
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1846{
Chuck Lever55296802005-08-18 11:24:09 -07001847 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001848 struct nfs_access_entry *cache;
1849 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001851 spin_lock(&inode->i_lock);
1852 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1853 goto out_zap;
1854 cache = nfs_access_search_rbtree(inode, cred);
1855 if (cache == NULL)
1856 goto out;
1857 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1858 goto out_stale;
1859 res->jiffies = cache->jiffies;
1860 res->cred = cache->cred;
1861 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001862 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001863 err = 0;
1864out:
1865 spin_unlock(&inode->i_lock);
1866 return err;
1867out_stale:
1868 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001869 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001870 spin_unlock(&inode->i_lock);
1871 nfs_access_free_entry(cache);
1872 return -ENOENT;
1873out_zap:
1874 /* This will release the spinlock */
1875 __nfs_access_zap_cache(inode);
1876 return -ENOENT;
1877}
1878
1879static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1880{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001881 struct nfs_inode *nfsi = NFS_I(inode);
1882 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001883 struct rb_node **p = &root_node->rb_node;
1884 struct rb_node *parent = NULL;
1885 struct nfs_access_entry *entry;
1886
1887 spin_lock(&inode->i_lock);
1888 while (*p != NULL) {
1889 parent = *p;
1890 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1891
1892 if (set->cred < entry->cred)
1893 p = &parent->rb_left;
1894 else if (set->cred > entry->cred)
1895 p = &parent->rb_right;
1896 else
1897 goto found;
1898 }
1899 rb_link_node(&set->rb_node, parent, p);
1900 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001901 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001902 spin_unlock(&inode->i_lock);
1903 return;
1904found:
1905 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001906 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1907 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001908 spin_unlock(&inode->i_lock);
1909 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910}
1911
1912void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1913{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001914 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1915 if (cache == NULL)
1916 return;
1917 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001919 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001921
1922 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001923
1924 /* Update accounting */
1925 smp_mb__before_atomic_inc();
1926 atomic_long_inc(&nfs_access_nr_entries);
1927 smp_mb__after_atomic_inc();
1928
1929 /* Add inode to global LRU list */
1930 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1931 spin_lock(&nfs_access_lru_lock);
1932 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1933 spin_unlock(&nfs_access_lru_lock);
1934 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
1936
1937static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1938{
1939 struct nfs_access_entry cache;
1940 int status;
1941
1942 status = nfs_access_get_cached(inode, cred, &cache);
1943 if (status == 0)
1944 goto out;
1945
1946 /* Be clever: ask server to check for all possible rights */
1947 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1948 cache.cred = cred;
1949 cache.jiffies = jiffies;
1950 status = NFS_PROTO(inode)->access(inode, &cache);
1951 if (status != 0)
1952 return status;
1953 nfs_access_add_cache(inode, &cache);
1954out:
1955 if ((cache.mask & mask) == mask)
1956 return 0;
1957 return -EACCES;
1958}
1959
1960int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1961{
1962 struct rpc_cred *cred;
1963 int res = 0;
1964
Chuck Lever91d5b472006-03-20 13:44:14 -05001965 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1966
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 if (mask == 0)
1968 goto out;
1969 /* Is this sys_access() ? */
1970 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1971 goto force_lookup;
1972
1973 switch (inode->i_mode & S_IFMT) {
1974 case S_IFLNK:
1975 goto out;
1976 case S_IFREG:
1977 /* NFSv4 has atomic_open... */
1978 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1979 && nd != NULL
1980 && (nd->flags & LOOKUP_OPEN))
1981 goto out;
1982 break;
1983 case S_IFDIR:
1984 /*
1985 * Optimize away all write operations, since the server
1986 * will check permissions when we perform the op.
1987 */
1988 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1989 goto out;
1990 }
1991
1992force_lookup:
1993 lock_kernel();
1994
1995 if (!NFS_PROTO(inode)->access)
1996 goto out_notsup;
1997
1998 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1999 if (!IS_ERR(cred)) {
2000 res = nfs_do_access(inode, cred, mask);
2001 put_rpccred(cred);
2002 } else
2003 res = PTR_ERR(cred);
2004 unlock_kernel();
2005out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05002006 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2007 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 return res;
2009out_notsup:
2010 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2011 if (res == 0)
2012 res = generic_permission(inode, mask, NULL);
2013 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05002014 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015}
2016
2017/*
2018 * Local variables:
2019 * version-control: t
2020 * kept-new-versions: 5
2021 * End:
2022 */