blob: c86a1ead47724b60fd5badb02cbe8945f4785b44 [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>
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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
41#define NFS_PARANOIA 1
42/* #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
68struct inode_operations nfs_dir_inode_operations = {
69 .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
84struct inode_operations nfs3_dir_inode_operations = {
85 .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 *);
107struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
110 .link = nfs_link,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
113 .mkdir = nfs_mkdir,
114 .rmdir = nfs_rmdir,
115 .mknod = nfs_mknod,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
J. Bruce Fields6b3b5492005-06-22 17:16:22 +0000120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123};
124
125#endif /* CONFIG_NFS_V4 */
126
127/*
128 * Open file
129 */
130static int
131nfs_opendir(struct inode *inode, struct file *filp)
132{
Carsten Otte7451c4f2006-04-19 13:06:37 -0400133 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 lock_kernel();
139 /* Call generic open code in order to cache credentials */
Carsten Otte7451c4f2006-04-19 13:06:37 -0400140 res = nfs_open(inode, filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 unlock_kernel();
142 return res;
143}
144
145typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
146typedef struct {
147 struct file *file;
148 struct page *page;
149 unsigned long page_index;
150 u32 *ptr;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000151 u64 *dir_cookie;
152 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 struct nfs_entry *entry;
154 decode_dirent_t decode;
155 int plus;
156 int error;
157} nfs_readdir_descriptor_t;
158
159/* Now we cache directories properly, by stuffing the dirent
160 * data directly in the page cache.
161 *
162 * Inode invalidation due to refresh etc. takes care of
163 * _everything_, no sloppy entry flushing logic, no extraneous
164 * copying, network direct to page cache, the way it was meant
165 * to be.
166 *
167 * NOTE: Dirent information verification is done always by the
168 * page-in of the RPC reply, nowhere else, this simplies
169 * things substantially.
170 */
171static
172int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
173{
174 struct file *file = desc->file;
175 struct inode *inode = file->f_dentry->d_inode;
176 struct rpc_cred *cred = nfs_file_cred(file);
177 unsigned long timestamp;
178 int error;
179
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500180 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
181 __FUNCTION__, (long long)desc->entry->cookie,
182 page->index);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
184 again:
185 timestamp = jiffies;
186 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
187 NFS_SERVER(inode)->dtsize, desc->plus);
188 if (error < 0) {
189 /* We requested READDIRPLUS, but the server doesn't grok it */
190 if (error == -ENOTSUPP && desc->plus) {
191 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Chuck Lever412d5822005-08-18 11:24:11 -0700192 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 desc->plus = 0;
194 goto again;
195 }
196 goto error;
197 }
198 SetPageUptodate(page);
Chuck Leverdc592502005-08-18 11:24:12 -0700199 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700200 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700201 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 /* Ensure consistent page alignment of the data.
203 * Note: assumes we have exclusive access to this mapping either
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800204 * through inode->i_mutex or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 */
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700206 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
207 /* Should never happen */
208 nfs_zap_mapping(inode, inode->i_mapping);
209 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 unlock_page(page);
211 return 0;
212 error:
213 SetPageError(page);
214 unlock_page(page);
215 nfs_zap_caches(inode);
216 desc->error = error;
217 return -EIO;
218}
219
220static inline
221int dir_decode(nfs_readdir_descriptor_t *desc)
222{
223 u32 *p = desc->ptr;
224 p = desc->decode(p, desc->entry, desc->plus);
225 if (IS_ERR(p))
226 return PTR_ERR(p);
227 desc->ptr = p;
228 return 0;
229}
230
231static inline
232void dir_page_release(nfs_readdir_descriptor_t *desc)
233{
234 kunmap(desc->page);
235 page_cache_release(desc->page);
236 desc->page = NULL;
237 desc->ptr = NULL;
238}
239
240/*
241 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000242 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 *
244 * If the end of the buffer has been reached, return -EAGAIN, if not,
245 * return the offset within the buffer of the next entry to be
246 * read.
247 */
248static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000249int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 struct nfs_entry *entry = desc->entry;
252 int loop_count = 0,
253 status;
254
255 while((status = dir_decode(desc)) == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500256 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
257 __FUNCTION__, (unsigned long long)entry->cookie);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000258 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 break;
260 if (loop_count++ > 200) {
261 loop_count = 0;
262 schedule();
263 }
264 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 return status;
266}
267
268/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000269 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000270 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000271 *
272 * If the end of the buffer has been reached, return -EAGAIN, if not,
273 * return the offset within the buffer of the next entry to be
274 * read.
275 */
276static inline
277int find_dirent_index(nfs_readdir_descriptor_t *desc)
278{
279 struct nfs_entry *entry = desc->entry;
280 int loop_count = 0,
281 status;
282
283 for(;;) {
284 status = dir_decode(desc);
285 if (status)
286 break;
287
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500288 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
289 (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000290
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000291 if (desc->file->f_pos == desc->current_index) {
292 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000293 break;
294 }
295 desc->current_index++;
296 if (loop_count++ > 200) {
297 loop_count = 0;
298 schedule();
299 }
300 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000301 return status;
302}
303
304/*
305 * Find the given page, and call find_dirent() or find_dirent_index in
306 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 */
308static inline
309int find_dirent_page(nfs_readdir_descriptor_t *desc)
310{
311 struct inode *inode = desc->file->f_dentry->d_inode;
312 struct page *page;
313 int status;
314
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500315 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
316 __FUNCTION__, desc->page_index,
317 (long long) *desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
319 page = read_cache_page(inode->i_mapping, desc->page_index,
320 (filler_t *)nfs_readdir_filler, desc);
321 if (IS_ERR(page)) {
322 status = PTR_ERR(page);
323 goto out;
324 }
325 if (!PageUptodate(page))
326 goto read_error;
327
328 /* NOTE: Someone else may have changed the READDIRPLUS flag */
329 desc->page = page;
330 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000331 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000332 status = find_dirent(desc);
333 else
334 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 if (status < 0)
336 dir_page_release(desc);
337 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500338 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 return status;
340 read_error:
341 page_cache_release(page);
342 return -EIO;
343}
344
345/*
346 * Recurse through the page cache pages, and return a
347 * filled nfs_entry structure of the next directory entry if possible.
348 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000349 * The target for the search is '*desc->dir_cookie' if non-0,
350 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700351 */
352static inline
353int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
354{
355 int loop_count = 0;
356 int res;
357
Olivier Galibert00a92642005-06-22 17:16:29 +0000358 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000359 if (*desc->dir_cookie == 0) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500360 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
361 (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000362 desc->page_index = 0;
363 desc->entry->cookie = desc->entry->prev_cookie = 0;
364 desc->entry->eof = 0;
365 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000366 } else
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500367 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
368 (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 for (;;) {
371 res = find_dirent_page(desc);
372 if (res != -EAGAIN)
373 break;
374 /* Align to beginning of next page */
375 desc->page_index ++;
376 if (loop_count++ > 200) {
377 loop_count = 0;
378 schedule();
379 }
380 }
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500381
382 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return res;
384}
385
386static inline unsigned int dt_type(struct inode *inode)
387{
388 return (inode->i_mode >> 12) & 15;
389}
390
391static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
392
393/*
394 * Once we've found the start of the dirent within a page: fill 'er up...
395 */
396static
397int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
398 filldir_t filldir)
399{
400 struct file *file = desc->file;
401 struct nfs_entry *entry = desc->entry;
402 struct dentry *dentry = NULL;
403 unsigned long fileid;
404 int loop_count = 0,
405 res;
406
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500407 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
408 (unsigned long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409
410 for(;;) {
411 unsigned d_type = DT_UNKNOWN;
412 /* Note: entry->prev_cookie contains the cookie for
413 * retrieving the current dirent on the server */
414 fileid = nfs_fileid_to_ino_t(entry->ino);
415
416 /* Get a dentry if we have one */
417 if (dentry != NULL)
418 dput(dentry);
419 dentry = nfs_readdir_lookup(desc);
420
421 /* Use readdirplus info */
422 if (dentry != NULL && dentry->d_inode != NULL) {
423 d_type = dt_type(dentry->d_inode);
424 fileid = dentry->d_inode->i_ino;
425 }
426
427 res = filldir(dirent, entry->name, entry->len,
Olivier Galibert00a92642005-06-22 17:16:29 +0000428 file->f_pos, fileid, d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 if (res < 0)
430 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000431 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000432 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 if (dir_decode(desc) != 0) {
434 desc->page_index ++;
435 break;
436 }
437 if (loop_count++ > 200) {
438 loop_count = 0;
439 schedule();
440 }
441 }
442 dir_page_release(desc);
443 if (dentry != NULL)
444 dput(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500445 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
446 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 return res;
448}
449
450/*
451 * If we cannot find a cookie in our cache, we suspect that this is
452 * because it points to a deleted file, so we ask the server to return
453 * whatever it thinks is the next entry. We then feed this to filldir.
454 * If all goes well, we should then be able to find our way round the
455 * cache on the next call to readdir_search_pagecache();
456 *
457 * NOTE: we cannot add the anonymous page to the pagecache because
458 * the data it contains might not be page aligned. Besides,
459 * we should already have a complete representation of the
460 * directory in the page cache by the time we get here.
461 */
462static inline
463int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
464 filldir_t filldir)
465{
466 struct file *file = desc->file;
467 struct inode *inode = file->f_dentry->d_inode;
468 struct rpc_cred *cred = nfs_file_cred(file);
469 struct page *page = NULL;
470 int status;
471
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500472 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
473 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474
475 page = alloc_page(GFP_HIGHUSER);
476 if (!page) {
477 status = -ENOMEM;
478 goto out;
479 }
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000480 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 page,
482 NFS_SERVER(inode)->dtsize,
483 desc->plus);
Chuck Leverdc592502005-08-18 11:24:12 -0700484 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700485 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700486 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 desc->page = page;
488 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
489 if (desc->error >= 0) {
490 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000491 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 } else
493 status = -EIO;
494 if (status < 0)
495 goto out_release;
496
497 status = nfs_do_filldir(desc, dirent, filldir);
498
499 /* Reset read descriptor so it searches the page cache from
500 * the start upon the next call to readdir_search_pagecache() */
501 desc->page_index = 0;
502 desc->entry->cookie = desc->entry->prev_cookie = 0;
503 desc->entry->eof = 0;
504 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500505 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
506 __FUNCTION__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 return status;
508 out_release:
509 dir_page_release(desc);
510 goto out;
511}
512
Olivier Galibert00a92642005-06-22 17:16:29 +0000513/* The file offset position represents the dirent entry number. A
514 last cookie cache takes care of the common case of reading the
515 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 */
517static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
518{
519 struct dentry *dentry = filp->f_dentry;
520 struct inode *inode = dentry->d_inode;
521 nfs_readdir_descriptor_t my_desc,
522 *desc = &my_desc;
523 struct nfs_entry my_entry;
524 struct nfs_fh fh;
525 struct nfs_fattr fattr;
526 long res;
527
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500528 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
529 dentry->d_parent->d_name.name, dentry->d_name.name,
530 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500531 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
532
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 lock_kernel();
534
Trond Myklebust44b11872006-05-25 01:40:59 -0400535 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 if (res < 0) {
537 unlock_kernel();
538 return res;
539 }
540
541 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000542 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000543 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000544 * to either find the entry with the appropriate number or
545 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 */
547 memset(desc, 0, sizeof(*desc));
548
549 desc->file = filp;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000550 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 desc->decode = NFS_PROTO(inode)->decode_dirent;
552 desc->plus = NFS_USE_READDIRPLUS(inode);
553
554 my_entry.cookie = my_entry.prev_cookie = 0;
555 my_entry.eof = 0;
556 my_entry.fh = &fh;
557 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400558 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 desc->entry = &my_entry;
560
561 while(!desc->entry->eof) {
562 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000563
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 if (res == -EBADCOOKIE) {
565 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000566 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 /* Or that the server has 'lost' a cookie */
568 res = uncached_readdir(desc, dirent, filldir);
569 if (res >= 0)
570 continue;
571 }
572 res = 0;
573 break;
574 }
575 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700576 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 nfs_zap_caches(inode);
578 desc->plus = 0;
579 desc->entry->eof = 0;
580 continue;
581 }
582 if (res < 0)
583 break;
584
585 res = nfs_do_filldir(desc, dirent, filldir);
586 if (res < 0) {
587 res = 0;
588 break;
589 }
590 }
591 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500592 if (res > 0)
593 res = 0;
594 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
595 dentry->d_parent->d_name.name, dentry->d_name.name,
596 res);
597 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598}
599
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000600loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
601{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800602 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000603 switch (origin) {
604 case 1:
605 offset += filp->f_pos;
606 case 0:
607 if (offset >= 0)
608 break;
609 default:
610 offset = -EINVAL;
611 goto out;
612 }
613 if (offset != filp->f_pos) {
614 filp->f_pos = offset;
615 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
616 }
617out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800618 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000619 return offset;
620}
621
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622/*
623 * All directory operations under NFS are synchronous, so fsync()
624 * is a dummy operation.
625 */
626int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
627{
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500628 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
629 dentry->d_parent->d_name.name, dentry->d_name.name,
630 datasync);
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 return 0;
633}
634
635/*
636 * A check for whether or not the parent directory has changed.
637 * In the case it has, we assume that the dentries are untrustworthy
638 * and may need to be looked up again.
639 */
640static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
641{
642 if (IS_ROOT(dentry))
643 return 1;
Chuck Lever55296802005-08-18 11:24:09 -0700644 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 || nfs_attribute_timeout(dir))
646 return 0;
647 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
648}
649
650static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
651{
652 dentry->d_fsdata = (void *)verf;
653}
654
655/*
656 * Whenever an NFS operation succeeds, we know that the dentry
657 * is valid, so we update the revalidation timestamp.
658 */
659static inline void nfs_renew_times(struct dentry * dentry)
660{
661 dentry->d_time = jiffies;
662}
663
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400664/*
665 * Return the intent data that applies to this particular path component
666 *
667 * Note that the current set of intents only apply to the very last
668 * component of the path.
669 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
670 */
671static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
672{
673 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
674 return 0;
675 return nd->flags & mask;
676}
677
678/*
679 * Inode and filehandle revalidation for lookups.
680 *
681 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
682 * or if the intent information indicates that we're about to open this
683 * particular file and the "nocto" mount flag is not set.
684 *
685 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686static inline
687int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
688{
689 struct nfs_server *server = NFS_SERVER(inode);
690
691 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400693 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 goto out_force;
695 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400696 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -0400697 !(server->flags & NFS_MOUNT_NOCTO) &&
698 (S_ISREG(inode->i_mode) ||
699 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 goto out_force;
701 }
702 return nfs_revalidate_inode(server, inode);
703out_force:
704 return __nfs_revalidate_inode(server, inode);
705}
706
707/*
708 * We judge how long we want to trust negative
709 * dentries by looking at the parent inode mtime.
710 *
711 * If parent mtime has changed, we revalidate, else we wait for a
712 * period corresponding to the parent's attribute cache timeout value.
713 */
714static inline
715int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
716 struct nameidata *nd)
717{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400719 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 return 0;
721 return !nfs_check_verifier(dir, dentry);
722}
723
724/*
725 * This is called every time the dcache has a lookup hit,
726 * and we should check whether we can really trust that
727 * lookup.
728 *
729 * NOTE! The hit can be a negative hit too, don't assume
730 * we have an inode!
731 *
732 * If the parent directory is seen to have changed, we throw out the
733 * cached dentry and do a new lookup.
734 */
735static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
736{
737 struct inode *dir;
738 struct inode *inode;
739 struct dentry *parent;
740 int error;
741 struct nfs_fh fhandle;
742 struct nfs_fattr fattr;
743 unsigned long verifier;
744
745 parent = dget_parent(dentry);
746 lock_kernel();
747 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -0500748 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 inode = dentry->d_inode;
750
751 if (!inode) {
752 if (nfs_neg_need_reval(dir, dentry, nd))
753 goto out_bad;
754 goto out_valid;
755 }
756
757 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500758 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
759 __FUNCTION__, dentry->d_parent->d_name.name,
760 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 goto out_bad;
762 }
763
764 /* Revalidate parent directory attribute cache */
765 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
766 goto out_zap_parent;
767
768 /* Force a full look up iff the parent directory has changed */
769 if (nfs_check_verifier(dir, dentry)) {
770 if (nfs_lookup_verify_inode(inode, nd))
771 goto out_zap_parent;
772 goto out_valid;
773 }
774
775 if (NFS_STALE(inode))
776 goto out_bad;
777
778 verifier = nfs_save_change_attribute(dir);
779 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
780 if (error)
781 goto out_bad;
782 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
783 goto out_bad;
784 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
785 goto out_bad;
786
787 nfs_renew_times(dentry);
788 nfs_set_verifier(dentry, verifier);
789 out_valid:
790 unlock_kernel();
791 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500792 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
793 __FUNCTION__, dentry->d_parent->d_name.name,
794 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 return 1;
796out_zap_parent:
797 nfs_zap_caches(dir);
798 out_bad:
799 NFS_CACHEINV(dir);
800 if (inode && S_ISDIR(inode->i_mode)) {
801 /* Purge readdir caches. */
802 nfs_zap_caches(inode);
803 /* If we have submounts, don't unhash ! */
804 if (have_submounts(dentry))
805 goto out_valid;
806 shrink_dcache_parent(dentry);
807 }
808 d_drop(dentry);
809 unlock_kernel();
810 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500811 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
812 __FUNCTION__, dentry->d_parent->d_name.name,
813 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 return 0;
815}
816
817/*
818 * This is called from dput() when d_count is going to 0.
819 */
820static int nfs_dentry_delete(struct dentry *dentry)
821{
822 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
823 dentry->d_parent->d_name.name, dentry->d_name.name,
824 dentry->d_flags);
825
826 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
827 /* Unhash it, so that ->d_iput() would be called */
828 return 1;
829 }
830 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
831 /* Unhash it, so that ancestors of killed async unlink
832 * files will be cleaned up during umount */
833 return 1;
834 }
835 return 0;
836
837}
838
839/*
840 * Called when the dentry loses inode.
841 * We use it to clean up silly-renamed files.
842 */
843static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
844{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700845 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
847 lock_kernel();
Dave Hansen9a53c3a2006-09-30 23:29:03 -0700848 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 nfs_complete_unlink(dentry);
850 unlock_kernel();
851 }
852 /* When creating a negative dentry, we want to renew d_time */
853 nfs_renew_times(dentry);
854 iput(inode);
855}
856
857struct dentry_operations nfs_dentry_operations = {
858 .d_revalidate = nfs_lookup_revalidate,
859 .d_delete = nfs_dentry_delete,
860 .d_iput = nfs_dentry_iput,
861};
862
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400863/*
864 * Use intent information to check whether or not we're going to do
865 * an O_EXCL create using this path component.
866 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867static inline
868int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
869{
870 if (NFS_PROTO(dir)->version == 2)
871 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400872 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873 return 0;
874 return (nd->intent.open.flags & O_EXCL) != 0;
875}
876
David Howells54ceac42006-08-22 20:06:13 -0400877static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
878 struct nfs_fh *fh, struct nfs_fattr *fattr)
Trond Myklebust55a97592006-06-09 09:34:19 -0400879{
880 struct nfs_server *server = NFS_SERVER(dir);
881
882 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
883 /* Revalidate fsid on root dir */
David Howells54ceac42006-08-22 20:06:13 -0400884 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
Trond Myklebust55a97592006-06-09 09:34:19 -0400885 return 0;
886}
887
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
889{
890 struct dentry *res;
891 struct inode *inode = NULL;
892 int error;
893 struct nfs_fh fhandle;
894 struct nfs_fattr fattr;
895
896 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
897 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -0500898 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899
900 res = ERR_PTR(-ENAMETOOLONG);
901 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
902 goto out;
903
904 res = ERR_PTR(-ENOMEM);
905 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
906
907 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
Trond Myklebustfd684072006-09-05 12:27:44 -0400909 /*
910 * If we're doing an exclusive create, optimize away the lookup
911 * but don't hash the dentry.
912 */
913 if (nfs_is_exclusive_create(dir, nd)) {
914 d_instantiate(dentry, NULL);
915 res = NULL;
916 goto out_unlock;
917 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918
919 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
920 if (error == -ENOENT)
921 goto no_entry;
922 if (error < 0) {
923 res = ERR_PTR(error);
924 goto out_unlock;
925 }
David Howells54ceac42006-08-22 20:06:13 -0400926 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
Trond Myklebust55a97592006-06-09 09:34:19 -0400927 if (error < 0) {
928 res = ERR_PTR(error);
929 goto out_unlock;
930 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -0500932 res = (struct dentry *)inode;
933 if (IS_ERR(res))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934 goto out_unlock;
David Howells54ceac42006-08-22 20:06:13 -0400935
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936no_entry:
David Howells54ceac42006-08-22 20:06:13 -0400937 res = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 if (res != NULL)
939 dentry = res;
940 nfs_renew_times(dentry);
941 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
942out_unlock:
943 unlock_kernel();
944out:
945 return res;
946}
947
948#ifdef CONFIG_NFS_V4
949static int nfs_open_revalidate(struct dentry *, struct nameidata *);
950
951struct dentry_operations nfs4_dentry_operations = {
952 .d_revalidate = nfs_open_revalidate,
953 .d_delete = nfs_dentry_delete,
954 .d_iput = nfs_dentry_iput,
955};
956
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400957/*
958 * Use intent information to determine whether we need to substitute
959 * the NFSv4-style stateful OPEN for the LOOKUP call
960 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961static int is_atomic_open(struct inode *dir, struct nameidata *nd)
962{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400963 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 return 0;
965 /* NFS does not (yet) have a stateful open for directories */
966 if (nd->flags & LOOKUP_DIRECTORY)
967 return 0;
968 /* Are we trying to write to a read only partition? */
969 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
970 return 0;
971 return 1;
972}
973
974static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
975{
976 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 int error;
978
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500979 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
980 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 /* Check that we are indeed trying to open this file */
983 if (!is_atomic_open(dir, nd))
984 goto no_open;
985
986 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
987 res = ERR_PTR(-ENAMETOOLONG);
988 goto out;
989 }
990 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
991
992 /* Let vfs_create() deal with O_EXCL */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700993 if (nd->intent.open.flags & O_EXCL) {
994 d_add(dentry, NULL);
995 goto out;
996 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
998 /* Open the file on the server */
999 lock_kernel();
1000 /* Revalidate parent directory attribute cache */
1001 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1002 if (error < 0) {
1003 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -07001004 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 goto out;
1006 }
1007
1008 if (nd->intent.open.flags & O_CREAT) {
1009 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001010 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 nfs_end_data_update(dir);
1012 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -07001013 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -07001015 if (IS_ERR(res)) {
1016 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 switch (error) {
1018 /* Make a negative dentry */
1019 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -07001020 res = NULL;
1021 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001023 case -EISDIR:
1024 case -ENOTDIR:
1025 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 case -ELOOP:
1027 if (!(nd->intent.open.flags & O_NOFOLLOW))
1028 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 /* case -EINVAL: */
1030 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031 goto out;
1032 }
Trond Myklebust02a913a2005-10-18 14:20:17 -07001033 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 dentry = res;
1035 nfs_renew_times(dentry);
1036 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1037out:
1038 return res;
1039no_open:
1040 return nfs_lookup(dir, dentry, nd);
1041}
1042
1043static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1044{
1045 struct dentry *parent = NULL;
1046 struct inode *inode = dentry->d_inode;
1047 struct inode *dir;
1048 unsigned long verifier;
1049 int openflags, ret = 0;
1050
1051 parent = dget_parent(dentry);
1052 dir = parent->d_inode;
1053 if (!is_atomic_open(dir, nd))
1054 goto no_open;
1055 /* We can't create new files in nfs_open_revalidate(), so we
1056 * optimize away revalidation of negative dentries.
1057 */
1058 if (inode == NULL)
1059 goto out;
1060 /* NFS only supports OPEN on regular files */
1061 if (!S_ISREG(inode->i_mode))
1062 goto no_open;
1063 openflags = nd->intent.open.flags;
1064 /* We cannot do exclusive creation on a positive dentry */
1065 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1066 goto no_open;
1067 /* We can't create new files, or truncate existing ones here */
1068 openflags &= ~(O_CREAT|O_TRUNC);
1069
1070 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001071 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 * operations that change the directory. We therefore save the
1073 * change attribute *before* we do the RPC call.
1074 */
1075 lock_kernel();
1076 verifier = nfs_save_change_attribute(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001077 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 if (!ret)
1079 nfs_set_verifier(dentry, verifier);
1080 unlock_kernel();
1081out:
1082 dput(parent);
1083 if (!ret)
1084 d_drop(dentry);
1085 return ret;
1086no_open:
1087 dput(parent);
1088 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1089 return 1;
1090 return nfs_lookup_revalidate(dentry, nd);
1091}
1092#endif /* CONFIG_NFSV4 */
1093
1094static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1095{
1096 struct dentry *parent = desc->file->f_dentry;
1097 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;
1105
1106 switch (name.len) {
1107 case 2:
1108 if (name.name[0] == '.' && name.name[1] == '.')
1109 return dget_parent(parent);
1110 break;
1111 case 1:
1112 if (name.name[0] == '.')
1113 return dget(parent);
1114 }
1115 name.hash = full_name_hash(name.name, name.len);
1116 dentry = d_lookup(parent, &name);
1117 if (dentry != NULL)
1118 return dentry;
1119 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1120 return NULL;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001121 /* Note: caller is already holding the dir->i_mutex! */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 dentry = d_alloc(parent, &name);
1123 if (dentry == NULL)
1124 return NULL;
1125 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1126 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001127 if (IS_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 dput(dentry);
1129 return NULL;
1130 }
David Howells54ceac42006-08-22 20:06:13 -04001131
1132 alias = d_materialise_unique(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133 if (alias != NULL) {
1134 dput(dentry);
1135 dentry = alias;
1136 }
David Howells54ceac42006-08-22 20:06:13 -04001137
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 nfs_renew_times(dentry);
1139 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1140 return dentry;
1141}
1142
1143/*
1144 * Code common to create, mkdir, and mknod.
1145 */
1146int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1147 struct nfs_fattr *fattr)
1148{
1149 struct inode *inode;
1150 int error = -EACCES;
1151
1152 /* We may have been initialized further down */
1153 if (dentry->d_inode)
1154 return 0;
1155 if (fhandle->size == 0) {
1156 struct inode *dir = dentry->d_parent->d_inode;
1157 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1158 if (error)
Chuck Lever4f390c12006-08-22 20:06:22 -04001159 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 }
1161 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1162 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001163 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (error < 0)
Chuck Lever4f390c12006-08-22 20:06:22 -04001165 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001168 error = PTR_ERR(inode);
1169 if (IS_ERR(inode))
Chuck Lever4f390c12006-08-22 20:06:22 -04001170 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171 d_instantiate(dentry, inode);
Trond Myklebustfd684072006-09-05 12:27:44 -04001172 if (d_unhashed(dentry))
1173 d_rehash(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175}
1176
1177/*
1178 * Following a failed create operation, we drop the dentry rather
1179 * than retain a negative dentry. This avoids a problem in the event
1180 * that the operation succeeded on the server, but an error in the
1181 * reply path made it appear to have failed.
1182 */
1183static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1184 struct nameidata *nd)
1185{
1186 struct iattr attr;
1187 int error;
1188 int open_flags = 0;
1189
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001190 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1191 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
1193 attr.ia_mode = mode;
1194 attr.ia_valid = ATTR_MODE;
1195
1196 if (nd && (nd->flags & LOOKUP_CREATE))
1197 open_flags = nd->intent.open.flags;
1198
1199 lock_kernel();
1200 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001201 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 nfs_end_data_update(dir);
1203 if (error != 0)
1204 goto out_err;
1205 nfs_renew_times(dentry);
1206 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1207 unlock_kernel();
1208 return 0;
1209out_err:
1210 unlock_kernel();
1211 d_drop(dentry);
1212 return error;
1213}
1214
1215/*
1216 * See comments for nfs_proc_create regarding failed operations.
1217 */
1218static int
1219nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1220{
1221 struct iattr attr;
1222 int status;
1223
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001224 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1225 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226
1227 if (!new_valid_dev(rdev))
1228 return -EINVAL;
1229
1230 attr.ia_mode = mode;
1231 attr.ia_valid = ATTR_MODE;
1232
1233 lock_kernel();
1234 nfs_begin_data_update(dir);
1235 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1236 nfs_end_data_update(dir);
1237 if (status != 0)
1238 goto out_err;
1239 nfs_renew_times(dentry);
1240 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1241 unlock_kernel();
1242 return 0;
1243out_err:
1244 unlock_kernel();
1245 d_drop(dentry);
1246 return status;
1247}
1248
1249/*
1250 * See comments for nfs_proc_create regarding failed operations.
1251 */
1252static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1253{
1254 struct iattr attr;
1255 int error;
1256
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001257 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1258 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259
1260 attr.ia_valid = ATTR_MODE;
1261 attr.ia_mode = mode | S_IFDIR;
1262
1263 lock_kernel();
1264 nfs_begin_data_update(dir);
1265 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1266 nfs_end_data_update(dir);
1267 if (error != 0)
1268 goto out_err;
1269 nfs_renew_times(dentry);
1270 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1271 unlock_kernel();
1272 return 0;
1273out_err:
1274 d_drop(dentry);
1275 unlock_kernel();
1276 return error;
1277}
1278
1279static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1280{
1281 int error;
1282
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001283 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1284 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285
1286 lock_kernel();
1287 nfs_begin_data_update(dir);
1288 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1289 /* Ensure the VFS deletes this inode */
1290 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001291 clear_nlink(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 nfs_end_data_update(dir);
1293 unlock_kernel();
1294
1295 return error;
1296}
1297
1298static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1299{
1300 static unsigned int sillycounter;
1301 const int i_inosize = sizeof(dir->i_ino)*2;
1302 const int countersize = sizeof(sillycounter)*2;
1303 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1304 char silly[slen+1];
1305 struct qstr qsilly;
1306 struct dentry *sdentry;
1307 int error = -EIO;
1308
1309 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1310 dentry->d_parent->d_name.name, dentry->d_name.name,
1311 atomic_read(&dentry->d_count));
Chuck Lever91d5b472006-03-20 13:44:14 -05001312 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314#ifdef NFS_PARANOIA
1315if (!dentry->d_inode)
1316printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1317dentry->d_parent->d_name.name, dentry->d_name.name);
1318#endif
1319 /*
1320 * We don't allow a dentry to be silly-renamed twice.
1321 */
1322 error = -EBUSY;
1323 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1324 goto out;
1325
1326 sprintf(silly, ".nfs%*.*lx",
1327 i_inosize, i_inosize, dentry->d_inode->i_ino);
1328
Trond Myklebust34ea8182005-11-04 15:35:02 -05001329 /* Return delegation in anticipation of the rename */
1330 nfs_inode_return_delegation(dentry->d_inode);
1331
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 sdentry = NULL;
1333 do {
1334 char *suffix = silly + slen - countersize;
1335
1336 dput(sdentry);
1337 sillycounter++;
1338 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1339
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001340 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1341 dentry->d_name.name, silly);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342
1343 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1344 /*
1345 * N.B. Better to return EBUSY here ... it could be
1346 * dangerous to delete the file while it's in use.
1347 */
1348 if (IS_ERR(sdentry))
1349 goto out;
1350 } while(sdentry->d_inode != NULL); /* need negative lookup */
1351
1352 qsilly.name = silly;
1353 qsilly.len = strlen(silly);
1354 nfs_begin_data_update(dir);
1355 if (dentry->d_inode) {
1356 nfs_begin_data_update(dentry->d_inode);
1357 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1358 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001359 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 nfs_end_data_update(dentry->d_inode);
1361 } else
1362 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1363 dir, &qsilly);
1364 nfs_end_data_update(dir);
1365 if (!error) {
1366 nfs_renew_times(dentry);
1367 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1368 d_move(dentry, sdentry);
1369 error = nfs_async_unlink(dentry);
1370 /* If we return 0 we don't unlink */
1371 }
1372 dput(sdentry);
1373out:
1374 return error;
1375}
1376
1377/*
1378 * Remove a file after making sure there are no pending writes,
1379 * and after checking that the file has only one user.
1380 *
1381 * We invalidate the attribute cache and free the inode prior to the operation
1382 * to avoid possible races if the server reuses the inode.
1383 */
1384static int nfs_safe_remove(struct dentry *dentry)
1385{
1386 struct inode *dir = dentry->d_parent->d_inode;
1387 struct inode *inode = dentry->d_inode;
1388 int error = -EBUSY;
1389
1390 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1391 dentry->d_parent->d_name.name, dentry->d_name.name);
1392
1393 /* If the dentry was sillyrenamed, we simply call d_delete() */
1394 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1395 error = 0;
1396 goto out;
1397 }
1398
1399 nfs_begin_data_update(dir);
1400 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001401 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 nfs_begin_data_update(inode);
1403 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1404 /* The VFS may want to delete this inode */
1405 if (error == 0)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001406 drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001407 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 nfs_end_data_update(inode);
1409 } else
1410 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1411 nfs_end_data_update(dir);
1412out:
1413 return error;
1414}
1415
1416/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1417 * belongs to an active ".nfs..." file and we return -EBUSY.
1418 *
1419 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1420 */
1421static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1422{
1423 int error;
1424 int need_rehash = 0;
1425
1426 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1427 dir->i_ino, dentry->d_name.name);
1428
1429 lock_kernel();
1430 spin_lock(&dcache_lock);
1431 spin_lock(&dentry->d_lock);
1432 if (atomic_read(&dentry->d_count) > 1) {
1433 spin_unlock(&dentry->d_lock);
1434 spin_unlock(&dcache_lock);
1435 error = nfs_sillyrename(dir, dentry);
1436 unlock_kernel();
1437 return error;
1438 }
1439 if (!d_unhashed(dentry)) {
1440 __d_drop(dentry);
1441 need_rehash = 1;
1442 }
1443 spin_unlock(&dentry->d_lock);
1444 spin_unlock(&dcache_lock);
1445 error = nfs_safe_remove(dentry);
1446 if (!error) {
1447 nfs_renew_times(dentry);
1448 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1449 } else if (need_rehash)
1450 d_rehash(dentry);
1451 unlock_kernel();
1452 return error;
1453}
1454
Chuck Lever873101b2006-08-22 20:06:23 -04001455/*
1456 * To create a symbolic link, most file systems instantiate a new inode,
1457 * add a page to it containing the path, then write it out to the disk
1458 * using prepare_write/commit_write.
1459 *
1460 * Unfortunately the NFS client can't create the in-core inode first
1461 * because it needs a file handle to create an in-core inode (see
1462 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1463 * symlink request has completed on the server.
1464 *
1465 * So instead we allocate a raw page, copy the symname into it, then do
1466 * the SYMLINK request with the page as the buffer. If it succeeds, we
1467 * now have a new file handle and can instantiate an in-core NFS inode
1468 * and move the raw page into its mapping.
1469 */
1470static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
Chuck Lever873101b2006-08-22 20:06:23 -04001472 struct pagevec lru_pvec;
1473 struct page *page;
1474 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001476 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 int error;
1478
1479 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1480 dir->i_ino, dentry->d_name.name, symname);
1481
Chuck Lever873101b2006-08-22 20:06:23 -04001482 if (pathlen > PAGE_SIZE)
1483 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Chuck Lever873101b2006-08-22 20:06:23 -04001485 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1486 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
1488 lock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001489
1490 page = alloc_page(GFP_KERNEL);
1491 if (!page) {
1492 unlock_kernel();
1493 return -ENOMEM;
1494 }
1495
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
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 nfs_begin_data_update(dir);
Chuck Lever94a6d752006-08-22 20:06:23 -04001503 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 nfs_end_data_update(dir);
Chuck Lever873101b2006-08-22 20:06:23 -04001505 if (error != 0) {
1506 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1507 dir->i_sb->s_id, dir->i_ino,
1508 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001510 __free_page(page);
1511 unlock_kernel();
1512 return error;
1513 }
1514
1515 /*
1516 * No big deal if we can't add this page to the page cache here.
1517 * READLINK will get the missing page from the server if needed.
1518 */
1519 pagevec_init(&lru_pvec, 0);
1520 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1521 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001522 pagevec_add(&lru_pvec, page);
1523 pagevec_lru_add(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001524 SetPageUptodate(page);
1525 unlock_page(page);
1526 } else
1527 __free_page(page);
1528
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 unlock_kernel();
Chuck Lever873101b2006-08-22 20:06:23 -04001530 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001531}
1532
1533static int
1534nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1535{
1536 struct inode *inode = old_dentry->d_inode;
1537 int error;
1538
1539 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1540 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1541 dentry->d_parent->d_name.name, dentry->d_name.name);
1542
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 nfs_begin_data_update(dir);
1545 nfs_begin_data_update(inode);
1546 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001547 if (error == 0) {
1548 atomic_inc(&inode->i_count);
1549 d_instantiate(dentry, inode);
1550 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 nfs_end_data_update(inode);
1552 nfs_end_data_update(dir);
1553 unlock_kernel();
1554 return error;
1555}
1556
1557/*
1558 * RENAME
1559 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1560 * different file handle for the same inode after a rename (e.g. when
1561 * moving to a different directory). A fail-safe method to do so would
1562 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1563 * rename the old file using the sillyrename stuff. This way, the original
1564 * file in old_dir will go away when the last process iput()s the inode.
1565 *
1566 * FIXED.
1567 *
1568 * It actually works quite well. One needs to have the possibility for
1569 * at least one ".nfs..." file in each directory the file ever gets
1570 * moved or linked to which happens automagically with the new
1571 * implementation that only depends on the dcache stuff instead of
1572 * using the inode layer
1573 *
1574 * Unfortunately, things are a little more complicated than indicated
1575 * above. For a cross-directory move, we want to make sure we can get
1576 * rid of the old inode after the operation. This means there must be
1577 * no pending writes (if it's a file), and the use count must be 1.
1578 * If these conditions are met, we can drop the dentries before doing
1579 * the rename.
1580 */
1581static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1582 struct inode *new_dir, struct dentry *new_dentry)
1583{
1584 struct inode *old_inode = old_dentry->d_inode;
1585 struct inode *new_inode = new_dentry->d_inode;
1586 struct dentry *dentry = NULL, *rehash = NULL;
1587 int error = -EBUSY;
1588
1589 /*
1590 * To prevent any new references to the target during the rename,
1591 * we unhash the dentry and free the inode in advance.
1592 */
1593 lock_kernel();
1594 if (!d_unhashed(new_dentry)) {
1595 d_drop(new_dentry);
1596 rehash = new_dentry;
1597 }
1598
1599 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1600 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1601 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1602 atomic_read(&new_dentry->d_count));
1603
1604 /*
1605 * First check whether the target is busy ... we can't
1606 * safely do _any_ rename if the target is in use.
1607 *
1608 * For files, make a copy of the dentry and then do a
1609 * silly-rename. If the silly-rename succeeds, the
1610 * copied dentry is hashed and becomes the new target.
1611 */
1612 if (!new_inode)
1613 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001614 if (S_ISDIR(new_inode->i_mode)) {
1615 error = -EISDIR;
1616 if (!S_ISDIR(old_inode->i_mode))
1617 goto out;
1618 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 int err;
1620 /* copy the target dentry's name */
1621 dentry = d_alloc(new_dentry->d_parent,
1622 &new_dentry->d_name);
1623 if (!dentry)
1624 goto out;
1625
1626 /* silly-rename the existing target ... */
1627 err = nfs_sillyrename(new_dir, new_dentry);
1628 if (!err) {
1629 new_dentry = rehash = dentry;
1630 new_inode = NULL;
1631 /* instantiate the replacement target */
1632 d_instantiate(new_dentry, NULL);
1633 } else if (atomic_read(&new_dentry->d_count) > 1) {
1634 /* dentry still busy? */
1635#ifdef NFS_PARANOIA
1636 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1637 new_dentry->d_parent->d_name.name,
1638 new_dentry->d_name.name,
1639 atomic_read(&new_dentry->d_count));
1640#endif
1641 goto out;
1642 }
Trond Myklebust20509f12005-08-25 16:25:34 -07001643 } else
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001644 drop_nlink(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
1646go_ahead:
1647 /*
1648 * ... prune child dentries and writebacks if needed.
1649 */
1650 if (atomic_read(&old_dentry->d_count) > 1) {
1651 nfs_wb_all(old_inode);
1652 shrink_dcache_parent(old_dentry);
1653 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001654 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
Trond Myklebust24174112006-01-03 09:55:33 +01001656 if (new_inode != NULL) {
1657 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001659 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661 nfs_begin_data_update(old_dir);
1662 nfs_begin_data_update(new_dir);
1663 nfs_begin_data_update(old_inode);
1664 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1665 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001666 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001667 nfs_end_data_update(old_inode);
1668 nfs_end_data_update(new_dir);
1669 nfs_end_data_update(old_dir);
1670out:
1671 if (rehash)
1672 d_rehash(rehash);
1673 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07001674 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 nfs_renew_times(new_dentry);
1676 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1677 }
1678
1679 /* new dentry created? */
1680 if (dentry)
1681 dput(dentry);
1682 unlock_kernel();
1683 return error;
1684}
1685
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001686static DEFINE_SPINLOCK(nfs_access_lru_lock);
1687static LIST_HEAD(nfs_access_lru_list);
1688static atomic_long_t nfs_access_nr_entries;
1689
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001690static void nfs_access_free_entry(struct nfs_access_entry *entry)
1691{
1692 put_rpccred(entry->cred);
1693 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001694 smp_mb__before_atomic_dec();
1695 atomic_long_dec(&nfs_access_nr_entries);
1696 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001697}
1698
Trond Myklebust979df722006-07-25 11:28:19 -04001699int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1700{
1701 LIST_HEAD(head);
1702 struct nfs_inode *nfsi;
1703 struct nfs_access_entry *cache;
1704
1705 spin_lock(&nfs_access_lru_lock);
1706restart:
1707 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1708 struct inode *inode;
1709
1710 if (nr_to_scan-- == 0)
1711 break;
1712 inode = igrab(&nfsi->vfs_inode);
1713 if (inode == NULL)
1714 continue;
1715 spin_lock(&inode->i_lock);
1716 if (list_empty(&nfsi->access_cache_entry_lru))
1717 goto remove_lru_entry;
1718 cache = list_entry(nfsi->access_cache_entry_lru.next,
1719 struct nfs_access_entry, lru);
1720 list_move(&cache->lru, &head);
1721 rb_erase(&cache->rb_node, &nfsi->access_cache);
1722 if (!list_empty(&nfsi->access_cache_entry_lru))
1723 list_move_tail(&nfsi->access_cache_inode_lru,
1724 &nfs_access_lru_list);
1725 else {
1726remove_lru_entry:
1727 list_del_init(&nfsi->access_cache_inode_lru);
1728 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1729 }
1730 spin_unlock(&inode->i_lock);
1731 iput(inode);
1732 goto restart;
1733 }
1734 spin_unlock(&nfs_access_lru_lock);
1735 while (!list_empty(&head)) {
1736 cache = list_entry(head.next, struct nfs_access_entry, lru);
1737 list_del(&cache->lru);
1738 nfs_access_free_entry(cache);
1739 }
1740 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1741}
1742
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001743static void __nfs_access_zap_cache(struct inode *inode)
1744{
1745 struct nfs_inode *nfsi = NFS_I(inode);
1746 struct rb_root *root_node = &nfsi->access_cache;
1747 struct rb_node *n, *dispose = NULL;
1748 struct nfs_access_entry *entry;
1749
1750 /* Unhook entries from the cache */
1751 while ((n = rb_first(root_node)) != NULL) {
1752 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1753 rb_erase(n, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001754 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001755 n->rb_left = dispose;
1756 dispose = n;
1757 }
1758 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1759 spin_unlock(&inode->i_lock);
1760
1761 /* Now kill them all! */
1762 while (dispose != NULL) {
1763 n = dispose;
1764 dispose = n->rb_left;
1765 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1766 }
1767}
1768
1769void nfs_access_zap_cache(struct inode *inode)
1770{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001771 /* Remove from global LRU init */
1772 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1773 spin_lock(&nfs_access_lru_lock);
1774 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1775 spin_unlock(&nfs_access_lru_lock);
1776 }
1777
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001778 spin_lock(&inode->i_lock);
1779 /* This will release the spinlock */
1780 __nfs_access_zap_cache(inode);
1781}
1782
1783static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1784{
1785 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1786 struct nfs_access_entry *entry;
1787
1788 while (n != NULL) {
1789 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1790
1791 if (cred < entry->cred)
1792 n = n->rb_left;
1793 else if (cred > entry->cred)
1794 n = n->rb_right;
1795 else
1796 return entry;
1797 }
1798 return NULL;
1799}
1800
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1802{
Chuck Lever55296802005-08-18 11:24:09 -07001803 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001804 struct nfs_access_entry *cache;
1805 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001807 spin_lock(&inode->i_lock);
1808 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1809 goto out_zap;
1810 cache = nfs_access_search_rbtree(inode, cred);
1811 if (cache == NULL)
1812 goto out;
1813 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1814 goto out_stale;
1815 res->jiffies = cache->jiffies;
1816 res->cred = cache->cred;
1817 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001818 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001819 err = 0;
1820out:
1821 spin_unlock(&inode->i_lock);
1822 return err;
1823out_stale:
1824 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001825 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001826 spin_unlock(&inode->i_lock);
1827 nfs_access_free_entry(cache);
1828 return -ENOENT;
1829out_zap:
1830 /* This will release the spinlock */
1831 __nfs_access_zap_cache(inode);
1832 return -ENOENT;
1833}
1834
1835static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1836{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001837 struct nfs_inode *nfsi = NFS_I(inode);
1838 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001839 struct rb_node **p = &root_node->rb_node;
1840 struct rb_node *parent = NULL;
1841 struct nfs_access_entry *entry;
1842
1843 spin_lock(&inode->i_lock);
1844 while (*p != NULL) {
1845 parent = *p;
1846 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1847
1848 if (set->cred < entry->cred)
1849 p = &parent->rb_left;
1850 else if (set->cred > entry->cred)
1851 p = &parent->rb_right;
1852 else
1853 goto found;
1854 }
1855 rb_link_node(&set->rb_node, parent, p);
1856 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001857 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001858 spin_unlock(&inode->i_lock);
1859 return;
1860found:
1861 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001862 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1863 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001864 spin_unlock(&inode->i_lock);
1865 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866}
1867
1868void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1869{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001870 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1871 if (cache == NULL)
1872 return;
1873 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001875 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04001877
1878 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04001879
1880 /* Update accounting */
1881 smp_mb__before_atomic_inc();
1882 atomic_long_inc(&nfs_access_nr_entries);
1883 smp_mb__after_atomic_inc();
1884
1885 /* Add inode to global LRU list */
1886 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1887 spin_lock(&nfs_access_lru_lock);
1888 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1889 spin_unlock(&nfs_access_lru_lock);
1890 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891}
1892
1893static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1894{
1895 struct nfs_access_entry cache;
1896 int status;
1897
1898 status = nfs_access_get_cached(inode, cred, &cache);
1899 if (status == 0)
1900 goto out;
1901
1902 /* Be clever: ask server to check for all possible rights */
1903 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1904 cache.cred = cred;
1905 cache.jiffies = jiffies;
1906 status = NFS_PROTO(inode)->access(inode, &cache);
1907 if (status != 0)
1908 return status;
1909 nfs_access_add_cache(inode, &cache);
1910out:
1911 if ((cache.mask & mask) == mask)
1912 return 0;
1913 return -EACCES;
1914}
1915
1916int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1917{
1918 struct rpc_cred *cred;
1919 int res = 0;
1920
Chuck Lever91d5b472006-03-20 13:44:14 -05001921 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1922
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 if (mask == 0)
1924 goto out;
1925 /* Is this sys_access() ? */
1926 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1927 goto force_lookup;
1928
1929 switch (inode->i_mode & S_IFMT) {
1930 case S_IFLNK:
1931 goto out;
1932 case S_IFREG:
1933 /* NFSv4 has atomic_open... */
1934 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1935 && nd != NULL
1936 && (nd->flags & LOOKUP_OPEN))
1937 goto out;
1938 break;
1939 case S_IFDIR:
1940 /*
1941 * Optimize away all write operations, since the server
1942 * will check permissions when we perform the op.
1943 */
1944 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1945 goto out;
1946 }
1947
1948force_lookup:
1949 lock_kernel();
1950
1951 if (!NFS_PROTO(inode)->access)
1952 goto out_notsup;
1953
1954 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1955 if (!IS_ERR(cred)) {
1956 res = nfs_do_access(inode, cred, mask);
1957 put_rpccred(cred);
1958 } else
1959 res = PTR_ERR(cred);
1960 unlock_kernel();
1961out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001962 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
1963 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 return res;
1965out_notsup:
1966 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1967 if (res == 0)
1968 res = generic_permission(inode, mask, NULL);
1969 unlock_kernel();
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001970 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971}
1972
1973/*
1974 * Local variables:
1975 * version-control: t
1976 * kept-new-versions: 5
1977 * End:
1978 */