blob: 880052cadbcd4fdeb1dc05f5018b0b5e587f6824 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/namei.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7/*
8 * Some corrections by tytso.
9 */
10
11/* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12 * lookup logic.
13 */
14/* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15 */
16
17#include <linux/init.h>
18#include <linux/module.h>
19#include <linux/slab.h>
20#include <linux/fs.h>
21#include <linux/namei.h>
22#include <linux/quotaops.h>
23#include <linux/pagemap.h>
Robert Love0eeca282005-07-12 17:06:03 -040024#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/smp_lock.h>
26#include <linux/personality.h>
27#include <linux/security.h>
28#include <linux/syscalls.h>
29#include <linux/mount.h>
30#include <linux/audit.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080031#include <linux/capability.h>
Trond Myklebust834f2a42005-10-18 14:20:16 -070032#include <linux/file.h>
Ulrich Drepper5590ff02006-01-18 17:43:53 -080033#include <linux/fcntl.h>
34#include <linux/namei.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/namei.h>
36#include <asm/uaccess.h>
37
38#define ACC_MODE(x) ("\000\004\002\006"[(x)&O_ACCMODE])
39
40/* [Feb-1997 T. Schoebel-Theuer]
41 * Fundamental changes in the pathname lookup mechanisms (namei)
42 * were necessary because of omirr. The reason is that omirr needs
43 * to know the _real_ pathname, not the user-supplied one, in case
44 * of symlinks (and also when transname replacements occur).
45 *
46 * The new code replaces the old recursive symlink resolution with
47 * an iterative one (in case of non-nested symlink chains). It does
48 * this with calls to <fs>_follow_link().
49 * As a side effect, dir_namei(), _namei() and follow_link() are now
50 * replaced with a single function lookup_dentry() that can handle all
51 * the special cases of the former code.
52 *
53 * With the new dcache, the pathname is stored at each inode, at least as
54 * long as the refcount of the inode is positive. As a side effect, the
55 * size of the dcache depends on the inode cache and thus is dynamic.
56 *
57 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
58 * resolution to correspond with current state of the code.
59 *
60 * Note that the symlink resolution is not *completely* iterative.
61 * There is still a significant amount of tail- and mid- recursion in
62 * the algorithm. Also, note that <fs>_readlink() is not used in
63 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
64 * may return different results than <fs>_follow_link(). Many virtual
65 * filesystems (including /proc) exhibit this behavior.
66 */
67
68/* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
69 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
70 * and the name already exists in form of a symlink, try to create the new
71 * name indicated by the symlink. The old code always complained that the
72 * name already exists, due to not following the symlink even if its target
73 * is nonexistent. The new semantics affects also mknod() and link() when
74 * the name is a symlink pointing to a non-existant name.
75 *
76 * I don't know which semantics is the right one, since I have no access
77 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
78 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
79 * "old" one. Personally, I think the new semantics is much more logical.
80 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
81 * file does succeed in both HP-UX and SunOs, but not in Solaris
82 * and in the old Linux semantics.
83 */
84
85/* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
86 * semantics. See the comments in "open_namei" and "do_link" below.
87 *
88 * [10-Sep-98 Alan Modra] Another symlink change.
89 */
90
91/* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
92 * inside the path - always follow.
93 * in the last component in creation/removal/renaming - never follow.
94 * if LOOKUP_FOLLOW passed - follow.
95 * if the pathname has trailing slashes - follow.
96 * otherwise - don't follow.
97 * (applied in that order).
98 *
99 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
100 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
101 * During the 2.4 we need to fix the userland stuff depending on it -
102 * hopefully we will be able to get rid of that wart in 2.5. So far only
103 * XEmacs seems to be relying on it...
104 */
105/*
106 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
Arjan van de Vena11f3a02006-03-23 03:00:33 -0800107 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * any extra contention...
109 */
110
111/* In order to reduce some races, while at the same time doing additional
112 * checking and hopefully speeding things up, we copy filenames to the
113 * kernel data space before using them..
114 *
115 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
116 * PATH_MAX includes the nul terminator --RR.
117 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800118static int do_getname(const char __user *filename, char *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119{
120 int retval;
121 unsigned long len = PATH_MAX;
122
123 if (!segment_eq(get_fs(), KERNEL_DS)) {
124 if ((unsigned long) filename >= TASK_SIZE)
125 return -EFAULT;
126 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
127 len = TASK_SIZE - (unsigned long) filename;
128 }
129
130 retval = strncpy_from_user(page, filename, len);
131 if (retval > 0) {
132 if (retval < len)
133 return 0;
134 return -ENAMETOOLONG;
135 } else if (!retval)
136 retval = -ENOENT;
137 return retval;
138}
139
140char * getname(const char __user * filename)
141{
142 char *tmp, *result;
143
144 result = ERR_PTR(-ENOMEM);
145 tmp = __getname();
146 if (tmp) {
147 int retval = do_getname(filename, tmp);
148
149 result = tmp;
150 if (retval < 0) {
151 __putname(tmp);
152 result = ERR_PTR(retval);
153 }
154 }
155 audit_getname(result);
156 return result;
157}
158
159#ifdef CONFIG_AUDITSYSCALL
160void putname(const char *name)
161{
Al Viro5ac3a9c2006-07-16 06:38:45 -0400162 if (unlikely(!audit_dummy_context()))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 audit_putname(name);
164 else
165 __putname(name);
166}
167EXPORT_SYMBOL(putname);
168#endif
169
170
171/**
172 * generic_permission - check for access rights on a Posix-like filesystem
173 * @inode: inode to check access rights for
174 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
175 * @check_acl: optional callback to check for Posix ACLs
176 *
177 * Used to check for read/write/execute permissions on a file.
178 * We use "fsuid" for this, letting us set arbitrary permissions
179 * for filesystem access without changing the "normal" uids which
180 * are used for other things..
181 */
182int generic_permission(struct inode *inode, int mask,
183 int (*check_acl)(struct inode *inode, int mask))
184{
185 umode_t mode = inode->i_mode;
186
187 if (current->fsuid == inode->i_uid)
188 mode >>= 6;
189 else {
190 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
191 int error = check_acl(inode, mask);
192 if (error == -EACCES)
193 goto check_capabilities;
194 else if (error != -EAGAIN)
195 return error;
196 }
197
198 if (in_group_p(inode->i_gid))
199 mode >>= 3;
200 }
201
202 /*
203 * If the DACs are ok we don't need any capability check.
204 */
205 if (((mode & mask & (MAY_READ|MAY_WRITE|MAY_EXEC)) == mask))
206 return 0;
207
208 check_capabilities:
209 /*
210 * Read/write DACs are always overridable.
211 * Executable DACs are overridable if at least one exec bit is set.
212 */
213 if (!(mask & MAY_EXEC) ||
214 (inode->i_mode & S_IXUGO) || S_ISDIR(inode->i_mode))
215 if (capable(CAP_DAC_OVERRIDE))
216 return 0;
217
218 /*
219 * Searching includes executable on directories, else just read.
220 */
221 if (mask == MAY_READ || (S_ISDIR(inode->i_mode) && !(mask & MAY_WRITE)))
222 if (capable(CAP_DAC_READ_SEARCH))
223 return 0;
224
225 return -EACCES;
226}
227
228int permission(struct inode *inode, int mask, struct nameidata *nd)
229{
Trond Myklebusta343bb72006-08-22 20:06:03 -0400230 umode_t mode = inode->i_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 int retval, submask;
232
233 if (mask & MAY_WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 /*
236 * Nobody gets write access to a read-only fs.
237 */
238 if (IS_RDONLY(inode) &&
239 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
240 return -EROFS;
241
242 /*
243 * Nobody gets write access to an immutable file.
244 */
245 if (IS_IMMUTABLE(inode))
246 return -EACCES;
247 }
248
249
Trond Myklebusta343bb72006-08-22 20:06:03 -0400250 /*
251 * MAY_EXEC on regular files requires special handling: We override
Stas Sergeev317a40a2006-12-06 20:35:25 -0800252 * filesystem execute permissions if the mode bits aren't set or
253 * the fs is mounted with the "noexec" flag.
Trond Myklebusta343bb72006-08-22 20:06:03 -0400254 */
Stas Sergeev317a40a2006-12-06 20:35:25 -0800255 if ((mask & MAY_EXEC) && S_ISREG(mode) && (!(mode & S_IXUGO) ||
256 (nd && nd->mnt && (nd->mnt->mnt_flags & MNT_NOEXEC))))
Trond Myklebusta343bb72006-08-22 20:06:03 -0400257 return -EACCES;
258
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 /* Ordinary permission routines do not understand MAY_APPEND. */
260 submask = mask & ~MAY_APPEND;
261 if (inode->i_op && inode->i_op->permission)
262 retval = inode->i_op->permission(inode, submask, nd);
263 else
264 retval = generic_permission(inode, submask, NULL);
265 if (retval)
266 return retval;
267
268 return security_inode_permission(inode, mask, nd);
269}
270
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800271/**
272 * vfs_permission - check for access rights to a given path
273 * @nd: lookup result that describes the path
274 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
275 *
276 * Used to check for read/write/execute permissions on a path.
277 * We use "fsuid" for this, letting us set arbitrary permissions
278 * for filesystem access without changing the "normal" uids which
279 * are used for other things.
280 */
281int vfs_permission(struct nameidata *nd, int mask)
282{
283 return permission(nd->dentry->d_inode, mask, nd);
284}
285
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800286/**
287 * file_permission - check for additional access rights to a given file
288 * @file: file to check access rights for
289 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
290 *
291 * Used to check for read/write/execute permissions on an already opened
292 * file.
293 *
294 * Note:
295 * Do not use this function in new code. All access checks should
296 * be done using vfs_permission().
297 */
298int file_permission(struct file *file, int mask)
299{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800300 return permission(file->f_path.dentry->d_inode, mask, NULL);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -0800301}
302
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303/*
304 * get_write_access() gets write permission for a file.
305 * put_write_access() releases this write permission.
306 * This is used for regular files.
307 * We cannot support write (and maybe mmap read-write shared) accesses and
308 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
309 * can have the following values:
310 * 0: no writers, no VM_DENYWRITE mappings
311 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
312 * > 0: (i_writecount) users are writing to the file.
313 *
314 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
315 * except for the cases where we don't hold i_writecount yet. Then we need to
316 * use {get,deny}_write_access() - these functions check the sign and refuse
317 * to do the change if sign is wrong. Exclusion between them is provided by
318 * the inode->i_lock spinlock.
319 */
320
321int get_write_access(struct inode * inode)
322{
323 spin_lock(&inode->i_lock);
324 if (atomic_read(&inode->i_writecount) < 0) {
325 spin_unlock(&inode->i_lock);
326 return -ETXTBSY;
327 }
328 atomic_inc(&inode->i_writecount);
329 spin_unlock(&inode->i_lock);
330
331 return 0;
332}
333
334int deny_write_access(struct file * file)
335{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800336 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337
338 spin_lock(&inode->i_lock);
339 if (atomic_read(&inode->i_writecount) > 0) {
340 spin_unlock(&inode->i_lock);
341 return -ETXTBSY;
342 }
343 atomic_dec(&inode->i_writecount);
344 spin_unlock(&inode->i_lock);
345
346 return 0;
347}
348
349void path_release(struct nameidata *nd)
350{
351 dput(nd->dentry);
352 mntput(nd->mnt);
353}
354
355/*
356 * umount() mustn't call path_release()/mntput() as that would clear
357 * mnt_expiry_mark
358 */
359void path_release_on_umount(struct nameidata *nd)
360{
361 dput(nd->dentry);
Miklos Szeredi751c4042005-07-07 17:57:30 -0700362 mntput_no_expire(nd->mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363}
364
Trond Myklebust834f2a42005-10-18 14:20:16 -0700365/**
366 * release_open_intent - free up open intent resources
367 * @nd: pointer to nameidata
368 */
369void release_open_intent(struct nameidata *nd)
370{
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -0800371 if (nd->intent.open.file->f_path.dentry == NULL)
Trond Myklebust834f2a42005-10-18 14:20:16 -0700372 put_filp(nd->intent.open.file);
373 else
374 fput(nd->intent.open.file);
375}
376
Ian Kentbcdc5e02006-09-27 01:50:44 -0700377static inline struct dentry *
378do_revalidate(struct dentry *dentry, struct nameidata *nd)
379{
380 int status = dentry->d_op->d_revalidate(dentry, nd);
381 if (unlikely(status <= 0)) {
382 /*
383 * The dentry failed validation.
384 * If d_revalidate returned 0 attempt to invalidate
385 * the dentry otherwise d_revalidate is asking us
386 * to return a fail status.
387 */
388 if (!status) {
389 if (!d_invalidate(dentry)) {
390 dput(dentry);
391 dentry = NULL;
392 }
393 } else {
394 dput(dentry);
395 dentry = ERR_PTR(status);
396 }
397 }
398 return dentry;
399}
400
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401/*
402 * Internal lookup() using the new generic dcache.
403 * SMP-safe
404 */
405static struct dentry * cached_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
406{
407 struct dentry * dentry = __d_lookup(parent, name);
408
409 /* lockess __d_lookup may fail due to concurrent d_move()
410 * in some unrelated directory, so try with d_lookup
411 */
412 if (!dentry)
413 dentry = d_lookup(parent, name);
414
Ian Kentbcdc5e02006-09-27 01:50:44 -0700415 if (dentry && dentry->d_op && dentry->d_op->d_revalidate)
416 dentry = do_revalidate(dentry, nd);
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 return dentry;
419}
420
421/*
422 * Short-cut version of permission(), for calling by
423 * path_walk(), when dcache lock is held. Combines parts
424 * of permission() and generic_permission(), and tests ONLY for
425 * MAY_EXEC permission.
426 *
427 * If appropriate, check DAC only. If not appropriate, or
428 * short-cut DAC fails, then call permission() to do more
429 * complete permission check.
430 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800431static int exec_permission_lite(struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 struct nameidata *nd)
433{
434 umode_t mode = inode->i_mode;
435
436 if (inode->i_op && inode->i_op->permission)
437 return -EAGAIN;
438
439 if (current->fsuid == inode->i_uid)
440 mode >>= 6;
441 else if (in_group_p(inode->i_gid))
442 mode >>= 3;
443
444 if (mode & MAY_EXEC)
445 goto ok;
446
447 if ((inode->i_mode & S_IXUGO) && capable(CAP_DAC_OVERRIDE))
448 goto ok;
449
450 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_OVERRIDE))
451 goto ok;
452
453 if (S_ISDIR(inode->i_mode) && capable(CAP_DAC_READ_SEARCH))
454 goto ok;
455
456 return -EACCES;
457ok:
458 return security_inode_permission(inode, MAY_EXEC, nd);
459}
460
461/*
462 * This is called when everything else fails, and we actually have
463 * to go to the low-level filesystem to find out what we should do..
464 *
465 * We get the directory semaphore, and after getting that we also
466 * make sure that nobody added the entry to the dcache in the meantime..
467 * SMP-safe
468 */
469static struct dentry * real_lookup(struct dentry * parent, struct qstr * name, struct nameidata *nd)
470{
471 struct dentry * result;
472 struct inode *dir = parent->d_inode;
473
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800474 mutex_lock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 /*
476 * First re-do the cached lookup just in case it was created
477 * while we waited for the directory semaphore..
478 *
479 * FIXME! This could use version numbering or similar to
480 * avoid unnecessary cache lookups.
481 *
482 * The "dcache_lock" is purely to protect the RCU list walker
483 * from concurrent renames at this point (we mustn't get false
484 * negatives from the RCU list walk here, unlike the optimistic
485 * fast walk).
486 *
487 * so doing d_lookup() (with seqlock), instead of lockfree __d_lookup
488 */
489 result = d_lookup(parent, name);
490 if (!result) {
491 struct dentry * dentry = d_alloc(parent, name);
492 result = ERR_PTR(-ENOMEM);
493 if (dentry) {
494 result = dir->i_op->lookup(dir, dentry, nd);
495 if (result)
496 dput(dentry);
497 else
498 result = dentry;
499 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800500 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 return result;
502 }
503
504 /*
505 * Uhhuh! Nasty case: the cache was re-populated while
506 * we waited on the semaphore. Need to revalidate.
507 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -0800508 mutex_unlock(&dir->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 if (result->d_op && result->d_op->d_revalidate) {
Ian Kentbcdc5e02006-09-27 01:50:44 -0700510 result = do_revalidate(result, nd);
511 if (!result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 result = ERR_PTR(-ENOENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 }
514 return result;
515}
516
517static int __emul_lookup_dentry(const char *, struct nameidata *);
518
519/* SMP-safe */
Arjan van de Venf1662352006-01-14 13:21:31 -0800520static __always_inline int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521walk_init_root(const char *name, struct nameidata *nd)
522{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700523 struct fs_struct *fs = current->fs;
524
525 read_lock(&fs->lock);
526 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
527 nd->mnt = mntget(fs->altrootmnt);
528 nd->dentry = dget(fs->altroot);
529 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 if (__emul_lookup_dentry(name,nd))
531 return 0;
Andreas Mohre518ddb2006-09-29 02:01:22 -0700532 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700534 nd->mnt = mntget(fs->rootmnt);
535 nd->dentry = dget(fs->root);
536 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 return 1;
538}
539
Arjan van de Venf1662352006-01-14 13:21:31 -0800540static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541{
542 int res = 0;
543 char *name;
544 if (IS_ERR(link))
545 goto fail;
546
547 if (*link == '/') {
548 path_release(nd);
549 if (!walk_init_root(link, nd))
550 /* weird __emul_prefix() stuff did it */
551 goto out;
552 }
553 res = link_path_walk(link, nd);
554out:
555 if (nd->depth || res || nd->last_type!=LAST_NORM)
556 return res;
557 /*
558 * If it is an iterative symlinks resolution in open_namei() we
559 * have to copy the last component. And all that crap because of
560 * bloody create() on broken symlinks. Furrfu...
561 */
562 name = __getname();
563 if (unlikely(!name)) {
564 path_release(nd);
565 return -ENOMEM;
566 }
567 strcpy(name, nd->last.name);
568 nd->last.name = name;
569 return 0;
570fail:
571 path_release(nd);
572 return PTR_ERR(link);
573}
574
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700575static inline void dput_path(struct path *path, struct nameidata *nd)
576{
577 dput(path->dentry);
578 if (path->mnt != nd->mnt)
579 mntput(path->mnt);
580}
581
582static inline void path_to_nameidata(struct path *path, struct nameidata *nd)
583{
584 dput(nd->dentry);
585 if (nd->mnt != path->mnt)
586 mntput(nd->mnt);
587 nd->mnt = path->mnt;
588 nd->dentry = path->dentry;
589}
590
Ian Kent051d3812006-03-27 01:14:53 -0800591static __always_inline int __do_follow_link(struct path *path, struct nameidata *nd)
592{
593 int error;
594 void *cookie;
595 struct dentry *dentry = path->dentry;
596
597 touch_atime(path->mnt, dentry);
598 nd_set_link(nd, NULL);
599
600 if (path->mnt != nd->mnt) {
601 path_to_nameidata(path, nd);
602 dget(dentry);
603 }
604 mntget(path->mnt);
605 cookie = dentry->d_inode->i_op->follow_link(dentry, nd);
606 error = PTR_ERR(cookie);
607 if (!IS_ERR(cookie)) {
608 char *s = nd_get_link(nd);
609 error = 0;
610 if (s)
611 error = __vfs_follow_link(nd, s);
612 if (dentry->d_inode->i_op->put_link)
613 dentry->d_inode->i_op->put_link(dentry, nd, cookie);
614 }
615 dput(dentry);
616 mntput(path->mnt);
617
618 return error;
619}
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621/*
622 * This limits recursive symlink follows to 8, while
623 * limiting consecutive symlinks to 40.
624 *
625 * Without that kind of total limit, nasty chains of consecutive
626 * symlinks can cause almost arbitrarily long lookups.
627 */
Al Viro90ebe562005-06-06 13:35:58 -0700628static inline int do_follow_link(struct path *path, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
630 int err = -ELOOP;
631 if (current->link_count >= MAX_NESTED_LINKS)
632 goto loop;
633 if (current->total_link_count >= 40)
634 goto loop;
635 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
636 cond_resched();
Al Viro90ebe562005-06-06 13:35:58 -0700637 err = security_inode_follow_link(path->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (err)
639 goto loop;
640 current->link_count++;
641 current->total_link_count++;
642 nd->depth++;
Al Virocd4e91d2005-06-06 13:36:03 -0700643 err = __do_follow_link(path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 current->link_count--;
645 nd->depth--;
646 return err;
647loop:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700648 dput_path(path, nd);
Al Viro839d9f92005-06-06 13:36:02 -0700649 path_release(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650 return err;
651}
652
653int follow_up(struct vfsmount **mnt, struct dentry **dentry)
654{
655 struct vfsmount *parent;
656 struct dentry *mountpoint;
657 spin_lock(&vfsmount_lock);
658 parent=(*mnt)->mnt_parent;
659 if (parent == *mnt) {
660 spin_unlock(&vfsmount_lock);
661 return 0;
662 }
663 mntget(parent);
664 mountpoint=dget((*mnt)->mnt_mountpoint);
665 spin_unlock(&vfsmount_lock);
666 dput(*dentry);
667 *dentry = mountpoint;
668 mntput(*mnt);
669 *mnt = parent;
670 return 1;
671}
672
673/* no need for dcache_lock, as serialization is taken care in
674 * namespace.c
675 */
Al Viro463ffb22005-06-06 13:36:05 -0700676static int __follow_mount(struct path *path)
677{
678 int res = 0;
679 while (d_mountpoint(path->dentry)) {
680 struct vfsmount *mounted = lookup_mnt(path->mnt, path->dentry);
681 if (!mounted)
682 break;
683 dput(path->dentry);
684 if (res)
685 mntput(path->mnt);
686 path->mnt = mounted;
687 path->dentry = dget(mounted->mnt_root);
688 res = 1;
689 }
690 return res;
691}
692
Al Viro58c465e2005-06-06 13:36:13 -0700693static void follow_mount(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 while (d_mountpoint(*dentry)) {
696 struct vfsmount *mounted = lookup_mnt(*mnt, *dentry);
697 if (!mounted)
698 break;
Al Viro58c465e2005-06-06 13:36:13 -0700699 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 mntput(*mnt);
701 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 *dentry = dget(mounted->mnt_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704}
705
706/* no need for dcache_lock, as serialization is taken care in
707 * namespace.c
708 */
Al Viroe13b2102005-06-06 13:36:06 -0700709int follow_down(struct vfsmount **mnt, struct dentry **dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710{
711 struct vfsmount *mounted;
712
713 mounted = lookup_mnt(*mnt, *dentry);
714 if (mounted) {
Al Viroe13b2102005-06-06 13:36:06 -0700715 dput(*dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 mntput(*mnt);
717 *mnt = mounted;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 *dentry = dget(mounted->mnt_root);
719 return 1;
720 }
721 return 0;
722}
723
Arjan van de Venf1662352006-01-14 13:21:31 -0800724static __always_inline void follow_dotdot(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Andreas Mohre518ddb2006-09-29 02:01:22 -0700726 struct fs_struct *fs = current->fs;
727
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 while(1) {
729 struct vfsmount *parent;
Al Viro58c465e2005-06-06 13:36:13 -0700730 struct dentry *old = nd->dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
Andreas Mohre518ddb2006-09-29 02:01:22 -0700732 read_lock(&fs->lock);
733 if (nd->dentry == fs->root &&
734 nd->mnt == fs->rootmnt) {
735 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 break;
737 }
Andreas Mohre518ddb2006-09-29 02:01:22 -0700738 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739 spin_lock(&dcache_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700740 if (nd->dentry != nd->mnt->mnt_root) {
741 nd->dentry = dget(nd->dentry->d_parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 spin_unlock(&dcache_lock);
743 dput(old);
744 break;
745 }
746 spin_unlock(&dcache_lock);
747 spin_lock(&vfsmount_lock);
Al Viro58c465e2005-06-06 13:36:13 -0700748 parent = nd->mnt->mnt_parent;
749 if (parent == nd->mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 spin_unlock(&vfsmount_lock);
751 break;
752 }
753 mntget(parent);
Al Viro58c465e2005-06-06 13:36:13 -0700754 nd->dentry = dget(nd->mnt->mnt_mountpoint);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 spin_unlock(&vfsmount_lock);
756 dput(old);
Al Viro58c465e2005-06-06 13:36:13 -0700757 mntput(nd->mnt);
758 nd->mnt = parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 }
Al Viro58c465e2005-06-06 13:36:13 -0700760 follow_mount(&nd->mnt, &nd->dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
762
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763/*
764 * It's more convoluted than I'd like it to be, but... it's still fairly
765 * small and for now I'd prefer to have fast path as straight as possible.
766 * It _is_ time-critical.
767 */
768static int do_lookup(struct nameidata *nd, struct qstr *name,
769 struct path *path)
770{
771 struct vfsmount *mnt = nd->mnt;
772 struct dentry *dentry = __d_lookup(nd->dentry, name);
773
774 if (!dentry)
775 goto need_lookup;
776 if (dentry->d_op && dentry->d_op->d_revalidate)
777 goto need_revalidate;
778done:
779 path->mnt = mnt;
780 path->dentry = dentry;
Al Viro634ee702005-06-06 13:36:13 -0700781 __follow_mount(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 return 0;
783
784need_lookup:
785 dentry = real_lookup(nd->dentry, name, nd);
786 if (IS_ERR(dentry))
787 goto fail;
788 goto done;
789
790need_revalidate:
Ian Kentbcdc5e02006-09-27 01:50:44 -0700791 dentry = do_revalidate(dentry, nd);
792 if (!dentry)
793 goto need_lookup;
794 if (IS_ERR(dentry))
795 goto fail;
796 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797
798fail:
799 return PTR_ERR(dentry);
800}
801
802/*
803 * Name resolution.
Prasanna Medaea3834d2005-04-29 16:00:17 +0100804 * This is the basic name resolution function, turning a pathname into
805 * the final dentry. We expect 'base' to be positive and a directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 *
Prasanna Medaea3834d2005-04-29 16:00:17 +0100807 * Returns 0 and nd will have valid dentry and mnt on success.
808 * Returns error and drops reference to input namei data on failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 */
810static fastcall int __link_path_walk(const char * name, struct nameidata *nd)
811{
812 struct path next;
813 struct inode *inode;
814 int err;
815 unsigned int lookup_flags = nd->flags;
816
817 while (*name=='/')
818 name++;
819 if (!*name)
820 goto return_reval;
821
822 inode = nd->dentry->d_inode;
823 if (nd->depth)
Trond Myklebustf55eab82006-02-04 23:28:01 -0800824 lookup_flags = LOOKUP_FOLLOW | (nd->flags & LOOKUP_CONTINUE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 /* At this point we know we have a real path component. */
827 for(;;) {
828 unsigned long hash;
829 struct qstr this;
830 unsigned int c;
831
Trond Myklebustcdce5d62005-10-18 14:20:18 -0700832 nd->flags |= LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 err = exec_permission_lite(inode, nd);
Christoph Hellwige4543ed2005-11-08 21:35:04 -0800834 if (err == -EAGAIN)
835 err = vfs_permission(nd, MAY_EXEC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 if (err)
837 break;
838
839 this.name = name;
840 c = *(const unsigned char *)name;
841
842 hash = init_name_hash();
843 do {
844 name++;
845 hash = partial_name_hash(c, hash);
846 c = *(const unsigned char *)name;
847 } while (c && (c != '/'));
848 this.len = name - (const char *) this.name;
849 this.hash = end_name_hash(hash);
850
851 /* remove trailing slashes? */
852 if (!c)
853 goto last_component;
854 while (*++name == '/');
855 if (!*name)
856 goto last_with_slashes;
857
858 /*
859 * "." and ".." are special - ".." especially so because it has
860 * to be able to know about the current root directory and
861 * parent relationships.
862 */
863 if (this.name[0] == '.') switch (this.len) {
864 default:
865 break;
866 case 2:
867 if (this.name[1] != '.')
868 break;
Al Viro58c465e2005-06-06 13:36:13 -0700869 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700870 inode = nd->dentry->d_inode;
871 /* fallthrough */
872 case 1:
873 continue;
874 }
875 /*
876 * See if the low-level filesystem might want
877 * to use its own hash..
878 */
879 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
880 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
881 if (err < 0)
882 break;
883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 /* This does the actual lookups.. */
885 err = do_lookup(nd, &this, &next);
886 if (err)
887 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
889 err = -ENOENT;
890 inode = next.dentry->d_inode;
891 if (!inode)
892 goto out_dput;
893 err = -ENOTDIR;
894 if (!inode->i_op)
895 goto out_dput;
896
897 if (inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700898 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 if (err)
900 goto return_err;
901 err = -ENOENT;
902 inode = nd->dentry->d_inode;
903 if (!inode)
904 break;
905 err = -ENOTDIR;
906 if (!inode->i_op)
907 break;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700908 } else
909 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 err = -ENOTDIR;
911 if (!inode->i_op->lookup)
912 break;
913 continue;
914 /* here ends the main loop */
915
916last_with_slashes:
917 lookup_flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
918last_component:
Trond Myklebustf55eab82006-02-04 23:28:01 -0800919 /* Clear LOOKUP_CONTINUE iff it was previously unset */
920 nd->flags &= lookup_flags | ~LOOKUP_CONTINUE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 if (lookup_flags & LOOKUP_PARENT)
922 goto lookup_parent;
923 if (this.name[0] == '.') switch (this.len) {
924 default:
925 break;
926 case 2:
927 if (this.name[1] != '.')
928 break;
Al Viro58c465e2005-06-06 13:36:13 -0700929 follow_dotdot(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 inode = nd->dentry->d_inode;
931 /* fallthrough */
932 case 1:
933 goto return_reval;
934 }
935 if (nd->dentry->d_op && nd->dentry->d_op->d_hash) {
936 err = nd->dentry->d_op->d_hash(nd->dentry, &this);
937 if (err < 0)
938 break;
939 }
940 err = do_lookup(nd, &this, &next);
941 if (err)
942 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 inode = next.dentry->d_inode;
944 if ((lookup_flags & LOOKUP_FOLLOW)
945 && inode && inode->i_op && inode->i_op->follow_link) {
Al Viro90ebe562005-06-06 13:35:58 -0700946 err = do_follow_link(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 if (err)
948 goto return_err;
949 inode = nd->dentry->d_inode;
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700950 } else
951 path_to_nameidata(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 err = -ENOENT;
953 if (!inode)
954 break;
955 if (lookup_flags & LOOKUP_DIRECTORY) {
956 err = -ENOTDIR;
957 if (!inode->i_op || !inode->i_op->lookup)
958 break;
959 }
960 goto return_base;
961lookup_parent:
962 nd->last = this;
963 nd->last_type = LAST_NORM;
964 if (this.name[0] != '.')
965 goto return_base;
966 if (this.len == 1)
967 nd->last_type = LAST_DOT;
968 else if (this.len == 2 && this.name[1] == '.')
969 nd->last_type = LAST_DOTDOT;
970 else
971 goto return_base;
972return_reval:
973 /*
974 * We bypassed the ordinary revalidation routines.
975 * We may need to check the cached dentry for staleness.
976 */
977 if (nd->dentry && nd->dentry->d_sb &&
978 (nd->dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)) {
979 err = -ESTALE;
980 /* Note: we do not d_invalidate() */
981 if (!nd->dentry->d_op->d_revalidate(nd->dentry, nd))
982 break;
983 }
984return_base:
985 return 0;
986out_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -0700987 dput_path(&next, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 break;
989 }
990 path_release(nd);
991return_err:
992 return err;
993}
994
995/*
996 * Wrapper to retry pathname resolution whenever the underlying
997 * file system returns an ESTALE.
998 *
999 * Retry the whole path once, forcing real lookup requests
1000 * instead of relying on the dcache.
1001 */
1002int fastcall link_path_walk(const char *name, struct nameidata *nd)
1003{
1004 struct nameidata save = *nd;
1005 int result;
1006
1007 /* make sure the stuff we saved doesn't go away */
1008 dget(save.dentry);
1009 mntget(save.mnt);
1010
1011 result = __link_path_walk(name, nd);
1012 if (result == -ESTALE) {
1013 *nd = save;
1014 dget(nd->dentry);
1015 mntget(nd->mnt);
1016 nd->flags |= LOOKUP_REVAL;
1017 result = __link_path_walk(name, nd);
1018 }
1019
1020 dput(save.dentry);
1021 mntput(save.mnt);
1022
1023 return result;
1024}
1025
1026int fastcall path_walk(const char * name, struct nameidata *nd)
1027{
1028 current->total_link_count = 0;
1029 return link_path_walk(name, nd);
1030}
1031
Prasanna Medaea3834d2005-04-29 16:00:17 +01001032/*
1033 * SMP-safe: Returns 1 and nd will have valid dentry and mnt, if
1034 * everything is done. Returns 0 and drops input nd, if lookup failed;
1035 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036static int __emul_lookup_dentry(const char *name, struct nameidata *nd)
1037{
1038 if (path_walk(name, nd))
1039 return 0; /* something went wrong... */
1040
1041 if (!nd->dentry->d_inode || S_ISDIR(nd->dentry->d_inode->i_mode)) {
1042 struct dentry *old_dentry = nd->dentry;
1043 struct vfsmount *old_mnt = nd->mnt;
1044 struct qstr last = nd->last;
1045 int last_type = nd->last_type;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001046 struct fs_struct *fs = current->fs;
1047
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 /*
Andreas Mohre518ddb2006-09-29 02:01:22 -07001049 * NAME was not found in alternate root or it's a directory.
1050 * Try to find it in the normal root:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
1052 nd->last_type = LAST_ROOT;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001053 read_lock(&fs->lock);
1054 nd->mnt = mntget(fs->rootmnt);
1055 nd->dentry = dget(fs->root);
1056 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (path_walk(name, nd) == 0) {
1058 if (nd->dentry->d_inode) {
1059 dput(old_dentry);
1060 mntput(old_mnt);
1061 return 1;
1062 }
1063 path_release(nd);
1064 }
1065 nd->dentry = old_dentry;
1066 nd->mnt = old_mnt;
1067 nd->last = last;
1068 nd->last_type = last_type;
1069 }
1070 return 1;
1071}
1072
1073void set_fs_altroot(void)
1074{
1075 char *emul = __emul_prefix();
1076 struct nameidata nd;
1077 struct vfsmount *mnt = NULL, *oldmnt;
1078 struct dentry *dentry = NULL, *olddentry;
1079 int err;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001080 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
1082 if (!emul)
1083 goto set_it;
1084 err = path_lookup(emul, LOOKUP_FOLLOW|LOOKUP_DIRECTORY|LOOKUP_NOALT, &nd);
1085 if (!err) {
1086 mnt = nd.mnt;
1087 dentry = nd.dentry;
1088 }
1089set_it:
Andreas Mohre518ddb2006-09-29 02:01:22 -07001090 write_lock(&fs->lock);
1091 oldmnt = fs->altrootmnt;
1092 olddentry = fs->altroot;
1093 fs->altrootmnt = mnt;
1094 fs->altroot = dentry;
1095 write_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 if (olddentry) {
1097 dput(olddentry);
1098 mntput(oldmnt);
1099 }
1100}
1101
Prasanna Medaea3834d2005-04-29 16:00:17 +01001102/* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001103static int fastcall do_path_lookup(int dfd, const char *name,
1104 unsigned int flags, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105{
Prasanna Medaea3834d2005-04-29 16:00:17 +01001106 int retval = 0;
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001107 int fput_needed;
1108 struct file *file;
Andreas Mohre518ddb2006-09-29 02:01:22 -07001109 struct fs_struct *fs = current->fs;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110
1111 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1112 nd->flags = flags;
1113 nd->depth = 0;
1114
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 if (*name=='/') {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001116 read_lock(&fs->lock);
1117 if (fs->altroot && !(nd->flags & LOOKUP_NOALT)) {
1118 nd->mnt = mntget(fs->altrootmnt);
1119 nd->dentry = dget(fs->altroot);
1120 read_unlock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 if (__emul_lookup_dentry(name,nd))
Prasanna Medaea3834d2005-04-29 16:00:17 +01001122 goto out; /* found in altroot */
Andreas Mohre518ddb2006-09-29 02:01:22 -07001123 read_lock(&fs->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
Andreas Mohre518ddb2006-09-29 02:01:22 -07001125 nd->mnt = mntget(fs->rootmnt);
1126 nd->dentry = dget(fs->root);
1127 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001128 } else if (dfd == AT_FDCWD) {
Andreas Mohre518ddb2006-09-29 02:01:22 -07001129 read_lock(&fs->lock);
1130 nd->mnt = mntget(fs->pwdmnt);
1131 nd->dentry = dget(fs->pwd);
1132 read_unlock(&fs->lock);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001133 } else {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001134 struct dentry *dentry;
1135
1136 file = fget_light(dfd, &fput_needed);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001137 retval = -EBADF;
1138 if (!file)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001139 goto out_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001140
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001141 dentry = file->f_path.dentry;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001142
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001143 retval = -ENOTDIR;
1144 if (!S_ISDIR(dentry->d_inode->i_mode))
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001145 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001146
1147 retval = file_permission(file, MAY_EXEC);
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001148 if (retval)
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001149 goto fput_fail;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001150
Josef "Jeff" Sipek0f7fc9e2006-12-08 02:36:35 -08001151 nd->mnt = mntget(file->f_path.mnt);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001152 nd->dentry = dget(dentry);
1153
1154 fput_light(file, fput_needed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156 current->total_link_count = 0;
1157 retval = link_path_walk(name, nd);
Prasanna Medaea3834d2005-04-29 16:00:17 +01001158out:
Suzuki3bc84142006-02-07 12:58:36 -08001159 if (likely(retval == 0)) {
Al Viro5ac3a9c2006-07-16 06:38:45 -04001160 if (unlikely(!audit_dummy_context() && nd && nd->dentry &&
Suzuki3bc84142006-02-07 12:58:36 -08001161 nd->dentry->d_inode))
Amy Griffis9c937dc2006-06-08 23:19:31 -04001162 audit_inode(name, nd->dentry->d_inode);
Suzuki3bc84142006-02-07 12:58:36 -08001163 }
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001164out_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001165 return retval;
1166
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001167fput_fail:
Ulrich Drepper170aa3d2006-02-04 23:28:02 -08001168 fput_light(file, fput_needed);
Trond Myklebust6d09bb62006-06-04 02:51:37 -07001169 goto out_fail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170}
1171
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001172int fastcall path_lookup(const char *name, unsigned int flags,
1173 struct nameidata *nd)
1174{
1175 return do_path_lookup(AT_FDCWD, name, flags, nd);
1176}
1177
1178static int __path_lookup_intent_open(int dfd, const char *name,
1179 unsigned int lookup_flags, struct nameidata *nd,
1180 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001181{
1182 struct file *filp = get_empty_filp();
1183 int err;
1184
1185 if (filp == NULL)
1186 return -ENFILE;
1187 nd->intent.open.file = filp;
1188 nd->intent.open.flags = open_flags;
1189 nd->intent.open.create_mode = create_mode;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001190 err = do_path_lookup(dfd, name, lookup_flags|LOOKUP_OPEN, nd);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001191 if (IS_ERR(nd->intent.open.file)) {
1192 if (err == 0) {
1193 err = PTR_ERR(nd->intent.open.file);
1194 path_release(nd);
1195 }
1196 } else if (err != 0)
1197 release_open_intent(nd);
1198 return err;
1199}
1200
1201/**
1202 * path_lookup_open - lookup a file path with open intent
Martin Waitz7045f372006-02-01 03:06:57 -08001203 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001204 * @name: pointer to file name
1205 * @lookup_flags: lookup intent flags
1206 * @nd: pointer to nameidata
1207 * @open_flags: open intent flags
1208 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001209int path_lookup_open(int dfd, const char *name, unsigned int lookup_flags,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001210 struct nameidata *nd, int open_flags)
1211{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001212 return __path_lookup_intent_open(dfd, name, lookup_flags, nd,
Trond Myklebust834f2a42005-10-18 14:20:16 -07001213 open_flags, 0);
1214}
1215
1216/**
1217 * path_lookup_create - lookup a file path with open + create intent
Martin Waitz7045f372006-02-01 03:06:57 -08001218 * @dfd: the directory to use as base, or AT_FDCWD
Trond Myklebust834f2a42005-10-18 14:20:16 -07001219 * @name: pointer to file name
1220 * @lookup_flags: lookup intent flags
1221 * @nd: pointer to nameidata
1222 * @open_flags: open intent flags
1223 * @create_mode: create intent flags
1224 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001225static int path_lookup_create(int dfd, const char *name,
1226 unsigned int lookup_flags, struct nameidata *nd,
1227 int open_flags, int create_mode)
Trond Myklebust834f2a42005-10-18 14:20:16 -07001228{
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001229 return __path_lookup_intent_open(dfd, name, lookup_flags|LOOKUP_CREATE,
1230 nd, open_flags, create_mode);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001231}
1232
1233int __user_path_lookup_open(const char __user *name, unsigned int lookup_flags,
1234 struct nameidata *nd, int open_flags)
1235{
1236 char *tmp = getname(name);
1237 int err = PTR_ERR(tmp);
1238
1239 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001240 err = __path_lookup_intent_open(AT_FDCWD, tmp, lookup_flags, nd, open_flags, 0);
Trond Myklebust834f2a42005-10-18 14:20:16 -07001241 putname(tmp);
1242 }
1243 return err;
1244}
1245
James Morris057f6c02007-04-26 00:12:05 -07001246static inline struct dentry *__lookup_hash_kern(struct qstr *name, struct dentry *base, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
James Morris057f6c02007-04-26 00:12:05 -07001248 struct dentry *dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 struct inode *inode;
1250 int err;
1251
1252 inode = base->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254 /*
1255 * See if the low-level filesystem might want
1256 * to use its own hash..
1257 */
1258 if (base->d_op && base->d_op->d_hash) {
1259 err = base->d_op->d_hash(base, name);
1260 dentry = ERR_PTR(err);
1261 if (err < 0)
1262 goto out;
1263 }
1264
1265 dentry = cached_lookup(base, name, nd);
1266 if (!dentry) {
1267 struct dentry *new = d_alloc(base, name);
1268 dentry = ERR_PTR(-ENOMEM);
1269 if (!new)
1270 goto out;
1271 dentry = inode->i_op->lookup(inode, new, nd);
1272 if (!dentry)
1273 dentry = new;
1274 else
1275 dput(new);
1276 }
1277out:
1278 return dentry;
1279}
1280
James Morris057f6c02007-04-26 00:12:05 -07001281/*
1282 * Restricted form of lookup. Doesn't follow links, single-component only,
1283 * needs parent already locked. Doesn't follow mounts.
1284 * SMP-safe.
1285 */
1286static inline struct dentry * __lookup_hash(struct qstr *name, struct dentry *base, struct nameidata *nd)
1287{
1288 struct dentry *dentry;
1289 struct inode *inode;
1290 int err;
1291
1292 inode = base->d_inode;
1293
1294 err = permission(inode, MAY_EXEC, nd);
1295 dentry = ERR_PTR(err);
1296 if (err)
1297 goto out;
1298
1299 dentry = __lookup_hash_kern(name, base, nd);
1300out:
1301 return dentry;
1302}
1303
Adrian Bunka244e162006-03-31 02:32:11 -08001304static struct dentry *lookup_hash(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305{
Christoph Hellwig49705b72005-11-08 21:35:06 -08001306 return __lookup_hash(&nd->last, nd->dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307}
1308
1309/* SMP-safe */
James Morris057f6c02007-04-26 00:12:05 -07001310static inline int __lookup_one_len(const char *name, struct qstr *this, struct dentry *base, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311{
1312 unsigned long hash;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 unsigned int c;
1314
James Morris057f6c02007-04-26 00:12:05 -07001315 this->name = name;
1316 this->len = len;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 if (!len)
James Morris057f6c02007-04-26 00:12:05 -07001318 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
1320 hash = init_name_hash();
1321 while (len--) {
1322 c = *(const unsigned char *)name++;
1323 if (c == '/' || c == '\0')
James Morris057f6c02007-04-26 00:12:05 -07001324 return -EACCES;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 hash = partial_name_hash(c, hash);
1326 }
James Morris057f6c02007-04-26 00:12:05 -07001327 this->hash = end_name_hash(hash);
1328 return 0;
1329}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330
James Morris057f6c02007-04-26 00:12:05 -07001331struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1332{
1333 int err;
1334 struct qstr this;
1335
1336 err = __lookup_one_len(name, &this, base, len);
1337 if (err)
1338 return ERR_PTR(err);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001339 return __lookup_hash(&this, base, NULL);
James Morris057f6c02007-04-26 00:12:05 -07001340}
1341
1342struct dentry *lookup_one_len_kern(const char *name, struct dentry *base, int len)
1343{
1344 int err;
1345 struct qstr this;
1346
1347 err = __lookup_one_len(name, &this, base, len);
1348 if (err)
1349 return ERR_PTR(err);
1350 return __lookup_hash_kern(&this, base, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351}
1352
1353/*
1354 * namei()
1355 *
1356 * is used by most simple commands to get the inode of a specified name.
1357 * Open, link etc use their own routines, but this is enough for things
1358 * like 'chmod' etc.
1359 *
1360 * namei exists in two versions: namei/lnamei. The only difference is
1361 * that namei follows links, while lnamei does not.
1362 * SMP-safe
1363 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001364int fastcall __user_walk_fd(int dfd, const char __user *name, unsigned flags,
1365 struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 char *tmp = getname(name);
1368 int err = PTR_ERR(tmp);
1369
1370 if (!IS_ERR(tmp)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001371 err = do_path_lookup(dfd, tmp, flags, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 putname(tmp);
1373 }
1374 return err;
1375}
1376
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001377int fastcall __user_walk(const char __user *name, unsigned flags, struct nameidata *nd)
1378{
1379 return __user_walk_fd(AT_FDCWD, name, flags, nd);
1380}
1381
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382/*
1383 * It's inline, so penalty for filesystems that don't use sticky bit is
1384 * minimal.
1385 */
1386static inline int check_sticky(struct inode *dir, struct inode *inode)
1387{
1388 if (!(dir->i_mode & S_ISVTX))
1389 return 0;
1390 if (inode->i_uid == current->fsuid)
1391 return 0;
1392 if (dir->i_uid == current->fsuid)
1393 return 0;
1394 return !capable(CAP_FOWNER);
1395}
1396
1397/*
1398 * Check whether we can remove a link victim from directory dir, check
1399 * whether the type of victim is right.
1400 * 1. We can't do it if dir is read-only (done in permission())
1401 * 2. We should have write and exec permissions on dir
1402 * 3. We can't remove anything from append-only dir
1403 * 4. We can't do anything with immutable dir (done in permission())
1404 * 5. If the sticky bit on dir is set we should either
1405 * a. be owner of dir, or
1406 * b. be owner of victim, or
1407 * c. have CAP_FOWNER capability
1408 * 6. If the victim is append-only or immutable we can't do antyhing with
1409 * links pointing to it.
1410 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1411 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1412 * 9. We can't remove a root or mountpoint.
1413 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1414 * nfs_async_unlink().
1415 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001416static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417{
1418 int error;
1419
1420 if (!victim->d_inode)
1421 return -ENOENT;
1422
1423 BUG_ON(victim->d_parent->d_inode != dir);
Amy Griffis73d3ec52006-07-13 13:16:39 -04001424 audit_inode_child(victim->d_name.name, victim->d_inode, dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
1426 error = permission(dir,MAY_WRITE | MAY_EXEC, NULL);
1427 if (error)
1428 return error;
1429 if (IS_APPEND(dir))
1430 return -EPERM;
1431 if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1432 IS_IMMUTABLE(victim->d_inode))
1433 return -EPERM;
1434 if (isdir) {
1435 if (!S_ISDIR(victim->d_inode->i_mode))
1436 return -ENOTDIR;
1437 if (IS_ROOT(victim))
1438 return -EBUSY;
1439 } else if (S_ISDIR(victim->d_inode->i_mode))
1440 return -EISDIR;
1441 if (IS_DEADDIR(dir))
1442 return -ENOENT;
1443 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1444 return -EBUSY;
1445 return 0;
1446}
1447
1448/* Check whether we can create an object with dentry child in directory
1449 * dir.
1450 * 1. We can't do it if child already exists (open has special treatment for
1451 * this case, but since we are inlined it's OK)
1452 * 2. We can't do it if dir is read-only (done in permission())
1453 * 3. We should have write and exec permissions on dir
1454 * 4. We can't do it if dir is immutable (done in permission())
1455 */
1456static inline int may_create(struct inode *dir, struct dentry *child,
1457 struct nameidata *nd)
1458{
1459 if (child->d_inode)
1460 return -EEXIST;
1461 if (IS_DEADDIR(dir))
1462 return -ENOENT;
1463 return permission(dir,MAY_WRITE | MAY_EXEC, nd);
1464}
1465
1466/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 * O_DIRECTORY translates into forcing a directory lookup.
1468 */
1469static inline int lookup_flags(unsigned int f)
1470{
1471 unsigned long retval = LOOKUP_FOLLOW;
1472
1473 if (f & O_NOFOLLOW)
1474 retval &= ~LOOKUP_FOLLOW;
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (f & O_DIRECTORY)
1477 retval |= LOOKUP_DIRECTORY;
1478
1479 return retval;
1480}
1481
1482/*
1483 * p1 and p2 should be directories on the same fs.
1484 */
1485struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1486{
1487 struct dentry *p;
1488
1489 if (p1 == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001490 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 return NULL;
1492 }
1493
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001494 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495
1496 for (p = p1; p->d_parent != p; p = p->d_parent) {
1497 if (p->d_parent == p2) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001498 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1499 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 return p;
1501 }
1502 }
1503
1504 for (p = p2; p->d_parent != p; p = p->d_parent) {
1505 if (p->d_parent == p1) {
Ingo Molnarf2eace22006-07-03 00:25:05 -07001506 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1507 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return p;
1509 }
1510 }
1511
Ingo Molnarf2eace22006-07-03 00:25:05 -07001512 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1513 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 return NULL;
1515}
1516
1517void unlock_rename(struct dentry *p1, struct dentry *p2)
1518{
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001519 mutex_unlock(&p1->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 if (p1 != p2) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001521 mutex_unlock(&p2->d_inode->i_mutex);
Arjan van de Vena11f3a02006-03-23 03:00:33 -08001522 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 }
1524}
1525
1526int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1527 struct nameidata *nd)
1528{
1529 int error = may_create(dir, dentry, nd);
1530
1531 if (error)
1532 return error;
1533
1534 if (!dir->i_op || !dir->i_op->create)
1535 return -EACCES; /* shouldn't it be ENOSYS? */
1536 mode &= S_IALLUGO;
1537 mode |= S_IFREG;
1538 error = security_inode_create(dir, dentry, mode);
1539 if (error)
1540 return error;
1541 DQUOT_INIT(dir);
1542 error = dir->i_op->create(dir, dentry, mode, nd);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001543 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001544 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545 return error;
1546}
1547
1548int may_open(struct nameidata *nd, int acc_mode, int flag)
1549{
1550 struct dentry *dentry = nd->dentry;
1551 struct inode *inode = dentry->d_inode;
1552 int error;
1553
1554 if (!inode)
1555 return -ENOENT;
1556
1557 if (S_ISLNK(inode->i_mode))
1558 return -ELOOP;
1559
1560 if (S_ISDIR(inode->i_mode) && (flag & FMODE_WRITE))
1561 return -EISDIR;
1562
Christoph Hellwige4543ed2005-11-08 21:35:04 -08001563 error = vfs_permission(nd, acc_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 if (error)
1565 return error;
1566
1567 /*
1568 * FIFO's, sockets and device files are special: they don't
1569 * actually live on the filesystem itself, and as such you
1570 * can write to them even if the filesystem is read-only.
1571 */
1572 if (S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
1573 flag &= ~O_TRUNC;
1574 } else if (S_ISBLK(inode->i_mode) || S_ISCHR(inode->i_mode)) {
1575 if (nd->mnt->mnt_flags & MNT_NODEV)
1576 return -EACCES;
1577
1578 flag &= ~O_TRUNC;
1579 } else if (IS_RDONLY(inode) && (flag & FMODE_WRITE))
1580 return -EROFS;
1581 /*
1582 * An append-only file must be opened in append mode for writing.
1583 */
1584 if (IS_APPEND(inode)) {
1585 if ((flag & FMODE_WRITE) && !(flag & O_APPEND))
1586 return -EPERM;
1587 if (flag & O_TRUNC)
1588 return -EPERM;
1589 }
1590
1591 /* O_NOATIME can only be set by the owner or superuser */
1592 if (flag & O_NOATIME)
1593 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
1594 return -EPERM;
1595
1596 /*
1597 * Ensure there are no outstanding leases on the file.
1598 */
1599 error = break_lease(inode, flag);
1600 if (error)
1601 return error;
1602
1603 if (flag & O_TRUNC) {
1604 error = get_write_access(inode);
1605 if (error)
1606 return error;
1607
1608 /*
1609 * Refuse to truncate files with mandatory locks held on them.
1610 */
1611 error = locks_verify_locked(inode);
1612 if (!error) {
1613 DQUOT_INIT(inode);
1614
NeilBrown4a301312006-01-08 01:02:39 -08001615 error = do_truncate(dentry, 0, ATTR_MTIME|ATTR_CTIME, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616 }
1617 put_write_access(inode);
1618 if (error)
1619 return error;
1620 } else
1621 if (flag & FMODE_WRITE)
1622 DQUOT_INIT(inode);
1623
1624 return 0;
1625}
1626
Dave Hansenaab520e2006-09-30 23:29:02 -07001627static int open_namei_create(struct nameidata *nd, struct path *path,
1628 int flag, int mode)
1629{
1630 int error;
1631 struct dentry *dir = nd->dentry;
1632
1633 if (!IS_POSIXACL(dir->d_inode))
1634 mode &= ~current->fs->umask;
1635 error = vfs_create(dir->d_inode, path->dentry, mode, nd);
1636 mutex_unlock(&dir->d_inode->i_mutex);
1637 dput(nd->dentry);
1638 nd->dentry = path->dentry;
1639 if (error)
1640 return error;
1641 /* Don't check for write permission, don't truncate */
1642 return may_open(nd, 0, flag & ~O_TRUNC);
1643}
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645/*
1646 * open_namei()
1647 *
1648 * namei for open - this is in fact almost the whole open-routine.
1649 *
1650 * Note that the low bits of "flag" aren't the same as in the open
1651 * system call - they are 00 - no permissions needed
1652 * 01 - read permission needed
1653 * 10 - write permission needed
1654 * 11 - read/write permissions needed
1655 * which is a lot more logical, and also allows the "no perm" needed
1656 * for symlinks (where the permissions are checked later).
1657 * SMP-safe
1658 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001659int open_namei(int dfd, const char *pathname, int flag,
1660 int mode, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661{
Trond Myklebust834f2a42005-10-18 14:20:16 -07001662 int acc_mode, error;
Al Viro4e7506e2005-06-06 13:36:00 -07001663 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664 struct dentry *dir;
1665 int count = 0;
1666
1667 acc_mode = ACC_MODE(flag);
1668
Trond Myklebust834f2a42005-10-18 14:20:16 -07001669 /* O_TRUNC implies we need access checks for write permissions */
1670 if (flag & O_TRUNC)
1671 acc_mode |= MAY_WRITE;
1672
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 /* Allow the LSM permission hook to distinguish append
1674 access from general write access. */
1675 if (flag & O_APPEND)
1676 acc_mode |= MAY_APPEND;
1677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 /*
1679 * The simplest case - just a plain lookup.
1680 */
1681 if (!(flag & O_CREAT)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001682 error = path_lookup_open(dfd, pathname, lookup_flags(flag),
1683 nd, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684 if (error)
1685 return error;
1686 goto ok;
1687 }
1688
1689 /*
1690 * Create - we need to know the parent.
1691 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001692 error = path_lookup_create(dfd,pathname,LOOKUP_PARENT,nd,flag,mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693 if (error)
1694 return error;
1695
1696 /*
1697 * We have the parent and last component. First of all, check
1698 * that we are not asked to creat(2) an obvious directory - that
1699 * will not do.
1700 */
1701 error = -EISDIR;
1702 if (nd->last_type != LAST_NORM || nd->last.name[nd->last.len])
1703 goto exit;
1704
1705 dir = nd->dentry;
1706 nd->flags &= ~LOOKUP_PARENT;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001707 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001708 path.dentry = lookup_hash(nd);
Al Virod73ffe12005-06-06 13:36:01 -07001709 path.mnt = nd->mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711do_last:
Al Viro4e7506e2005-06-06 13:36:00 -07001712 error = PTR_ERR(path.dentry);
1713 if (IS_ERR(path.dentry)) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001714 mutex_unlock(&dir->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 goto exit;
1716 }
1717
Oleg Drokin4af4c522006-03-25 03:06:54 -08001718 if (IS_ERR(nd->intent.open.file)) {
1719 mutex_unlock(&dir->d_inode->i_mutex);
1720 error = PTR_ERR(nd->intent.open.file);
1721 goto exit_dput;
1722 }
1723
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 /* Negative dentry, just create the file */
Al Viro4e7506e2005-06-06 13:36:00 -07001725 if (!path.dentry->d_inode) {
Dave Hansenaab520e2006-09-30 23:29:02 -07001726 error = open_namei_create(nd, &path, flag, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 if (error)
1728 goto exit;
Dave Hansenaab520e2006-09-30 23:29:02 -07001729 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 }
1731
1732 /*
1733 * It already exists.
1734 */
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001735 mutex_unlock(&dir->d_inode->i_mutex);
Amy Griffis3e2efce2006-07-13 13:16:02 -04001736 audit_inode_update(path.dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737
1738 error = -EEXIST;
1739 if (flag & O_EXCL)
1740 goto exit_dput;
1741
Al Viroe13b2102005-06-06 13:36:06 -07001742 if (__follow_mount(&path)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 error = -ELOOP;
Al Viroba7a4c12005-06-06 13:36:08 -07001744 if (flag & O_NOFOLLOW)
1745 goto exit_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 }
Amy Griffis3e2efce2006-07-13 13:16:02 -04001747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 error = -ENOENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001749 if (!path.dentry->d_inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 goto exit_dput;
Al Viro4e7506e2005-06-06 13:36:00 -07001751 if (path.dentry->d_inode->i_op && path.dentry->d_inode->i_op->follow_link)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 goto do_link;
1753
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001754 path_to_nameidata(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 error = -EISDIR;
Al Viro4e7506e2005-06-06 13:36:00 -07001756 if (path.dentry->d_inode && S_ISDIR(path.dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 goto exit;
1758ok:
1759 error = may_open(nd, acc_mode, flag);
1760 if (error)
1761 goto exit;
1762 return 0;
1763
1764exit_dput:
Miklos Szeredi09dd17d2005-09-06 15:18:21 -07001765 dput_path(&path, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766exit:
Trond Myklebust834f2a42005-10-18 14:20:16 -07001767 if (!IS_ERR(nd->intent.open.file))
1768 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 path_release(nd);
1770 return error;
1771
1772do_link:
1773 error = -ELOOP;
1774 if (flag & O_NOFOLLOW)
1775 goto exit_dput;
1776 /*
1777 * This is subtle. Instead of calling do_follow_link() we do the
1778 * thing by hands. The reason is that this way we have zero link_count
1779 * and path_walk() (called from ->follow_link) honoring LOOKUP_PARENT.
1780 * After that we have the parent and last component, i.e.
1781 * we are in the same situation as after the first path_walk().
1782 * Well, almost - if the last component is normal we get its copy
1783 * stored in nd->last.name and we will have to putname() it when we
1784 * are done. Procfs-like symlinks just set LAST_BIND.
1785 */
1786 nd->flags |= LOOKUP_PARENT;
Al Viro4e7506e2005-06-06 13:36:00 -07001787 error = security_inode_follow_link(path.dentry, nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001788 if (error)
1789 goto exit_dput;
Al Virocd4e91d2005-06-06 13:36:03 -07001790 error = __do_follow_link(&path, nd);
Kirill Korotaevde459212006-07-14 00:23:49 -07001791 if (error) {
1792 /* Does someone understand code flow here? Or it is only
1793 * me so stupid? Anathema to whoever designed this non-sense
1794 * with "intent.open".
1795 */
1796 release_open_intent(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 return error;
Kirill Korotaevde459212006-07-14 00:23:49 -07001798 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 nd->flags &= ~LOOKUP_PARENT;
Al Virod671d5e2005-06-06 13:36:04 -07001800 if (nd->last_type == LAST_BIND)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 goto ok;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 error = -EISDIR;
1803 if (nd->last_type != LAST_NORM)
1804 goto exit;
1805 if (nd->last.name[nd->last.len]) {
Linus Torvalds82984112005-10-06 21:54:21 -07001806 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 goto exit;
1808 }
1809 error = -ELOOP;
1810 if (count++==32) {
Linus Torvalds82984112005-10-06 21:54:21 -07001811 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 goto exit;
1813 }
1814 dir = nd->dentry;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001815 mutex_lock(&dir->d_inode->i_mutex);
Christoph Hellwig49705b72005-11-08 21:35:06 -08001816 path.dentry = lookup_hash(nd);
Al Virod671d5e2005-06-06 13:36:04 -07001817 path.mnt = nd->mnt;
Linus Torvalds82984112005-10-06 21:54:21 -07001818 __putname(nd->last.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 goto do_last;
1820}
1821
1822/**
1823 * lookup_create - lookup a dentry, creating it if it doesn't exist
1824 * @nd: nameidata info
1825 * @is_dir: directory flag
1826 *
1827 * Simple function to lookup and return a dentry and create it
1828 * if it doesn't exist. Is SMP-safe.
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001829 *
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001830 * Returns with nd->dentry->d_inode->i_mutex locked.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 */
1832struct dentry *lookup_create(struct nameidata *nd, int is_dir)
1833{
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001834 struct dentry *dentry = ERR_PTR(-EEXIST);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835
Ingo Molnarf2eace22006-07-03 00:25:05 -07001836 mutex_lock_nested(&nd->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001837 /*
1838 * Yucky last component or no last component at all?
1839 * (foo/., foo/.., /////)
1840 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001841 if (nd->last_type != LAST_NORM)
1842 goto fail;
1843 nd->flags &= ~LOOKUP_PARENT;
ASANO Masahiroa6349042006-08-22 20:06:02 -04001844 nd->flags |= LOOKUP_CREATE;
1845 nd->intent.open.flags = O_EXCL;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001846
1847 /*
1848 * Do the final lookup.
1849 */
Christoph Hellwig49705b72005-11-08 21:35:06 -08001850 dentry = lookup_hash(nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 if (IS_ERR(dentry))
1852 goto fail;
Christoph Hellwigc663e5d2005-06-23 00:09:49 -07001853
1854 /*
1855 * Special case - lookup gave negative, but... we had foo/bar/
1856 * From the vfs_mknod() POV we just have a negative dentry -
1857 * all is fine. Let's be bastards - you had / on the end, you've
1858 * been asking for (non-existent) directory. -ENOENT for you.
1859 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 if (!is_dir && nd->last.name[nd->last.len] && !dentry->d_inode)
1861 goto enoent;
1862 return dentry;
1863enoent:
1864 dput(dentry);
1865 dentry = ERR_PTR(-ENOENT);
1866fail:
1867 return dentry;
1868}
Christoph Hellwigf81a0bf2005-05-19 12:26:43 -07001869EXPORT_SYMBOL_GPL(lookup_create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
1871int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1872{
1873 int error = may_create(dir, dentry, NULL);
1874
1875 if (error)
1876 return error;
1877
1878 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
1879 return -EPERM;
1880
1881 if (!dir->i_op || !dir->i_op->mknod)
1882 return -EPERM;
1883
1884 error = security_inode_mknod(dir, dentry, mode, dev);
1885 if (error)
1886 return error;
1887
1888 DQUOT_INIT(dir);
1889 error = dir->i_op->mknod(dir, dentry, mode, dev);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001890 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001891 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 return error;
1893}
1894
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001895asmlinkage long sys_mknodat(int dfd, const char __user *filename, int mode,
1896 unsigned dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897{
1898 int error = 0;
1899 char * tmp;
1900 struct dentry * dentry;
1901 struct nameidata nd;
1902
1903 if (S_ISDIR(mode))
1904 return -EPERM;
1905 tmp = getname(filename);
1906 if (IS_ERR(tmp))
1907 return PTR_ERR(tmp);
1908
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001909 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 if (error)
1911 goto out;
1912 dentry = lookup_create(&nd, 0);
1913 error = PTR_ERR(dentry);
1914
1915 if (!IS_POSIXACL(nd.dentry->d_inode))
1916 mode &= ~current->fs->umask;
1917 if (!IS_ERR(dentry)) {
1918 switch (mode & S_IFMT) {
1919 case 0: case S_IFREG:
1920 error = vfs_create(nd.dentry->d_inode,dentry,mode,&nd);
1921 break;
1922 case S_IFCHR: case S_IFBLK:
1923 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,
1924 new_decode_dev(dev));
1925 break;
1926 case S_IFIFO: case S_IFSOCK:
1927 error = vfs_mknod(nd.dentry->d_inode,dentry,mode,0);
1928 break;
1929 case S_IFDIR:
1930 error = -EPERM;
1931 break;
1932 default:
1933 error = -EINVAL;
1934 }
1935 dput(dentry);
1936 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001937 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001938 path_release(&nd);
1939out:
1940 putname(tmp);
1941
1942 return error;
1943}
1944
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001945asmlinkage long sys_mknod(const char __user *filename, int mode, unsigned dev)
1946{
1947 return sys_mknodat(AT_FDCWD, filename, mode, dev);
1948}
1949
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1951{
1952 int error = may_create(dir, dentry, NULL);
1953
1954 if (error)
1955 return error;
1956
1957 if (!dir->i_op || !dir->i_op->mkdir)
1958 return -EPERM;
1959
1960 mode &= (S_IRWXUGO|S_ISVTX);
1961 error = security_inode_mkdir(dir, dentry, mode);
1962 if (error)
1963 return error;
1964
1965 DQUOT_INIT(dir);
1966 error = dir->i_op->mkdir(dir, dentry, mode);
Stephen Smalleya74574a2005-09-09 13:01:44 -07001967 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00001968 fsnotify_mkdir(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 return error;
1970}
1971
Ulrich Drepper5590ff02006-01-18 17:43:53 -08001972asmlinkage long sys_mkdirat(int dfd, const char __user *pathname, int mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973{
1974 int error = 0;
1975 char * tmp;
Dave Hansen6902d922006-09-30 23:29:01 -07001976 struct dentry *dentry;
1977 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978
1979 tmp = getname(pathname);
1980 error = PTR_ERR(tmp);
Dave Hansen6902d922006-09-30 23:29:01 -07001981 if (IS_ERR(tmp))
1982 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983
Dave Hansen6902d922006-09-30 23:29:01 -07001984 error = do_path_lookup(dfd, tmp, LOOKUP_PARENT, &nd);
1985 if (error)
1986 goto out;
1987 dentry = lookup_create(&nd, 1);
1988 error = PTR_ERR(dentry);
1989 if (IS_ERR(dentry))
1990 goto out_unlock;
1991
1992 if (!IS_POSIXACL(nd.dentry->d_inode))
1993 mode &= ~current->fs->umask;
1994 error = vfs_mkdir(nd.dentry->d_inode, dentry, mode);
1995 dput(dentry);
1996out_unlock:
1997 mutex_unlock(&nd.dentry->d_inode->i_mutex);
1998 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999out:
Dave Hansen6902d922006-09-30 23:29:01 -07002000 putname(tmp);
2001out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 return error;
2003}
2004
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002005asmlinkage long sys_mkdir(const char __user *pathname, int mode)
2006{
2007 return sys_mkdirat(AT_FDCWD, pathname, mode);
2008}
2009
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010/*
2011 * We try to drop the dentry early: we should have
2012 * a usage count of 2 if we're the only user of this
2013 * dentry, and if that is true (possibly after pruning
2014 * the dcache), then we drop the dentry now.
2015 *
2016 * A low-level filesystem can, if it choses, legally
2017 * do a
2018 *
2019 * if (!d_unhashed(dentry))
2020 * return -EBUSY;
2021 *
2022 * if it cannot handle the case of removing a directory
2023 * that is still in use by something else..
2024 */
2025void dentry_unhash(struct dentry *dentry)
2026{
2027 dget(dentry);
Vasily Averindc168422006-12-06 20:37:07 -08002028 shrink_dcache_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 spin_lock(&dcache_lock);
2030 spin_lock(&dentry->d_lock);
2031 if (atomic_read(&dentry->d_count) == 2)
2032 __d_drop(dentry);
2033 spin_unlock(&dentry->d_lock);
2034 spin_unlock(&dcache_lock);
2035}
2036
2037int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2038{
2039 int error = may_delete(dir, dentry, 1);
2040
2041 if (error)
2042 return error;
2043
2044 if (!dir->i_op || !dir->i_op->rmdir)
2045 return -EPERM;
2046
2047 DQUOT_INIT(dir);
2048
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002049 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 dentry_unhash(dentry);
2051 if (d_mountpoint(dentry))
2052 error = -EBUSY;
2053 else {
2054 error = security_inode_rmdir(dir, dentry);
2055 if (!error) {
2056 error = dir->i_op->rmdir(dir, dentry);
2057 if (!error)
2058 dentry->d_inode->i_flags |= S_DEAD;
2059 }
2060 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002061 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 if (!error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 d_delete(dentry);
2064 }
2065 dput(dentry);
2066
2067 return error;
2068}
2069
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002070static long do_rmdir(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071{
2072 int error = 0;
2073 char * name;
2074 struct dentry *dentry;
2075 struct nameidata nd;
2076
2077 name = getname(pathname);
2078 if(IS_ERR(name))
2079 return PTR_ERR(name);
2080
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002081 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 if (error)
2083 goto exit;
2084
2085 switch(nd.last_type) {
2086 case LAST_DOTDOT:
2087 error = -ENOTEMPTY;
2088 goto exit1;
2089 case LAST_DOT:
2090 error = -EINVAL;
2091 goto exit1;
2092 case LAST_ROOT:
2093 error = -EBUSY;
2094 goto exit1;
2095 }
Ingo Molnarf2eace22006-07-03 00:25:05 -07002096 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002097 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 error = PTR_ERR(dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002099 if (IS_ERR(dentry))
2100 goto exit2;
2101 error = vfs_rmdir(nd.dentry->d_inode, dentry);
2102 dput(dentry);
2103exit2:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002104 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105exit1:
2106 path_release(&nd);
2107exit:
2108 putname(name);
2109 return error;
2110}
2111
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002112asmlinkage long sys_rmdir(const char __user *pathname)
2113{
2114 return do_rmdir(AT_FDCWD, pathname);
2115}
2116
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117int vfs_unlink(struct inode *dir, struct dentry *dentry)
2118{
2119 int error = may_delete(dir, dentry, 0);
2120
2121 if (error)
2122 return error;
2123
2124 if (!dir->i_op || !dir->i_op->unlink)
2125 return -EPERM;
2126
2127 DQUOT_INIT(dir);
2128
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002129 mutex_lock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 if (d_mountpoint(dentry))
2131 error = -EBUSY;
2132 else {
2133 error = security_inode_unlink(dir, dentry);
2134 if (!error)
2135 error = dir->i_op->unlink(dir, dentry);
2136 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002137 mutex_unlock(&dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138
2139 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2140 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2141 d_delete(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142 }
Robert Love0eeca282005-07-12 17:06:03 -04002143
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 return error;
2145}
2146
2147/*
2148 * Make sure that the actual truncation of the file will occur outside its
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002149 * directory's i_mutex. Truncate can take a long time if there is a lot of
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 * writeout happening, and we don't want to prevent access to the directory
2151 * while waiting on the I/O.
2152 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002153static long do_unlinkat(int dfd, const char __user *pathname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154{
2155 int error = 0;
2156 char * name;
2157 struct dentry *dentry;
2158 struct nameidata nd;
2159 struct inode *inode = NULL;
2160
2161 name = getname(pathname);
2162 if(IS_ERR(name))
2163 return PTR_ERR(name);
2164
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002165 error = do_path_lookup(dfd, name, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 if (error)
2167 goto exit;
2168 error = -EISDIR;
2169 if (nd.last_type != LAST_NORM)
2170 goto exit1;
Ingo Molnarf2eace22006-07-03 00:25:05 -07002171 mutex_lock_nested(&nd.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
Christoph Hellwig49705b72005-11-08 21:35:06 -08002172 dentry = lookup_hash(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173 error = PTR_ERR(dentry);
2174 if (!IS_ERR(dentry)) {
2175 /* Why not before? Because we want correct error value */
2176 if (nd.last.name[nd.last.len])
2177 goto slashes;
2178 inode = dentry->d_inode;
2179 if (inode)
2180 atomic_inc(&inode->i_count);
2181 error = vfs_unlink(nd.dentry->d_inode, dentry);
2182 exit2:
2183 dput(dentry);
2184 }
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002185 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 if (inode)
2187 iput(inode); /* truncate the inode here */
2188exit1:
2189 path_release(&nd);
2190exit:
2191 putname(name);
2192 return error;
2193
2194slashes:
2195 error = !dentry->d_inode ? -ENOENT :
2196 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2197 goto exit2;
2198}
2199
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002200asmlinkage long sys_unlinkat(int dfd, const char __user *pathname, int flag)
2201{
2202 if ((flag & ~AT_REMOVEDIR) != 0)
2203 return -EINVAL;
2204
2205 if (flag & AT_REMOVEDIR)
2206 return do_rmdir(dfd, pathname);
2207
2208 return do_unlinkat(dfd, pathname);
2209}
2210
2211asmlinkage long sys_unlink(const char __user *pathname)
2212{
2213 return do_unlinkat(AT_FDCWD, pathname);
2214}
2215
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname, int mode)
2217{
2218 int error = may_create(dir, dentry, NULL);
2219
2220 if (error)
2221 return error;
2222
2223 if (!dir->i_op || !dir->i_op->symlink)
2224 return -EPERM;
2225
2226 error = security_inode_symlink(dir, dentry, oldname);
2227 if (error)
2228 return error;
2229
2230 DQUOT_INIT(dir);
2231 error = dir->i_op->symlink(dir, dentry, oldname);
Stephen Smalleya74574a2005-09-09 13:01:44 -07002232 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002233 fsnotify_create(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 return error;
2235}
2236
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002237asmlinkage long sys_symlinkat(const char __user *oldname,
2238 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239{
2240 int error = 0;
2241 char * from;
2242 char * to;
Dave Hansen6902d922006-09-30 23:29:01 -07002243 struct dentry *dentry;
2244 struct nameidata nd;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
2246 from = getname(oldname);
2247 if(IS_ERR(from))
2248 return PTR_ERR(from);
2249 to = getname(newname);
2250 error = PTR_ERR(to);
Dave Hansen6902d922006-09-30 23:29:01 -07002251 if (IS_ERR(to))
2252 goto out_putname;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
Dave Hansen6902d922006-09-30 23:29:01 -07002254 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
2255 if (error)
2256 goto out;
2257 dentry = lookup_create(&nd, 0);
2258 error = PTR_ERR(dentry);
2259 if (IS_ERR(dentry))
2260 goto out_unlock;
2261
2262 error = vfs_symlink(nd.dentry->d_inode, dentry, from, S_IALLUGO);
2263 dput(dentry);
2264out_unlock:
2265 mutex_unlock(&nd.dentry->d_inode->i_mutex);
2266 path_release(&nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267out:
Dave Hansen6902d922006-09-30 23:29:01 -07002268 putname(to);
2269out_putname:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 putname(from);
2271 return error;
2272}
2273
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002274asmlinkage long sys_symlink(const char __user *oldname, const char __user *newname)
2275{
2276 return sys_symlinkat(oldname, AT_FDCWD, newname);
2277}
2278
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2280{
2281 struct inode *inode = old_dentry->d_inode;
2282 int error;
2283
2284 if (!inode)
2285 return -ENOENT;
2286
2287 error = may_create(dir, new_dentry, NULL);
2288 if (error)
2289 return error;
2290
2291 if (dir->i_sb != inode->i_sb)
2292 return -EXDEV;
2293
2294 /*
2295 * A link to an append-only or immutable file cannot be created.
2296 */
2297 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2298 return -EPERM;
2299 if (!dir->i_op || !dir->i_op->link)
2300 return -EPERM;
2301 if (S_ISDIR(old_dentry->d_inode->i_mode))
2302 return -EPERM;
2303
2304 error = security_inode_link(old_dentry, dir, new_dentry);
2305 if (error)
2306 return error;
2307
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002308 mutex_lock(&old_dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 DQUOT_INIT(dir);
2310 error = dir->i_op->link(old_dentry, dir, new_dentry);
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002311 mutex_unlock(&old_dentry->d_inode->i_mutex);
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002312 if (!error)
Amy Griffisf38aa942005-11-03 15:57:06 +00002313 fsnotify_create(dir, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314 return error;
2315}
2316
2317/*
2318 * Hardlinks are often used in delicate situations. We avoid
2319 * security-related surprises by not following symlinks on the
2320 * newname. --KAB
2321 *
2322 * We don't follow them on the oldname either to be compatible
2323 * with linux 2.0, and to avoid hard-linking to directories
2324 * and other special files. --ADM
2325 */
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002326asmlinkage long sys_linkat(int olddfd, const char __user *oldname,
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002327 int newdfd, const char __user *newname,
2328 int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
2330 struct dentry *new_dentry;
2331 struct nameidata nd, old_nd;
2332 int error;
2333 char * to;
2334
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002335 if ((flags & ~AT_SYMLINK_FOLLOW) != 0)
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002336 return -EINVAL;
2337
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 to = getname(newname);
2339 if (IS_ERR(to))
2340 return PTR_ERR(to);
2341
Ulrich Drepper45c9b112006-06-25 05:49:11 -07002342 error = __user_walk_fd(olddfd, oldname,
2343 flags & AT_SYMLINK_FOLLOW ? LOOKUP_FOLLOW : 0,
2344 &old_nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345 if (error)
2346 goto exit;
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002347 error = do_path_lookup(newdfd, to, LOOKUP_PARENT, &nd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 if (error)
2349 goto out;
2350 error = -EXDEV;
2351 if (old_nd.mnt != nd.mnt)
2352 goto out_release;
2353 new_dentry = lookup_create(&nd, 0);
2354 error = PTR_ERR(new_dentry);
Dave Hansen6902d922006-09-30 23:29:01 -07002355 if (IS_ERR(new_dentry))
2356 goto out_unlock;
2357 error = vfs_link(old_nd.dentry, nd.dentry->d_inode, new_dentry);
2358 dput(new_dentry);
2359out_unlock:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002360 mutex_unlock(&nd.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361out_release:
2362 path_release(&nd);
2363out:
2364 path_release(&old_nd);
2365exit:
2366 putname(to);
2367
2368 return error;
2369}
2370
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002371asmlinkage long sys_link(const char __user *oldname, const char __user *newname)
2372{
Ulrich Drepperc04030e2006-02-24 13:04:21 -08002373 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002374}
2375
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376/*
2377 * The worst of all namespace operations - renaming directory. "Perverted"
2378 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2379 * Problems:
2380 * a) we can get into loop creation. Check is done in is_subdir().
2381 * b) race potential - two innocent renames can create a loop together.
2382 * That's where 4.4 screws up. Current fix: serialization on
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002383 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 * story.
2385 * c) we have to lock _three_ objects - parents and victim (if it exists).
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002386 * And that - after we got ->i_mutex on parents (until then we don't know
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 * whether the target exists). Solution: try to be smart with locking
2388 * order for inodes. We rely on the fact that tree topology may change
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002389 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
Linus Torvalds1da177e2005-04-16 15:20:36 -07002390 * move will be locked. Thus we can rank directories by the tree
2391 * (ancestors first) and rank all non-directories after them.
2392 * That works since everybody except rename does "lock parent, lookup,
Arjan van de Vena11f3a02006-03-23 03:00:33 -08002393 * lock child" and rename is under ->s_vfs_rename_mutex.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 * HOWEVER, it relies on the assumption that any object with ->lookup()
2395 * has no more than 1 dentry. If "hybrid" objects will ever appear,
2396 * we'd better make sure that there's no link(2) for them.
2397 * d) some filesystems don't support opened-but-unlinked directories,
2398 * either because of layout or because they are not ready to deal with
2399 * all cases correctly. The latter will be fixed (taking this sort of
2400 * stuff into VFS), but the former is not going away. Solution: the same
2401 * trick as in rmdir().
2402 * e) conversion from fhandle to dentry may come in the wrong moment - when
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002403 * we are removing the target. Solution: we will have to grab ->i_mutex
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002405 * ->i_mutex on parents, which works but leads to some truely excessive
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406 * locking].
2407 */
Adrian Bunk75c96f82005-05-05 16:16:09 -07002408static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2409 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410{
2411 int error = 0;
2412 struct inode *target;
2413
2414 /*
2415 * If we are going to change the parent - check write permissions,
2416 * we'll need to flip '..'.
2417 */
2418 if (new_dir != old_dir) {
2419 error = permission(old_dentry->d_inode, MAY_WRITE, NULL);
2420 if (error)
2421 return error;
2422 }
2423
2424 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2425 if (error)
2426 return error;
2427
2428 target = new_dentry->d_inode;
2429 if (target) {
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002430 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431 dentry_unhash(new_dentry);
2432 }
2433 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2434 error = -EBUSY;
2435 else
2436 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2437 if (target) {
2438 if (!error)
2439 target->i_flags |= S_DEAD;
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002440 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 if (d_unhashed(new_dentry))
2442 d_rehash(new_dentry);
2443 dput(new_dentry);
2444 }
Stephen Smalleye31e14e2005-09-09 13:01:45 -07002445 if (!error)
Mark Fasheh349457c2006-09-08 14:22:21 -07002446 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2447 d_move(old_dentry,new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 return error;
2449}
2450
Adrian Bunk75c96f82005-05-05 16:16:09 -07002451static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2452 struct inode *new_dir, struct dentry *new_dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453{
2454 struct inode *target;
2455 int error;
2456
2457 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2458 if (error)
2459 return error;
2460
2461 dget(new_dentry);
2462 target = new_dentry->d_inode;
2463 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002464 mutex_lock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2466 error = -EBUSY;
2467 else
2468 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2469 if (!error) {
Mark Fasheh349457c2006-09-08 14:22:21 -07002470 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471 d_move(old_dentry, new_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002472 }
2473 if (target)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002474 mutex_unlock(&target->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002475 dput(new_dentry);
2476 return error;
2477}
2478
2479int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2480 struct inode *new_dir, struct dentry *new_dentry)
2481{
2482 int error;
2483 int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
Robert Love0eeca282005-07-12 17:06:03 -04002484 const char *old_name;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485
2486 if (old_dentry->d_inode == new_dentry->d_inode)
2487 return 0;
2488
2489 error = may_delete(old_dir, old_dentry, is_dir);
2490 if (error)
2491 return error;
2492
2493 if (!new_dentry->d_inode)
2494 error = may_create(new_dir, new_dentry, NULL);
2495 else
2496 error = may_delete(new_dir, new_dentry, is_dir);
2497 if (error)
2498 return error;
2499
2500 if (!old_dir->i_op || !old_dir->i_op->rename)
2501 return -EPERM;
2502
2503 DQUOT_INIT(old_dir);
2504 DQUOT_INIT(new_dir);
2505
Robert Love0eeca282005-07-12 17:06:03 -04002506 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
2507
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 if (is_dir)
2509 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
2510 else
2511 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
2512 if (!error) {
Robert Love0eeca282005-07-12 17:06:03 -04002513 const char *new_name = old_dentry->d_name.name;
John McCutchan89204c42005-08-15 12:13:28 -04002514 fsnotify_move(old_dir, new_dir, old_name, new_name, is_dir,
2515 new_dentry->d_inode, old_dentry->d_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002516 }
Robert Love0eeca282005-07-12 17:06:03 -04002517 fsnotify_oldname_free(old_name);
2518
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519 return error;
2520}
2521
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002522static int do_rename(int olddfd, const char *oldname,
2523 int newdfd, const char *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524{
2525 int error = 0;
2526 struct dentry * old_dir, * new_dir;
2527 struct dentry * old_dentry, *new_dentry;
2528 struct dentry * trap;
2529 struct nameidata oldnd, newnd;
2530
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002531 error = do_path_lookup(olddfd, oldname, LOOKUP_PARENT, &oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 if (error)
2533 goto exit;
2534
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002535 error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 if (error)
2537 goto exit1;
2538
2539 error = -EXDEV;
2540 if (oldnd.mnt != newnd.mnt)
2541 goto exit2;
2542
2543 old_dir = oldnd.dentry;
2544 error = -EBUSY;
2545 if (oldnd.last_type != LAST_NORM)
2546 goto exit2;
2547
2548 new_dir = newnd.dentry;
2549 if (newnd.last_type != LAST_NORM)
2550 goto exit2;
2551
2552 trap = lock_rename(new_dir, old_dir);
2553
Christoph Hellwig49705b72005-11-08 21:35:06 -08002554 old_dentry = lookup_hash(&oldnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002555 error = PTR_ERR(old_dentry);
2556 if (IS_ERR(old_dentry))
2557 goto exit3;
2558 /* source must exist */
2559 error = -ENOENT;
2560 if (!old_dentry->d_inode)
2561 goto exit4;
2562 /* unless the source is a directory trailing slashes give -ENOTDIR */
2563 if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
2564 error = -ENOTDIR;
2565 if (oldnd.last.name[oldnd.last.len])
2566 goto exit4;
2567 if (newnd.last.name[newnd.last.len])
2568 goto exit4;
2569 }
2570 /* source should not be ancestor of target */
2571 error = -EINVAL;
2572 if (old_dentry == trap)
2573 goto exit4;
Christoph Hellwig49705b72005-11-08 21:35:06 -08002574 new_dentry = lookup_hash(&newnd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 error = PTR_ERR(new_dentry);
2576 if (IS_ERR(new_dentry))
2577 goto exit4;
2578 /* target should not be an ancestor of source */
2579 error = -ENOTEMPTY;
2580 if (new_dentry == trap)
2581 goto exit5;
2582
2583 error = vfs_rename(old_dir->d_inode, old_dentry,
2584 new_dir->d_inode, new_dentry);
2585exit5:
2586 dput(new_dentry);
2587exit4:
2588 dput(old_dentry);
2589exit3:
2590 unlock_rename(new_dir, old_dir);
2591exit2:
2592 path_release(&newnd);
2593exit1:
2594 path_release(&oldnd);
2595exit:
2596 return error;
2597}
2598
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002599asmlinkage long sys_renameat(int olddfd, const char __user *oldname,
2600 int newdfd, const char __user *newname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601{
2602 int error;
2603 char * from;
2604 char * to;
2605
2606 from = getname(oldname);
2607 if(IS_ERR(from))
2608 return PTR_ERR(from);
2609 to = getname(newname);
2610 error = PTR_ERR(to);
2611 if (!IS_ERR(to)) {
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002612 error = do_rename(olddfd, from, newdfd, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 putname(to);
2614 }
2615 putname(from);
2616 return error;
2617}
2618
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002619asmlinkage long sys_rename(const char __user *oldname, const char __user *newname)
2620{
2621 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
2622}
2623
Linus Torvalds1da177e2005-04-16 15:20:36 -07002624int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
2625{
2626 int len;
2627
2628 len = PTR_ERR(link);
2629 if (IS_ERR(link))
2630 goto out;
2631
2632 len = strlen(link);
2633 if (len > (unsigned) buflen)
2634 len = buflen;
2635 if (copy_to_user(buffer, link, len))
2636 len = -EFAULT;
2637out:
2638 return len;
2639}
2640
2641/*
2642 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
2643 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
2644 * using) it for any given inode is up to filesystem.
2645 */
2646int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2647{
2648 struct nameidata nd;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002649 void *cookie;
2650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 nd.depth = 0;
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002652 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
2653 if (!IS_ERR(cookie)) {
2654 int res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655 if (dentry->d_inode->i_op->put_link)
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002656 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
2657 cookie = ERR_PTR(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 }
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002659 return PTR_ERR(cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660}
2661
2662int vfs_follow_link(struct nameidata *nd, const char *link)
2663{
2664 return __vfs_follow_link(nd, link);
2665}
2666
2667/* get the link contents into pagecache */
2668static char *page_getlink(struct dentry * dentry, struct page **ppage)
2669{
2670 struct page * page;
2671 struct address_space *mapping = dentry->d_inode->i_mapping;
Pekka Enberg090d2b12006-06-23 02:05:08 -07002672 page = read_mapping_page(mapping, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (IS_ERR(page))
2674 goto sync_fail;
2675 wait_on_page_locked(page);
2676 if (!PageUptodate(page))
2677 goto async_fail;
2678 *ppage = page;
2679 return kmap(page);
2680
2681async_fail:
2682 page_cache_release(page);
2683 return ERR_PTR(-EIO);
2684
2685sync_fail:
2686 return (char*)page;
2687}
2688
2689int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
2690{
2691 struct page *page = NULL;
2692 char *s = page_getlink(dentry, &page);
2693 int res = vfs_readlink(dentry,buffer,buflen,s);
2694 if (page) {
2695 kunmap(page);
2696 page_cache_release(page);
2697 }
2698 return res;
2699}
2700
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002701void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002703 struct page *page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704 nd_set_link(nd, page_getlink(dentry, &page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002705 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706}
2707
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002708void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709{
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002710 struct page *page = cookie;
2711
2712 if (page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 kunmap(page);
2714 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 }
2716}
2717
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002718int __page_symlink(struct inode *inode, const char *symname, int len,
2719 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720{
2721 struct address_space *mapping = inode->i_mapping;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002722 struct page *page;
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002723 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724 char *kaddr;
2725
NeilBrown7e53cac2006-03-25 03:07:57 -08002726retry:
Dmitriy Monakhovbeb497a2007-02-16 01:27:18 -08002727 err = -ENOMEM;
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002728 page = find_or_create_page(mapping, 0, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 if (!page)
2730 goto fail;
2731 err = mapping->a_ops->prepare_write(NULL, page, 0, len-1);
NeilBrown7e53cac2006-03-25 03:07:57 -08002732 if (err == AOP_TRUNCATED_PAGE) {
2733 page_cache_release(page);
2734 goto retry;
2735 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736 if (err)
2737 goto fail_map;
2738 kaddr = kmap_atomic(page, KM_USER0);
2739 memcpy(kaddr, symname, len-1);
2740 kunmap_atomic(kaddr, KM_USER0);
NeilBrown7e53cac2006-03-25 03:07:57 -08002741 err = mapping->a_ops->commit_write(NULL, page, 0, len-1);
2742 if (err == AOP_TRUNCATED_PAGE) {
2743 page_cache_release(page);
2744 goto retry;
2745 }
2746 if (err)
2747 goto fail_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002748 /*
2749 * Notice that we are _not_ going to block here - end of page is
2750 * unmapped, so this will only try to map the rest of page, see
2751 * that it is unmapped (typically even will not look into inode -
2752 * ->i_size will be enough for everything) and zero it out.
2753 * OTOH it's obviously correct and should make the page up-to-date.
2754 */
2755 if (!PageUptodate(page)) {
2756 err = mapping->a_ops->readpage(NULL, page);
NeilBrown7e53cac2006-03-25 03:07:57 -08002757 if (err != AOP_TRUNCATED_PAGE)
2758 wait_on_page_locked(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 } else {
2760 unlock_page(page);
2761 }
2762 page_cache_release(page);
2763 if (err < 0)
2764 goto fail;
2765 mark_inode_dirty(inode);
2766 return 0;
2767fail_map:
2768 unlock_page(page);
2769 page_cache_release(page);
2770fail:
2771 return err;
2772}
2773
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002774int page_symlink(struct inode *inode, const char *symname, int len)
2775{
2776 return __page_symlink(inode, symname, len,
2777 mapping_gfp_mask(inode->i_mapping));
2778}
2779
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002780const struct inode_operations page_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 .readlink = generic_readlink,
2782 .follow_link = page_follow_link_light,
2783 .put_link = page_put_link,
2784};
2785
2786EXPORT_SYMBOL(__user_walk);
Ulrich Drepper5590ff02006-01-18 17:43:53 -08002787EXPORT_SYMBOL(__user_walk_fd);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788EXPORT_SYMBOL(follow_down);
2789EXPORT_SYMBOL(follow_up);
2790EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
2791EXPORT_SYMBOL(getname);
2792EXPORT_SYMBOL(lock_rename);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793EXPORT_SYMBOL(lookup_one_len);
2794EXPORT_SYMBOL(page_follow_link_light);
2795EXPORT_SYMBOL(page_put_link);
2796EXPORT_SYMBOL(page_readlink);
Kirill Korotaev0adb25d2006-03-11 03:27:13 -08002797EXPORT_SYMBOL(__page_symlink);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798EXPORT_SYMBOL(page_symlink);
2799EXPORT_SYMBOL(page_symlink_inode_operations);
2800EXPORT_SYMBOL(path_lookup);
2801EXPORT_SYMBOL(path_release);
2802EXPORT_SYMBOL(path_walk);
2803EXPORT_SYMBOL(permission);
Christoph Hellwige4543ed2005-11-08 21:35:04 -08002804EXPORT_SYMBOL(vfs_permission);
Christoph Hellwig8c744fb2005-11-08 21:35:04 -08002805EXPORT_SYMBOL(file_permission);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806EXPORT_SYMBOL(unlock_rename);
2807EXPORT_SYMBOL(vfs_create);
2808EXPORT_SYMBOL(vfs_follow_link);
2809EXPORT_SYMBOL(vfs_link);
2810EXPORT_SYMBOL(vfs_mkdir);
2811EXPORT_SYMBOL(vfs_mknod);
2812EXPORT_SYMBOL(generic_permission);
2813EXPORT_SYMBOL(vfs_readlink);
2814EXPORT_SYMBOL(vfs_rename);
2815EXPORT_SYMBOL(vfs_rmdir);
2816EXPORT_SYMBOL(vfs_symlink);
2817EXPORT_SYMBOL(vfs_unlink);
2818EXPORT_SYMBOL(dentry_unhash);
2819EXPORT_SYMBOL(generic_readlink);