blob: e9255198f767b8af7b964a66f66e8be39fc597c6 [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>
33#include <linux/namei.h>
34
Trond Myklebust4ce79712005-06-22 17:16:21 +000035#include "nfs4_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include "delegation.h"
37
38#define NFS_PARANOIA 1
39/* #define NFS_DEBUG_VERBOSE 1 */
40
41static int nfs_opendir(struct inode *, struct file *);
42static int nfs_readdir(struct file *, void *, filldir_t);
43static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
44static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
45static int nfs_mkdir(struct inode *, struct dentry *, int);
46static int nfs_rmdir(struct inode *, struct dentry *);
47static int nfs_unlink(struct inode *, struct dentry *);
48static int nfs_symlink(struct inode *, struct dentry *, const char *);
49static int nfs_link(struct dentry *, struct inode *, struct dentry *);
50static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
51static int nfs_rename(struct inode *, struct dentry *,
52 struct inode *, struct dentry *);
53static int nfs_fsync_dir(struct file *, struct dentry *, int);
Trond Myklebustf0dd2132005-06-22 17:16:29 +000054static loff_t nfs_llseek_dir(struct file *, loff_t, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
56struct file_operations nfs_dir_operations = {
Trond Myklebustf0dd2132005-06-22 17:16:29 +000057 .llseek = nfs_llseek_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 .read = generic_read_dir,
59 .readdir = nfs_readdir,
60 .open = nfs_opendir,
61 .release = nfs_release,
62 .fsync = nfs_fsync_dir,
63};
64
65struct inode_operations nfs_dir_inode_operations = {
66 .create = nfs_create,
67 .lookup = nfs_lookup,
68 .link = nfs_link,
69 .unlink = nfs_unlink,
70 .symlink = nfs_symlink,
71 .mkdir = nfs_mkdir,
72 .rmdir = nfs_rmdir,
73 .mknod = nfs_mknod,
74 .rename = nfs_rename,
75 .permission = nfs_permission,
76 .getattr = nfs_getattr,
77 .setattr = nfs_setattr,
78};
79
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000080#ifdef CONFIG_NFS_V3
81struct inode_operations nfs3_dir_inode_operations = {
82 .create = nfs_create,
83 .lookup = nfs_lookup,
84 .link = nfs_link,
85 .unlink = nfs_unlink,
86 .symlink = nfs_symlink,
87 .mkdir = nfs_mkdir,
88 .rmdir = nfs_rmdir,
89 .mknod = nfs_mknod,
90 .rename = nfs_rename,
91 .permission = nfs_permission,
92 .getattr = nfs_getattr,
93 .setattr = nfs_setattr,
94 .listxattr = nfs3_listxattr,
95 .getxattr = nfs3_getxattr,
96 .setxattr = nfs3_setxattr,
97 .removexattr = nfs3_removexattr,
98};
99#endif /* CONFIG_NFS_V3 */
100
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101#ifdef CONFIG_NFS_V4
102
103static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
104struct inode_operations nfs4_dir_inode_operations = {
105 .create = nfs_create,
106 .lookup = nfs_atomic_lookup,
107 .link = nfs_link,
108 .unlink = nfs_unlink,
109 .symlink = nfs_symlink,
110 .mkdir = nfs_mkdir,
111 .rmdir = nfs_rmdir,
112 .mknod = nfs_mknod,
113 .rename = nfs_rename,
114 .permission = nfs_permission,
115 .getattr = nfs_getattr,
116 .setattr = nfs_setattr,
J. Bruce Fields6b3b5492005-06-22 17:16:22 +0000117 .getxattr = nfs4_getxattr,
118 .setxattr = nfs4_setxattr,
119 .listxattr = nfs4_listxattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120};
121
122#endif /* CONFIG_NFS_V4 */
123
124/*
125 * Open file
126 */
127static int
128nfs_opendir(struct inode *inode, struct file *filp)
129{
130 int res = 0;
131
132 lock_kernel();
133 /* Call generic open code in order to cache credentials */
134 if (!res)
135 res = nfs_open(inode, filp);
136 unlock_kernel();
137 return res;
138}
139
140typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
141typedef struct {
142 struct file *file;
143 struct page *page;
144 unsigned long page_index;
145 u32 *ptr;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000146 u64 *dir_cookie;
147 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 struct nfs_entry *entry;
149 decode_dirent_t decode;
150 int plus;
151 int error;
152} nfs_readdir_descriptor_t;
153
154/* Now we cache directories properly, by stuffing the dirent
155 * data directly in the page cache.
156 *
157 * Inode invalidation due to refresh etc. takes care of
158 * _everything_, no sloppy entry flushing logic, no extraneous
159 * copying, network direct to page cache, the way it was meant
160 * to be.
161 *
162 * NOTE: Dirent information verification is done always by the
163 * page-in of the RPC reply, nowhere else, this simplies
164 * things substantially.
165 */
166static
167int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
168{
169 struct file *file = desc->file;
170 struct inode *inode = file->f_dentry->d_inode;
171 struct rpc_cred *cred = nfs_file_cred(file);
172 unsigned long timestamp;
173 int error;
174
175 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
176
177 again:
178 timestamp = jiffies;
179 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
180 NFS_SERVER(inode)->dtsize, desc->plus);
181 if (error < 0) {
182 /* We requested READDIRPLUS, but the server doesn't grok it */
183 if (error == -ENOTSUPP && desc->plus) {
184 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Chuck Lever412d5822005-08-18 11:24:11 -0700185 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 desc->plus = 0;
187 goto again;
188 }
189 goto error;
190 }
191 SetPageUptodate(page);
Chuck Leverdc592502005-08-18 11:24:12 -0700192 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700193 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700194 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 /* Ensure consistent page alignment of the data.
196 * Note: assumes we have exclusive access to this mapping either
Trond Myklebusta656db92005-06-22 17:16:21 +0000197 * through inode->i_sem or some other mechanism.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 */
Trond Myklebusta656db92005-06-22 17:16:21 +0000199 if (page->index == 0)
200 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 unlock_page(page);
202 return 0;
203 error:
204 SetPageError(page);
205 unlock_page(page);
206 nfs_zap_caches(inode);
207 desc->error = error;
208 return -EIO;
209}
210
211static inline
212int dir_decode(nfs_readdir_descriptor_t *desc)
213{
214 u32 *p = desc->ptr;
215 p = desc->decode(p, desc->entry, desc->plus);
216 if (IS_ERR(p))
217 return PTR_ERR(p);
218 desc->ptr = p;
219 return 0;
220}
221
222static inline
223void dir_page_release(nfs_readdir_descriptor_t *desc)
224{
225 kunmap(desc->page);
226 page_cache_release(desc->page);
227 desc->page = NULL;
228 desc->ptr = NULL;
229}
230
231/*
232 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000233 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 *
235 * If the end of the buffer has been reached, return -EAGAIN, if not,
236 * return the offset within the buffer of the next entry to be
237 * read.
238 */
239static inline
Olivier Galibert00a92642005-06-22 17:16:29 +0000240int find_dirent(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241{
242 struct nfs_entry *entry = desc->entry;
243 int loop_count = 0,
244 status;
245
246 while((status = dir_decode(desc)) == 0) {
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000247 dfprintk(VFS, "NFS: found cookie %Lu\n", (unsigned long long)entry->cookie);
248 if (entry->prev_cookie == *desc->dir_cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 break;
250 if (loop_count++ > 200) {
251 loop_count = 0;
252 schedule();
253 }
254 }
255 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
256 return status;
257}
258
259/*
Olivier Galibert00a92642005-06-22 17:16:29 +0000260 * Given a pointer to a buffer that has already been filled by a call
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000261 * to readdir, find the entry at offset 'desc->file->f_pos'.
Olivier Galibert00a92642005-06-22 17:16:29 +0000262 *
263 * If the end of the buffer has been reached, return -EAGAIN, if not,
264 * return the offset within the buffer of the next entry to be
265 * read.
266 */
267static inline
268int find_dirent_index(nfs_readdir_descriptor_t *desc)
269{
270 struct nfs_entry *entry = desc->entry;
271 int loop_count = 0,
272 status;
273
274 for(;;) {
275 status = dir_decode(desc);
276 if (status)
277 break;
278
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000279 dfprintk(VFS, "NFS: found cookie %Lu at index %Ld\n", (unsigned long long)entry->cookie, desc->current_index);
Olivier Galibert00a92642005-06-22 17:16:29 +0000280
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000281 if (desc->file->f_pos == desc->current_index) {
282 *desc->dir_cookie = entry->cookie;
Olivier Galibert00a92642005-06-22 17:16:29 +0000283 break;
284 }
285 desc->current_index++;
286 if (loop_count++ > 200) {
287 loop_count = 0;
288 schedule();
289 }
290 }
291 dfprintk(VFS, "NFS: find_dirent_index() returns %d\n", status);
292 return status;
293}
294
295/*
296 * Find the given page, and call find_dirent() or find_dirent_index in
297 * order to try to return the next entry.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298 */
299static inline
300int find_dirent_page(nfs_readdir_descriptor_t *desc)
301{
302 struct inode *inode = desc->file->f_dentry->d_inode;
303 struct page *page;
304 int status;
305
306 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
307
308 page = read_cache_page(inode->i_mapping, desc->page_index,
309 (filler_t *)nfs_readdir_filler, desc);
310 if (IS_ERR(page)) {
311 status = PTR_ERR(page);
312 goto out;
313 }
314 if (!PageUptodate(page))
315 goto read_error;
316
317 /* NOTE: Someone else may have changed the READDIRPLUS flag */
318 desc->page = page;
319 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000320 if (*desc->dir_cookie != 0)
Olivier Galibert00a92642005-06-22 17:16:29 +0000321 status = find_dirent(desc);
322 else
323 status = find_dirent_index(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 if (status < 0)
325 dir_page_release(desc);
326 out:
327 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
328 return status;
329 read_error:
330 page_cache_release(page);
331 return -EIO;
332}
333
334/*
335 * Recurse through the page cache pages, and return a
336 * filled nfs_entry structure of the next directory entry if possible.
337 *
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000338 * The target for the search is '*desc->dir_cookie' if non-0,
339 * 'desc->file->f_pos' otherwise
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 */
341static inline
342int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
343{
344 int loop_count = 0;
345 int res;
346
Olivier Galibert00a92642005-06-22 17:16:29 +0000347 /* Always search-by-index from the beginning of the cache */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000348 if (*desc->dir_cookie == 0) {
349 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for offset %Ld\n", (long long)desc->file->f_pos);
Olivier Galibert00a92642005-06-22 17:16:29 +0000350 desc->page_index = 0;
351 desc->entry->cookie = desc->entry->prev_cookie = 0;
352 desc->entry->eof = 0;
353 desc->current_index = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000354 } else
355 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
Olivier Galibert00a92642005-06-22 17:16:29 +0000356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 for (;;) {
358 res = find_dirent_page(desc);
359 if (res != -EAGAIN)
360 break;
361 /* Align to beginning of next page */
362 desc->page_index ++;
363 if (loop_count++ > 200) {
364 loop_count = 0;
365 schedule();
366 }
367 }
368 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
369 return res;
370}
371
372static inline unsigned int dt_type(struct inode *inode)
373{
374 return (inode->i_mode >> 12) & 15;
375}
376
377static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
378
379/*
380 * Once we've found the start of the dirent within a page: fill 'er up...
381 */
382static
383int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
384 filldir_t filldir)
385{
386 struct file *file = desc->file;
387 struct nfs_entry *entry = desc->entry;
388 struct dentry *dentry = NULL;
389 unsigned long fileid;
390 int loop_count = 0,
391 res;
392
Olivier Galibert00a92642005-06-22 17:16:29 +0000393 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)entry->cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700394
395 for(;;) {
396 unsigned d_type = DT_UNKNOWN;
397 /* Note: entry->prev_cookie contains the cookie for
398 * retrieving the current dirent on the server */
399 fileid = nfs_fileid_to_ino_t(entry->ino);
400
401 /* Get a dentry if we have one */
402 if (dentry != NULL)
403 dput(dentry);
404 dentry = nfs_readdir_lookup(desc);
405
406 /* Use readdirplus info */
407 if (dentry != NULL && dentry->d_inode != NULL) {
408 d_type = dt_type(dentry->d_inode);
409 fileid = dentry->d_inode->i_ino;
410 }
411
412 res = filldir(dirent, entry->name, entry->len,
Olivier Galibert00a92642005-06-22 17:16:29 +0000413 file->f_pos, fileid, d_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (res < 0)
415 break;
Olivier Galibert00a92642005-06-22 17:16:29 +0000416 file->f_pos++;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000417 *desc->dir_cookie = entry->cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 if (dir_decode(desc) != 0) {
419 desc->page_index ++;
420 break;
421 }
422 if (loop_count++ > 200) {
423 loop_count = 0;
424 schedule();
425 }
426 }
427 dir_page_release(desc);
428 if (dentry != NULL)
429 dput(dentry);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000430 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 return res;
432}
433
434/*
435 * If we cannot find a cookie in our cache, we suspect that this is
436 * because it points to a deleted file, so we ask the server to return
437 * whatever it thinks is the next entry. We then feed this to filldir.
438 * If all goes well, we should then be able to find our way round the
439 * cache on the next call to readdir_search_pagecache();
440 *
441 * NOTE: we cannot add the anonymous page to the pagecache because
442 * the data it contains might not be page aligned. Besides,
443 * we should already have a complete representation of the
444 * directory in the page cache by the time we get here.
445 */
446static inline
447int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
448 filldir_t filldir)
449{
450 struct file *file = desc->file;
451 struct inode *inode = file->f_dentry->d_inode;
452 struct rpc_cred *cred = nfs_file_cred(file);
453 struct page *page = NULL;
454 int status;
455
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000456 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457
458 page = alloc_page(GFP_HIGHUSER);
459 if (!page) {
460 status = -ENOMEM;
461 goto out;
462 }
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000463 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 page,
465 NFS_SERVER(inode)->dtsize,
466 desc->plus);
Chuck Leverdc592502005-08-18 11:24:12 -0700467 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -0700468 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
Chuck Leverdc592502005-08-18 11:24:12 -0700469 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 desc->page = page;
471 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
472 if (desc->error >= 0) {
473 if ((status = dir_decode(desc)) == 0)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000474 desc->entry->prev_cookie = *desc->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 } else
476 status = -EIO;
477 if (status < 0)
478 goto out_release;
479
480 status = nfs_do_filldir(desc, dirent, filldir);
481
482 /* Reset read descriptor so it searches the page cache from
483 * the start upon the next call to readdir_search_pagecache() */
484 desc->page_index = 0;
485 desc->entry->cookie = desc->entry->prev_cookie = 0;
486 desc->entry->eof = 0;
487 out:
488 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
489 return status;
490 out_release:
491 dir_page_release(desc);
492 goto out;
493}
494
Olivier Galibert00a92642005-06-22 17:16:29 +0000495/* The file offset position represents the dirent entry number. A
496 last cookie cache takes care of the common case of reading the
497 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 */
499static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
500{
501 struct dentry *dentry = filp->f_dentry;
502 struct inode *inode = dentry->d_inode;
503 nfs_readdir_descriptor_t my_desc,
504 *desc = &my_desc;
505 struct nfs_entry my_entry;
506 struct nfs_fh fh;
507 struct nfs_fattr fattr;
508 long res;
509
510 lock_kernel();
511
512 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
513 if (res < 0) {
514 unlock_kernel();
515 return res;
516 }
517
518 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000519 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000520 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000521 * to either find the entry with the appropriate number or
522 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523 */
524 memset(desc, 0, sizeof(*desc));
525
526 desc->file = filp;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000527 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 desc->decode = NFS_PROTO(inode)->decode_dirent;
529 desc->plus = NFS_USE_READDIRPLUS(inode);
530
531 my_entry.cookie = my_entry.prev_cookie = 0;
532 my_entry.eof = 0;
533 my_entry.fh = &fh;
534 my_entry.fattr = &fattr;
Trond Myklebust0e574af2005-10-27 22:12:38 -0400535 nfs_fattr_init(&fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 desc->entry = &my_entry;
537
538 while(!desc->entry->eof) {
539 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000540
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541 if (res == -EBADCOOKIE) {
542 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000543 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 /* Or that the server has 'lost' a cookie */
545 res = uncached_readdir(desc, dirent, filldir);
546 if (res >= 0)
547 continue;
548 }
549 res = 0;
550 break;
551 }
552 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700553 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 nfs_zap_caches(inode);
555 desc->plus = 0;
556 desc->entry->eof = 0;
557 continue;
558 }
559 if (res < 0)
560 break;
561
562 res = nfs_do_filldir(desc, dirent, filldir);
563 if (res < 0) {
564 res = 0;
565 break;
566 }
567 }
568 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 if (res < 0)
570 return res;
571 return 0;
572}
573
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000574loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
575{
576 down(&filp->f_dentry->d_inode->i_sem);
577 switch (origin) {
578 case 1:
579 offset += filp->f_pos;
580 case 0:
581 if (offset >= 0)
582 break;
583 default:
584 offset = -EINVAL;
585 goto out;
586 }
587 if (offset != filp->f_pos) {
588 filp->f_pos = offset;
589 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
590 }
591out:
592 up(&filp->f_dentry->d_inode->i_sem);
593 return offset;
594}
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596/*
597 * All directory operations under NFS are synchronous, so fsync()
598 * is a dummy operation.
599 */
600int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
601{
602 return 0;
603}
604
605/*
606 * A check for whether or not the parent directory has changed.
607 * In the case it has, we assume that the dentries are untrustworthy
608 * and may need to be looked up again.
609 */
610static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
611{
612 if (IS_ROOT(dentry))
613 return 1;
Chuck Lever55296802005-08-18 11:24:09 -0700614 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 || nfs_attribute_timeout(dir))
616 return 0;
617 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
618}
619
620static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
621{
622 dentry->d_fsdata = (void *)verf;
623}
624
625/*
626 * Whenever an NFS operation succeeds, we know that the dentry
627 * is valid, so we update the revalidation timestamp.
628 */
629static inline void nfs_renew_times(struct dentry * dentry)
630{
631 dentry->d_time = jiffies;
632}
633
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400634/*
635 * Return the intent data that applies to this particular path component
636 *
637 * Note that the current set of intents only apply to the very last
638 * component of the path.
639 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
640 */
641static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
642{
643 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
644 return 0;
645 return nd->flags & mask;
646}
647
648/*
649 * Inode and filehandle revalidation for lookups.
650 *
651 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
652 * or if the intent information indicates that we're about to open this
653 * particular file and the "nocto" mount flag is not set.
654 *
655 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656static inline
657int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
658{
659 struct nfs_server *server = NFS_SERVER(inode);
660
661 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400663 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 goto out_force;
665 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400666 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 !(server->flags & NFS_MOUNT_NOCTO))
668 goto out_force;
669 }
670 return nfs_revalidate_inode(server, inode);
671out_force:
672 return __nfs_revalidate_inode(server, inode);
673}
674
675/*
676 * We judge how long we want to trust negative
677 * dentries by looking at the parent inode mtime.
678 *
679 * If parent mtime has changed, we revalidate, else we wait for a
680 * period corresponding to the parent's attribute cache timeout value.
681 */
682static inline
683int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
684 struct nameidata *nd)
685{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400687 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688 return 0;
689 return !nfs_check_verifier(dir, dentry);
690}
691
692/*
693 * This is called every time the dcache has a lookup hit,
694 * and we should check whether we can really trust that
695 * lookup.
696 *
697 * NOTE! The hit can be a negative hit too, don't assume
698 * we have an inode!
699 *
700 * If the parent directory is seen to have changed, we throw out the
701 * cached dentry and do a new lookup.
702 */
703static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
704{
705 struct inode *dir;
706 struct inode *inode;
707 struct dentry *parent;
708 int error;
709 struct nfs_fh fhandle;
710 struct nfs_fattr fattr;
711 unsigned long verifier;
712
713 parent = dget_parent(dentry);
714 lock_kernel();
715 dir = parent->d_inode;
716 inode = dentry->d_inode;
717
718 if (!inode) {
719 if (nfs_neg_need_reval(dir, dentry, nd))
720 goto out_bad;
721 goto out_valid;
722 }
723
724 if (is_bad_inode(inode)) {
725 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
726 dentry->d_parent->d_name.name, dentry->d_name.name);
727 goto out_bad;
728 }
729
730 /* Revalidate parent directory attribute cache */
731 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
732 goto out_zap_parent;
733
734 /* Force a full look up iff the parent directory has changed */
735 if (nfs_check_verifier(dir, dentry)) {
736 if (nfs_lookup_verify_inode(inode, nd))
737 goto out_zap_parent;
738 goto out_valid;
739 }
740
741 if (NFS_STALE(inode))
742 goto out_bad;
743
744 verifier = nfs_save_change_attribute(dir);
745 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
746 if (error)
747 goto out_bad;
748 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
749 goto out_bad;
750 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
751 goto out_bad;
752
753 nfs_renew_times(dentry);
754 nfs_set_verifier(dentry, verifier);
755 out_valid:
756 unlock_kernel();
757 dput(parent);
758 return 1;
759out_zap_parent:
760 nfs_zap_caches(dir);
761 out_bad:
762 NFS_CACHEINV(dir);
763 if (inode && S_ISDIR(inode->i_mode)) {
764 /* Purge readdir caches. */
765 nfs_zap_caches(inode);
766 /* If we have submounts, don't unhash ! */
767 if (have_submounts(dentry))
768 goto out_valid;
769 shrink_dcache_parent(dentry);
770 }
771 d_drop(dentry);
772 unlock_kernel();
773 dput(parent);
774 return 0;
775}
776
777/*
778 * This is called from dput() when d_count is going to 0.
779 */
780static int nfs_dentry_delete(struct dentry *dentry)
781{
782 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
783 dentry->d_parent->d_name.name, dentry->d_name.name,
784 dentry->d_flags);
785
786 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
787 /* Unhash it, so that ->d_iput() would be called */
788 return 1;
789 }
790 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
791 /* Unhash it, so that ancestors of killed async unlink
792 * files will be cleaned up during umount */
793 return 1;
794 }
795 return 0;
796
797}
798
799/*
800 * Called when the dentry loses inode.
801 * We use it to clean up silly-renamed files.
802 */
803static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
804{
Trond Myklebustcae7a072005-10-18 14:20:19 -0700805 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
807 lock_kernel();
808 inode->i_nlink--;
809 nfs_complete_unlink(dentry);
810 unlock_kernel();
811 }
812 /* When creating a negative dentry, we want to renew d_time */
813 nfs_renew_times(dentry);
814 iput(inode);
815}
816
817struct dentry_operations nfs_dentry_operations = {
818 .d_revalidate = nfs_lookup_revalidate,
819 .d_delete = nfs_dentry_delete,
820 .d_iput = nfs_dentry_iput,
821};
822
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400823/*
824 * Use intent information to check whether or not we're going to do
825 * an O_EXCL create using this path component.
826 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827static inline
828int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
829{
830 if (NFS_PROTO(dir)->version == 2)
831 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400832 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 return 0;
834 return (nd->intent.open.flags & O_EXCL) != 0;
835}
836
837static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
838{
839 struct dentry *res;
840 struct inode *inode = NULL;
841 int error;
842 struct nfs_fh fhandle;
843 struct nfs_fattr fattr;
844
845 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
846 dentry->d_parent->d_name.name, dentry->d_name.name);
847
848 res = ERR_PTR(-ENAMETOOLONG);
849 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
850 goto out;
851
852 res = ERR_PTR(-ENOMEM);
853 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
854
855 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856
857 /* If we're doing an exclusive create, optimize away the lookup */
858 if (nfs_is_exclusive_create(dir, nd))
859 goto no_entry;
860
861 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
862 if (error == -ENOENT)
863 goto no_entry;
864 if (error < 0) {
865 res = ERR_PTR(error);
866 goto out_unlock;
867 }
868 res = ERR_PTR(-EACCES);
869 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
870 if (!inode)
871 goto out_unlock;
872no_entry:
873 res = d_add_unique(dentry, inode);
874 if (res != NULL)
875 dentry = res;
876 nfs_renew_times(dentry);
877 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
878out_unlock:
879 unlock_kernel();
880out:
881 return res;
882}
883
884#ifdef CONFIG_NFS_V4
885static int nfs_open_revalidate(struct dentry *, struct nameidata *);
886
887struct dentry_operations nfs4_dentry_operations = {
888 .d_revalidate = nfs_open_revalidate,
889 .d_delete = nfs_dentry_delete,
890 .d_iput = nfs_dentry_iput,
891};
892
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400893/*
894 * Use intent information to determine whether we need to substitute
895 * the NFSv4-style stateful OPEN for the LOOKUP call
896 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700897static int is_atomic_open(struct inode *dir, struct nameidata *nd)
898{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400899 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 return 0;
901 /* NFS does not (yet) have a stateful open for directories */
902 if (nd->flags & LOOKUP_DIRECTORY)
903 return 0;
904 /* Are we trying to write to a read only partition? */
905 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
906 return 0;
907 return 1;
908}
909
910static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
911{
912 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 int error;
914
915 /* Check that we are indeed trying to open this file */
916 if (!is_atomic_open(dir, nd))
917 goto no_open;
918
919 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
920 res = ERR_PTR(-ENAMETOOLONG);
921 goto out;
922 }
923 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
924
925 /* Let vfs_create() deal with O_EXCL */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700926 if (nd->intent.open.flags & O_EXCL) {
927 d_add(dentry, NULL);
928 goto out;
929 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930
931 /* Open the file on the server */
932 lock_kernel();
933 /* Revalidate parent directory attribute cache */
934 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
935 if (error < 0) {
936 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -0700937 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 goto out;
939 }
940
941 if (nd->intent.open.flags & O_CREAT) {
942 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -0700943 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 nfs_end_data_update(dir);
945 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -0700946 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -0700948 if (IS_ERR(res)) {
949 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 switch (error) {
951 /* Make a negative dentry */
952 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -0700953 res = NULL;
954 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -0700956 case -EISDIR:
957 case -ENOTDIR:
958 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 case -ELOOP:
960 if (!(nd->intent.open.flags & O_NOFOLLOW))
961 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 /* case -EINVAL: */
963 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964 goto out;
965 }
Trond Myklebust02a913a2005-10-18 14:20:17 -0700966 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 dentry = res;
968 nfs_renew_times(dentry);
969 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
970out:
971 return res;
972no_open:
973 return nfs_lookup(dir, dentry, nd);
974}
975
976static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
977{
978 struct dentry *parent = NULL;
979 struct inode *inode = dentry->d_inode;
980 struct inode *dir;
981 unsigned long verifier;
982 int openflags, ret = 0;
983
984 parent = dget_parent(dentry);
985 dir = parent->d_inode;
986 if (!is_atomic_open(dir, nd))
987 goto no_open;
988 /* We can't create new files in nfs_open_revalidate(), so we
989 * optimize away revalidation of negative dentries.
990 */
991 if (inode == NULL)
992 goto out;
993 /* NFS only supports OPEN on regular files */
994 if (!S_ISREG(inode->i_mode))
995 goto no_open;
996 openflags = nd->intent.open.flags;
997 /* We cannot do exclusive creation on a positive dentry */
998 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
999 goto no_open;
1000 /* We can't create new files, or truncate existing ones here */
1001 openflags &= ~(O_CREAT|O_TRUNC);
1002
1003 /*
1004 * Note: we're not holding inode->i_sem and so may be racing with
1005 * operations that change the directory. We therefore save the
1006 * change attribute *before* we do the RPC call.
1007 */
1008 lock_kernel();
1009 verifier = nfs_save_change_attribute(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001010 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 if (!ret)
1012 nfs_set_verifier(dentry, verifier);
1013 unlock_kernel();
1014out:
1015 dput(parent);
1016 if (!ret)
1017 d_drop(dentry);
1018 return ret;
1019no_open:
1020 dput(parent);
1021 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1022 return 1;
1023 return nfs_lookup_revalidate(dentry, nd);
1024}
1025#endif /* CONFIG_NFSV4 */
1026
1027static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1028{
1029 struct dentry *parent = desc->file->f_dentry;
1030 struct inode *dir = parent->d_inode;
1031 struct nfs_entry *entry = desc->entry;
1032 struct dentry *dentry, *alias;
1033 struct qstr name = {
1034 .name = entry->name,
1035 .len = entry->len,
1036 };
1037 struct inode *inode;
1038
1039 switch (name.len) {
1040 case 2:
1041 if (name.name[0] == '.' && name.name[1] == '.')
1042 return dget_parent(parent);
1043 break;
1044 case 1:
1045 if (name.name[0] == '.')
1046 return dget(parent);
1047 }
1048 name.hash = full_name_hash(name.name, name.len);
1049 dentry = d_lookup(parent, &name);
1050 if (dentry != NULL)
1051 return dentry;
1052 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1053 return NULL;
1054 /* Note: caller is already holding the dir->i_sem! */
1055 dentry = d_alloc(parent, &name);
1056 if (dentry == NULL)
1057 return NULL;
1058 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1059 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1060 if (!inode) {
1061 dput(dentry);
1062 return NULL;
1063 }
1064 alias = d_add_unique(dentry, inode);
1065 if (alias != NULL) {
1066 dput(dentry);
1067 dentry = alias;
1068 }
1069 nfs_renew_times(dentry);
1070 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1071 return dentry;
1072}
1073
1074/*
1075 * Code common to create, mkdir, and mknod.
1076 */
1077int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1078 struct nfs_fattr *fattr)
1079{
1080 struct inode *inode;
1081 int error = -EACCES;
1082
1083 /* We may have been initialized further down */
1084 if (dentry->d_inode)
1085 return 0;
1086 if (fhandle->size == 0) {
1087 struct inode *dir = dentry->d_parent->d_inode;
1088 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1089 if (error)
1090 goto out_err;
1091 }
1092 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1093 struct nfs_server *server = NFS_SB(dentry->d_sb);
1094 error = server->rpc_ops->getattr(server, fhandle, fattr);
1095 if (error < 0)
1096 goto out_err;
1097 }
1098 error = -ENOMEM;
1099 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1100 if (inode == NULL)
1101 goto out_err;
1102 d_instantiate(dentry, inode);
1103 return 0;
1104out_err:
1105 d_drop(dentry);
1106 return error;
1107}
1108
1109/*
1110 * Following a failed create operation, we drop the dentry rather
1111 * than retain a negative dentry. This avoids a problem in the event
1112 * that the operation succeeded on the server, but an error in the
1113 * reply path made it appear to have failed.
1114 */
1115static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1116 struct nameidata *nd)
1117{
1118 struct iattr attr;
1119 int error;
1120 int open_flags = 0;
1121
1122 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1123 dir->i_ino, dentry->d_name.name);
1124
1125 attr.ia_mode = mode;
1126 attr.ia_valid = ATTR_MODE;
1127
1128 if (nd && (nd->flags & LOOKUP_CREATE))
1129 open_flags = nd->intent.open.flags;
1130
1131 lock_kernel();
1132 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001133 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 nfs_end_data_update(dir);
1135 if (error != 0)
1136 goto out_err;
1137 nfs_renew_times(dentry);
1138 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1139 unlock_kernel();
1140 return 0;
1141out_err:
1142 unlock_kernel();
1143 d_drop(dentry);
1144 return error;
1145}
1146
1147/*
1148 * See comments for nfs_proc_create regarding failed operations.
1149 */
1150static int
1151nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1152{
1153 struct iattr attr;
1154 int status;
1155
1156 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1157 dir->i_ino, dentry->d_name.name);
1158
1159 if (!new_valid_dev(rdev))
1160 return -EINVAL;
1161
1162 attr.ia_mode = mode;
1163 attr.ia_valid = ATTR_MODE;
1164
1165 lock_kernel();
1166 nfs_begin_data_update(dir);
1167 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1168 nfs_end_data_update(dir);
1169 if (status != 0)
1170 goto out_err;
1171 nfs_renew_times(dentry);
1172 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1173 unlock_kernel();
1174 return 0;
1175out_err:
1176 unlock_kernel();
1177 d_drop(dentry);
1178 return status;
1179}
1180
1181/*
1182 * See comments for nfs_proc_create regarding failed operations.
1183 */
1184static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1185{
1186 struct iattr attr;
1187 int error;
1188
1189 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1190 dir->i_ino, dentry->d_name.name);
1191
1192 attr.ia_valid = ATTR_MODE;
1193 attr.ia_mode = mode | S_IFDIR;
1194
1195 lock_kernel();
1196 nfs_begin_data_update(dir);
1197 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1198 nfs_end_data_update(dir);
1199 if (error != 0)
1200 goto out_err;
1201 nfs_renew_times(dentry);
1202 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1203 unlock_kernel();
1204 return 0;
1205out_err:
1206 d_drop(dentry);
1207 unlock_kernel();
1208 return error;
1209}
1210
1211static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1212{
1213 int error;
1214
1215 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1216 dir->i_ino, dentry->d_name.name);
1217
1218 lock_kernel();
1219 nfs_begin_data_update(dir);
1220 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1221 /* Ensure the VFS deletes this inode */
1222 if (error == 0 && dentry->d_inode != NULL)
1223 dentry->d_inode->i_nlink = 0;
1224 nfs_end_data_update(dir);
1225 unlock_kernel();
1226
1227 return error;
1228}
1229
1230static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1231{
1232 static unsigned int sillycounter;
1233 const int i_inosize = sizeof(dir->i_ino)*2;
1234 const int countersize = sizeof(sillycounter)*2;
1235 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1236 char silly[slen+1];
1237 struct qstr qsilly;
1238 struct dentry *sdentry;
1239 int error = -EIO;
1240
1241 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1242 dentry->d_parent->d_name.name, dentry->d_name.name,
1243 atomic_read(&dentry->d_count));
1244
1245#ifdef NFS_PARANOIA
1246if (!dentry->d_inode)
1247printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1248dentry->d_parent->d_name.name, dentry->d_name.name);
1249#endif
1250 /*
1251 * We don't allow a dentry to be silly-renamed twice.
1252 */
1253 error = -EBUSY;
1254 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1255 goto out;
1256
1257 sprintf(silly, ".nfs%*.*lx",
1258 i_inosize, i_inosize, dentry->d_inode->i_ino);
1259
Trond Myklebust34ea8182005-11-04 15:35:02 -05001260 /* Return delegation in anticipation of the rename */
1261 nfs_inode_return_delegation(dentry->d_inode);
1262
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 sdentry = NULL;
1264 do {
1265 char *suffix = silly + slen - countersize;
1266
1267 dput(sdentry);
1268 sillycounter++;
1269 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1270
1271 dfprintk(VFS, "trying to rename %s to %s\n",
1272 dentry->d_name.name, silly);
1273
1274 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1275 /*
1276 * N.B. Better to return EBUSY here ... it could be
1277 * dangerous to delete the file while it's in use.
1278 */
1279 if (IS_ERR(sdentry))
1280 goto out;
1281 } while(sdentry->d_inode != NULL); /* need negative lookup */
1282
1283 qsilly.name = silly;
1284 qsilly.len = strlen(silly);
1285 nfs_begin_data_update(dir);
1286 if (dentry->d_inode) {
1287 nfs_begin_data_update(dentry->d_inode);
1288 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1289 dir, &qsilly);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001290 nfs_mark_for_revalidate(dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 nfs_end_data_update(dentry->d_inode);
1292 } else
1293 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1294 dir, &qsilly);
1295 nfs_end_data_update(dir);
1296 if (!error) {
1297 nfs_renew_times(dentry);
1298 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1299 d_move(dentry, sdentry);
1300 error = nfs_async_unlink(dentry);
1301 /* If we return 0 we don't unlink */
1302 }
1303 dput(sdentry);
1304out:
1305 return error;
1306}
1307
1308/*
1309 * Remove a file after making sure there are no pending writes,
1310 * and after checking that the file has only one user.
1311 *
1312 * We invalidate the attribute cache and free the inode prior to the operation
1313 * to avoid possible races if the server reuses the inode.
1314 */
1315static int nfs_safe_remove(struct dentry *dentry)
1316{
1317 struct inode *dir = dentry->d_parent->d_inode;
1318 struct inode *inode = dentry->d_inode;
1319 int error = -EBUSY;
1320
1321 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1322 dentry->d_parent->d_name.name, dentry->d_name.name);
1323
1324 /* If the dentry was sillyrenamed, we simply call d_delete() */
1325 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1326 error = 0;
1327 goto out;
1328 }
1329
1330 nfs_begin_data_update(dir);
1331 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001332 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 nfs_begin_data_update(inode);
1334 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1335 /* The VFS may want to delete this inode */
1336 if (error == 0)
1337 inode->i_nlink--;
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001338 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 nfs_end_data_update(inode);
1340 } else
1341 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1342 nfs_end_data_update(dir);
1343out:
1344 return error;
1345}
1346
1347/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1348 * belongs to an active ".nfs..." file and we return -EBUSY.
1349 *
1350 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1351 */
1352static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1353{
1354 int error;
1355 int need_rehash = 0;
1356
1357 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1358 dir->i_ino, dentry->d_name.name);
1359
1360 lock_kernel();
1361 spin_lock(&dcache_lock);
1362 spin_lock(&dentry->d_lock);
1363 if (atomic_read(&dentry->d_count) > 1) {
1364 spin_unlock(&dentry->d_lock);
1365 spin_unlock(&dcache_lock);
1366 error = nfs_sillyrename(dir, dentry);
1367 unlock_kernel();
1368 return error;
1369 }
1370 if (!d_unhashed(dentry)) {
1371 __d_drop(dentry);
1372 need_rehash = 1;
1373 }
1374 spin_unlock(&dentry->d_lock);
1375 spin_unlock(&dcache_lock);
1376 error = nfs_safe_remove(dentry);
1377 if (!error) {
1378 nfs_renew_times(dentry);
1379 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1380 } else if (need_rehash)
1381 d_rehash(dentry);
1382 unlock_kernel();
1383 return error;
1384}
1385
1386static int
1387nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1388{
1389 struct iattr attr;
1390 struct nfs_fattr sym_attr;
1391 struct nfs_fh sym_fh;
1392 struct qstr qsymname;
1393 int error;
1394
1395 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1396 dir->i_ino, dentry->d_name.name, symname);
1397
1398#ifdef NFS_PARANOIA
1399if (dentry->d_inode)
1400printk("nfs_proc_symlink: %s/%s not negative!\n",
1401dentry->d_parent->d_name.name, dentry->d_name.name);
1402#endif
1403 /*
1404 * Fill in the sattr for the call.
1405 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1406 */
1407 attr.ia_valid = ATTR_MODE;
1408 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1409
1410 qsymname.name = symname;
1411 qsymname.len = strlen(symname);
1412
1413 lock_kernel();
1414 nfs_begin_data_update(dir);
1415 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1416 &attr, &sym_fh, &sym_attr);
1417 nfs_end_data_update(dir);
1418 if (!error) {
1419 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1420 } else {
1421 if (error == -EEXIST)
1422 printk("nfs_proc_symlink: %s/%s already exists??\n",
1423 dentry->d_parent->d_name.name, dentry->d_name.name);
1424 d_drop(dentry);
1425 }
1426 unlock_kernel();
1427 return error;
1428}
1429
1430static int
1431nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1432{
1433 struct inode *inode = old_dentry->d_inode;
1434 int error;
1435
1436 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1437 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1438 dentry->d_parent->d_name.name, dentry->d_name.name);
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 nfs_begin_data_update(dir);
1442 nfs_begin_data_update(inode);
1443 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001444 if (error == 0) {
1445 atomic_inc(&inode->i_count);
1446 d_instantiate(dentry, inode);
1447 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 nfs_end_data_update(inode);
1449 nfs_end_data_update(dir);
1450 unlock_kernel();
1451 return error;
1452}
1453
1454/*
1455 * RENAME
1456 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1457 * different file handle for the same inode after a rename (e.g. when
1458 * moving to a different directory). A fail-safe method to do so would
1459 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1460 * rename the old file using the sillyrename stuff. This way, the original
1461 * file in old_dir will go away when the last process iput()s the inode.
1462 *
1463 * FIXED.
1464 *
1465 * It actually works quite well. One needs to have the possibility for
1466 * at least one ".nfs..." file in each directory the file ever gets
1467 * moved or linked to which happens automagically with the new
1468 * implementation that only depends on the dcache stuff instead of
1469 * using the inode layer
1470 *
1471 * Unfortunately, things are a little more complicated than indicated
1472 * above. For a cross-directory move, we want to make sure we can get
1473 * rid of the old inode after the operation. This means there must be
1474 * no pending writes (if it's a file), and the use count must be 1.
1475 * If these conditions are met, we can drop the dentries before doing
1476 * the rename.
1477 */
1478static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1479 struct inode *new_dir, struct dentry *new_dentry)
1480{
1481 struct inode *old_inode = old_dentry->d_inode;
1482 struct inode *new_inode = new_dentry->d_inode;
1483 struct dentry *dentry = NULL, *rehash = NULL;
1484 int error = -EBUSY;
1485
1486 /*
1487 * To prevent any new references to the target during the rename,
1488 * we unhash the dentry and free the inode in advance.
1489 */
1490 lock_kernel();
1491 if (!d_unhashed(new_dentry)) {
1492 d_drop(new_dentry);
1493 rehash = new_dentry;
1494 }
1495
1496 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1497 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1498 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1499 atomic_read(&new_dentry->d_count));
1500
1501 /*
1502 * First check whether the target is busy ... we can't
1503 * safely do _any_ rename if the target is in use.
1504 *
1505 * For files, make a copy of the dentry and then do a
1506 * silly-rename. If the silly-rename succeeds, the
1507 * copied dentry is hashed and becomes the new target.
1508 */
1509 if (!new_inode)
1510 goto go_ahead;
Trond Myklebust6fe43f92005-10-18 14:20:22 -07001511 if (S_ISDIR(new_inode->i_mode)) {
1512 error = -EISDIR;
1513 if (!S_ISDIR(old_inode->i_mode))
1514 goto out;
1515 } else if (atomic_read(&new_dentry->d_count) > 2) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 int err;
1517 /* copy the target dentry's name */
1518 dentry = d_alloc(new_dentry->d_parent,
1519 &new_dentry->d_name);
1520 if (!dentry)
1521 goto out;
1522
1523 /* silly-rename the existing target ... */
1524 err = nfs_sillyrename(new_dir, new_dentry);
1525 if (!err) {
1526 new_dentry = rehash = dentry;
1527 new_inode = NULL;
1528 /* instantiate the replacement target */
1529 d_instantiate(new_dentry, NULL);
1530 } else if (atomic_read(&new_dentry->d_count) > 1) {
1531 /* dentry still busy? */
1532#ifdef NFS_PARANOIA
1533 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1534 new_dentry->d_parent->d_name.name,
1535 new_dentry->d_name.name,
1536 atomic_read(&new_dentry->d_count));
1537#endif
1538 goto out;
1539 }
Trond Myklebust20509f12005-08-25 16:25:34 -07001540 } else
1541 new_inode->i_nlink--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
1543go_ahead:
1544 /*
1545 * ... prune child dentries and writebacks if needed.
1546 */
1547 if (atomic_read(&old_dentry->d_count) > 1) {
1548 nfs_wb_all(old_inode);
1549 shrink_dcache_parent(old_dentry);
1550 }
Trond Myklebustcae7a072005-10-18 14:20:19 -07001551 nfs_inode_return_delegation(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552
Trond Myklebust24174112006-01-03 09:55:33 +01001553 if (new_inode != NULL) {
1554 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 d_delete(new_dentry);
Trond Myklebust24174112006-01-03 09:55:33 +01001556 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557
1558 nfs_begin_data_update(old_dir);
1559 nfs_begin_data_update(new_dir);
1560 nfs_begin_data_update(old_inode);
1561 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1562 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001563 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 nfs_end_data_update(old_inode);
1565 nfs_end_data_update(new_dir);
1566 nfs_end_data_update(old_dir);
1567out:
1568 if (rehash)
1569 d_rehash(rehash);
1570 if (!error) {
1571 if (!S_ISDIR(old_inode->i_mode))
1572 d_move(old_dentry, new_dentry);
1573 nfs_renew_times(new_dentry);
1574 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1575 }
1576
1577 /* new dentry created? */
1578 if (dentry)
1579 dput(dentry);
1580 unlock_kernel();
1581 return error;
1582}
1583
1584int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1585{
Chuck Lever55296802005-08-18 11:24:09 -07001586 struct nfs_inode *nfsi = NFS_I(inode);
1587 struct nfs_access_entry *cache = &nfsi->cache_access;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588
1589 if (cache->cred != cred
1590 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
Chuck Lever55296802005-08-18 11:24:09 -07001591 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 return -ENOENT;
1593 memcpy(res, cache, sizeof(*res));
1594 return 0;
1595}
1596
1597void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1598{
Chuck Lever55296802005-08-18 11:24:09 -07001599 struct nfs_inode *nfsi = NFS_I(inode);
1600 struct nfs_access_entry *cache = &nfsi->cache_access;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601
1602 if (cache->cred != set->cred) {
1603 if (cache->cred)
1604 put_rpccred(cache->cred);
1605 cache->cred = get_rpccred(set->cred);
1606 }
Chuck Leverdc592502005-08-18 11:24:12 -07001607 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1608 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -07001609 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
Chuck Leverdc592502005-08-18 11:24:12 -07001610 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 cache->jiffies = set->jiffies;
1612 cache->mask = set->mask;
1613}
1614
1615static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1616{
1617 struct nfs_access_entry cache;
1618 int status;
1619
1620 status = nfs_access_get_cached(inode, cred, &cache);
1621 if (status == 0)
1622 goto out;
1623
1624 /* Be clever: ask server to check for all possible rights */
1625 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1626 cache.cred = cred;
1627 cache.jiffies = jiffies;
1628 status = NFS_PROTO(inode)->access(inode, &cache);
1629 if (status != 0)
1630 return status;
1631 nfs_access_add_cache(inode, &cache);
1632out:
1633 if ((cache.mask & mask) == mask)
1634 return 0;
1635 return -EACCES;
1636}
1637
1638int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1639{
1640 struct rpc_cred *cred;
1641 int res = 0;
1642
1643 if (mask == 0)
1644 goto out;
1645 /* Is this sys_access() ? */
1646 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1647 goto force_lookup;
1648
1649 switch (inode->i_mode & S_IFMT) {
1650 case S_IFLNK:
1651 goto out;
1652 case S_IFREG:
1653 /* NFSv4 has atomic_open... */
1654 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1655 && nd != NULL
1656 && (nd->flags & LOOKUP_OPEN))
1657 goto out;
1658 break;
1659 case S_IFDIR:
1660 /*
1661 * Optimize away all write operations, since the server
1662 * will check permissions when we perform the op.
1663 */
1664 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1665 goto out;
1666 }
1667
1668force_lookup:
1669 lock_kernel();
1670
1671 if (!NFS_PROTO(inode)->access)
1672 goto out_notsup;
1673
1674 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1675 if (!IS_ERR(cred)) {
1676 res = nfs_do_access(inode, cred, mask);
1677 put_rpccred(cred);
1678 } else
1679 res = PTR_ERR(cred);
1680 unlock_kernel();
1681out:
1682 return res;
1683out_notsup:
1684 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1685 if (res == 0)
1686 res = generic_permission(inode, mask, NULL);
1687 unlock_kernel();
1688 return res;
1689}
1690
1691/*
1692 * Local variables:
1693 * version-control: t
1694 * kept-new-versions: 5
1695 * End:
1696 */