blob: 58d43daec0842ad2be1e16ac8e97e38ea38b474f [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/dir.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs directory handling functions
7 *
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
18 */
19
20#include <linux/time.h>
21#include <linux/errno.h>
22#include <linux/stat.h>
23#include <linux/fcntl.h>
24#include <linux/string.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
27#include <linux/mm.h>
28#include <linux/sunrpc/clnt.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/pagemap.h>
32#include <linux/smp_lock.h>
Chuck Lever873101b2006-08-22 20:06:23 -040033#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/namei.h>
David Howells54ceac42006-08-22 20:06:13 -040035#include <linux/mount.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040036#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Trond Myklebust4ce79712005-06-22 17:16:21 +000038#include "nfs4_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include "delegation.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050040#include "iostat.h"
Adrian Bunk4c30d562007-11-21 15:04:31 -080041#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/* #define NFS_DEBUG_VERBOSE 1 */
44
45static int nfs_opendir(struct inode *, struct file *);
46static int nfs_readdir(struct file *, void *, filldir_t);
47static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
48static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
49static int nfs_mkdir(struct inode *, struct dentry *, int);
50static int nfs_rmdir(struct inode *, struct dentry *);
51static int nfs_unlink(struct inode *, struct dentry *);
52static int nfs_symlink(struct inode *, struct dentry *, const char *);
53static int nfs_link(struct dentry *, struct inode *, struct dentry *);
54static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
55static int nfs_rename(struct inode *, struct dentry *,
56 struct inode *, struct dentry *);
57static int nfs_fsync_dir(struct file *, struct dentry *, int);
Trond Myklebustf0dd2132005-06-22 17:16:29 +000058static loff_t nfs_llseek_dir(struct file *, loff_t, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080060const struct file_operations nfs_dir_operations = {
Trond Myklebustf0dd2132005-06-22 17:16:29 +000061 .llseek = nfs_llseek_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070062 .read = generic_read_dir,
63 .readdir = nfs_readdir,
64 .open = nfs_opendir,
65 .release = nfs_release,
66 .fsync = nfs_fsync_dir,
67};
68
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080069const struct inode_operations nfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070070 .create = nfs_create,
71 .lookup = nfs_lookup,
72 .link = nfs_link,
73 .unlink = nfs_unlink,
74 .symlink = nfs_symlink,
75 .mkdir = nfs_mkdir,
76 .rmdir = nfs_rmdir,
77 .mknod = nfs_mknod,
78 .rename = nfs_rename,
79 .permission = nfs_permission,
80 .getattr = nfs_getattr,
81 .setattr = nfs_setattr,
82};
83
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000084#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080085const struct inode_operations nfs3_dir_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000086 .create = nfs_create,
87 .lookup = nfs_lookup,
88 .link = nfs_link,
89 .unlink = nfs_unlink,
90 .symlink = nfs_symlink,
91 .mkdir = nfs_mkdir,
92 .rmdir = nfs_rmdir,
93 .mknod = nfs_mknod,
94 .rename = nfs_rename,
95 .permission = nfs_permission,
96 .getattr = nfs_getattr,
97 .setattr = nfs_setattr,
98 .listxattr = nfs3_listxattr,
99 .getxattr = nfs3_getxattr,
100 .setxattr = nfs3_setxattr,
101 .removexattr = nfs3_removexattr,
102};
103#endif /* CONFIG_NFS_V3 */
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105#ifdef CONFIG_NFS_V4
106
107static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800108const struct inode_operations nfs4_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109 .create = nfs_create,
110 .lookup = nfs_atomic_lookup,
111 .link = nfs_link,
112 .unlink = nfs_unlink,
113 .symlink = nfs_symlink,
114 .mkdir = nfs_mkdir,
115 .rmdir = nfs_rmdir,
116 .mknod = nfs_mknod,
117 .rename = nfs_rename,
118 .permission = nfs_permission,
119 .getattr = nfs_getattr,
120 .setattr = nfs_setattr,
J. Bruce Fields6b3b5492005-06-22 17:16:22 +0000121 .getxattr = nfs4_getxattr,
122 .setxattr = nfs4_setxattr,
123 .listxattr = nfs4_listxattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124};
125
126#endif /* CONFIG_NFS_V4 */
127
128/*
129 * Open file
130 */
131static int
132nfs_opendir(struct inode *inode, struct file *filp)
133{
Carsten Otte7451c4f2006-04-19 13:06:37 -0400134 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500136 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
137 inode->i_sb->s_id, inode->i_ino);
138
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 lock_kernel();
140 /* Call generic open code in order to cache credentials */
Carsten Otte7451c4f2006-04-19 13:06:37 -0400141 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 unlock_kernel();
143 return res;
144}
145
Al Viro0dbb4c62006-10-19 23:28:49 -0700146typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147typedef struct {
148 struct file *file;
149 struct page *page;
150 unsigned long page_index;
Al Viro0dbb4c62006-10-19 23:28:49 -0700151 __be32 *ptr;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000152 u64 *dir_cookie;
153 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 struct nfs_entry *entry;
155 decode_dirent_t decode;
156 int plus;
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",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700183 __func__, (long long)desc->entry->cookie,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500184 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;
Benny Halevy3a10c302008-01-23 08:58:59 +0200194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
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:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return -EIO;
216}
217
218static inline
219int dir_decode(nfs_readdir_descriptor_t *desc)
220{
Al Viro0dbb4c62006-10-19 23:28:49 -0700221 __be32 *p = desc->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 p = desc->decode(p, desc->entry, desc->plus);
223 if (IS_ERR(p))
224 return PTR_ERR(p);
225 desc->ptr = p;
Neil Brown1f4eab72007-04-16 09:35:27 +1000226 if (desc->timestamp_valid)
227 desc->entry->fattr->time_start = desc->timestamp;
228 else
229 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 return 0;
231}
232
233static inline
234void dir_page_release(nfs_readdir_descriptor_t *desc)
235{
236 kunmap(desc->page);
237 page_cache_release(desc->page);
238 desc->page = NULL;
239 desc->ptr = NULL;
240}
241
242/*
243 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000244 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 *
246 * If the end of the buffer has been reached, return -EAGAIN, if not,
247 * return the offset within the buffer of the next entry to be
248 * read.
249 */
250static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000251int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700252{
253 struct nfs_entry *entry = desc->entry;
254 int loop_count = 0,
255 status;
256
257 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500258 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700259 __func__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000260 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 break;
262 if (loop_count++ > 200) {
263 loop_count = 0;
264 schedule();
265 }
266 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return status;
268}
269
270/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000271 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000272 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000273 *
274 * If the end of the buffer has been reached, return -EAGAIN, if not,
275 * return the offset within the buffer of the next entry to be
276 * read.
277 */
278static inline
279int find_dirent_index(nfs_readdir_descriptor_t *desc)
280{
281 struct nfs_entry *entry = desc->entry;
282 int loop_count = 0,
283 status;
284
285 for(;;) {
286 status = dir_decode(desc);
287 if (status)
288 break;
289
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500290 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
291 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000292
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000293 if (desc->file->f_pos == desc->current_index) {
294 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000295 break;
296 }
297 desc->current_index++;
298 if (loop_count++ > 200) {
299 loop_count = 0;
300 schedule();
301 }
302 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000303 return status;
304}
305
306/*
307 * Find the given page, and call find_dirent() or find_dirent_index in
308 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 */
310static inline
311int find_dirent_page(nfs_readdir_descriptor_t *desc)
312{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800313 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 struct page *page;
315 int status;
316
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500317 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700318 __func__, desc->page_index,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500319 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
Neil Brown1f4eab72007-04-16 09:35:27 +1000321 /* If we find the page in the page_cache, we cannot be sure
322 * how fresh the data is, so we will ignore readdir_plus attributes.
323 */
324 desc->timestamp_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 page = read_cache_page(inode->i_mapping, desc->page_index,
326 (filler_t *)nfs_readdir_filler, desc);
327 if (IS_ERR(page)) {
328 status = PTR_ERR(page);
329 goto out;
330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /* NOTE: Someone else may have changed the READDIRPLUS flag */
333 desc->page = page;
334 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000335 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000336 status = find_dirent(desc);
337 else
338 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 if (status < 0)
340 dir_page_release(desc);
341 out:
Harvey Harrison3110ff82008-05-02 13:42:44 -0700342 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
346/*
347 * Recurse through the page cache pages, and return a
348 * filled nfs_entry structure of the next directory entry if possible.
349 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000350 * The target for the search is '*desc->dir_cookie' if non-0,
351 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 */
353static inline
354int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
355{
356 int loop_count = 0;
357 int res;
358
Olivier Galibert00a92642005-06-22 17:16:29 +0000359 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000360 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500361 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
362 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000363 desc->page_index = 0;
364 desc->entry->cookie = desc->entry->prev_cookie = 0;
365 desc->entry->eof = 0;
366 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000367 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500368 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
369 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 for (;;) {
372 res = find_dirent_page(desc);
373 if (res != -EAGAIN)
374 break;
375 /* Align to beginning of next page */
376 desc->page_index ++;
377 if (loop_count++ > 200) {
378 loop_count = 0;
379 schedule();
380 }
381 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500382
Harvey Harrison3110ff82008-05-02 13:42:44 -0700383 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384 return res;
385}
386
387static inline unsigned int dt_type(struct inode *inode)
388{
389 return (inode->i_mode >> 12) & 15;
390}
391
392static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
393
394/*
395 * Once we've found the start of the dirent within a page: fill 'er up...
396 */
397static
398int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
399 filldir_t filldir)
400{
401 struct file *file = desc->file;
402 struct nfs_entry *entry = desc->entry;
403 struct dentry *dentry = NULL;
Peter Staubach4e769b92007-08-03 15:07:10 -0400404 u64 fileid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 int loop_count = 0,
406 res;
407
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500408 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
409 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
411 for(;;) {
412 unsigned d_type = DT_UNKNOWN;
413 /* Note: entry->prev_cookie contains the cookie for
414 * retrieving the current dirent on the server */
Peter Staubach4e769b92007-08-03 15:07:10 -0400415 fileid = entry->ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416
417 /* Get a dentry if we have one */
418 if (dentry != NULL)
419 dput(dentry);
420 dentry = nfs_readdir_lookup(desc);
421
422 /* Use readdirplus info */
423 if (dentry != NULL && dentry->d_inode != NULL) {
424 d_type = dt_type(dentry->d_inode);
Peter Staubach4e769b92007-08-03 15:07:10 -0400425 fileid = NFS_FILEID(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 }
427
428 res = filldir(dirent, entry->name, entry->len,
Trond Myklebustf43bf0b2007-10-09 12:01:04 -0400429 file->f_pos, nfs_compat_user_ino64(fileid),
430 d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 if (res < 0)
432 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000433 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000434 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (dir_decode(desc) != 0) {
436 desc->page_index ++;
437 break;
438 }
439 if (loop_count++ > 200) {
440 loop_count = 0;
441 schedule();
442 }
443 }
444 dir_page_release(desc);
445 if (dentry != NULL)
446 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500447 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
448 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 return res;
450}
451
452/*
453 * If we cannot find a cookie in our cache, we suspect that this is
454 * because it points to a deleted file, so we ask the server to return
455 * whatever it thinks is the next entry. We then feed this to filldir.
456 * If all goes well, we should then be able to find our way round the
457 * cache on the next call to readdir_search_pagecache();
458 *
459 * NOTE: we cannot add the anonymous page to the pagecache because
460 * the data it contains might not be page aligned. Besides,
461 * we should already have a complete representation of the
462 * directory in the page cache by the time we get here.
463 */
464static inline
465int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
466 filldir_t filldir)
467{
468 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800469 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 struct rpc_cred *cred = nfs_file_cred(file);
471 struct page *page = NULL;
472 int status;
Neil Brown1f4eab72007-04-16 09:35:27 +1000473 unsigned long timestamp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500475 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
476 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477
478 page = alloc_page(GFP_HIGHUSER);
479 if (!page) {
480 status = -ENOMEM;
481 goto out;
482 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000483 timestamp = jiffies;
Jeff Layton25606652008-02-12 06:49:01 -0500484 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
485 *desc->dir_cookie, page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 NFS_SERVER(inode)->dtsize,
487 desc->plus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 desc->page = page;
489 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Jeff Layton25606652008-02-12 06:49:01 -0500490 if (status >= 0) {
Neil Brown1f4eab72007-04-16 09:35:27 +1000491 desc->timestamp = timestamp;
492 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000494 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 } else
496 status = -EIO;
497 if (status < 0)
498 goto out_release;
499
500 status = nfs_do_filldir(desc, dirent, filldir);
501
502 /* Reset read descriptor so it searches the page cache from
503 * the start upon the next call to readdir_search_pagecache() */
504 desc->page_index = 0;
505 desc->entry->cookie = desc->entry->prev_cookie = 0;
506 desc->entry->eof = 0;
507 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500508 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700509 __func__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 return status;
511 out_release:
512 dir_page_release(desc);
513 goto out;
514}
515
Olivier Galibert00a92642005-06-22 17:16:29 +0000516/* The file offset position represents the dirent entry number. A
517 last cookie cache takes care of the common case of reading the
518 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 */
520static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
521{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800522 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 struct inode *inode = dentry->d_inode;
524 nfs_readdir_descriptor_t my_desc,
525 *desc = &my_desc;
526 struct nfs_entry my_entry;
527 struct nfs_fh fh;
528 struct nfs_fattr fattr;
529 long res;
530
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500531 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
532 dentry->d_parent->d_name.name, dentry->d_name.name,
533 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500534 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
535
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 lock_kernel();
537
Linus Torvalds1da177e2005-04-16 15:20:36 -0700538 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000539 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000540 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000541 * to either find the entry with the appropriate number or
542 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
544 memset(desc, 0, sizeof(*desc));
545
546 desc->file = filp;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400547 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548 desc->decode = NFS_PROTO(inode)->decode_dirent;
549 desc->plus = NFS_USE_READDIRPLUS(inode);
550
551 my_entry.cookie = my_entry.prev_cookie = 0;
552 my_entry.eof = 0;
553 my_entry.fh = &fh;
554 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400555 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 desc->entry = &my_entry;
557
Trond Myklebust565277f2007-10-15 18:17:53 -0400558 nfs_block_sillyrename(dentry);
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500559 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
560 if (res < 0)
561 goto out;
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 while(!desc->entry->eof) {
564 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000565
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 if (res == -EBADCOOKIE) {
567 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000568 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 /* Or that the server has 'lost' a cookie */
570 res = uncached_readdir(desc, dirent, filldir);
571 if (res >= 0)
572 continue;
573 }
574 res = 0;
575 break;
576 }
577 if (res == -ETOOSMALL && desc->plus) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200578 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 nfs_zap_caches(inode);
580 desc->plus = 0;
581 desc->entry->eof = 0;
582 continue;
583 }
584 if (res < 0)
585 break;
586
587 res = nfs_do_filldir(desc, dirent, filldir);
588 if (res < 0) {
589 res = 0;
590 break;
591 }
592 }
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500593out:
Trond Myklebust565277f2007-10-15 18:17:53 -0400594 nfs_unblock_sillyrename(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500596 if (res > 0)
597 res = 0;
598 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
599 dentry->d_parent->d_name.name, dentry->d_name.name,
600 res);
601 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602}
603
Trond Myklebust10afec92007-05-14 17:16:04 -0400604static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000605{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800606 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000607 switch (origin) {
608 case 1:
609 offset += filp->f_pos;
610 case 0:
611 if (offset >= 0)
612 break;
613 default:
614 offset = -EINVAL;
615 goto out;
616 }
617 if (offset != filp->f_pos) {
618 filp->f_pos = offset;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400619 nfs_file_open_context(filp)->dir_cookie = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000620 }
621out:
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800622 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000623 return offset;
624}
625
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626/*
627 * All directory operations under NFS are synchronous, so fsync()
628 * is a dummy operation.
629 */
Trond Myklebust10afec92007-05-14 17:16:04 -0400630static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500632 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
633 dentry->d_parent->d_name.name, dentry->d_name.name,
634 datasync);
635
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 return 0;
637}
638
Trond Myklebustbfc69a42007-10-15 18:18:29 -0400639/**
640 * nfs_force_lookup_revalidate - Mark the directory as having changed
641 * @dir - pointer to directory inode
642 *
643 * This forces the revalidation code in nfs_lookup_revalidate() to do a
644 * full lookup on all child dentries of 'dir' whenever a change occurs
645 * on the server that might have invalidated our dcache.
646 *
647 * The caller should be holding dir->i_lock
648 */
649void nfs_force_lookup_revalidate(struct inode *dir)
650{
651 NFS_I(dir)->cache_change_attribute = jiffies;
652}
653
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654/*
655 * A check for whether or not the parent directory has changed.
656 * In the case it has, we assume that the dentries are untrustworthy
657 * and may need to be looked up again.
658 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500659static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660{
661 if (IS_ROOT(dentry))
662 return 1;
Trond Myklebustf2c77f42007-10-02 12:54:39 -0400663 if (!nfs_verify_change_attribute(dir, dentry->d_time))
664 return 0;
665 /* Revalidate nfsi->cache_change_attribute before we declare a match */
666 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
667 return 0;
668 if (!nfs_verify_change_attribute(dir, dentry->d_time))
669 return 0;
670 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671}
672
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400674 * Return the intent data that applies to this particular path component
675 *
676 * Note that the current set of intents only apply to the very last
677 * component of the path.
678 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
679 */
680static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
681{
682 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
683 return 0;
684 return nd->flags & mask;
685}
686
687/*
Trond Myklebusta12802c2007-10-02 19:13:04 -0400688 * Use intent information to check whether or not we're going to do
689 * an O_EXCL create using this path component.
690 */
691static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
692{
693 if (NFS_PROTO(dir)->version == 2)
694 return 0;
695 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
696 return 0;
697 return (nd->intent.open.flags & O_EXCL) != 0;
698}
699
700/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400701 * Inode and filehandle revalidation for lookups.
702 *
703 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
704 * or if the intent information indicates that we're about to open this
705 * particular file and the "nocto" mount flag is not set.
706 *
707 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708static inline
709int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
710{
711 struct nfs_server *server = NFS_SERVER(inode);
712
Trond Myklebust4e99a1f2008-03-06 12:34:59 -0500713 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
714 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400717 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 goto out_force;
719 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400720 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400721 !(server->flags & NFS_MOUNT_NOCTO) &&
722 (S_ISREG(inode->i_mode) ||
723 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 goto out_force;
Trond Myklebust4f48af42007-10-02 23:13:32 -0400725 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726 }
727 return nfs_revalidate_inode(server, inode);
728out_force:
729 return __nfs_revalidate_inode(server, inode);
730}
731
732/*
733 * We judge how long we want to trust negative
734 * dentries by looking at the parent inode mtime.
735 *
736 * If parent mtime has changed, we revalidate, else we wait for a
737 * period corresponding to the parent's attribute cache timeout value.
738 */
739static inline
740int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
741 struct nameidata *nd)
742{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400744 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 return 0;
746 return !nfs_check_verifier(dir, dentry);
747}
748
749/*
750 * This is called every time the dcache has a lookup hit,
751 * and we should check whether we can really trust that
752 * lookup.
753 *
754 * NOTE! The hit can be a negative hit too, don't assume
755 * we have an inode!
756 *
757 * If the parent directory is seen to have changed, we throw out the
758 * cached dentry and do a new lookup.
759 */
760static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
761{
762 struct inode *dir;
763 struct inode *inode;
764 struct dentry *parent;
765 int error;
766 struct nfs_fh fhandle;
767 struct nfs_fattr fattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768
769 parent = dget_parent(dentry);
770 lock_kernel();
771 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500772 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 inode = dentry->d_inode;
774
775 if (!inode) {
776 if (nfs_neg_need_reval(dir, dentry, nd))
777 goto out_bad;
778 goto out_valid;
779 }
780
781 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500782 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700783 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500784 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 goto out_bad;
786 }
787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 /* Force a full look up iff the parent directory has changed */
Trond Myklebusta12802c2007-10-02 19:13:04 -0400789 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 if (nfs_lookup_verify_inode(inode, nd))
791 goto out_zap_parent;
792 goto out_valid;
793 }
794
795 if (NFS_STALE(inode))
796 goto out_bad;
797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 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
Trond Myklebustcf8ba452007-10-01 13:46:53 -0400806 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807 out_valid:
808 unlock_kernel();
809 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500810 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700811 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500812 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813 return 1;
814out_zap_parent:
815 nfs_zap_caches(dir);
816 out_bad:
Trond Myklebusta1643a92007-09-29 17:25:43 -0400817 nfs_mark_for_revalidate(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 if (inode && S_ISDIR(inode->i_mode)) {
819 /* Purge readdir caches. */
820 nfs_zap_caches(inode);
821 /* If we have submounts, don't unhash ! */
822 if (have_submounts(dentry))
823 goto out_valid;
824 shrink_dcache_parent(dentry);
825 }
826 d_drop(dentry);
827 unlock_kernel();
828 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500829 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700830 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500831 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832 return 0;
833}
834
835/*
836 * This is called from dput() when d_count is going to 0.
837 */
838static int nfs_dentry_delete(struct dentry *dentry)
839{
840 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
841 dentry->d_parent->d_name.name, dentry->d_name.name,
842 dentry->d_flags);
843
Trond Myklebust77f11192008-01-28 19:43:19 -0500844 /* Unhash any dentry with a stale inode */
845 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
846 return 1;
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
849 /* Unhash it, so that ->d_iput() would be called */
850 return 1;
851 }
852 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
853 /* Unhash it, so that ancestors of killed async unlink
854 * files will be cleaned up during umount */
855 return 1;
856 }
857 return 0;
858
859}
860
861/*
862 * Called when the dentry loses inode.
863 * We use it to clean up silly-renamed files.
864 */
865static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
866{
Neil Brown83672d32007-02-26 12:48:25 +1100867 if (S_ISDIR(inode->i_mode))
868 /* drop any readdir cache as it could easily be old */
869 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
872 lock_kernel();
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700873 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -0400874 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 unlock_kernel();
876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
887{
888 struct dentry *res;
Trond Myklebust565277f2007-10-15 18:17:53 -0400889 struct dentry *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 struct inode *inode = NULL;
891 int error;
892 struct nfs_fh fhandle;
893 struct nfs_fattr fattr;
894
895 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
896 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500897 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 res = ERR_PTR(-ENAMETOOLONG);
900 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
901 goto out;
902
903 res = ERR_PTR(-ENOMEM);
904 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
905
906 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907
Trond Myklebustfd684072006-09-05 12:27:44 -0400908 /*
909 * If we're doing an exclusive create, optimize away the lookup
910 * but don't hash the dentry.
911 */
912 if (nfs_is_exclusive_create(dir, nd)) {
913 d_instantiate(dentry, NULL);
914 res = NULL;
915 goto out_unlock;
916 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917
Trond Myklebust565277f2007-10-15 18:17:53 -0400918 parent = dentry->d_parent;
919 /* Protect against concurrent sillydeletes */
920 nfs_block_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
922 if (error == -ENOENT)
923 goto no_entry;
924 if (error < 0) {
925 res = ERR_PTR(error);
Trond Myklebust565277f2007-10-15 18:17:53 -0400926 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500929 res = (struct dentry *)inode;
930 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400931 goto out_unblock_sillyrename;
David Howells54ceac42006-08-22 20:06:13 -0400932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400934 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -0700935 if (res != NULL) {
936 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400937 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700939 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Trond Myklebust565277f2007-10-15 18:17:53 -0400941out_unblock_sillyrename:
942 nfs_unblock_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943out_unlock:
944 unlock_kernel();
945out:
946 return res;
947}
948
949#ifdef CONFIG_NFS_V4
950static int nfs_open_revalidate(struct dentry *, struct nameidata *);
951
952struct dentry_operations nfs4_dentry_operations = {
953 .d_revalidate = nfs_open_revalidate,
954 .d_delete = nfs_dentry_delete,
955 .d_iput = nfs_dentry_iput,
956};
957
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400958/*
959 * Use intent information to determine whether we need to substitute
960 * the NFSv4-style stateful OPEN for the LOOKUP call
961 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962static int is_atomic_open(struct inode *dir, struct nameidata *nd)
963{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400964 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 return 0;
966 /* NFS does not (yet) have a stateful open for directories */
967 if (nd->flags & LOOKUP_DIRECTORY)
968 return 0;
969 /* Are we trying to write to a read only partition? */
Dave Hansen2c463e92008-02-15 14:37:56 -0800970 if (__mnt_is_readonly(nd->path.mnt) &&
971 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return 0;
973 return 1;
974}
975
976static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
977{
978 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979 int error;
980
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500981 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
982 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 /* Check that we are indeed trying to open this file */
985 if (!is_atomic_open(dir, nd))
986 goto no_open;
987
988 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
989 res = ERR_PTR(-ENAMETOOLONG);
990 goto out;
991 }
992 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
993
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400994 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
995 * the dentry. */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700996 if (nd->intent.open.flags & O_EXCL) {
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -0400997 d_instantiate(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -0700998 goto out;
999 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000
1001 /* Open the file on the server */
1002 lock_kernel();
Trond Myklebust60ccd4e2007-09-29 17:48:19 -04001003 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001005 if (IS_ERR(res)) {
1006 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 switch (error) {
1008 /* Make a negative dentry */
1009 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001010 res = NULL;
1011 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001013 case -EISDIR:
1014 case -ENOTDIR:
1015 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016 case -ELOOP:
1017 if (!(nd->intent.open.flags & O_NOFOLLOW))
1018 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 /* case -EINVAL: */
1020 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 goto out;
1022 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001023 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024 dentry = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025out:
1026 return res;
1027no_open:
1028 return nfs_lookup(dir, dentry, nd);
1029}
1030
1031static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1032{
1033 struct dentry *parent = NULL;
1034 struct inode *inode = dentry->d_inode;
1035 struct inode *dir;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 int openflags, ret = 0;
1037
1038 parent = dget_parent(dentry);
1039 dir = parent->d_inode;
1040 if (!is_atomic_open(dir, nd))
1041 goto no_open;
1042 /* We can't create new files in nfs_open_revalidate(), so we
1043 * optimize away revalidation of negative dentries.
1044 */
Trond Myklebust216d5d02007-10-01 20:10:12 -04001045 if (inode == NULL) {
1046 if (!nfs_neg_need_reval(dir, dentry, nd))
1047 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 goto out;
Trond Myklebust216d5d02007-10-01 20:10:12 -04001049 }
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 /* NFS only supports OPEN on regular files */
1052 if (!S_ISREG(inode->i_mode))
1053 goto no_open;
1054 openflags = nd->intent.open.flags;
1055 /* We cannot do exclusive creation on a positive dentry */
1056 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1057 goto no_open;
1058 /* We can't create new files, or truncate existing ones here */
1059 openflags &= ~(O_CREAT|O_TRUNC);
1060
1061 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001062 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 * operations that change the directory. We therefore save the
1064 * change attribute *before* we do the RPC call.
1065 */
1066 lock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001067 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 unlock_kernel();
1069out:
1070 dput(parent);
1071 if (!ret)
1072 d_drop(dentry);
1073 return ret;
1074no_open:
1075 dput(parent);
1076 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1077 return 1;
1078 return nfs_lookup_revalidate(dentry, nd);
1079}
1080#endif /* CONFIG_NFSV4 */
1081
1082static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1083{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -08001084 struct dentry *parent = desc->file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 struct inode *dir = parent->d_inode;
1086 struct nfs_entry *entry = desc->entry;
1087 struct dentry *dentry, *alias;
1088 struct qstr name = {
1089 .name = entry->name,
1090 .len = entry->len,
1091 };
1092 struct inode *inode;
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001093 unsigned long verf = nfs_save_change_attribute(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094
1095 switch (name.len) {
1096 case 2:
1097 if (name.name[0] == '.' && name.name[1] == '.')
1098 return dget_parent(parent);
1099 break;
1100 case 1:
1101 if (name.name[0] == '.')
1102 return dget(parent);
1103 }
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001104
1105 spin_lock(&dir->i_lock);
1106 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1107 spin_unlock(&dir->i_lock);
1108 return NULL;
1109 }
1110 spin_unlock(&dir->i_lock);
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 name.hash = full_name_hash(name.name, name.len);
1113 dentry = d_lookup(parent, &name);
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001114 if (dentry != NULL) {
Trond Myklebustef75c792007-01-16 10:09:44 -05001115 /* Is this a positive dentry that matches the readdir info? */
1116 if (dentry->d_inode != NULL &&
1117 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1118 d_mountpoint(dentry))) {
1119 if (!desc->plus || entry->fh->size == 0)
1120 return dentry;
1121 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1122 entry->fh) == 0)
1123 goto out_renew;
1124 }
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001125 /* No, so d_drop to allow one to be created */
1126 d_drop(dentry);
1127 dput(dentry);
1128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1130 return NULL;
Trond Myklebust54af3bb2007-09-28 12:27:41 -04001131 if (name.len > NFS_SERVER(dir)->namelen)
1132 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001133 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 dentry = d_alloc(parent, &name);
1135 if (dentry == NULL)
1136 return NULL;
1137 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1138 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001139 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 dput(dentry);
1141 return NULL;
1142 }
David Howells54ceac42006-08-22 20:06:13 -04001143
1144 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (alias != NULL) {
1146 dput(dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001147 if (IS_ERR(alias))
1148 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 dentry = alias;
1150 }
David Howells54ceac42006-08-22 20:06:13 -04001151
Trond Myklebustc79ba782007-01-31 08:16:24 -05001152out_renew:
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001153 nfs_set_verifier(dentry, verf);
Trond Myklebustc79ba782007-01-31 08:16:24 -05001154 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155}
1156
1157/*
1158 * Code common to create, mkdir, and mknod.
1159 */
1160int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1161 struct nfs_fattr *fattr)
1162{
Trond Myklebustfab728e2007-09-29 17:41:33 -04001163 struct dentry *parent = dget_parent(dentry);
1164 struct inode *dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 struct inode *inode;
1166 int error = -EACCES;
1167
Trond Myklebustfab728e2007-09-29 17:41:33 -04001168 d_drop(dentry);
1169
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 /* We may have been initialized further down */
1171 if (dentry->d_inode)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001172 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 if (fhandle->size == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1175 if (error)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001176 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 }
Trond Myklebust5724ab32007-10-01 21:51:38 -04001178 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1180 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001181 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 if (error < 0)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001183 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001186 error = PTR_ERR(inode);
1187 if (IS_ERR(inode))
Trond Myklebustfab728e2007-09-29 17:41:33 -04001188 goto out_error;
1189 d_add(dentry, inode);
1190out:
1191 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return 0;
Trond Myklebustfab728e2007-09-29 17:41:33 -04001193out_error:
1194 nfs_mark_for_revalidate(dir);
1195 dput(parent);
1196 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
1198
1199/*
1200 * Following a failed create operation, we drop the dentry rather
1201 * than retain a negative dentry. This avoids a problem in the event
1202 * that the operation succeeded on the server, but an error in the
1203 * reply path made it appear to have failed.
1204 */
1205static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1206 struct nameidata *nd)
1207{
1208 struct iattr attr;
1209 int error;
1210 int open_flags = 0;
1211
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001212 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1213 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214
1215 attr.ia_mode = mode;
1216 attr.ia_valid = ATTR_MODE;
1217
Trond Myklebustad389da2007-06-05 12:30:00 -04001218 if ((nd->flags & LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 open_flags = nd->intent.open.flags;
1220
1221 lock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001222 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (error != 0)
1224 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001225 unlock_kernel();
1226 return 0;
1227out_err:
1228 unlock_kernel();
1229 d_drop(dentry);
1230 return error;
1231}
1232
1233/*
1234 * See comments for nfs_proc_create regarding failed operations.
1235 */
1236static int
1237nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1238{
1239 struct iattr attr;
1240 int status;
1241
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001242 dfprintk(VFS, "NFS: mknod(%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 if (!new_valid_dev(rdev))
1246 return -EINVAL;
1247
1248 attr.ia_mode = mode;
1249 attr.ia_valid = ATTR_MODE;
1250
1251 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253 if (status != 0)
1254 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 unlock_kernel();
1256 return 0;
1257out_err:
1258 unlock_kernel();
1259 d_drop(dentry);
1260 return status;
1261}
1262
1263/*
1264 * See comments for nfs_proc_create regarding failed operations.
1265 */
1266static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1267{
1268 struct iattr attr;
1269 int error;
1270
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001271 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1272 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273
1274 attr.ia_valid = ATTR_MODE;
1275 attr.ia_mode = mode | S_IFDIR;
1276
1277 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 if (error != 0)
1280 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 unlock_kernel();
1282 return 0;
1283out_err:
1284 d_drop(dentry);
1285 unlock_kernel();
1286 return error;
1287}
1288
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001289static void nfs_dentry_handle_enoent(struct dentry *dentry)
1290{
1291 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1292 d_delete(dentry);
1293}
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1296{
1297 int error;
1298
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001299 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1300 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301
1302 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1304 /* Ensure the VFS deletes this inode */
1305 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001306 clear_nlink(dentry->d_inode);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001307 else if (error == -ENOENT)
1308 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 unlock_kernel();
1310
1311 return error;
1312}
1313
1314static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1315{
1316 static unsigned int sillycounter;
Peter Staubach4e769b92007-08-03 15:07:10 -04001317 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 const int countersize = sizeof(sillycounter)*2;
Peter Staubach4e769b92007-08-03 15:07:10 -04001319 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 char silly[slen+1];
1321 struct qstr qsilly;
1322 struct dentry *sdentry;
1323 int error = -EIO;
1324
1325 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1326 dentry->d_parent->d_name.name, dentry->d_name.name,
1327 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001328 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 /*
1331 * We don't allow a dentry to be silly-renamed twice.
1332 */
1333 error = -EBUSY;
1334 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1335 goto out;
1336
Peter Staubach4e769b92007-08-03 15:07:10 -04001337 sprintf(silly, ".nfs%*.*Lx",
1338 fileidsize, fileidsize,
1339 (unsigned long long)NFS_FILEID(dentry->d_inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340
Trond Myklebust34ea81882005-11-04 15:35:02 -05001341 /* Return delegation in anticipation of the rename */
1342 nfs_inode_return_delegation(dentry->d_inode);
1343
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 sdentry = NULL;
1345 do {
1346 char *suffix = silly + slen - countersize;
1347
1348 dput(sdentry);
1349 sillycounter++;
1350 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1351
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001352 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1353 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354
1355 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1356 /*
1357 * N.B. Better to return EBUSY here ... it could be
1358 * dangerous to delete the file while it's in use.
1359 */
1360 if (IS_ERR(sdentry))
1361 goto out;
1362 } while(sdentry->d_inode != NULL); /* need negative lookup */
1363
1364 qsilly.name = silly;
1365 qsilly.len = strlen(silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 if (dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1368 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001369 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 } else
1371 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1372 dir, &qsilly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1375 d_move(dentry, sdentry);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001376 error = nfs_async_unlink(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 /* If we return 0 we don't unlink */
1378 }
1379 dput(sdentry);
1380out:
1381 return error;
1382}
1383
1384/*
1385 * Remove a file after making sure there are no pending writes,
1386 * and after checking that the file has only one user.
1387 *
1388 * We invalidate the attribute cache and free the inode prior to the operation
1389 * to avoid possible races if the server reuses the inode.
1390 */
1391static int nfs_safe_remove(struct dentry *dentry)
1392{
1393 struct inode *dir = dentry->d_parent->d_inode;
1394 struct inode *inode = dentry->d_inode;
1395 int error = -EBUSY;
1396
1397 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1398 dentry->d_parent->d_name.name, dentry->d_name.name);
1399
1400 /* If the dentry was sillyrenamed, we simply call d_delete() */
1401 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1402 error = 0;
1403 goto out;
1404 }
1405
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001407 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1409 /* The VFS may want to delete this inode */
1410 if (error == 0)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001411 drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001412 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 } else
1414 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001415 if (error == -ENOENT)
1416 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417out:
1418 return error;
1419}
1420
1421/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1422 * belongs to an active ".nfs..." file and we return -EBUSY.
1423 *
1424 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1425 */
1426static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1427{
1428 int error;
1429 int need_rehash = 0;
1430
1431 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1432 dir->i_ino, dentry->d_name.name);
1433
1434 lock_kernel();
1435 spin_lock(&dcache_lock);
1436 spin_lock(&dentry->d_lock);
1437 if (atomic_read(&dentry->d_count) > 1) {
1438 spin_unlock(&dentry->d_lock);
1439 spin_unlock(&dcache_lock);
Trond Myklebustccfeb502007-01-13 02:28:12 -05001440 /* Start asynchronous writeout of the inode */
1441 write_inode_now(dentry->d_inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 error = nfs_sillyrename(dir, dentry);
1443 unlock_kernel();
1444 return error;
1445 }
1446 if (!d_unhashed(dentry)) {
1447 __d_drop(dentry);
1448 need_rehash = 1;
1449 }
1450 spin_unlock(&dentry->d_lock);
1451 spin_unlock(&dcache_lock);
1452 error = nfs_safe_remove(dentry);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001453 if (!error || error == -ENOENT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1455 } else if (need_rehash)
1456 d_rehash(dentry);
1457 unlock_kernel();
1458 return error;
1459}
1460
Chuck Lever873101b2006-08-22 20:06:23 -04001461/*
1462 * To create a symbolic link, most file systems instantiate a new inode,
1463 * add a page to it containing the path, then write it out to the disk
1464 * using prepare_write/commit_write.
1465 *
1466 * Unfortunately the NFS client can't create the in-core inode first
1467 * because it needs a file handle to create an in-core inode (see
1468 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1469 * symlink request has completed on the server.
1470 *
1471 * So instead we allocate a raw page, copy the symname into it, then do
1472 * the SYMLINK request with the page as the buffer. If it succeeds, we
1473 * now have a new file handle and can instantiate an in-core NFS inode
1474 * and move the raw page into its mapping.
1475 */
1476static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477{
Chuck Lever873101b2006-08-22 20:06:23 -04001478 struct pagevec lru_pvec;
1479 struct page *page;
1480 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001482 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 int error;
1484
1485 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1486 dir->i_ino, dentry->d_name.name, symname);
1487
Chuck Lever873101b2006-08-22 20:06:23 -04001488 if (pathlen > PAGE_SIZE)
1489 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490
Chuck Lever873101b2006-08-22 20:06:23 -04001491 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1492 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493
1494 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001495
Jeff Layton83d93f22007-06-07 09:58:08 -04001496 page = alloc_page(GFP_HIGHUSER);
Chuck Lever873101b2006-08-22 20:06:23 -04001497 if (!page) {
1498 unlock_kernel();
1499 return -ENOMEM;
1500 }
1501
1502 kaddr = kmap_atomic(page, KM_USER0);
1503 memcpy(kaddr, symname, pathlen);
1504 if (pathlen < PAGE_SIZE)
1505 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1506 kunmap_atomic(kaddr, KM_USER0);
1507
Chuck Lever94a6d752006-08-22 20:06:23 -04001508 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Chuck Lever873101b2006-08-22 20:06:23 -04001509 if (error != 0) {
1510 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1511 dir->i_sb->s_id, dir->i_ino,
1512 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001514 __free_page(page);
1515 unlock_kernel();
1516 return error;
1517 }
1518
1519 /*
1520 * No big deal if we can't add this page to the page cache here.
1521 * READLINK will get the missing page from the server if needed.
1522 */
1523 pagevec_init(&lru_pvec, 0);
1524 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1525 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001526 pagevec_add(&lru_pvec, page);
1527 pagevec_lru_add(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001528 SetPageUptodate(page);
1529 unlock_page(page);
1530 } else
1531 __free_page(page);
1532
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001534 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535}
1536
1537static int
1538nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1539{
1540 struct inode *inode = old_dentry->d_inode;
1541 int error;
1542
1543 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1544 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1545 dentry->d_parent->d_name.name, dentry->d_name.name);
1546
Linus Torvalds1da177e2005-04-16 15:20:36 -07001547 lock_kernel();
Trond Myklebust9697d232007-10-02 21:58:05 -04001548 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001550 if (error == 0) {
1551 atomic_inc(&inode->i_count);
Trond Myklebust9697d232007-10-02 21:58:05 -04001552 d_add(dentry, inode);
Trond Myklebustcf809552005-10-27 22:12:42 -04001553 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 unlock_kernel();
1555 return error;
1556}
1557
1558/*
1559 * RENAME
1560 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1561 * different file handle for the same inode after a rename (e.g. when
1562 * moving to a different directory). A fail-safe method to do so would
1563 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1564 * rename the old file using the sillyrename stuff. This way, the original
1565 * file in old_dir will go away when the last process iput()s the inode.
1566 *
1567 * FIXED.
1568 *
1569 * It actually works quite well. One needs to have the possibility for
1570 * at least one ".nfs..." file in each directory the file ever gets
1571 * moved or linked to which happens automagically with the new
1572 * implementation that only depends on the dcache stuff instead of
1573 * using the inode layer
1574 *
1575 * Unfortunately, things are a little more complicated than indicated
1576 * above. For a cross-directory move, we want to make sure we can get
1577 * rid of the old inode after the operation. This means there must be
1578 * no pending writes (if it's a file), and the use count must be 1.
1579 * If these conditions are met, we can drop the dentries before doing
1580 * the rename.
1581 */
1582static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1583 struct inode *new_dir, struct dentry *new_dentry)
1584{
1585 struct inode *old_inode = old_dentry->d_inode;
1586 struct inode *new_inode = new_dentry->d_inode;
1587 struct dentry *dentry = NULL, *rehash = NULL;
1588 int error = -EBUSY;
1589
1590 /*
1591 * To prevent any new references to the target during the rename,
1592 * we unhash the dentry and free the inode in advance.
1593 */
1594 lock_kernel();
1595 if (!d_unhashed(new_dentry)) {
1596 d_drop(new_dentry);
1597 rehash = new_dentry;
1598 }
1599
1600 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1601 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1602 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1603 atomic_read(&new_dentry->d_count));
1604
1605 /*
1606 * First check whether the target is busy ... we can't
1607 * safely do _any_ rename if the target is in use.
1608 *
1609 * For files, make a copy of the dentry and then do a
1610 * silly-rename. If the silly-rename succeeds, the
1611 * copied dentry is hashed and becomes the new target.
1612 */
1613 if (!new_inode)
1614 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001615 if (S_ISDIR(new_inode->i_mode)) {
1616 error = -EISDIR;
1617 if (!S_ISDIR(old_inode->i_mode))
1618 goto out;
1619 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 int err;
1621 /* copy the target dentry's name */
1622 dentry = d_alloc(new_dentry->d_parent,
1623 &new_dentry->d_name);
1624 if (!dentry)
1625 goto out;
1626
1627 /* silly-rename the existing target ... */
1628 err = nfs_sillyrename(new_dir, new_dentry);
1629 if (!err) {
1630 new_dentry = rehash = dentry;
1631 new_inode = NULL;
1632 /* instantiate the replacement target */
1633 d_instantiate(new_dentry, NULL);
Jesper Juhl7a13e932007-04-26 00:29:02 -07001634 } else if (atomic_read(&new_dentry->d_count) > 1)
1635 /* dentry still busy? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 goto out;
Trond Myklebust20509f12005-08-25 16:25:34 -07001637 } else
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001638 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639
1640go_ahead:
1641 /*
1642 * ... prune child dentries and writebacks if needed.
1643 */
1644 if (atomic_read(&old_dentry->d_count) > 1) {
Trond Myklebuste1552e12007-04-14 19:07:28 -04001645 if (S_ISREG(old_inode->i_mode))
1646 nfs_wb_all(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 shrink_dcache_parent(old_dentry);
1648 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001649 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
Trond Myklebust24174112006-01-03 09:55:33 +01001651 if (new_inode != NULL) {
1652 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1657 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001658 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659out:
1660 if (rehash)
1661 d_rehash(rehash);
1662 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07001663 d_move(old_dentry, new_dentry);
Chuck Lever8fb559f2007-09-24 15:40:16 -04001664 nfs_set_verifier(new_dentry,
1665 nfs_save_change_attribute(new_dir));
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001666 } else if (error == -ENOENT)
1667 nfs_dentry_handle_enoent(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 /* new dentry created? */
1670 if (dentry)
1671 dput(dentry);
1672 unlock_kernel();
1673 return error;
1674}
1675
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001676static DEFINE_SPINLOCK(nfs_access_lru_lock);
1677static LIST_HEAD(nfs_access_lru_list);
1678static atomic_long_t nfs_access_nr_entries;
1679
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001680static void nfs_access_free_entry(struct nfs_access_entry *entry)
1681{
1682 put_rpccred(entry->cred);
1683 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001684 smp_mb__before_atomic_dec();
1685 atomic_long_dec(&nfs_access_nr_entries);
1686 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001687}
1688
Trond Myklebust979df722006-07-25 11:28:19 -04001689int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1690{
1691 LIST_HEAD(head);
1692 struct nfs_inode *nfsi;
1693 struct nfs_access_entry *cache;
1694
Trond Myklebust979df722006-07-25 11:28:19 -04001695restart:
Trond Myklebusta50f7952007-06-05 19:23:43 -04001696 spin_lock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001697 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
Trond Myklebust6f23e382008-01-25 16:38:17 -05001698 struct rw_semaphore *s_umount;
Trond Myklebust979df722006-07-25 11:28:19 -04001699 struct inode *inode;
1700
1701 if (nr_to_scan-- == 0)
1702 break;
Trond Myklebust6f23e382008-01-25 16:38:17 -05001703 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1704 if (!down_read_trylock(s_umount))
Trond Myklebust979df722006-07-25 11:28:19 -04001705 continue;
Trond Myklebust6f23e382008-01-25 16:38:17 -05001706 inode = igrab(&nfsi->vfs_inode);
1707 if (inode == NULL) {
1708 up_read(s_umount);
1709 continue;
1710 }
Trond Myklebust979df722006-07-25 11:28:19 -04001711 spin_lock(&inode->i_lock);
1712 if (list_empty(&nfsi->access_cache_entry_lru))
1713 goto remove_lru_entry;
1714 cache = list_entry(nfsi->access_cache_entry_lru.next,
1715 struct nfs_access_entry, lru);
1716 list_move(&cache->lru, &head);
1717 rb_erase(&cache->rb_node, &nfsi->access_cache);
1718 if (!list_empty(&nfsi->access_cache_entry_lru))
1719 list_move_tail(&nfsi->access_cache_inode_lru,
1720 &nfs_access_lru_list);
1721 else {
1722remove_lru_entry:
1723 list_del_init(&nfsi->access_cache_inode_lru);
1724 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1725 }
1726 spin_unlock(&inode->i_lock);
Trond Myklebusta50f7952007-06-05 19:23:43 -04001727 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001728 iput(inode);
Trond Myklebust6f23e382008-01-25 16:38:17 -05001729 up_read(s_umount);
Trond Myklebust979df722006-07-25 11:28:19 -04001730 goto restart;
1731 }
1732 spin_unlock(&nfs_access_lru_lock);
1733 while (!list_empty(&head)) {
1734 cache = list_entry(head.next, struct nfs_access_entry, lru);
1735 list_del(&cache->lru);
1736 nfs_access_free_entry(cache);
1737 }
1738 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1739}
1740
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001741static void __nfs_access_zap_cache(struct inode *inode)
1742{
1743 struct nfs_inode *nfsi = NFS_I(inode);
1744 struct rb_root *root_node = &nfsi->access_cache;
1745 struct rb_node *n, *dispose = NULL;
1746 struct nfs_access_entry *entry;
1747
1748 /* Unhook entries from the cache */
1749 while ((n = rb_first(root_node)) != NULL) {
1750 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1751 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001752 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001753 n->rb_left = dispose;
1754 dispose = n;
1755 }
1756 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1757 spin_unlock(&inode->i_lock);
1758
1759 /* Now kill them all! */
1760 while (dispose != NULL) {
1761 n = dispose;
1762 dispose = n->rb_left;
1763 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1764 }
1765}
1766
1767void nfs_access_zap_cache(struct inode *inode)
1768{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001769 /* Remove from global LRU init */
Benny Halevy3a10c302008-01-23 08:58:59 +02001770 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001771 spin_lock(&nfs_access_lru_lock);
1772 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1773 spin_unlock(&nfs_access_lru_lock);
1774 }
1775
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001776 spin_lock(&inode->i_lock);
1777 /* This will release the spinlock */
1778 __nfs_access_zap_cache(inode);
1779}
1780
1781static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1782{
1783 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1784 struct nfs_access_entry *entry;
1785
1786 while (n != NULL) {
1787 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1788
1789 if (cred < entry->cred)
1790 n = n->rb_left;
1791 else if (cred > entry->cred)
1792 n = n->rb_right;
1793 else
1794 return entry;
1795 }
1796 return NULL;
1797}
1798
Trond Myklebustaf22f942007-08-10 17:45:10 -04001799static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
Chuck Lever55296802005-08-18 11:24:09 -07001801 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001802 struct nfs_access_entry *cache;
1803 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001805 spin_lock(&inode->i_lock);
1806 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1807 goto out_zap;
1808 cache = nfs_access_search_rbtree(inode, cred);
1809 if (cache == NULL)
1810 goto out;
Trond Myklebustc7c20972007-09-28 19:22:40 -04001811 if (!time_in_range(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001812 goto out_stale;
1813 res->jiffies = cache->jiffies;
1814 res->cred = cache->cred;
1815 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001816 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001817 err = 0;
1818out:
1819 spin_unlock(&inode->i_lock);
1820 return err;
1821out_stale:
1822 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001823 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001824 spin_unlock(&inode->i_lock);
1825 nfs_access_free_entry(cache);
1826 return -ENOENT;
1827out_zap:
1828 /* This will release the spinlock */
1829 __nfs_access_zap_cache(inode);
1830 return -ENOENT;
1831}
1832
1833static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1834{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001835 struct nfs_inode *nfsi = NFS_I(inode);
1836 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001837 struct rb_node **p = &root_node->rb_node;
1838 struct rb_node *parent = NULL;
1839 struct nfs_access_entry *entry;
1840
1841 spin_lock(&inode->i_lock);
1842 while (*p != NULL) {
1843 parent = *p;
1844 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1845
1846 if (set->cred < entry->cred)
1847 p = &parent->rb_left;
1848 else if (set->cred > entry->cred)
1849 p = &parent->rb_right;
1850 else
1851 goto found;
1852 }
1853 rb_link_node(&set->rb_node, parent, p);
1854 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001855 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001856 spin_unlock(&inode->i_lock);
1857 return;
1858found:
1859 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001860 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1861 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001862 spin_unlock(&inode->i_lock);
1863 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864}
1865
Trond Myklebustaf22f942007-08-10 17:45:10 -04001866static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001868 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1869 if (cache == NULL)
1870 return;
1871 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001873 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001875
1876 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001877
1878 /* Update accounting */
1879 smp_mb__before_atomic_inc();
1880 atomic_long_inc(&nfs_access_nr_entries);
1881 smp_mb__after_atomic_inc();
1882
1883 /* Add inode to global LRU list */
Benny Halevy3a10c302008-01-23 08:58:59 +02001884 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001885 spin_lock(&nfs_access_lru_lock);
1886 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1887 spin_unlock(&nfs_access_lru_lock);
1888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889}
1890
1891static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1892{
1893 struct nfs_access_entry cache;
1894 int status;
1895
1896 status = nfs_access_get_cached(inode, cred, &cache);
1897 if (status == 0)
1898 goto out;
1899
1900 /* Be clever: ask server to check for all possible rights */
1901 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1902 cache.cred = cred;
1903 cache.jiffies = jiffies;
1904 status = NFS_PROTO(inode)->access(inode, &cache);
1905 if (status != 0)
1906 return status;
1907 nfs_access_add_cache(inode, &cache);
1908out:
1909 if ((cache.mask & mask) == mask)
1910 return 0;
1911 return -EACCES;
1912}
1913
Trond Myklebustaf22f942007-08-10 17:45:10 -04001914static int nfs_open_permission_mask(int openflags)
1915{
1916 int mask = 0;
1917
1918 if (openflags & FMODE_READ)
1919 mask |= MAY_READ;
1920 if (openflags & FMODE_WRITE)
1921 mask |= MAY_WRITE;
1922 if (openflags & FMODE_EXEC)
1923 mask |= MAY_EXEC;
1924 return mask;
1925}
1926
1927int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1928{
1929 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1930}
1931
Linus Torvalds1da177e2005-04-16 15:20:36 -07001932int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1933{
1934 struct rpc_cred *cred;
1935 int res = 0;
1936
Chuck Lever91d5b472006-03-20 13:44:14 -05001937 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1938
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 if (mask == 0)
1940 goto out;
1941 /* Is this sys_access() ? */
1942 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1943 goto force_lookup;
1944
1945 switch (inode->i_mode & S_IFMT) {
1946 case S_IFLNK:
1947 goto out;
1948 case S_IFREG:
1949 /* NFSv4 has atomic_open... */
1950 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1951 && nd != NULL
1952 && (nd->flags & LOOKUP_OPEN))
1953 goto out;
1954 break;
1955 case S_IFDIR:
1956 /*
1957 * Optimize away all write operations, since the server
1958 * will check permissions when we perform the op.
1959 */
1960 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1961 goto out;
1962 }
1963
1964force_lookup:
1965 lock_kernel();
1966
1967 if (!NFS_PROTO(inode)->access)
1968 goto out_notsup;
1969
Trond Myklebust98a8e322008-03-12 12:25:28 -04001970 cred = rpc_lookup_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 if (!IS_ERR(cred)) {
1972 res = nfs_do_access(inode, cred, mask);
1973 put_rpccred(cred);
1974 } else
1975 res = PTR_ERR(cred);
1976 unlock_kernel();
1977out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001978 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1979 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 return res;
1981out_notsup:
1982 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1983 if (res == 0)
1984 res = generic_permission(inode, mask, NULL);
1985 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001986 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987}
1988
1989/*
1990 * Local variables:
1991 * version-control: t
1992 * kept-new-versions: 5
1993 * End:
1994 */