blob: ac331d2d4c4ae047eec5bc00b7c5df303c9c1386 [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;
535 desc->entry = &my_entry;
536
537 while(!desc->entry->eof) {
538 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000539
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 if (res == -EBADCOOKIE) {
541 /* This means either end of directory */
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000542 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 /* Or that the server has 'lost' a cookie */
544 res = uncached_readdir(desc, dirent, filldir);
545 if (res >= 0)
546 continue;
547 }
548 res = 0;
549 break;
550 }
551 if (res == -ETOOSMALL && desc->plus) {
Chuck Lever412d5822005-08-18 11:24:11 -0700552 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 nfs_zap_caches(inode);
554 desc->plus = 0;
555 desc->entry->eof = 0;
556 continue;
557 }
558 if (res < 0)
559 break;
560
561 res = nfs_do_filldir(desc, dirent, filldir);
562 if (res < 0) {
563 res = 0;
564 break;
565 }
566 }
567 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 if (res < 0)
569 return res;
570 return 0;
571}
572
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000573loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
574{
575 down(&filp->f_dentry->d_inode->i_sem);
576 switch (origin) {
577 case 1:
578 offset += filp->f_pos;
579 case 0:
580 if (offset >= 0)
581 break;
582 default:
583 offset = -EINVAL;
584 goto out;
585 }
586 if (offset != filp->f_pos) {
587 filp->f_pos = offset;
588 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
589 }
590out:
591 up(&filp->f_dentry->d_inode->i_sem);
592 return offset;
593}
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595/*
596 * All directory operations under NFS are synchronous, so fsync()
597 * is a dummy operation.
598 */
599int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
600{
601 return 0;
602}
603
604/*
605 * A check for whether or not the parent directory has changed.
606 * In the case it has, we assume that the dentries are untrustworthy
607 * and may need to be looked up again.
608 */
609static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
610{
611 if (IS_ROOT(dentry))
612 return 1;
Chuck Lever55296802005-08-18 11:24:09 -0700613 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 || nfs_attribute_timeout(dir))
615 return 0;
616 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
617}
618
619static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
620{
621 dentry->d_fsdata = (void *)verf;
622}
623
624/*
625 * Whenever an NFS operation succeeds, we know that the dentry
626 * is valid, so we update the revalidation timestamp.
627 */
628static inline void nfs_renew_times(struct dentry * dentry)
629{
630 dentry->d_time = jiffies;
631}
632
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400633/*
634 * Return the intent data that applies to this particular path component
635 *
636 * Note that the current set of intents only apply to the very last
637 * component of the path.
638 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
639 */
640static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
641{
642 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
643 return 0;
644 return nd->flags & mask;
645}
646
647/*
648 * Inode and filehandle revalidation for lookups.
649 *
650 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
651 * or if the intent information indicates that we're about to open this
652 * particular file and the "nocto" mount flag is not set.
653 *
654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655static inline
656int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
657{
658 struct nfs_server *server = NFS_SERVER(inode);
659
660 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400662 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 goto out_force;
664 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400665 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 !(server->flags & NFS_MOUNT_NOCTO))
667 goto out_force;
668 }
669 return nfs_revalidate_inode(server, inode);
670out_force:
671 return __nfs_revalidate_inode(server, inode);
672}
673
674/*
675 * We judge how long we want to trust negative
676 * dentries by looking at the parent inode mtime.
677 *
678 * If parent mtime has changed, we revalidate, else we wait for a
679 * period corresponding to the parent's attribute cache timeout value.
680 */
681static inline
682int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
683 struct nameidata *nd)
684{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400686 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 return 0;
688 return !nfs_check_verifier(dir, dentry);
689}
690
691/*
692 * This is called every time the dcache has a lookup hit,
693 * and we should check whether we can really trust that
694 * lookup.
695 *
696 * NOTE! The hit can be a negative hit too, don't assume
697 * we have an inode!
698 *
699 * If the parent directory is seen to have changed, we throw out the
700 * cached dentry and do a new lookup.
701 */
702static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
703{
704 struct inode *dir;
705 struct inode *inode;
706 struct dentry *parent;
707 int error;
708 struct nfs_fh fhandle;
709 struct nfs_fattr fattr;
710 unsigned long verifier;
711
712 parent = dget_parent(dentry);
713 lock_kernel();
714 dir = parent->d_inode;
715 inode = dentry->d_inode;
716
717 if (!inode) {
718 if (nfs_neg_need_reval(dir, dentry, nd))
719 goto out_bad;
720 goto out_valid;
721 }
722
723 if (is_bad_inode(inode)) {
724 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
725 dentry->d_parent->d_name.name, dentry->d_name.name);
726 goto out_bad;
727 }
728
729 /* Revalidate parent directory attribute cache */
730 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
731 goto out_zap_parent;
732
733 /* Force a full look up iff the parent directory has changed */
734 if (nfs_check_verifier(dir, dentry)) {
735 if (nfs_lookup_verify_inode(inode, nd))
736 goto out_zap_parent;
737 goto out_valid;
738 }
739
740 if (NFS_STALE(inode))
741 goto out_bad;
742
743 verifier = nfs_save_change_attribute(dir);
744 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
745 if (error)
746 goto out_bad;
747 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
748 goto out_bad;
749 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
750 goto out_bad;
751
752 nfs_renew_times(dentry);
753 nfs_set_verifier(dentry, verifier);
754 out_valid:
755 unlock_kernel();
756 dput(parent);
757 return 1;
758out_zap_parent:
759 nfs_zap_caches(dir);
760 out_bad:
761 NFS_CACHEINV(dir);
762 if (inode && S_ISDIR(inode->i_mode)) {
763 /* Purge readdir caches. */
764 nfs_zap_caches(inode);
765 /* If we have submounts, don't unhash ! */
766 if (have_submounts(dentry))
767 goto out_valid;
768 shrink_dcache_parent(dentry);
769 }
770 d_drop(dentry);
771 unlock_kernel();
772 dput(parent);
773 return 0;
774}
775
776/*
777 * This is called from dput() when d_count is going to 0.
778 */
779static int nfs_dentry_delete(struct dentry *dentry)
780{
781 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
782 dentry->d_parent->d_name.name, dentry->d_name.name,
783 dentry->d_flags);
784
785 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
786 /* Unhash it, so that ->d_iput() would be called */
787 return 1;
788 }
789 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
790 /* Unhash it, so that ancestors of killed async unlink
791 * files will be cleaned up during umount */
792 return 1;
793 }
794 return 0;
795
796}
797
798/*
799 * Called when the dentry loses inode.
800 * We use it to clean up silly-renamed files.
801 */
802static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
803{
804 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
805 lock_kernel();
806 inode->i_nlink--;
807 nfs_complete_unlink(dentry);
808 unlock_kernel();
809 }
810 /* When creating a negative dentry, we want to renew d_time */
811 nfs_renew_times(dentry);
812 iput(inode);
813}
814
815struct dentry_operations nfs_dentry_operations = {
816 .d_revalidate = nfs_lookup_revalidate,
817 .d_delete = nfs_dentry_delete,
818 .d_iput = nfs_dentry_iput,
819};
820
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400821/*
822 * Use intent information to check whether or not we're going to do
823 * an O_EXCL create using this path component.
824 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825static inline
826int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
827{
828 if (NFS_PROTO(dir)->version == 2)
829 return 0;
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400830 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return 0;
832 return (nd->intent.open.flags & O_EXCL) != 0;
833}
834
835static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
836{
837 struct dentry *res;
838 struct inode *inode = NULL;
839 int error;
840 struct nfs_fh fhandle;
841 struct nfs_fattr fattr;
842
843 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
844 dentry->d_parent->d_name.name, dentry->d_name.name);
845
846 res = ERR_PTR(-ENAMETOOLONG);
847 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
848 goto out;
849
850 res = ERR_PTR(-ENOMEM);
851 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
852
853 lock_kernel();
854 /* Revalidate parent directory attribute cache */
855 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
856 if (error < 0) {
857 res = ERR_PTR(error);
858 goto out_unlock;
859 }
860
861 /* If we're doing an exclusive create, optimize away the lookup */
862 if (nfs_is_exclusive_create(dir, nd))
863 goto no_entry;
864
865 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
866 if (error == -ENOENT)
867 goto no_entry;
868 if (error < 0) {
869 res = ERR_PTR(error);
870 goto out_unlock;
871 }
872 res = ERR_PTR(-EACCES);
873 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
874 if (!inode)
875 goto out_unlock;
876no_entry:
877 res = d_add_unique(dentry, inode);
878 if (res != NULL)
879 dentry = res;
880 nfs_renew_times(dentry);
881 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
882out_unlock:
883 unlock_kernel();
884out:
885 return res;
886}
887
888#ifdef CONFIG_NFS_V4
889static int nfs_open_revalidate(struct dentry *, struct nameidata *);
890
891struct dentry_operations nfs4_dentry_operations = {
892 .d_revalidate = nfs_open_revalidate,
893 .d_delete = nfs_dentry_delete,
894 .d_iput = nfs_dentry_iput,
895};
896
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400897/*
898 * Use intent information to determine whether we need to substitute
899 * the NFSv4-style stateful OPEN for the LOOKUP call
900 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901static int is_atomic_open(struct inode *dir, struct nameidata *nd)
902{
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400903 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 return 0;
905 /* NFS does not (yet) have a stateful open for directories */
906 if (nd->flags & LOOKUP_DIRECTORY)
907 return 0;
908 /* Are we trying to write to a read only partition? */
909 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
910 return 0;
911 return 1;
912}
913
914static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
915{
916 struct dentry *res = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 int error;
918
919 /* Check that we are indeed trying to open this file */
920 if (!is_atomic_open(dir, nd))
921 goto no_open;
922
923 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
924 res = ERR_PTR(-ENAMETOOLONG);
925 goto out;
926 }
927 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
928
929 /* Let vfs_create() deal with O_EXCL */
Trond Myklebust02a913a2005-10-18 14:20:17 -0700930 if (nd->intent.open.flags & O_EXCL) {
931 d_add(dentry, NULL);
932 goto out;
933 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934
935 /* Open the file on the server */
936 lock_kernel();
937 /* Revalidate parent directory attribute cache */
938 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
939 if (error < 0) {
940 res = ERR_PTR(error);
Steve Dickson01c314a2005-08-19 17:57:48 -0700941 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 goto out;
943 }
944
945 if (nd->intent.open.flags & O_CREAT) {
946 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -0700947 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 nfs_end_data_update(dir);
949 } else
Trond Myklebust02a913a2005-10-18 14:20:17 -0700950 res = nfs4_atomic_open(dir, dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 unlock_kernel();
Trond Myklebust02a913a2005-10-18 14:20:17 -0700952 if (IS_ERR(res)) {
953 error = PTR_ERR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 switch (error) {
955 /* Make a negative dentry */
956 case -ENOENT:
Trond Myklebust02a913a2005-10-18 14:20:17 -0700957 res = NULL;
958 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -0700960 case -EISDIR:
961 case -ENOTDIR:
962 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 case -ELOOP:
964 if (!(nd->intent.open.flags & O_NOFOLLOW))
965 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966 /* case -EINVAL: */
967 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 goto out;
969 }
Trond Myklebust02a913a2005-10-18 14:20:17 -0700970 } else if (res != NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 dentry = res;
972 nfs_renew_times(dentry);
973 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
974out:
975 return res;
976no_open:
977 return nfs_lookup(dir, dentry, nd);
978}
979
980static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
981{
982 struct dentry *parent = NULL;
983 struct inode *inode = dentry->d_inode;
984 struct inode *dir;
985 unsigned long verifier;
986 int openflags, ret = 0;
987
988 parent = dget_parent(dentry);
989 dir = parent->d_inode;
990 if (!is_atomic_open(dir, nd))
991 goto no_open;
992 /* We can't create new files in nfs_open_revalidate(), so we
993 * optimize away revalidation of negative dentries.
994 */
995 if (inode == NULL)
996 goto out;
997 /* NFS only supports OPEN on regular files */
998 if (!S_ISREG(inode->i_mode))
999 goto no_open;
1000 openflags = nd->intent.open.flags;
1001 /* We cannot do exclusive creation on a positive dentry */
1002 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1003 goto no_open;
1004 /* We can't create new files, or truncate existing ones here */
1005 openflags &= ~(O_CREAT|O_TRUNC);
1006
1007 /*
1008 * Note: we're not holding inode->i_sem and so may be racing with
1009 * operations that change the directory. We therefore save the
1010 * change attribute *before* we do the RPC call.
1011 */
1012 lock_kernel();
1013 verifier = nfs_save_change_attribute(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001014 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 if (!ret)
1016 nfs_set_verifier(dentry, verifier);
1017 unlock_kernel();
1018out:
1019 dput(parent);
1020 if (!ret)
1021 d_drop(dentry);
1022 return ret;
1023no_open:
1024 dput(parent);
1025 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1026 return 1;
1027 return nfs_lookup_revalidate(dentry, nd);
1028}
1029#endif /* CONFIG_NFSV4 */
1030
1031static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1032{
1033 struct dentry *parent = desc->file->f_dentry;
1034 struct inode *dir = parent->d_inode;
1035 struct nfs_entry *entry = desc->entry;
1036 struct dentry *dentry, *alias;
1037 struct qstr name = {
1038 .name = entry->name,
1039 .len = entry->len,
1040 };
1041 struct inode *inode;
1042
1043 switch (name.len) {
1044 case 2:
1045 if (name.name[0] == '.' && name.name[1] == '.')
1046 return dget_parent(parent);
1047 break;
1048 case 1:
1049 if (name.name[0] == '.')
1050 return dget(parent);
1051 }
1052 name.hash = full_name_hash(name.name, name.len);
1053 dentry = d_lookup(parent, &name);
1054 if (dentry != NULL)
1055 return dentry;
1056 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1057 return NULL;
1058 /* Note: caller is already holding the dir->i_sem! */
1059 dentry = d_alloc(parent, &name);
1060 if (dentry == NULL)
1061 return NULL;
1062 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1063 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1064 if (!inode) {
1065 dput(dentry);
1066 return NULL;
1067 }
1068 alias = d_add_unique(dentry, inode);
1069 if (alias != NULL) {
1070 dput(dentry);
1071 dentry = alias;
1072 }
1073 nfs_renew_times(dentry);
1074 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1075 return dentry;
1076}
1077
1078/*
1079 * Code common to create, mkdir, and mknod.
1080 */
1081int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1082 struct nfs_fattr *fattr)
1083{
1084 struct inode *inode;
1085 int error = -EACCES;
1086
1087 /* We may have been initialized further down */
1088 if (dentry->d_inode)
1089 return 0;
1090 if (fhandle->size == 0) {
1091 struct inode *dir = dentry->d_parent->d_inode;
1092 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1093 if (error)
1094 goto out_err;
1095 }
1096 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1097 struct nfs_server *server = NFS_SB(dentry->d_sb);
1098 error = server->rpc_ops->getattr(server, fhandle, fattr);
1099 if (error < 0)
1100 goto out_err;
1101 }
1102 error = -ENOMEM;
1103 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1104 if (inode == NULL)
1105 goto out_err;
1106 d_instantiate(dentry, inode);
1107 return 0;
1108out_err:
1109 d_drop(dentry);
1110 return error;
1111}
1112
1113/*
1114 * Following a failed create operation, we drop the dentry rather
1115 * than retain a negative dentry. This avoids a problem in the event
1116 * that the operation succeeded on the server, but an error in the
1117 * reply path made it appear to have failed.
1118 */
1119static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1120 struct nameidata *nd)
1121{
1122 struct iattr attr;
1123 int error;
1124 int open_flags = 0;
1125
1126 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1127 dir->i_ino, dentry->d_name.name);
1128
1129 attr.ia_mode = mode;
1130 attr.ia_valid = ATTR_MODE;
1131
1132 if (nd && (nd->flags & LOOKUP_CREATE))
1133 open_flags = nd->intent.open.flags;
1134
1135 lock_kernel();
1136 nfs_begin_data_update(dir);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001137 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138 nfs_end_data_update(dir);
1139 if (error != 0)
1140 goto out_err;
1141 nfs_renew_times(dentry);
1142 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1143 unlock_kernel();
1144 return 0;
1145out_err:
1146 unlock_kernel();
1147 d_drop(dentry);
1148 return error;
1149}
1150
1151/*
1152 * See comments for nfs_proc_create regarding failed operations.
1153 */
1154static int
1155nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1156{
1157 struct iattr attr;
1158 int status;
1159
1160 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1161 dir->i_ino, dentry->d_name.name);
1162
1163 if (!new_valid_dev(rdev))
1164 return -EINVAL;
1165
1166 attr.ia_mode = mode;
1167 attr.ia_valid = ATTR_MODE;
1168
1169 lock_kernel();
1170 nfs_begin_data_update(dir);
1171 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1172 nfs_end_data_update(dir);
1173 if (status != 0)
1174 goto out_err;
1175 nfs_renew_times(dentry);
1176 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1177 unlock_kernel();
1178 return 0;
1179out_err:
1180 unlock_kernel();
1181 d_drop(dentry);
1182 return status;
1183}
1184
1185/*
1186 * See comments for nfs_proc_create regarding failed operations.
1187 */
1188static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1189{
1190 struct iattr attr;
1191 int error;
1192
1193 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1194 dir->i_ino, dentry->d_name.name);
1195
1196 attr.ia_valid = ATTR_MODE;
1197 attr.ia_mode = mode | S_IFDIR;
1198
1199 lock_kernel();
1200 nfs_begin_data_update(dir);
1201 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1202 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 d_drop(dentry);
1211 unlock_kernel();
1212 return error;
1213}
1214
1215static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1216{
1217 int error;
1218
1219 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1220 dir->i_ino, dentry->d_name.name);
1221
1222 lock_kernel();
1223 nfs_begin_data_update(dir);
1224 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1225 /* Ensure the VFS deletes this inode */
1226 if (error == 0 && dentry->d_inode != NULL)
1227 dentry->d_inode->i_nlink = 0;
1228 nfs_end_data_update(dir);
1229 unlock_kernel();
1230
1231 return error;
1232}
1233
1234static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1235{
1236 static unsigned int sillycounter;
1237 const int i_inosize = sizeof(dir->i_ino)*2;
1238 const int countersize = sizeof(sillycounter)*2;
1239 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1240 char silly[slen+1];
1241 struct qstr qsilly;
1242 struct dentry *sdentry;
1243 int error = -EIO;
1244
1245 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1246 dentry->d_parent->d_name.name, dentry->d_name.name,
1247 atomic_read(&dentry->d_count));
1248
1249#ifdef NFS_PARANOIA
1250if (!dentry->d_inode)
1251printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1252dentry->d_parent->d_name.name, dentry->d_name.name);
1253#endif
1254 /*
1255 * We don't allow a dentry to be silly-renamed twice.
1256 */
1257 error = -EBUSY;
1258 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1259 goto out;
1260
1261 sprintf(silly, ".nfs%*.*lx",
1262 i_inosize, i_inosize, dentry->d_inode->i_ino);
1263
1264 sdentry = NULL;
1265 do {
1266 char *suffix = silly + slen - countersize;
1267
1268 dput(sdentry);
1269 sillycounter++;
1270 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1271
1272 dfprintk(VFS, "trying to rename %s to %s\n",
1273 dentry->d_name.name, silly);
1274
1275 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1276 /*
1277 * N.B. Better to return EBUSY here ... it could be
1278 * dangerous to delete the file while it's in use.
1279 */
1280 if (IS_ERR(sdentry))
1281 goto out;
1282 } while(sdentry->d_inode != NULL); /* need negative lookup */
1283
1284 qsilly.name = silly;
1285 qsilly.len = strlen(silly);
1286 nfs_begin_data_update(dir);
1287 if (dentry->d_inode) {
1288 nfs_begin_data_update(dentry->d_inode);
1289 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1290 dir, &qsilly);
1291 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) {
1332 nfs_begin_data_update(inode);
1333 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1334 /* The VFS may want to delete this inode */
1335 if (error == 0)
1336 inode->i_nlink--;
1337 nfs_end_data_update(inode);
1338 } else
1339 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1340 nfs_end_data_update(dir);
1341out:
1342 return error;
1343}
1344
1345/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1346 * belongs to an active ".nfs..." file and we return -EBUSY.
1347 *
1348 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1349 */
1350static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1351{
1352 int error;
1353 int need_rehash = 0;
1354
1355 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1356 dir->i_ino, dentry->d_name.name);
1357
1358 lock_kernel();
1359 spin_lock(&dcache_lock);
1360 spin_lock(&dentry->d_lock);
1361 if (atomic_read(&dentry->d_count) > 1) {
1362 spin_unlock(&dentry->d_lock);
1363 spin_unlock(&dcache_lock);
1364 error = nfs_sillyrename(dir, dentry);
1365 unlock_kernel();
1366 return error;
1367 }
1368 if (!d_unhashed(dentry)) {
1369 __d_drop(dentry);
1370 need_rehash = 1;
1371 }
1372 spin_unlock(&dentry->d_lock);
1373 spin_unlock(&dcache_lock);
1374 error = nfs_safe_remove(dentry);
1375 if (!error) {
1376 nfs_renew_times(dentry);
1377 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1378 } else if (need_rehash)
1379 d_rehash(dentry);
1380 unlock_kernel();
1381 return error;
1382}
1383
1384static int
1385nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1386{
1387 struct iattr attr;
1388 struct nfs_fattr sym_attr;
1389 struct nfs_fh sym_fh;
1390 struct qstr qsymname;
1391 int error;
1392
1393 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1394 dir->i_ino, dentry->d_name.name, symname);
1395
1396#ifdef NFS_PARANOIA
1397if (dentry->d_inode)
1398printk("nfs_proc_symlink: %s/%s not negative!\n",
1399dentry->d_parent->d_name.name, dentry->d_name.name);
1400#endif
1401 /*
1402 * Fill in the sattr for the call.
1403 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1404 */
1405 attr.ia_valid = ATTR_MODE;
1406 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1407
1408 qsymname.name = symname;
1409 qsymname.len = strlen(symname);
1410
1411 lock_kernel();
1412 nfs_begin_data_update(dir);
1413 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1414 &attr, &sym_fh, &sym_attr);
1415 nfs_end_data_update(dir);
1416 if (!error) {
1417 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1418 } else {
1419 if (error == -EEXIST)
1420 printk("nfs_proc_symlink: %s/%s already exists??\n",
1421 dentry->d_parent->d_name.name, dentry->d_name.name);
1422 d_drop(dentry);
1423 }
1424 unlock_kernel();
1425 return error;
1426}
1427
1428static int
1429nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1430{
1431 struct inode *inode = old_dentry->d_inode;
1432 int error;
1433
1434 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1435 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1436 dentry->d_parent->d_name.name, dentry->d_name.name);
1437
1438 /*
1439 * Drop the dentry in advance to force a new lookup.
1440 * Since nfs_proc_link doesn't return a file handle,
1441 * we can't use the existing dentry.
1442 */
1443 lock_kernel();
1444 d_drop(dentry);
1445
1446 nfs_begin_data_update(dir);
1447 nfs_begin_data_update(inode);
1448 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1449 nfs_end_data_update(inode);
1450 nfs_end_data_update(dir);
1451 unlock_kernel();
1452 return error;
1453}
1454
1455/*
1456 * RENAME
1457 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1458 * different file handle for the same inode after a rename (e.g. when
1459 * moving to a different directory). A fail-safe method to do so would
1460 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1461 * rename the old file using the sillyrename stuff. This way, the original
1462 * file in old_dir will go away when the last process iput()s the inode.
1463 *
1464 * FIXED.
1465 *
1466 * It actually works quite well. One needs to have the possibility for
1467 * at least one ".nfs..." file in each directory the file ever gets
1468 * moved or linked to which happens automagically with the new
1469 * implementation that only depends on the dcache stuff instead of
1470 * using the inode layer
1471 *
1472 * Unfortunately, things are a little more complicated than indicated
1473 * above. For a cross-directory move, we want to make sure we can get
1474 * rid of the old inode after the operation. This means there must be
1475 * no pending writes (if it's a file), and the use count must be 1.
1476 * If these conditions are met, we can drop the dentries before doing
1477 * the rename.
1478 */
1479static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1480 struct inode *new_dir, struct dentry *new_dentry)
1481{
1482 struct inode *old_inode = old_dentry->d_inode;
1483 struct inode *new_inode = new_dentry->d_inode;
1484 struct dentry *dentry = NULL, *rehash = NULL;
1485 int error = -EBUSY;
1486
1487 /*
1488 * To prevent any new references to the target during the rename,
1489 * we unhash the dentry and free the inode in advance.
1490 */
1491 lock_kernel();
1492 if (!d_unhashed(new_dentry)) {
1493 d_drop(new_dentry);
1494 rehash = new_dentry;
1495 }
1496
1497 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1498 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1499 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1500 atomic_read(&new_dentry->d_count));
1501
1502 /*
1503 * First check whether the target is busy ... we can't
1504 * safely do _any_ rename if the target is in use.
1505 *
1506 * For files, make a copy of the dentry and then do a
1507 * silly-rename. If the silly-rename succeeds, the
1508 * copied dentry is hashed and becomes the new target.
1509 */
1510 if (!new_inode)
1511 goto go_ahead;
1512 if (S_ISDIR(new_inode->i_mode))
1513 goto out;
1514 else if (atomic_read(&new_dentry->d_count) > 2) {
1515 int err;
1516 /* copy the target dentry's name */
1517 dentry = d_alloc(new_dentry->d_parent,
1518 &new_dentry->d_name);
1519 if (!dentry)
1520 goto out;
1521
1522 /* silly-rename the existing target ... */
1523 err = nfs_sillyrename(new_dir, new_dentry);
1524 if (!err) {
1525 new_dentry = rehash = dentry;
1526 new_inode = NULL;
1527 /* instantiate the replacement target */
1528 d_instantiate(new_dentry, NULL);
1529 } else if (atomic_read(&new_dentry->d_count) > 1) {
1530 /* dentry still busy? */
1531#ifdef NFS_PARANOIA
1532 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1533 new_dentry->d_parent->d_name.name,
1534 new_dentry->d_name.name,
1535 atomic_read(&new_dentry->d_count));
1536#endif
1537 goto out;
1538 }
Trond Myklebust20509f12005-08-25 16:25:34 -07001539 } else
1540 new_inode->i_nlink--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541
1542go_ahead:
1543 /*
1544 * ... prune child dentries and writebacks if needed.
1545 */
1546 if (atomic_read(&old_dentry->d_count) > 1) {
1547 nfs_wb_all(old_inode);
1548 shrink_dcache_parent(old_dentry);
1549 }
1550
1551 if (new_inode)
1552 d_delete(new_dentry);
1553
1554 nfs_begin_data_update(old_dir);
1555 nfs_begin_data_update(new_dir);
1556 nfs_begin_data_update(old_inode);
1557 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1558 new_dir, &new_dentry->d_name);
1559 nfs_end_data_update(old_inode);
1560 nfs_end_data_update(new_dir);
1561 nfs_end_data_update(old_dir);
1562out:
1563 if (rehash)
1564 d_rehash(rehash);
1565 if (!error) {
1566 if (!S_ISDIR(old_inode->i_mode))
1567 d_move(old_dentry, new_dentry);
1568 nfs_renew_times(new_dentry);
1569 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1570 }
1571
1572 /* new dentry created? */
1573 if (dentry)
1574 dput(dentry);
1575 unlock_kernel();
1576 return error;
1577}
1578
1579int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1580{
Chuck Lever55296802005-08-18 11:24:09 -07001581 struct nfs_inode *nfsi = NFS_I(inode);
1582 struct nfs_access_entry *cache = &nfsi->cache_access;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
1584 if (cache->cred != cred
1585 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
Chuck Lever55296802005-08-18 11:24:09 -07001586 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001587 return -ENOENT;
1588 memcpy(res, cache, sizeof(*res));
1589 return 0;
1590}
1591
1592void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1593{
Chuck Lever55296802005-08-18 11:24:09 -07001594 struct nfs_inode *nfsi = NFS_I(inode);
1595 struct nfs_access_entry *cache = &nfsi->cache_access;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 if (cache->cred != set->cred) {
1598 if (cache->cred)
1599 put_rpccred(cache->cred);
1600 cache->cred = get_rpccred(set->cred);
1601 }
Chuck Leverdc592502005-08-18 11:24:12 -07001602 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1603 spin_lock(&inode->i_lock);
Chuck Lever55296802005-08-18 11:24:09 -07001604 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
Chuck Leverdc592502005-08-18 11:24:12 -07001605 spin_unlock(&inode->i_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 cache->jiffies = set->jiffies;
1607 cache->mask = set->mask;
1608}
1609
1610static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1611{
1612 struct nfs_access_entry cache;
1613 int status;
1614
1615 status = nfs_access_get_cached(inode, cred, &cache);
1616 if (status == 0)
1617 goto out;
1618
1619 /* Be clever: ask server to check for all possible rights */
1620 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1621 cache.cred = cred;
1622 cache.jiffies = jiffies;
1623 status = NFS_PROTO(inode)->access(inode, &cache);
1624 if (status != 0)
1625 return status;
1626 nfs_access_add_cache(inode, &cache);
1627out:
1628 if ((cache.mask & mask) == mask)
1629 return 0;
1630 return -EACCES;
1631}
1632
1633int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1634{
1635 struct rpc_cred *cred;
1636 int res = 0;
1637
1638 if (mask == 0)
1639 goto out;
1640 /* Is this sys_access() ? */
1641 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1642 goto force_lookup;
1643
1644 switch (inode->i_mode & S_IFMT) {
1645 case S_IFLNK:
1646 goto out;
1647 case S_IFREG:
1648 /* NFSv4 has atomic_open... */
1649 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1650 && nd != NULL
1651 && (nd->flags & LOOKUP_OPEN))
1652 goto out;
1653 break;
1654 case S_IFDIR:
1655 /*
1656 * Optimize away all write operations, since the server
1657 * will check permissions when we perform the op.
1658 */
1659 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1660 goto out;
1661 }
1662
1663force_lookup:
1664 lock_kernel();
1665
1666 if (!NFS_PROTO(inode)->access)
1667 goto out_notsup;
1668
1669 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1670 if (!IS_ERR(cred)) {
1671 res = nfs_do_access(inode, cred, mask);
1672 put_rpccred(cred);
1673 } else
1674 res = PTR_ERR(cred);
1675 unlock_kernel();
1676out:
1677 return res;
1678out_notsup:
1679 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1680 if (res == 0)
1681 res = generic_permission(inode, mask, NULL);
1682 unlock_kernel();
1683 return res;
1684}
1685
1686/*
1687 * Local variables:
1688 * version-control: t
1689 * kept-new-versions: 5
1690 * End:
1691 */