blob: 32062c33c859496198395db435117c7c33105d3d [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>
Chuck Lever873101b2006-08-22 20:06:23 -040032#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/namei.h>
David Howells54ceac42006-08-22 20:06:13 -040034#include <linux/mount.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
Trond Myklebust4ce79712005-06-22 17:16:21 +000037#include "nfs4_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "delegation.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050039#include "iostat.h"
Adrian Bunk4c30d562007-11-21 15:04:31 -080040#include "internal.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 Lever6da24bc2008-06-11 17:55:58 -0400135 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400136 filp->f_path.dentry->d_parent->d_name.name,
137 filp->f_path.dentry->d_name.name);
138
139 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 /* Call generic open code in order to cache credentials */
Carsten Otte7451c4f2006-04-19 13:06:37 -0400142 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 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;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400158 unsigned long gencount;
Neil Brown1f4eab72007-04-16 09:35:27 +1000159 int timestamp_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160} nfs_readdir_descriptor_t;
161
162/* Now we cache directories properly, by stuffing the dirent
163 * data directly in the page cache.
164 *
165 * Inode invalidation due to refresh etc. takes care of
166 * _everything_, no sloppy entry flushing logic, no extraneous
167 * copying, network direct to page cache, the way it was meant
168 * to be.
169 *
170 * NOTE: Dirent information verification is done always by the
171 * page-in of the RPC reply, nowhere else, this simplies
172 * things substantially.
173 */
174static
175int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176{
177 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800178 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 struct rpc_cred *cred = nfs_file_cred(file);
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400180 unsigned long timestamp, gencount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 int error;
182
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500183 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700184 __func__, (long long)desc->entry->cookie,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500185 page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 again:
188 timestamp = jiffies;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400189 gencount = nfs_inc_attr_generation_counter();
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800190 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 NFS_SERVER(inode)->dtsize, desc->plus);
192 if (error < 0) {
193 /* We requested READDIRPLUS, but the server doesn't grok it */
194 if (error == -ENOTSUPP && desc->plus) {
195 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Benny Halevy3a10c302008-01-23 08:58:59 +0200196 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 desc->plus = 0;
198 goto again;
199 }
200 goto error;
201 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000202 desc->timestamp = timestamp;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400203 desc->gencount = gencount;
Neil Brown1f4eab72007-04-16 09:35:27 +1000204 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 SetPageUptodate(page);
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 Myklebust2aac05a2008-07-07 13:26:10 -0400210 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700211 /* 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:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 return -EIO;
219}
220
221static inline
222int dir_decode(nfs_readdir_descriptor_t *desc)
223{
Al Viro0dbb4c62006-10-19 23:28:49 -0700224 __be32 *p = desc->ptr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 p = desc->decode(p, desc->entry, desc->plus);
226 if (IS_ERR(p))
227 return PTR_ERR(p);
228 desc->ptr = p;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400229 if (desc->timestamp_valid) {
Neil Brown1f4eab72007-04-16 09:35:27 +1000230 desc->entry->fattr->time_start = desc->timestamp;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400231 desc->entry->fattr->gencount = desc->gencount;
232 } else
Neil Brown1f4eab72007-04-16 09:35:27 +1000233 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 return 0;
235}
236
237static inline
238void dir_page_release(nfs_readdir_descriptor_t *desc)
239{
240 kunmap(desc->page);
241 page_cache_release(desc->page);
242 desc->page = NULL;
243 desc->ptr = NULL;
244}
245
246/*
247 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000248 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 *
250 * If the end of the buffer has been reached, return -EAGAIN, if not,
251 * return the offset within the buffer of the next entry to be
252 * read.
253 */
254static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000255int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256{
257 struct nfs_entry *entry = desc->entry;
258 int loop_count = 0,
259 status;
260
261 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500262 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700263 __func__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000264 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 break;
266 if (loop_count++ > 200) {
267 loop_count = 0;
268 schedule();
269 }
270 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 return status;
272}
273
274/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000275 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000276 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000277 *
278 * If the end of the buffer has been reached, return -EAGAIN, if not,
279 * return the offset within the buffer of the next entry to be
280 * read.
281 */
282static inline
283int find_dirent_index(nfs_readdir_descriptor_t *desc)
284{
285 struct nfs_entry *entry = desc->entry;
286 int loop_count = 0,
287 status;
288
289 for(;;) {
290 status = dir_decode(desc);
291 if (status)
292 break;
293
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500294 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
295 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000296
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000297 if (desc->file->f_pos == desc->current_index) {
298 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000299 break;
300 }
301 desc->current_index++;
302 if (loop_count++ > 200) {
303 loop_count = 0;
304 schedule();
305 }
306 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000307 return status;
308}
309
310/*
311 * Find the given page, and call find_dirent() or find_dirent_index in
312 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 */
314static inline
315int find_dirent_page(nfs_readdir_descriptor_t *desc)
316{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800317 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 struct page *page;
319 int status;
320
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500321 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700322 __func__, desc->page_index,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500323 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
Neil Brown1f4eab72007-04-16 09:35:27 +1000325 /* If we find the page in the page_cache, we cannot be sure
326 * how fresh the data is, so we will ignore readdir_plus attributes.
327 */
328 desc->timestamp_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 page = read_cache_page(inode->i_mapping, desc->page_index,
330 (filler_t *)nfs_readdir_filler, desc);
331 if (IS_ERR(page)) {
332 status = PTR_ERR(page);
333 goto out;
334 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335
336 /* NOTE: Someone else may have changed the READDIRPLUS flag */
337 desc->page = page;
338 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000339 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000340 status = find_dirent(desc);
341 else
342 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 if (status < 0)
344 dir_page_release(desc);
345 out:
Harvey Harrison3110ff82008-05-02 13:42:44 -0700346 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347 return status;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
350/*
351 * Recurse through the page cache pages, and return a
352 * filled nfs_entry structure of the next directory entry if possible.
353 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000354 * The target for the search is '*desc->dir_cookie' if non-0,
355 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 */
357static inline
358int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
359{
360 int loop_count = 0;
361 int res;
362
Olivier Galibert00a92642005-06-22 17:16:29 +0000363 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000364 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500365 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
366 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000367 desc->page_index = 0;
368 desc->entry->cookie = desc->entry->prev_cookie = 0;
369 desc->entry->eof = 0;
370 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000371 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500372 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
373 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000374
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 for (;;) {
376 res = find_dirent_page(desc);
377 if (res != -EAGAIN)
378 break;
379 /* Align to beginning of next page */
380 desc->page_index ++;
381 if (loop_count++ > 200) {
382 loop_count = 0;
383 schedule();
384 }
385 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500386
Harvey Harrison3110ff82008-05-02 13:42:44 -0700387 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 return res;
389}
390
391static inline unsigned int dt_type(struct inode *inode)
392{
393 return (inode->i_mode >> 12) & 15;
394}
395
396static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
397
398/*
399 * Once we've found the start of the dirent within a page: fill 'er up...
400 */
401static
402int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
403 filldir_t filldir)
404{
405 struct file *file = desc->file;
406 struct nfs_entry *entry = desc->entry;
407 struct dentry *dentry = NULL;
Peter Staubach4e769b92007-08-03 15:07:10 -0400408 u64 fileid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 int loop_count = 0,
410 res;
411
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500412 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
413 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
415 for(;;) {
416 unsigned d_type = DT_UNKNOWN;
417 /* Note: entry->prev_cookie contains the cookie for
418 * retrieving the current dirent on the server */
Peter Staubach4e769b92007-08-03 15:07:10 -0400419 fileid = entry->ino;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420
421 /* Get a dentry if we have one */
422 if (dentry != NULL)
423 dput(dentry);
424 dentry = nfs_readdir_lookup(desc);
425
426 /* Use readdirplus info */
427 if (dentry != NULL && dentry->d_inode != NULL) {
428 d_type = dt_type(dentry->d_inode);
Peter Staubach4e769b92007-08-03 15:07:10 -0400429 fileid = NFS_FILEID(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 }
431
432 res = filldir(dirent, entry->name, entry->len,
Trond Myklebustf43bf0b2007-10-09 12:01:04 -0400433 file->f_pos, nfs_compat_user_ino64(fileid),
434 d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 if (res < 0)
436 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000437 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000438 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 if (dir_decode(desc) != 0) {
440 desc->page_index ++;
441 break;
442 }
443 if (loop_count++ > 200) {
444 loop_count = 0;
445 schedule();
446 }
447 }
448 dir_page_release(desc);
449 if (dentry != NULL)
450 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500451 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
452 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 return res;
454}
455
456/*
457 * If we cannot find a cookie in our cache, we suspect that this is
458 * because it points to a deleted file, so we ask the server to return
459 * whatever it thinks is the next entry. We then feed this to filldir.
460 * If all goes well, we should then be able to find our way round the
461 * cache on the next call to readdir_search_pagecache();
462 *
463 * NOTE: we cannot add the anonymous page to the pagecache because
464 * the data it contains might not be page aligned. Besides,
465 * we should already have a complete representation of the
466 * directory in the page cache by the time we get here.
467 */
468static inline
469int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
470 filldir_t filldir)
471{
472 struct file *file = desc->file;
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800473 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 struct rpc_cred *cred = nfs_file_cred(file);
475 struct page *page = NULL;
476 int status;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400477 unsigned long timestamp, gencount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500479 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
480 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 page = alloc_page(GFP_HIGHUSER);
483 if (!page) {
484 status = -ENOMEM;
485 goto out;
486 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000487 timestamp = jiffies;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400488 gencount = nfs_inc_attr_generation_counter();
Jeff Layton25606652008-02-12 06:49:01 -0500489 status = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred,
490 *desc->dir_cookie, page,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 NFS_SERVER(inode)->dtsize,
492 desc->plus);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 desc->page = page;
494 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Jeff Layton25606652008-02-12 06:49:01 -0500495 if (status >= 0) {
Neil Brown1f4eab72007-04-16 09:35:27 +1000496 desc->timestamp = timestamp;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400497 desc->gencount = gencount;
Neil Brown1f4eab72007-04-16 09:35:27 +1000498 desc->timestamp_valid = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000500 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 } else
502 status = -EIO;
503 if (status < 0)
504 goto out_release;
505
506 status = nfs_do_filldir(desc, dirent, filldir);
507
508 /* Reset read descriptor so it searches the page cache from
509 * the start upon the next call to readdir_search_pagecache() */
510 desc->page_index = 0;
511 desc->entry->cookie = desc->entry->prev_cookie = 0;
512 desc->entry->eof = 0;
513 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500514 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700515 __func__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 return status;
517 out_release:
518 dir_page_release(desc);
519 goto out;
520}
521
Olivier Galibert00a92642005-06-22 17:16:29 +0000522/* The file offset position represents the dirent entry number. A
523 last cookie cache takes care of the common case of reading the
524 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525 */
526static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
527{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800528 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 struct inode *inode = dentry->d_inode;
530 nfs_readdir_descriptor_t my_desc,
531 *desc = &my_desc;
532 struct nfs_entry my_entry;
533 struct nfs_fh fh;
534 struct nfs_fattr fattr;
535 long res;
536
Chuck Lever6da24bc2008-06-11 17:55:58 -0400537 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500538 dentry->d_parent->d_name.name, dentry->d_name.name,
539 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500540 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000543 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000544 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000545 * to either find the entry with the appropriate number or
546 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 */
548 memset(desc, 0, sizeof(*desc));
549
550 desc->file = filp;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400551 desc->dir_cookie = &nfs_file_open_context(filp)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 desc->decode = NFS_PROTO(inode)->decode_dirent;
553 desc->plus = NFS_USE_READDIRPLUS(inode);
554
555 my_entry.cookie = my_entry.prev_cookie = 0;
556 my_entry.eof = 0;
557 my_entry.fh = &fh;
558 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400559 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 desc->entry = &my_entry;
561
Trond Myklebust565277f2007-10-15 18:17:53 -0400562 nfs_block_sillyrename(dentry);
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500563 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
564 if (res < 0)
565 goto out;
566
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 while(!desc->entry->eof) {
568 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 if (res == -EBADCOOKIE) {
571 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000572 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 /* Or that the server has 'lost' a cookie */
574 res = uncached_readdir(desc, dirent, filldir);
575 if (res >= 0)
576 continue;
577 }
578 res = 0;
579 break;
580 }
581 if (res == -ETOOSMALL && desc->plus) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200582 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 nfs_zap_caches(inode);
584 desc->plus = 0;
585 desc->entry->eof = 0;
586 continue;
587 }
588 if (res < 0)
589 break;
590
591 res = nfs_do_filldir(desc, dirent, filldir);
592 if (res < 0) {
593 res = 0;
594 break;
595 }
596 }
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500597out:
Trond Myklebust565277f2007-10-15 18:17:53 -0400598 nfs_unblock_sillyrename(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500599 if (res > 0)
600 res = 0;
Chuck Lever6da24bc2008-06-11 17:55:58 -0400601 dfprintk(FILE, "NFS: readdir(%s/%s) returns %ld\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500602 dentry->d_parent->d_name.name, dentry->d_name.name,
603 res);
604 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
Trond Myklebust10afec92007-05-14 17:16:04 -0400607static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000608{
Chuck Leverb84e06c2008-06-11 17:55:34 -0400609 struct dentry *dentry = filp->f_path.dentry;
610 struct inode *inode = dentry->d_inode;
611
Chuck Lever6da24bc2008-06-11 17:55:58 -0400612 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400613 dentry->d_parent->d_name.name,
614 dentry->d_name.name,
615 offset, origin);
616
617 mutex_lock(&inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000618 switch (origin) {
619 case 1:
620 offset += filp->f_pos;
621 case 0:
622 if (offset >= 0)
623 break;
624 default:
625 offset = -EINVAL;
626 goto out;
627 }
628 if (offset != filp->f_pos) {
629 filp->f_pos = offset;
Trond Myklebustcd3758e2007-08-10 17:44:32 -0400630 nfs_file_open_context(filp)->dir_cookie = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000631 }
632out:
Chuck Leverb84e06c2008-06-11 17:55:34 -0400633 mutex_unlock(&inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000634 return offset;
635}
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637/*
638 * All directory operations under NFS are synchronous, so fsync()
639 * is a dummy operation.
640 */
Trond Myklebust10afec92007-05-14 17:16:04 -0400641static int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642{
Chuck Lever6da24bc2008-06-11 17:55:58 -0400643 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500644 dentry->d_parent->d_name.name, dentry->d_name.name,
645 datasync);
646
Chuck Lever54917782008-05-27 16:29:07 -0400647 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 return 0;
649}
650
Trond Myklebustbfc69a42007-10-15 18:18:29 -0400651/**
652 * nfs_force_lookup_revalidate - Mark the directory as having changed
653 * @dir - pointer to directory inode
654 *
655 * This forces the revalidation code in nfs_lookup_revalidate() to do a
656 * full lookup on all child dentries of 'dir' whenever a change occurs
657 * on the server that might have invalidated our dcache.
658 *
659 * The caller should be holding dir->i_lock
660 */
661void nfs_force_lookup_revalidate(struct inode *dir)
662{
Trond Myklebust011935a02008-10-14 19:24:50 -0400663 NFS_I(dir)->cache_change_attribute++;
Trond Myklebustbfc69a42007-10-15 18:18:29 -0400664}
665
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666/*
667 * A check for whether or not the parent directory has changed.
668 * In the case it has, we assume that the dentries are untrustworthy
669 * and may need to be looked up again.
670 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500671static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672{
673 if (IS_ROOT(dentry))
674 return 1;
Trond Myklebust4eec9522008-07-15 17:58:13 -0400675 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
676 return 0;
Trond Myklebustf2c77f42007-10-02 12:54:39 -0400677 if (!nfs_verify_change_attribute(dir, dentry->d_time))
678 return 0;
679 /* Revalidate nfsi->cache_change_attribute before we declare a match */
680 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
681 return 0;
682 if (!nfs_verify_change_attribute(dir, dentry->d_time))
683 return 0;
684 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685}
686
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400688 * Return the intent data that applies to this particular path component
689 *
690 * Note that the current set of intents only apply to the very last
691 * component of the path.
692 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
693 */
694static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
695{
696 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
697 return 0;
698 return nd->flags & mask;
699}
700
701/*
Trond Myklebusta12802c2007-10-02 19:13:04 -0400702 * Use intent information to check whether or not we're going to do
703 * an O_EXCL create using this path component.
704 */
705static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
706{
707 if (NFS_PROTO(dir)->version == 2)
708 return 0;
Al Viro35165862008-08-05 03:00:49 -0400709 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
Trond Myklebusta12802c2007-10-02 19:13:04 -0400710}
711
712/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400713 * Inode and filehandle revalidation for lookups.
714 *
715 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
716 * or if the intent information indicates that we're about to open this
717 * particular file and the "nocto" mount flag is not set.
718 *
719 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720static inline
721int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
722{
723 struct nfs_server *server = NFS_SERVER(inode);
724
Trond Myklebust4e99a1f2008-03-06 12:34:59 -0500725 if (test_bit(NFS_INO_MOUNTPOINT, &NFS_I(inode)->flags))
726 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400729 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 goto out_force;
731 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400732 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400733 !(server->flags & NFS_MOUNT_NOCTO) &&
734 (S_ISREG(inode->i_mode) ||
735 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 goto out_force;
Trond Myklebust4f48af42007-10-02 23:13:32 -0400737 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
739 return nfs_revalidate_inode(server, inode);
740out_force:
741 return __nfs_revalidate_inode(server, inode);
742}
743
744/*
745 * We judge how long we want to trust negative
746 * dentries by looking at the parent inode mtime.
747 *
748 * If parent mtime has changed, we revalidate, else we wait for a
749 * period corresponding to the parent's attribute cache timeout value.
750 */
751static inline
752int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
753 struct nameidata *nd)
754{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400756 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 return 0;
Trond Myklebust4eec9522008-07-15 17:58:13 -0400758 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
759 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 return !nfs_check_verifier(dir, dentry);
761}
762
763/*
764 * This is called every time the dcache has a lookup hit,
765 * and we should check whether we can really trust that
766 * lookup.
767 *
768 * NOTE! The hit can be a negative hit too, don't assume
769 * we have an inode!
770 *
771 * If the parent directory is seen to have changed, we throw out the
772 * cached dentry and do a new lookup.
773 */
774static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
775{
776 struct inode *dir;
777 struct inode *inode;
778 struct dentry *parent;
779 int error;
780 struct nfs_fh fhandle;
781 struct nfs_fattr fattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 parent = dget_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500785 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 inode = dentry->d_inode;
787
788 if (!inode) {
789 if (nfs_neg_need_reval(dir, dentry, nd))
790 goto out_bad;
791 goto out_valid;
792 }
793
794 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500795 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700796 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500797 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 goto out_bad;
799 }
800
Trond Myklebust15860ab2008-12-23 15:21:54 -0500801 if (nfs_have_delegation(inode, FMODE_READ))
802 goto out_set_verifier;
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 /* Force a full look up iff the parent directory has changed */
Trond Myklebusta12802c2007-10-02 19:13:04 -0400805 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 if (nfs_lookup_verify_inode(inode, nd))
807 goto out_zap_parent;
808 goto out_valid;
809 }
810
811 if (NFS_STALE(inode))
812 goto out_bad;
813
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
815 if (error)
816 goto out_bad;
817 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
818 goto out_bad;
819 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
820 goto out_bad;
821
Trond Myklebust15860ab2008-12-23 15:21:54 -0500822out_set_verifier:
Trond Myklebustcf8ba452007-10-01 13:46:53 -0400823 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700824 out_valid:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500826 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700827 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500828 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 return 1;
830out_zap_parent:
831 nfs_zap_caches(dir);
832 out_bad:
Trond Myklebusta1643a92007-09-29 17:25:43 -0400833 nfs_mark_for_revalidate(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 if (inode && S_ISDIR(inode->i_mode)) {
835 /* Purge readdir caches. */
836 nfs_zap_caches(inode);
837 /* If we have submounts, don't unhash ! */
838 if (have_submounts(dentry))
839 goto out_valid;
840 shrink_dcache_parent(dentry);
841 }
842 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500844 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700845 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500846 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 return 0;
848}
849
850/*
851 * This is called from dput() when d_count is going to 0.
852 */
853static int nfs_dentry_delete(struct dentry *dentry)
854{
855 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
856 dentry->d_parent->d_name.name, dentry->d_name.name,
857 dentry->d_flags);
858
Trond Myklebust77f11192008-01-28 19:43:19 -0500859 /* Unhash any dentry with a stale inode */
860 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
861 return 1;
862
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
864 /* Unhash it, so that ->d_iput() would be called */
865 return 1;
866 }
867 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
868 /* Unhash it, so that ancestors of killed async unlink
869 * files will be cleaned up during umount */
870 return 1;
871 }
872 return 0;
873
874}
875
Trond Myklebust1b83d702008-06-11 15:44:04 -0400876static void nfs_drop_nlink(struct inode *inode)
877{
878 spin_lock(&inode->i_lock);
879 if (inode->i_nlink > 0)
880 drop_nlink(inode);
881 spin_unlock(&inode->i_lock);
882}
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884/*
885 * Called when the dentry loses inode.
886 * We use it to clean up silly-renamed files.
887 */
888static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
889{
Neil Brown83672d32007-02-26 12:48:25 +1100890 if (S_ISDIR(inode->i_mode))
891 /* drop any readdir cache as it could easily be old */
892 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
893
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700895 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -0400896 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898 iput(inode);
899}
900
Al Virof786aa92009-02-20 05:51:22 +0000901const struct dentry_operations nfs_dentry_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 .d_revalidate = nfs_lookup_revalidate,
903 .d_delete = nfs_dentry_delete,
904 .d_iput = nfs_dentry_iput,
905};
906
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
908{
909 struct dentry *res;
Trond Myklebust565277f2007-10-15 18:17:53 -0400910 struct dentry *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 struct inode *inode = NULL;
912 int error;
913 struct nfs_fh fhandle;
914 struct nfs_fattr fattr;
915
916 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
917 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500918 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
920 res = ERR_PTR(-ENAMETOOLONG);
921 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
922 goto out;
923
924 res = ERR_PTR(-ENOMEM);
925 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
926
Trond Myklebustfd684072006-09-05 12:27:44 -0400927 /*
928 * If we're doing an exclusive create, optimize away the lookup
929 * but don't hash the dentry.
930 */
931 if (nfs_is_exclusive_create(dir, nd)) {
932 d_instantiate(dentry, NULL);
933 res = NULL;
Trond Myklebustfc0f6842008-06-11 15:44:20 -0400934 goto out;
Trond Myklebustfd684072006-09-05 12:27:44 -0400935 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936
Trond Myklebust565277f2007-10-15 18:17:53 -0400937 parent = dentry->d_parent;
938 /* Protect against concurrent sillydeletes */
939 nfs_block_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
941 if (error == -ENOENT)
942 goto no_entry;
943 if (error < 0) {
944 res = ERR_PTR(error);
Trond Myklebust565277f2007-10-15 18:17:53 -0400945 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500948 res = (struct dentry *)inode;
949 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400950 goto out_unblock_sillyrename;
David Howells54ceac42006-08-22 20:06:13 -0400951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400953 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -0700954 if (res != NULL) {
955 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -0400956 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -0700958 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Trond Myklebust565277f2007-10-15 18:17:53 -0400960out_unblock_sillyrename:
961 nfs_unblock_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962out:
963 return res;
964}
965
966#ifdef CONFIG_NFS_V4
967static int nfs_open_revalidate(struct dentry *, struct nameidata *);
968
Al Virof786aa92009-02-20 05:51:22 +0000969const struct dentry_operations nfs4_dentry_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970 .d_revalidate = nfs_open_revalidate,
971 .d_delete = nfs_dentry_delete,
972 .d_iput = nfs_dentry_iput,
973};
974
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400975/*
976 * Use intent information to determine whether we need to substitute
977 * the NFSv4-style stateful OPEN for the LOOKUP call
978 */
Trond Myklebust5584c302008-12-23 15:21:54 -0500979static int is_atomic_open(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400981 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 return 0;
983 /* NFS does not (yet) have a stateful open for directories */
984 if (nd->flags & LOOKUP_DIRECTORY)
985 return 0;
986 /* Are we trying to write to a read only partition? */
Dave Hansen2c463e92008-02-15 14:37:56 -0800987 if (__mnt_is_readonly(nd->path.mnt) &&
988 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989 return 0;
990 return 1;
991}
992
993static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
994{
995 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 int error;
997
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500998 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
999 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1000
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 /* Check that we are indeed trying to open this file */
Trond Myklebust5584c302008-12-23 15:21:54 -05001002 if (!is_atomic_open(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 goto no_open;
1004
1005 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1006 res = ERR_PTR(-ENAMETOOLONG);
1007 goto out;
1008 }
1009 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1010
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -04001011 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1012 * the dentry. */
Al Viro35165862008-08-05 03:00:49 -04001013 if (nd->flags & LOOKUP_EXCL) {
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -04001014 d_instantiate(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001015 goto out;
1016 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 /* Open the file on the server */
Trond Myklebust60ccd4e2007-09-29 17:48:19 -04001019 res = nfs4_atomic_open(dir, dentry, nd);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001020 if (IS_ERR(res)) {
1021 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 switch (error) {
1023 /* Make a negative dentry */
1024 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001025 res = NULL;
1026 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001028 case -ENOTDIR:
1029 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 case -ELOOP:
1031 if (!(nd->intent.open.flags & O_NOFOLLOW))
1032 goto no_open;
Trond Myklebustd9531262009-07-21 19:22:38 -04001033 /* case -EISDIR: */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 /* case -EINVAL: */
1035 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 goto out;
1037 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001038 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 dentry = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040out:
1041 return res;
1042no_open:
1043 return nfs_lookup(dir, dentry, nd);
1044}
1045
1046static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1047{
1048 struct dentry *parent = NULL;
1049 struct inode *inode = dentry->d_inode;
1050 struct inode *dir;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 int openflags, ret = 0;
1052
Trond Myklebust5584c302008-12-23 15:21:54 -05001053 if (!is_atomic_open(nd))
1054 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 parent = dget_parent(dentry);
1056 dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 /* We can't create new files in nfs_open_revalidate(), so we
1058 * optimize away revalidation of negative dentries.
1059 */
Trond Myklebust216d5d02007-10-01 20:10:12 -04001060 if (inode == NULL) {
1061 if (!nfs_neg_need_reval(dir, dentry, nd))
1062 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 goto out;
Trond Myklebust216d5d02007-10-01 20:10:12 -04001064 }
1065
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 /* NFS only supports OPEN on regular files */
1067 if (!S_ISREG(inode->i_mode))
Trond Myklebust5584c302008-12-23 15:21:54 -05001068 goto no_open_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 openflags = nd->intent.open.flags;
1070 /* We cannot do exclusive creation on a positive dentry */
1071 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
Trond Myklebust5584c302008-12-23 15:21:54 -05001072 goto no_open_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 /* We can't create new files, or truncate existing ones here */
1074 openflags &= ~(O_CREAT|O_TRUNC);
1075
1076 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001077 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 * operations that change the directory. We therefore save the
1079 * change attribute *before* we do the RPC call.
1080 */
Trond Myklebust02a913a2005-10-18 14:20:17 -07001081 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082out:
1083 dput(parent);
1084 if (!ret)
1085 d_drop(dentry);
1086 return ret;
Trond Myklebust5584c302008-12-23 15:21:54 -05001087no_open_dput:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 dput(parent);
Trond Myklebust5584c302008-12-23 15:21:54 -05001089no_open:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 return nfs_lookup_revalidate(dentry, nd);
1091}
1092#endif /* CONFIG_NFSV4 */
1093
1094static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1095{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -08001096 struct dentry *parent = desc->file->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 struct inode *dir = parent->d_inode;
1098 struct nfs_entry *entry = desc->entry;
1099 struct dentry *dentry, *alias;
1100 struct qstr name = {
1101 .name = entry->name,
1102 .len = entry->len,
1103 };
1104 struct inode *inode;
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001105 unsigned long verf = nfs_save_change_attribute(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 switch (name.len) {
1108 case 2:
1109 if (name.name[0] == '.' && name.name[1] == '.')
1110 return dget_parent(parent);
1111 break;
1112 case 1:
1113 if (name.name[0] == '.')
1114 return dget(parent);
1115 }
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001116
1117 spin_lock(&dir->i_lock);
1118 if (NFS_I(dir)->cache_validity & NFS_INO_INVALID_DATA) {
1119 spin_unlock(&dir->i_lock);
1120 return NULL;
1121 }
1122 spin_unlock(&dir->i_lock);
1123
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 name.hash = full_name_hash(name.name, name.len);
1125 dentry = d_lookup(parent, &name);
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001126 if (dentry != NULL) {
Trond Myklebustef75c792007-01-16 10:09:44 -05001127 /* Is this a positive dentry that matches the readdir info? */
1128 if (dentry->d_inode != NULL &&
1129 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1130 d_mountpoint(dentry))) {
1131 if (!desc->plus || entry->fh->size == 0)
1132 return dentry;
1133 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1134 entry->fh) == 0)
1135 goto out_renew;
1136 }
Trond Myklebustdf1d5d22007-01-15 13:56:29 -05001137 /* No, so d_drop to allow one to be created */
1138 d_drop(dentry);
1139 dput(dentry);
1140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1142 return NULL;
Trond Myklebust54af3bb2007-09-28 12:27:41 -04001143 if (name.len > NFS_SERVER(dir)->namelen)
1144 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001145 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 dentry = d_alloc(parent, &name);
1147 if (dentry == NULL)
1148 return NULL;
1149 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1150 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001151 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001152 dput(dentry);
1153 return NULL;
1154 }
David Howells54ceac42006-08-22 20:06:13 -04001155
1156 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (alias != NULL) {
1158 dput(dentry);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001159 if (IS_ERR(alias))
1160 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 dentry = alias;
1162 }
David Howells54ceac42006-08-22 20:06:13 -04001163
Trond Myklebustc79ba782007-01-31 08:16:24 -05001164out_renew:
Trond Myklebust57fa76f2007-09-30 18:01:13 -04001165 nfs_set_verifier(dentry, verf);
Trond Myklebustc79ba782007-01-31 08:16:24 -05001166 return dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167}
1168
1169/*
1170 * Code common to create, mkdir, and mknod.
1171 */
1172int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1173 struct nfs_fattr *fattr)
1174{
Trond Myklebustfab728e2007-09-29 17:41:33 -04001175 struct dentry *parent = dget_parent(dentry);
1176 struct inode *dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177 struct inode *inode;
1178 int error = -EACCES;
1179
Trond Myklebustfab728e2007-09-29 17:41:33 -04001180 d_drop(dentry);
1181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /* We may have been initialized further down */
1183 if (dentry->d_inode)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001184 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185 if (fhandle->size == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1187 if (error)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001188 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 }
Trond Myklebust5724ab32007-10-01 21:51:38 -04001190 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1192 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001193 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 if (error < 0)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001195 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001198 error = PTR_ERR(inode);
1199 if (IS_ERR(inode))
Trond Myklebustfab728e2007-09-29 17:41:33 -04001200 goto out_error;
1201 d_add(dentry, inode);
1202out:
1203 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 return 0;
Trond Myklebustfab728e2007-09-29 17:41:33 -04001205out_error:
1206 nfs_mark_for_revalidate(dir);
1207 dput(parent);
1208 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209}
1210
1211/*
1212 * Following a failed create operation, we drop the dentry rather
1213 * than retain a negative dentry. This avoids a problem in the event
1214 * that the operation succeeded on the server, but an error in the
1215 * reply path made it appear to have failed.
1216 */
1217static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1218 struct nameidata *nd)
1219{
1220 struct iattr attr;
1221 int error;
1222 int open_flags = 0;
1223
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001224 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1225 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 attr.ia_mode = mode;
1228 attr.ia_valid = ATTR_MODE;
1229
Trond Myklebustad389da2007-06-05 12:30:00 -04001230 if ((nd->flags & LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 open_flags = nd->intent.open.flags;
1232
Trond Myklebust02a913a2005-10-18 14:20:17 -07001233 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 if (error != 0)
1235 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236 return 0;
1237out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 d_drop(dentry);
1239 return error;
1240}
1241
1242/*
1243 * See comments for nfs_proc_create regarding failed operations.
1244 */
1245static int
1246nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1247{
1248 struct iattr attr;
1249 int status;
1250
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001251 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1252 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 if (!new_valid_dev(rdev))
1255 return -EINVAL;
1256
1257 attr.ia_mode = mode;
1258 attr.ia_valid = ATTR_MODE;
1259
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (status != 0)
1262 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 return 0;
1264out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265 d_drop(dentry);
1266 return status;
1267}
1268
1269/*
1270 * See comments for nfs_proc_create regarding failed operations.
1271 */
1272static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1273{
1274 struct iattr attr;
1275 int error;
1276
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001277 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1278 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 attr.ia_valid = ATTR_MODE;
1281 attr.ia_mode = mode | S_IFDIR;
1282
Linus Torvalds1da177e2005-04-16 15:20:36 -07001283 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 if (error != 0)
1285 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 return 0;
1287out_err:
1288 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 return error;
1290}
1291
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001292static void nfs_dentry_handle_enoent(struct dentry *dentry)
1293{
1294 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1295 d_delete(dentry);
1296}
1297
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1299{
1300 int error;
1301
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001302 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1303 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1306 /* Ensure the VFS deletes this inode */
1307 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001308 clear_nlink(dentry->d_inode);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001309 else if (error == -ENOENT)
1310 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311
1312 return error;
1313}
1314
1315static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1316{
1317 static unsigned int sillycounter;
Peter Staubach4e769b92007-08-03 15:07:10 -04001318 const int fileidsize = sizeof(NFS_FILEID(dentry->d_inode))*2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 const int countersize = sizeof(sillycounter)*2;
Peter Staubach4e769b92007-08-03 15:07:10 -04001320 const int slen = sizeof(".nfs")+fileidsize+countersize-1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 char silly[slen+1];
1322 struct qstr qsilly;
1323 struct dentry *sdentry;
1324 int error = -EIO;
1325
1326 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1327 dentry->d_parent->d_name.name, dentry->d_name.name,
1328 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001329 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 /*
1332 * We don't allow a dentry to be silly-renamed twice.
1333 */
1334 error = -EBUSY;
1335 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1336 goto out;
1337
Peter Staubach4e769b92007-08-03 15:07:10 -04001338 sprintf(silly, ".nfs%*.*Lx",
1339 fileidsize, fileidsize,
1340 (unsigned long long)NFS_FILEID(dentry->d_inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Trond Myklebust34ea8182005-11-04 15:35:02 -05001342 /* Return delegation in anticipation of the rename */
1343 nfs_inode_return_delegation(dentry->d_inode);
1344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 sdentry = NULL;
1346 do {
1347 char *suffix = silly + slen - countersize;
1348
1349 dput(sdentry);
1350 sillycounter++;
1351 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1352
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001353 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1354 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355
1356 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1357 /*
1358 * N.B. Better to return EBUSY here ... it could be
1359 * dangerous to delete the file while it's in use.
1360 */
1361 if (IS_ERR(sdentry))
1362 goto out;
1363 } while(sdentry->d_inode != NULL); /* need negative lookup */
1364
1365 qsilly.name = silly;
1366 qsilly.len = strlen(silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 if (dentry->d_inode) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1369 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001370 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 } else
1372 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1373 dir, &qsilly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1376 d_move(dentry, sdentry);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001377 error = nfs_async_unlink(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 /* If we return 0 we don't unlink */
1379 }
1380 dput(sdentry);
1381out:
1382 return error;
1383}
1384
1385/*
1386 * Remove a file after making sure there are no pending writes,
1387 * and after checking that the file has only one user.
1388 *
1389 * We invalidate the attribute cache and free the inode prior to the operation
1390 * to avoid possible races if the server reuses the inode.
1391 */
1392static int nfs_safe_remove(struct dentry *dentry)
1393{
1394 struct inode *dir = dentry->d_parent->d_inode;
1395 struct inode *inode = dentry->d_inode;
1396 int error = -EBUSY;
1397
1398 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1399 dentry->d_parent->d_name.name, dentry->d_name.name);
1400
1401 /* If the dentry was sillyrenamed, we simply call d_delete() */
1402 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1403 error = 0;
1404 goto out;
1405 }
1406
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001408 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1410 /* The VFS may want to delete this inode */
1411 if (error == 0)
Trond Myklebust1b83d702008-06-11 15:44:04 -04001412 nfs_drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001413 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 } else
1415 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001416 if (error == -ENOENT)
1417 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418out:
1419 return error;
1420}
1421
1422/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1423 * belongs to an active ".nfs..." file and we return -EBUSY.
1424 *
1425 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1426 */
1427static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1428{
1429 int error;
1430 int need_rehash = 0;
1431
1432 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1433 dir->i_ino, dentry->d_name.name);
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return error;
1444 }
1445 if (!d_unhashed(dentry)) {
1446 __d_drop(dentry);
1447 need_rehash = 1;
1448 }
1449 spin_unlock(&dentry->d_lock);
1450 spin_unlock(&dcache_lock);
1451 error = nfs_safe_remove(dentry);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001452 if (!error || error == -ENOENT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1454 } else if (need_rehash)
1455 d_rehash(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 return error;
1457}
1458
Chuck Lever873101b2006-08-22 20:06:23 -04001459/*
1460 * To create a symbolic link, most file systems instantiate a new inode,
1461 * add a page to it containing the path, then write it out to the disk
1462 * using prepare_write/commit_write.
1463 *
1464 * Unfortunately the NFS client can't create the in-core inode first
1465 * because it needs a file handle to create an in-core inode (see
1466 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1467 * symlink request has completed on the server.
1468 *
1469 * So instead we allocate a raw page, copy the symname into it, then do
1470 * the SYMLINK request with the page as the buffer. If it succeeds, we
1471 * now have a new file handle and can instantiate an in-core NFS inode
1472 * and move the raw page into its mapping.
1473 */
1474static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475{
Chuck Lever873101b2006-08-22 20:06:23 -04001476 struct pagevec lru_pvec;
1477 struct page *page;
1478 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001480 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 int error;
1482
1483 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1484 dir->i_ino, dentry->d_name.name, symname);
1485
Chuck Lever873101b2006-08-22 20:06:23 -04001486 if (pathlen > PAGE_SIZE)
1487 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488
Chuck Lever873101b2006-08-22 20:06:23 -04001489 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1490 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491
Jeff Layton83d93f22007-06-07 09:58:08 -04001492 page = alloc_page(GFP_HIGHUSER);
Trond Myklebust76566992008-06-11 15:44:22 -04001493 if (!page)
Chuck Lever873101b2006-08-22 20:06:23 -04001494 return -ENOMEM;
Chuck Lever873101b2006-08-22 20:06:23 -04001495
1496 kaddr = kmap_atomic(page, KM_USER0);
1497 memcpy(kaddr, symname, pathlen);
1498 if (pathlen < PAGE_SIZE)
1499 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1500 kunmap_atomic(kaddr, KM_USER0);
1501
Chuck Lever94a6d752006-08-22 20:06:23 -04001502 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Chuck Lever873101b2006-08-22 20:06:23 -04001503 if (error != 0) {
1504 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1505 dir->i_sb->s_id, dir->i_ino,
1506 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001508 __free_page(page);
Chuck Lever873101b2006-08-22 20:06:23 -04001509 return error;
1510 }
1511
1512 /*
1513 * No big deal if we can't add this page to the page cache here.
1514 * READLINK will get the missing page from the server if needed.
1515 */
1516 pagevec_init(&lru_pvec, 0);
1517 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1518 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001519 pagevec_add(&lru_pvec, page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001520 pagevec_lru_add_file(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001521 SetPageUptodate(page);
1522 unlock_page(page);
1523 } else
1524 __free_page(page);
1525
Chuck Lever873101b2006-08-22 20:06:23 -04001526 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527}
1528
1529static int
1530nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1531{
1532 struct inode *inode = old_dentry->d_inode;
1533 int error;
1534
1535 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1536 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1537 dentry->d_parent->d_name.name, dentry->d_name.name);
1538
Trond Myklebust9697d232007-10-02 21:58:05 -04001539 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001541 if (error == 0) {
1542 atomic_inc(&inode->i_count);
Trond Myklebust9697d232007-10-02 21:58:05 -04001543 d_add(dentry, inode);
Trond Myklebustcf809552005-10-27 22:12:42 -04001544 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return error;
1546}
1547
1548/*
1549 * RENAME
1550 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1551 * different file handle for the same inode after a rename (e.g. when
1552 * moving to a different directory). A fail-safe method to do so would
1553 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1554 * rename the old file using the sillyrename stuff. This way, the original
1555 * file in old_dir will go away when the last process iput()s the inode.
1556 *
1557 * FIXED.
1558 *
1559 * It actually works quite well. One needs to have the possibility for
1560 * at least one ".nfs..." file in each directory the file ever gets
1561 * moved or linked to which happens automagically with the new
1562 * implementation that only depends on the dcache stuff instead of
1563 * using the inode layer
1564 *
1565 * Unfortunately, things are a little more complicated than indicated
1566 * above. For a cross-directory move, we want to make sure we can get
1567 * rid of the old inode after the operation. This means there must be
1568 * no pending writes (if it's a file), and the use count must be 1.
1569 * If these conditions are met, we can drop the dentries before doing
1570 * the rename.
1571 */
1572static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1573 struct inode *new_dir, struct dentry *new_dentry)
1574{
1575 struct inode *old_inode = old_dentry->d_inode;
1576 struct inode *new_inode = new_dentry->d_inode;
1577 struct dentry *dentry = NULL, *rehash = NULL;
1578 int error = -EBUSY;
1579
1580 /*
1581 * To prevent any new references to the target during the rename,
1582 * we unhash the dentry and free the inode in advance.
1583 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001584 if (!d_unhashed(new_dentry)) {
1585 d_drop(new_dentry);
1586 rehash = new_dentry;
1587 }
1588
1589 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1590 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1591 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1592 atomic_read(&new_dentry->d_count));
1593
1594 /*
1595 * First check whether the target is busy ... we can't
1596 * safely do _any_ rename if the target is in use.
1597 *
1598 * For files, make a copy of the dentry and then do a
1599 * silly-rename. If the silly-rename succeeds, the
1600 * copied dentry is hashed and becomes the new target.
1601 */
1602 if (!new_inode)
1603 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001604 if (S_ISDIR(new_inode->i_mode)) {
1605 error = -EISDIR;
1606 if (!S_ISDIR(old_inode->i_mode))
1607 goto out;
1608 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 int err;
1610 /* copy the target dentry's name */
1611 dentry = d_alloc(new_dentry->d_parent,
1612 &new_dentry->d_name);
1613 if (!dentry)
1614 goto out;
1615
1616 /* silly-rename the existing target ... */
1617 err = nfs_sillyrename(new_dir, new_dentry);
1618 if (!err) {
1619 new_dentry = rehash = dentry;
1620 new_inode = NULL;
1621 /* instantiate the replacement target */
1622 d_instantiate(new_dentry, NULL);
Jesper Juhl7a13e932007-04-26 00:29:02 -07001623 } else if (atomic_read(&new_dentry->d_count) > 1)
1624 /* dentry still busy? */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 goto out;
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001626 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
1628go_ahead:
1629 /*
1630 * ... prune child dentries and writebacks if needed.
1631 */
1632 if (atomic_read(&old_dentry->d_count) > 1) {
Trond Myklebuste1552e12007-04-14 19:07:28 -04001633 if (S_ISREG(old_inode->i_mode))
1634 nfs_wb_all(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 shrink_dcache_parent(old_dentry);
1636 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001637 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001639 if (new_inode != NULL)
Trond Myklebust24174112006-01-03 09:55:33 +01001640 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1643 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001644 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645out:
1646 if (rehash)
1647 d_rehash(rehash);
1648 if (!error) {
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001649 if (new_inode != NULL)
1650 nfs_drop_nlink(new_inode);
Mark Fasheh349457c2006-09-08 14:22:21 -07001651 d_move(old_dentry, new_dentry);
Chuck Lever8fb559f2007-09-24 15:40:16 -04001652 nfs_set_verifier(new_dentry,
1653 nfs_save_change_attribute(new_dir));
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001654 } else if (error == -ENOENT)
1655 nfs_dentry_handle_enoent(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657 /* new dentry created? */
1658 if (dentry)
1659 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 return error;
1661}
1662
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001663static DEFINE_SPINLOCK(nfs_access_lru_lock);
1664static LIST_HEAD(nfs_access_lru_list);
1665static atomic_long_t nfs_access_nr_entries;
1666
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001667static void nfs_access_free_entry(struct nfs_access_entry *entry)
1668{
1669 put_rpccred(entry->cred);
1670 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001671 smp_mb__before_atomic_dec();
1672 atomic_long_dec(&nfs_access_nr_entries);
1673 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001674}
1675
Trond Myklebust979df722006-07-25 11:28:19 -04001676int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1677{
1678 LIST_HEAD(head);
1679 struct nfs_inode *nfsi;
1680 struct nfs_access_entry *cache;
1681
Trond Myklebust979df722006-07-25 11:28:19 -04001682restart:
Trond Myklebusta50f7952007-06-05 19:23:43 -04001683 spin_lock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001684 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
Trond Myklebust6f23e382008-01-25 16:38:17 -05001685 struct rw_semaphore *s_umount;
Trond Myklebust979df722006-07-25 11:28:19 -04001686 struct inode *inode;
1687
1688 if (nr_to_scan-- == 0)
1689 break;
Trond Myklebust6f23e382008-01-25 16:38:17 -05001690 s_umount = &nfsi->vfs_inode.i_sb->s_umount;
1691 if (!down_read_trylock(s_umount))
Trond Myklebust979df722006-07-25 11:28:19 -04001692 continue;
Trond Myklebust6f23e382008-01-25 16:38:17 -05001693 inode = igrab(&nfsi->vfs_inode);
1694 if (inode == NULL) {
1695 up_read(s_umount);
1696 continue;
1697 }
Trond Myklebust979df722006-07-25 11:28:19 -04001698 spin_lock(&inode->i_lock);
1699 if (list_empty(&nfsi->access_cache_entry_lru))
1700 goto remove_lru_entry;
1701 cache = list_entry(nfsi->access_cache_entry_lru.next,
1702 struct nfs_access_entry, lru);
1703 list_move(&cache->lru, &head);
1704 rb_erase(&cache->rb_node, &nfsi->access_cache);
1705 if (!list_empty(&nfsi->access_cache_entry_lru))
1706 list_move_tail(&nfsi->access_cache_inode_lru,
1707 &nfs_access_lru_list);
1708 else {
1709remove_lru_entry:
1710 list_del_init(&nfsi->access_cache_inode_lru);
1711 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1712 }
1713 spin_unlock(&inode->i_lock);
Trond Myklebusta50f7952007-06-05 19:23:43 -04001714 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04001715 iput(inode);
Trond Myklebust6f23e382008-01-25 16:38:17 -05001716 up_read(s_umount);
Trond Myklebust979df722006-07-25 11:28:19 -04001717 goto restart;
1718 }
1719 spin_unlock(&nfs_access_lru_lock);
1720 while (!list_empty(&head)) {
1721 cache = list_entry(head.next, struct nfs_access_entry, lru);
1722 list_del(&cache->lru);
1723 nfs_access_free_entry(cache);
1724 }
1725 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1726}
1727
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001728static void __nfs_access_zap_cache(struct inode *inode)
1729{
1730 struct nfs_inode *nfsi = NFS_I(inode);
1731 struct rb_root *root_node = &nfsi->access_cache;
1732 struct rb_node *n, *dispose = NULL;
1733 struct nfs_access_entry *entry;
1734
1735 /* Unhook entries from the cache */
1736 while ((n = rb_first(root_node)) != NULL) {
1737 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1738 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001739 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001740 n->rb_left = dispose;
1741 dispose = n;
1742 }
1743 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1744 spin_unlock(&inode->i_lock);
1745
1746 /* Now kill them all! */
1747 while (dispose != NULL) {
1748 n = dispose;
1749 dispose = n->rb_left;
1750 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1751 }
1752}
1753
1754void nfs_access_zap_cache(struct inode *inode)
1755{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001756 /* Remove from global LRU init */
Benny Halevy3a10c302008-01-23 08:58:59 +02001757 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001758 spin_lock(&nfs_access_lru_lock);
1759 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1760 spin_unlock(&nfs_access_lru_lock);
1761 }
1762
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001763 spin_lock(&inode->i_lock);
1764 /* This will release the spinlock */
1765 __nfs_access_zap_cache(inode);
1766}
1767
1768static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1769{
1770 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1771 struct nfs_access_entry *entry;
1772
1773 while (n != NULL) {
1774 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1775
1776 if (cred < entry->cred)
1777 n = n->rb_left;
1778 else if (cred > entry->cred)
1779 n = n->rb_right;
1780 else
1781 return entry;
1782 }
1783 return NULL;
1784}
1785
Trond Myklebustaf22f942007-08-10 17:45:10 -04001786static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787{
Chuck Lever55296802005-08-18 11:24:09 -07001788 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001789 struct nfs_access_entry *cache;
1790 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001792 spin_lock(&inode->i_lock);
1793 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1794 goto out_zap;
1795 cache = nfs_access_search_rbtree(inode, cred);
1796 if (cache == NULL)
1797 goto out;
Trond Myklebust7a50c602008-12-23 15:21:55 -05001798 if (!nfs_have_delegation(inode, FMODE_READ) &&
Peter Staubach64672d52008-12-23 15:21:56 -05001799 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001800 goto out_stale;
1801 res->jiffies = cache->jiffies;
1802 res->cred = cache->cred;
1803 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001804 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001805 err = 0;
1806out:
1807 spin_unlock(&inode->i_lock);
1808 return err;
1809out_stale:
1810 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001811 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001812 spin_unlock(&inode->i_lock);
1813 nfs_access_free_entry(cache);
1814 return -ENOENT;
1815out_zap:
1816 /* This will release the spinlock */
1817 __nfs_access_zap_cache(inode);
1818 return -ENOENT;
1819}
1820
1821static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1822{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001823 struct nfs_inode *nfsi = NFS_I(inode);
1824 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001825 struct rb_node **p = &root_node->rb_node;
1826 struct rb_node *parent = NULL;
1827 struct nfs_access_entry *entry;
1828
1829 spin_lock(&inode->i_lock);
1830 while (*p != NULL) {
1831 parent = *p;
1832 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1833
1834 if (set->cred < entry->cred)
1835 p = &parent->rb_left;
1836 else if (set->cred > entry->cred)
1837 p = &parent->rb_right;
1838 else
1839 goto found;
1840 }
1841 rb_link_node(&set->rb_node, parent, p);
1842 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001843 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001844 spin_unlock(&inode->i_lock);
1845 return;
1846found:
1847 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001848 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1849 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001850 spin_unlock(&inode->i_lock);
1851 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001852}
1853
Trond Myklebustaf22f942007-08-10 17:45:10 -04001854static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001856 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1857 if (cache == NULL)
1858 return;
1859 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001861 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001863
1864 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001865
1866 /* Update accounting */
1867 smp_mb__before_atomic_inc();
1868 atomic_long_inc(&nfs_access_nr_entries);
1869 smp_mb__after_atomic_inc();
1870
1871 /* Add inode to global LRU list */
Benny Halevy3a10c302008-01-23 08:58:59 +02001872 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001873 spin_lock(&nfs_access_lru_lock);
1874 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1875 spin_unlock(&nfs_access_lru_lock);
1876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877}
1878
1879static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1880{
1881 struct nfs_access_entry cache;
1882 int status;
1883
1884 status = nfs_access_get_cached(inode, cred, &cache);
1885 if (status == 0)
1886 goto out;
1887
1888 /* Be clever: ask server to check for all possible rights */
1889 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1890 cache.cred = cred;
1891 cache.jiffies = jiffies;
1892 status = NFS_PROTO(inode)->access(inode, &cache);
Suresh Jayaramana71ee332009-03-10 20:33:21 -04001893 if (status != 0) {
1894 if (status == -ESTALE) {
1895 nfs_zap_caches(inode);
1896 if (!S_ISDIR(inode->i_mode))
1897 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
1898 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 return status;
Suresh Jayaramana71ee332009-03-10 20:33:21 -04001900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 nfs_access_add_cache(inode, &cache);
1902out:
Al Viroe6305c42008-07-15 21:03:57 -04001903 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001904 return 0;
1905 return -EACCES;
1906}
1907
Trond Myklebustaf22f942007-08-10 17:45:10 -04001908static int nfs_open_permission_mask(int openflags)
1909{
1910 int mask = 0;
1911
1912 if (openflags & FMODE_READ)
1913 mask |= MAY_READ;
1914 if (openflags & FMODE_WRITE)
1915 mask |= MAY_WRITE;
1916 if (openflags & FMODE_EXEC)
1917 mask |= MAY_EXEC;
1918 return mask;
1919}
1920
1921int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
1922{
1923 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
1924}
1925
Al Viroe6305c42008-07-15 21:03:57 -04001926int nfs_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
1928 struct rpc_cred *cred;
1929 int res = 0;
1930
Chuck Lever91d5b472006-03-20 13:44:14 -05001931 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1932
Al Viroe6305c42008-07-15 21:03:57 -04001933 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 goto out;
1935 /* Is this sys_access() ? */
Al Viroe6305c42008-07-15 21:03:57 -04001936 if (mask & MAY_ACCESS)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001937 goto force_lookup;
1938
1939 switch (inode->i_mode & S_IFMT) {
1940 case S_IFLNK:
1941 goto out;
1942 case S_IFREG:
1943 /* NFSv4 has atomic_open... */
1944 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
Frank Filz7ee2cb72009-05-18 17:41:40 -04001945 && (mask & MAY_OPEN)
1946 && !(mask & MAY_EXEC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 goto out;
1948 break;
1949 case S_IFDIR:
1950 /*
1951 * Optimize away all write operations, since the server
1952 * will check permissions when we perform the op.
1953 */
1954 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1955 goto out;
1956 }
1957
1958force_lookup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 if (!NFS_PROTO(inode)->access)
1960 goto out_notsup;
1961
Trond Myklebust98a8e322008-03-12 12:25:28 -04001962 cred = rpc_lookup_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 if (!IS_ERR(cred)) {
1964 res = nfs_do_access(inode, cred, mask);
1965 put_rpccred(cred);
1966 } else
1967 res = PTR_ERR(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968out:
Miklos Szeredif696a362008-07-31 13:41:58 +02001969 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
1970 res = -EACCES;
1971
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001972 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1973 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 return res;
1975out_notsup:
1976 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1977 if (res == 0)
1978 res = generic_permission(inode, mask, NULL);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001979 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980}
1981
1982/*
1983 * Local variables:
1984 * version-control: t
1985 * kept-new-versions: 5
1986 * End:
1987 */